diff options
73 files changed, 8276 insertions, 3189 deletions
@@ -25,6 +25,8 @@ /vsproject/*.depend /vsproject/*.opendb +/run_rustc/output + /gitstats /bnf/*.o @@ -1,6 +1,7 @@ set expandtab set sts=4 ts=4 sw=4 set list +set listchars=eol:$,tab:>- highlight ExtraWhitespace ctermbg=darkgreen guibg=darkgreen autocmd Syntax * syn match ExtraWhitespace /\s\+$\| \+\ze\t/ diff --git a/.travis.yml b/.travis.yml index c9fd9118..d38cd4b5 100644 --- a/.travis.yml +++ b/.travis.yml @@ -6,20 +6,22 @@ addons: apt: sources: - ubuntu-toolchain-r-test + - george-edison55-precise-backports packages: - g++-6 - zlib1g-dev + - cmake cmake-data install: # Build mrustc - make RUSTCSRC - CC=gcc-6 CXX=g++-6 make -# Tests! (check that they parse, and keep going) script: -# libstd +# libstd and hello_world - CC=gcc-6 make test TAIL_COUNT=2 -# rustc (not the actual binary, because that doesn't emit a file yet) - - CC=gcc-6 make output/librustc_driver.hir TAIL_COUNT=15 +# rustc +# - CC=gcc-6 make output/rustc TAIL_COUNT=15 # Tests -# - CC=gcc-6 make rust_tests-run-pass RUST_TESTS_FINAL_STAGE=expand -k + - CC=gcc-6 make local_tests -k + - CC=gcc-6 make rust_tests -k @@ -44,6 +44,11 @@ CXXFLAGS += -Wno-pessimizing-move CXXFLAGS += -Wno-misleading-indentation #CXXFLAGS += -Wno-unused-private-field + +# - Flags to pass to all mrustc invocations +RUST_FLAGS := --cfg debug_assertions +RUST_FLAGS += -g + SHELL = bash ifeq ($(DBGTPL),) @@ -79,6 +84,7 @@ OBJ += expand/include.o OBJ += expand/env.o OBJ += expand/test.o OBJ += expand/rustc_diagnostics.o +OBJ += expand/test_harness.o OBJ += macro_rules/mod.o macro_rules/eval.o macro_rules/parse.o OBJ += resolve/use.o resolve/index.o resolve/absolute.o OBJ += hir/from_ast.o hir/from_ast_expr.o @@ -92,16 +98,19 @@ OBJ += hir_typeck/outer.o hir_typeck/common.o hir_typeck/helpers.o hir_typeck/st OBJ += hir_typeck/expr_visit.o OBJ += hir_typeck/expr_cs.o OBJ += hir_typeck/expr_check.o -OBJ += hir_expand/annotate_value_usage.o hir_expand/closures.o hir_expand/ufcs_everything.o +OBJ += hir_expand/annotate_value_usage.o hir_expand/closures.o +OBJ += hir_expand/ufcs_everything.o OBJ += hir_expand/reborrow.o hir_expand/erased_types.o hir_expand/vtable.o OBJ += hir_expand/const_eval_full.o OBJ += mir/mir.o mir/mir_ptr.o OBJ += mir/dump.o mir/helpers.o mir/visit_crate_mir.o OBJ += mir/from_hir.o mir/from_hir_match.o mir/mir_builder.o OBJ += mir/check.o mir/cleanup.o mir/optimise.o +OBJ += mir/check_full.o OBJ += hir/serialise.o hir/deserialise.o hir/serialise_lowlevel.o OBJ += trans/trans_list.o trans/mangling.o OBJ += trans/enumerate.o trans/monomorphise.o trans/codegen.o trans/codegen_c.o +OBJ += trans/target.o PCHS := ast/ast.hpp @@ -127,17 +136,30 @@ output/lib%.hir: $(RUSTCSRC)src/lib%/lib.rs $(RUSTCSRC) $(BIN) @echo "--- [MRUSTC] $@" @mkdir -p output/ @rm -f $@ - $(DBG) $(ENV_$@) $(BIN) $< -o $@ $(ARGS_$@) $(PIPECMD) + $(DBG) $(ENV_$@) $(BIN) $< -o $@ $(RUST_FLAGS) $(ARGS_$@) $(PIPECMD) # # HACK: Work around gdb returning success even if the program crashed @test -e $@ output/lib%.hir: $(RUSTCSRC)src/lib%/src/lib.rs $(RUSTCSRC) $(BIN) @echo "--- [MRUSTC] $@" @mkdir -p output/ @rm -f $@ - $(DBG) $(BIN) $< -o $@ $(PIPECMD) + $(DBG) $(ENV_$@) $(BIN) $< -o $@ $(RUST_FLAGS) $(ARGS_$@) $(PIPECMD) +# # HACK: Work around gdb returning success even if the program crashed + @test -e $@ +output/lib%-test: $(RUSTCSRC)src/lib%/lib.rs $(RUSTCSRC) $(BIN) output/libtest.hir + @echo "--- [MRUSTC] --test -o $@" + @mkdir -p output/ + @rm -f $@ + $(DBG) $(ENV_$@) $(BIN) --test $< -o $@ -L output/libs $(RUST_FLAGS) $(ARGS_$@) $(PIPECMD) +# # HACK: Work around gdb returning success even if the program crashed + @test -e $@ +output/lib%-test: $(RUSTCSRC)src/lib%/src/lib.rs $(RUSTCSRC) $(BIN) output/libtest.hir + @echo "--- [MRUSTC] $@" + @mkdir -p output/ + @rm -f $@ + $(DBG) $(ENV_$@) $(BIN) --test $< -o $@ -L output/libs $(RUST_FLAGS) $(ARGS_$@) $(PIPECMD) # # HACK: Work around gdb returning success even if the program crashed @test -e $@ - fcn_extcrate = $(patsubst %,output/lib%.hir,$(1)) fn_getdeps = \ @@ -163,13 +185,13 @@ output/librustc_llvm.hir: $(LLVM_LINKAGE_FILE) RUSTC_LLVM_LINKAGE: $(LLVM_LINKAGE_FILE) output/librustc_llvm_build: rustc-nightly/src/librustc_llvm/build.rs $(call fcn_extcrate, std gcc build_helper alloc_system panic_abort) @echo "--- [MRUSTC] $@" - $(BIN) $< -o $@ -L output/libs $(PIPECMD) + $(BIN) $< -o $@ -L output/libs $(RUST_FLAGS) $(PIPECMD) output/libgcc.hir: crates.io/gcc-0.3.28/src/lib.rs $(BIN) output/libstd.hir @echo "--- [MRUSTC] $@" - $(BIN) $< -o $@ --crate-type rlib --crate-name gcc $(PIPECMD) + $(BIN) $< -o $@ --crate-type rlib --crate-name gcc $(RUST_FLAGS) $(PIPECMD) output/libbuild_helper.hir: rustc-nightly/src/build_helper/lib.rs $(BIN) output/libstd.hir @echo "--- [MRUSTC] $@" - $(BIN) $< -o $@ --crate-type rlib --crate-name build_helper $(PIPECMD) + $(BIN) $< -o $@ --crate-type rlib --crate-name build_helper $(RUST_FLAGS) $(PIPECMD) crates.io/%/src/lib.rs: crates.io/%.tar.gz tar -xf $< -C crates.io/ @@ -199,7 +221,7 @@ output/cargo_libflate/libminiz.a: output/libflate_build output/libflate_build: rustc-nightly/src/libflate/build.rs $(call fcn_extcrate, std gcc alloc_system panic_abort) @echo "--- [MRUSTC] $@" - $(BIN) $< -o $@ -L output/libs $(PIPECMD) + $(BIN) $< -o $@ -L output/libs $(RUST_FLAGS) $(PIPECMD) ARGS_output/librustc_llvm.hir := --cfg llvm_component=x86 --cfg cargobuild ENV_output/librustc_llvm.hir := CFG_LLVM_LINKAGE_FILE=$(LLVM_LINKAGE_FILE) @@ -268,14 +290,14 @@ output/rustc: $(RUSTCSRC)src/rustc/rustc.rs output/librustc_driver.hir output/ru @echo "--- [MRUSTC] $@" @mkdir -p output/ @rm -f $@ - $V$(DBG) $(BIN) $< -o $@ -L output/libs $$(cat output/rustc_link_opts.txt output/rustc_link_opts-libflate.txt) -l stdc++ $(PIPECMD) + $V$(DBG) $(BIN) $< -o $@ -L output/libs $$(cat output/rustc_link_opts.txt output/rustc_link_opts-libflate.txt) -l stdc++ $(RUST_FLAGS) $(PIPECMD) # # HACK: Work around gdb returning success even if the program crashed @test -e $@ .PHONY: RUSTCSRC RUSTCSRC: $(RUSTCSRC) -$(RUSTCSRC): rust-nightly-date +$(RUSTCSRC): rust-nightly-date rust_src.patch @export DL_RUST_DATE=$$(cat rust-nightly-date); \ export DISK_RUST_DATE=$$([ -f $(RUSTC_SRC_DL) ] && cat $(RUSTC_SRC_DL)); \ if [ "$$DL_RUST_DATE" != "$$DISK_RUST_DATE" ]; then \ @@ -284,6 +306,7 @@ $(RUSTCSRC): rust-nightly-date rm -rf rustc-nightly; \ curl -sS https://static.rust-lang.org/dist/$${DL_RUST_DATE}/rustc-nightly-src.tar.gz -o rustc-nightly-src.tar.gz; \ tar -xf rustc-nightly-src.tar.gz --transform 's~^rustc-nightly-src~rustc-nightly~'; \ + patch -p0 < rust_src.patch; \ echo "$$DL_RUST_DATE" > $(RUSTC_SRC_DL); \ fi @@ -311,55 +334,503 @@ $(RUSTCSRC)build/Makefile: $(RUSTCSRC)src/llvm/CMakeLists.txt $Vcd $(RUSTCSRC)build && cmake $(addprefix -D , $(LLVM_CMAKE_OPTS)) ../src/llvm +# MRUSTC-specific tests +.PHONY: local_tests +local_tests: $(patsubst samples/test/%.rs,output/local_test/%_out.txt,$(wildcard samples/test/*.rs)) + +output/local_test/%_out.txt: output/local_test/% + ./$< > $@ +output/local_test/%: samples/test/%.rs $(BIN) + mkdir -p $(dir $@) + $(BIN) -L output/libs -g $< -o $@ --test $(PIPECMD) + # # RUSTC TESTS # -.PHONY: rust_tests +.PHONY: rust_tests local_tests RUST_TESTS_DIR := $(RUSTCSRC)src/test/ -rust_tests: rust_tests-run-pass rust_tests-run-fail +rust_tests: rust_tests-run-pass +# rust_tests-run-fail # rust_tests-compile-fail -# - Require external symbols that aren't generated. -DISABLED_TESTS = run-pass/abi-sysv64-arg-passing run-pass/abi-sysv64-register-usage run-pass/anon-extern-mod run-pass/anon-extern-mod-cross-crate-2 +DISABLED_TESTS := # - NOT A TEST DISABLED_TESTS += run-pass/backtrace-debuginfo-aux -# - asm! is hard to trnaslate -DISABLED_TESTS += run-pass/asm-in-out-operand run-pass/asm-indirect-memory run-pass/asm-out-assign +DISABLED_TESTS += run-pass/mod_file_aux +# - asm! is hard to translate +DISABLED_TESTS += run-pass/abi-sysv64-register-usage +DISABLED_TESTS += run-pass/asm-in-out-operand +DISABLED_TESTS += run-pass/asm-indirect-memory +DISABLED_TESTS += run-pass/asm-out-assign +DISABLED_TESTS += run-pass/i128 # Unknown leader 'r' +DISABLED_TESTS += run-pass/issue-14936 +DISABLED_TESTS += run-pass/issue-32947 +DISABLED_TESTS += run-pass/num-wrapping +DISABLED_TESTS += run-pass/out-of-stack # - Requires jemalloc -DISABLED_TESTS += run-pass/allocator-default run-pass/allocator-override +DISABLED_TESTS += run-pass/allocator-default +DISABLED_TESTS += run-pass/allocator-override # - Bug in inferrence order. DISABLED_TESTS += run-pass/associated-types-conditional-dispatch # - Lazy. -DISABLED_TESTS += run-pass/associated-types-projection-in-where-clause run-pass/autoderef-privacy +DISABLED_TESTS += run-pass/associated-types-projection-in-where-clause # Not normalizing bounds +DISABLED_TESTS += run-pass/cast # Disallows cast from char to i32 +DISABLED_TESTS += run-pass/empty-struct-braces # Empty struct support +DISABLED_TESTS += run-pass/explicit-self-generic # Tries to use HashMap as an iterator +DISABLED_TESTS += run-pass/extern-compare-with-return-type # Specialisation with function pointers +DISABLED_TESTS += run-pass/issue-14399 # Inferrence ran though a coercion point. +DISABLED_TESTS += run-pass/issue-26709 # ^ (integer literal) +DISABLED_TESTS += run-pass/issue-20797 # Failed to find impl with associated type, possible incorrect coerce? +DISABLED_TESTS += run-pass/issue-21245 # IntoIterator on core::slice::Iterator ? +DISABLED_TESTS += run-pass/issue-21486 # Type mismatch +DISABLED_TESTS += run-pass/issue-21410 # Infinite recursion +DISABLED_TESTS += run-pass/issue-25439 # ^ +DISABLED_TESTS += run-pass/issue-22629 # Auto trait + UFCS todo +DISABLED_TESTS += run-pass/send-is-not-static-par-for # ^ +DISABLED_TESTS += run-pass/issue-22828 # ^ +DISABLED_TESTS += run-pass/issue-23699 # fn() inferrence +DISABLED_TESTS += run-pass/issue-30371 # destructuring pattern on ! +DISABLED_TESTS += run-pass/issue-33687 # Unit struct implementing FnOnce call +DISABLED_TESTS += run-pass/issue-38033 # Not equating associated type of type param. +DISABLED_TESTS += run-pass/issue-7784 # PartialEq impl +DISABLED_TESTS += run-pass/traits-issue-26339 # ^ +DISABLED_TESTS += run-pass/builtin-superkinds-self-type # ^ +DISABLED_TESTS += run-pass/intrinsic-move-val # ^ +DISABLED_TESTS += run-pass/issue-9951 # Trait impled for i32 +DISABLED_TESTS += run-pass/trait-default-method-xc # ^ +DISABLED_TESTS += run-pass/trait-impl # ^ +DISABLED_TESTS += run-pass/issue-11205 # ^ +DISABLED_TESTS += run-pass/mir_coercions # Coercion to unsafe fn +DISABLED_TESTS += run-pass/typeck-fn-to-unsafe-fn-ptr # ^ +DISABLED_TESTS += run-pass/unsafe-coercion # ^ +DISABLED_TESTS += run-pass/mir_misc_casts # Cast fn to *const isize +DISABLED_TESTS += run-pass/never-result # ! not correctly unifiying +DISABLED_TESTS += run-pass/reachable-unnameable-items # assert Struct::is_Named() +DISABLED_TESTS += run-pass/self-impl # Unable to infer +DISABLED_TESTS += run-pass/trait-copy-guessing # ^ +DISABLED_TESTS += run-pass/issue-20575 # ^ +DISABLED_TESTS += run-pass/sync-send-iterators-in-libcore # Send for Range<_> +DISABLED_TESTS += run-pass/const-bound # Sync for Box<_> +DISABLED_TESTS += run-pass/traits-repeated-supertrait # Type mismatch, i64 and u64 +DISABLED_TESTS += run-pass/trans-object-shim # fn cast to other fn as type annotation +DISABLED_TESTS += run-pass/variadic-ffi # variadics not supported +DISABLED_TESTS += run-pass/weird-exprs # Line 17, let _ = return; result type +DISABLED_TESTS += run-pass/where-for-self # Failed deref coercion? +DISABLED_TESTS += run-pass/union/union-backcomp # ? discarded value? +# - Typecheck - `_` type with no ivar index assigned +DISABLED_TESTS += run-pass/coerce-expect-unsized +DISABLED_TESTS += run-pass/coerce-unify-return +DISABLED_TESTS += run-pass/issue-23485 +DISABLED_TESTS += run-pass/issue-26805 +DISABLED_TESTS += run-pass/match-vec-alternatives +DISABLED_TESTS += run-pass/pure-sum +DISABLED_TESTS += run-pass/struct-aliases +# - Lazy (Typecheck - Leftover rules) +DISABLED_TESTS += run-pass/regions-infer-borrow-scope-addr-of # Didn't unify literal ivar +DISABLED_TESTS += run-pass/swap-2 # ^ +DISABLED_TESTS += run-pass/slice_binary_search # Didn't detect infer possiblity (&str, &String) +# - Lazy (Typecheck - Array unsize) +DISABLED_TESTS += run-pass/byte-literals # Over-eager inferrence +DISABLED_TESTS += run-pass/cast-rfc0401-vtable-kinds # Spare rules +DISABLED_TESTS += run-pass/cast-rfc0401 # Skipped coerce unsized +DISABLED_TESTS += run-pass/dst-struct-sole # Spare rules +DISABLED_TESTS += run-pass/dst-struct # Spare rules +DISABLED_TESTS += run-pass/issue-23261 # Spare rules +DISABLED_TESTS += run-pass/fat-ptr-cast # Skiped coerce unsized +DISABLED_TESTS += run-pass/issue-21562 # ERROR - Borrow->Pointer and Unsize in one +DISABLED_TESTS += run-pass/raw-fat-ptr # ^ +DISABLED_TESTS += run-pass/issue-9382 # Missing coercion point (struct field) +DISABLED_TESTS += run-pass/overloaded-autoderef-indexing # Missing coercion point (struct field) +DISABLED_TESTS += run-pass/match-byte-array-patterns # Byte string match +DISABLED_TESTS += run-pass/mir_raw_fat_ptr # Byte string match +DISABLED_TESTS += run-pass/never_coercions # Missed coerce +## - Lazy (Typecheck + Trait unsize) +#DISABLED_TESTS += run-pass/issue-27105 +#DISABLED_TESTS += run-pass/dst-coerce-rc +DISABLED_TESTS += run-pass/dst-coercions # Skipped CoerceUnsize +DISABLED_TESTS += run-pass/dst-raw # Skipped CoerceUnsize +DISABLED_TESTS += run-pass/issue-11677 # Skipped +#DISABLED_TESTS += run-pass/dst-trait +# - Lazy (MIR) +DISABLED_TESTS += run-pass/if-ret # If condition wasn't a bool +DISABLED_TESTS += run-pass/intrinsics-integer # todo - bswap<i8> +DISABLED_TESTS += run-pass/issue-11940 # todo: Match literal Borrow +DISABLED_TESTS += run-pass/mir_build_match_comparisons # - ^ +DISABLED_TESTS += run-pass/issue-13620 # - Todo in cleanup +DISABLED_TESTS += run-pass/vec-matching-fold # todo: Match SplitSlice with tailing (rule gen) +DISABLED_TESTS += run-pass/issue-17877 # - SplitSlice + array +DISABLED_TESTS += run-pass/vec-matching-fixed # ^ +DISABLED_TESTS += run-pass/vec-tail-matching # SplitSlice destructure array +DISABLED_TESTS += run-pass/zero_sized_subslice_match # ^ +DISABLED_TESTS += run-pass/issue-18352 # - Match+const +DISABLED_TESTS += run-pass/issue-28839 # - Move &mut ? +DISABLED_TESTS += run-pass/union/union-inherent-method # ^ ? +DISABLED_TESTS += run-pass/issue-21306 # ^ +DISABLED_TESTS += run-pass/issue-28950 # - Stack overflow in vec! +DISABLED_TESTS += run-pass/mir_heavy_promoted # Stack overflow in array constant +DISABLED_TESTS += run-pass/issue-29227 # - Excessive time in MIR lowering +DISABLED_TESTS += run-pass/issue-15763 # No value avaliable +DISABLED_TESTS += run-pass/issue-18110 # ^ +DISABLED_TESTS += run-pass/issue-30018-nopanic # ^ +DISABLED_TESTS += run-pass/match-bot-2 # ^ +DISABLED_TESTS += run-pass/unreachable-code # ^ +DISABLED_TESTS += run-pass/issue-36936 # - Cast removed +DISABLED_TESTS += run-pass/struct-order-of-eval-1 # Struct init order (fails validation) +DISABLED_TESTS += run-pass/struct-order-of-eval-3 # ^ +DISABLED_TESTS += run-pass/const-enum-vec-index # This is valid code? +# - Lazy (trans) +DISABLED_TESTS += run-pass/issue-21058 # Empty trait object vtable +DISABLED_TESTS += run-pass/issue-25515 # ^ +DISABLED_TESTS += run-pass/issue-35815 # ^ +DISABLED_TESTS += run-pass/issue-29663 # Missing volatile_(load|store) intrinsic +DISABLED_TESTS += run-pass/intrinsic-alignment # Missing pref_align_of intrinsic +DISABLED_TESTS += run-pass/volatile-fat-ptr # ^ +DISABLED_TESTS += run-pass/newtype # Can't handle mutally recursive definitions +DISABLED_TESTS += run-pass/transmute-specialization # Opaque type hit? +DISABLED_TESTS += run-pass/unit-fallback # ! didn't default to () +DISABLED_TESTS += run-pass/issue-33387 # Missing vtable for array +# - HIR resolve +DISABLED_TESTS += run-pass/union/union-generic # Can't find associated type on type param +# - Lazy (misc) +DISABLED_TESTS += run-pass/issue-13494 +DISABLED_TESTS += run-pass/issue-6919 # Literal function pointer +DISABLED_TESTS += run-pass/item-attributes # Attributed function after last statement leads to last statement yielded +DISABLED_TESTS += run-pass/new-box-syntax # todo - placement syntax +DISABLED_TESTS += run-pass/placement-in-syntax # ^ +DISABLED_TESTS += run-pass/pat-tuple-1 # assertion in "Annotate Value Usage" +DISABLED_TESTS += run-pass/pat-tuple-2 # ^ +DISABLED_TESTS += run-pass/pat-tuple-3 # ^ +DISABLED_TESTS += run-pass/pat-tuple-4 # ^ +DISABLED_TESTS += run-pass/paths-in-macro-invocations # MISSING: qualified macro paths +DISABLED_TESTS += run-pass/struct-path-associated-type # non-absolute path for HIR::GenericPath +DISABLED_TESTS += run-pass/struct-path-self # ^ +DISABLED_TESTS += run-pass/ufcs-polymorphic-paths # ^ +# - Resolve +DISABLED_TESTS += run-pass/issue-22546 # None::<u8> handling in patterns +DISABLED_TESTS += run-pass/issue-29540 # Infinite recursion +DISABLED_TESTS += run-pass/issue-38002 # Enum::StructVariant +DISABLED_TESTS += run-pass/match-arm-statics # ^ +DISABLED_TESTS += run-pass/mir_ascription_coercion # Missed item +DISABLED_TESTS += run-pass/type-ascription # Relative path in lowering +DISABLED_TESTS += run-pass/issue-15221 # Macros in patterns +# - Overly-restrictive consteval +DISABLED_TESTS += run-pass/const-cast # Cast from fn() to pointer +DISABLED_TESTS += run-pass/const-autoderef # Expected [u8] +DISABLED_TESTS += run-pass/check-static-mut-slices +DISABLED_TESTS += run-pass/check-static-slice +DISABLED_TESTS += run-pass/const-binops +DISABLED_TESTS += run-pass/const-contents +DISABLED_TESTS += run-pass/const-deref +DISABLED_TESTS += run-pass/const-enum-cast +DISABLED_TESTS += run-pass/const-err +DISABLED_TESTS += run-pass/const-fields-and-indexing +DISABLED_TESTS += run-pass/const-fn-method +DISABLED_TESTS += run-pass/const-fn +DISABLED_TESTS += run-pass/const-str-ptr +DISABLED_TESTS += run-pass/const-vec-of-fns +DISABLED_TESTS += run-pass/diverging-fn-tail-35849 +DISABLED_TESTS += run-pass/enum-vec-initializer +DISABLED_TESTS += run-pass/huge-largest-array +DISABLED_TESTS += run-pass/issue-17233 +DISABLED_TESTS += run-pass/issue-19244 # Missing type info +DISABLED_TESTS += run-pass/issue-22894 # TODO: Deref +DISABLED_TESTS += run-pass/issue-25180 # Closure in const +DISABLED_TESTS += run-pass/issue-27268 # ^ +DISABLED_TESTS += run-pass/issue-28189 # ^ +DISABLED_TESTS += run-pass/issue-25757 # UFCS function pointer +DISABLED_TESTS += run-pass/mir_refs_correct +DISABLED_TESTS += run-pass/vec-fixed-length # Overflow in costeval +DISABLED_TESTS += run-pass/union/union-const-trans # Union literal +# - Type defaults not supported +DISABLED_TESTS += run-pass/default-associated-types +DISABLED_TESTS += run-pass/default_ty_param_default_dependent_associated_type +DISABLED_TESTS += run-pass/default_ty_param_dependent_defaults +DISABLED_TESTS += run-pass/default_ty_param_method_call_test +DISABLED_TESTS += run-pass/default_ty_param_struct_and_type_alias +DISABLED_TESTS += run-pass/default_ty_param_struct +DISABLED_TESTS += run-pass/default_ty_param_trait_impl +DISABLED_TESTS += run-pass/default_ty_param_trait_impl_simple +DISABLED_TESTS += run-pass/default_ty_param_type_alias +DISABLED_TESTS += run-pass/generic-default-type-params-cross-crate +DISABLED_TESTS += run-pass/generic-default-type-params +# - ERROR: Function pointers in consants/statics don't trigger calls +DISABLED_TESTS += run-pass/issue-17718 +DISABLED_TESTS += run-pass/rfc1623 +DISABLED_TESTS += run-pass/static-function-pointer-xc +DISABLED_TESTS += run-pass/static-function-pointer +DISABLED_TESTS += run-pass/const-block-cross-crate-fn +DISABLED_TESTS += run-pass/const-block-item +DISABLED_TESTS += run-pass/const-block +# - Quirks +DISABLED_TESTS += run-pass/autoderef-privacy # No privacy with autoderef +DISABLED_TESTS += run-pass/fn-item-type-zero-sized # fn() items are not ZSTs +DISABLED_TESTS += run-pass/int-abs-overflow # No overflow checks +DISABLED_TESTS += run-pass/issue-18859 # module_path output is differend +DISABLED_TESTS += run-pass/issue-8709 # stringify! output +DISABLED_TESTS += run-pass/tydesc-name # ^ +DISABLED_TESTS += run-pass/concat # ^ +DISABLED_TESTS += run-pass/type-id-higher-rank-2 # lifetimes don't apply in type_id +DISABLED_TESTS += run-pass/type-id-higher-rank # ^ +# - BUG-Expand: macro_rules +DISABLED_TESTS += run-pass/macro-of-higher-order +DISABLED_TESTS += run-pass/macro-pat # :pat doesn't allow MACRO +DISABLED_TESTS += run-pass/macro-reexport-no-intermediate-use # macro_reexport failed +DISABLED_TESTS += run-pass/macro-with-attrs1 # cfg on macro_rules +DISABLED_TESTS += run-pass/shift-near-oflo # Scoping rules +DISABLED_TESTS += run-pass/type-macros-simple # ^ +DISABLED_TESTS += run-pass/stmt_expr_attr_macro_parse # Orderign with :expr and #[] +DISABLED_TESTS += run-pass/sync-send-iterators-in-libcollections # .. should match :expr +DISABLED_TESTS += run-pass/type-macros-hlist # Mismatched arms +# - BUG-Expand: format_args! +DISABLED_TESTS += run-pass/ifmt # Unknown formatting type specifier '*' +# - BUG-Expand: line/column macros don't work properly +DISABLED_TESTS += run-pass/issue-26322 +# - Expand +DISABLED_TESTS += run-pass/issue-11085 # No support for cfg() on enum variants +DISABLED_TESTS += run-pass/lexer-crlf-line-endings-string-literal-doc-comment # Missing include_str! +DISABLED_TESTS += run-pass/syntax-extension-source-utils # ^ +DISABLED_TESTS += run-pass/link-cfg-works # cfg in #[link] +DISABLED_TESTS += run-pass/linkage1 # #[linkage] +DISABLED_TESTS += run-pass/log_syntax-trace_macros-macro-locations # no trace_macros! +DISABLED_TESTS += run-pass/macro-use-all-and-none # missing macro_use feature +DISABLED_TESTS += run-pass/macro-use-both # ^ +DISABLED_TESTS += run-pass/macro-use-one # ^ +DISABLED_TESTS += run-pass/two-macro-use # ^ +DISABLED_TESTS += run-pass/simd-intrinsic-generic-cast # Missing concat_idents! +DISABLED_TESTS += run-pass/simd-intrinsic-generic-comparison # ^ +DISABLED_TESTS += run-pass/smallest-hello-world # missing lang item +DISABLED_TESTS += run-pass/trait-item-inside-macro # macro invocations in traits +DISABLED_TESTS += run-pass/try-operator-custom # `?` carrier +DISABLED_TESTS += run-pass/wrapping-int-api # cfg on match arms +DISABLED_TESTS += run-pass/union/union-c-interop # union derive +DISABLED_TESTS += run-pass/union/union-derive # ^ +DISABLED_TESTS += run-pass/union/union-overwrite # ? MetaItem::as_String() +DISABLED_TESTS += run-pass/union/union-packed # ^ +DISABLED_TESTS += run-pass/union/union-pat-refutability # ^ +# - Parse +DISABLED_TESTS += run-pass/issue-37733 # for<'a,> +DISABLED_TESTS += run-pass/loop-break-value # `break value` +DISABLED_TESTS += run-pass/macro-attribute-expansion # No handling of $expr in attributes +DISABLED_TESTS += run-pass/macro-doc-escapes # Doc comments aren't attributes +DISABLED_TESTS += run-pass/macro-doc-raw-str-hashes # ^ +DISABLED_TESTS += run-pass/macro-interpolation # $block not allowed in place of function body +DISABLED_TESTS += run-pass/macro-stmt # ^ +DISABLED_TESTS += run-pass/macro-tt-followed-by-seq # Mismatched arms? +DISABLED_TESTS += run-pass/struct-field-shorthand # Struct field shorthand +DISABLED_TESTS += run-pass/vec-matching # [a, [b,..].., c] +# HIR Lowering +DISABLED_TESTS += run-pass/union/union-basic # Union struct pattern +# - BUG-Parse: `use *` +DISABLED_TESTS += run-pass/import-glob-crate +DISABLED_TESTS += run-pass/import-prefix-macro +# - BUG-CODEGEN: Missing symbol +DISABLED_TESTS += run-pass/const-enum-ptr +DISABLED_TESTS += run-pass/const-enum-vec-ptr +DISABLED_TESTS += run-pass/const-vecs-and-slices +DISABLED_TESTS += run-pass/issue-5688 +DISABLED_TESTS += run-pass/issue-5917 +DISABLED_TESTS += run-pass/issue-7012 +DISABLED_TESTS += run-pass/issue-29147 # Missing type +DISABLED_TESTS += run-pass/issue-30081 # ^ +DISABLED_TESTS += run-pass/issue-3447 # ^ +DISABLED_TESTS += run-pass/issue-34796 # Missing vtable type (in dep) +DISABLED_TESTS += run-pass/simd-generics # "platform-intrinsics" +DISABLED_TESTS += run-pass/simd-intrinsic-generic-arithmetic # ^ +DISABLED_TESTS += run-pass/simd-intrinsic-generic-elements # ^ +DISABLED_TESTS += run-pass/thread-local-extern-static # Extern static not generated? +# - BUG: Codegen +DISABLED_TESTS += run-pass/unsized3 # Pointer instead of fat pointer +DISABLED_TESTS += run-pass/utf8_idents # C backend doesn't support utf8 idents +# - BUG: Hygine +DISABLED_TESTS += run-pass/hygiene +DISABLED_TESTS += run-pass/hygienic-labels-in-let +DISABLED_TESTS += run-pass/hygienic-labels +DISABLED_TESTS += run-pass/macro-nested_stmt_macros # hygine fires when it shouldn't +# - Test framework required +DISABLED_TESTS += run-pass/core-run-destroy +DISABLED_TESTS += run-pass/exec-env +DISABLED_TESTS += run-pass/issue-16597-empty +DISABLED_TESTS += run-pass/issue-16597 # NOTE: Crashes in resolve +DISABLED_TESTS += run-pass/issue-20823 +DISABLED_TESTS += run-pass/issue-34932 +DISABLED_TESTS += run-pass/issue-36768 +DISABLED_TESTS += run-pass/reexport-test-harness-main +DISABLED_TESTS += run-pass/test-fn-signature-verification-for-explicit-return-type +DISABLED_TESTS += run-pass/test-main-not-dead-attr +DISABLED_TESTS += run-pass/test-main-not-dead +DISABLED_TESTS += run-pass/test-runner-hides-buried-main +DISABLED_TESTS += run-pass/test-runner-hides-main +DISABLED_TESTS += run-pass/test-runner-hides-start +DISABLED_TESTS += run-pass/test-should-fail-good-message +DISABLED_TESTS += run-pass/test-should-panic-attr +# - Makefile test framework quirks +DISABLED_TESTS += run-pass/issue-18913 +DISABLED_TESTS += run-pass/issue-2380-b +DISABLED_TESTS += run-pass/issue-29485 +DISABLED_TESTS += run-pass/svh-add-comment +DISABLED_TESTS += run-pass/svh-add-doc +DISABLED_TESTS += run-pass/svh-add-macro +DISABLED_TESTS += run-pass/svh-add-nothing +DISABLED_TESTS += run-pass/svh-add-redundant-cfg +DISABLED_TESTS += run-pass/svh-add-whitespace +# - Target Features +DISABLED_TESTS += run-pass/crt-static-on-works +DISABLED_TESTS += run-pass/sse2 +# - Infinite loops +DISABLED_TESTS += run-pass/issue-27890 # - Stack exhausted : Resolve +DISABLED_TESTS += run-pass/project-cache-issue-31849 +# - Impl selection +DISABLED_TESTS += run-pass/issue-23208 # Couldn't find an impl for <T/*M:0*/ as ::TheSuperTrait<u32,>>::get +DISABLED_TESTS += run-pass/xcrate-associated-type-defaults # Failed to find an impl +# --- Runtime Errors --- # - Line information that isn't avaliable due to codegen -DISABLED_TESTS += run-pass/backtrace-debuginfo run-pass/backtrace +DISABLED_TESTS += run-pass/backtrace-debuginfo +DISABLED_TESTS += run-pass/backtrace # - No unwind catching support -DISABLED_TESTS += run-pass/binary-heap-panic-safe run-pass/box-of-array-of-drop-1 run-pass/box-of-array-of-drop-2 -# - Infinite loops -DISABLED_TESTS += run-pass/issue-16671 +DISABLED_TESTS += run-pass/binary-heap-panic-safe +DISABLED_TESTS += run-pass/box-of-array-of-drop-1 +DISABLED_TESTS += run-pass/box-of-array-of-drop-2 +DISABLED_TESTS += run-pass/cleanup-rvalue-temp-during-incomplete-alloc +DISABLED_TESTS += run-pass/drop-trait-enum +DISABLED_TESTS += run-pass/intrinsic-move-val-cleanups +DISABLED_TESTS += run-pass/issue-14875 +DISABLED_TESTS += run-pass/issue-25089 +DISABLED_TESTS += run-pass/issue-26655 +DISABLED_TESTS += run-pass/issue-30018-panic +DISABLED_TESTS += run-pass/issue-8460 +DISABLED_TESTS += run-pass/iter-step-overflow-debug +DISABLED_TESTS += run-pass/iter-sum-overflow-debug +DISABLED_TESTS += run-pass/multi-panic +DISABLED_TESTS += run-pass/nested-vec-3 +DISABLED_TESTS += run-pass/no-landing-pads +DISABLED_TESTS += run-pass/panic-handler-chain +DISABLED_TESTS += run-pass/panic-handler-flail-wildly +DISABLED_TESTS += run-pass/panic-handler-set-twice +DISABLED_TESTS += run-pass/panic-in-dtor-drops-fields +DISABLED_TESTS += run-pass/panic-recover-propagate +DISABLED_TESTS += run-pass/sepcomp-unwind +DISABLED_TESTS += run-pass/slice-panic-1 +DISABLED_TESTS += run-pass/slice-panic-2 +DISABLED_TESTS += run-pass/task-stderr +DISABLED_TESTS += run-pass/terminate-in-initializer +DISABLED_TESTS += run-pass/unit-like-struct-drop-run +DISABLED_TESTS += run-pass/unwind-resource +DISABLED_TESTS += run-pass/unwind-unique +DISABLED_TESTS += run-pass/vector-sort-panic-safe +DISABLED_TESTS += run-pass/dynamic-drop +# - Misc +DISABLED_TESTS += run-pass/issue-16671 # Blocks forever +DISABLED_TESTS += run-pass/issue-13027 # Infinite loop (match?) +# - BUG: Incorrect drop order of ? +DISABLED_TESTS += run-pass/issue-23338-ensure-param-drop-order +# - BUG: Incorrect consteval +DISABLED_TESTS += run-pass/issue-23968-const-not-overflow # !0 / 2 incorrect value +# - BUG: Incorrect ordering of read in binops +DISABLED_TESTS += run-pass/issue-27054-primitive-binary-ops +# - BUG: Enum variants not getting correct integer values (negatives) +DISABLED_TESTS += run-pass/discriminant_value +DISABLED_TESTS += run-pass/enum-discr +DISABLED_TESTS += run-pass/enum-disr-val-pretty +DISABLED_TESTS += run-pass/issue-15523-big +DISABLED_TESTS += run-pass/issue-9837 +DISABLED_TESTS += run-pass/signed-shift-const-eval +DISABLED_TESTS += run-pass/tag-variant-disr-val +# - BUG: repr(size) not working +DISABLED_TESTS += run-pass/enum-univariant-repr +DISABLED_TESTS += run-pass/issue-15523 +# - ConstEval: Handling of enum variant casts +DISABLED_TESTS += run-pass/issue-23304-1 +DISABLED_TESTS += run-pass/issue-23304-2 +#DISABLED_TESTS += run-pass/issue-2428 +#DISABLED_TESTS += run-pass/resolve-issue-2428 +#DISABLED_TESTS += run-pass/small-enum-range-edge +# - BUG: Null pointer opt not fully correct +DISABLED_TESTS += run-pass/enum-null-pointer-opt +DISABLED_TESTS += run-pass/nonzero-enum +DISABLED_TESTS += run-pass/nullable-pointer-opt-closures +DISABLED_TESTS += run-pass/nullable-pointer-size +# - BUG: Incorrect enum sizing +DISABLED_TESTS += run-pass/enum-discrim-autosizing +DISABLED_TESTS += run-pass/enum-discrim-manual-sizing +DISABLED_TESTS += run-pass/enum-discrim-width-stuff +DISABLED_TESTS += run-pass/multiple-reprs # no repr handling +DISABLED_TESTS += run-pass/small-enums-with-fields +DISABLED_TESTS += run-pass/type-sizes +DISABLED_TESTS += run-pass/discrim-explicit-23030 +DISABLED_TESTS += run-pass/issue-13902 +# - BUG: Bad floats +DISABLED_TESTS += run-pass/float-nan +DISABLED_TESTS += run-pass/float_math +DISABLED_TESTS += run-pass/floatlits +DISABLED_TESTS += run-pass/intrinsics-math +# - BUG: MIR Generation +DISABLED_TESTS += run-pass/union/union-drop-assign # No drop when assiging to union field +DISABLED_TESTS += run-pass/issue-4734 # Destructor on unused rvalue +DISABLED_TESTS += run-pass/issue-8860 # No drop of un-moved arguments +DISABLED_TESTS += run-pass/issue-15080 # Inifinte loop from incorrect match generation +# - BUG: Codegen +DISABLED_TESTS += run-pass/union/union-transmute # Incorrect union behavior, likey backend UB +DISABLED_TESTS += run-pass/mir_overflow_off # out-of-range shift behavior +DISABLED_TESTS += run-pass/dst-field-align # DST Fields aren't aligned correctly +# - BUG: Codegen - No handling of repr() +DISABLED_TESTS += run-pass/packed-struct-generic-layout +DISABLED_TESTS += run-pass/packed-struct-generic-size +DISABLED_TESTS += run-pass/packed-struct-layout +DISABLED_TESTS += run-pass/packed-struct-size-xc +DISABLED_TESTS += run-pass/packed-struct-size +DISABLED_TESTS += run-pass/packed-struct-vec +DISABLED_TESTS += run-pass/packed-tuple-struct-layout +DISABLED_TESTS += run-pass/packed-tuple-struct-size +# - BUG-Expand: format_args! +DISABLED_TESTS += run-pass/format-ref-cell +# - BUG-Expand: Copy,Clone calls Clone for inner values instead of copying +DISABLED_TESTS += run-pass/deriving-copyclone +# - BUG: Unknown +DISABLED_TESTS += run-pass/process-spawn-with-unicode-params # Bad path for process spawn +DISABLED_TESTS += run-pass/u128 # u128 not very good, unknown where error is DEF_RUST_TESTS = $(sort $(patsubst $(RUST_TESTS_DIR)%.rs,output/rust/%_out.txt,$(wildcard $(RUST_TESTS_DIR)$1/*.rs))) -rust_tests-run-pass: $(filter-out $(patsubst %,output/rust/%_out.txt,$(DISABLED_TESTS)), $(call DEF_RUST_TESTS,run-pass)) +rust_tests-run-pass: $(filter-out $(patsubst %,output/rust/%_out.txt,$(DISABLED_TESTS)), $(call DEF_RUST_TESTS,run-pass) $(call DEF_RUST_TESTS,run-pass/union)) rust_tests-run-fail: $(call DEF_RUST_TESTS,run-fail) + +LIB_TESTS := collections collectionstest rustc_data_structures +RUNTIME_ARGS_output/libcollectionstest-test := --test-threads 1 --skip linked_list::test_ord_nan --skip ::slice::test_box_slice_clone_panics +RUNTIME_ARGS_output/libstd-test := --test-threads 1 --skip :collections::hash::map::test_map::test_index_nonexistent +RUNTIME_ARGS_output/libstd-test += --skip ::collections::hash::map::test_map::test_drops +rust_tests-libs: $(patsubst %,output/lib%-test_out.txt, $(LIB_TESTS)) + #rust_tests-compile-fail: $(call DEF_RUST_TESTS,compile-fail) output/rust/test_run-pass_hello: $(RUST_TESTS_DIR)run-pass/hello.rs output/libstd.hir $(BIN) output/liballoc_system.hir output/libpanic_abort.hir @mkdir -p $(dir $@) @echo "--- [MRUSTC] -o $@" - $(DBG) $(BIN) $< -L output/libs -o $@ $(PIPECMD) - @echo "--- [$@]" - @./$@ + $(DBG) $(BIN) $< -L output/libs -o $@ $(RUST_FLAGS) $(PIPECMD) +output/rust/test_run-pass_hello_out.txt: output/rust/test_run-pass_hello + @echo "--- [$<]" + @./$< | tee $@ +TEST_ARGS_run-pass/cfg-in-crate-1 := --cfg bar TEST_ARGS_run-pass/cfgs-on-items := --cfg fooA --cfg fooB - -output/rust/%: $(RUST_TESTS_DIR)%.rs $(RUSTCSRC) $(BIN) output/libstd.hir output/libtest.hir +TEST_ARGS_run-pass/cfg-macros-foo := --cfg foo +TEST_ARGS_run-pass/cfg_attr := --cfg set1 --cfg set2 +TEST_ARGS_run-pass/issue-11085 := --cfg foo +TEST_ARGS_run-pass/macro-meta-items := --cfg foo +TEST_ARGS_run-pass/issue-21361 := -g +TEST_ARGS_run-pass/syntax-extension-cfg := --cfg foo --cfg 'qux=foo' + +output/rust/%: $(RUST_TESTS_DIR)%.rs $(RUSTCSRC) $(BIN) output/libstd.hir output/libtest.hir output/test_deps/librust_test_helpers.a @mkdir -p $(dir $@) @echo "=== TEST $(patsubst output/rust/%,%,$@)" @echo "--- [MRUSTC] -o $@" - $V$(BIN) $< -o $@ -L output/libs --stop-after $(RUST_TESTS_FINAL_STAGE) $(TEST_ARGS_$*) > $@.txt 2>&1 || (tail -n 1 $@.txt; false) -output/rust/%_out.txt: output/rust/% + $V$(BIN) $< -o $@ -L output/libs -L output/test_deps --stop-after $(RUST_TESTS_FINAL_STAGE) $(TEST_ARGS_$*) > $@.txt 2>&1 || (tail -n 1 $@.txt; false) +output/%_out.txt: output/% @echo "--- [$<]" - @./$< > $@ || (tail -n 1 $@; false) + @./$< $(RUNTIME_ARGS_$<) > $@ || (tail -n 1 $@; mv $@ $@_fail; false) + +output/test_deps/librust_test_helpers.a: output/test_deps/rust_test_helpers.o + ar cur $@ $< +output/test_deps/rust_test_helpers.o: $(RUSTCSRC)src/rt/rust_test_helpers.c + $(CC) -c $< -o $@ output/rust/run-pass/allocator-default.o: output/libstd.hir output/liballoc_jemalloc.hir output/rust/run-pass/allocator-system.o: output/liballoc_system.hir @@ -386,7 +857,7 @@ test_deps_run-pass.mk: Makefile $(wildcard $(RUST_TESTS_DIR)run_pass/*.rs) # # TEST: Rust standard library and the "hello, world" run-pass test # -test: $(RUSTCSRC) output/libcore.hir output/liballoc.hir output/libcollections.hir output/libstd.hir output/rust/test_run-pass_hello $(BIN) +test: $(RUSTCSRC) output/libcore.hir output/liballoc.hir output/libcollections.hir output/libstd.hir output/rust/test_run-pass_hello_out.txt $(BIN) # # TEST: Attempt to compile rust_os (Tifflin) from ../rust_os @@ -413,6 +884,9 @@ $(BIN): $(OBJ) @mkdir -p $(dir $@) @echo [CXX] -o $@ $V$(CXX) -o $@ $(LINKFLAGS) $(OBJ) $(LIBS) + objcopy --only-keep-debug $(BIN) $(BIN).debug + objcopy --add-gnu-debuglink=$(BIN).debug $(BIN) + strip $(BIN) $(OBJDIR)%.o: src/%.cpp @mkdir -p $(dir $@) diff --git a/Notes/BigChanges/Variant.md b/Notes/BigChanges/Variant.md new file mode 100644 index 00000000..55bb3bb0 --- /dev/null +++ b/Notes/BigChanges/Variant.md @@ -0,0 +1,70 @@ + +(the below is from @ubsan) + +``` +// one can define their own variant, but I'm lazy +#include <variant> +#include <type_traits> +#include <iostream> + +template <typename F, typename T, typename Void = void> +struct return_type {}; + +template <typename F, typename T> +struct return_type<T, F, std::void_t< + decltype(std::declval<F>()(std::declval<T>())) +>> { + using type = decltype(std::declval<F>()(std::declval<T>())); +}; + +template <typename F, typename T> +using return_type_t = typename return_type<T, F>::type; + +// not sure what to call this +template <typename F, typename... Ts> +struct common_return_type { + using type = std::common_type_t<return_type_t<F, Ts>...>; +}; + +template <typename F, typename... Ts> +using common_return_type_t = typename common_return_type<F, Ts...>::type; + +template <typename F, typename... Ts> +auto match( + std::variant<Ts...>& variant, F&& functor +) -> common_return_type_t<F, Ts...> { + // you could also use static_assert to make it SFINAE-unfriendly + return std::visit(functor, variant); +} + +template <typename... Fs> +struct overloaded : Fs... { + using Fs::operator()...; + overloaded(Fs&&... fs) : Fs(std::forward<Fs>(fs))... { } +}; + +int main() { + auto var = std::variant<int, std::string>(0); + std::cout << match(var, overloaded( + [](int i) { return i; }, + [](std::string s) { return 0; } + )); +} +``` + + + +ALTERNATIVE +=========== + +Could just update TU to have: +`` +stmt.if_let( +stmt.match_def( +stmt.match( + [](::MIR::LValue::Data_Assign& se) { + }, + [](::MIR::LValue::Data_Drop& se) { + } + ); +``` diff --git a/Notes/BorrowChecker.md b/Notes/BorrowChecker.md new file mode 100644 index 00000000..78b46ff5 --- /dev/null +++ b/Notes/BorrowChecker.md @@ -0,0 +1,7 @@ +- On borrow, calculate lifetime of asignment (using existing lifetime code) + - Ignore reborrows? +- Visit all statements in that lifetime and locate places where the borrow is propagated/stored + - Requires lifetime parameters on functions/&-ptrs to be present +- Assignment of the source value during the lifetime of the borrow is an error +- Dropping of the source value is an error +- Returning the borrow is an error diff --git a/Notes/MIR-Match.md b/Notes/MIR-Match.md index f47d2c88..cc95522c 100644 --- a/Notes/MIR-Match.md +++ b/Notes/MIR-Match.md @@ -20,3 +20,22 @@ For each index in the rule (all rules must be the same length) - Ranges sort after? + +Alternative Generator 2 +======================= + +Maintains match ordering + +1. Calculate branch rulesets (as existing) +1. While rules to process: + 1. Group based on shared values. + 1. Generate dispatch arm for each group + 1. Recurse into group, passing local _ as fallback (or parent _ if none) + +``` + +for + +``` + + diff --git a/Notes/MIR-Validation.txt b/Notes/MIR-Validation.txt new file mode 100644 index 00000000..337c65cd --- /dev/null +++ b/Notes/MIR-Validation.txt @@ -0,0 +1,9 @@ + +Requirements: +- Know full state (and linked states) for each path through the function. + + +Idea: +- Lifetime calculations (existing code) + - Mask out known invalid values at loopback. + diff --git a/Notes/todo.txt b/Notes/todo.txt index 35526889..54956718 100644 --- a/Notes/todo.txt +++ b/Notes/todo.txt @@ -19,5 +19,7 @@ TODO: - Optimise typecheck. -- MIR: Add a Parameter type that is either LValue, Constant - > For any place a LValue is currently used, but a constant is valid +- MIR: Unify Variable/Argument/Temporary LValue types + - This should reduce the amount of code needed for validation, but will be a + BIG change. + @@ -8,36 +8,34 @@ This project is an attempt at creating a simple rust compiler in C++, with the u The short-term goal is to compile pre-borrowchecked rust code into some intermediate form (e.g. LLVM IR, x86-64 assembly, or C code). Thankfully, (from what I have seen), the borrow checker is not needed to compile rust code (just to ensure that it's valid) + +Building Requirements +===================== +- C++14-compatible compiler (tested with gcc 5.4 and gcc 6) +- C11 compatible C compiler (for output, see above) +- `curl` (for downloading the rust source) +- `cmake` (at least 3.4.3, required for building llvm in rustc) + Current Features === -- Attribute and macro expansion -- Resolves all paths to absolute forms -- Converts name-resolved AST into a more compact "HIR" (simplified module and expression tree) -- Hackily evaluates constants - - Constant evaluation is done by using duck-typing, which is then validated by the Type Check pass - - This is how rustc did (or still does?) const eval before MIR -- Type inference and checking -- Closure and operator desugaring -- MIR generation (with partial validation pass) -- HIR/MIR (de)serialisation, allowing for `extern crate` handling -- C-based code generation -- Basic MIR optimisations (including inlining) +- Full compilation chain including HIR and MIR stages (outputting to C) +- Supports just x86-64 linux +- MIR optimisations +- Optionally-enablable exhaustive MIR validation (set the `MRUSTC_FULL_VALIDATE` environment variable) Short-Term Plans === -- Parse and Typecheck all run-pass tests -- Compile a running rustc +- Fix currently-failing tests (mostly in type inferrence) +- Fix all known TODOs in MIR generation (still some possible leaks) Medium-Term Goals === -- Extensive MIR optimisations - Propagate lifetime annotations so that MIR can include a borrow checker Progress === -- Compiles the standard library into loadable MIR -- Compiles the "hello, world" test into compilable and running C code -- Compiles `rustc` - - Generated code is likely not correct +- Compiles static libraries into loadable HIR tree and MIR code +- Generates working executables (most of the test suite) +- Compiles `rustc` that can compile the standard library and "hello, world" diff --git a/run_rustc/Makefile b/run_rustc/Makefile new file mode 100644 index 00000000..9a7a2ed2 --- /dev/null +++ b/run_rustc/Makefile @@ -0,0 +1,31 @@ +RUSTC := ../output/rustc +RUST_SRC := ../rustc-nightly/src/ + +all: output/libstd.rlib output/hello_world + +RUSTFLAGS_output/liblibc.rlib := --cfg stdbuild + +$(RUSTC): + make -C ../ output/rustc -j 3 + +output/%.rlib: $(RUST_SRC)%/lib.rs $(RUSTC) + mkdir -p $(dir $@) + time $(RUSTC) $(RUSTFLAGS_$@) -L output/ -L ../output/libs $< -o $@ +output/%.rlib: $(RUST_SRC)%/src/lib.rs $(RUSTC) + mkdir -p $(dir $@) + time $(RUSTC) $(RUSTFLAGS_$@) -L output/ -L ../output/libs $< -o $@ + +output/hello_world: $(RUST_SRC)test/run-pass/hello.rs output/libstd.rlib $(RUSTC) + time $(RUSTC) $(RUSTFLAGS_$@) -L output/ -L ../output/libs $< -o $@ + +fcn_extcrate = $(patsubst %,output/lib%.rlib,$(1)) + +output/libarena.rlib: output/libstd.rlib +output/liballoc.rlib: output/libcore.rlib +output/libstd_unicode.rlib: output/libcore.rlib +output/libcollections.rlib: output/libcore.rlib output/liballoc.rlib output/libstd_unicode.rlib +output/librand.rlib: output/libcore.rlib +output/liblibc.rlib: output/libcore.rlib +output/libcompiler_builtins.rlib: output/libcore.rlib +output/libstd.rlib: $(call fcn_extcrate, core collections rand libc unwind compiler_builtins alloc_system panic_unwind) +output/libunwind.rlib: $(call fcn_extcrate, core libc) diff --git a/rust_src.patch b/rust_src.patch new file mode 100644 index 00000000..ac479382 --- /dev/null +++ b/rust_src.patch @@ -0,0 +1,26 @@ +--- rustc-nightly/src/libcore/intrinsics.rs ++++ rustc-nightly/src/libcore/intrinsics.rs +@@ -643,5 +643,9 @@ + pub fn drop_in_place<T: ?Sized>(to_drop: *mut T); + ++ /// Obtain the length of a slice pointer ++ #[cfg(rust_compiler="mrustc")] ++ pub fn mrustc_slice_len<T>(pointer: *const [T]) -> usize; ++ + /// Gets a static string slice containing the name of a type. + pub fn type_name<T: ?Sized>() -> &'static str; + +--- rustc-nightly/src/libcore/slice.rs ++++ rustc-nightly/src/libcore/slice.rs +@@ -340,6 +340,8 @@ + #[inline] + fn len(&self) -> usize { +- unsafe { +- mem::transmute::<&[T], Repr<T>>(self).len +- } ++ #[cfg(not(rust_compiler="mrustc"))] ++ let rv = unsafe { mem::transmute::<&[T], Repr<T>>(self).len }; ++ #[cfg(rust_compiler="mrustc")] ++ let rv = unsafe { ::intrinsics::mrustc_slice_len(self) }; ++ rv + } diff --git a/samples/test/for_underscore_drop.rs b/samples/test/for_underscore_drop.rs new file mode 100644 index 00000000..e590fee0 --- /dev/null +++ b/samples/test/for_underscore_drop.rs @@ -0,0 +1,18 @@ + +struct DropFlag<'a>(&'a mut i32); +impl<'a> ::std::ops::Drop for DropFlag<'a> +{ + fn drop(&mut self) { + *self.0 += 1; + } +} + +#[test] +fn values_in_for_loop_dropped() +{ + let mut foo = 0; + for _ in Some(DropFlag(&mut foo)) + { + } + assert_eq!(foo, 1); +} diff --git a/samples/test/scoping_rules.rs b/samples/test/scoping_rules.rs new file mode 100644 index 00000000..9878fd16 --- /dev/null +++ b/samples/test/scoping_rules.rs @@ -0,0 +1,21 @@ + +struct DropFlag<'a>(&'a ::std::cell::Cell<i32>); +impl<'a> ::std::ops::Drop for DropFlag<'a> +{ + fn drop(&mut self) { + self.0.set( self.0.get() + 1 ); + } +} + +// Any temporaries defined in the expression part of a statement (i.e. in the yeilded part of a +// block) are stored in the parent scope. +#[test] +fn temporaries_in_yielded_expr() +{ + let drop_count = ::std::cell::Cell::new(0); + let _foo = ({ DropFlag(&drop_count).0 }, assert_eq!(drop_count.get(), 0) ); + drop(_foo); + assert_eq!(drop_count.get(), 1); +} + + diff --git a/src/ast/crate.hpp b/src/ast/crate.hpp index b74012a3..f9594a83 100644 --- a/src/ast/crate.hpp +++ b/src/ast/crate.hpp @@ -10,6 +10,23 @@ namespace AST { class ExternCrate; +class TestDesc +{ +public: + ::AST::Path path; + ::std::string name; + bool ignore = false; + bool is_benchmark = false; + + enum class ShouldPanic { + No, + Yes, + YesWithMessage, + } panic_type = ShouldPanic::No; + + ::std::string expected_panic_message; +}; + class Crate { public: @@ -22,6 +39,10 @@ public: // Mapping filled by searching for (?visible) macros with is_pub=true ::std::map< ::std::string, const MacroRules*> m_exported_macros; + // List of tests (populated in expand if --test is passed) + bool m_test_harness = false; + ::std::vector<TestDesc> m_tests; + enum class Type { Unknown, RustLib, diff --git a/src/ast/expr.cpp b/src/ast/expr.cpp index 65048183..dd586683 100644 --- a/src/ast/expr.cpp +++ b/src/ast/expr.cpp @@ -85,9 +85,7 @@ NODE(ExprNode_Block, { ::std::vector<ExprNodeP> nodes; for(const auto& n : m_nodes) nodes.push_back( n->clone() ); - if( m_local_mod ) - TODO(get_pos(), "Handle cloning ExprNode_Block with a module"); - return NEWNODE(ExprNode_Block, m_is_unsafe, m_yields_final_value, mv$(nodes), nullptr); + return NEWNODE(ExprNode_Block, m_is_unsafe, m_yields_final_value, mv$(nodes), m_local_mod); }) NODE(ExprNode_Macro, { @@ -266,7 +264,16 @@ NODE(ExprNode_ByteString, { }) NODE(ExprNode_Closure, { - os << "/* todo: closure */"; + if( m_is_move ) + os << "move "; + os << "|"; + for(const auto& a : m_args) + { + os << a.first << ": " << a.second << ","; + } + os << "|"; + os << "->" << m_return; + os << " " << *m_code; },{ ExprNode_Closure::args_t args; for(const auto& a : m_args) { @@ -276,7 +283,16 @@ NODE(ExprNode_Closure, { }); NODE(ExprNode_StructLiteral, { - os << "/* todo: sl */"; + os << m_path << " { "; + for(const auto& v : m_values) + { + os << v.first << ": " << *v.second << ", "; + } + if(m_base_value) + { + os << ".." << *m_base_value; + } + os << "}"; },{ ExprNode_StructLiteral::t_values vals; diff --git a/src/common.hpp b/src/common.hpp index 76940d25..55e0f833 100644 --- a/src/common.hpp +++ b/src/common.hpp @@ -72,26 +72,56 @@ static inline Ordering ord(bool l, bool r) else return OrdLess; } +static inline Ordering ord(char l, char r) +{ + return (l == r ? OrdEqual : (l > r ? OrdGreater : OrdLess)); +} + +static inline Ordering ord(unsigned char l, unsigned char r) +{ + return (l == r ? OrdEqual : (l > r ? OrdGreater : OrdLess)); +} +static inline Ordering ord(unsigned short l, unsigned short r) +{ + return (l == r ? OrdEqual : (l > r ? OrdGreater : OrdLess)); +} static inline Ordering ord(unsigned l, unsigned r) { - if(l == r) - return OrdEqual; - else if( l > r ) - return OrdGreater; - else - return OrdLess; + return (l == r ? OrdEqual : (l > r ? OrdGreater : OrdLess)); } -#if UINTPTR_MAX != UINT_MAX -static inline Ordering ord(::std::uintptr_t l, ::std::uintptr_t r) +static inline Ordering ord(unsigned long l, unsigned long r) { - if(l == r) - return OrdEqual; - else if( l > r ) - return OrdGreater; - else - return OrdLess; + return (l == r ? OrdEqual : (l > r ? OrdGreater : OrdLess)); } -#endif +static inline Ordering ord(unsigned long long l, unsigned long long r) +{ + return (l == r ? OrdEqual : (l > r ? OrdGreater : OrdLess)); +} +static inline Ordering ord(signed char l, signed char r) +{ + return (l == r ? OrdEqual : (l > r ? OrdGreater : OrdLess)); +} +static inline Ordering ord(short l, short r) +{ + return (l == r ? OrdEqual : (l > r ? OrdGreater : OrdLess)); +} +static inline Ordering ord(long l, long r) +{ + return (l == r ? OrdEqual : (l > r ? OrdGreater : OrdLess)); +} +static inline Ordering ord(long long l, long long r) +{ + return (l == r ? OrdEqual : (l > r ? OrdGreater : OrdLess)); +} +static inline Ordering ord(float l, float r) +{ + return (l == r ? OrdEqual : (l > r ? OrdGreater : OrdLess)); +} +static inline Ordering ord(double l, double r) +{ + return (l == r ? OrdEqual : (l > r ? OrdGreater : OrdLess)); +} + static inline Ordering ord(const ::std::string& l, const ::std::string& r) { if(l == r) @@ -302,19 +332,8 @@ struct FmtEscaped { FmtEscaped(const ::std::string& s): s(s.c_str()) {} - friend ::std::ostream& operator<<(::std::ostream& os, const FmtEscaped& x) { - for(auto s = x.s; *s != '\0'; s ++) - { - switch(*s) - { - case '\n': os << "\\n"; break; - case '\\': os << "\\\\"; break; - case '"': os << "\\\""; break; - default: os << *s; break; - } - } - return os; - } + // See main.cpp + friend ::std::ostream& operator<<(::std::ostream& os, const FmtEscaped& x); }; // ------------------------------------------------------------------- @@ -333,4 +352,52 @@ auto end (reversion_wrapper<T> w) { return w.iterable.rend(); } template <typename T> reversion_wrapper<T> reverse (T&& iterable) { return { iterable }; } + +template<typename T> +struct RunIterable { + const ::std::vector<T>& list; + unsigned int ofs; + ::std::pair<size_t,size_t> cur; + RunIterable(const ::std::vector<T>& list): + list(list), ofs(0) + { + advance(); + } + void advance() { + if( ofs < list.size() ) + { + auto start = ofs; + while(ofs < list.size() && list[ofs] == list[start]) + ofs ++; + cur = ::std::make_pair(start, ofs-1); + } + else + { + ofs = list.size()+1; + } + } + RunIterable<T> begin() { return *this; } + RunIterable<T> end() { auto rv = *this; rv.ofs = list.size()+1; return rv; } + bool operator==(const RunIterable<T>& x) { + return x.ofs == ofs; + } + bool operator!=(const RunIterable<T>& x) { + return !(*this == x); + } + void operator++() { + advance(); + } + const ::std::pair<size_t,size_t>& operator*() const { + return this->cur; + } + const ::std::pair<size_t,size_t>* operator->() const { + return &this->cur; + } +}; +template<typename T> +RunIterable<T> runs(const ::std::vector<T>& x) { + return RunIterable<T>(x); +} + + #endif diff --git a/src/expand/derive.cpp b/src/expand/derive.cpp index 58eb1f96..c363b689 100644 --- a/src/expand/derive.cpp +++ b/src/expand/derive.cpp @@ -369,43 +369,56 @@ public: pat_a = AST::Pattern(AST::Pattern::TagValue(), AST::Pattern::Value::make_Named(base_path + v.m_name)); ), (Tuple, - // TODO: Complete this. ::std::vector<AST::Pattern> pats_a; - //::std::vector<AST::ExprNodeP> nodes; + AST::ExprNodeP node; + + node = NEWNODE(NamedValue, AST::Path("f")); + node = NEWNODE(CallMethod, + mv$(node), AST::PathNode("debug_tuple",{}), + vec$( NEWNODE(String, v.m_name) ) + ); for( unsigned int idx = 0; idx < e.m_sub_types.size(); idx ++ ) { auto name_a = FMT("a" << idx); pats_a.push_back( ::AST::Pattern(::AST::Pattern::TagBind(), name_a, ::AST::PatternBinding::Type::REF) ); - //nodes.push_back( this->assert_is_eq(assert_method_path, NEWNODE(NamedValue, AST::Path(name_a))) ); - } - //code = NEWNODE(Block, mv$(nodes)); - code = NEWNODE(CallMethod, - NEWNODE(NamedValue, AST::Path("f")), - AST::PathNode("write_str",{}), - vec$( NEWNODE(String, v.m_name + "(...)") ) - ); + node = NEWNODE(CallMethod, + mv$(node), AST::PathNode("field",{}), + vec$( + NEWNODE(NamedValue, AST::Path(name_a)) + ) + ); + } + code = NEWNODE(CallMethod, mv$(node), AST::PathNode("finish",{}), {}); pat_a = AST::Pattern(AST::Pattern::TagNamedTuple(), base_path + v.m_name, mv$(pats_a)); ), (Struct, ::std::vector< ::std::pair<std::string, AST::Pattern> > pats_a; - //::std::vector<AST::ExprNodeP> nodes; + AST::ExprNodeP node; + + node = NEWNODE(NamedValue, AST::Path("f")); + node = NEWNODE(CallMethod, + mv$(node), AST::PathNode("debug_struct",{}), + vec$( NEWNODE(String, v.m_name) ) + ); for( const auto& fld : e.m_fields ) { auto name_a = FMT("a" << fld.m_name); pats_a.push_back( ::std::make_pair(fld.m_name, ::AST::Pattern(::AST::Pattern::TagBind(), name_a, ::AST::PatternBinding::Type::REF)) ); - //nodes.push_back( this->assert_is_eq(assert_method_path, NEWNODE(NamedValue, AST::Path(name_a))) ); + + node = NEWNODE(CallMethod, + mv$(node), AST::PathNode("field",{}), + vec$( + NEWNODE(String, fld.m_name), + NEWNODE(NamedValue, AST::Path(name_a)) + ) + ); } - //code = NEWNODE(Block, mv$(nodes) ); - code = NEWNODE(CallMethod, - NEWNODE(NamedValue, AST::Path("f")), - AST::PathNode("write_str",{}), - vec$( NEWNODE(String, v.m_name + "{...}") ) - ); + code = NEWNODE(CallMethod, mv$(node), AST::PathNode("finish",{}), {}); pat_a = AST::Pattern(AST::Pattern::TagStruct(), base_path + v.m_name, mv$(pats_a), true); ) ) diff --git a/src/expand/file_line.cpp b/src/expand/file_line.cpp index a485fe61..8dfb7e6d 100644 --- a/src/expand/file_line.cpp +++ b/src/expand/file_line.cpp @@ -27,6 +27,15 @@ class CExpanderLine: } }; +class CExpanderColumn: + public ExpandProcMacro +{ + ::std::unique_ptr<TokenStream> expand(const Span& sp, const AST::Crate& crate, const ::std::string& ident, const TokenTree& tt, AST::Module& mod) override + { + return box$( TTStreamO(TokenTree(Token((uint64_t)sp.start_ofs, CORETYPE_U32))) ); + } +}; + class CExpanderModulePath: public ExpandProcMacro { @@ -44,5 +53,6 @@ class CExpanderModulePath: STATIC_MACRO("file", CExpanderFile); STATIC_MACRO("line", CExpanderLine); +STATIC_MACRO("column", CExpanderColumn); STATIC_MACRO("module_path", CExpanderModulePath); diff --git a/src/expand/format_args.cpp b/src/expand/format_args.cpp index deda6968..92afcbb2 100644 --- a/src/expand/format_args.cpp +++ b/src/expand/format_args.cpp @@ -140,6 +140,11 @@ namespace { if( *s != '{' ) { if( *s == '}' ) { + s ++; + if( *s != '}' ) { + // TODO: Error? Warning? + s --; // Step backwards, just in case + } // Doesn't need escaping cur_literal += '}'; } @@ -247,6 +252,7 @@ namespace { if( *s == '0' ) { args.zero_pad = true; + args.align_char = '0'; s ++; } else { @@ -407,6 +413,24 @@ namespace { toks.push_back( Token(TOK_IDENT, ent) ); } } + void push_toks(::std::vector<TokenTree>& toks, Token t1) { + toks.push_back( mv$(t1) ); + } + void push_toks(::std::vector<TokenTree>& toks, Token t1, Token t2) { + toks.push_back( mv$(t1) ); + toks.push_back( mv$(t2) ); + } + //void push_toks(::std::vector<TokenTree>& toks, Token t1, Token t2, Token t3) { + // toks.push_back( mv$(t1) ); + // toks.push_back( mv$(t2) ); + // toks.push_back( mv$(t3) ); + //} + void push_toks(::std::vector<TokenTree>& toks, Token t1, Token t2, Token t3, Token t4) { + toks.push_back( mv$(t1) ); + toks.push_back( mv$(t2) ); + toks.push_back( mv$(t3) ); + toks.push_back( mv$(t4) ); + } } class CFormatArgsExpander: @@ -576,10 +600,12 @@ class CFormatArgsExpander: } else // if(is_simple) { + // 1. Generate a set of arguments+formatters + // > Each combination of argument index and fragment type needs a unique entry in the `args` array + // Use new_v1_formatted // - requires creating more entries in the `args` list to cover multiple formatters for one value - //push_path(toks, crate, {"fmt", "Arguments", "new_v1_formatted"}); - push_path(toks, crate, {"fmt", "Arguments", "new_v1"}); + push_path(toks, crate, {"fmt", "Arguments", "new_v1_formatted"}); // ( toks.push_back( TokenTree(TOK_PAREN_OPEN) ); { @@ -587,14 +613,100 @@ class CFormatArgsExpander: toks.push_back( Token(TOK_IDENT, "FRAGMENTS") ); toks.push_back( TokenTree(TOK_COMMA) ); - // 1. Generate a set of arguments+formatters - // TODO: Fragments to format // - The format stored by mrustc doesn't quite work with how rustc (and fmt::rt::v1) works toks.push_back( TokenTree(TOK_AMP) ); toks.push_back( TokenTree(TOK_SQUARE_OPEN) ); - //for(const auto& frag : fragments ) { - //} + for(const auto& frag : fragments ) + { + push_path(toks, crate, {"fmt", "ArgumentV1", "new"}); + toks.push_back( Token(TOK_PAREN_OPEN) ); + toks.push_back( Token(TOK_IDENT, FMT("a" << frag.arg_index)) ); + + toks.push_back( TokenTree(TOK_COMMA) ); + + push_path(toks, crate, {"fmt", frag.trait_name, "fmt"}); + toks.push_back( TokenTree(TOK_PAREN_CLOSE) ); + toks.push_back( TokenTree(TOK_COMMA) ); + } + toks.push_back( TokenTree(TOK_SQUARE_CLOSE) ); + toks.push_back( TokenTree(TOK_COMMA) ); + + toks.push_back( TokenTree(TOK_AMP) ); + toks.push_back( TokenTree(TOK_SQUARE_OPEN) ); + for(const auto& frag : fragments) + { + push_path(toks, crate, {"fmt", "rt", "v1", "Argument"}); + toks.push_back( TokenTree(TOK_BRACE_OPEN) ); + + push_toks(toks, Token(TOK_IDENT, "position"), TOK_COLON ); + push_path(toks, crate, {"fmt", "rt", "v1", "Position", "Next"}); + push_toks(toks, TOK_COMMA); + + push_toks(toks, Token(TOK_IDENT, "format"), TOK_COLON ); + push_path(toks, crate, {"fmt", "rt", "v1", "FormatSpec"}); + toks.push_back( TokenTree(TOK_BRACE_OPEN) ); + { + push_toks(toks, Token(TOK_IDENT, "fill"), TOK_COLON, Token(uint64_t(frag.args.align_char), CORETYPE_CHAR), TOK_COMMA ); + + push_toks(toks, Token(TOK_IDENT, "align"), TOK_COLON); + const char* align_var_name = nullptr; + switch( frag.args.align ) + { + case FmtArgs::Align::Unspec: align_var_name = "Unknown"; break; + case FmtArgs::Align::Left: align_var_name = "Left"; break; + case FmtArgs::Align::Center: align_var_name = "Center"; break; + case FmtArgs::Align::Right: align_var_name = "Right"; break; + } + push_path(toks, crate, {"fmt", "rt", "v1", "Alignment", align_var_name}); + push_toks(toks, TOK_COMMA); + + push_toks(toks, Token(TOK_IDENT, "flags"), TOK_COLON); + uint64_t flags = 0; + if(frag.args.alternate) + flags |= 1 << 2; + push_toks(toks, Token(uint64_t(flags), CORETYPE_U32)); + push_toks(toks, TOK_COMMA); + + push_toks(toks, Token(TOK_IDENT, "precision"), TOK_COLON ); + if( frag.args.prec_is_arg || frag.args.prec != 0 ) { + push_path(toks, crate, {"fmt", "rt", "v1", "Count", "Is"}); + push_toks(toks, TOK_PAREN_OPEN); + if( frag.args.prec_is_arg ) { + push_toks(toks, TOK_STAR, Token(TOK_IDENT, FMT("a" << frag.args.prec)) ); + } + else { + push_toks(toks, Token(uint64_t(frag.args.prec), CORETYPE_UINT) ); + } + toks.push_back( TokenTree(TOK_PAREN_CLOSE) ); + } + else { + push_path(toks, crate, {"fmt", "rt", "v1", "Count", "Implied"}); + } + toks.push_back( TokenTree(TOK_COMMA) ); + + push_toks(toks, Token(TOK_IDENT, "width"), TOK_COLON ); + if( frag.args.width_is_arg || frag.args.width != 0 ) { + push_path(toks, crate, {"fmt", "rt", "v1", "Count", "Is"}); + push_toks(toks, TOK_PAREN_OPEN); + if( frag.args.width_is_arg ) { + push_toks(toks, TOK_STAR, Token(TOK_IDENT, FMT("a" << frag.args.width)) ); + } + else { + push_toks(toks, Token(uint64_t(frag.args.width), CORETYPE_UINT) ); + } + toks.push_back( TokenTree(TOK_PAREN_CLOSE) ); + } + else { + push_path(toks, crate, {"fmt", "rt", "v1", "Count", "Implied"}); + } + toks.push_back( TokenTree(TOK_COMMA) ); + } + toks.push_back( TokenTree(TOK_BRACE_CLOSE) ); + + toks.push_back( TokenTree(TOK_BRACE_CLOSE) ); + toks.push_back( TokenTree(TOK_COMMA) ); + } toks.push_back( TokenTree(TOK_SQUARE_CLOSE) ); } // ) diff --git a/src/expand/lang_item.cpp b/src/expand/lang_item.cpp index 0a5c87f5..988dc0e0 100644 --- a/src/expand/lang_item.cpp +++ b/src/expand/lang_item.cpp @@ -187,7 +187,10 @@ public: AttrStage stage() const override { return AttrStage::Post; } void handle(const Span& sp, const AST::MetaItem& attr, AST::Crate& crate, const AST::Path& path, AST::Module& mod, AST::Item& i) const override { - TU_IFLET(::AST::Item, i, Function, e, + if( i.is_None() ) { + // Ignore. + } + else TU_IFLET(::AST::Item, i, Function, e, auto rv = crate.m_lang_items.insert(::std::make_pair( ::std::string("mrustc-main"), ::AST::Path(path) )); if( !rv.second ) { diff --git a/src/expand/mod.cpp b/src/expand/mod.cpp index f152b1ab..788b2109 100644 --- a/src/expand/mod.cpp +++ b/src/expand/mod.cpp @@ -67,7 +67,7 @@ void Expand_Attrs(/*const */::AST::MetaItems& attrs, AttrStage stage, ::std::fu } void Expand_Attrs(::AST::MetaItems& attrs, AttrStage stage, ::AST::Crate& crate, const ::AST::Path& path, ::AST::Module& mod, ::AST::Item& item) { - Expand_Attrs(attrs, stage, [&](const auto& sp, const auto& d, const auto& a){ d.handle(sp, a, crate, path, mod, item); }); + Expand_Attrs(attrs, stage, [&](const auto& sp, const auto& d, const auto& a){ if(!item.is_None()) d.handle(sp, a, crate, path, mod, item); }); } void Expand_Attrs(::AST::MetaItems& attrs, AttrStage stage, ::AST::Crate& crate, ::AST::Module& mod, ::AST::ImplDef& impl) { @@ -109,7 +109,8 @@ void Expand_Attrs(::AST::MetaItems& attrs, AttrStage stage, ::AST::Crate& crate return e; } } - // TODO: Shouldn't this use the _last_ located macro? Allowing later (local) defininitions to override it? + // Find the last macro of this name (allows later #[macro_use] definitions to override) + const MacroRules* last_mac = nullptr; for( const auto& mri : mac_mod.macro_imports_res() ) { //DEBUG("- " << mri.name); @@ -118,10 +119,14 @@ void Expand_Attrs(::AST::MetaItems& attrs, AttrStage stage, ::AST::Crate& crate if( input_ident != "" ) ERROR(mi_span, E0000, "macro_rules! macros can't take an ident"); - auto e = Macro_Invoke(name.c_str(), *mri.data, mv$(input_tt), mod); - return e; + last_mac = mri.data; } } + if( last_mac ) + { + auto e = Macro_Invoke(name.c_str(), *last_mac, mv$(input_tt), mod); + return e; + } } // Error - Unknown macro name @@ -307,11 +312,9 @@ struct CExpandExpr: assert( ! this->replacement ); } void visit_vector(::std::vector< ::std::unique_ptr<AST::ExprNode> >& cnodes) { - for( auto& child : cnodes ) { - this->visit(child); - } - // Delete null children for( auto it = cnodes.begin(); it != cnodes.end(); ) { + assert( it->get() ); + this->visit(*it); if( it->get() == nullptr ) { it = cnodes.erase( it ); } @@ -321,53 +324,94 @@ struct CExpandExpr: } } - void visit(::AST::ExprNode_Macro& node) override + ::AST::ExprNodeP visit_macro(::AST::ExprNode_Macro& node, ::std::vector< ::AST::ExprNodeP>* nodes_out) { - TRACE_FUNCTION_F("ExprNode_Macro - name = " << node.m_name); + TRACE_FUNCTION_F(node.m_name << "!"); if( node.m_name == "" ) { - return ; + return ::AST::ExprNodeP(); } + ::AST::ExprNodeP rv; auto& mod = this->cur_mod(); - auto ttl = Expand_Macro( - crate, modstack, mod, - Span(node.get_pos()), - node.m_name, node.m_ident, node.m_tokens - ); - if( ttl.get() != nullptr ) + auto ttl = Expand_Macro( crate, modstack, mod, Span(node.get_pos()), node.m_name, node.m_ident, node.m_tokens ); + if( !ttl.get() ) + { + // No expansion + } + else { - if( ttl->lookahead(0) != TOK_EOF ) + while( ttl->lookahead(0) != TOK_EOF ) { SET_MODULE( (*ttl), mod ); + // Reparse as expression / item bool add_silence_if_end = false; ::std::shared_ptr< AST::Module> tmp_local_mod; auto& local_mod_ptr = (this->current_block ? this->current_block->m_local_mod : tmp_local_mod); - DEBUG("-- Parsing as expression line (legacy)"); + DEBUG("-- Parsing as expression line"); auto newexpr = Parse_ExprBlockLine_WithItems(*ttl, local_mod_ptr, add_silence_if_end); + + if( tmp_local_mod ) + TODO(node.get_pos(), "Handle edge case where a macro expansion outside of a _Block creates an item"); + if( newexpr ) { - // TODO: use add_silence_if_end - Applies if this node is the last node in the block. - - // Then call visit on it again - DEBUG("--- Visiting new node"); - this->visit(newexpr); - // And schedule it to replace the previous - replacement = mv$(newexpr); + if( nodes_out ) { + nodes_out->push_back( mv$(newexpr) ); + } + else { + assert( !rv ); + rv = mv$(newexpr); + } } else { - // TODO: Delete this node somehow? Or just leave it for later. + // Expansion line just added a new item } - ASSERT_BUG(node.get_pos(), !tmp_local_mod, "TODO: Handle edge case where a macro expansion outside of a _Block creates an item"); + + if( ttl->lookahead(0) != TOK_EOF ) + { + if( !nodes_out ) { + ERROR(node.get_pos(), E0000, "Unused tokens at the end of macro expansion - " << ttl->getToken()); + } + } + } + } + + node.m_name = ""; + return mv$(rv); + } + + void visit(::AST::ExprNode_Macro& node) override + { + TRACE_FUNCTION_F("ExprNode_Macro - name = " << node.m_name); + if( node.m_name == "" ) { + return ; + } + + replacement = this->visit_macro(node, nullptr); + + if( this->replacement ) + { + DEBUG("--- Visiting new node"); + auto n = mv$(this->replacement); + this->visit(n); + if( n ) + { + assert( !this->replacement ); + this->replacement = mv$(n); } - DEBUG("ExprNode_Macro - replacement = " << replacement.get()); - node.m_name = ""; } } void visit(::AST::ExprNode_Block& node) override { unsigned int mod_item_count = 0; + + auto prev_modstack = this->modstack; + if( node.m_local_mod ) { + this->modstack = LList<const ::AST::Module*>(&prev_modstack, node.m_local_mod.get()); + } + // TODO: macro_rules! invocations within the expression list influence this. // > Solution: Defer creation of the local module until during expand. if( node.m_local_mod ) { @@ -377,13 +421,48 @@ struct CExpandExpr: auto saved = this->current_block; this->current_block = &node; - this->visit_vector(node.m_nodes); + + for( auto it = node.m_nodes.begin(); it != node.m_nodes.end(); ) + { + assert( it->get() ); + + if( auto* node_mac = dynamic_cast<::AST::ExprNode_Macro*>(it->get()) ) + { + Expand_Attrs((*it)->attrs(), AttrStage::Pre, [&](const auto& sp, const auto& d, const auto& a){ d.handle(sp, a, this->crate, *it); }); + if( !it->get() ) { + it = node.m_nodes.erase( it ); + continue ; + } + + assert(it->get() == node_mac); + + ::std::vector< ::AST::ExprNodeP> new_nodes; + this->visit_macro(*node_mac, &new_nodes); + + it = node.m_nodes.erase(it); + it = node.m_nodes.insert(it, ::std::make_move_iterator(new_nodes.begin()), ::std::make_move_iterator(new_nodes.end())); + // NOTE: Doesn't advance the iterator above, we want to re-visit the new node + } + else + { + this->visit(*it); + if( it->get() == nullptr ) { + it = node.m_nodes.erase( it ); + } + else { + ++ it; + } + } + } + this->current_block = saved; // HACK! Run Expand_Mod twice on local modules. if( node.m_local_mod ) { Expand_Mod(crate, modstack, node.m_local_mod->path(), *node.m_local_mod, mod_item_count); } + + this->modstack = mv$(prev_modstack); } void visit(::AST::ExprNode_Asm& node) override { for(auto& v : node.m_output) @@ -611,11 +690,13 @@ struct CExpandExpr: void Expand_Expr(::AST::Crate& crate, LList<const AST::Module*> modstack, ::std::unique_ptr<AST::ExprNode>& node) { + TRACE_FUNCTION_F("unique_ptr"); auto visitor = CExpandExpr(crate, modstack); visitor.visit(node); } void Expand_Expr(::AST::Crate& crate, LList<const AST::Module*> modstack, ::std::shared_ptr<AST::ExprNode>& node) { + TRACE_FUNCTION_F("shared_ptr"); auto visitor = CExpandExpr(crate, modstack); node->visit(visitor); if( visitor.replacement ) { @@ -624,6 +705,7 @@ void Expand_Expr(::AST::Crate& crate, LList<const AST::Module*> modstack, ::std: } void Expand_Expr(::AST::Crate& crate, LList<const AST::Module*> modstack, AST::Expr& node) { + TRACE_FUNCTION_F("AST::Expr"); auto visitor = CExpandExpr(crate, modstack); node.visit_nodes(visitor); if( visitor.replacement ) { @@ -750,7 +832,7 @@ void Expand_Mod(::AST::Crate& crate, LList<const AST::Module*> modstack, ::AST:: { auto& i = mod.items()[idx]; - DEBUG("- " << i.name << " (" << ::AST::Item::tag_to_str(i.data.tag()) << ") :: " << i.data.attrs); + DEBUG("- " << modpath << "::" << i.name << " (" << ::AST::Item::tag_to_str(i.data.tag()) << ") :: " << i.data.attrs); ::AST::Path path = modpath + i.name; auto attrs = mv$(i.data.attrs); diff --git a/src/expand/stringify.cpp b/src/expand/stringify.cpp index f43d896f..cbb5c65c 100644 --- a/src/expand/stringify.cpp +++ b/src/expand/stringify.cpp @@ -20,8 +20,9 @@ class CExpander: auto lex = TTStream(tt); while( GET_TOK(tok, lex) != TOK_EOF ) { + if(!rv.empty()) + rv += " "; rv += tok.to_str(); - rv += " "; } return box$( TTStreamO(TokenTree(Token(TOK_STRING, mv$(rv)))) ); diff --git a/src/expand/test.cpp b/src/expand/test.cpp index fba6556f..12bfbb7d 100644 --- a/src/expand/test.cpp +++ b/src/expand/test.cpp @@ -7,21 +7,95 @@ */ #include <synext_decorator.hpp> #include <ast/ast.hpp> +#include <ast/crate.hpp> class CTestHandler: public ExpandDecorator { - AttrStage stage() const override { return AttrStage::Pre; } + AttrStage stage() const override { return AttrStage::Post; } void handle(const Span& sp, const AST::MetaItem& mi, ::AST::Crate& crate, const AST::Path& path, AST::Module& mod, AST::Item&i) const override { if( ! i.is_Function() ) { ERROR(sp, E0000, "#[test] can only be put on functions - found on " << i.tag_str()); } - // TODO: Proper #[test] support, for now just remove them - i = AST::Item::make_None({}); + if( crate.m_test_harness ) + { + ::AST::TestDesc td; + for(const auto& node : path.nodes()) + { + td.name += "::"; + td.name += node.name(); + } + td.path = ::AST::Path(path); + + crate.m_tests.push_back( mv$(td) ); + } + else + { + i = AST::Item::make_None({}); + } + } +}; +class CTestHandler_SP: + public ExpandDecorator +{ + AttrStage stage() const override { return AttrStage::Pre; } + + void handle(const Span& sp, const AST::MetaItem& mi, ::AST::Crate& crate, const AST::Path& path, AST::Module& mod, AST::Item&i) const override { + if( ! i.is_Function() ) { + ERROR(sp, E0000, "#[should_panic] can only be put on functions - found on " << i.tag_str()); + } + + if( crate.m_test_harness ) + { + for(auto& td : crate.m_tests) + { + if( td.path != path ) + continue ; + + if( mi.has_sub_items() ) + { + td.panic_type = ::AST::TestDesc::ShouldPanic::YesWithMessage; + // TODO: Check that name is correct and that it is a string + td.expected_panic_message = mi.items().at(0).string(); + } + else + { + td.panic_type = ::AST::TestDesc::ShouldPanic::Yes; + } + return ; + } + //ERROR() + } + } +}; +class CTestHandler_Ignore: + public ExpandDecorator +{ + AttrStage stage() const override { return AttrStage::Pre; } + + void handle(const Span& sp, const AST::MetaItem& mi, ::AST::Crate& crate, const AST::Path& path, AST::Module& mod, AST::Item&i) const override { + if( ! i.is_Function() ) { + ERROR(sp, E0000, "#[should_panic] can only be put on functions - found on " << i.tag_str()); + } + + if( crate.m_test_harness ) + { + for(auto& td : crate.m_tests) + { + if( td.path != path ) + continue ; + + td.ignore = true; + return ; + } + //ERROR() + } } }; STATIC_DECORATOR("test", CTestHandler); +STATIC_DECORATOR("should_panic", CTestHandler_SP); +STATIC_DECORATOR("ignore", CTestHandler_Ignore); diff --git a/src/expand/test_harness.cpp b/src/expand/test_harness.cpp new file mode 100644 index 00000000..12d32121 --- /dev/null +++ b/src/expand/test_harness.cpp @@ -0,0 +1,120 @@ +/* + * MRustC - Rust Compiler + * - By John Hodge (Mutabah/thePowersGang) + * + * expand/mod.cpp + * - Expand pass core code + */ +#include <ast/ast.hpp> +#include <ast/expr.hpp> +#include <ast/crate.hpp> +#include <main_bindings.hpp> +#include <hir/hir.hpp> // ABI_RUST + +#define NEWNODE(_ty, ...) ::AST::ExprNodeP(new ::AST::ExprNode##_ty(__VA_ARGS__)) + +void Expand_TestHarness(::AST::Crate& crate) +{ + // Create the following module: + // ``` + // mod `#test` { + // extern crate std; + // extern crate test; + // fn main() { + // self::test::test_main_static(&::`#test`::TESTS); + // } + // static TESTS: [test::TestDescAndFn; _] = [ + // test::TestDescAndFn { desc: test::TestDesc { name: "foo", ignore: false, should_panic: test::ShouldPanic::No }, testfn: ::path::to::foo }, + // ]; + // } + // ``` + + // ---- main function ---- + auto main_fn = ::AST::Function { Span(), {}, ABI_RUST, false, false, false, TypeRef(TypeRef::TagUnit(), Span()), {} }; + { + auto call_node = NEWNODE(_CallPath, + ::AST::Path("test", { ::AST::PathNode("test_main_static") }), + ::make_vec1( + NEWNODE(_UniOp, ::AST::ExprNode_UniOp::REF, + NEWNODE(_NamedValue, ::AST::Path("", { ::AST::PathNode("test#"), ::AST::PathNode("TESTS") })) + ) + ) + ); + main_fn.set_code( mv$(call_node) ); + } + + + // ---- test list ---- + ::std::vector< ::AST::ExprNodeP> test_nodes; + + for(const auto& test : crate.m_tests) + { + // HACK: Don't emit should_panic tests + if( test.panic_type != ::AST::TestDesc::ShouldPanic::No ) + continue ; + + ::AST::ExprNode_StructLiteral::t_values desc_vals; + // `name: "foo",` + desc_vals.push_back( ::std::make_pair("name", NEWNODE(_CallPath, + ::AST::Path("test", { ::AST::PathNode("StaticTestName") }), + ::make_vec1( NEWNODE(_String, test.name) ) + ) )); + // `ignore: false,` + desc_vals.push_back( ::std::make_pair("ignore", NEWNODE(_Bool, test.ignore)) ); + // `should_panic: ShouldPanic::No,` + { + ::AST::ExprNodeP should_panic_val; + switch(test.panic_type) + { + case ::AST::TestDesc::ShouldPanic::No: + should_panic_val = NEWNODE(_NamedValue, ::AST::Path("test", { ::AST::PathNode("ShouldPanic"), ::AST::PathNode("No") })); + break; + case ::AST::TestDesc::ShouldPanic::Yes: + should_panic_val = NEWNODE(_NamedValue, ::AST::Path("test", { ::AST::PathNode("ShouldPanic"), ::AST::PathNode("Yes") })); + break; + case ::AST::TestDesc::ShouldPanic::YesWithMessage: + should_panic_val = NEWNODE(_CallPath, + ::AST::Path("test", { ::AST::PathNode("ShouldPanic"), ::AST::PathNode("YesWithMessage") }), + make_vec1( NEWNODE(_String, test.expected_panic_message) ) + ); + break; + } + desc_vals.push_back( ::std::make_pair("should_panic", mv$(should_panic_val)) ); + } + auto desc_expr = NEWNODE(_StructLiteral, ::AST::Path("test", { ::AST::PathNode("TestDesc")}), nullptr, mv$(desc_vals)); + + ::AST::ExprNode_StructLiteral::t_values descandfn_vals; + descandfn_vals.push_back( ::std::make_pair(::std::string("desc"), mv$(desc_expr)) ); + + auto test_type_var_name = test.is_benchmark ? "StaticBenchFn" : "StaticTestFn"; + descandfn_vals.push_back( ::std::make_pair(::std::string("testfn"), NEWNODE(_CallPath, + ::AST::Path("test", { ::AST::PathNode(test_type_var_name) }), + ::make_vec1( NEWNODE(_NamedValue, AST::Path(test.path)) ) + ) ) ); + + test_nodes.push_back( NEWNODE(_StructLiteral, ::AST::Path("test", { ::AST::PathNode("TestDescAndFn")}), nullptr, mv$(descandfn_vals) ) ); + } + auto* tests_array = new ::AST::ExprNode_Array(mv$(test_nodes)); + + size_t test_count = tests_array->m_values.size(); + auto tests_list = ::AST::Static { ::AST::Static::Class::STATIC, + TypeRef(TypeRef::TagSizedArray(), Span(), + TypeRef(Span(), ::AST::Path("test", { ::AST::PathNode("TestDescAndFn") })), + ::std::shared_ptr<::AST::ExprNode>( new ::AST::ExprNode_Integer(test_count, CORETYPE_UINT) ) + ), + ::AST::Expr( mv$(tests_array) ) + }; + + // ---- module ---- + auto newmod = ::AST::Module { ::AST::Path("", { ::AST::PathNode("test#") }) }; + // - TODO: These need to be loaded too. + // > They don't actually need to exist here, just be loaded (and use absolute paths) + newmod.add_ext_crate(false, "std", "std", {}); + newmod.add_ext_crate(false, "test", "test", {}); + + newmod.add_item(false, "main", mv$(main_fn), {}); + newmod.add_item(false, "TESTS", mv$(tests_list), {}); + + crate.m_root_module.add_item(false, "test#", mv$(newmod), {}); + crate.m_lang_items["mrustc-main"] = ::AST::Path("", { AST::PathNode("test#"), AST::PathNode("main") }); +} diff --git a/src/hir/deserialise.cpp b/src/hir/deserialise.cpp index c08f886d..0ff4d63b 100644 --- a/src/hir/deserialise.cpp +++ b/src/hir/deserialise.cpp @@ -586,6 +586,7 @@ namespace { m.dst_type = static_cast< ::HIR::TraitMarkings::DstType>( m_in.read_tag() ); m.coerce_unsized_index = m_in.read_count( ); m.unsized_field = m_in.read_count( ); + m.unsized_param = m_in.read_count(); // TODO: auto_impls return m; } @@ -1009,6 +1010,11 @@ namespace { sdf.other = static_cast<unsigned int>(m_in.read_count()); return ::MIR::Statement::make_SetDropFlag(sdf); } + case 4: + return ::MIR::Statement::make_ScopeEnd({ + deserialise_vec<unsigned int>(), + deserialise_vec<unsigned int>() + }); default: ::std::cerr << "Bad tag for a MIR Statement" << ::std::endl; throw ""; diff --git a/src/hir/dump.cpp b/src/hir/dump.cpp index f00efcd5..8bcffa4e 100644 --- a/src/hir/dump.cpp +++ b/src/hir/dump.cpp @@ -165,7 +165,13 @@ namespace { (Unit, ), (Value, - m_os << " = ?";// << + m_os << " = "; + if( e.val.is_Invalid() ) { + m_os << "?"; + } + else { + m_os << e.val; + } ), (Tuple, m_os << "("; @@ -296,28 +302,28 @@ namespace { void visit(::HIR::ExprNode_Block& node) override { if( node.m_nodes.size() == 0 ) { - m_os << "{ }"; + m_os << "{"; + if( node.m_value_node ) + { + m_os << " "; + this->visit_node_ptr(node.m_value_node); + } + m_os << " }"; } - //else if( node.m_nodes.size() == 1) { - // m_os << "{ "; - // this->visit_node_ptr(node.m_nodes.front()); - // m_os << " }"; - //} else { m_os << "{\n"; inc_indent(); for(auto& sn : node.m_nodes) { + m_os << "\n"; m_os << indent(); this->visit_node_ptr(sn); - if( &sn != &node.m_nodes.back() ) { - m_os << ";\n"; - } - else if( !node.m_yields_final ) { - m_os << ";\n"; - } - else { - m_os << "\n"; - } + m_os << ";\n"; + } + if( node.m_value_node ) + { + m_os << indent(); + this->visit_node_ptr(node.m_value_node); + m_os << "\n"; } dec_indent(); m_os << indent() << "}"; @@ -418,7 +424,7 @@ namespace { void visit(::HIR::ExprNode_Assign& node) override { this->visit_node_ptr(node.m_slot); - m_os << " = "; + m_os << " " << ::HIR::ExprNode_Assign::opname(node.m_op) << "= "; this->visit_node_ptr(node.m_value); } void visit(::HIR::ExprNode_BinOp& node) override diff --git a/src/hir/expr.cpp b/src/hir/expr.cpp index e19e7df6..53f85c42 100644 --- a/src/hir/expr.cpp +++ b/src/hir/expr.cpp @@ -23,6 +23,8 @@ DEF_VISIT(ExprNode_Block, node, for(auto& subnode : node.m_nodes) { visit_node_ptr(subnode); } + if( node.m_value_node ) + visit_node_ptr(node.m_value_node); ) DEF_VISIT(ExprNode_Asm, node, for(auto& v : node.m_outputs) diff --git a/src/hir/expr.hpp b/src/hir/expr.hpp index 81e89b28..3c37a930 100644 --- a/src/hir/expr.hpp +++ b/src/hir/expr.hpp @@ -68,21 +68,20 @@ struct ExprNode_Block: { bool m_is_unsafe; ::std::vector< ExprNodeP > m_nodes; - bool m_yields_final; + ExprNodeP m_value_node; // can be null ::HIR::SimplePath m_local_mod; t_trait_list m_traits; ExprNode_Block(Span sp): ExprNode(mv$(sp)), - m_is_unsafe(false), - m_yields_final(false) + m_is_unsafe(false) {} - ExprNode_Block(Span sp, bool is_unsafe, ::std::vector<ExprNodeP> nodes): + ExprNode_Block(Span sp, bool is_unsafe, ::std::vector<ExprNodeP> nodes, ExprNodeP value_node): ExprNode( mv$(sp) ), m_is_unsafe(is_unsafe), m_nodes( mv$(nodes) ), - m_yields_final(false) + m_value_node( mv$(value_node) ) {} NODE_METHODS(); diff --git a/src/hir/from_ast_expr.cpp b/src/hir/from_ast_expr.cpp index 25f6eade..cd2b27fa 100644 --- a/src/hir/from_ast_expr.cpp +++ b/src/hir/from_ast_expr.cpp @@ -33,7 +33,11 @@ struct LowerHIR_ExprNode_Visitor: ASSERT_BUG(v.get_pos(), n, "NULL node encountered in block"); rv->m_nodes.push_back( LowerHIR_ExprNode_Inner( *n ) ); } - rv->m_yields_final = v.m_yields_final_value; + if( v.m_yields_final_value && ! rv->m_nodes.empty() ) + { + rv->m_value_node = mv$(rv->m_nodes.back()); + rv->m_nodes.pop_back(); + } if( v.m_local_mod ) { @@ -89,8 +93,8 @@ struct LowerHIR_ExprNode_Visitor: case ::AST::ExprNode_Assign::ADD: return ::HIR::ExprNode_Assign::Op::Add; case ::AST::ExprNode_Assign::SUB: return ::HIR::ExprNode_Assign::Op::Sub; - case ::AST::ExprNode_Assign::DIV: return ::HIR::ExprNode_Assign::Op::Mul; - case ::AST::ExprNode_Assign::MUL: return ::HIR::ExprNode_Assign::Op::Div; + case ::AST::ExprNode_Assign::MUL: return ::HIR::ExprNode_Assign::Op::Mul; + case ::AST::ExprNode_Assign::DIV: return ::HIR::ExprNode_Assign::Op::Div; case ::AST::ExprNode_Assign::MOD: return ::HIR::ExprNode_Assign::Op::Mod; case ::AST::ExprNode_Assign::AND: return ::HIR::ExprNode_Assign::Op::And; @@ -328,7 +332,7 @@ struct LowerHIR_ExprNode_Visitor: m_rv.reset( new ::HIR::ExprNode_Loop( v.span(), v.m_label, - ::HIR::ExprNodeP(new ::HIR::ExprNode_Block( v.span(), false, mv$(code))) + ::HIR::ExprNodeP(new ::HIR::ExprNode_Block( v.span(), false, mv$(code), {} )) ) ); break; } case ::AST::ExprNode_Loop::WHILELET: { diff --git a/src/hir/hir.cpp b/src/hir/hir.cpp index 9d1fcf97..d50c3e49 100644 --- a/src/hir/hir.cpp +++ b/src/hir/hir.cpp @@ -93,6 +93,30 @@ const ::HIR::Enum::Variant* ::HIR::Enum::get_variant(const ::std::string& name) return nullptr; return &it->second; } +bool HIR::Enum::is_value() const +{ + return this->m_repr != ::HIR::Enum::Repr::Rust || ::std::all_of(m_variants.begin(), m_variants.end(), [](const auto& x){return x.second.is_Unit() || x.second.is_Value();}); +} +uint32_t HIR::Enum::get_value(size_t idx) const +{ + assert(idx < m_variants.size()); + + if( const auto* e = m_variants[idx].second.opt_Value() ) + { + return e->val.as_Integer(); + } + + uint32_t val = 0; + for(size_t i = 0; i < idx; i ++) + { + if( const auto* e = m_variants[i].second.opt_Value() ) + { + val = e->val.as_Integer(); + } + val ++; + } + return val; +} namespace { bool matches_genericpath(const ::HIR::GenericParams& params, const ::HIR::GenericPath& left, const ::HIR::GenericPath& right, ::HIR::t_cb_resolve_type ty_res, bool expand_generic); diff --git a/src/hir/hir.hpp b/src/hir/hir.hpp index 2f6b37df..304bdd64 100644 --- a/src/hir/hir.hpp +++ b/src/hir/hir.hpp @@ -178,6 +178,7 @@ struct TraitMarkings TraitObject, // (Trait) } dst_type; unsigned int unsized_field = ~0u; + unsigned int unsized_param = ~0u; /// `true` if there is a Copy impl bool is_copy = false; @@ -219,6 +220,11 @@ public: TraitMarkings m_markings; const Variant* get_variant(const ::std::string& ) const; + + /// Returns true if this enum is a C-like enum (has values only) + bool is_value() const; + /// Returns the value for the given variant + uint32_t get_value(size_t variant) const; }; class Struct { diff --git a/src/hir/serialise.cpp b/src/hir/serialise.cpp index 81319c2e..77e17dba 100644 --- a/src/hir/serialise.cpp +++ b/src/hir/serialise.cpp @@ -495,6 +495,11 @@ namespace { m_out.write_count(e.idx); m_out.write_bool(e.new_val); m_out.write_count(e.other); + ), + (ScopeEnd, + m_out.write_tag(4); + serialise_vec(e.vars); + serialise_vec(e.tmps); ) ) } @@ -859,6 +864,7 @@ namespace { m_out.write_tag( static_cast<unsigned int>(m.dst_type) ); m_out.write_count( m.coerce_unsized_index ); m_out.write_count( m.unsized_field ); + m_out.write_count( m.unsized_param ); // TODO: auto_impls } diff --git a/src/hir/type.cpp b/src/hir/type.cpp index dd655ccc..7dd8dc80 100644 --- a/src/hir/type.cpp +++ b/src/hir/type.cpp @@ -606,7 +606,26 @@ bool ::HIR::TypeRef::match_test_generics(const Span& sp, const ::HIR::TypeRef& x return Compare::Unequal; } TU_MATCH(::HIR::TypeRef::Data, (v.m_data, x.m_data), (te, xe), - (Infer, throw "";), + (Infer, + // Both sides are infer + switch(te.ty_class) + { + case ::HIR::InferClass::None: + case ::HIR::InferClass::Diverge: + return Compare::Fuzzy; + default: + switch(xe.ty_class) + { + case ::HIR::InferClass::None: + case ::HIR::InferClass::Diverge: + return Compare::Fuzzy; + default: + if( te.ty_class != xe.ty_class ) + return Compare::Unequal; + return Compare::Fuzzy; + } + } + ), (Generic, throw "";), (Primitive, return (te == xe ? Compare::Equal : Compare::Unequal); diff --git a/src/hir_conv/constant_evaluation.cpp b/src/hir_conv/constant_evaluation.cpp index 1ebdbb11..a950a3dd 100644 --- a/src/hir_conv/constant_evaluation.cpp +++ b/src/hir_conv/constant_evaluation.cpp @@ -276,6 +276,10 @@ namespace { { e->visit(*this); } + if( node.m_value_node ) + node.m_value_node->visit(*this); + else + ; } void visit(::HIR::ExprNode_Asm& node) override { badnode(node); diff --git a/src/hir_conv/markings.cpp b/src/hir_conv/markings.cpp index a294f47b..4550bef5 100644 --- a/src/hir_conv/markings.cpp +++ b/src/hir_conv/markings.cpp @@ -43,6 +43,33 @@ public: { str.m_markings.unsized_field = (str.m_data.is_Tuple() ? str.m_data.as_Tuple().size()-1 : str.m_data.as_Named().size()-1); } + + // Rules: + // - A type parameter must be ?Sized + // - That type parameter must only be used as part of the last field, and only once + // - If the final field isn't the parameter, it must also impl Unsize + + // HACK: Just determine what ?Sized parameter is controlling the sized-ness + if( str.m_markings.dst_type == ::HIR::TraitMarkings::DstType::Possible ) + { + auto& last_field_ty = (str.m_data.is_Tuple() ? str.m_data.as_Tuple().back().ent : str.m_data.as_Named().back().second.ent); + auto ty = ::HIR::TypeRef("", 0); + for(size_t i = 0; i < str.m_params.m_types.size(); i++) + { + const auto& param = str.m_params.m_types[i]; + auto ty = ::HIR::TypeRef(param.m_name, i); + if( !param.m_is_sized ) + { + if( visit_ty_with(last_field_ty, [&](const auto& t){ return t == ty; }) ) + { + assert(str.m_markings.unsized_param == ~0u); + str.m_markings.unsized_param = i; + } + } + } + ASSERT_BUG(Span(), str.m_markings.unsized_param != ~0u, "No unsized param for type " << ip); + str.m_markings.can_unsize = true; + } } void visit_trait(::HIR::ItemPath ip, ::HIR::Trait& tr) override @@ -212,6 +239,7 @@ public: ::HIR::TraitMarkings::DstType get_field_dst_type(const ::HIR::TypeRef& ty, const ::HIR::GenericParams& inner_def, const ::HIR::GenericParams& params_def, const ::HIR::PathParams* params) { + TRACE_FUNCTION_F("ty=" << ty); // If the type is generic, and the pointed-to parameters is ?Sized, record as needing unsize if( const auto* te = ty.m_data.opt_Generic() ) { @@ -242,7 +270,7 @@ public: // If the type is a struct, check it (recursively) if( ! te->path.m_data.is_Generic() ) { // Associated type, TODO: Check this better. - return ::HIR::TraitMarkings::DstType::Possible; + return ::HIR::TraitMarkings::DstType::None; } else if( te->binding.is_Struct() ) { const auto& params_tpl = te->path.m_data.as_Generic().m_params; diff --git a/src/hir_expand/annotate_value_usage.cpp b/src/hir_expand/annotate_value_usage.cpp index da4c897f..dbcd6b13 100644 --- a/src/hir_expand/annotate_value_usage.cpp +++ b/src/hir_expand/annotate_value_usage.cpp @@ -84,6 +84,8 @@ namespace { for( auto& subnode : node.m_nodes ) { this->visit_node_ptr(subnode); } + if( node.m_value_node ) + this->visit_node_ptr(node.m_value_node); } void visit(::HIR::ExprNode_Asm& node) override @@ -463,6 +465,7 @@ namespace { return get_usage_for_pattern(sp, *pe.sub, *ty.m_data.as_Borrow().inner); ), (Tuple, + ASSERT_BUG(sp, ty.m_data.is_Tuple(), "Tuple pattern with non-tuple type - " << ty); const auto& subtys = ty.m_data.as_Tuple(); assert(pe.sub_patterns.size() == subtys.size()); auto rv = ::HIR::ValueUsage::Borrow; @@ -471,6 +474,7 @@ namespace { return rv; ), (SplitTuple, + ASSERT_BUG(sp, ty.m_data.is_Tuple(), "SplitTuple pattern with non-tuple type - " << ty); const auto& subtys = ty.m_data.as_Tuple(); assert(pe.leading.size() + pe.trailing.size() < subtys.size()); auto rv = ::HIR::ValueUsage::Borrow; @@ -486,6 +490,7 @@ namespace { (StructTuple, // TODO: Avoid monomorphising all the time. const auto& str = *pe.binding; + ASSERT_BUG(sp, str.m_data.is_Tuple(), "StructTuple pattern with non-tuple struct - " << str.m_data.tag_str()); const auto& flds = str.m_data.as_Tuple(); assert(pe.sub_patterns.size() == flds.size()); auto monomorph_cb = monomorphise_type_get_cb(sp, nullptr, &pe.path.m_params, nullptr); diff --git a/src/hir_expand/reborrow.cpp b/src/hir_expand/reborrow.cpp index 688dcf97..b6f6aeb4 100644 --- a/src/hir_expand/reborrow.cpp +++ b/src/hir_expand/reborrow.cpp @@ -73,8 +73,8 @@ namespace { // Recurse into blocks - Neater this way else if( auto p = dynamic_cast< ::HIR::ExprNode_Block*>(node_ptr.get()) ) { - auto& last_node = p->m_nodes.back(); - last_node = do_reborrow(mv$(last_node)); + ASSERT_BUG( node_ptr->span(), p->m_value_node, "reborrow into block that doesn't yield" ); + p->m_value_node = do_reborrow(mv$(p->m_value_node)); } else { diff --git a/src/hir_expand/ufcs_everything.cpp b/src/hir_expand/ufcs_everything.cpp index 60d503aa..23b9c288 100644 --- a/src/hir_expand/ufcs_everything.cpp +++ b/src/hir_expand/ufcs_everything.cpp @@ -410,7 +410,7 @@ namespace { break; case ::HIR::ExprNode_BinOp::Op::Shr: langitem = method = "shr"; if(0) - case ::HIR::ExprNode_BinOp::Op::Shl: langitem = method = "shr"; + case ::HIR::ExprNode_BinOp::Op::Shl: langitem = method = "shl"; if( is_op_valid_shift(ty_l, ty_r) ) { return ; } @@ -611,6 +611,7 @@ namespace { m_replacement = NEWNODE( mv$(node.m_res_type), Deref, sp, mv$(m_replacement) ); } +#if 0 void visit(::HIR::ExprNode_Deref& node) override { const auto& sp = node.span(); @@ -689,6 +690,7 @@ namespace { // - Dereference the result (which is an &-ptr) m_replacement = NEWNODE( mv$(node.m_res_type), Deref, sp, mv$(m_replacement) ); } +#endif diff --git a/src/hir_typeck/expr_check.cpp b/src/hir_typeck/expr_check.cpp index 6cbb61f4..1b4317a6 100644 --- a/src/hir_typeck/expr_check.cpp +++ b/src/hir_typeck/expr_check.cpp @@ -59,9 +59,10 @@ namespace { { n->visit(*this); } - if( node.m_nodes.size() > 0 && node.m_yields_final ) + if( node.m_value_node ) { - check_types_equal(node.span(), node.m_res_type, node.m_nodes.back()->m_res_type); + node.m_value_node->visit(*this); + check_types_equal(node.span(), node.m_res_type, node.m_value_node->m_res_type); } } void visit(::HIR::ExprNode_Asm& node) override diff --git a/src/hir_typeck/expr_cs.cpp b/src/hir_typeck/expr_cs.cpp index 66baba4d..2388d078 100644 --- a/src/hir_typeck/expr_cs.cpp +++ b/src/hir_typeck/expr_cs.cpp @@ -543,13 +543,12 @@ namespace { return ty.m_data.is_Diverge();// || (ty.m_data.is_Infer() && ty.m_data.as_Infer().ty_class == ::HIR::InferClass::Diverge); }; + bool diverges = false; + this->push_traits( node.m_traits ); if( node.m_nodes.size() > 0 ) { - bool diverges = false; - this->push_traits( node.m_traits ); - this->push_inner_coerce(false); - for( unsigned int i = 0; i < node.m_nodes.size()-1; i ++ ) + for( unsigned int i = 0; i < node.m_nodes.size(); i ++ ) { auto& snp = node.m_nodes[i]; this->context.add_ivars( snp->m_res_type ); @@ -561,62 +560,56 @@ namespace { } } this->pop_inner_coerce(); + } - if( node.m_yields_final ) - { - auto& snp = node.m_nodes.back(); - DEBUG("Block yields final value"); - this->context.add_ivars( snp->m_res_type ); - this->context.equate_types(snp->span(), node.m_res_type, snp->m_res_type); - snp->visit(*this); - } - else + if( node.m_value_node ) + { + auto& snp = node.m_value_node; + DEBUG("Block yields final value"); + this->context.add_ivars( snp->m_res_type ); + this->context.equate_types(snp->span(), node.m_res_type, snp->m_res_type); + snp->visit(*this); + } + else if( node.m_nodes.size() > 0 ) + { + // NOTE: If the final statement in the block diverges, mark this as diverging + const auto& snp = node.m_nodes.back(); + bool defer = false; + if( !diverges ) { - auto& snp = node.m_nodes.back(); - this->context.add_ivars( snp->m_res_type ); - // - Not yielded - so don't equate the return - snp->visit(*this); - - // NOTE: If the final statement in the block diverges, mark this as diverging - bool defer = false; - if( !diverges ) - { - TU_IFLET(::HIR::TypeRef::Data, this->context.get_type(snp->m_res_type).m_data, Infer, e, - switch(e.ty_class) - { - case ::HIR::InferClass::Integer: - case ::HIR::InferClass::Float: - diverges = false; - break; - default: - defer = true; - break; - } - ) - else if( is_diverge(snp->m_res_type) ) { - diverges = true; - } - else { + TU_IFLET(::HIR::TypeRef::Data, this->context.get_type(snp->m_res_type).m_data, Infer, e, + switch(e.ty_class) + { + case ::HIR::InferClass::Integer: + case ::HIR::InferClass::Float: diverges = false; + break; + default: + defer = true; + break; } - } - - // If a statement in this block diverges - if( defer ) { - DEBUG("Block final node returns _, derfer diverge check"); - this->context.add_revisit(node); - } - else if( diverges ) { - DEBUG("Block diverges, yield !"); - this->context.equate_types(node.span(), node.m_res_type, ::HIR::TypeRef::new_diverge()); + ) + else if( is_diverge(snp->m_res_type) ) { + diverges = true; } else { - DEBUG("Block doesn't diverge but doesn't yield a value, yield ()"); - this->context.equate_types(node.span(), node.m_res_type, ::HIR::TypeRef::new_unit()); + diverges = false; } } - this->pop_traits( node.m_traits ); + // If a statement in this block diverges + if( defer ) { + DEBUG("Block final node returns _, derfer diverge check"); + this->context.add_revisit(node); + } + else if( diverges ) { + DEBUG("Block diverges, yield !"); + this->context.equate_types(node.span(), node.m_res_type, ::HIR::TypeRef::new_diverge()); + } + else { + DEBUG("Block doesn't diverge but doesn't yield a value, yield ()"); + this->context.equate_types(node.span(), node.m_res_type, ::HIR::TypeRef::new_unit()); + } } else { @@ -624,6 +617,7 @@ namespace { DEBUG("Block is empty, yield ()"); this->context.equate_types(node.span(), node.m_res_type, ::HIR::TypeRef::new_unit()); } + this->pop_traits( node.m_traits ); } void visit(::HIR::ExprNode_Asm& node) override { @@ -1199,7 +1193,7 @@ namespace { { const auto& name = val.first; auto it = ::std::find_if(fields.begin(), fields.end(), [&](const auto& v)->bool{ return v.first == name; }); - assert(it != fields.end()); + ASSERT_BUG(node.span(), it != fields.end(), "Field '" << name << "' not found in struct " << node.m_path); const auto& des_ty_r = it->second.ent; auto& des_ty_cache = node.m_value_types[it - fields.begin()]; const auto* des_ty = &des_ty_r; @@ -1899,6 +1893,7 @@ namespace { return ty.m_data.is_Diverge();// || (ty.m_data.is_Infer() && ty.m_data.as_Infer().ty_class == ::HIR::InferClass::Diverge); }; + assert( !node.m_nodes.empty() ); const auto& last_ty = this->context.get_type( node.m_nodes.back()->m_res_type ); DEBUG("_Block: last_ty = " << last_ty); @@ -2915,11 +2910,11 @@ namespace { void visit(::HIR::ExprNode_Literal& node) override { TU_MATCH(::HIR::ExprNode_Literal::Data, (node.m_data), (e), (Integer, - ASSERT_BUG(node.span(), node.m_res_type.m_data.is_Primitive(), "Float Literal didn't return primitive"); + ASSERT_BUG(node.span(), node.m_res_type.m_data.is_Primitive(), "Integer _Literal didn't return primitive - " << node.m_res_type); e.m_type = node.m_res_type.m_data.as_Primitive(); ), (Float, - ASSERT_BUG(node.span(), node.m_res_type.m_data.is_Primitive(), "Float Literal didn't return primitive"); + ASSERT_BUG(node.span(), node.m_res_type.m_data.is_Primitive(), "Float Literal didn't return primitive - " << node.m_res_type); e.m_type = node.m_res_type.m_data.as_Primitive(); ), (Boolean, @@ -4057,11 +4052,12 @@ namespace { while( auto* p = dynamic_cast< ::HIR::ExprNode_Block*>(&**node_ptr_ptr) ) { DEBUG("- Moving into block"); - ASSERT_BUG( p->span(), context.m_ivars.types_equal(p->m_res_type, p->m_nodes.back()->m_res_type), - "Block and result mismatch - " << context.m_ivars.fmt_type(p->m_res_type) << " != " << context.m_ivars.fmt_type(p->m_nodes.back()->m_res_type)); + assert( p->m_value_node ); + ASSERT_BUG( p->span(), context.m_ivars.types_equal(p->m_res_type, p->m_value_node->m_res_type), + "Block and result mismatch - " << context.m_ivars.fmt_type(p->m_res_type) << " != " << context.m_ivars.fmt_type(p->m_value_node->m_res_type)); // - Override the the result type to the desired result p->m_res_type = ::HIR::TypeRef::new_borrow(borrow_type, des_borrow_inner.clone()); - node_ptr_ptr = &p->m_nodes.back(); + node_ptr_ptr = &p->m_value_node; } #endif auto& node_ptr = *node_ptr_ptr; @@ -4187,38 +4183,40 @@ namespace { // Deref coercions // - If right can be dereferenced to left + DEBUG("-- Deref coercions"); { ::HIR::TypeRef tmp_ty; - const ::HIR::TypeRef* out_ty = &ty_src; + const ::HIR::TypeRef* out_ty_p = &ty_src; unsigned int count = 0; ::std::vector< ::HIR::TypeRef> types; - while( (out_ty = context.m_resolve.autoderef(sp, *out_ty, tmp_ty)) ) + while( (out_ty_p = context.m_resolve.autoderef(sp, *out_ty_p, tmp_ty)) ) { + const auto& out_ty = context.m_ivars.get_type(*out_ty_p); count += 1; - if( out_ty->m_data.is_Infer() && out_ty->m_data.as_Infer().ty_class == ::HIR::InferClass::None ) { + if( out_ty.m_data.is_Infer() && out_ty.m_data.as_Infer().ty_class == ::HIR::InferClass::None ) { // Hit a _, so can't keep going break; } - types.push_back( out_ty->clone() ); + types.push_back( out_ty.clone() ); - if( context.m_ivars.types_equal(ty_dst, *out_ty) == false ) { + if( context.m_ivars.types_equal(ty_dst, out_ty) == false ) { // Check equivalence - if( ty_dst.m_data.tag() == out_ty->m_data.tag() ) { - TU_MATCH_DEF( ::HIR::TypeRef::Data, (ty_dst.m_data, out_ty->m_data), (d_e, s_e), + if( ty_dst.m_data.tag() == out_ty.m_data.tag() ) { + TU_MATCH_DEF( ::HIR::TypeRef::Data, (ty_dst.m_data, out_ty.m_data), (d_e, s_e), ( - if( ty_dst .compare_with_placeholders(sp, *out_ty, context.m_ivars.callback_resolve_infer()) == ::HIR::Compare::Unequal ) { + if( ty_dst .compare_with_placeholders(sp, out_ty, context.m_ivars.callback_resolve_infer()) == ::HIR::Compare::Unequal ) { DEBUG("Same tag, but not fuzzy match"); continue ; } - DEBUG("Same tag and fuzzy match - assuming " << ty_dst << " == " << *out_ty); - context.equate_types(sp, ty_dst, *out_ty); + DEBUG("Same tag and fuzzy match - assuming " << ty_dst << " == " << out_ty); + context.equate_types(sp, ty_dst, out_ty); ), (Slice, // Equate! - context.equate_types(sp, ty_dst, *out_ty); + context.equate_types(sp, ty_dst, out_ty); // - Fall through ) ) @@ -4338,15 +4336,13 @@ namespace { // Search for Unsize // - If `right`: ::core::marker::Unsize<`left`> + DEBUG("-- Unsize trait"); { - const auto& lang_Unsize = context.m_crate.get_lang_item_path(sp, "unsize"); - ::HIR::PathParams pp; - pp.m_types.push_back( ty_dst.clone() ); - bool found = context.m_resolve.find_trait_impls(sp, lang_Unsize, pp, ty_src, [&](auto impl, auto cmp) { - // TODO: Allow fuzzy match if only match - return cmp == ::HIR::Compare::Equal; + auto cmp = context.m_resolve.can_unsize(sp, ty_dst, ty_src, [&](auto new_dst) { + // Equate these two types }); - if( found ) { + if(cmp == ::HIR::Compare::Equal) + { DEBUG("- Unsize " << &*node_ptr << " -> " << ty_dst); auto ty_dst_b = ::HIR::TypeRef::new_borrow(bt, ty_dst.clone()); auto ty_dst_b2 = ty_dst_b.clone(); @@ -4355,18 +4351,27 @@ namespace { return true; } - } - - if( ty_dst.m_data.is_Path() && ty_dst.m_data.as_Path().binding.is_Unbound() ) - { - } - else if( ty_src.m_data.is_Path() && ty_src.m_data.as_Path().binding.is_Unbound() ) - { - } - else if( ty_dst.compare_with_placeholders(sp, ty_src, context.m_ivars.callback_resolve_infer()) != ::HIR::Compare::Unequal ) - { - context.equate_types(sp, ty_dst, ty_src); - return true; + if(cmp == ::HIR::Compare::Unequal) + { + // No unsize possible, equate types + // - Only if they're not already fixed as unequal (that gets handled elsewhere) + if( ty_dst.m_data.is_Path() && ty_dst.m_data.as_Path().binding.is_Unbound() ) + { + } + else if( ty_src.m_data.is_Path() && ty_src.m_data.as_Path().binding.is_Unbound() ) + { + } + else if( ty_dst.compare_with_placeholders(sp, ty_src, context.m_ivars.callback_resolve_infer()) != ::HIR::Compare::Unequal ) + { + context.equate_types(sp, ty_dst, ty_src); + return true; + } + } + if(cmp == ::HIR::Compare::Fuzzy) + { + // Not sure yet + return false; + } } // Keep trying @@ -4635,13 +4640,13 @@ namespace { while( auto* p = dynamic_cast< ::HIR::ExprNode_Block*>(&**npp) ) { DEBUG("- Propagate to the last node of a _Block"); - ASSERT_BUG( p->span(), context.m_ivars.types_equal(p->m_res_type, p->m_nodes.back()->m_res_type), - "Block and result mismatch - " << context.m_ivars.fmt_type(p->m_res_type) << " != " << context.m_ivars.fmt_type(p->m_nodes.back()->m_res_type)); + ASSERT_BUG( p->span(), context.m_ivars.types_equal(p->m_res_type, p->m_value_node->m_res_type), + "Block and result mismatch - " << context.m_ivars.fmt_type(p->m_res_type) << " != " << context.m_ivars.fmt_type(p->m_value_node->m_res_type)); ASSERT_BUG( p->span(), context.m_ivars.types_equal(p->m_res_type, ty_src), "Block and result mismatch - " << context.m_ivars.fmt_type(p->m_res_type) << " != " << context.m_ivars.fmt_type(ty_src) ); p->m_res_type = new_type.clone(); - npp = &p->m_nodes.back(); + npp = &p->m_value_node; } ::HIR::ExprNodeP& node_ptr = *npp; @@ -4872,10 +4877,16 @@ namespace { // - If either is an ivar, add the other as a possibility TU_IFLET( ::HIR::TypeRef::Data, src_ty.m_data, Infer, se, // TODO: Update for InferClass::Diverge ? - if( se.ty_class != ::HIR::InferClass::None ) { - context.equate_types(sp, dst_ty, src_ty); - } - else { + switch(se.ty_class) + { + case ::HIR::InferClass::Integer: + case ::HIR::InferClass::Float: + if( dst_ty.m_data.is_Primitive() ) { + context.equate_types(sp, dst_ty, src_ty); + } + break; + case ::HIR::InferClass::None: + case ::HIR::InferClass::Diverge: TU_IFLET(::HIR::TypeRef::Data, dst_ty.m_data, Infer, de, context.possible_equate_type_unsize_to(se.index, dst_ty); context.possible_equate_type_unsize_from(de.index, src_ty); @@ -4898,6 +4909,7 @@ namespace { // No equivalence added } // - Fall through and search for the impl + DEBUG("- Unsize, no ivar equivalence"); } if( v.trait == context.m_crate.get_lang_item_path(sp, "coerce_unsized") ) { diff --git a/src/hir_typeck/helpers.cpp b/src/hir_typeck/helpers.cpp index 1bf2cafd..8f7dd2f2 100644 --- a/src/hir_typeck/helpers.cpp +++ b/src/hir_typeck/helpers.cpp @@ -1020,6 +1020,41 @@ bool TraitResolution::iterate_bounds( ::std::function<bool(const ::HIR::GenericB } return false; } +bool TraitResolution::iterate_aty_bounds(const Span& sp, const ::HIR::Path::Data::Data_UfcsKnown& pe, ::std::function<bool(const ::HIR::TraitPath&)> cb) const +{ + const auto& trait_ref = m_crate.get_trait_by_path(sp, pe.trait.m_path); + ASSERT_BUG(sp, trait_ref.m_types.count( pe.item ) != 0, "Trait " << pe.trait.m_path << " doesn't contain an associated type " << pe.item); + const auto& aty_def = trait_ref.m_types.find(pe.item)->second; + + for(const auto& bound : aty_def.m_trait_bounds) + { + if( cb(bound) ) + return true; + } + // Search `<Self as Trait>::Name` bounds on the trait itself + for(const auto& bound : trait_ref.m_params.m_bounds) + { + if( ! bound.is_TraitBound() ) continue ; + const auto& be = bound.as_TraitBound(); + + if( ! be.type.m_data.is_Path() ) continue ; + if( ! be.type.m_data.as_Path().binding.is_Opaque() ) continue ; + + const auto& be_type_pe = be.type.m_data.as_Path().path.m_data.as_UfcsKnown(); + if( *be_type_pe.type != ::HIR::TypeRef("Self", 0xFFFF) ) + continue ; + if( be_type_pe.trait.m_path != pe.trait.m_path ) + continue ; + if( be_type_pe.item != pe.item ) + continue ; + + if( cb(be.trait) ) + return true; + } + + return false; +} + bool TraitResolution::find_trait_impls(const Span& sp, const ::HIR::SimplePath& trait, const ::HIR::PathParams& params, const ::HIR::TypeRef& ty, @@ -1075,66 +1110,26 @@ bool TraitResolution::find_trait_impls(const Span& sp, } // Magic Unsize impls to trait objects - if( trait == lang_Unsize ) { + if( trait == lang_Unsize ) + { ASSERT_BUG(sp, params.m_types.size() == 1, "Unsize trait requires a single type param"); const auto& dst_ty = this->m_ivars.get_type(params.m_types[0]); - TU_IFLET( ::HIR::TypeRef::Data, dst_ty.m_data, TraitObject, e, - // Magic impl if T: ThisTrait - bool good; - ::HIR::TypeRef::Data::Data_TraitObject tmp_e; - tmp_e.m_trait.m_path = e.m_trait.m_path.m_path; - - ::HIR::Compare total_cmp = ::HIR::Compare::Equal; - if( e.m_trait.m_path.m_path == ::HIR::SimplePath() ) { - ASSERT_BUG(sp, e.m_markers.size() > 0, "TraitObject with no traits - " << dst_ty); - good = true; - } - else { - good = find_trait_impls(sp, e.m_trait.m_path.m_path, e.m_trait.m_path.m_params, ty, - [&](const auto impl, auto cmp){ - if( cmp == ::HIR::Compare::Unequal ) - return false; - total_cmp &= cmp; - tmp_e.m_trait.m_path.m_params = impl.get_trait_params(); - for(const auto& aty : e.m_trait.m_type_bounds) { - auto atyv = impl.get_type(aty.first.c_str()); - if( atyv == ::HIR::TypeRef() ) - { - // Get the trait from which this associated type comes. - // Insert a UfcsKnown path for that - auto p = ::HIR::Path( ty.clone(), e.m_trait.m_path.clone(), aty.first ); - // Run EAT - atyv = this->expand_associated_types( sp, ::HIR::TypeRef::new_path( mv$(p), {} ) ); - } - tmp_e.m_trait.m_type_bounds[aty.first] = mv$(atyv); - } - return true; - }); - } - auto cb = [&](const auto impl, auto cmp){ - if( cmp == ::HIR::Compare::Unequal ) - return false; - total_cmp &= cmp; - tmp_e.m_markers.back().m_params = impl.get_trait_params(); - return true; - }; - for(const auto& marker : e.m_markers) - { - if(!good) break; - tmp_e.m_markers.push_back( marker.m_path ); - good &= find_trait_impls(sp, marker.m_path, marker.m_params, ty, cb); - } - if( good ) { - ::HIR::PathParams real_params { ::HIR::TypeRef( ::HIR::TypeRef::Data(mv$(tmp_e)) ) }; - return callback( ImplRef(type.clone(), mv$(real_params), {}), total_cmp ); - } - else { - return false; - } - ) + if( find_trait_impls_bound(sp, trait, params, type, callback) ) + return true; - // [T;N] -> [T] is handled down with array indexing + bool rv = false; + auto cb = [&](auto new_dst) { + ::HIR::PathParams real_params { mv$(new_dst) }; + rv = callback( ImplRef(type.clone(), mv$(real_params), {}), ::HIR::Compare::Fuzzy ); + }; + auto cmp = this->can_unsize(sp, dst_ty, type, cb); + if( cmp == ::HIR::Compare::Equal ) + { + assert(!rv); + rv = callback( ImplRef(type.clone(), params.clone(), {}), ::HIR::Compare::Equal ); + } + return rv; } // Magical CoerceUnsized impls for various types @@ -1278,23 +1273,6 @@ bool TraitResolution::find_trait_impls(const Span& sp, */ return false; } - - // Unsize impl for arrays - if( trait == lang_Unsize ) - { - ASSERT_BUG(sp, params.m_types.size() == 1, ""); - const auto& dst_ty = m_ivars.get_type( params.m_types[0] ); - - TU_IFLET(::HIR::TypeRef::Data, dst_ty.m_data, Slice, e2, - auto cmp = e.inner->compare_with_placeholders(sp, *e2.inner, m_ivars.callback_resolve_infer()); - if( cmp != ::HIR::Compare::Unequal ) { - ::HIR::PathParams pp; - // - <[`array_inner`]> so it can be matched with the param by the caller - pp.m_types.push_back( ::HIR::TypeRef::new_slice(e.inner->clone()) ); - return callback( ImplRef(type.clone(), mv$(pp), {}), cmp ); - } - ) - } ) @@ -1344,44 +1322,6 @@ bool TraitResolution::find_trait_impls(const Span& sp, return rv; } - // Trait objects can unsize to a subset of their traits. - if( trait == lang_Unsize ) - { - ASSERT_BUG(sp, params.m_types.size() == 1, ""); - const auto& dst_ty = m_ivars.get_type( params.m_types[0] ); - if( ! dst_ty.m_data.is_TraitObject() ) { - // If the destination isn't a trait object, don't even bother - return false; - } - const auto& e2 = dst_ty.m_data.as_TraitObject(); - - auto cmp = ::HIR::Compare::Equal; - - // TODO: Fuzzy compare - if( e2.m_trait != e.m_trait ) { - return false; - } - // The destination must have a strict subset of marker traits. - const auto& src_markers = e.m_markers; - const auto& dst_markers = e2.m_markers; - for(const auto& mt : dst_markers) - { - // TODO: Fuzzy match - bool found = false; - for(const auto& omt : src_markers) { - if( omt == mt ) { - found = true; - break; - } - } - if( !found ) { - // Return early. - return false; - } - } - - return callback( ImplRef(&type, &e.m_trait.m_path.m_params, &e.m_trait.m_type_bounds), cmp ); - } ) TU_IFLET(::HIR::TypeRef::Data, type.m_data, ErasedType, e, @@ -1438,15 +1378,9 @@ bool TraitResolution::find_trait_impls(const Span& sp, ASSERT_BUG(sp, e.path.m_data.is_UfcsKnown(), "Opaque bound type wasn't UfcsKnown - " << type); const auto& pe = e.path.m_data.as_UfcsKnown(); - // If this associated type has a bound of the desired trait, return it. - const auto& trait_ref = m_crate.get_trait_by_path(sp, pe.trait.m_path); - ASSERT_BUG(sp, trait_ref.m_types.count( pe.item ) != 0, "Trait " << pe.trait.m_path << " doesn't contain an associated type " << pe.item); - const auto& aty_def = trait_ref.m_types.find(pe.item)->second; - auto monomorph_cb = monomorphise_type_get_cb(sp, &*pe.type, &pe.trait.m_params, nullptr, nullptr); - for(const auto& bound : aty_def.m_trait_bounds) - { + auto rv = this->iterate_aty_bounds(sp, pe, [&](const auto& bound) { const auto& b_params = bound.m_path.m_params; ::HIR::PathParams params_mono_o; const auto& b_params_mono = (monomorphise_pathparams_needed(b_params) ? params_mono_o = monomorphise_path_params_with(sp, b_params, monomorph_cb, false) : b_params); @@ -1477,58 +1411,14 @@ bool TraitResolution::find_trait_impls(const Span& sp, return cmp != ::HIR::Compare::Unequal && callback( ImplRef(i_ty.clone(), i_params.clone(), {}), cmp ); }); if( ret ) - return true; - } - // TODO: Search `<Self as Trait>::Name` bounds on the trait itself - for(const auto& bound : trait_ref.m_params.m_bounds) - { - if( ! bound.is_TraitBound() ) continue ; - const auto& be = bound.as_TraitBound(); - - if( ! be.type.m_data.is_Path() ) continue ; - if( ! be.type.m_data.as_Path().binding.is_Opaque() ) continue ; - - const auto& be_type_pe = be.type.m_data.as_Path().path.m_data.as_UfcsKnown(); - if( *be_type_pe.type != ::HIR::TypeRef("Self", 0xFFFF) ) - continue ; - if( be_type_pe.trait.m_path != pe.trait.m_path ) - continue ; - if( be_type_pe.item != pe.item ) - continue ; - - // TODO: Merge the below code with the code from the above loop. - const auto& b_params = be.trait.m_path.m_params; - ::HIR::PathParams params_mono_o; - const auto& b_params_mono = (monomorphise_pathparams_needed(b_params) ? params_mono_o = monomorphise_path_params_with(sp, b_params, monomorph_cb, false) : b_params); - - if( be.trait.m_path.m_path == trait ) { - auto cmp = this->compare_pp(sp, b_params_mono, params); - if( cmp != ::HIR::Compare::Unequal ) - { - if( &b_params_mono == ¶ms_mono_o ) - { - if( callback( ImplRef(type.clone(), mv$(params_mono_o), {}), cmp ) ) - return true; - params_mono_o = monomorphise_path_params_with(sp, b_params, monomorph_cb, false); - } - else - { - if( callback( ImplRef(&type, &b_params, &null_assoc), cmp ) ) - return true; - } - } - } - - bool ret = this->find_named_trait_in_trait(sp, trait, params, *be.trait.m_trait_ptr, be.trait.m_path.m_path, b_params_mono, type, - [&](const auto& i_ty, const auto& i_params, const auto& i_assoc) { - auto cmp = this->compare_pp(sp, i_params, params); - DEBUG("cmp=" << cmp << ", impl " << trait << i_params << " for " << i_ty << " -- desired " << trait << params); - return cmp != ::HIR::Compare::Unequal && callback( ImplRef(i_ty.clone(), i_params.clone(), {}), cmp ); - }); - if( ret ) + // NOTE: Callback called in closure's return statement return true; - } + } + return false; + }); + if( rv ) + return true; } ) @@ -2406,7 +2296,7 @@ bool TraitResolution::find_trait_impls_crate(const Span& sp, // NOTE: Expected behavior is for Ivars to return false // TODO: Should they return Compare::Fuzzy instead? if( type.m_data.is_Infer() ) { - return false; + return callback( ImplRef(&type, params_ptr, &null_assoc), ::HIR::Compare::Fuzzy ); } const ::HIR::TraitMarkings* markings = nullptr; @@ -3096,6 +2986,270 @@ bool TraitResolution::trait_contains_type(const Span& sp, const ::HIR::GenericPa ) ) } +// Checks if a type can unsize to another +// - Returns Compare::Equal if the unsize is possible and fully known +// - Returns Compare::Fuzzy if the unsize is possible, but still unknown. +// - Returns Compare::Unequal if the unsize is impossibe (for any reason) +// +// Closure is called `get_new_type` is true, and the unsize is possible +// +// usecases: +// - Checking for an impl as part of impl selection (return True/False/Maybe with required match for Maybe) +// - Checking for an impl as part of typeck (return True/False/Maybe with unsize possibility OR required equality) +::HIR::Compare TraitResolution::can_unsize( + const Span& sp, const ::HIR::TypeRef& dst_ty, const ::HIR::TypeRef& src_ty, + ::std::function<void(::HIR::TypeRef new_dst)>* new_type_callback, + ::std::function<void(const ::HIR::TypeRef& dst, const ::HIR::TypeRef& src)>* infer_callback + ) const +{ + TRACE_FUNCTION_F(dst_ty << " <- " << src_ty); + const auto& lang_Unsize = this->m_crate.get_lang_item_path(sp, "unsize"); + + // 1. Test for type equality + { + auto cmp = dst_ty.compare_with_placeholders(sp, src_ty, m_ivars.callback_resolve_infer()); + if( cmp == ::HIR::Compare::Equal ) + { + return ::HIR::Compare::Unequal; + } + } + + // 2. If either side is an ivar, fuzzy. + if( dst_ty.m_data.is_Infer() || src_ty.m_data.is_Infer() ) + { + // Inform the caller that these two types could unsize to each other + // - This allows the coercions code to move the coercion rule up + if( infer_callback ) + { + (*infer_callback)(dst_ty, src_ty); + } + return ::HIR::Compare::Fuzzy; + } + + { + bool found_bound = this->iterate_bounds([&](const auto& gb){ + if(!gb.is_TraitBound()) + return false; + const auto& be = gb.as_TraitBound(); + if(be.trait.m_path.m_path != lang_Unsize) + return false; + const auto& be_dst = be.trait.m_path.m_params.m_types.at(0); + + auto cmp = src_ty.compare_with_placeholders(sp, be.type, m_ivars.callback_resolve_infer()); + if(cmp == ::HIR::Compare::Unequal) return false; + + cmp &= dst_ty.compare_with_placeholders(sp, be_dst, m_ivars.callback_resolve_infer()); + if(cmp == ::HIR::Compare::Unequal) return false; + + if( cmp != ::HIR::Compare::Equal ) + { + TODO(sp, "Found bound " << dst_ty << "=" << be_dst << " <- " << src_ty << "=" << be.type); + } + return true; + }); + if( found_bound ) + { + return ::HIR::Compare::Equal; + } + } + + // Associated types, check the bounds in the trait. + if( src_ty.m_data.is_Path() && src_ty.m_data.as_Path().path.m_data.is_UfcsKnown() ) + { + ::HIR::Compare rv = ::HIR::Compare::Equal; + const auto& pe = src_ty.m_data.as_Path().path.m_data.as_UfcsKnown(); + auto monomorph_cb = monomorphise_type_get_cb(sp, &*pe.type, &pe.trait.m_params, nullptr, nullptr); + auto found_bound = this->iterate_aty_bounds(sp, pe, [&](const ::HIR::TraitPath& bound) { + if( bound.m_path.m_path != lang_Unsize ) + return false; + const auto& be_dst_tpl = bound.m_path.m_params.m_types.at(0); + ::HIR::TypeRef tmp_ty; + const auto& be_dst = (monomorphise_type_needed(be_dst_tpl) ? tmp_ty = monomorphise_type_with(sp, be_dst_tpl, monomorph_cb) : be_dst_tpl); + + auto cmp = dst_ty.compare_with_placeholders(sp, be_dst, m_ivars.callback_resolve_infer()); + if(cmp == ::HIR::Compare::Unequal) return false; + + if( cmp != ::HIR::Compare::Equal ) + { + DEBUG("> Found bound (fuzzy) " << dst_ty << "=" << be_dst << " <- " << src_ty); + rv = ::HIR::Compare::Fuzzy; + } + return true; + }); + if( found_bound ) + { + return rv; + } + } + + // Struct<..., T, ...>: Unsize<Struct<..., U, ...>> + if( dst_ty.m_data.is_Path() && src_ty.m_data.is_Path() ) + { + bool dst_is_unsizable = dst_ty.m_data.as_Path().binding.is_Struct() && dst_ty.m_data.as_Path().binding.as_Struct()->m_markings.can_unsize; + bool src_is_unsizable = src_ty.m_data.as_Path().binding.is_Struct() && src_ty.m_data.as_Path().binding.as_Struct()->m_markings.can_unsize; + if( dst_is_unsizable || src_is_unsizable ) + { + DEBUG("Struct unsize? " << dst_ty << " <- " << src_ty); + const auto& str = *dst_ty.m_data.as_Path().binding.as_Struct(); + const auto& dst_gp = dst_ty.m_data.as_Path().path.m_data.as_Generic(); + const auto& src_gp = src_ty.m_data.as_Path().path.m_data.as_Generic(); + + if( dst_gp == src_gp ) + { + DEBUG("Can't Unsize, destination and source are identical"); + return ::HIR::Compare::Unequal; + } + else if( dst_gp.m_path == src_gp.m_path ) + { + DEBUG("Checking for Unsize " << dst_gp << " <- " << src_gp); + // Structures are equal, add the requirement that the ?Sized parameter also impl Unsize + const auto& dst_inner = m_ivars.get_type( dst_gp.m_params.m_types.at(str.m_markings.unsized_param) ); + const auto& src_inner = m_ivars.get_type( src_gp.m_params.m_types.at(str.m_markings.unsized_param) ); + + auto cb = [&](auto d){ + assert(new_type_callback); + + // Re-create structure with s/d + auto dst_gp_new = dst_gp.clone(); + dst_gp_new.m_params.m_types.at(str.m_markings.unsized_param) = mv$(d); + (*new_type_callback)( ::HIR::TypeRef::new_path(mv$(dst_gp_new), &str) ); + }; + if( new_type_callback ) + { + ::std::function<void(::HIR::TypeRef)> cb_p = cb; + return this->can_unsize(sp, dst_inner, src_inner, &cb_p, infer_callback); + } + else + { + return this->can_unsize(sp, dst_inner, src_inner, nullptr, infer_callback); + } + } + else + { + DEBUG("Can't Unsize, destination and source are different structs"); + return ::HIR::Compare::Unequal; + } + } + } + + // (Trait) <- Foo + if( const auto* de = dst_ty.m_data.opt_TraitObject() ) + { + // TODO: Check if src_ty is !Sized + // - Only allowed if the source is a trait object with the same data trait and lesser bounds + + DEBUG("TraitObject unsize? " << dst_ty << " <- " << src_ty); + + // (Trait) <- (Trait+Foo) + if( const auto* se = src_ty.m_data.opt_TraitObject() ) + { + auto rv = ::HIR::Compare::Equal; + // 1. Data trait must be the same (TODO: Fuzzy) + if( de->m_trait != se->m_trait ) + { + return ::HIR::Compare::Unequal; + } + + // 2. Destination markers must be a strict subset + for(const auto& mt : de->m_markers) + { + // TODO: Fuzzy match + bool found = false; + for(const auto& omt : se->m_markers) { + if( omt == mt ) { + found = true; + break; + } + } + if( !found ) { + // Return early. + return ::HIR::Compare::Unequal; + } + } + + if( rv == ::HIR::Compare::Fuzzy && new_type_callback ) + { + // TODO: Inner type + } + return ::HIR::Compare::Equal; + } + + bool good; + ::HIR::Compare total_cmp = ::HIR::Compare::Equal; + + ::HIR::TypeRef::Data::Data_TraitObject tmp_e; + tmp_e.m_trait.m_path = de->m_trait.m_path.m_path; + + // Check data trait first. + if( de->m_trait.m_path.m_path == ::HIR::SimplePath() ) { + ASSERT_BUG(sp, de->m_markers.size() > 0, "TraitObject with no traits - " << dst_ty); + good = true; + } + else { + good = find_trait_impls(sp, de->m_trait.m_path.m_path, de->m_trait.m_path.m_params, src_ty, + [&](const auto impl, auto cmp) { + if( cmp == ::HIR::Compare::Unequal ) + return false; + total_cmp &= cmp; + tmp_e.m_trait.m_path.m_params = impl.get_trait_params(); + for(const auto& aty : de->m_trait.m_type_bounds) { + auto atyv = impl.get_type(aty.first.c_str()); + if( atyv == ::HIR::TypeRef() ) + { + // Get the trait from which this associated type comes. + // Insert a UfcsKnown path for that + auto p = ::HIR::Path( src_ty.clone(), de->m_trait.m_path.clone(), aty.first ); + // Run EAT + atyv = this->expand_associated_types( sp, ::HIR::TypeRef::new_path( mv$(p), {} ) ); + } + tmp_e.m_trait.m_type_bounds[aty.first] = mv$(atyv); + } + return true; + }); + } + + // Then markers + auto cb = [&](const auto impl, auto cmp){ + if( cmp == ::HIR::Compare::Unequal ) + return false; + total_cmp &= cmp; + tmp_e.m_markers.back().m_params = impl.get_trait_params(); + return true; + }; + for(const auto& marker : de->m_markers) + { + if(!good) break; + tmp_e.m_markers.push_back( marker.m_path ); + good &= find_trait_impls(sp, marker.m_path, marker.m_params, src_ty, cb); + } + + if( good && total_cmp == ::HIR::Compare::Fuzzy && new_type_callback ) + { + (*new_type_callback)( ::HIR::TypeRef(mv$(tmp_e)) ); + } + return total_cmp; + } + + // [T] <- [T; n] + if( const auto* de = dst_ty.m_data.opt_Slice() ) + { + if( const auto* se = src_ty.m_data.opt_Array() ) + { + DEBUG("Array unsize? " << *de->inner << " <- " << *se->inner); + auto cmp = de->inner->compare_with_placeholders(sp, *se->inner, m_ivars.callback_resolve_infer()); + // TODO: Indicate to caller that for this to be true, these two must be the same. + // - I.E. if true, equate these types + if(cmp == ::HIR::Compare::Fuzzy && new_type_callback) + { + (*new_type_callback)( ::HIR::TypeRef::new_slice( se->inner->clone() ) ); + } + return cmp; + } + } + + DEBUG("Can't unsize, no rules matched"); + return ::HIR::Compare::Unequal; +} const ::HIR::TypeRef* TraitResolution::type_is_owned_box(const Span& sp, const ::HIR::TypeRef& ty) const { TU_IFLET(::HIR::TypeRef::Data, ty.m_data, Path, e, @@ -3118,7 +3272,7 @@ const ::HIR::TypeRef* TraitResolution::autoderef(const Span& sp, const ::HIR::Ty const auto& ty = this->m_ivars.get_type(ty_in); TU_IFLET(::HIR::TypeRef::Data, ty.m_data, Borrow, e, DEBUG("Deref " << ty << " into " << *e.inner); - return &*e.inner; + return &this->m_ivars.get_type(*e.inner); ) // TODO: Just doing `*[1,2,3]` doesn't work, but this is needed to allow `[1,2,3].iter()` to work else TU_IFLET(::HIR::TypeRef::Data, ty.m_data, Array, e, diff --git a/src/hir_typeck/helpers.hpp b/src/hir_typeck/helpers.hpp index a1db0af9..b7e6ca38 100644 --- a/src/hir_typeck/helpers.hpp +++ b/src/hir_typeck/helpers.hpp @@ -201,6 +201,7 @@ public: /// Iterate over in-scope bounds (function then top) bool iterate_bounds( ::std::function<bool(const ::HIR::GenericBound&)> cb) const; + bool iterate_aty_bounds(const Span& sp, const ::HIR::Path::Data::Data_UfcsKnown& pe, ::std::function<bool(const ::HIR::TraitPath&)> cb) const; typedef ::std::function<bool(const ::HIR::TypeRef&, const ::HIR::PathParams&, const ::std::map< ::std::string,::HIR::TypeRef>&)> t_cb_trait_impl; typedef ::std::function<bool(ImplRef, ::HIR::Compare)> t_cb_trait_impl_r; @@ -265,6 +266,13 @@ public: bool trait_contains_type(const Span& sp, const ::HIR::GenericPath& trait_path, const ::HIR::Trait& trait_ptr, const ::std::string& name, ::HIR::GenericPath& out_path) const; ::HIR::Compare type_is_copy(const Span& sp, const ::HIR::TypeRef& ty) const; + // If `new_type_callback` is populated, it will be called with the actual/possible dst_type + // If `infer_callback` is populated, it will be called when either side is an ivar + ::HIR::Compare can_unsize(const Span& sp, const ::HIR::TypeRef& dst_ty, const ::HIR::TypeRef& src_ty, ::std::function<void(::HIR::TypeRef new_dst)> new_type_callback) const { + return can_unsize(sp, dst_ty, src_ty, &new_type_callback); + } + ::HIR::Compare can_unsize(const Span& sp, const ::HIR::TypeRef& dst_ty, const ::HIR::TypeRef& src_ty, ::std::function<void(::HIR::TypeRef new_dst)>* new_type_callback, ::std::function<void(const ::HIR::TypeRef& dst, const ::HIR::TypeRef& src)>* infer_callback=nullptr) const; + const ::HIR::TypeRef* type_is_owned_box(const Span& sp, const ::HIR::TypeRef& ty) const; private: diff --git a/src/hir_typeck/static.cpp b/src/hir_typeck/static.cpp index 3bae5ba8..8baa2557 100644 --- a/src/hir_typeck/static.cpp +++ b/src/hir_typeck/static.cpp @@ -237,12 +237,11 @@ bool StaticTraitResolve::find_impl( auto monomorph_cb = monomorphise_type_get_cb(sp, &*pe.type, &pe.trait.m_params, nullptr, nullptr); - for(const auto& bound : aty_def.m_trait_bounds) - { + auto check_bound = [&](const ::HIR::TraitPath& bound) { const auto& b_params = bound.m_path.m_params; ::HIR::PathParams params_mono_o; const auto& b_params_mono = (monomorphise_pathparams_needed(b_params) ? params_mono_o = monomorphise_path_params_with(sp, b_params, monomorph_cb, false) : b_params); - DEBUG(": " << bound.m_path.m_path << b_params_mono); + DEBUG("[find_impl] : " << bound.m_path.m_path << b_params_mono); if( bound.m_path.m_path == trait_path ) { @@ -279,9 +278,40 @@ bool StaticTraitResolve::find_impl( DEBUG("impl " << trait_path << i_params << " for " << type << " -- desired " << trait_path << *trait_params); return found_cb( ImplRef(type.clone(), i_params.clone(), {}), false ); }); - if( ret ) + return ret; + }; + + for(const auto& bound : aty_def.m_trait_bounds) + { + if( check_bound(bound) ) + return true; + } + + // Check `where` clauses on the trait too + for(const auto& bound : trait_ref.m_params.m_bounds) + { + if( !bound.is_TraitBound() ) continue; + const auto& be = bound.as_TraitBound(); + + DEBUG("be.type = " << be.type); + if( !be.type.m_data.is_Path() ) + continue; + if( !be.type.m_data.as_Path().path.m_data.is_UfcsKnown() ) + continue ; + { + const auto& pe2 = be.type.m_data.as_Path().path.m_data.as_UfcsKnown(); + if( *pe2.type != ::HIR::TypeRef("Self",GENERIC_Self) ) + continue ; + if( pe2.trait.m_path != pe.trait.m_path ) + continue ; + if( pe2.item != pe.item ) + continue ; + } + + if( check_bound(be.trait) ) return true; } + DEBUG("- No bounds matched"); } ) @@ -564,7 +594,7 @@ bool StaticTraitResolve::find_impl__check_crate_raw( { const auto& e = *ep; - DEBUG("[find_impl] Trait bound " << e.type << " : " << e.trait); + DEBUG("Trait bound " << e.type << " : " << e.trait); auto b_ty_mono = monomorphise_type_with(sp, e.type, cb_monomorph); this->expand_associated_types(sp, b_ty_mono); auto b_tp_mono = monomorphise_traitpath_with(sp, e.trait, cb_monomorph, false); @@ -575,14 +605,15 @@ bool StaticTraitResolve::find_impl__check_crate_raw( // TODO: These should be tagged with the source trait and that source trait used for expansion. this->expand_associated_types(sp, assoc_bound.second); } - DEBUG("[find_impl] - b_ty_mono = " << b_ty_mono << ", b_tp_mono = " << b_tp_mono); + DEBUG("- b_ty_mono = " << b_ty_mono << ", b_tp_mono = " << b_tp_mono); // HACK: If the type is '_', assume the bound passes if( b_ty_mono.m_data.is_Infer() ) { continue ; } // TODO: This is extrememly inefficient (looks up the trait impl 1+N times) - if( b_tp_mono.m_type_bounds.size() > 0 ) { + if( b_tp_mono.m_type_bounds.size() > 0 ) + { // for(const auto& assoc_bound : b_tp_mono.m_type_bounds) { const auto& aty_name = assoc_bound.first; @@ -603,7 +634,7 @@ bool StaticTraitResolve::find_impl__check_crate_raw( //auto cmp = have .match_test_generics_fuzz(sp, exp, cb_ident, cb_match); auto cmp = exp .match_test_generics_fuzz(sp, have, cb_ident, cb_match); - ASSERT_BUG(sp, cmp == ::HIR::Compare::Equal, "Assoc ty " << aty_name << " mismatch, " << have << " != des " << exp); + ASSERT_BUG(sp, cmp != ::HIR::Compare::Unequal, "Assoc ty " << aty_name << " mismatch, " << have << " != des " << exp); return true; }); } @@ -613,10 +644,13 @@ bool StaticTraitResolve::find_impl__check_crate_raw( } } } - else + + // TODO: Detect if the associated type bound above is from directly the bounded trait, and skip this if it's the case + //else { bool rv = false; if( b_ty_mono.m_data.is_Generic() && (b_ty_mono.m_data.as_Generic().binding >> 8) == 2 ) { + DEBUG("- Placeholder param " << b_ty_mono << ", magic success"); rv = true; } else { @@ -1401,6 +1435,151 @@ bool StaticTraitResolve::type_is_sized(const Span& sp, const ::HIR::TypeRef& ty) throw ""; } +bool StaticTraitResolve::type_needs_drop_glue(const Span& sp, const ::HIR::TypeRef& ty) const +{ + // If `T: Copy`, then it can't need drop glue + if( type_is_copy(sp, ty) ) + return false; + + TU_MATCH(::HIR::TypeRef::Data, (ty.m_data), (e), + (Generic, + return true; + ), + (Path, + if( e.binding.is_Opaque() ) + return true; + + auto pp = ::HIR::PathParams(); + bool has_direct_drop = this->find_impl(sp, m_lang_Drop, &pp, ty, [&](auto , bool){ return true; }, true); + if( has_direct_drop ) + return true; + + ::HIR::TypeRef tmp_ty; + const auto& pe = e.path.m_data.as_Generic(); + auto monomorph_cb = monomorphise_type_get_cb(sp, nullptr, &pe.m_params, nullptr, nullptr); + auto monomorph = [&](const auto& tpl)->const ::HIR::TypeRef& { + if( monomorphise_type_needed(tpl) ) { + tmp_ty = monomorphise_type_with(sp, tpl, monomorph_cb, false); + this->expand_associated_types(sp, tmp_ty); + return tmp_ty; + } + else { + return tpl; + } + }; + TU_MATCHA( (e.binding), (pbe), + (Unbound, + BUG(sp, "Unbound path"); + ), + (Opaque, + // Technically a bug, checked above + return true; + ), + (Struct, + TU_MATCHA( (pbe->m_data), (se), + (Unit, + ), + (Tuple, + for(const auto& e : se) + { + if( type_needs_drop_glue(sp, monomorph(e.ent)) ) + return true; + } + ), + (Named, + for(const auto& e : se) + { + if( type_needs_drop_glue(sp, monomorph(e.second.ent)) ) + return true; + } + ) + ) + return false; + ), + (Enum, + for(const auto& e : pbe->m_variants) + { + TU_MATCHA( (e.second), (ve), + (Unit, + ), + (Value, + ), + (Tuple, + for(const auto& e : ve) + { + if( type_needs_drop_glue(sp, monomorph(e.ent)) ) + return true; + } + ), + (Struct, + for(const auto& e : ve) + { + if( type_needs_drop_glue(sp, monomorph(e.second.ent)) ) + return true; + } + ) + ) + } + return false; + ), + (Union, + // Unions don't have drop glue unless they impl Drop + return false; + ) + ) + ), + (Diverge, + return false; + ), + (Closure, + // TODO: Destructure? + return true; + ), + (Infer, + BUG(sp, "type_needs_drop_glue on _"); + return false; + ), + (Borrow, + // &-ptrs don't have drop glue + if( e.type != ::HIR::BorrowType::Owned ) + return false; + return type_needs_drop_glue(sp, *e.inner); + ), + (Pointer, + return false; + ), + (Function, + return false; + ), + (Primitive, + return false; + ), + (Array, + return type_needs_drop_glue(sp, *e.inner); + ), + (Slice, + return type_needs_drop_glue(sp, *e.inner); + ), + (TraitObject, + return true; + ), + (ErasedType, + // Is this an error? + return true; + ), + (Tuple, + for(const auto& ty : e) + { + if( !type_needs_drop_glue(sp, ty) ) + return true; + } + return false; + ) + ) + assert(!"Fell off the end of type_needs_drop_glue"); + throw ""; +} + const ::HIR::TypeRef* StaticTraitResolve::is_type_owned_box(const ::HIR::TypeRef& ty) const { if( ! ty.m_data.is_Path() ) { @@ -1482,7 +1661,44 @@ StaticTraitResolve::ValuePtr StaticTraitResolve::get_value(const Span& sp, const } else { - TODO(sp, "Search for trait impl"); + ImplRef best_impl; + this->find_impl(sp, pe.trait.m_path, &pe.trait.m_params, *pe.type, [&](auto impl, bool is_fuzz)->bool{ + if( ! impl.m_data.is_TraitImpl() ) + return false; + const auto& ti = *impl.m_data.as_TraitImpl().impl; + auto it = ti.m_constants.find(pe.item); + if(it == ti.m_constants.end()) + return false; + + if( impl.more_specific_than(best_impl) ) + { + best_impl = mv$(impl); + // If this value is specialisable, keep searching (return false) + return !it->second.is_specialisable; + } + // Keep searching + return false; + }); + if( !best_impl.is_valid() ) + { + TODO(sp, "What should be done when an impl can't be found? " << p); + } + + if( ! best_impl.m_data.is_TraitImpl() ) + TODO(sp, "Use bounded constant values for " << p); + auto& ie = best_impl.m_data.as_TraitImpl(); + out_params.pp_impl = &out_params.pp_impl_data; + for(auto ptr : ie.params) + { + // TODO: Avoid cloning when the params are in the placeholder array + out_params.pp_impl_data.m_types.push_back( ptr->clone() ); + } + + const auto& ti = *ie.impl; + const auto& c = ti.m_constants.at(pe.item); + + // TODO: What if the type requires monomorphisation? Leave it up to the caller + return &c.data; } throw ""; ), @@ -1494,6 +1710,7 @@ StaticTraitResolve::ValuePtr StaticTraitResolve::get_value(const Span& sp, const m_crate.find_type_impls(*pe.type, [](const auto&x)->const ::HIR::TypeRef& { return x; }, [&](const auto& impl) { DEBUG("Found impl" << impl.m_params.fmt_args() << " " << impl.m_type); // TODO: Populate pp_impl + // TODO: Specialisation { auto fit = impl.m_methods.find(pe.item); if( fit != impl.m_methods.end() ) diff --git a/src/hir_typeck/static.hpp b/src/hir_typeck/static.hpp index cde9797c..16302218 100644 --- a/src/hir_typeck/static.hpp +++ b/src/hir_typeck/static.hpp @@ -178,6 +178,9 @@ public: bool type_is_copy(const Span& sp, const ::HIR::TypeRef& ty) const; bool type_is_sized(const Span& sp, const ::HIR::TypeRef& ty) const; + /// Returns `true` if the passed type either implements Drop, or contains a type that implements Drop + bool type_needs_drop_glue(const Span& sp, const ::HIR::TypeRef& ty) const; + const ::HIR::TypeRef* is_type_owned_box(const ::HIR::TypeRef& ty) const; const ::HIR::TypeRef* is_type_phantom_data(const ::HIR::TypeRef& ty) const; diff --git a/src/include/main_bindings.hpp b/src/include/main_bindings.hpp index c01ff86d..184f266f 100644 --- a/src/include/main_bindings.hpp +++ b/src/include/main_bindings.hpp @@ -16,6 +16,7 @@ extern AST::Crate Parse_Crate(::std::string mainfile); extern void Expand(::AST::Crate& crate); +extern void Expand_TestHarness(::AST::Crate& crate); /// Process #[] decorators extern void Process_Decorators(AST::Crate& crate); diff --git a/src/main.cpp b/src/main.cpp index 4c8d3104..8616ff2f 100644 --- a/src/main.cpp +++ b/src/main.cpp @@ -68,6 +68,7 @@ void init_debug_list() g_debug_disable_map.insert( "MIR Cleanup" ); g_debug_disable_map.insert( "MIR Optimise" ); g_debug_disable_map.insert( "MIR Validate PO" ); + g_debug_disable_map.insert( "MIR Validate Full" ); g_debug_disable_map.insert( "HIR Serialise" ); g_debug_disable_map.insert( "Trans Enumerate" ); @@ -131,6 +132,11 @@ struct ProgramParams ::AST::Crate::Type crate_type = ::AST::Crate::Type::Unknown; ::std::string crate_name; + unsigned opt_level = 0; + bool emit_debug_info = false; + + bool test_harness = false; + ::std::vector<const char*> lib_search_dirs; ::std::vector<const char*> libraries; @@ -167,10 +173,12 @@ int main(int argc, char *argv[]) ProgramParams params(argc, argv); // Set up cfg values + Cfg_SetValue("rust_compiler", "mrustc"); // TODO: Target spec Cfg_SetFlag("unix"); Cfg_SetFlag("linux"); Cfg_SetValue("target_os", "linux"); + Cfg_SetValue("target_family", "unix"); Cfg_SetValue("target_pointer_width", "64"); Cfg_SetValue("target_endian", "little"); Cfg_SetValue("target_arch", "x86_64"); @@ -188,6 +196,11 @@ int main(int argc, char *argv[]) }); + if( params.test_harness ) + { + Cfg_SetFlag("test"); + } + try { @@ -195,6 +208,7 @@ int main(int argc, char *argv[]) AST::Crate crate = CompilePhase<AST::Crate>("Parse", [&]() { return Parse_Crate(params.infile); }); + crate.m_test_harness = params.test_harness; if( params.last_stage == ProgramParams::STAGE_PARSE ) { return 0; @@ -210,6 +224,11 @@ int main(int argc, char *argv[]) Expand(crate); }); + if( params.test_harness ) + { + Expand_TestHarness(crate); + } + // Extract the crate type and name from the crate attributes auto crate_type = params.crate_type; if( crate_type == ::AST::Crate::Type::Unknown ) { @@ -254,6 +273,10 @@ int main(int argc, char *argv[]) } } crate.m_crate_name = crate_name; + if( params.test_harness ) + { + crate.m_crate_name += "$test"; + } if( params.outfile == "" ) { switch( crate.m_crate_type ) @@ -281,7 +304,7 @@ int main(int argc, char *argv[]) } // Allocator and panic strategies - if( crate.m_crate_type == ::AST::Crate::Type::Executable ) + if( crate.m_crate_type == ::AST::Crate::Type::Executable || params.test_harness ) { // TODO: Detect if an allocator crate is already present. crate.load_extern_crate(Span(), "alloc_system"); @@ -290,6 +313,10 @@ int main(int argc, char *argv[]) // - `mrustc-main` lang item default crate.m_lang_items.insert(::std::make_pair( ::std::string("mrustc-main"), ::AST::Path("", {AST::PathNode("main")}) )); } + if( params.test_harness ) + { + crate.load_extern_crate(Span(), "test"); + } // Resolve names to be absolute names (include references to the relevant struct/global/function) // - This does name checking on types and free functions. @@ -421,6 +448,13 @@ int main(int argc, char *argv[]) CompilePhaseV("MIR Cleanup", [&]() { MIR_CleanupCrate(*hir_crate); }); + if( getenv("MRUSTC_FULL_VALIDATE_PREOPT") ) + { + CompilePhaseV("MIR Validate Full", [&]() { + MIR_CheckCrate_Full(*hir_crate); + }); + } + // Optimise the MIR CompilePhaseV("MIR Optimise", [&]() { MIR_OptimiseCrate(*hir_crate); @@ -433,6 +467,12 @@ int main(int argc, char *argv[]) CompilePhaseV("MIR Validate PO", [&]() { MIR_CheckCrate(*hir_crate); }); + // - Exhaustive MIR validation (follows every code path and checks variable validity) + // > DEBUGGING ONLY + CompilePhaseV("MIR Validate Full", [&]() { + if( getenv("MRUSTC_FULL_VALIDATE") ) + MIR_CheckCrate_Full(*hir_crate); + }); if( params.last_stage == ProgramParams::STAGE_MIR ) { return 0; @@ -449,8 +489,14 @@ int main(int argc, char *argv[]) for(const char* libdir : params.libraries ) { trans_opt.libraries.push_back( libdir ); } + trans_opt.emit_debug_info = params.emit_debug_info; // Generate code for non-generic public items (if requested) + if( params.test_harness ) + { + // If the test harness is enabled, override crate type to "Executable" + crate_type = ::AST::Crate::Type::Executable; + } switch( crate_type ) { case ::AST::Crate::Type::Unknown: @@ -580,6 +626,12 @@ ProgramParams::ProgramParams(int argc, char *argv[]) } this->outfile = argv[++i]; break; + case 'O': + this->opt_level = 2; + break; + case 'g': + this->emit_debug_info = true; + break; default: exit(1); } @@ -672,6 +724,9 @@ ProgramParams::ProgramParams(int argc, char *argv[]) exit(1); } } + else if( strcmp(arg, "--test") == 0 ) { + this->test_harness = true; + } else { ::std::cerr << "Unknown option '" << arg << "'" << ::std::endl; exit(1); @@ -680,3 +735,54 @@ ProgramParams::ProgramParams(int argc, char *argv[]) } } + +::std::ostream& operator<<(::std::ostream& os, const FmtEscaped& x) +{ + os << ::std::hex; + for(auto s = x.s; *s != '\0'; s ++) + { + switch(*s) + { + case '\0': os << "\\0"; break; + case '\n': os << "\\n"; break; + case '\\': os << "\\\\"; break; + case '"': os << "\\\""; break; + default: + uint8_t v = *s; + if( v < 0x80 ) + { + if( v < ' ' || v > 0x7F ) + os << "\\u{" << ::std::hex << (unsigned int)v << "}"; + else + os << v; + } + else if( v < 0xC0 ) + ; + else if( v < 0xE0 ) + { + uint32_t val = (uint32_t)(v & 0x1F) << 6; + v = (uint8_t)*++s; if( (v & 0xC0) != 0x80 ) { s--; continue ; } val |= (uint32_t)v << 6; + os << "\\u{" << ::std::hex << val << "}"; + } + else if( v < 0xF0 ) + { + uint32_t val = (uint32_t)(v & 0x0F) << 12; + v = (uint8_t)*++s; if( (v & 0xC0) != 0x80 ) { s--; continue ; } val |= (uint32_t)v << 12; + v = (uint8_t)*++s; if( (v & 0xC0) != 0x80 ) { s--; continue ; } val |= (uint32_t)v << 6; + os << "\\u{" << ::std::hex << val << "}"; + } + else if( v < 0xF8 ) + { + uint32_t val = (uint32_t)(v & 0x07) << 18; + v = (uint8_t)*++s; if( (v & 0xC0) != 0x80 ) { s--; continue ; } val |= (uint32_t)v << 18; + v = (uint8_t)*++s; if( (v & 0xC0) != 0x80 ) { s--; continue ; } val |= (uint32_t)v << 12; + v = (uint8_t)*++s; if( (v & 0xC0) != 0x80 ) { s--; continue ; } val |= (uint32_t)v << 6; + os << "\\u{" << ::std::hex << val << "}"; + } + break; + } + } + os << ::std::dec; + return os; +} + diff --git a/src/mir/check.cpp b/src/mir/check.cpp index 807f08aa..4b9dfd8b 100644 --- a/src/mir/check.cpp +++ b/src/mir/check.cpp @@ -84,53 +84,6 @@ namespace { } } -namespace { - template<typename T> - struct RunIterable { - const ::std::vector<T>& list; - unsigned int ofs; - ::std::pair<size_t,size_t> cur; - RunIterable(const ::std::vector<T>& list): - list(list), ofs(0) - { - advance(); - } - void advance() { - if( ofs < list.size() ) - { - auto start = ofs; - while(ofs < list.size() && list[ofs] == list[start]) - ofs ++; - cur = ::std::make_pair(start, ofs-1); - } - else - { - ofs = list.size()+1; - } - } - RunIterable<T> begin() { return *this; } - RunIterable<T> end() { auto rv = *this; rv.ofs = list.size()+1; return rv; } - bool operator==(const RunIterable<T>& x) { - return x.ofs == ofs; - } - bool operator!=(const RunIterable<T>& x) { - return !(*this == x); - } - void operator++() { - advance(); - } - const ::std::pair<size_t,size_t>& operator*() const { - return this->cur; - } - const ::std::pair<size_t,size_t>* operator->() const { - return &this->cur; - } - }; - template<typename T> - RunIterable<T> runs(const ::std::vector<T>& x) { - return RunIterable<T>(x); - } -} //template<typename T> //::std::ostream& operator<<(::std::ostream& os, const T& v) { // v.fmt(os); @@ -209,9 +162,9 @@ void MIR_Validate_ValState(::MIR::TypeResolve& state, const ::MIR::Function& fcn return arguments.empty() && temporaries.empty() && variables.empty(); } - bool merge(ValStates& other) + bool merge(unsigned bb_idx, ValStates& other) { - DEBUG("this=" << FMT_CB(ss,this->fmt(ss);) << ", other=" << FMT_CB(ss,other.fmt(ss);)); + DEBUG("bb" << bb_idx << " this=" << FMT_CB(ss,this->fmt(ss);) << ", other=" << FMT_CB(ss,other.fmt(ss);)); if( this->empty() ) { *this = other; @@ -242,14 +195,17 @@ void MIR_Validate_ValState(::MIR::TypeResolve& state, const ::MIR::Function& fcn ), (Argument, MIR_ASSERT(state, e.idx < this->arguments.size(), ""); + DEBUG("arg" << e.idx << " = " << (is_valid ? "Valid" : "Invalid")); this->arguments[e.idx] = is_valid ? State::Valid : State::Invalid; ), (Variable, MIR_ASSERT(state, e < this->variables.size(), ""); + DEBUG("var" << e << " = " << (is_valid ? "Valid" : "Invalid")); this->variables[e] = is_valid ? State::Valid : State::Invalid; ), (Temporary, MIR_ASSERT(state, e.idx < this->temporaries.size(), ""); + DEBUG("tmp" << e.idx << " = " << (is_valid ? "Valid" : "Invalid")); this->temporaries[e.idx] = is_valid ? State::Valid : State::Invalid; ) ) @@ -364,7 +320,7 @@ void MIR_Validate_ValState(::MIR::TypeResolve& state, const ::MIR::Function& fcn // 1. Apply current state to `block_start_states` (merging if needed) // - If no change happened, skip. - if( ! block_start_states.at(block).merge( val_state ) ) { + if( ! block_start_states.at(block).merge(block, val_state) ) { continue ; } DEBUG("BB" << block << " via [" << path << "]"); @@ -376,6 +332,7 @@ void MIR_Validate_ValState(::MIR::TypeResolve& state, const ::MIR::Function& fcn const auto& stmt = bb.statements[stmt_idx]; state.set_cur_stmt(block, stmt_idx); + DEBUG(state << stmt); switch( stmt.tag() ) { case ::MIR::Statement::TAGDEAD: @@ -453,11 +410,18 @@ void MIR_Validate_ValState(::MIR::TypeResolve& state, const ::MIR::Function& fcn // Mark destination as valid val_state.mark_validity( state, stmt.as_Assign().dst, true ); break; + case ::MIR::Statement::TAG_ScopeEnd: + //for(auto idx : stmt.as_ScopeEnd().vars) + // val_state.mark_validity(state, ::MIR::LValue::make_Variable(idx), false); + //for(auto idx : stmt.as_ScopeEnd().tmps) + // val_state.mark_validity(state, ::MIR::LValue::make_Temporary({idx}), false); + break; } } // 3. Pass new state on to destination blocks state.set_cur_stmt_term(block); + DEBUG(state << bb.terminator); TU_MATCH(::MIR::Terminator, (bb.terminator), (e), (Incomplete, // Should be impossible here. @@ -850,6 +814,9 @@ void MIR_Validate(const StaticTraitResolve& resolve, const ::HIR::ItemPath& path case ::MIR::Statement::TAG_Drop: // TODO: Anything need checking here? break; + case ::MIR::Statement::TAG_ScopeEnd: + // TODO: Mark listed values as descoped + break; } } diff --git a/src/mir/check_full.cpp b/src/mir/check_full.cpp new file mode 100644 index 00000000..52c0aaf5 --- /dev/null +++ b/src/mir/check_full.cpp @@ -0,0 +1,981 @@ +/* + * MRustC - Rust Compiler + * - By John Hodge (Mutabah/thePowersGang) + * + * mir/check_full.cpp + * - Full MIR correctness checks (expensive value state checks) + */ +#include "main_bindings.hpp" +#include "mir.hpp" +#include <hir/visitor.hpp> +#include <hir_typeck/static.hpp> +#include <mir/helpers.hpp> +#include <mir/visit_crate_mir.hpp> + +// DISABLED: Unsizing intentionally leaks +#define ENABLE_LEAK_DETECTOR 0 + +namespace +{ + struct State + { + // 0 = invalid + // -1 = valid + // other = 1-based index into `inner_states` + unsigned int index; + + State(): index(0) {} + State(bool valid): index(valid ? ~0u : 0) {} + State(size_t idx): + index(idx+1) + { + } + + bool is_composite() const { + return index != 0 && index != ~0u; + } + bool is_valid() const { + return index != 0; + } + + bool operator==(const State& x) const { + return index == x.index; + } + bool operator!=(const State& x) const { + return !(*this == x); + } + }; + + struct ValueStates; +} + +struct StateFmt { + const ValueStates& vss; + State s; + StateFmt( const ValueStates& vss, State s ): + vss(vss), s(s) + {} +}; + +namespace +{ + struct ValueStates + { + ::std::vector<State> vars; + ::std::vector<State> temporaries; + ::std::vector<State> arguments; + State return_value; + ::std::vector<bool> drop_flags; + + ::std::vector< ::std::vector<State> > inner_states; + + ::std::vector<unsigned int> bb_path; + + ValueStates clone() const + { + return *this; + } + bool is_equivalent_to(const ValueStates& x) const + { + struct H { + static bool equal(const ValueStates& vss_a, const State& a, const ValueStates& vss_b, const State& b) + { + if( a.index == 0 ) + { + return b.index == 0; + } + if( a.index == ~0u ) + { + return b.index == ~0u; + } + if( b.index == 0 || b.index == ~0u ) + { + return false; + } + + const auto& states_a = vss_a.inner_states.at( a.index - 1 ); + const auto& states_b = vss_b.inner_states.at( b.index - 1 ); + // NOTE: If there's two differen variants, this can happen. + if( states_a.size() != states_b.size() ) + return false; + + for(size_t i = 0; i < states_a.size(); i ++) + { + if( ! H::equal(vss_a, states_a[i], vss_b, states_b[i]) ) + return false; + } + // If the above loop didn't early exit, the two states are equal + return true; + } + }; + + if( this->drop_flags != x.drop_flags ) + return false; + if( ! H::equal(*this, return_value, x, x.return_value) ) + return false; + assert(vars.size() == x.vars.size()); + for(size_t i = 0; i < vars.size(); i ++) + { + if( ! H::equal(*this, vars[i], x, x.vars[i]) ) + return false; + } + assert(temporaries.size() == x.temporaries.size()); + for(size_t i = 0; i < temporaries.size(); i ++) + { + if( ! H::equal(*this, temporaries[i], x, x.temporaries[i]) ) + return false; + } + assert(arguments.size() == x.arguments.size()); + for(size_t i = 0; i < arguments.size(); i ++) + { + if( ! H::equal(*this, arguments[i], x, x.arguments[i]) ) + return false; + } + return true; + } + + StateFmt fmt_state(const ::MIR::TypeResolve& mir_res, const ::MIR::LValue& lv) const { + return StateFmt(*this, get_lvalue_state(mir_res, lv)); + } + + void ensure_param_valid(const ::MIR::TypeResolve& mir_res, const ::MIR::Param& lv) const + { + if(const auto* e = lv.opt_LValue()) + { + this->ensure_lvalue_valid(mir_res, *e); + } + } + void ensure_lvalue_valid(const ::MIR::TypeResolve& mir_res, const ::MIR::LValue& lv) const + { + auto vs = get_lvalue_state(mir_res, lv); + ::std::vector<unsigned int> path; + ensure_valid(mir_res, lv, vs, path); + } + private: + struct InvalidReason { + enum { + Unwritten, + Moved, + Invalidated, + } ty; + size_t bb; + size_t stmt; + + void fmt(::std::ostream& os) const { + switch(this->ty) + { + case Unwritten: os << "Not Written"; break; + case Moved: os << "Moved at BB" << bb << "/" << stmt; break; + case Invalidated: os << "Invalidated at BB" << bb << "/" << stmt; break; + } + } + }; + InvalidReason find_invalid_reason(const ::MIR::TypeResolve& mir_res, const ::MIR::LValue& root_lv) const + { + using ::MIR::visit::ValUsage; + using ::MIR::visit::visit_mir_lvalues; + + ::HIR::TypeRef tmp; + bool is_copy = mir_res.m_resolve.type_is_copy( mir_res.sp, mir_res.get_lvalue_type(tmp, root_lv) ); + size_t cur_stmt = mir_res.get_cur_stmt_ofs(); + + // Dump all statements + if(true) + { + for(size_t i = 0; i < this->bb_path.size()-1; i++) + { + size_t bb_idx = this->bb_path[i]; + const auto& bb = mir_res.m_fcn.blocks.at(bb_idx); + + for(size_t stmt_idx = 0; stmt_idx < bb.statements.size(); stmt_idx++) + { + DEBUG("BB" << bb_idx << "/" << stmt_idx << " - " << bb.statements[stmt_idx]); + } + DEBUG("BB" << bb_idx << "/TERM - " << bb.terminator); + } + + { + size_t bb_idx = this->bb_path.back(); + const auto& bb = mir_res.m_fcn.blocks.at(bb_idx); + for(size_t stmt_idx = 0; stmt_idx < cur_stmt; stmt_idx ++) + { + DEBUG("BB" << bb_idx << "/" << stmt_idx << " - " << bb.statements[stmt_idx]); + } + } + } + + if( !is_copy ) + { + // Walk backwards through the BBs and find where it's used by value + assert(this->bb_path.size() > 0); + size_t bb_idx; + size_t stmt_idx; + + bool was_moved = false; + size_t moved_bb, moved_stmt; + auto visit_cb = [&](const auto& lv, auto vu) { + if(lv == root_lv && vu == ValUsage::Move) { + was_moved = true; + moved_bb = bb_idx; + moved_stmt = stmt_idx; + return false; + } + return false; + }; + // Most recent block (incomplete) + { + bb_idx = this->bb_path.back(); + const auto& bb = mir_res.m_fcn.blocks.at(bb_idx); + for(stmt_idx = cur_stmt; stmt_idx -- && !was_moved; ) + { + visit_mir_lvalues(bb.statements[stmt_idx], visit_cb); + } + } + for(size_t i = this->bb_path.size()-1; i -- && !was_moved; ) + { + bb_idx = this->bb_path[i]; + const auto& bb = mir_res.m_fcn.blocks.at(bb_idx); + stmt_idx = bb.statements.size(); + + visit_mir_lvalues(bb.terminator, visit_cb); + + for(stmt_idx = bb.statements.size(); stmt_idx -- && !was_moved; ) + { + visit_mir_lvalues(bb.statements[stmt_idx], visit_cb); + } + } + + if( was_moved ) + { + // Reason found, the value was moved + DEBUG("- Moved in BB" << moved_bb << "/" << moved_stmt); + return InvalidReason { InvalidReason::Moved, moved_bb, moved_stmt }; + } + } + else + { + // Walk backwards to find assignment (if none, it's never initialized) + assert(this->bb_path.size() > 0); + size_t bb_idx; + size_t stmt_idx; + + bool assigned = false; + auto visit_cb = [&](const auto& lv, auto vu) { + if(lv == root_lv && vu == ValUsage::Write) { + assigned = true; + //assigned_bb = this->bb_path[i]; + //assigned_stmt = j; + return true; + } + return false; + }; + + // Most recent block (incomplete) + { + bb_idx = this->bb_path.back(); + const auto& bb = mir_res.m_fcn.blocks.at(bb_idx); + for(stmt_idx = cur_stmt; stmt_idx -- && !assigned; ) + { + visit_mir_lvalues(bb.statements[stmt_idx], visit_cb); + } + } + for(size_t i = this->bb_path.size()-1; i -- && !assigned; ) + { + bb_idx = this->bb_path[i]; + const auto& bb = mir_res.m_fcn.blocks.at(bb_idx); + stmt_idx = bb.statements.size(); + + visit_mir_lvalues(bb.terminator, visit_cb); + + for(stmt_idx = bb.statements.size(); stmt_idx -- && !assigned; ) + { + visit_mir_lvalues(bb.statements[stmt_idx], visit_cb); + } + } + + if( !assigned ) + { + // Value wasn't ever assigned, that's why it's not valid. + DEBUG("- Not assigned"); + return InvalidReason { InvalidReason::Unwritten, 0, 0 }; + } + } + // If neither of the above return a reason, check for blocks that don't have the value valid. + // TODO: This requires access to the lifetime bitmaps to know where it was invalidated + DEBUG("- (assume) lifetime invalidated [is_copy=" << is_copy << "]"); + return InvalidReason { InvalidReason::Invalidated, 0, 0 }; + } + void ensure_valid(const ::MIR::TypeResolve& mir_res, const ::MIR::LValue& root_lv, const State& vs, ::std::vector<unsigned int>& path) const + { + if( vs.is_composite() ) + { + MIR_ASSERT(mir_res, vs.index-1 < this->inner_states.size(), ""); + const auto& states = this->inner_states.at( vs.index - 1 ); + + path.push_back(0); + for(const auto& inner_vs : states) + { + ensure_valid(mir_res,root_lv, inner_vs, path); + path.back() ++; + } + path.pop_back(); + } + else if( !vs.is_valid() ) + { + // Locate where it was invalidated. + auto reason = find_invalid_reason(mir_res, root_lv); + MIR_BUG(mir_res, "Accessing invalidated lvalue - " << root_lv << " - " << FMT_CB(s,reason.fmt(s);) << " - field path=[" << path << "], BBs=[" << this->bb_path << "]"); + } + else + { + } + } + + public: + void move_lvalue(const ::MIR::TypeResolve& mir_res, const ::MIR::LValue& lv) + { + this->ensure_lvalue_valid(mir_res, lv); + + ::HIR::TypeRef tmp; + const auto& ty = mir_res.get_lvalue_type(tmp, lv); + if( mir_res.m_resolve.type_is_copy(mir_res.sp, ty) ) + { + // NOTE: Copy types aren't moved. + } + else + { + this->set_lvalue_state(mir_res, lv, State(false)); + } + } + void mark_lvalue_valid(const ::MIR::TypeResolve& mir_res, const ::MIR::LValue& lv) + { + this->set_lvalue_state(mir_res, lv, State(true)); + } + + // Scan states and clear unused composite slots + void garbage_collect() + { + struct Marker { + ::std::vector<bool> used; + + void mark_from_state(const ValueStates& vss, const State& s) { + if(s.is_composite()) { + used.at(s.index-1) = true; + for(const auto& s : vss.inner_states.at(s.index-1)) + mark_from_state(vss, s); + + // TODO: Should this compact composites with all-equal inner states? + } + } + }; + Marker m; + m.used.resize(this->inner_states.size(), false); + + for(const auto& s : this->vars) + m.mark_from_state(*this, s); + for(const auto& s : this->temporaries) + m.mark_from_state(*this, s); + for(const auto& s : this->arguments) + m.mark_from_state(*this, s); + m.mark_from_state(*this, this->return_value); + } + private: + State allocate_composite(unsigned int n_fields, State basis) + { + assert(n_fields > 0); + for(size_t i = 0; i < this->inner_states.size(); i ++) + { + if( this->inner_states[i].size() == 0 ) + { + inner_states[i] = ::std::vector<State>(n_fields, basis); + return State(i); + } + } + auto idx = inner_states.size(); + inner_states.push_back( ::std::vector<State>(n_fields, basis) ); + return State(idx); + } + + public: + ::std::vector<State>& get_composite(const ::MIR::TypeResolve& mir_res, const State& vs) + { + MIR_ASSERT(mir_res, vs.index-1 < this->inner_states.size(), ""); + return this->inner_states.at( vs.index - 1 ); + } + const ::std::vector<State>& get_composite(const ::MIR::TypeResolve& mir_res, const State& vs) const + { + MIR_ASSERT(mir_res, vs.index-1 < this->inner_states.size(), ""); + return this->inner_states.at( vs.index - 1 ); + } + State get_lvalue_state(const ::MIR::TypeResolve& mir_res, const ::MIR::LValue& lv) const + { + TU_MATCHA( (lv), (e), + (Variable, + return vars.at(e); + ), + (Temporary, + return temporaries.at(e.idx); + ), + (Argument, + return arguments.at(e.idx); + ), + (Static, + return State(true); + ), + (Return, + return return_value; + ), + (Field, + auto vs = get_lvalue_state(mir_res, *e.val); + if( vs.is_composite() ) + { + const auto& states = this->get_composite(mir_res, vs); + MIR_ASSERT(mir_res, e.field_index < states.size(), "Field index out of range"); + return states[e.field_index]; + } + else + { + return vs; + } + ), + (Deref, + auto vs = get_lvalue_state(mir_res, *e.val); + if( vs.is_composite() ) + { + MIR_TODO(mir_res, "Deref with composite state"); + } + else + { + return vs; + } + ), + (Index, + auto vs_v = get_lvalue_state(mir_res, *e.val); + auto vs_i = get_lvalue_state(mir_res, *e.idx); + MIR_ASSERT(mir_res, !vs_v.is_composite(), ""); + MIR_ASSERT(mir_res, !vs_i.is_composite(), ""); + return State(vs_v.is_valid() && vs_i.is_valid()); + ), + (Downcast, + auto vs_v = get_lvalue_state(mir_res, *e.val); + if( vs_v.is_composite() ) + { + const auto& states = this->get_composite(mir_res, vs_v); + MIR_ASSERT(mir_res, states.size() == 1, "Downcast on composite of invalid size"); + return states[0]; + } + else + { + return vs_v; + } + ) + ) + throw ""; + } + + void set_lvalue_state(const ::MIR::TypeResolve& mir_res, const ::MIR::LValue& lv, State new_vs) + { + TU_MATCHA( (lv), (e), + (Variable, + vars.at(e) = new_vs; + ), + (Temporary, + temporaries.at(e.idx) = new_vs; + ), + (Argument, + arguments.at(e.idx) = new_vs; + ), + (Static, + // Ignore. + ), + (Return, + return_value = new_vs; + ), + (Field, + auto cur_vs = get_lvalue_state(mir_res, *e.val); + if( !cur_vs.is_composite() && cur_vs == new_vs ) + { + // Not a composite, and no state change + } + else + { + if( !cur_vs.is_composite() ) + { + ::HIR::TypeRef tmp; + const auto& ty = mir_res.get_lvalue_type(tmp, *e.val); + unsigned int n_fields = 0; + if( const auto* e = ty.m_data.opt_Tuple() ) + { + n_fields = e->size(); + } + else if( ty.m_data.is_Path() && ty.m_data.as_Path().binding.is_Struct() ) + { + const auto& e = ty.m_data.as_Path().binding.as_Struct(); + TU_MATCHA( (e->m_data), (se), + (Unit, + n_fields = 0; + ), + (Tuple, + n_fields = se.size(); + ), + (Named, + n_fields = se.size(); + ) + ) + } + else { + MIR_BUG(mir_res, "Unknown type being accessed with Field - " << ty); + } + cur_vs = this->allocate_composite(n_fields, cur_vs); + set_lvalue_state(mir_res, *e.val, cur_vs); + } + // Get composite state and assign into it + auto& states = this->get_composite(mir_res, cur_vs); + MIR_ASSERT(mir_res, e.field_index < states.size(), "Field index out of range"); + states[e.field_index] = new_vs; + } + ), + (Deref, + auto cur_vs = get_lvalue_state(mir_res, *e.val); + if( !cur_vs.is_composite() && cur_vs == new_vs ) + { + // Not a composite, and no state change + } + else + { + if( !cur_vs.is_composite() ) + { + //::HIR::TypeRef tmp; + //const auto& ty = mir_res.get_lvalue_type(tmp, *e.val); + // TODO: Should this check if the type is Box? + + cur_vs = this->allocate_composite(2, cur_vs); + set_lvalue_state(mir_res, *e.val, cur_vs); + } + // Get composite state and assign into it + auto& states = this->get_composite(mir_res, cur_vs); + MIR_ASSERT(mir_res, states.size() == 2, "Deref with invalid state list size"); + states[1] = new_vs; + } + ), + (Index, + auto vs_v = get_lvalue_state(mir_res, *e.val); + auto vs_i = get_lvalue_state(mir_res, *e.idx); + MIR_ASSERT(mir_res, !vs_v.is_composite(), ""); + MIR_ASSERT(mir_res, !vs_i.is_composite(), ""); + + MIR_ASSERT(mir_res, vs_v.is_valid(), "Indexing an invalid value"); + MIR_ASSERT(mir_res, vs_i.is_valid(), "Indexing with an invalid index"); + + // NOTE: Ignore + ), + (Downcast, + auto cur_vs = get_lvalue_state(mir_res, *e.val); + if( !cur_vs.is_composite() && cur_vs == new_vs ) + { + // Not a composite, and no state change + } + else + { + if( !cur_vs.is_composite() ) + { + cur_vs = this->allocate_composite(1, cur_vs); + set_lvalue_state(mir_res, *e.val, cur_vs); + } + // Get composite state and assign into it + auto& states = this->get_composite(mir_res, cur_vs); + MIR_ASSERT(mir_res, states.size() == 1, "Downcast on composite of invalid size"); + states[0] = new_vs; + } + ) + ) + } + }; + + + struct StateSet + { + ::std::vector<ValueStates> known_state_sets; + + bool add_state(const ValueStates& state_set) + { + for(const auto& s : this->known_state_sets) + { + if( s.is_equivalent_to(state_set) ) + { + return false; + } + } + this->known_state_sets.push_back( state_set.clone() ); + this->known_state_sets.back().bb_path = ::std::vector<unsigned int>(); + return true; + } + }; +} + +::std::ostream& operator<<(::std::ostream& os, const StateFmt& x) +{ + if(x.s.index == 0) { + os << "_"; + } + else if( x.s.index == ~0u ) { + os << "X"; + } + else { + assert(x.s.index-1 < x.vss.inner_states.size()); + const auto& is = x.vss.inner_states[x.s.index-1]; + os << "["; + for(const auto& s : is) + os << StateFmt(x.vss, s); + os << "]"; + } + return os; +} + +namespace std { + ostream& operator<<(ostream& os, const ValueStates& x) + { + auto print_val = [&](auto tag, const State& s) { + if(s.is_composite()) { + os << tag << "=" << StateFmt(x,s); + } + else if( s.is_valid() ) { + os << tag; + } + else { + } + }; + + os << "ValueStates(path=[" << x.bb_path << "]"; + print_val(",rv", x.return_value); + for(unsigned int i = 0; i < x.arguments.size(); i ++) + print_val(FMT_CB(ss, ss << ",a" << i;), x.arguments[i]); + for(unsigned int i = 0; i < x.vars.size(); i ++) + print_val(FMT_CB(ss, ss << ",_" << i;), x.vars[i]); + for(unsigned int i = 0; i < x.temporaries.size(); i ++) + print_val(FMT_CB(ss, ss << ",t" << i;), x.temporaries[i]); + for(unsigned int i = 0; i < x.drop_flags.size(); i++) + if(x.drop_flags[i]) + os << ",df" << i; + os << ")"; + return os; + } +} + + +// "Executes" the function, keeping track of drop flags and variable validities +void MIR_Validate_FullValState(::MIR::TypeResolve& mir_res, const ::MIR::Function& fcn) +{ + // TODO: Use a timer to check elapsed CPU time in this function, and check on each iteration + // - If more than `n` (10?) seconds passes on one function, warn and abort + //ElapsedTimeCounter timer; + ::std::vector<StateSet> block_entry_states( fcn.blocks.size() ); + + // Determine value lifetimes (BBs in which Copy values are valid) + // - Used to mask out Copy value (prevents combinatorial explosion) + auto lifetimes = MIR_Helper_GetLifetimes(mir_res, fcn, /*dump_debug=*/true); + DEBUG(lifetimes.m_block_offsets); + + ValueStates state; + state.arguments.resize( mir_res.m_args.size(), State(true) ); + state.vars.resize( fcn.named_variables.size() ); + state.temporaries.resize( fcn.temporaries.size() ); + state.drop_flags = fcn.drop_flags; + + ::std::vector< ::std::pair<unsigned int, ValueStates> > todo_queue; + todo_queue.push_back( ::std::make_pair(0, mv$(state)) ); + while( ! todo_queue.empty() ) + { + auto cur_block = todo_queue.back().first; + auto state = mv$(todo_queue.back().second); + todo_queue.pop_back(); + + // Mask off any values which aren't valid in the first statement of this block + { + for(unsigned i = 0; i < state.vars.size(); i ++) + { + /*if( !variables_copy[i] ) + { + // Not Copy, don't apply masking + } + else*/ if( ! state.vars[i].is_valid() ) + { + // Already invalid + } + else if( lifetimes.var_valid(i, cur_block, 0) ) + { + // Expected to be valid in this block, leave as-is + } + else + { + // Copy value not used at/after this block, mask to false + DEBUG("BB" << cur_block << " - var$" << i << " - Outside lifetime, discard"); + state.vars[i] = State(false); + } + } + for(unsigned i = 0; i < state.temporaries.size(); i ++) + { + /*if( !variables_copy[i] ) + { + // Not Copy, don't apply masking + } + else*/ if( ! state.temporaries[i].is_valid() ) + { + // Already invalid + } + else if( lifetimes.tmp_valid(i, cur_block, 0) ) + { + // Expected to be valid in this block, leave as-is + } + else + { + // Copy value not used at/after this block, mask to false + DEBUG("BB" << cur_block << " - tmp$" << i << " - Outside lifetime, discard"); + state.temporaries[i] = State(false); + } + } + } + + // If this state already exists in the map, skip + if( ! block_entry_states[cur_block].add_state(state) ) + { + DEBUG("BB" << cur_block << " - Nothing new"); + continue ; + } + DEBUG("BB" << cur_block << " - " << state); + state.bb_path.push_back( cur_block ); + + const auto& blk = fcn.blocks.at(cur_block); + for(size_t i = 0; i < blk.statements.size(); i++) + { + mir_res.set_cur_stmt(cur_block, i); + + DEBUG(mir_res << blk.statements[i]); + + TU_MATCHA( (blk.statements[i]), (se), + (Assign, + #if ENABLE_LEAK_DETECTOR + // TODO: Check if the target isn't valid. Allow if either invaid, or too complex to know. + #endif + TU_MATCHA( (se.src), (ve), + (Use, + state.move_lvalue(mir_res, ve); + ), + (Constant, + ), + (SizedArray, + state.ensure_param_valid(mir_res, ve.val); + ), + (Borrow, + state.ensure_lvalue_valid(mir_res, ve.val); + ), + // Cast on primitives + (Cast, + state.ensure_lvalue_valid(mir_res, ve.val); + ), + // Binary operation on primitives + (BinOp, + state.ensure_param_valid(mir_res, ve.val_l); + state.ensure_param_valid(mir_res, ve.val_r); + ), + // Unary operation on primitives + (UniOp, + state.ensure_lvalue_valid(mir_res, ve.val); + ), + // Extract the metadata from a DST pointer + // NOTE: If used on an array, this yields the array size (for generics) + (DstMeta, + state.ensure_lvalue_valid(mir_res, ve.val); + ), + // Extract the pointer from a DST pointer (as *const ()) + (DstPtr, + state.ensure_lvalue_valid(mir_res, ve.val); + ), + // Construct a DST pointer from a thin pointer and metadata + (MakeDst, + state.ensure_param_valid(mir_res, ve.ptr_val); + state.ensure_param_valid(mir_res, ve.meta_val); + ), + (Tuple, + for(const auto& v : ve.vals) + if(const auto* e = v.opt_LValue()) + state.move_lvalue(mir_res, *e); + ), + // Array literal + (Array, + for(const auto& v : ve.vals) + if(const auto* e = v.opt_LValue()) + state.move_lvalue(mir_res, *e); + ), + // Create a new instance of a union (and eventually enum) + (Variant, + if(const auto* e = ve.val.opt_LValue()) + state.move_lvalue(mir_res, *e); + ), + // Create a new instance of a struct (or enum) + (Struct, + for(const auto& v : ve.vals) + if(const auto* e = v.opt_LValue()) + state.move_lvalue(mir_res, *e); + ) + ) + state.mark_lvalue_valid(mir_res, se.dst); + ), + (Asm, + for(const auto& v : se.inputs) + state.ensure_lvalue_valid(mir_res, v.second); + for(const auto& v : se.outputs) + state.mark_lvalue_valid(mir_res, v.second); + ), + (SetDropFlag, + if( se.other == ~0u ) + { + state.drop_flags[se.idx] = se.new_val; + } + else + { + state.drop_flags[se.idx] = (se.new_val != state.drop_flags[se.other]); + } + ), + (Drop, + if( se.flag_idx == ~0u || state.drop_flags.at(se.flag_idx) ) + { + if( se.kind == ::MIR::eDropKind::SHALLOW ) + { + // HACK: A move out of a Box generates the following pattern: `[[[[X_]]X]]` + // - Ensure that that is the pattern we're seeing here. + auto vs = state.get_lvalue_state(mir_res, se.slot); + + MIR_ASSERT(mir_res, vs.index != ~0u, "Shallow drop on fully-valid value - " << se.slot); + + // Box<T> - Wrapper around Unique<T> + MIR_ASSERT(mir_res, vs.is_composite(), "Shallow drop on non-composite state - " << se.slot << " (state=" << StateFmt(state,vs) << ")"); + const auto& sub_states = state.get_composite(mir_res, vs); + MIR_ASSERT(mir_res, sub_states.size() == 1, ""); + // Unique<T> - NonZero<*const T>, PhantomData<T> + MIR_ASSERT(mir_res, sub_states[0].is_composite(), ""); + const auto& sub_states2 = state.get_composite(mir_res, sub_states[0]); + MIR_ASSERT(mir_res, sub_states2.size() == 2, "- " << StateFmt(state, sub_states[0])); + MIR_ASSERT(mir_res, sub_states2[0].is_composite(), ""); + MIR_ASSERT(mir_res, sub_states2[1].is_valid(), ""); + // `NonZero<*const T>` - *const T + const auto& sub_states3 = state.get_composite(mir_res, sub_states2[0]); + MIR_ASSERT(mir_res, sub_states3.size() == 1, "- " << StateFmt(state, sub_states2[0])); + MIR_ASSERT(mir_res, sub_states3[0].is_composite(), ""); + // `*const T` - Moved out of, so has a composite state + const auto& sub_states4 = state.get_composite(mir_res, sub_states3[0]); + MIR_ASSERT(mir_res, sub_states4.size() == 2, "- " << StateFmt(state, sub_states3[0])); + MIR_ASSERT(mir_res, sub_states4[0].is_valid(), "Shallow drop on deallocated Box - " << se.slot << " (state=" << StateFmt(state,vs) << ")"); + // TODO: This is leak protection, enable it once the rest works + if( ENABLE_LEAK_DETECTOR ) + { + MIR_ASSERT(mir_res, !sub_states4[1].is_valid(), "Shallow drop on populated Box - " << se.slot << " (state=" << StateFmt(state,vs) << ")"); + } + + state.set_lvalue_state(mir_res, se.slot, State(false)); + } + else + { + state.move_lvalue(mir_res, se.slot); + } + } + ), + (ScopeEnd, + // TODO: Mark all mentioned variables as invalid + ) + ) + } + + state.garbage_collect(); + + mir_res.set_cur_stmt_term(cur_block); + DEBUG(mir_res << " " << blk.terminator); + TU_MATCHA( (blk.terminator), (te), + (Incomplete, + ), + (Return, + state.ensure_lvalue_valid(mir_res, ::MIR::LValue::make_Return({})); + if( ENABLE_LEAK_DETECTOR ) + { + auto ensure_dropped = [&](const State& s, const ::MIR::LValue& lv) { + if( s.is_valid() ) { + // Check if !Copy + ::HIR::TypeRef tmp; + const auto& ty = mir_res.get_lvalue_type(tmp, lv); + if( mir_res.m_resolve.type_is_copy(mir_res.sp, ty) ) { + } + else { + MIR_BUG(mir_res, "Value " << lv << " was not dropped at end of function"); + } + } + }; + for(unsigned i = 0; i < state.arguments.size(); i ++ ) { + ensure_dropped(state.arguments[i], ::MIR::LValue::make_Argument({i})); + } + for(unsigned i = 0; i < state.vars.size(); i ++ ) { + ensure_dropped(state.vars[i], ::MIR::LValue::make_Variable(i)); + } + } + ), + (Diverge, + ), + (Goto, // Jump to another block + todo_queue.push_back( ::std::make_pair(te, mv$(state)) ); + ), + (Panic, + todo_queue.push_back( ::std::make_pair(te.dst, mv$(state)) ); + ), + (If, + state.ensure_lvalue_valid(mir_res, te.cond); + todo_queue.push_back( ::std::make_pair(te.bb0, state.clone()) ); + todo_queue.push_back( ::std::make_pair(te.bb1, mv$(state)) ); + ), + (Switch, + state.ensure_lvalue_valid(mir_res, te.val); + for(size_t i = 0; i < te.targets.size(); i ++) + { + todo_queue.push_back( ::std::make_pair(te.targets[i], i == te.targets.size()-1 ? mv$(state) : state.clone()) ); + } + ), + (Call, + if(const auto* e = te.fcn.opt_Value()) + { + state.ensure_lvalue_valid(mir_res, *e); + } + for(auto& arg : te.args) + { + if(const auto* e = arg.opt_LValue()) + { + state.move_lvalue(mir_res, *e); + } + } + todo_queue.push_back( ::std::make_pair(te.panic_block, state.clone()) ); + state.mark_lvalue_valid(mir_res, te.ret_val); + todo_queue.push_back( ::std::make_pair(te.ret_block, mv$(state)) ); + ) + ) + } +} + +void MIR_Validate_Full(const StaticTraitResolve& resolve, const ::HIR::ItemPath& path, const ::MIR::Function& fcn, const ::HIR::Function::args_t& args, const ::HIR::TypeRef& ret_type) +{ + TRACE_FUNCTION_F(path); + Span sp; + ::MIR::TypeResolve state { sp, resolve, FMT_CB(ss, ss << path;), ret_type, args, fcn }; + // Validation rules: + + MIR_Validate_FullValState(state, fcn); +} + +// -------------------------------------------------------------------- + +void MIR_CheckCrate_Full(/*const*/ ::HIR::Crate& crate) +{ + ::MIR::OuterVisitor ov(crate, [](const auto& res, const auto& p, auto& expr, const auto& args, const auto& ty) + { + MIR_Validate_Full(res, p, *expr.m_mir, args, ty); + } + ); + ov.visit_crate( crate ); +} + diff --git a/src/mir/cleanup.cpp b/src/mir/cleanup.cpp index 6516dd60..17dca948 100644 --- a/src/mir/cleanup.cpp +++ b/src/mir/cleanup.cpp @@ -932,6 +932,8 @@ void MIR_Cleanup(const StaticTraitResolve& resolve, const ::HIR::ItemPath& path, ), (SetDropFlag, ), + (ScopeEnd, + ), (Asm, for(auto& v : se.inputs) MIR_Cleanup_LValue(state, mutator, v.second); diff --git a/src/mir/dump.cpp b/src/mir/dump.cpp index f3dee478..b4177295 100644 --- a/src/mir/dump.cpp +++ b/src/mir/dump.cpp @@ -9,6 +9,7 @@ #include <hir/visitor.hpp> #include "mir.hpp" #include "operations.hpp" +#include <iomanip> namespace { @@ -98,6 +99,14 @@ namespace { m_os << " IF df$" << e.flag_idx; } m_os << ");\n"; + ), + (ScopeEnd, + m_os << "// Scope End: "; + for(auto idx : e.vars) + m_os << "var$" << idx << ","; + for(auto idx : e.tmps) + m_os << "tmp$" << idx << ","; + m_os << "\n"; ) ) } @@ -208,7 +217,19 @@ namespace { os << (ce.v ? "true" : "false"); ), (Bytes, - os << "b\"" << ce << "\""; + os << ::std::hex << "b\""; + for(auto b : ce) + { + if( b == '\\' ) + os << "\\\\"; + else if( b == '"' ) + os << "\\\""; + else if( ' ' <= b && b < 0x7F ) + os << b; + else + os << "\\x" << ::std::setw(2) << ::std::setfill('0') << (int)b; + } + os << ::std::dec << "\""; ), (StaticString, os << "\"" << ce << "\""; @@ -494,6 +515,30 @@ namespace { m_os << indent() << " ;\n"; } } + void visit_constant(::HIR::ItemPath p, ::HIR::Constant& item) override + { + m_os << indent(); + m_os << "const "; + if( m_short_item_name ) + m_os << p.get_name(); + else + m_os << p; + m_os << ": " << item.m_type; + if( item.m_value ) + { + inc_indent(); + m_os << " = {\n"; + inc_indent(); + dump_mir(m_os, m_indent_level, *item.m_value.m_mir); + dec_indent(); + m_os << indent() << "} /* = " << item.m_value_res << "*/;\n"; + dec_indent(); + } + else + { + m_os << ";\n"; + } + } void visit_static(::HIR::ItemPath p, ::HIR::Static& item) override { m_os << indent(); @@ -506,7 +551,7 @@ namespace { if( item.m_value ) { inc_indent(); - m_os << "= {\n"; + m_os << " = {\n"; inc_indent(); dump_mir(m_os, m_indent_level, *item.m_value.m_mir); dec_indent(); diff --git a/src/mir/from_hir.cpp b/src/mir/from_hir.cpp index e62722ef..ceda0a87 100644 --- a/src/mir/from_hir.cpp +++ b/src/mir/from_hir.cpp @@ -21,6 +21,26 @@ namespace { + template<typename T> + struct SaveAndEditVal { + T& m_dst; + T m_saved; + SaveAndEditVal(T& dst, T newval): + m_dst(dst), + m_saved(dst) + { + m_dst = mv$(newval); + } + ~SaveAndEditVal() + { + this->m_dst = this->m_saved; + } + }; + template<typename T> + SaveAndEditVal<T> save_and_edit(T& dst, typename ::std::remove_reference<T&>::type newval) { + return SaveAndEditVal<T> { dst, mv$(newval) }; + } + class ExprVisitor_Conv: public MirConverter { @@ -36,6 +56,11 @@ namespace { }; ::std::vector<LoopDesc> m_loop_stack; + const ScopeHandle* m_block_tmp_scope = nullptr; + const ScopeHandle* m_borrow_raise_target = nullptr; + const ScopeHandle* m_stmt_scope = nullptr; + bool m_in_borrow = false; + public: ExprVisitor_Conv(MirBuilder& builder, const ::std::vector< ::HIR::TypeRef>& var_types): m_builder(builder), @@ -153,11 +178,22 @@ namespace { m_builder.push_stmt_assign( sp, ::MIR::LValue::make_Variable(pat.m_binding.m_slot), mv$(lval) ); break; case ::HIR::PatternBinding::Type::Ref: + if(m_borrow_raise_target) + { + DEBUG("- Raising destructure borrow of " << lval << " to scope " << *m_borrow_raise_target); + m_builder.raise_variables(sp, lval, *m_borrow_raise_target); + } + m_builder.push_stmt_assign( sp, ::MIR::LValue::make_Variable(pat.m_binding.m_slot), ::MIR::RValue::make_Borrow({ 0, ::HIR::BorrowType::Shared, mv$(lval) }) ); break; case ::HIR::PatternBinding::Type::MutRef: + if(m_borrow_raise_target) + { + DEBUG("- Raising destructure borrow of " << lval << " to scope " << *m_borrow_raise_target); + m_builder.raise_variables(sp, lval, *m_borrow_raise_target); + } m_builder.push_stmt_assign( sp, ::MIR::LValue::make_Variable(pat.m_binding.m_slot), ::MIR::RValue::make_Borrow({ 0, ::HIR::BorrowType::Unique, mv$(lval) }) ); @@ -331,7 +367,7 @@ namespace { if( e.extra_bind.is_valid() ) { // 1. Obtain remaining length - auto sub_val = m_builder.lvalue_or_temp(sp, ::HIR::CoreType::Usize, ::MIR::Constant::make_Uint({ e.leading.size() + e.trailing.size(), ::HIR::CoreType::Usize })); + auto sub_val = ::MIR::Param(::MIR::Constant::make_Uint({ e.leading.size() + e.trailing.size(), ::HIR::CoreType::Usize })); ::MIR::LValue len_val = m_builder.lvalue_or_temp(sp, ::HIR::CoreType::Usize, ::MIR::RValue::make_BinOp({ len_lval.clone(), ::MIR::eBinOp::SUB, mv$(sub_val) }) ); // 2. Obtain pointer to element @@ -370,79 +406,88 @@ namespace { { TRACE_FUNCTION_F("_Block"); // NOTE: This doesn't create a BB, as BBs are not needed for scoping - if( node.m_nodes.size() > 0 ) - { - bool diverged = false; + bool diverged = false; - auto scope = m_builder.new_scope_var(node.span()); + auto res_val = (node.m_value_node ? m_builder.new_temporary(node.m_res_type) : ::MIR::LValue()); + auto scope = m_builder.new_scope_var(node.span()); + auto tmp_scope = m_builder.new_scope_temp(node.span()); + auto _block_tmp_scope = save_and_edit(m_block_tmp_scope, &tmp_scope); - for(unsigned int i = 0; i < node.m_nodes.size() - (node.m_yields_final ? 1 : 0); i ++) - { - auto& subnode = node.m_nodes[i]; - const Span& sp = subnode->span(); + for(unsigned int i = 0; i < node.m_nodes.size(); i ++) + { + auto _ = save_and_edit(m_borrow_raise_target, nullptr); + auto& subnode = node.m_nodes[i]; + const Span& sp = subnode->span(); - auto stmt_scope = m_builder.new_scope_temp(sp); - this->visit_node_ptr(subnode); + auto stmt_scope = m_builder.new_scope_temp(sp); + auto _stmt_scope_push = save_and_edit(m_stmt_scope, &stmt_scope); + this->visit_node_ptr(subnode); - if( m_builder.block_active() || m_builder.has_result() ) { - // TODO: Emit a drop - m_builder.get_result(sp); - m_builder.terminate_scope(sp, mv$(stmt_scope)); - } - else { - m_builder.terminate_scope(sp, mv$(stmt_scope), false); + if( m_builder.block_active() || m_builder.has_result() ) { + // TODO: Emit a drop + m_builder.get_result(sp); + m_builder.terminate_scope(sp, mv$(stmt_scope)); + } + else { + m_builder.terminate_scope(sp, mv$(stmt_scope), false); - m_builder.set_cur_block( m_builder.new_bb_unlinked() ); - diverged = true; - } + m_builder.set_cur_block( m_builder.new_bb_unlinked() ); + diverged = true; } + } - // For the last node, specially handle. - if( node.m_yields_final ) - { - auto& subnode = node.m_nodes.back(); - const Span& sp = subnode->span(); + // For the last node, specially handle. + // TODO: Any temporaries defined within this node must be elevated into the parent scope + if( node.m_value_node ) + { + auto& subnode = node.m_value_node; + const Span& sp = subnode->span(); - auto res_val = m_builder.new_temporary(node.m_res_type); - auto stmt_scope = m_builder.new_scope_temp(sp); - this->visit_node_ptr(subnode); - if( m_builder.has_result() || m_builder.block_active() ) - { - ASSERT_BUG(sp, m_builder.block_active(), "Result yielded, but no active block"); - ASSERT_BUG(sp, m_builder.has_result(), "Active block but no result yeilded"); - // PROBLEM: This can drop the result before we want to use it. + auto stmt_scope = m_builder.new_scope_temp(sp); + this->visit_node_ptr(subnode); + if( m_builder.has_result() || m_builder.block_active() ) + { + ASSERT_BUG(sp, m_builder.block_active(), "Result yielded, but no active block"); + ASSERT_BUG(sp, m_builder.has_result(), "Active block but no result yeilded"); + // PROBLEM: This can drop the result before we want to use it. - m_builder.push_stmt_assign(sp, res_val.clone(), m_builder.get_result(sp)); + m_builder.push_stmt_assign(sp, res_val.clone(), m_builder.get_result(sp)); - m_builder.terminate_scope(sp, mv$(stmt_scope)); - m_builder.terminate_scope( node.span(), mv$(scope) ); - m_builder.set_result( node.span(), mv$(res_val) ); + // If this block is part of a statement, raise all temporaries from this final scope to the enclosing scope + if( m_stmt_scope ) + { + m_builder.raise_all(sp, mv$(stmt_scope), *m_stmt_scope); + //m_builder.terminate_scope(sp, mv$(stmt_scope)); } else { - m_builder.terminate_scope( node.span(), mv$(stmt_scope), false ); - m_builder.terminate_scope( node.span(), mv$(scope), false ); - // Block diverged in final node. + m_builder.terminate_scope(sp, mv$(stmt_scope)); } + m_builder.set_result( node.span(), mv$(res_val) ); } else { - if( diverged ) - { - m_builder.terminate_scope( node.span(), mv$(scope), false ); - m_builder.end_block( ::MIR::Terminator::make_Diverge({}) ); - // Don't set a result if there's no block. - } - else - { - m_builder.terminate_scope( node.span(), mv$(scope) ); - m_builder.set_result(node.span(), ::MIR::RValue::make_Tuple({})); - } + m_builder.terminate_scope( sp, mv$(stmt_scope), false ); + // Block diverged in final node. } + m_builder.terminate_scope( node.span(), mv$(tmp_scope), m_builder.block_active() ); + m_builder.terminate_scope( node.span(), mv$(scope), m_builder.block_active() ); } else { - m_builder.set_result(node.span(), ::MIR::RValue::make_Tuple({})); + if( diverged ) + { + m_builder.terminate_scope( node.span(), mv$(tmp_scope), false ); + m_builder.terminate_scope( node.span(), mv$(scope), false ); + m_builder.end_block( ::MIR::Terminator::make_Diverge({}) ); + // Don't set a result if there's no block. + } + else + { + m_builder.terminate_scope( node.span(), mv$(tmp_scope) ); + m_builder.terminate_scope( node.span(), mv$(scope) ); + m_builder.set_result(node.span(), ::MIR::RValue::make_Tuple({})); + } } } void visit(::HIR::ExprNode_Asm& node) override @@ -477,23 +522,25 @@ namespace { } void visit(::HIR::ExprNode_Let& node) override { - TRACE_FUNCTION_F("_Let"); + TRACE_FUNCTION_F("_Let " << node.m_pattern); this->define_vars_from(node.span(), node.m_pattern); if( node.m_value ) { + auto _ = save_and_edit(m_borrow_raise_target, m_block_tmp_scope); this->visit_node_ptr(node.m_value); if( ! m_builder.block_active() ) { return ; } + auto res = m_builder.get_result(node.span()); if( node.m_pattern.m_binding.is_valid() && node.m_pattern.m_data.is_Any() && node.m_pattern.m_binding.m_type == ::HIR::PatternBinding::Type::Move ) { - m_builder.push_stmt_assign( node.span(), ::MIR::LValue::make_Variable(node.m_pattern.m_binding.m_slot), m_builder.get_result(node.span()) ); + m_builder.push_stmt_assign( node.span(), ::MIR::LValue::make_Variable(node.m_pattern.m_binding.m_slot), mv$(res) ); } else { - this->destructure_from(node.span(), node.m_pattern, m_builder.get_result_in_lvalue(node.m_value->span(), node.m_type)); + this->destructure_from(node.span(), node.m_pattern, m_builder.lvalue_or_temp(node.m_value->span(), node.m_type, mv$(res))); } } m_builder.set_result(node.span(), ::MIR::RValue::make_Tuple({})); @@ -505,6 +552,10 @@ namespace { auto loop_block = m_builder.new_bb_linked(); auto loop_next = m_builder.new_bb_unlinked(); + auto loop_tmp_scope = m_builder.new_scope_temp(node.span()); + auto _ = save_and_edit(m_stmt_scope, &loop_tmp_scope); + + // TODO: `continue` in a loop should jump to the cleanup, not the top m_loop_stack.push_back( LoopDesc { mv$(loop_body_scope), node.m_label, loop_block, loop_next } ); this->visit_node_ptr(node.m_code); auto loop_scope = mv$(m_loop_stack.back().scope); @@ -513,7 +564,7 @@ namespace { // If there's a stray result, drop it if( m_builder.has_result() ) { assert( m_builder.block_active() ); - // TODO: Properly drop this? Or just discard it? + // TODO: Properly drop this? Or just discard it? It should be () m_builder.get_result(node.span()); } // Terminate block with a jump back to the start @@ -522,12 +573,14 @@ namespace { { DEBUG("- Reached end, loop back"); // Insert drop of all scopes within the current scope + m_builder.terminate_scope( node.span(), mv$(loop_tmp_scope) ); m_builder.terminate_scope( node.span(), mv$(loop_scope) ); m_builder.end_block( ::MIR::Terminator::make_Goto(loop_block) ); } else { // Terminate scope without emitting cleanup (cleanup was handled by `break`) + m_builder.terminate_scope( node.span(), mv$(loop_tmp_scope), false ); m_builder.terminate_scope( node.span(), mv$(loop_scope), false ); } @@ -579,6 +632,8 @@ namespace { void visit(::HIR::ExprNode_Match& node) override { TRACE_FUNCTION_FR("_Match", "_Match"); + auto _ = save_and_edit(m_borrow_raise_target, nullptr); + //auto stmt_scope = m_builder.new_scope_temp(node.span()); this->visit_node_ptr(node.m_value); auto match_val = m_builder.get_result_in_lvalue(node.m_value->span(), node.m_value->m_res_type); @@ -608,7 +663,7 @@ namespace { if( m_builder.block_active() ) { auto res = m_builder.get_result(arm.m_code->span()); - m_builder.raise_variables( arm.m_code->span(), res, scope ); + m_builder.raise_variables( arm.m_code->span(), res, scope, /*to_above=*/true); m_builder.set_result(arm.m_code->span(), mv$(res)); m_builder.terminate_scope( node.span(), mv$(tmp_scope) ); @@ -622,6 +677,19 @@ namespace { else { MIR_LowerHIR_Match(m_builder, *this, node, mv$(match_val)); } + + if( m_builder.block_active() ) { + const auto& sp = node.span(); + + auto res = m_builder.get_result(sp); + //m_builder.raise_variables(sp, res, stmt_scope, /*to_above=*/true); + m_builder.set_result(sp, mv$(res)); + + //m_builder.terminate_scope( node.span(), mv$(stmt_scope) ); + } + else { + //m_builder.terminate_scope( node.span(), mv$(stmt_scope), false ); + } } // ExprNode_Match void emit_if(/*const*/ ::HIR::ExprNodeP& cond, ::MIR::BasicBlockId true_branch, ::MIR::BasicBlockId false_branch) @@ -691,7 +759,7 @@ namespace { auto scope = m_builder.new_scope_temp( cond->span() ); this->visit_node_ptr(*cond_p); ASSERT_BUG(cond->span(), cond->m_res_type == ::HIR::CoreType::Bool, "If condition wasn't a bool"); - decision_val = m_builder.get_result_in_lvalue(cond->span(), ::HIR::CoreType::Bool); + decision_val = m_builder.get_result_in_if_cond(cond->span()); m_builder.terminate_scope(cond->span(), mv$(scope)); } @@ -760,7 +828,7 @@ namespace { m_builder.set_result( node.span(), mv$(result_val) ); } - void generate_checked_binop(const Span& sp, ::MIR::LValue res_slot, ::MIR::eBinOp op, ::MIR::LValue val_l, const ::HIR::TypeRef& ty_l, ::MIR::LValue val_r, const ::HIR::TypeRef& ty_r) + void generate_checked_binop(const Span& sp, ::MIR::LValue res_slot, ::MIR::eBinOp op, ::MIR::Param val_l, const ::HIR::TypeRef& ty_l, ::MIR::Param val_r, const ::HIR::TypeRef& ty_r) { switch(op) { @@ -870,7 +938,16 @@ namespace { if( node.m_op != ::HIR::ExprNode_Assign::Op::None ) { auto dst_clone = dst.clone(); - auto val_lv = m_builder.lvalue_or_temp( node.span(), ty_val, mv$(val) ); + ::MIR::Param val_p; + if( auto* e = val.opt_Use() ) { + val_p = mv$(*e); + } + else if( auto* e = val.opt_Constant() ) { + val_p = mv$(*e); + } + else { + val_p = m_builder.lvalue_or_temp( node.span(), ty_val, mv$(val) ); + } ASSERT_BUG(sp, ty_slot.m_data.is_Primitive(), "Assignment operator overloads are only valid on primitives - ty_slot="<<ty_slot); ASSERT_BUG(sp, ty_val.m_data.is_Primitive(), "Assignment operator overloads are only valid on primitives - ty_val="<<ty_val); @@ -885,16 +962,16 @@ namespace { case _(Mul): op = ::MIR::eBinOp::MUL; if(0) case _(Div): op = ::MIR::eBinOp::DIV; if(0) case _(Mod): op = ::MIR::eBinOp::MOD; - this->generate_checked_binop(sp, mv$(dst), op, mv$(dst_clone), ty_slot, mv$(val_lv), ty_val); + this->generate_checked_binop(sp, mv$(dst), op, mv$(dst_clone), ty_slot, mv$(val_p), ty_val); break; case _(Xor): op = ::MIR::eBinOp::BIT_XOR; if(0) case _(Or ): op = ::MIR::eBinOp::BIT_OR ; if(0) case _(And): op = ::MIR::eBinOp::BIT_AND; - this->generate_checked_binop(sp, mv$(dst), op, mv$(dst_clone), ty_slot, mv$(val_lv), ty_val); + this->generate_checked_binop(sp, mv$(dst), op, mv$(dst_clone), ty_slot, mv$(val_p), ty_val); break; case _(Shl): op = ::MIR::eBinOp::BIT_SHL; if(0) case _(Shr): op = ::MIR::eBinOp::BIT_SHR; - this->generate_checked_binop(sp, mv$(dst), op, mv$(dst_clone), ty_slot, mv$(val_lv), ty_val); + this->generate_checked_binop(sp, mv$(dst), op, mv$(dst_clone), ty_slot, mv$(val_p), ty_val); break; } #undef _ @@ -916,12 +993,12 @@ namespace { const auto& ty_r = node.m_right->m_res_type; auto res = m_builder.new_temporary(node.m_res_type); - this->visit_node_ptr(node.m_left); - auto left = m_builder.get_result_in_lvalue(node.m_left->span(), ty_l); - // Short-circuiting boolean operations if( node.m_op == ::HIR::ExprNode_BinOp::Op::BoolAnd || node.m_op == ::HIR::ExprNode_BinOp::Op::BoolOr ) { + this->visit_node_ptr(node.m_left); + auto left = m_builder.get_result_in_lvalue(node.m_left->span(), ty_l); + auto bb_next = m_builder.new_bb_unlinked(); auto bb_true = m_builder.new_bb_unlinked(); auto bb_false = m_builder.new_bb_unlinked(); @@ -963,8 +1040,10 @@ namespace { { } + this->visit_node_ptr(node.m_left); + auto left = m_builder.get_result_in_param(node.m_left->span(), ty_l); this->visit_node_ptr(node.m_right); - auto right = m_builder.get_result_in_lvalue(node.m_right->span(), ty_r); + auto right = m_builder.get_result_in_param(node.m_right->span(), ty_r); ::MIR::eBinOp op; switch(node.m_op) @@ -1069,13 +1148,19 @@ namespace { { TRACE_FUNCTION_F("_Borrow"); + auto _ = save_and_edit(m_in_borrow, true); + const auto& ty_val = node.m_value->m_res_type; this->visit_node_ptr(node.m_value); auto val = m_builder.get_result_in_lvalue(node.m_value->span(), ty_val); - auto res = m_builder.new_temporary(node.m_res_type); - m_builder.push_stmt_assign( node.span(), res.as_Temporary(), ::MIR::RValue::make_Borrow({ 0, node.m_type, mv$(val) })); - m_builder.set_result( node.span(), mv$(res) ); + if( m_borrow_raise_target ) + { + DEBUG("- Raising borrow to scope " << *m_borrow_raise_target); + m_builder.raise_variables(node.span(), val, *m_borrow_raise_target); + } + + m_builder.set_result( node.span(), ::MIR::RValue::make_Borrow({ 0, node.m_type, mv$(val) }) ); } void visit(::HIR::ExprNode_Cast& node) override { @@ -1233,8 +1318,9 @@ namespace { } else { - // Probably an error. - TODO(node.span(), "MIR _Unsize to " << ty_out); + // Probably an error? + m_builder.set_result( node.span(), ::MIR::RValue::make_Cast({ mv$(ptr_lval), node.m_res_type.clone() }) ); + //TODO(node.span(), "MIR _Unsize to " << ty_out); } ), (Slice, @@ -1345,10 +1431,63 @@ namespace { if( m_builder.is_type_owned_box( ty_val ) ) { // Box magically derefs. - // HACK: Break out of the switch used for TU_MATCH_DEF - break; } - BUG(sp, "Deref on unsupported type - " << ty_val); + else + { + // TODO: Do operator replacement here after handling scope-raising for _Borrow + if( m_borrow_raise_target && m_in_borrow ) + { + DEBUG("- Raising deref in borrow to scope " << *m_borrow_raise_target); + m_builder.raise_variables(node.span(), val, *m_borrow_raise_target); + } + + + const char* langitem = nullptr; + const char* method = nullptr; + ::HIR::BorrowType bt; + // - Uses the value's usage beacuse for T: Copy node.m_value->m_usage is Borrow, but node.m_usage is Move + switch( node.m_value->m_usage ) + { + case ::HIR::ValueUsage::Unknown: + BUG(sp, "Unknown usage type of deref value"); + break; + case ::HIR::ValueUsage::Borrow: + bt = ::HIR::BorrowType::Shared; + langitem = method = "deref"; + break; + case ::HIR::ValueUsage::Mutate: + bt = ::HIR::BorrowType::Unique; + langitem = method = "deref_mut"; + break; + case ::HIR::ValueUsage::Move: + TODO(sp, "ValueUsage::Move for desugared Deref of " << node.m_value->m_res_type); + break; + } + // Needs replacement, continue + assert(langitem); + assert(method); + + // - Construct trait path - Index*<IdxTy> + auto method_path = ::HIR::Path(ty_val.clone(), ::HIR::GenericPath(m_builder.resolve().m_crate.get_lang_item_path(node.span(), langitem), {}), method); + + // Store a borrow of the input value + ::std::vector<::MIR::Param> args; + args.push_back( m_builder.lvalue_or_temp(sp, + ::HIR::TypeRef::new_borrow(bt, node.m_value->m_res_type.clone()), + ::MIR::RValue::make_Borrow({0, bt, mv$(val)}) + ) ); + m_builder.moved_lvalue(node.span(), args[0].as_LValue()); + val = m_builder.new_temporary(::HIR::TypeRef::new_borrow(bt, node.m_res_type.clone())); + // Call the above trait method + // Store result of that call in `val` (which will be derefed below) + auto ok_block = m_builder.new_bb_unlinked(); + auto panic_block = m_builder.new_bb_unlinked(); + m_builder.end_block(::MIR::Terminator::make_Call({ ok_block, panic_block, val.clone(), mv$(method_path), mv$(args) })); + m_builder.set_cur_block(panic_block); + m_builder.end_block(::MIR::Terminator::make_Diverge({})); + + m_builder.set_cur_block(ok_block); + } ), (Pointer, // Deref on a pointer - TODO: Requires unsafe @@ -1413,6 +1552,7 @@ namespace { // TODO: Proper panic handling, including scope destruction m_builder.set_cur_block(place__panic); + //m_builder.terminate_scope_early( node.span(), m_builder.fcn_scope() ); // TODO: Drop `place` m_builder.end_block( ::MIR::Terminator::make_Diverge({}) ); m_builder.set_cur_block(place__ok); @@ -1437,6 +1577,7 @@ namespace { // TODO: Proper panic handling, including scope destruction m_builder.set_cur_block(place_raw__panic); + //m_builder.terminate_scope_early( node.span(), m_builder.fcn_scope() ); // TODO: Drop `place` m_builder.end_block( ::MIR::Terminator::make_Diverge({}) ); m_builder.set_cur_block(place_raw__ok); @@ -1473,11 +1614,13 @@ namespace { // TODO: Proper panic handling, including scope destruction m_builder.set_cur_block(res__panic); + //m_builder.terminate_scope_early( node.span(), m_builder.fcn_scope() ); // TODO: Should this drop the value written to the rawptr? // - No, becuase it's likely invalid now. Goodbye! m_builder.end_block( ::MIR::Terminator::make_Diverge({}) ); m_builder.set_cur_block(res__ok); + m_builder.mark_value_assigned(node.span(), res); m_builder.set_result( node.span(), mv$(res) ); } @@ -1521,20 +1664,29 @@ namespace { for(auto& arg : args) { this->visit_node_ptr(arg); - - if( args.size() == 1 ) + if( !m_builder.block_active() ) + { + auto tmp = m_builder.new_temporary(arg->m_res_type); + values.push_back( mv$(tmp) ); + } + else if( args.size() == 1 ) { values.push_back( m_builder.get_result_in_param(arg->span(), arg->m_res_type, /*allow_missing_value=*/true) ); } else { - // NOTE: Have to allocate a new temporary because ordering matters - auto tmp = m_builder.new_temporary(arg->m_res_type); - if( m_builder.block_active() ) + auto res = m_builder.get_result(arg->span()); + if( auto* e = res.opt_Constant() ) { - m_builder.push_stmt_assign( arg->span(), tmp.clone(), m_builder.get_result(arg->span()) ); + values.push_back( mv$(*e) ); + } + else + { + // NOTE: Have to allocate a new temporary because ordering matters + auto tmp = m_builder.new_temporary(arg->m_res_type); + m_builder.push_stmt_assign( arg->span(), tmp.clone(), mv$(res) ); + values.push_back( mv$(tmp) ); } - values.push_back( mv$(tmp) ); } if(const auto* e = values.back().opt_LValue() ) @@ -1548,6 +1700,7 @@ namespace { void visit(::HIR::ExprNode_CallPath& node) override { TRACE_FUNCTION_F("_CallPath " << node.m_path); + auto _ = save_and_edit(m_borrow_raise_target, nullptr); auto values = get_args(node.m_args); auto panic_block = m_builder.new_bb_unlinked(); @@ -1615,6 +1768,7 @@ namespace { void visit(::HIR::ExprNode_CallValue& node) override { TRACE_FUNCTION_F("_CallValue " << node.m_value->m_res_type); + auto _ = save_and_edit(m_borrow_raise_target, nullptr); // _CallValue is ONLY valid on function pointers (all others must be desugared) ASSERT_BUG(node.span(), node.m_value->m_res_type.m_data.is_Function(), "Leftover _CallValue on a non-fn()"); @@ -1638,6 +1792,8 @@ namespace { m_builder.end_block( ::MIR::Terminator::make_Diverge({}) ); m_builder.set_cur_block( next_block ); + // TODO: Support diverging value calls + m_builder.mark_value_assigned(node.span(), res); m_builder.set_result( node.span(), mv$(res) ); } void visit(::HIR::ExprNode_CallMethod& node) override @@ -1897,6 +2053,7 @@ namespace { ::MIR::LValue base_val; if( node.m_base_value ) { + DEBUG("_StructLiteral - base"); this->visit_node_ptr(node.m_base_value); base_val = m_builder.get_result_in_lvalue(node.m_base_value->span(), node.m_base_value->m_res_type); } @@ -1935,6 +2092,7 @@ namespace { auto idx = ::std::find_if(fields.begin(), fields.end(), [&](const auto&x){ return x.first == ent.first; }) - fields.begin(); assert( !values_set[idx] ); values_set[idx] = true; + DEBUG("_StructLiteral - fld '" << ent.first << "' (idx " << idx << ")"); this->visit_node_ptr(valnode); auto res = m_builder.get_result(valnode->span()); @@ -2035,6 +2193,7 @@ namespace { void visit(::HIR::ExprNode_Closure& node) override { TRACE_FUNCTION_F("_Closure - " << node.m_obj_path); + auto _ = save_and_edit(m_borrow_raise_target, nullptr); ::std::vector< ::MIR::Param> vals; vals.reserve( node.m_captures.size() ); @@ -2082,6 +2241,8 @@ namespace { root_node.visit( ev ); } + MIR_Validate(resolve, path, fcn, args, ptr->m_res_type); + return ::MIR::FunctionPointer(new ::MIR::Function(mv$(fcn))); } diff --git a/src/mir/from_hir.hpp b/src/mir/from_hir.hpp index 1b57a7f4..6b10d5bd 100644 --- a/src/mir/from_hir.hpp +++ b/src/mir/from_hir.hpp @@ -36,6 +36,11 @@ public: ScopeHandle& operator=(const ScopeHandle& x) = delete; ScopeHandle& operator=(ScopeHandle&& x) = delete; ~ScopeHandle(); + + friend ::std::ostream& operator<<(::std::ostream& os, const ScopeHandle& x) { + os << x.idx; + return os; + } }; // - Needs to handle future DerefMove (which can't use the Box hack) @@ -48,11 +53,15 @@ enum class InvalidType { TAGGED_UNION_EX(VarState, (), Invalid, ( // Currently invalid (Invalid, InvalidType), - // Partially valid (Map of field states, Box is assumed to have one field) + // Partially valid (Map of field states) (Partial, struct { ::std::vector<VarState> inner_states; unsigned int outer_flag; // If ~0u there's no condition on the outer }), + (MovedOut, struct { + ::std::unique_ptr<VarState> inner_state; + unsigned int outer_flag; + }), // Optionally valid (integer indicates the drop flag index) (Optional, unsigned int), // Fully valid @@ -99,6 +108,14 @@ TAGGED_UNION(ScopeType, Variables, }) ); +enum class VarGroup +{ + Return, + Argument, + Variable, + Temporary, +}; + /// Helper class to construct MIR class MirBuilder { @@ -118,7 +135,8 @@ class MirBuilder bool m_result_valid; // TODO: Extra information. - //::std::vector<VarState> m_arg_states; + VarState m_return_state; + ::std::vector<VarState> m_arg_states; ::std::vector<VarState> m_variable_states; ::std::vector<VarState> m_temporary_states; @@ -142,6 +160,11 @@ class MirBuilder ::std::vector<ScopeDef> m_scopes; ::std::vector<unsigned int> m_scope_stack; ScopeHandle m_fcn_scope; + + // LValue used only for the condition of `if` + // - Using a fixed temporary simplifies parts of lowering (scope related) and reduces load on + // the optimiser. + ::MIR::LValue m_if_cond_lval; public: MirBuilder(const Span& sp, const StaticTraitResolve& resolve, const ::HIR::Function::args_t& args, ::MIR::Function& output); ~MirBuilder(); @@ -169,6 +192,17 @@ public: /// Obtains a result in a param (or a lvalue) ::MIR::Param get_result_in_param(const Span& sp, const ::HIR::TypeRef& ty, bool allow_missing_value=false); + ::MIR::LValue get_if_cond() const { + return m_if_cond_lval.clone(); + } + ::MIR::LValue get_rval_in_if_cond(const Span& sp, ::MIR::RValue val) { + push_stmt_assign(sp, m_if_cond_lval.clone(), mv$(val)); + return m_if_cond_lval.clone(); + } + ::MIR::LValue get_result_in_if_cond(const Span& sp) { + return get_rval_in_if_cond(sp, get_result(sp)); + } + // - Statements // Push an assignment. NOTE: This also marks the rvalue as moved void push_stmt_assign(const Span& sp, ::MIR::LValue dst, ::MIR::RValue val); @@ -181,6 +215,9 @@ public: // Push a setting/clearing of a drop flag void push_stmt_set_dropflag_val(const Span& sp, unsigned int index, bool value); void push_stmt_set_dropflag_other(const Span& sp, unsigned int index, unsigned int other); + void push_stmt_set_dropflag_default(const Span& sp, unsigned int index); + + void push_stmt(const Span& sp, ::MIR::Statement stmt); // - Block management bool block_active() const { @@ -190,12 +227,13 @@ public: // Mark a value as initialised (used for Call, because it has to be done after the panic block is populated) void mark_value_assigned(const Span& sp, const ::MIR::LValue& val); - // Moves control of temporaries up to the next scope - void raise_variables(const Span& sp, const ::MIR::LValue& val, const ScopeHandle& scope); - void raise_variables(const Span& sp, const ::MIR::RValue& rval, const ScopeHandle& scope); + // Moves control of temporaries up to the specified scope (or to above it) + void raise_variables(const Span& sp, const ::MIR::LValue& val, const ScopeHandle& scope, bool to_above=false); + void raise_variables(const Span& sp, const ::MIR::RValue& rval, const ScopeHandle& scope, bool to_above=false); void set_cur_block(unsigned int new_block); ::MIR::BasicBlockId pause_cur_block(); + void end_block(::MIR::Terminator term); ::MIR::BasicBlockId new_bb_linked(); @@ -210,9 +248,16 @@ public: ScopeHandle new_scope_temp(const Span& sp); ScopeHandle new_scope_split(const Span& sp); ScopeHandle new_scope_loop(const Span& sp); + + /// Raises every variable defined in the source scope into the target scope + void raise_all(const Span& sp, ScopeHandle src, const ScopeHandle& target); + /// Drop all defined values in the scope (emits the drops if `cleanup` is set) void terminate_scope(const Span& sp, ScopeHandle , bool cleanup=true); + /// Terminates a scope early (e.g. via return/break/...) void terminate_scope_early(const Span& sp, const ScopeHandle& , bool loop_exit=false); + /// Marks the end of a split arm (end match arm, if body, ...) void end_split_arm(const Span& sp, const ScopeHandle& , bool reachable); + /// Terminates the current split early (TODO: What does this mean?) void end_split_arm_early(const Span& sp); const ScopeHandle& fcn_scope() const { @@ -224,11 +269,17 @@ public: // Helper - Marks a variable/... as moved (and checks if the move is valid) void moved_lvalue(const Span& sp, const ::MIR::LValue& lv); private: + const VarState& get_slot_state(const Span& sp, VarGroup ty, unsigned int idx, unsigned int skip_count=0) const; + VarState& get_slot_state_mut(const Span& sp, VarGroup ty, unsigned int idx); + const VarState& get_variable_state(const Span& sp, unsigned int idx, unsigned int skip_count=0) const; VarState& get_variable_state_mut(const Span& sp, unsigned int idx); const VarState& get_temp_state(const Span& sp, unsigned int idx, unsigned int skip_count=0) const; VarState& get_temp_state_mut(const Span& sp, unsigned int idx); + const VarState& get_val_state(const Span& sp, const ::MIR::LValue& lv, unsigned int skip_count=0); + VarState& get_val_state_mut(const Span& sp, const ::MIR::LValue& lv); + void terminate_loop_early(const Span& sp, ScopeType::Data_Loop& sd_loop); void drop_value_from_state(const Span& sp, const VarState& vs, ::MIR::LValue lv); diff --git a/src/mir/from_hir_match.cpp b/src/mir/from_hir_match.cpp index aa9825cd..46331ca6 100644 --- a/src/mir/from_hir_match.cpp +++ b/src/mir/from_hir_match.cpp @@ -33,24 +33,35 @@ struct field_path_t }; TAGGED_UNION_EX(PatternRule, (), Any,( - // _ pattern - (Any, struct {}), // Enum variant (Variant, struct { unsigned int idx; ::std::vector<PatternRule> sub_rules; }), // Slice (includes desired length) (Slice, struct { unsigned int len; ::std::vector<PatternRule> sub_rules; }), // SplitSlice // TODO: How can the negative offsets in the `trailing` be handled correctly? (both here and in the destructure) - (SplitSlice, struct { unsigned int min_len; ::std::vector<PatternRule> leading, trailing; }), + (SplitSlice, struct { unsigned int min_len; unsigned int trailing_len; ::std::vector<PatternRule> leading, trailing; }), // Boolean (different to Constant because of how restricted it is) (Bool, bool), // General value (Value, ::MIR::Constant), - (ValueRange, struct { ::MIR::Constant first, last; }) + (ValueRange, struct { ::MIR::Constant first, last; }), + // _ pattern + (Any, struct {}) ), ( , field_path(mv$(x.field_path)) ), (field_path = mv$(x.field_path);), ( field_path_t field_path; + + bool operator<(const PatternRule& x) const { + return this->ord(x) == OrdLess; + } + bool operator==(const PatternRule& x) const { + return this->ord(x) == OrdEqual; + } + bool operator!=(const PatternRule& x) const { + return this->ord(x) != OrdEqual; + } + Ordering ord(const PatternRule& x) const; ) ); ::std::ostream& operator<<(::std::ostream& os, const PatternRule& x); @@ -74,11 +85,14 @@ struct ArmCode { ::MIR::BasicBlockId cond_end; ::MIR::LValue cond_lval; ::std::vector< ::MIR::BasicBlockId> destructures; // NOTE: Incomplete + + mutable ::MIR::BasicBlockId cond_fail_tgt = 0; }; typedef ::std::vector<PatternRuleset> t_arm_rules; void MIR_LowerHIR_Match_Simple( MirBuilder& builder, MirConverter& conv, ::HIR::ExprNode_Match& node, ::MIR::LValue match_val, t_arm_rules arm_rules, ::std::vector<ArmCode> arm_code, ::MIR::BasicBlockId first_cmp_block); +void MIR_LowerHIR_Match_Grouped( MirBuilder& builder, MirConverter& conv, ::HIR::ExprNode_Match& node, ::MIR::LValue match_val, t_arm_rules arm_rules, ::std::vector<ArmCode> arms_code, ::MIR::BasicBlockId first_cmp_block ); void MIR_LowerHIR_Match_DecisionTree( MirBuilder& builder, MirConverter& conv, ::HIR::ExprNode_Match& node, ::MIR::LValue match_val, t_arm_rules arm_rules, ::std::vector<ArmCode> arm_code , ::MIR::BasicBlockId first_cmp_block); /// Helper to construct rules from a passed pattern struct PatternRulesetBuilder @@ -103,6 +117,82 @@ struct PatternRulesetBuilder void push_rule(PatternRule r); }; +class RulesetRef +{ + ::std::vector<PatternRuleset>* m_rules_vec = nullptr; + RulesetRef* m_parent = nullptr; + size_t m_parent_ofs=0; // If len == 0, this is the innner index, else it's the base + size_t m_parent_len=0; +public: + RulesetRef(::std::vector<PatternRuleset>& rules): + m_rules_vec(&rules) + { + } + RulesetRef(RulesetRef& parent, size_t start, size_t n): + m_parent(&parent), + m_parent_ofs(start), + m_parent_len(n) + { + } + RulesetRef(RulesetRef& parent, size_t idx): + m_parent(&parent), + m_parent_ofs(idx) + { + } + + size_t size() const { + if( m_rules_vec ) { + return m_rules_vec->size(); + } + else if( m_parent_len ) { + return m_parent_len; + } + else { + return m_parent->size(); + } + } + RulesetRef slice(size_t s, size_t n) { + return RulesetRef(*this, s, n); + } + + const ::std::vector<PatternRule>& operator[](size_t i) const { + if( m_rules_vec ) { + return (*m_rules_vec)[i].m_rules; + } + else if( m_parent_len ) { + return (*m_parent)[m_parent_ofs + i]; + } + else { + // Fun part - Indexes into inner patterns + const auto& parent_rule = (*m_parent)[i][m_parent_ofs]; + if(const auto* re = parent_rule.opt_Variant()) { + return re->sub_rules; + } + else { + throw "TODO"; + } + } + } + void swap(size_t a, size_t b) { + TRACE_FUNCTION_F(a << ", " << b); + if( m_rules_vec ) { + ::std::swap( (*m_rules_vec)[a], (*m_rules_vec)[b] ); + } + else { + assert(m_parent); + if( m_parent_len ) { + m_parent->swap(m_parent_ofs + a, m_parent_ofs + b); + } + else { + m_parent->swap(a, b); + } + } + } +}; + +void sort_rulesets(RulesetRef rulesets, size_t idx=0); +void sort_rulesets_inner(RulesetRef rulesets, size_t idx); + // -------------------------------------------------------------------- // CODE // -------------------------------------------------------------------- @@ -282,8 +372,7 @@ void MIR_LowerHIR_Match( MirBuilder& builder, MirConverter& conv, ::HIR::ExprNod auto tmp_scope = builder.new_scope_temp(arm.m_cond->span()); conv.visit_node_ptr( arm.m_cond ); - ac.cond_lval = builder.get_result_in_lvalue(arm.m_cond->span(), ::HIR::TypeRef(::HIR::CoreType::Bool)); - // NOTE: Terminating the scope slightly early is safe, because the resulting boolean temp isn't invalidated. + ac.cond_lval = builder.get_result_in_if_cond(arm.m_cond->span()); builder.terminate_scope( arm.m_code->span(), mv$(tmp_scope) ); ac.cond_end = builder.pause_cur_block(); @@ -366,20 +455,55 @@ void MIR_LowerHIR_Match( MirBuilder& builder, MirConverter& conv, ::HIR::ExprNod for(const auto& arm_rule : arm_rules) { - DEBUG("> (" << arm_rule.arm_idx << ", " << arm_rule.pat_idx << ") - " << arm_rule.m_rules); + DEBUG("> (" << arm_rule.arm_idx << ", " << arm_rule.pat_idx << ") - " << arm_rule.m_rules + << (arm_code[arm_rule.arm_idx].has_condition ? " (cond)" : "")); } - // TODO: Don't generate inner code until decisions are generated (keeps MIR flow nice) + // TODO: Remove columns that are all `_`? + // - Ideally, only accessible structures would be fully destructured like this, making this check redundant + + // Sort rules using the following restrictions: + // - A rule cannot be reordered across an item that has an overlapping match set + // > e.g. nothing can cross _ + // > equal rules cannot be reordered + // > Values cannot cross ranges that contain the value + // > This will have to be a bubble sort to ensure that it's correctly stable. + sort_rulesets(arm_rules); + DEBUG("Post-sort"); + for(const auto& arm_rule : arm_rules) + { + DEBUG("> (" << arm_rule.arm_idx << ", " << arm_rule.pat_idx << ") - " << arm_rule.m_rules + << (arm_code[arm_rule.arm_idx].has_condition ? " (cond)" : "")); + } + // De-duplicate arms (emitting a warning when it happens) + // - This allows later code to assume that duplicate arms are a codegen bug. + if( ! arm_rules.empty() ) + { + for(auto it = arm_rules.begin()+1; it != arm_rules.end(); ) + { + // If duplicate rule, (and neither is conditional) + if( (it-1)->m_rules == it->m_rules && !arm_code[it->arm_idx].has_condition && !arm_code[(it-1)->arm_idx].has_condition ) + { + // Remove + it = arm_rules.erase(it); + WARNING(node.m_arms[it->arm_idx].m_code->span(), W0000, "Duplicate match pattern, unreachable code"); + } + else + { + ++ it; + } + } + } - // TODO: Detect if a rule is ordering-dependent. In this case we currently have to fall back on the simple match code - // - A way would be to search for `_` rules with non _ rules following. Would false-positive in some cases, but shouldn't false negative - // TODO: Merge equal rulesets if there's one with no condition. + // TODO: Don't generate inner code until decisions are generated (keeps MIR flow nice) + // - Challenging, as the decision code needs somewhere to jump to. + // - Allocating a BB and then rewriting references to it is a possibility. if( fall_back_on_simple ) { MIR_LowerHIR_Match_Simple( builder, conv, node, mv$(match_val), mv$(arm_rules), mv$(arm_code), first_cmp_block ); } else { - MIR_LowerHIR_Match_DecisionTree( builder, conv, node, mv$(match_val), mv$(arm_rules), mv$(arm_code), first_cmp_block ); + MIR_LowerHIR_Match_Grouped( builder, conv, node, mv$(match_val), mv$(arm_rules), mv$(arm_code), first_cmp_block ); } builder.set_cur_block( next_block ); @@ -424,6 +548,56 @@ void MIR_LowerHIR_Match( MirBuilder& builder, MirConverter& conv, ::HIR::ExprNod return os; } +::Ordering PatternRule::ord(const PatternRule& x) const +{ + if(tag() != x.tag()) + { + return tag() < x.tag() ? ::OrdLess : ::OrdGreater; + } + TU_MATCHA( (*this, x), (te, xe), + (Any, return OrdEqual;), + (Variant, + if(te.idx != xe.idx) return ::ord(te.idx, xe.idx); + assert( te.sub_rules.size() == xe.sub_rules.size() ); + for(unsigned int i = 0; i < te.sub_rules.size(); i ++) + { + auto cmp = te.sub_rules[i].ord( xe.sub_rules[i] ); + if( cmp != ::OrdEqual ) + return cmp; + } + return ::OrdEqual; + ), + (Slice, + if(te.len != xe.len) return ::ord(te.len, xe.len); + // Wait? Why would the rule count be the same? + assert( te.sub_rules.size() == xe.sub_rules.size() ); + for(unsigned int i = 0; i < te.sub_rules.size(); i ++) + { + auto cmp = te.sub_rules[i].ord( xe.sub_rules[i] ); + if( cmp != ::OrdEqual ) + return cmp; + } + return ::OrdEqual; + ), + (SplitSlice, + auto rv = ::ord( te.leading, xe.leading ); + if(rv != OrdEqual) return rv; + return ::ord(te.trailing, xe.trailing); + ), + (Bool, + return ::ord( te, xe ); + ), + (Value, + return ::ord( te, xe ); + ), + (ValueRange, + if( te.first != xe.first ) + return ::ord(te.first, xe.first); + return ::ord(te.last, xe.last); + ) + ) + throw ""; +} ::Ordering PatternRuleset::rule_is_before(const PatternRule& l, const PatternRule& r) { if( l.tag() != r.tag() ) { @@ -635,6 +809,7 @@ void PatternRulesetBuilder::append_from_lit(const Span& sp, const ::HIR::Literal PatternRulesetBuilder sub_builder { this->m_resolve }; sub_builder.m_field_path = m_field_path; + sub_builder.m_field_path.push_back(var_idx); sub_builder.m_field_path.push_back(0); TU_MATCH( ::HIR::Enum::Variant, (var_def.second), (fields_def), @@ -1074,6 +1249,7 @@ void PatternRulesetBuilder::append_from(const Span& sp, const ::HIR::Pattern& pa const auto& fields_def = var_def.second.as_Tuple(); PatternRulesetBuilder sub_builder { this->m_resolve }; sub_builder.m_field_path = m_field_path; + sub_builder.m_field_path.push_back(pe.binding_idx); sub_builder.m_field_path.push_back(0); for( unsigned int i = 0; i < pe.sub_patterns.size(); i ++ ) { @@ -1107,6 +1283,7 @@ void PatternRulesetBuilder::append_from(const Span& sp, const ::HIR::Pattern& pa // 2. Iterate this list and recurse on the patterns PatternRulesetBuilder sub_builder { this->m_resolve }; sub_builder.m_field_path = m_field_path; + sub_builder.m_field_path.push_back(pe.binding_idx); sub_builder.m_field_path.push_back(0); for( unsigned int i = 0; i < tmp.size(); i ++ ) { @@ -1212,12 +1389,14 @@ void PatternRulesetBuilder::append_from(const Span& sp, const ::HIR::Pattern& pa sub_builder.m_field_path.back() = 0; if( pe.trailing.size() ) { + // Needs a way of encoding the negative offset in the field path TODO(sp, "SplitSlice on [T] with trailing - " << pat); } auto trailing = mv$(sub_builder.m_rules); this->push_rule( PatternRule::make_SplitSlice({ static_cast<unsigned int>(pe.leading.size() + pe.trailing.size()), + static_cast<unsigned int>(pe.trailing.size()), mv$(leading), mv$(trailing) }) ); ) @@ -1281,6 +1460,227 @@ void PatternRulesetBuilder::append_from(const Span& sp, const ::HIR::Pattern& pa } namespace { + // Order rules ignoring inner rules + Ordering ord_rule_compatible(const PatternRule& a, const PatternRule& b) + { + if(a.tag() != b.tag()) + return ::ord( (unsigned)a.tag(), b.tag() ); + + TU_MATCHA( (a, b), (ae, be), + (Any, + return OrdEqual; + ), + (Variant, + return ::ord(ae.idx, be.idx); + ), + (Slice, + return ::ord(ae.len, be.len); + ), + (SplitSlice, + auto v = ::ord(ae.leading.size(), be.leading.size()); + if(v != OrdEqual) return v; + v = ::ord(ae.trailing.size(), be.trailing.size()); + if(v != OrdEqual) return v; + return OrdEqual; + ), + (Bool, + return ::ord(ae, be); + ), + (Value, + return ::ord(ae, be); + ), + (ValueRange, + auto v = ::ord(ae.first, be.first); + if(v != OrdEqual) return v; + return ::ord(ae.last, be.last); + ) + ) + throw ""; + } + bool rule_compatible(const PatternRule& a, const PatternRule& b) + { + return ord_rule_compatible(a,b) == OrdEqual; + } + + bool rules_overlap(const PatternRule& a, const PatternRule& b) + { + if( a.is_Any() || b.is_Any() ) + return true; + + // Defensive: If a constant is encountered, assume it overlaps with anything + if(const auto* ae = a.opt_Value()) { + if(ae->is_Const()) + return true; + } + if(const auto* be = b.opt_Value()) { + if(be->is_Const()) + return true; + } + + // Value Range: Overlaps with contained values. + if(const auto* ae = a.opt_ValueRange() ) + { + if(const auto* be = b.opt_Value() ) + { + return ( ae->first <= *be && *be <= ae->last ); + } + else if( const auto* be = b.opt_ValueRange() ) + { + // Start of B within A + if( ae->first <= be->first && be->first <= ae->last ) + return true; + // End of B within A + if( ae->first <= be->last && be->last <= ae->last ) + return true; + // Start of A within B + if( be->first <= ae->first && ae->first <= be->last ) + return true; + // End of A within B + if( be->first <= ae->last && ae->last <= be->last ) + return true; + + // Disjoint + return false; + } + else + { + TODO(Span(), "Check overlap of " << a << " and " << b); + } + } + if(const auto* be = b.opt_ValueRange()) + { + if(const auto* ae = a.opt_Value() ) + { + return (be->first <= *ae && *ae <= be->last); + } + // Note: A can't be ValueRange + else + { + TODO(Span(), "Check overlap of " << a << " and " << b); + } + } + + // SplitSlice patterns overlap with other SplitSlice patterns and larger slices + if(const auto* ae = a.opt_SplitSlice()) + { + if( b.is_SplitSlice() ) + { + return true; + } + else if( const auto* be = b.opt_Slice() ) + { + return be->len >= ae->min_len; + } + else + { + TODO(Span(), "Check overlap of " << a << " and " << b); + } + } + if(const auto* be = b.opt_SplitSlice()) + { + if( const auto* ae = a.opt_Slice() ) + { + return ae->len >= be->min_len; + } + else + { + TODO(Span(), "Check overlap of " << a << " and " << b); + } + } + + // Otherwise, If rules are approximately equal, they overlap + return ( ord_rule_compatible(a, b) == OrdEqual ); + } +} +void sort_rulesets(RulesetRef rulesets, size_t idx) +{ + if(rulesets.size() < 2) + return ; + + bool found_non_any = false; + for(size_t i = 0; i < rulesets.size(); i ++) + if( !rulesets[i][idx].is_Any() ) + found_non_any = true; + if( found_non_any ) + { + TRACE_FUNCTION_F(idx); + for(size_t i = 0; i < rulesets.size(); i ++) + DEBUG("- " << i << ": " << rulesets[i]); + + bool action_taken; + do + { + action_taken = false; + for(size_t i = 0; i < rulesets.size()-1; i ++) + { + if( rules_overlap(rulesets[i][idx], rulesets[i+1][idx]) ) + { + // Don't move + } + else if( ord_rule_compatible(rulesets[i][idx], rulesets[i+1][idx]) == OrdGreater ) + { + rulesets.swap(i, i+1); + action_taken = true; + } + else + { + } + } + } while(action_taken); + for(size_t i = 0; i < rulesets.size(); i ++) + DEBUG("- " << i << ": " << rulesets[i]); + + // TODO: Print sorted ruleset + + // Where compatible, sort insides + size_t start = 0; + for(size_t i = 1; i < rulesets.size(); i++) + { + if( ord_rule_compatible(rulesets[i][idx], rulesets[start][idx]) != OrdEqual ) + { + sort_rulesets_inner(rulesets.slice(start, i-start), idx); + start = i; + } + } + sort_rulesets_inner(rulesets.slice(start, rulesets.size()-start), idx); + + // Iterate onwards where rules are equal + if( idx + 1 < rulesets[0].size() ) + { + size_t start = 0; + for(size_t i = 1; i < rulesets.size(); i++) + { + if( rulesets[i][idx] != rulesets[start][idx] ) + { + sort_rulesets(rulesets.slice(start, i-start), idx+1); + start = i; + } + } + sort_rulesets(rulesets.slice(start, rulesets.size()-start), idx+1); + } + } + else + { + if( idx + 1 < rulesets[0].size() ) + { + sort_rulesets(rulesets, idx + 1); + } + } +} +void sort_rulesets_inner(RulesetRef rulesets, size_t idx) +{ + TRACE_FUNCTION_F(idx << " - " << rulesets[0][idx].tag_str()); + if( const auto* re = rulesets[0][idx].opt_Variant() ) + { + // Sort rules based on contents of enum + if( re->sub_rules.size() > 0 ) + { + sort_rulesets(RulesetRef(rulesets, idx), 0); + } + } +} + +namespace { void get_ty_and_val( const Span& sp, const StaticTraitResolve& resolve, const ::HIR::TypeRef& top_ty, const ::MIR::LValue& top_val, @@ -1296,7 +1696,7 @@ namespace { ASSERT_BUG(sp, field_path_ofs <= field_path.size(), "Field path offset " << field_path_ofs << " is larger than the path [" << field_path << "]"); for(unsigned int i = field_path_ofs; i < field_path.size(); i ++ ) { - auto idx = field_path.data[i]; + unsigned idx = field_path.data[i]; TU_MATCHA( (cur_ty->m_data), (e), (Infer, BUG(sp, "Ivar for in match type"); ), @@ -1378,7 +1778,43 @@ namespace { lval = ::MIR::LValue::make_Downcast({ box$(lval), idx }); ), (Enum, - BUG(sp, "Destructuring an enum - " << *cur_ty); + auto monomorph_to_ptr = [&](const auto& ty)->const auto* { + if( monomorphise_type_needed(ty) ) { + auto rv = monomorphise_type(sp, pbe->m_params, e.path.m_data.as_Generic().m_params, ty); + resolve.expand_associated_types(sp, rv); + tmp_ty = mv$(rv); + return &tmp_ty; + } + else { + return &ty; + } + }; + ASSERT_BUG(sp, idx < pbe->m_variants.size(), "Variant index (" << idx << ") out of range (" << pbe->m_variants.size() << ") for enum " << *cur_ty); + const auto& var = pbe->m_variants[idx]; + + i++; + assert(i < field_path.data.size()); + unsigned fld_idx = field_path.data[i]; + + TU_MATCHA( (var.second), (e), + (Unit, + BUG(sp, "Unit variant being destructured"); + ), + (Value, + BUG(sp, "Value variant being destructured"); + ), + (Tuple, + ASSERT_BUG(sp, fld_idx < e.size(), "Variant field index (" << fld_idx << ") out of range (" << e.size() << ") for enum " << *cur_ty << "::" << var.first); + cur_ty = monomorph_to_ptr(e[fld_idx].ent); + ), + (Struct, + ASSERT_BUG(sp, fld_idx < e.size(), "Variant field index (" << fld_idx << ") out of range (" << e.size() << ") for enum " << *cur_ty << "::" << var.first); + cur_ty = monomorph_to_ptr(e[fld_idx].second.ent); + ) + ) + DEBUG("*cur_ty = " << *cur_ty); + lval = ::MIR::LValue::make_Downcast({ box$(lval), idx }); + lval = ::MIR::LValue::make_Field({ box$(lval), fld_idx }); ) ) ), @@ -1402,12 +1838,15 @@ namespace { ), (Borrow, ASSERT_BUG(sp, idx == FIELD_DEREF, "Destructure of borrow doesn't correspond to a deref in the path"); + DEBUG(i << " " << *cur_ty << " - " << cur_ty << " " << &tmp_ty); if( cur_ty == &tmp_ty ) { - tmp_ty = mv$(*tmp_ty.m_data.as_Borrow().inner); + auto ip = mv$(tmp_ty.m_data.as_Borrow().inner); + tmp_ty = mv$(*ip); } else { cur_ty = &*e.inner; } + DEBUG(i << " " << *cur_ty); lval = ::MIR::LValue::make_Deref({ box$(lval) }); ), (Pointer, @@ -1437,7 +1876,6 @@ void MIR_LowerHIR_Match_Simple( MirBuilder& builder, MirConverter& conv, ::HIR:: TRACE_FUNCTION; // 1. Generate pattern matches - unsigned int rule_idx = 0; builder.set_cur_block( first_cmp_block ); for( unsigned int arm_idx = 0; arm_idx < node.m_arms.size(); arm_idx ++ ) { @@ -1451,6 +1889,10 @@ void MIR_LowerHIR_Match_Simple( MirBuilder& builder, MirConverter& conv, ::HIR:: if( arm_code.destructures[i] == 0 ) continue ; + size_t rule_idx = 0; + for(; rule_idx < arm_rules.size(); rule_idx++) + if( arm_rules[rule_idx].arm_idx == arm_idx && arm_rules[rule_idx].pat_idx == i ) + break; const auto& pat_rule = arm_rules[rule_idx]; bool is_last_pat = (i+1 == arm.m_patterns.size()); auto next_pattern_bb = (!is_last_pat ? builder.new_bb_unlinked() : next_arm_bb); @@ -1478,8 +1920,6 @@ void MIR_LowerHIR_Match_Simple( MirBuilder& builder, MirConverter& conv, ::HIR:: { builder.set_cur_block( next_pattern_bb ); } - - rule_idx ++; } if( arm_code.has_condition ) { @@ -1549,12 +1989,28 @@ int MIR_LowerHIR_Match_Simple__GeneratePattern(MirBuilder& builder, const Span& auto succ_bb = builder.new_bb_unlinked(); auto test_val = ::MIR::Param( ::MIR::Constant::make_Uint({ re.as_Uint().v, te })); - auto cmp_lval = builder.lvalue_or_temp(sp, ::HIR::CoreType::Bool, ::MIR::RValue::make_BinOp({ val.clone(), ::MIR::eBinOp::EQ, mv$(test_val) })); - builder.end_block( ::MIR::Terminator::make_If({ mv$(cmp_lval), succ_bb, fail_bb }) ); + builder.push_stmt_assign(sp, builder.get_if_cond(), ::MIR::RValue::make_BinOp({ val.clone(), ::MIR::eBinOp::EQ, mv$(test_val) })); + builder.end_block( ::MIR::Terminator::make_If({ builder.get_if_cond(), succ_bb, fail_bb }) ); builder.set_cur_block(succ_bb); ), (ValueRange, - TODO(sp, "Simple match over primitive - " << ty << " - ValueRange"); + auto succ_bb = builder.new_bb_unlinked(); + auto test_bb_2 = builder.new_bb_unlinked(); + + auto test_lt_val = ::MIR::Param(::MIR::Constant::make_Uint({ re.first.as_Uint().v, te })); + auto test_gt_val = ::MIR::Param(::MIR::Constant::make_Uint({ re.last.as_Uint().v, te })); + + // IF `val` < `first` : fail_bb + auto cmp_lt_lval = builder.lvalue_or_temp(sp, ::HIR::CoreType::Bool, ::MIR::RValue::make_BinOp({ ::MIR::Param(val.clone()), ::MIR::eBinOp::LT, mv$(test_lt_val) })); + builder.end_block( ::MIR::Terminator::make_If({ mv$(cmp_lt_lval), fail_bb, test_bb_2 }) ); + + builder.set_cur_block(test_bb_2); + + // IF `val` > `last` : fail_bb + auto cmp_gt_lval = builder.lvalue_or_temp(sp, ::HIR::CoreType::Bool, ::MIR::RValue::make_BinOp({ ::MIR::Param(val.clone()), ::MIR::eBinOp::GT, mv$(test_gt_val) })); + builder.end_block( ::MIR::Terminator::make_If({ mv$(cmp_gt_lval), fail_bb, succ_bb }) ); + + builder.set_cur_block(succ_bb); ) ) break; @@ -1577,7 +2033,23 @@ int MIR_LowerHIR_Match_Simple__GeneratePattern(MirBuilder& builder, const Span& builder.set_cur_block(succ_bb); ), (ValueRange, - TODO(sp, "Simple match over primitive - " << ty << " - ValueRange"); + auto succ_bb = builder.new_bb_unlinked(); + auto test_bb_2 = builder.new_bb_unlinked(); + + auto test_lt_val = ::MIR::Param(::MIR::Constant::make_Int({ re.first.as_Int().v, te })); + auto test_gt_val = ::MIR::Param(::MIR::Constant::make_Int({ re.last.as_Int().v, te })); + + // IF `val` < `first` : fail_bb + auto cmp_lt_lval = builder.lvalue_or_temp(sp, ::HIR::CoreType::Bool, ::MIR::RValue::make_BinOp({ ::MIR::Param(val.clone()), ::MIR::eBinOp::LT, mv$(test_lt_val) })); + builder.end_block( ::MIR::Terminator::make_If({ mv$(cmp_lt_lval), fail_bb, test_bb_2 }) ); + + builder.set_cur_block(test_bb_2); + + // IF `val` > `last` : fail_bb + auto cmp_gt_lval = builder.lvalue_or_temp(sp, ::HIR::CoreType::Bool, ::MIR::RValue::make_BinOp({ ::MIR::Param(val.clone()), ::MIR::eBinOp::GT, mv$(test_gt_val) })); + builder.end_block( ::MIR::Terminator::make_If({ mv$(cmp_gt_lval), fail_bb, succ_bb }) ); + + builder.set_cur_block(succ_bb); ) ) break; @@ -1617,7 +2089,38 @@ int MIR_LowerHIR_Match_Simple__GeneratePattern(MirBuilder& builder, const Span& break; case ::HIR::CoreType::F32: case ::HIR::CoreType::F64: - TODO(sp, "Simple match over float - " << ty); + TU_MATCH_DEF( PatternRule, (rule), (re), + ( + BUG(sp, "PatternRule for float is not Value or ValueRange"); + ), + (Value, + auto succ_bb = builder.new_bb_unlinked(); + + auto test_val = ::MIR::Param(::MIR::Constant::make_Float({ re.as_Float().v, te })); + auto cmp_lval = builder.lvalue_or_temp(sp, ::HIR::CoreType::Bool, ::MIR::RValue::make_BinOp({ val.clone(), ::MIR::eBinOp::EQ, mv$(test_val) })); + builder.end_block( ::MIR::Terminator::make_If({ mv$(cmp_lval), succ_bb, fail_bb }) ); + builder.set_cur_block(succ_bb); + ), + (ValueRange, + auto succ_bb = builder.new_bb_unlinked(); + auto test_bb_2 = builder.new_bb_unlinked(); + + auto test_lt_val = ::MIR::Param(::MIR::Constant::make_Float({ re.first.as_Float().v, te })); + auto test_gt_val = ::MIR::Param(::MIR::Constant::make_Float({ re.last.as_Float().v, te })); + + // IF `val` < `first` : fail_bb + auto cmp_lt_lval = builder.lvalue_or_temp(sp, ::HIR::CoreType::Bool, ::MIR::RValue::make_BinOp({ ::MIR::Param(val.clone()), ::MIR::eBinOp::LT, mv$(test_lt_val) })); + builder.end_block( ::MIR::Terminator::make_If({ mv$(cmp_lt_lval), fail_bb, test_bb_2 }) ); + + builder.set_cur_block(test_bb_2); + + // IF `val` > `last` : fail_bb + auto cmp_gt_lval = builder.lvalue_or_temp(sp, ::HIR::CoreType::Bool, ::MIR::RValue::make_BinOp({ ::MIR::Param(val.clone()), ::MIR::eBinOp::GT, mv$(test_gt_val) })); + builder.end_block( ::MIR::Terminator::make_If({ mv$(cmp_gt_lval), fail_bb, succ_bb }) ); + + builder.set_cur_block(succ_bb); + ) + ) break; case ::HIR::CoreType::Str: { ASSERT_BUG(sp, rule.is_Value() && rule.as_Value().is_StaticString(), ""); @@ -1658,7 +2161,11 @@ int MIR_LowerHIR_Match_Simple__GeneratePattern(MirBuilder& builder, const Span& TODO(sp, "Match over Union"); ), (Enum, - auto monomorph = [&](const auto& ty) { return monomorphise_type(sp, pbe->m_params, te.path.m_data.as_Generic().m_params, ty); }; + auto monomorph = [&](const auto& ty) { + auto rv = monomorphise_type(sp, pbe->m_params, te.path.m_data.as_Generic().m_params, ty); + builder.resolve().expand_associated_types(sp, rv); + return rv; + }; ASSERT_BUG(sp, rule.is_Variant(), "Rule for enum isn't Any or Variant"); const auto& re = rule.as_Variant(); unsigned int var_idx = re.idx; @@ -1696,7 +2203,7 @@ int MIR_LowerHIR_Match_Simple__GeneratePattern(MirBuilder& builder, const Span& // Recurse with the new ruleset MIR_LowerHIR_Match_Simple__GeneratePattern(builder, sp, re.sub_rules.data(), re.sub_rules.size(), - fake_tup, ::MIR::LValue::make_Downcast({ box$(val.clone()), var_idx }), rule.field_path.size(), + fake_tup, ::MIR::LValue::make_Downcast({ box$(val.clone()), var_idx }), rule.field_path.size()+1, fail_bb ); ), @@ -1713,7 +2220,7 @@ int MIR_LowerHIR_Match_Simple__GeneratePattern(MirBuilder& builder, const Span& // Recurse with the new ruleset MIR_LowerHIR_Match_Simple__GeneratePattern(builder, sp, re.sub_rules.data(), re.sub_rules.size(), - fake_tup, ::MIR::LValue::make_Downcast({ box$(val.clone()), var_idx }), rule.field_path.size(), + fake_tup, ::MIR::LValue::make_Downcast({ box$(val.clone()), var_idx }), rule.field_path.size()+1, fail_bb ); ) @@ -1733,7 +2240,7 @@ int MIR_LowerHIR_Match_Simple__GeneratePattern(MirBuilder& builder, const Span& ), (Array, TODO(sp, "Match directly on array?"); - /* + #if 0 unsigned int total = 0; for( unsigned int i = 0; i < te.size_val; i ++ ) { unsigned int cnt = MIR_LowerHIR_Match_Simple__GeneratePattern( @@ -1748,7 +2255,7 @@ int MIR_LowerHIR_Match_Simple__GeneratePattern(MirBuilder& builder, const Span& if( num_rules == 0 ) return total; } - */ + #endif ), (Slice, ASSERT_BUG(sp, rule.is_Slice() || rule.is_SplitSlice() || (rule.is_Value() && rule.as_Value().is_Bytes()), "Can only match slice with Bytes or Slice rules - " << rule); @@ -1833,1963 +2340,960 @@ int MIR_LowerHIR_Match_Simple__GeneratePattern(MirBuilder& builder, const Span& return 0; } -// -------------------------------------------------------------------- -// Decision Tree -// -------------------------------------------------------------------- +// -- +// Match v2 Algo - Grouped rules +// -- + -// ## Create descision tree in-memory based off the ruleset -// > Tree contains an lvalue and a set of possibilities (PatternRule) connected to another tree or to a branch index -struct DecisionTreeNode +class t_rules_subset { - TAGGED_UNION( Branch, Unset, - (Unset, struct{}), - (Subtree, ::std::unique_ptr<DecisionTreeNode>), - (Terminal, unsigned int) - ); - - template<typename T> - struct Range + ::std::vector<const ::std::vector<PatternRule>*> rule_sets; + bool is_arm_indexes; + ::std::vector<size_t> arm_idxes; +public: + t_rules_subset(size_t exp, bool is_arm_indexes): + is_arm_indexes(is_arm_indexes) { - T start; - T end; - - // `x` starts after this range ends - bool operator<(const Range<T>& x) const { - return (end < x.start); - } - // `x` falls above the end of this range - bool operator<(const T& x) const { - return (end < x); - } - - // `x` ends before this starts, or overlaps - bool operator>=(const Range<T>& x) const { - return start > x.end || ovelaps(x); - } - // `x` is before or within this range - bool operator>=(const T& x) const { - return start > x || contains(x); - } + rule_sets.reserve(exp); + arm_idxes.reserve(exp); + } - bool operator>(const Range<T>& x) const { - return (start > x.end); - } - bool operator>(const T& x) const { - return (start > x); - } + size_t size() const { + return rule_sets.size(); + } + const ::std::vector<PatternRule>& operator[](size_t n) const { + return *rule_sets[n]; + } + bool is_arm() const { return is_arm_indexes; } + ::std::pair<size_t,size_t> arm_idx(size_t n) const { + assert(is_arm_indexes); + auto v = arm_idxes.at(n); + return ::std::make_pair(v & 0xFFF, v >> 12); + } + ::MIR::BasicBlockId bb_idx(size_t n) const { + assert(!is_arm_indexes); + return arm_idxes.at(n); + } - bool operator==(const Range<T>& x) const { - return start == x.start && end == x.end; - } - bool operator!=(const Range<T>& x) const { - return start != x.start || end != x.end; - } + void sub_sort(size_t ofs, size_t start, size_t n) + { + ::std::vector<size_t> v; + for(size_t i = 0; i < n; i++) + v.push_back(start + i); + // Sort rules based on just the value (ignore inner rules) + ::std::stable_sort( v.begin(), v.end(), [&](auto a, auto b){ return ord_rule_compatible( (*rule_sets[a])[ofs], (*rule_sets[b])[ofs]) == OrdLess; } ); - bool contains(const T& x) const { - return (start <= x && x <= end); + // Reorder contents to above sorting + { + decltype(this->rule_sets) tmp; + for(auto i : v) + tmp.push_back(rule_sets[i]); + ::std::copy( tmp.begin(), tmp.end(), rule_sets.begin() + start ); } - bool overlaps(const Range<T>& x) const { - return (x.start <= start && start <= x.end) || (x.start <= end && end <= x.end); + { + decltype(this->arm_idxes) tmp; + for(auto i : v) + tmp.push_back(arm_idxes[i]); + ::std::copy( tmp.begin(), tmp.end(), arm_idxes.begin() + start ); } + } - friend ::std::ostream& operator<<(::std::ostream& os, const Range<T>& x) { - if( x.start == x.end ) { - return os << x.start; - } - else { - return os << x.start << " ... " << x.end; - } + t_rules_subset sub_slice(size_t ofs, size_t n) + { + t_rules_subset rv { n, this->is_arm_indexes }; + rv.rule_sets.reserve(n); + for(size_t i = 0; i < n; i++) + { + rv.rule_sets.push_back( this->rule_sets[ofs+i] ); + rv.arm_idxes.push_back( this->arm_idxes[ofs+i] ); } - }; - - TAGGED_UNION( Values, Unset, - (Unset, struct {}), - (Bool, struct { Branch false_branch, true_branch; }), - (Variant, ::std::vector< ::std::pair<unsigned int, Branch> >), - (Unsigned, ::std::vector< ::std::pair< Range<uint64_t>, Branch> >), - (Signed, ::std::vector< ::std::pair< Range<int64_t>, Branch> >), - (Float, ::std::vector< ::std::pair< Range<double>, Branch> >), - (String, ::std::vector< ::std::pair< ::std::string, Branch> >), - (Slice, struct { - ::std::vector< ::std::pair< unsigned int, Branch> > fixed_arms; - //::std::vector< ::std::pair< unsigned int, Branch> > variable_arms; - }) - ); - - // TODO: Arm specialisation? - field_path_t m_field_path; - Values m_branches; - Branch m_default; - - DecisionTreeNode( field_path_t field_path ): - // TODO: This is commented out fo a reason, but I don't know why. - //m_field_path( mv$(field_path) ), - m_branches(), - m_default() - {} - - static Branch clone(const Branch& b); - static Values clone(const Values& x); - DecisionTreeNode clone() const; - - void populate_tree_from_rule(const Span& sp, unsigned int arm_index, const PatternRule* first_rule, unsigned int rule_count) { - populate_tree_from_rule(sp, first_rule, rule_count, [sp,arm_index](Branch& branch){ - TU_MATCHA( (branch), (e), - (Unset, - // Good - ), - (Subtree, - if( e->m_branches.is_Unset() && e->m_default.is_Unset() ) { - // Good. - } - else { - BUG(sp, "Duplicate terminal - branch="<<branch); - } - ), - (Terminal, - // TODO: This is ok if it's due to overlapping rules (e.g. ranges) - //BUG(sp, "Duplicate terminal - Existing goes to arm " << e << ", new goes to arm " << arm_index ); - ) - ) - branch = Branch::make_Terminal(arm_index); - }); + return rv; } - // `and_then` - Closure called after processing the final rule - void populate_tree_from_rule(const Span& sp, const PatternRule* first_rule, unsigned int rule_count, ::std::function<void(Branch&)> and_then); - - /// Simplifies the tree by eliminating nodes that don't make a decision - void simplify(); - /// Propagate the m_default arm's contents to value arms, and vice-versa - void propagate_default(); - /// HELPER: Unfies the rules from the provided branch with this node - void unify_from(const Branch& b); - - ::MIR::LValue get_field(const ::MIR::LValue& base, unsigned int base_depth) const { - ::MIR::LValue cur = base.clone(); - for(unsigned int i = base_depth; i < m_field_path.size(); i ++ ) { - const auto idx = m_field_path.data[i]; - if( idx == FIELD_DEREF ) { - cur = ::MIR::LValue::make_Deref({ box$(cur) }); + void push_arm(const ::std::vector<PatternRule>& x, size_t arm_idx, size_t pat_idx) + { + assert(is_arm_indexes); + rule_sets.push_back(&x); + assert(arm_idx <= 0xFFF); + assert(pat_idx <= 0xFFF); + arm_idxes.push_back(arm_idx | (pat_idx << 12)); + } + void push_bb(const ::std::vector<PatternRule>& x, ::MIR::BasicBlockId bb) + { + assert(!is_arm_indexes); + rule_sets.push_back(&x); + arm_idxes.push_back(bb); + } + + friend ::std::ostream& operator<<(::std::ostream& os, const t_rules_subset& x) { + os << "t_rules_subset{"; + for(size_t i = 0; i < x.rule_sets.size(); i ++) + { + if(i != 0) + os << ", "; + os << "["; + if(x.is_arm_indexes) + { + os << (x.arm_idxes[i] & 0xFFF) << "," << (x.arm_idxes[i] >> 12); } - else { - cur = ::MIR::LValue::make_Field({ box$(cur), idx }); + else + { + os << "bb" << x.arm_idxes[i]; } + os << "]"; + os << ": " << *x.rule_sets[i]; } - return cur; + os << "}"; + return os; } - - friend ::std::ostream& operator<<(::std::ostream& os, const Branch& x); - friend ::std::ostream& operator<<(::std::ostream& os, const DecisionTreeNode& x); }; -struct DecisionTreeGen +class MatchGenGrouped { + const Span& sp; MirBuilder& m_builder; - const ::std::vector< ::MIR::BasicBlockId>& m_rule_blocks; - - DecisionTreeGen(MirBuilder& builder, const ::std::vector< ::MIR::BasicBlockId >& rule_blocks): - m_builder( builder ), - m_rule_blocks( rule_blocks ) - {} - - ::MIR::BasicBlockId get_block_for_rule(unsigned int rule_index) { - return m_rule_blocks.at( rule_index ); - } - - void generate_tree_code(const Span& sp, const DecisionTreeNode& node, const ::HIR::TypeRef& ty, const ::MIR::LValue& val) { - generate_tree_code(sp, node, ty, 0, val, [&](const auto& n){ - DEBUG("node = " << n); - // - Recurse on this method - this->generate_tree_code(sp, n, ty, val); - }); - } - void generate_tree_code( - const Span& sp, - const DecisionTreeNode& node, - const ::HIR::TypeRef& ty, unsigned int path_ofs, const ::MIR::LValue& base_val, - ::std::function<void(const DecisionTreeNode&)> and_then - ); - - void generate_branch(const DecisionTreeNode::Branch& branch, ::std::function<void(const DecisionTreeNode&)> cb); - - // HELPER - ::MIR::LValue push_compare(const Span& sp, ::MIR::LValue left, ::MIR::eBinOp op, ::MIR::Param right); - - void generate_branches_Signed( - const Span& sp, - const DecisionTreeNode::Branch& default_branch, - const DecisionTreeNode::Values::Data_Signed& branches, - const ::HIR::TypeRef& ty, ::MIR::LValue val, - ::std::function<void(const DecisionTreeNode&)> and_then - ); - void generate_branches_Unsigned( - const Span& sp, - const DecisionTreeNode::Branch& default_branch, - const DecisionTreeNode::Values::Data_Unsigned& branches, - const ::HIR::TypeRef& ty, ::MIR::LValue val, - ::std::function<void(const DecisionTreeNode&)> and_then - ); - void generate_branches_Float( - const Span& sp, - const DecisionTreeNode::Branch& default_branch, - const DecisionTreeNode::Values::Data_Float& branches, - const ::HIR::TypeRef& ty, ::MIR::LValue val, - ::std::function<void(const DecisionTreeNode&)> and_then - ); - void generate_branches_Char( - const Span& sp, - const DecisionTreeNode::Branch& default_branch, - const DecisionTreeNode::Values::Data_Unsigned& branches, - const ::HIR::TypeRef& ty, ::MIR::LValue val, - ::std::function<void(const DecisionTreeNode&)> and_then - ); - void generate_branches_Bool( - const Span& sp, - const DecisionTreeNode::Branch& default_branch, - const DecisionTreeNode::Values::Data_Bool& branches, - const ::HIR::TypeRef& ty, ::MIR::LValue val, - ::std::function<void(const DecisionTreeNode&)> and_then - ); - void generate_branches_Borrow_str( - const Span& sp, - const DecisionTreeNode::Branch& default_branch, - const DecisionTreeNode::Values::Data_String& branches, - const ::HIR::TypeRef& ty, ::MIR::LValue val, - ::std::function<void(const DecisionTreeNode&)> and_then - ); - - void generate_branches_Enum( - const Span& sp, - const DecisionTreeNode::Branch& default_branch, - const DecisionTreeNode::Values::Data_Variant& branches, - const field_path_t& field_path, // used to know when to stop handling sub-nodes - const ::HIR::TypeRef& ty, ::MIR::LValue val, - ::std::function<void(const DecisionTreeNode&)> and_then - ); - void generate_branches_Slice( - const Span& sp, - const DecisionTreeNode::Branch& default_branch, - const DecisionTreeNode::Values::Data_Slice& branches, - const field_path_t& field_path, - const ::HIR::TypeRef& ty, ::MIR::LValue val, - ::std::function<void(const DecisionTreeNode&)> and_then - ); - void generate_tree_code__enum( - const Span& sp, - const DecisionTreeNode& node, const ::HIR::TypeRef& fake_ty, const ::MIR::LValue& val, - const field_path_t& path_prefix, - ::std::function<void(const DecisionTreeNode&)> and_then - ); -}; - -void MIR_LowerHIR_Match_DecisionTree( MirBuilder& builder, MirConverter& conv, ::HIR::ExprNode_Match& node, ::MIR::LValue match_val, t_arm_rules arm_rules, ::std::vector<ArmCode> arms_code, ::MIR::BasicBlockId first_cmp_block ) -{ - TRACE_FUNCTION; - - // XXX XXX XXX: The current codegen (below) will generate incorrect code if ordering matters. - // ``` - // match ("foo", "bar") - // { - // (_, "bar") => {}, // Expected - // ("foo", _) => {}, // Actual - // _ => {}, - // } - // ``` - - // TODO: Sort the columns in `arm_rules` to ensure that the most specific rule is parsed first. - // - Ordering within a pattern doesn't matter, only the order of arms matters. - // - This sort could be designed such that the above case would match correctly? - - DEBUG("- Generating rule bindings"); - ::std::vector< ::MIR::BasicBlockId> rule_blocks; - for(const auto& rule : arm_rules) + const ::HIR::TypeRef& m_top_ty; + const ::MIR::LValue& m_top_val; + const ::std::vector<ArmCode>& m_arms_code; + + size_t m_field_path_ofs; +public: + MatchGenGrouped(MirBuilder& builder, const Span& sp, const ::HIR::TypeRef& top_ty, const ::MIR::LValue& top_val, const ::std::vector<ArmCode>& arms_code, size_t field_path_ofs): + sp(sp), + m_builder(builder), + m_top_ty(top_ty), + m_top_val(top_val), + m_arms_code(arms_code), + m_field_path_ofs(field_path_ofs) { - const auto& arm_code = arms_code[rule.arm_idx]; - ASSERT_BUG(node.span(), !arm_code.has_condition, "Decision tree doesn't (yet) support conditionals"); - - assert( rule.pat_idx < arm_code.destructures.size() ); - // Set the target for when a rule succeeds to the destructuring code for this rule - rule_blocks.push_back( arm_code.destructures[rule.pat_idx] ); - // - Tie the end of that block to the code block for this arm - builder.set_cur_block( rule_blocks.back() ); - builder.end_block( ::MIR::Terminator::make_Goto(arm_code.code) ); } + void gen_for_slice(t_rules_subset rules, size_t ofs, ::MIR::BasicBlockId default_arm); + void gen_dispatch(const ::std::vector<t_rules_subset>& rules, size_t ofs, const ::std::vector<::MIR::BasicBlockId>& arm_targets, ::MIR::BasicBlockId def_blk); + void gen_dispatch__primitive(::HIR::TypeRef ty, ::MIR::LValue val, const ::std::vector<t_rules_subset>& rules, size_t ofs, const ::std::vector<::MIR::BasicBlockId>& arm_targets, ::MIR::BasicBlockId def_blk); + void gen_dispatch__enum(::HIR::TypeRef ty, ::MIR::LValue val, const ::std::vector<t_rules_subset>& rules, size_t ofs, const ::std::vector<::MIR::BasicBlockId>& arm_targets, ::MIR::BasicBlockId def_blk); + void gen_dispatch__slice(::HIR::TypeRef ty, ::MIR::LValue val, const ::std::vector<t_rules_subset>& rules, size_t ofs, const ::std::vector<::MIR::BasicBlockId>& arm_targets, ::MIR::BasicBlockId def_blk); - // - Build tree by running each arm's pattern across it - DEBUG("- Building decision tree"); - DecisionTreeNode root_node({}); - unsigned int rule_idx = 0; - for( const auto& arm_rule : arm_rules ) - { - auto arm_idx = arm_rule.arm_idx; - DEBUG("(" << arm_idx << ", " << arm_rule.pat_idx << "): " << arm_rule.m_rules); - root_node.populate_tree_from_rule( node.m_arms[arm_idx].m_code->span(), rule_idx, arm_rule.m_rules.data(), arm_rule.m_rules.size() ); - rule_idx += 1; - } - DEBUG("root_node = " << root_node); - root_node.simplify(); - DEBUG("root_node = " << root_node); - root_node.propagate_default(); - DEBUG("root_node = " << root_node); - // TODO: Pretty print `root_node` - - // - Convert the above decision tree into MIR - DEBUG("- Emitting decision tree"); - DecisionTreeGen gen { builder, rule_blocks }; - builder.set_cur_block( first_cmp_block ); - gen.generate_tree_code( node.span(), root_node, node.m_value->m_res_type, mv$(match_val) ); - ASSERT_BUG(node.span(), !builder.block_active(), "Decision tree didn't terminate the final block"); -} + void gen_dispatch_range(const field_path_t& field_path, const ::MIR::Constant& first, const ::MIR::Constant& last, ::MIR::BasicBlockId def_blk); + void gen_dispatch_splitslice(const field_path_t& field_path, const PatternRule::Data_SplitSlice& e, ::MIR::BasicBlockId def_blk); -#if 0 -DecisionTreeNode MIR_LowerHIR_Match_DecisionTree__MakeTree(const Span& sp, t_arm_rules& arm_rules) -{ - ::std::vector<unsigned int> indexes; - ::std::vector< slice<PatternRule> > rules; - for(unsigned i = 0; i < arm_rules.size(); i ++) + ::MIR::LValue push_compare(::MIR::LValue left, ::MIR::eBinOp op, ::MIR::Param right) { - rules.push_back( arm_rules[i].m_rules ); - indexes.push_back(i); + return m_builder.lvalue_or_temp(sp, ::HIR::CoreType::Bool, + ::MIR::RValue::make_BinOp({ mv$(left), op, mv$(right) }) + ); } +}; - return MIR_LowerHIR_Match_DecisionTree__MakeTree_Node(sp, indexes, rules); -} -DecisionTreeNode MIR_LowerHIR_Match_DecisionTree__MakeTree_Node(const Span& sp, slice<unsigned int> arm_indexes, slice< slice<PaternRule>> arm_rules) -{ - assert( arm_indexes.size() == arm_rules.size() ); - assert( arm_rules.size() > 1 ); - assert( arm_rules[0].size() > 0 ); - - // 1. Sort list (should it already be sorted?) - for(const auto& rules : arm_rules) +namespace { + void push_flat_rules(::std::vector<PatternRule>& out_rules, PatternRule rule) { - ASSERT_BUG(sp, rules.size() != arm_rules[0].size(), ""); + TU_MATCHA( (rule), (e), + (Variant, + auto sub_rules = mv$(e.sub_rules); + out_rules.push_back( mv$(rule) ); + for(auto& sr : sub_rules) + push_flat_rules(out_rules, mv$(sr)); + ), + (Slice, + auto sub_rules = mv$(e.sub_rules); + out_rules.push_back( mv$(rule) ); + for(auto& sr : sub_rules) + push_flat_rules(out_rules, mv$(sr)); + ), + (SplitSlice, + auto leading = mv$(e.leading); + auto trailing = mv$(e.trailing); + out_rules.push_back( mv$(rule) ); + for(auto& sr : leading) + push_flat_rules(out_rules, mv$(sr)); + // TODO: the trailing rules need a special path format. + ASSERT_BUG(Span(), trailing.size() == 0, "TODO: Handle SplitSlice with trailing"); + for(auto& sr : trailing) + push_flat_rules(out_rules, mv$(sr)); + ), + (Bool, + out_rules.push_back( mv$(rule) ); + ), + (Value, + out_rules.push_back( mv$(rule) ); + ), + (ValueRange, + out_rules.push_back( mv$(rule) ); + ), + (Any, + out_rules.push_back( mv$(rule) ); + ) + ) } - - // 2. Detect all arms being `_` and move on to the next condition - while( ::std::all_of(arm_rules.begin(), arm_rules.end(), [](const auto& r){ return r.m_rules[0].is_Any(); }) ) + t_arm_rules flatten_rules(t_arm_rules rules) { - // Delete first rule from all and continue. - if( arm_rules[0].size() == 1 ) { - // No rules left? - BUG(sp, "Duplicate match arms"); - } - - for(auto& rules : arm_rules) + t_arm_rules rv; + rv.reserve(rules.size()); + for(auto& ruleset : rules) { - rules = rules.subslice_from(1); + ::std::vector<PatternRule> pattern_rules; + for( auto& r : ruleset.m_rules ) + { + push_flat_rules(pattern_rules, mv$(r)); + } + rv.push_back(PatternRuleset { ruleset.arm_idx, ruleset.pat_idx, mv$(pattern_rules) }); } + return rv; } +} - // We have a codition. - for(const auto& rules : arm_rules) - { - ASSERT_BUG(sp, rules[0].is_Any() || rules[0].tag() == arm_rules[0][0].tag(), "Mismatched rules in match"); - } - - bool has_any = arm_rules.back()[0].is_Any(); +void MIR_LowerHIR_Match_Grouped( + MirBuilder& builder, MirConverter& conv, ::HIR::ExprNode_Match& node, ::MIR::LValue match_val, + t_arm_rules arm_rules, ::std::vector<ArmCode> arms_code, ::MIR::BasicBlockId first_cmp_block + ) +{ + // TEMPORARY HACK: Grouped fails in complex matches (e.g. librustc_const_math Int::infer) + //MIR_LowerHIR_Match_Simple( builder, conv, node, mv$(match_val), mv$(arm_rules), mv$(arms_code), first_cmp_block ); + //return; - // All rules must either be _ or the same type, and can't all be _ - switch( arm_rules[0][0].tag() ) - { - case PatternRule::TAGDEAD: throw ""; - case PatternRule::TAG_Any: throw ""; + TRACE_FUNCTION_F(""); - case PatternRule::TAG_Variant: - break; - // TODO: Value and ValueRange can appear together. - // - They also overlap in non-trivial ways. - } -} -#endif - -// ---------------------------- -// DecisionTreeNode -// ---------------------------- -DecisionTreeNode::Branch DecisionTreeNode::clone(const DecisionTreeNode::Branch& b) { - TU_MATCHA( (b), (e), - (Unset, return Branch(e); ), - (Subtree, return Branch(box$( e->clone() )); ), - (Terminal, return Branch(e); ) - ) - throw ""; -} -DecisionTreeNode::Values DecisionTreeNode::clone(const DecisionTreeNode::Values& x) { - TU_MATCHA( (x), (e), - (Unset, return Values(e); ), - (Bool, - return Values::make_Bool({ clone(e.false_branch), clone(e.true_branch) }); - ), - (Variant, - Values::Data_Variant rv; - rv.reserve(e.size()); - for(const auto& v : e) - rv.push_back( ::std::make_pair(v.first, clone(v.second)) ); - return Values( mv$(rv) ); - ), - (Unsigned, - Values::Data_Unsigned rv; - rv.reserve(e.size()); - for(const auto& v : e) - rv.push_back( ::std::make_pair(v.first, clone(v.second)) ); - return Values( mv$(rv) ); - ), - (Signed, - Values::Data_Signed rv; - rv.reserve(e.size()); - for(const auto& v : e) - rv.push_back( ::std::make_pair(v.first, clone(v.second)) ); - return Values( mv$(rv) ); - ), - (Float, - Values::Data_Float rv; - rv.reserve(e.size()); - for(const auto& v : e) - rv.push_back( ::std::make_pair(v.first, clone(v.second)) ); - return Values( mv$(rv) ); - ), - (String, - Values::Data_String rv; - rv.reserve(e.size()); - for(const auto& v : e) - rv.push_back( ::std::make_pair(v.first, clone(v.second)) ); - return Values( mv$(rv) ); - ), - (Slice, - Values::Data_Slice rv; - rv.fixed_arms.reserve(e.fixed_arms.size()); - for(const auto& v : e.fixed_arms) - rv.fixed_arms.push_back( ::std::make_pair(v.first, clone(v.second)) ); - return Values( mv$(rv) ); - ) - ) - throw ""; -} -DecisionTreeNode DecisionTreeNode::clone() const { - DecisionTreeNode rv(m_field_path); - rv.m_field_path = m_field_path; - rv.m_branches = clone(m_branches); - rv.m_default = clone(m_default); - return rv; -} + // Flatten ruleset completely (remove grouping of enum/slice rules) + arm_rules = flatten_rules( mv$(arm_rules) ); -// Helpers for `populate_tree_from_rule` -namespace -{ - DecisionTreeNode::Branch new_branch_subtree(field_path_t path) + // - Create a "slice" of the passed rules, suitable for passing to the recursive part of the algo + t_rules_subset rules { arm_rules.size(), /*is_arm_indexes=*/true }; + for(const auto& r : arm_rules) { - return DecisionTreeNode::Branch( box$(DecisionTreeNode( mv$(path) )) ); + rules.push_arm( r.m_rules, r.arm_idx, r.pat_idx ); } - // Common code for numerics (Int, Uint, and Float) - template<typename T> - static void from_rule_value( - const Span& sp, - ::std::vector< ::std::pair< DecisionTreeNode::Range<T>, DecisionTreeNode::Branch> >& be, T ve, - const char* name, const field_path_t& field_path, ::std::function<void(DecisionTreeNode::Branch&)> and_then - ) - { - auto it = ::std::find_if(be.begin(), be.end(), [&](const auto& v){ return v.first.end >= ve; }); - if( it == be.end() || it->first.start > ve ) { - it = be.insert( it, ::std::make_pair( DecisionTreeNode::Range<T> { ve,ve }, new_branch_subtree(field_path) ) ); - } - else if( it->first.start == ve && it->first.end == ve ) { - // Equal, continue and add sub-pat - } - else { - // Collide or overlap! - TODO(sp, "Value patterns - " << name << " - Overlapping - " << it->first.start << " <= " << ve << " <= " << it->first.end); - } - and_then( it->second ); - } + auto inst = MatchGenGrouped { builder, node.span(), node.m_value->m_res_type, match_val, arms_code, 0 }; - template<typename T> - static void from_rule_valuerange( - const Span& sp, - ::std::vector< ::std::pair< DecisionTreeNode::Range<T>, DecisionTreeNode::Branch> >& be, T ve_start, T ve_end, - const char* name, const field_path_t& field_path, ::std::function<void(DecisionTreeNode::Branch&)> and_then - ) - { - TRACE_FUNCTION_F("be=[" << FMT_CB(os, for(const auto& i:be) os << i.first <<" , ";) << "], new=" << ve_start << "..." << ve_end); - ASSERT_BUG(sp, ve_start <= ve_end, "Range pattern with a start after the end - " << ve_start << "..." << ve_end); + // NOTE: This block should never be used + auto default_arm = builder.new_bb_unlinked(); - if( ve_start == ve_end ) { - from_rule_value(sp, be, ve_start, name, field_path, and_then); - return ; - } - // - Find the first entry that ends after the new one starts. - auto it = ::std::find_if(be.begin(), be.end(), [&](const auto& v){ return v.first.end >= ve_start; }); - while(ve_start < ve_end) - { - if( it == be.end() ) { - DEBUG("new = (" << ve_start << "..." << ve_end << "), exist=END"); - it = be.insert( it, ::std::make_pair( DecisionTreeNode::Range<T> { ve_start,ve_end }, new_branch_subtree(field_path) ) ); - and_then(it->second); - return ; - } - DEBUG("new = (" << ve_start << "..." << ve_end << "), exist=" << it->first); - // If the located entry starts after the end of this range - if( it->first.start >= ve_end ) { - DEBUG("- New free"); - it = be.insert( it, ::std::make_pair( DecisionTreeNode::Range<T> { ve_start,ve_end }, new_branch_subtree(field_path) ) ); - and_then(it->second); - return ; - } - // If this range is equal to the existing, just recurse into it - else if( it->first.start == ve_start && it->first.end == ve_end ) { - DEBUG("- Equal"); - and_then(it->second); - return ; - } - // If the new range starts before the start of this range, add a new entry before the existing one - else if( it->first.start > ve_start ) { - DEBUG("- New head, continue"); - it = be.insert( it, ::std::make_pair( DecisionTreeNode::Range<T> { ve_start,it->first.start-1 }, new_branch_subtree(field_path) ) ); - and_then(it->second); - ++ it; - ve_start = it->first.start; - } - // If the new range ends before the end of this range, split the existing range and recurse into the first - else if( it->first.end > ve_end ) { - DEBUG("- Inner"); - assert(ve_start == it->first.start); - it = be.insert( it, ::std::make_pair( DecisionTreeNode::Range<T> { ve_start, ve_end }, DecisionTreeNode::clone(it->second) ) ); - and_then(it->second); - (it+1)->first.start = ve_end+1; - return ; - } - // (else) if the new range ends after the end of this range, apply to the rest of this range and advance - else { - DEBUG("- Shared head, continue"); - //assert(it->first.start == ve_start); - assert((it->first.end) < ve_end); + builder.set_cur_block( first_cmp_block ); + inst.gen_for_slice( mv$(rules), 0, default_arm ); - if( it->first.start != it->first.end ) - and_then(it->second); - ve_start = it->first.end + 1; - ++ it; - } - } - } + // Make the default infinite loop. + // - Preferably, it'd abort. + builder.set_cur_block(default_arm); + builder.end_block( ::MIR::Terminator::make_Goto(default_arm) ); } -void DecisionTreeNode::populate_tree_from_rule(const Span& sp, const PatternRule* first_rule, unsigned int rule_count, ::std::function<void(Branch&)> and_then) +void MatchGenGrouped::gen_for_slice(t_rules_subset arm_rules, size_t ofs, ::MIR::BasicBlockId default_arm) { - assert( rule_count > 0 ); - const auto& rule = *first_rule; - - if( m_field_path.size() == 0 ) { - m_field_path = rule.field_path; - } - else { - ASSERT_BUG(sp, m_field_path == rule.field_path, "Patterns with mismatched field paths - " << m_field_path << " != " << rule.field_path); - } + TRACE_FUNCTION_F("arm_rules=" << arm_rules << ", ofs="<<ofs << ", default_arm=" << default_arm); + ASSERT_BUG(sp, arm_rules.size() > 0, ""); - struct GET_BRANCHES_H { - static void unexpected_type(const Span& sp, const char* exp, const char* have) { - BUG(sp, "Mismatched rules - have " << exp << ", but have seen " << have); - } - }; - #define GET_BRANCHES(fld, var) (fld.is_Unset()\ - ? (fld = Values::make_##var({})).as_##var() \ - : (fld.is_##var() \ - ? fld.as_##var() : (GET_BRANCHES_H::unexpected_type(sp, #var, fld.tag_str()), fld.as_##var()) \ - )\ - ) - - - TU_MATCHA( (rule), (e), - (Any, { - if( rule_count == 1 ) + // Quick hack: Skip any layers entirely made up of PatternRule::Any + for(;;) + { + bool is_all_any = true; + for(size_t i = 0; i < arm_rules.size() && is_all_any; i ++) { - ASSERT_BUG(sp, !m_default.is_Terminal(), "Duplicate terminal rule"); - and_then(m_default); + if( arm_rules[i].size() <= ofs ) + is_all_any = false; + else if( ! arm_rules[i][ofs].is_Any() ) + is_all_any = false; } - else + if( ! is_all_any ) { - if( m_default.is_Unset() ) { - m_default = new_branch_subtree(rule.field_path); - m_default.as_Subtree()->populate_tree_from_rule(sp, first_rule+1, rule_count-1, and_then); - } - else TU_IFLET( Branch, m_default, Subtree, be, - be->populate_tree_from_rule(sp, first_rule+1, rule_count-1, and_then); - ) - else { - // NOTE: All lists processed as part of the same tree should be the same length - BUG(sp, "Duplicate terminal rule"); - } - } - // TODO: Should this also recurse into branches? - }), - (Variant, { - auto& be = GET_BRANCHES(m_branches, Variant); - - auto it = ::std::find_if( be.begin(), be.end(), [&](const auto& x){ return x.first >= e.idx; }); - // If this variant isn't yet processed, add a new subtree for it - if( it == be.end() || it->first != e.idx ) { - it = be.insert(it, ::std::make_pair(e.idx, new_branch_subtree(rule.field_path))); - assert( it->second.is_Subtree() ); - } - else { - if( it->second.is_Terminal() ) { - BUG(sp, "Duplicate terminal rule - " << it->second.as_Terminal()); - } - assert( !it->second.is_Unset() ); - assert( it->second.is_Subtree() ); + break ; } - auto& subtree = *it->second.as_Subtree(); + ofs ++; + DEBUG("Skip to ofs=" << ofs); + } - if( e.sub_rules.size() > 0 && rule_count > 1 ) - { - subtree.populate_tree_from_rule(sp, e.sub_rules.data(), e.sub_rules.size(), [&](Branch& branch){ - TU_MATCH_DEF(Branch, (branch), (be), - ( - BUG(sp, "Duplicate terminator"); - ), - (Unset, - branch = new_branch_subtree(rule.field_path); - ), - (Subtree, - ) - ) - branch.as_Subtree()->populate_tree_from_rule(sp, first_rule+1, rule_count-1, and_then); - }); - } - else if( e.sub_rules.size() > 0) - { - subtree.populate_tree_from_rule(sp, e.sub_rules.data(), e.sub_rules.size(), and_then); - } - else if( rule_count > 1 ) - { - subtree.populate_tree_from_rule(sp, first_rule+1, rule_count-1, and_then); - } - else + // Split current set of rules into groups based on _ patterns + for(size_t idx = 0; idx < arm_rules.size(); ) + { + // Completed arms + while( idx < arm_rules.size() && arm_rules[idx].size() <= ofs ) { - and_then(it->second); - } - }), - (Slice, - auto& be = GET_BRANCHES(m_branches, Slice); + auto next = idx+1 == arm_rules.size() ? default_arm : m_builder.new_bb_unlinked(); + ASSERT_BUG(sp, arm_rules[idx].size() == ofs, "Offset too large for rule - ofs=" << ofs << ", rules=" << arm_rules[idx]); + DEBUG(idx << ": Complete"); + // Emit jump to either arm code, or arm condition + if( arm_rules.is_arm() ) + { + auto ai = arm_rules.arm_idx(idx); + ASSERT_BUG(sp, m_arms_code.size() > 0, "Bottom-level ruleset with no arm code information"); + const auto& ac = m_arms_code[ai.first]; - auto it = ::std::find_if( be.fixed_arms.begin(), be.fixed_arms.end(), [&](const auto& x){ return x.first >= e.len; } ); - if( it == be.fixed_arms.end() || it->first != e.len ) { - it = be.fixed_arms.insert(it, ::std::make_pair(e.len, new_branch_subtree(rule.field_path))); - } - else { - if( it->second.is_Terminal() ) { - BUG(sp, "Duplicate terminal rule - " << it->second.as_Terminal()); - } - assert( !it->second.is_Unset() ); - } - assert( it->second.is_Subtree() ); - auto& subtree = *it->second.as_Subtree(); + m_builder.end_block( ::MIR::Terminator::make_Goto(ac.destructures[ai.second]) ); + m_builder.set_cur_block( ac.destructures[ai.second] ); - if( e.sub_rules.size() > 0 && rule_count > 1 ) - { - subtree.populate_tree_from_rule(sp, e.sub_rules.data(), e.sub_rules.size(), [&](Branch& branch){ - TU_MATCH_DEF(Branch, (branch), (be), - ( - BUG(sp, "Duplicate terminator"); - ), - (Unset, - branch = new_branch_subtree(rule.field_path); - ), - (Subtree, - ) - ) - branch.as_Subtree()->populate_tree_from_rule(sp, first_rule+1, rule_count-1, and_then); - }); - } - else if( e.sub_rules.size() > 0) - { - subtree.populate_tree_from_rule(sp, e.sub_rules.data(), e.sub_rules.size(), and_then); - } - else if( rule_count > 1 ) - { - subtree.populate_tree_from_rule(sp, first_rule+1, rule_count-1, and_then); - } - else - { - and_then(it->second); - } - ), - (SplitSlice, - //auto& be = GET_BRANCHES(m_branches, Slice); - TODO(sp, "SplitSlice in DTN - " << rule); - ), - (Bool, - auto& be = GET_BRANCHES(m_branches, Bool); + if( ac.has_condition ) + { + TODO(sp, "Handle conditionals in Grouped"); + // TODO: If the condition fails, this should re-try the match on other rules that could have worked. + // - For now, conditionals are disabled. - auto& branch = (e ? be.true_branch : be.false_branch); - if( branch.is_Unset() ) { - branch = new_branch_subtree( rule.field_path ); - } - else if( branch.is_Terminal() ) { - BUG(sp, "Duplicate terminal rule - " << branch.as_Terminal()); - } - else { - // Good. - } - if( rule_count > 1 ) - { - auto& subtree = *branch.as_Subtree(); - subtree.populate_tree_from_rule(sp, first_rule+1, rule_count-1, and_then); - } - else - { - and_then(branch); - } - ), - (Value, - TU_MATCHA( (e), (ve), - (Int, - auto& be = GET_BRANCHES(m_branches, Signed); - - from_rule_value(sp, be, ve.v, "Signed", rule.field_path, - [&](auto& branch) { - if( rule_count > 1 ) { - assert( branch.as_Subtree() ); - auto& subtree = *branch.as_Subtree(); - subtree.populate_tree_from_rule(sp, first_rule+1, rule_count-1, and_then); - } - else - { - and_then(branch); - } - }); - ), - (Uint, - auto& be = GET_BRANCHES(m_branches, Unsigned); - - from_rule_value(sp, be, ve.v, "Unsigned", rule.field_path, - [&](auto& branch) { - if( rule_count > 1 ) { - assert( branch.as_Subtree() ); - auto& subtree = *branch.as_Subtree(); - subtree.populate_tree_from_rule(sp, first_rule+1, rule_count-1, and_then); - } - else + // TODO: What if there's multiple patterns on this condition? + // - For now, only the first pattern gets edited. + // - Maybe clone the blocks used for the condition? + m_builder.end_block( ::MIR::Terminator::make_Goto(ac.cond_start) ); + + // Check for marking in `ac` that the block has already been terminated, assert that target is `next` + if( ai.second == 0 ) { - and_then(branch); - } - }); - ), - (Float, - auto& be = GET_BRANCHES(m_branches, Float); - - from_rule_value(sp, be, ve.v, "Float", rule.field_path, - [&](auto& branch) { - if( rule_count > 1 ) { - assert( branch.as_Subtree() ); - auto& subtree = *branch.as_Subtree(); - subtree.populate_tree_from_rule(sp, first_rule+1, rule_count-1, and_then); - } - else { - and_then(branch); + if( ac.cond_fail_tgt != 0) + { + ASSERT_BUG(sp, ac.cond_fail_tgt == next, "Condition fail target already set with mismatching arm, set to bb" << ac.cond_fail_tgt << " cur is bb" << next); + } + else + { + ac.cond_fail_tgt = next; + + m_builder.set_cur_block( ac.cond_end ); + m_builder.end_block( ::MIR::Terminator::make_If({ ac.cond_lval.clone(), ac.code, next }) ); + } } - }); - ), - (Bool, - BUG(sp, "Hit Bool in PatternRule::Value - " << e); - ), - (Bytes, - TODO(sp, "Value patterns - Bytes"); - ), - (StaticString, - auto& be = GET_BRANCHES(m_branches, String); - auto it = ::std::find_if(be.begin(), be.end(), [&](const auto& v){ return v.first >= ve; }); - if( it == be.end() || it->first != ve ) { - it = be.insert( it, ::std::make_pair(ve, new_branch_subtree(rule.field_path) ) ); - } - auto& branch = it->second; - if( rule_count > 1 ) - { - assert( branch.as_Subtree() ); - auto& subtree = *branch.as_Subtree(); - subtree.populate_tree_from_rule(sp, first_rule+1, rule_count-1, and_then); + if( next != default_arm ) + m_builder.set_cur_block(next); + } + else + { + m_builder.end_block( ::MIR::Terminator::make_Goto(ac.code) ); + ASSERT_BUG(sp, idx+1 == arm_rules.size(), "Ended arm with other arms present"); + } } else { - and_then(branch); + auto bb = arm_rules.bb_idx(idx); + m_builder.end_block( ::MIR::Terminator::make_Goto(bb) ); + while( idx+1 < arm_rules.size() && bb == arm_rules.bb_idx(idx) && arm_rules[idx].size() == ofs ) + idx ++; + ASSERT_BUG(sp, idx+1 == arm_rules.size(), "Ended arm (inner) with other arms present"); } - ), - (Const, - BUG(sp, "Hit Const in PatternRule::Value - " << e); - ), - (ItemAddr, - BUG(sp, "Hit ItemAddr in PatternRule::Value - " << e); - ) - ) - ), - (ValueRange, - - ASSERT_BUG(sp, e.first.tag() == e.last.tag(), "Constant type mismatch in ValueRange - " << e.first << " and " << e.last); - TU_MATCHA( (e.first, e.last), (ve_start, ve_end), - (Int, - auto& be = GET_BRANCHES(m_branches, Signed); - from_rule_valuerange(sp, be, ve_start.v, ve_end.v, "Signed", rule.field_path, - [&](auto& branch) { - if( rule_count > 1 ) - { - assert( branch.as_Subtree() ); - auto& subtree = *branch.as_Subtree(); - subtree.populate_tree_from_rule(sp, first_rule+1, rule_count-1, and_then); - } - else - { - and_then(branch); - } - }); - ), - (Uint, - // TODO: Share code between the three numeric groups - auto& be = GET_BRANCHES(m_branches, Unsigned); - from_rule_valuerange(sp, be, ve_start.v, ve_end.v, "Unsigned", rule.field_path, - [&](auto& branch) { - if( rule_count > 1 ) - { - assert( branch.as_Subtree() ); - auto& subtree = *branch.as_Subtree(); - subtree.populate_tree_from_rule(sp, first_rule+1, rule_count-1, and_then); - } - else - { - and_then(branch); - } - }); - ), - (Float, - auto& be = GET_BRANCHES(m_branches, Float); - from_rule_valuerange(sp, be, ve_start.v, ve_end.v, "Float", rule.field_path, - [&](auto& branch) { - if( rule_count > 1 ) - { - assert( branch.as_Subtree() ); - auto& subtree = *branch.as_Subtree(); - subtree.populate_tree_from_rule(sp, first_rule+1, rule_count-1, and_then); - } - else - { - and_then(branch); - } - }); - ), - (Bool, - BUG(sp, "Hit Bool in PatternRule::ValueRange - " << e.first); - ), - (Bytes, - TODO(sp, "ValueRange patterns - Bytes"); - ), - (StaticString, - ERROR(sp, E0000, "Use of string in value range patter"); - ), - (Const, - BUG(sp, "Hit Const in PatternRule::ValueRange - " << e.first); - ), - (ItemAddr, - BUG(sp, "Hit ItemAddr in PatternRule::ValueRange - " << e.first); - ) - ) - ) - ) -} - -void DecisionTreeNode::simplify() -{ - struct H { - static void simplify_branch(Branch& b) - { - TU_IFLET(Branch, b, Subtree, be, - be->simplify(); - if( be->m_branches.is_Unset() ) { - auto v = mv$( be->m_default ); - b = mv$(v); - } - ) + idx ++; } - }; - TU_MATCHA( (m_branches), (e), - (Unset, - H::simplify_branch(m_default); - // Replace `this` with `m_default` if `m_default` is a subtree - // - Fixes the edge case for the top of the tree - if( m_default.is_Subtree() ) + // - Value arms + auto start = idx; + for(; idx < arm_rules.size() ; idx ++) { - *this = mv$(*m_default.as_Subtree()); - } - return ; - ), - (Bool, - H::simplify_branch(e.false_branch); - H::simplify_branch(e.true_branch); - ), - (Variant, - for(auto& branch : e) { - H::simplify_branch(branch.second); - } - ), - (Unsigned, - for(auto& branch : e) { - H::simplify_branch(branch.second); - } - ), - (Signed, - for(auto& branch : e) { - H::simplify_branch(branch.second); - } - ), - (Float, - for(auto& branch : e) { - H::simplify_branch(branch.second); - } - ), - (String, - for(auto& branch : e) { - H::simplify_branch(branch.second); - } - ), - (Slice, - for(auto& branch : e.fixed_arms) { - H::simplify_branch(branch.second); + if( arm_rules[idx].size() <= ofs ) + break; + if( arm_rules[idx][ofs].is_Any() ) + break; + if( arm_rules[idx][ofs].is_SplitSlice() ) + break; + // TODO: It would be nice if ValueRange could be combined with Value (if there's no overlap) + if( arm_rules[idx][ofs].is_ValueRange() ) + break; } - ) - ) + auto first_any = idx; - H::simplify_branch(m_default); -} + // Generate dispatch based on the above list + // - If there's value ranges they need special handling + // - Can sort arms within this group (ordering doesn't matter, as long as ranges are handled) + // - Sort must be stable. -void DecisionTreeNode::propagate_default() -{ - TRACE_FUNCTION_FR(*this, *this); - struct H { - static void handle_branch(Branch& b, const Branch& def) { - TU_IFLET(Branch, b, Subtree, be, - be->propagate_default(); - if( !def.is_Unset() ) + if( start < first_any ) + { + DEBUG(start << "+" << (first_any-start) << ": Values"); + bool has_default = (first_any < arm_rules.size()); + auto next = (has_default ? m_builder.new_bb_unlinked() : default_arm); + + // Sort rules before getting compatible runs + // TODO: Is this a valid operation? + arm_rules.sub_sort(ofs, start, first_any - start); + + // Create list of compatible arm slices (runs with the same selector value) + ::std::vector<t_rules_subset> slices; + auto cur_test = start; + for(auto i = start; i < first_any; i ++) + { + // Just check if the decision value differs (don't check nested rules) + if( ! rule_compatible(arm_rules[i][ofs], arm_rules[cur_test][ofs]) ) { - DEBUG("Unify " << *be << " with " << def); - be->unify_from(def); - be->propagate_default(); + slices.push_back( arm_rules.sub_slice(cur_test, i - cur_test) ); + cur_test = i; } - ) - } - }; + } + slices.push_back( arm_rules.sub_slice(cur_test, first_any - cur_test) ); + DEBUG("- " << slices.size() << " groupings"); + ::std::vector<::MIR::BasicBlockId> arm_blocks; + arm_blocks.reserve( slices.size() ); - TU_MATCHA( (m_branches), (e), - (Unset, - ), - (Bool, - DEBUG("- false"); - H::handle_branch(e.false_branch, m_default); - DEBUG("- true"); - H::handle_branch(e.true_branch, m_default); - ), - (Variant, - for(auto& branch : e) { - DEBUG("- V " << branch.first); - H::handle_branch(branch.second, m_default); - } - ), - (Unsigned, - for(auto& branch : e) { - DEBUG("- U " << branch.first); - H::handle_branch(branch.second, m_default); - } - ), - (Signed, - for(auto& branch : e) { - DEBUG("- S " << branch.first); - H::handle_branch(branch.second, m_default); - } - ), - (Float, - for(auto& branch : e) { - DEBUG("- " << branch.first); - H::handle_branch(branch.second, m_default); - } - ), - (String, - for(auto& branch : e) { - DEBUG("- '" << branch.first << "'"); - H::handle_branch(branch.second, m_default); - } - ), - (Slice, - for(auto& branch : e.fixed_arms) { - DEBUG("- [_;" << branch.first << "]"); - H::handle_branch(branch.second, m_default); - } - ) - ) - DEBUG("- default"); - TU_IFLET(Branch, m_default, Subtree, be, - be->propagate_default(); - - if( be->m_default.is_Unset() ) { - // Propagate default from value branches - TU_MATCHA( (m_branches), (e), - (Unset, - ), - (Bool, - be->unify_from(e.false_branch); - be->unify_from(e.true_branch); - ), - (Variant, - for(auto& branch : e) { - be->unify_from(branch.second); - } - ), - (Unsigned, - for(auto& branch : e) { - be->unify_from(branch.second); - } - ), - (Signed, - for(auto& branch : e) { - be->unify_from(branch.second); - } - ), - (Float, - for(auto& branch : e) { - be->unify_from(branch.second); - } - ), - (String, - for(auto& branch : e) { - be->unify_from(branch.second); - } - ), - (Slice, - for(auto& branch : e.fixed_arms) { - be->unify_from(branch.second); - } - ) - ) - } - ) -} + auto cur_blk = m_builder.pause_cur_block(); + // > Stable sort list + ::std::sort( slices.begin(), slices.end(), [&](const auto& a, const auto& b){ return a[0][ofs] < b[0][ofs]; } ); + // TODO: Should this do a stable sort of inner patterns too? + // - A sort of inner patterns such that `_` (and range?) patterns don't change position. -namespace { - static void unify_branch(DecisionTreeNode::Branch& dst, const DecisionTreeNode::Branch& src) { - if( dst.is_Unset() ) { - dst = DecisionTreeNode::clone(src); - } - else if( dst.is_Subtree() ) { - dst.as_Subtree()->unify_from(src); - } - else { - // Terminal, no unify - } - } + // > Get type of match, generate dispatch list. + for(size_t i = 0; i < slices.size(); i ++) + { + auto cur_block = m_builder.new_bb_unlinked(); + m_builder.set_cur_block(cur_block); - template<typename T> - void unify_from_vals_range(::std::vector< ::std::pair<T, DecisionTreeNode::Branch>>& dst, const ::std::vector< ::std::pair<T, DecisionTreeNode::Branch>>& src) - { - for(const auto& srcv : src) - { - // Find the first entry with an end greater than or equal to the start of this entry - auto it = ::std::find_if( dst.begin(), dst.end(), [&](const auto& x){ return x.first.end >= srcv.first.start; }); - // Not found? Insert a new branch - if( it == dst.end() ) { - it = dst.insert(it, ::std::make_pair(srcv.first, DecisionTreeNode::clone(srcv.second))); - } - // If the found entry doesn't overlap (the start of `*it` is after the end of `srcv`) - else if( it->first.start > srcv.first.end ) { - it = dst.insert(it, ::std::make_pair(srcv.first, DecisionTreeNode::clone(srcv.second))); - } - else if( it->first == srcv.first ) { - unify_branch( it->second, srcv.second ); - } - else { - // NOTE: Overlapping doesn't get handled here - } - } - } + for(size_t j = 0; j < slices[i].size(); j ++) + { + if(j > 0) + ASSERT_BUG(sp, slices[i][0][ofs] == slices[i][j][ofs], "Mismatched rules - " << slices[i][0][ofs] << " and " << slices[i][j][ofs]); + arm_blocks.push_back(cur_block); + } - template<typename T> - void unify_from_vals_pt(::std::vector< ::std::pair<T, DecisionTreeNode::Branch>>& dst, const ::std::vector< ::std::pair<T, DecisionTreeNode::Branch>>& src) - { - // Insert items not already present, merge present items - for(const auto& srcv : src) - { - auto it = ::std::find_if( dst.begin(), dst.end(), [&](const auto& x){ return x.first >= srcv.first; }); - // Not found? Insert a new branch - if( it == dst.end() || it->first != srcv.first ) { - it = dst.insert(it, ::std::make_pair(srcv.first, DecisionTreeNode::clone(srcv.second))); - } - else { - unify_branch( it->second, srcv.second ); + this->gen_for_slice(slices[i], ofs+1, next); } - } - } -} -void DecisionTreeNode::unify_from(const Branch& b) -{ - TRACE_FUNCTION_FR(*this << " with " << b, *this); + m_builder.set_cur_block(cur_blk); - assert( b.is_Terminal() || b.is_Subtree() ); + // Generate decision code + this->gen_dispatch(slices, ofs, arm_blocks, next); - if( m_default.is_Unset() ) { - if( b.is_Terminal() ) { - m_default = clone(b); - } - else { - m_default = clone(b.as_Subtree()->m_default); + if(has_default) + { + m_builder.set_cur_block(next); + } } - } - if( b.is_Subtree() && b.as_Subtree()->m_branches.tag() != m_branches.tag() ) { - // Is this a bug, or expected (and don't unify in?) - DEBUG("TODO - Unify mismatched arms? - " << b.as_Subtree()->m_branches.tag_str() << " and " << m_branches.tag_str()); - return ; - } - bool should_unify_subtree = b.is_Subtree() && this->m_field_path == b.as_Subtree()->m_field_path; - //if( b.is_Subtree() ) { - // ASSERT_BUG(Span(), this->m_field_path == b.as_Subtree()->m_field_path, "Unifiying DTNs with mismatched paths - " << this->m_field_path << " != " << b.as_Subtree()->m_field_path); - //} + // Collate matching blocks at `first_any` + assert(first_any == idx); + if( first_any < arm_rules.size() && arm_rules[idx].size() > ofs ) + { + // Collate all equal rules + while(idx < arm_rules.size() && arm_rules[idx][ofs] == arm_rules[first_any][ofs]) + idx ++; + DEBUG(first_any << "-" << idx << ": Multi-match"); - TU_MATCHA( (m_branches), (dst), - (Unset, - if( b.is_Subtree() ) { - assert( b.as_Subtree()->m_branches.is_Unset() ); - } - else { - // Huh? Terminal matching against an unset branch? - } - ), - (Bool, - auto* src = (b.is_Subtree() ? &b.as_Subtree()->m_branches.as_Bool() : nullptr); + bool has_next = idx < arm_rules.size(); + auto next = (has_next ? m_builder.new_bb_unlinked() : default_arm); - unify_branch( dst.false_branch, (src ? src->false_branch : b) ); - unify_branch( dst.true_branch , (src ? src->true_branch : b) ); - ), - (Variant, - if( should_unify_subtree ) { - auto& sb = b.as_Subtree()->m_branches; - ASSERT_BUG(Span(), sb.is_Variant(), "Unifying Variant with " << sb.tag_str()); - unify_from_vals_pt(dst, sb.as_Variant()); - } - else { - // Unify all with terminal branch - for(auto& dstv : dst) + const auto& rule = arm_rules[first_any][ofs]; + if(const auto* e = rule.opt_ValueRange()) { - unify_branch(dstv.second, b); + // Generate branch based on range + this->gen_dispatch_range(arm_rules[first_any][ofs].field_path, e->first, e->last, next); } - } - ), - (Unsigned, - if( should_unify_subtree ) { - auto& sb = b.as_Subtree()->m_branches; - ASSERT_BUG(Span(), sb.is_Unsigned(), "Unifying Unsigned with " << sb.tag_str()); - unify_from_vals_range(dst, sb.as_Unsigned()); - } - else { - for(auto& dstv : dst) + else if(const auto* e = rule.opt_SplitSlice()) { - unify_branch(dstv.second, b); + // Generate branch based on slice length being at least required. + this->gen_dispatch_splitslice(rule.field_path, *e, next); } - } - ), - (Signed, - if( should_unify_subtree ) { - auto& sb = b.as_Subtree()->m_branches; - ASSERT_BUG(Span(), sb.is_Signed(), "Unifying Signed with " << sb.tag_str()); - unify_from_vals_range(dst, sb.as_Signed()); - } - else { - for(auto& dstv : dst) + else { - unify_branch(dstv.second, b); - } - } - ), - (Float, - if( should_unify_subtree ) { - auto& sb = b.as_Subtree()->m_branches; - ASSERT_BUG(Span(), sb.is_Float(), "Unifying Float with " << sb.tag_str()); - unify_from_vals_range(dst, sb.as_Float()); - } - else { - for(auto& dstv : dst) { - unify_branch(dstv.second, b); - } - } - ), - (String, - if( should_unify_subtree ) { - auto& sb = b.as_Subtree()->m_branches; - ASSERT_BUG(Span(), sb.is_String(), "Unifying String with " << sb.tag_str()); - unify_from_vals_pt(dst, sb.as_String()); - } - else { - for(auto& dstv : dst) { - unify_branch( dstv.second, b ); + ASSERT_BUG(sp, rule.is_Any(), "Didn't expect non-Any rule here, got " << rule.tag_str() << " " << rule); } - } - ), - (Slice, - if( should_unify_subtree ) { - auto& sb = b.as_Subtree()->m_branches; - ASSERT_BUG(Span(), sb.is_Slice(), "Unifying Slice with " << sb.tag_str()); - const auto& src = sb.as_Slice(); - unify_from_vals_pt(dst.fixed_arms, src.fixed_arms); - } - else { - for(auto& dstv : dst.fixed_arms) { - unify_branch( dstv.second, b ); - } - } - ) - ) -} + // Step deeper into these arms + auto slice = arm_rules.sub_slice(first_any, idx - first_any); + this->gen_for_slice(mv$(slice), ofs+1, next); -::std::ostream& operator<<(::std::ostream& os, const DecisionTreeNode::Branch& x) { - TU_MATCHA( (x), (be), - (Unset, - os << "!"; - ), - (Terminal, - os << "ARM " << be; - ), - (Subtree, - os << *be; - ) - ) - return os; -} -::std::ostream& operator<<(::std::ostream& os, const DecisionTreeNode& x) { - os << "DTN [" << x.m_field_path << "] { "; - TU_MATCHA( (x.m_branches), (e), - (Unset, - os << "!, "; - ), - (Bool, - os << "false = " << e.false_branch << ", true = " << e.true_branch << ", "; - ), - (Variant, - os << "V "; - for(const auto& branch : e) { - os << branch.first << " = " << branch.second << ", "; - } - ), - (Unsigned, - os << "U "; - for(const auto& branch : e) { - const auto& range = branch.first; - if( range.start == range.end ) { - os << range.start; - } - else { - os << range.start << "..." << range.end; - } - os << " = " << branch.second << ", "; - } - ), - (Signed, - os << "S "; - for(const auto& branch : e) { - const auto& range = branch.first; - if( range.start == range.end ) { - os << range.start; - } - else { - os << range.start << "..." << range.end; - } - os << " = " << branch.second << ", "; - } - ), - (Float, - os << "F "; - for(const auto& branch : e) { - const auto& range = branch.first; - if( range.start == range.end ) { - os << range.start; - } - else { - os << range.start << "..." << range.end; + if(has_next) + { + m_builder.set_cur_block(next); } - os << " = " << branch.second << ", "; - } - ), - (String, - for(const auto& branch : e) { - os << "\"" << branch.first << "\"" << " = " << branch.second << ", "; } - ), - (Slice, - os << "len "; - for(const auto& branch : e.fixed_arms) { - os << "=" << branch.first << " = " << branch.second << ", "; - } - ) - ) + } - os << "* = " << x.m_default; - os << " }"; - return os; + ASSERT_BUG(sp, ! m_builder.block_active(), "Block left active after match group"); } - -// ---------------------------- -// DecisionTreeGen -// ---------------------------- - -void DecisionTreeGen::generate_tree_code( - const Span& sp, - const DecisionTreeNode& node, - const ::HIR::TypeRef& top_ty, unsigned int field_path_ofs, const ::MIR::LValue& top_val, - ::std::function<void(const DecisionTreeNode&)> and_then - ) +void MatchGenGrouped::gen_dispatch(const ::std::vector<t_rules_subset>& rules, size_t ofs, const ::std::vector<::MIR::BasicBlockId>& arm_targets, ::MIR::BasicBlockId def_blk) { - TRACE_FUNCTION_F("top_ty=" << top_ty << ", field_path_ofs=" << field_path_ofs << ", top_val=" << top_val << ", node=" << node); + const auto& field_path = rules[0][0][ofs].field_path; + TRACE_FUNCTION_F("rules=["<<rules <<"], ofs=" << ofs <<", field_path=" << field_path); + + { + size_t n = 0; + for(size_t i = 0; i < rules.size(); i++) + { + for(size_t j = 0; j < rules[i].size(); j++) + { + ASSERT_BUG(sp, rules[i][j][ofs].field_path == field_path, "Field path mismatch, " << rules[i][j][ofs].field_path << " != " << field_path); + n ++; + } + } + ASSERT_BUG(sp, arm_targets.size() == n, "Arm target count mismatch - " << n << " != " << arm_targets.size()); + } ::MIR::LValue val; ::HIR::TypeRef ty; - - get_ty_and_val(sp, m_builder.resolve(), top_ty, top_val, node.m_field_path, field_path_ofs, ty, val); + get_ty_and_val(sp, m_builder.resolve(), m_top_ty, m_top_val, field_path, m_field_path_ofs, ty, val); DEBUG("ty = " << ty << ", val = " << val); - TU_MATCHA( (ty.m_data), (e), - (Infer, BUG(sp, "Ivar for in match type"); ), - (Diverge, BUG(sp, "Diverge in match type"); ), - (Primitive, - switch(e) - { - case ::HIR::CoreType::Bool: - ASSERT_BUG(sp, node.m_branches.is_Bool(), "Tree for bool isn't a _Bool - node="<<node); - this->generate_branches_Bool(sp, node.m_default, node.m_branches.as_Bool(), ty, mv$(val), mv$(and_then)); - break; - case ::HIR::CoreType::U8: - case ::HIR::CoreType::U16: - case ::HIR::CoreType::U32: - case ::HIR::CoreType::U64: - case ::HIR::CoreType::U128: - case ::HIR::CoreType::Usize: - ASSERT_BUG(sp, node.m_branches.is_Unsigned(), "Tree for unsigned isn't a _Unsigned - node="<<node); - this->generate_branches_Unsigned(sp, node.m_default, node.m_branches.as_Unsigned(), ty, mv$(val), mv$(and_then)); - break; - case ::HIR::CoreType::I8: - case ::HIR::CoreType::I16: - case ::HIR::CoreType::I32: - case ::HIR::CoreType::I64: - case ::HIR::CoreType::I128: - case ::HIR::CoreType::Isize: - ASSERT_BUG(sp, node.m_branches.is_Signed(), "Tree for unsigned isn't a _Signed - node="<<node); - this->generate_branches_Signed(sp, node.m_default, node.m_branches.as_Signed(), ty, mv$(val), mv$(and_then)); - break; - case ::HIR::CoreType::Char: - ASSERT_BUG(sp, node.m_branches.is_Unsigned(), "Tree for char isn't a _Unsigned - node="<<node); - this->generate_branches_Char(sp, node.m_default, node.m_branches.as_Unsigned(), ty, mv$(val), mv$(and_then)); - break; - case ::HIR::CoreType::Str: - ASSERT_BUG(sp, node.m_branches.is_String(), "Tree for &str isn't a _String - node="<<node); - this->generate_branches_Borrow_str(sp, node.m_default, node.m_branches.as_String(), ty, mv$(val), mv$(and_then)); - break; - case ::HIR::CoreType::F32: - case ::HIR::CoreType::F64: - ASSERT_BUG(sp, node.m_branches.is_Float(), "Tree for float isn't a _Float - node="<<node); - this->generate_branches_Float(sp, node.m_default, node.m_branches.as_Float(), ty, mv$(val), mv$(and_then)); - break; - default: - TODO(sp, "Primitive - " << ty); - break; - } + TU_MATCHA( (ty.m_data), (te), + (Infer, + BUG(sp, "Hit _ in type - " << ty); ), - (Tuple, - BUG(sp, "Decision node on tuple - node=" << node); + (Diverge, + BUG(sp, "Matching over !"); + ), + (Primitive, + this->gen_dispatch__primitive(mv$(ty), mv$(val), rules, ofs, arm_targets, def_blk); ), (Path, - // This is either a struct destructure or an enum - TU_MATCHA( (e.binding), (pbe), - (Unbound, - BUG(sp, "Encounterd unbound path - " << e.path); - ), - (Opaque, - and_then(node); - ), - (Struct, - assert(pbe); - TU_MATCHA( (pbe->m_data), (fields), - (Unit, - and_then(node); + // Matching over a path can only happen with an enum. + // TODO: What about `box` destructures? + // - They're handled via hidden derefs + if( !te.binding.is_Enum() ) { + TU_MATCHA( (te.binding), (pbe), + (Unbound, + BUG(sp, "Encounterd unbound path - " << te.path); ), - (Tuple, - BUG(sp, "Decision node on tuple struct"); + (Opaque, + BUG(sp, "Attempting to match over opaque type - " << ty); ), - (Named, - BUG(sp, "Decision node on struct"); + (Struct, + const auto& str_data = pbe->m_data; + TU_MATCHA( (str_data), (sd), + (Unit, + BUG(sp, "Attempting to match over unit type - " << ty); + ), + (Tuple, + TODO(sp, "Matching on tuple-like struct?"); + ), + (Named, + TODO(sp, "Matching on struct?"); + ) + ) + ), + (Union, + TODO(sp, "Match over Union"); + ), + (Enum, ) ) - ), - (Union, - TODO(sp, "Decision node on Union"); - ), - (Enum, - ASSERT_BUG(sp, node.m_branches.is_Variant(), "Tree for enum isn't a Variant - node="<<node); - assert(pbe); - this->generate_branches_Enum(sp, node.m_default, node.m_branches.as_Variant(), node.m_field_path, ty, mv$(val), mv$(and_then)); - ) - ) + } + + this->gen_dispatch__enum(mv$(ty), mv$(val), rules, ofs, arm_targets, def_blk); ), (Generic, - and_then(node); + BUG(sp, "Attempting to match a generic"); ), (TraitObject, - ERROR(sp, E0000, "Attempting to match over a trait object"); + BUG(sp, "Attempting to match a trait object"); ), (ErasedType, - ERROR(sp, E0000, "Attempting to match over an erased type"); + BUG(sp, "Attempting to match an erased type"); ), (Array, - // TODO: Slice patterns, sequential comparison/sub-match - TODO(sp, "Match over array"); + BUG(sp, "Attempting to match on an Array (should have been destructured)"); ), (Slice, - ASSERT_BUG(sp, node.m_branches.is_Slice(), "Tree for [T] isn't a _Slice - node="<<node); - this->generate_branches_Slice(sp, node.m_default, node.m_branches.as_Slice(), node.m_field_path, ty, mv$(val), mv$(and_then)); + // TODO: Slice size matches! + this->gen_dispatch__slice(mv$(ty), mv$(val), rules, ofs, arm_targets, def_blk); + ), + (Tuple, + BUG(sp, "Match directly on tuple"); ), (Borrow, - if( *e.inner == ::HIR::CoreType::Str ) { - TODO(sp, "Match over &str"); - } - else { - BUG(sp, "Decision node on non-str/[T] borrow - " << ty); - } + BUG(sp, "Match directly on borrow"); ), (Pointer, - ERROR(sp, E0000, "Attempting to match over a pointer"); + // TODO: Could this actually be valid? + BUG(sp, "Attempting to match a pointer - " << ty); ), (Function, - ERROR(sp, E0000, "Attempting to match over a functon pointer"); + // TODO: Could this actually be valid? + BUG(sp, "Attempting to match a function pointer - " << ty); ), (Closure, - ERROR(sp, E0000, "Attempting to match over a closure"); + BUG(sp, "Attempting to match a closure"); ) ) } -void DecisionTreeGen::generate_branch(const DecisionTreeNode::Branch& branch, ::std::function<void(const DecisionTreeNode&)> cb) -{ - assert( !branch.is_Unset() ); - if( branch.is_Terminal() ) { - this->m_builder.end_block( ::MIR::Terminator::make_Goto( this->get_block_for_rule( branch.as_Terminal() ) ) ); - } - else { - assert( branch.is_Subtree() ); - const auto& subnode = *branch.as_Subtree(); - - cb(subnode); - } -} - -::MIR::LValue DecisionTreeGen::push_compare(const Span& sp, ::MIR::LValue left, ::MIR::eBinOp op, ::MIR::Param right) -{ - return m_builder.lvalue_or_temp(sp, ::HIR::CoreType::Bool, - ::MIR::RValue::make_BinOp({ mv$(left), op, mv$(right) }) - ); -} - -// TODO: Unify logic for these two, and support simpler checks for sequential values -void DecisionTreeGen::generate_branches_Signed( - const Span& sp, - const DecisionTreeNode::Branch& default_branch, - const DecisionTreeNode::Values::Data_Signed& branches, - const ::HIR::TypeRef& ty, ::MIR::LValue val, - ::std::function<void(const DecisionTreeNode&)> and_then - ) -{ - auto ity = ty.m_data.as_Primitive(); - auto default_block = m_builder.new_bb_unlinked(); - - // TODO: Convert into an integer switch w/ offset instead of chained comparisons - - for( const auto& branch : branches ) - { - auto next_block = (&branch == &branches.back() ? default_block : m_builder.new_bb_unlinked()); - - auto cmp_gt_block = m_builder.new_bb_unlinked(); - auto val_cmp_lt = push_compare(sp, val.clone(), ::MIR::eBinOp::LT, ::MIR::Constant::make_Int({ branch.first.start, ity })); - m_builder.end_block( ::MIR::Terminator::make_If({ mv$(val_cmp_lt), default_block, cmp_gt_block }) ); - m_builder.set_cur_block( cmp_gt_block ); - - auto success_block = m_builder.new_bb_unlinked(); - auto val_cmp_gt = push_compare(sp, val.clone(), ::MIR::eBinOp::GT, ::MIR::Constant::make_Int({ branch.first.end, ity })); - m_builder.end_block( ::MIR::Terminator::make_If({ mv$(val_cmp_gt), next_block, success_block }) ); - - m_builder.set_cur_block( success_block ); - this->generate_branch(branch.second, and_then); - - m_builder.set_cur_block( next_block ); - } - assert( m_builder.block_active() ); - - if( default_branch.is_Unset() ) { - // TODO: Emit error if non-exhaustive - m_builder.end_block( ::MIR::Terminator::make_Diverge({}) ); - } - else { - this->generate_branch(default_branch, and_then); - } -} - -void DecisionTreeGen::generate_branches_Unsigned( - const Span& sp, - const DecisionTreeNode::Branch& default_branch, - const DecisionTreeNode::Values::Data_Unsigned& branches, - const ::HIR::TypeRef& ty, ::MIR::LValue val, - ::std::function<void(const DecisionTreeNode&)> and_then - ) +void MatchGenGrouped::gen_dispatch__primitive(::HIR::TypeRef ty, ::MIR::LValue val, const ::std::vector<t_rules_subset>& rules, size_t ofs, const ::std::vector<::MIR::BasicBlockId>& arm_targets, ::MIR::BasicBlockId def_blk) { - auto ity = ty.m_data.as_Primitive(); - auto default_block = m_builder.new_bb_unlinked(); - - // TODO: Convert into an integer switch w/ offset instead of chained comparisons - - for( const auto& branch : branches ) + auto te = ty.m_data.as_Primitive(); + switch(te) { - auto next_block = (&branch == &branches.back() ? default_block : m_builder.new_bb_unlinked()); - - auto cmp_gt_block = m_builder.new_bb_unlinked(); - auto val_cmp_lt = push_compare(sp, val.clone(), ::MIR::eBinOp::LT, ::MIR::Constant::make_Uint({ branch.first.start, ity })); - m_builder.end_block( ::MIR::Terminator::make_If({ mv$(val_cmp_lt), default_block, cmp_gt_block }) ); - m_builder.set_cur_block( cmp_gt_block ); + case ::HIR::CoreType::Bool: { + ASSERT_BUG(sp, rules.size() <= 2, "More than 2 rules for boolean"); + for(size_t i = 0; i < rules.size(); i++) + { + ASSERT_BUG(sp, rules[i][0][ofs].is_Bool(), "PatternRule for bool isn't _Bool"); + } + + // False sorts before true. + auto fail_bb = rules.size() == 2 ? arm_targets[ 0] : (rules[0][0][ofs].as_Bool() ? def_blk : arm_targets[0]); + auto succ_bb = rules.size() == 2 ? arm_targets[rules[0].size()] : (rules[0][0][ofs].as_Bool() ? arm_targets[0] : def_blk); + + m_builder.end_block( ::MIR::Terminator::make_If({ mv$(val), succ_bb, fail_bb }) ); + } break; + case ::HIR::CoreType::U8: + case ::HIR::CoreType::U16: + case ::HIR::CoreType::U32: + case ::HIR::CoreType::U64: + case ::HIR::CoreType::U128: + case ::HIR::CoreType::Usize: + + case ::HIR::CoreType::I8: + case ::HIR::CoreType::I16: + case ::HIR::CoreType::I32: + case ::HIR::CoreType::I64: + case ::HIR::CoreType::I128: + case ::HIR::CoreType::Isize: + + case ::HIR::CoreType::Char: + if( rules.size() == 1 ) + { + // Special case, single option, equality only + const auto& r = rules[0][0][ofs]; + ASSERT_BUG(sp, r.is_Value(), "Matching without _Value pattern - " << r.tag_str()); + const auto& re = r.as_Value(); + auto test_val = ::MIR::Param(re.clone()); + auto cmp_lval = m_builder.get_rval_in_if_cond(sp, ::MIR::RValue::make_BinOp({ val.clone(), ::MIR::eBinOp::EQ, mv$(test_val) })); + m_builder.end_block( ::MIR::Terminator::make_If({ mv$(cmp_lval), arm_targets[0], def_blk }) ); + } + else + { + // TODO: Add a SwitchInt terminator for use with this. (Or just a SwitchVal terminator?) - auto success_block = m_builder.new_bb_unlinked(); - auto val_cmp_gt = push_compare(sp, val.clone(), ::MIR::eBinOp::GT, ::MIR::Constant::make_Uint({ branch.first.end, ity })); - m_builder.end_block( ::MIR::Terminator::make_If({ mv$(val_cmp_gt), next_block, success_block }) ); + // NOTE: Rules are currently sorted + // TODO: If there are Constant::Const values in the list, they need to come first! (with equality checks) - m_builder.set_cur_block( success_block ); - this->generate_branch(branch.second, and_then); + // Does a sorted linear search. Binary search would be nicer but is harder to implement. + size_t tgt_ofs = 0; + for(size_t i = 0; i < rules.size(); i++) + { + for(size_t j = 1; j < rules[i].size(); j ++) + ASSERT_BUG(sp, arm_targets[tgt_ofs] == arm_targets[tgt_ofs+j], "Mismatched target blocks for Value match"); + + const auto& r = rules[i][0][ofs]; + ASSERT_BUG(sp, r.is_Value(), "Matching without _Value pattern - " << r.tag_str()); + const auto& re = r.as_Value(); + if(re.is_Const()) + TODO(sp, "Handle Constant::Const in match"); + + // IF v < tst : def_blk + // Skip if the previous value was the imediat predecesor + bool is_succ = i != 0 && (re.is_Uint() + ? re.as_Uint().v == rules[i-1][0][ofs].as_Value().as_Uint().v + 1 + : re.as_Int().v == rules[i-1][0][ofs].as_Value().as_Int().v + 1 + ); + if( !is_succ ) + { + auto cmp_eq_blk = m_builder.new_bb_unlinked(); + auto cmp_lval_lt = this->push_compare(val.clone(), ::MIR::eBinOp::LT, ::MIR::Param(re.clone())); + m_builder.end_block( ::MIR::Terminator::make_If({ mv$(cmp_lval_lt), def_blk, cmp_eq_blk }) ); + m_builder.set_cur_block(cmp_eq_blk); + } - m_builder.set_cur_block( next_block ); - } - assert( m_builder.block_active() ); + // IF v == tst : target + { + auto next_cmp_blk = m_builder.new_bb_unlinked(); + auto cmp_lval_eq = this->push_compare( val.clone(), ::MIR::eBinOp::EQ, ::MIR::Param(re.clone()) ); + m_builder.end_block( ::MIR::Terminator::make_If({ mv$(cmp_lval_eq), arm_targets[tgt_ofs], next_cmp_blk }) ); + m_builder.set_cur_block(next_cmp_blk); + } - if( default_branch.is_Unset() ) { - // TODO: Emit error if non-exhaustive - m_builder.end_block( ::MIR::Terminator::make_Diverge({}) ); - } - else { - this->generate_branch(default_branch, and_then); - } -} + tgt_ofs += rules[i].size(); + } + m_builder.end_block( ::MIR::Terminator::make_Goto(def_blk) ); + } + break; + case ::HIR::CoreType::F32: + case ::HIR::CoreType::F64: { + // NOTE: Rules are currently sorted + // TODO: If there are Constant::Const values in the list, they need to come first! + size_t tgt_ofs = 0; + for(size_t i = 0; i < rules.size(); i++) + { + for(size_t j = 1; j < rules[i].size(); j ++) + ASSERT_BUG(sp, arm_targets[tgt_ofs] == arm_targets[tgt_ofs+j], "Mismatched target blocks for Value match"); -void DecisionTreeGen::generate_branches_Float( - const Span& sp, - const DecisionTreeNode::Branch& default_branch, - const DecisionTreeNode::Values::Data_Float& branches, - const ::HIR::TypeRef& ty, ::MIR::LValue val, - ::std::function<void(const DecisionTreeNode&)> and_then - ) -{ - auto ity = ty.m_data.as_Primitive(); - auto default_block = m_builder.new_bb_unlinked(); + const auto& r = rules[i][0][ofs]; + ASSERT_BUG(sp, r.is_Value(), "Matching without _Value pattern - " << r.tag_str()); + const auto& re = r.as_Value(); + if(re.is_Const()) + TODO(sp, "Handle Constant::Const in match"); - for( const auto& branch : branches ) - { - auto next_block = (&branch == &branches.back() ? default_block : m_builder.new_bb_unlinked()); + // IF v < tst : def_blk + { + auto cmp_eq_blk = m_builder.new_bb_unlinked(); + auto cmp_lval_lt = m_builder.lvalue_or_temp(sp, ::HIR::CoreType::Bool, ::MIR::RValue::make_BinOp({ val.clone(), ::MIR::eBinOp::LT, ::MIR::Param(re.clone()) })); + m_builder.end_block( ::MIR::Terminator::make_If({ mv$(cmp_lval_lt), def_blk, cmp_eq_blk }) ); + m_builder.set_cur_block(cmp_eq_blk); + } - auto cmp_gt_block = m_builder.new_bb_unlinked(); - auto val_cmp_lt = push_compare(sp, val.clone(), ::MIR::eBinOp::LT, ::MIR::Constant::make_Float({ branch.first.start, ity })); - m_builder.end_block( ::MIR::Terminator::make_If({ mv$(val_cmp_lt), default_block, cmp_gt_block }) ); - m_builder.set_cur_block( cmp_gt_block ); + // IF v == tst : target + { + auto next_cmp_blk = m_builder.new_bb_unlinked(); + auto cmp_lval_eq = m_builder.lvalue_or_temp(sp, ::HIR::CoreType::Bool, ::MIR::RValue::make_BinOp({ val.clone(), ::MIR::eBinOp::EQ, ::MIR::Param(re.clone()) })); + m_builder.end_block( ::MIR::Terminator::make_If({ mv$(cmp_lval_eq), arm_targets[tgt_ofs], next_cmp_blk }) ); + m_builder.set_cur_block(next_cmp_blk); + } + + tgt_ofs += rules[i].size(); + } + m_builder.end_block( ::MIR::Terminator::make_Goto(def_blk) ); + } break; + case ::HIR::CoreType::Str: + // Remove the deref on the &str + auto oval = mv$(val); + auto val = mv$(*oval.as_Deref().val); + // NOTE: Rules are currently sorted + // TODO: If there are Constant::Const values in the list, they need to come first! + size_t tgt_ofs = 0; + for(size_t i = 0; i < rules.size(); i++) + { + for(size_t j = 1; j < rules[i].size(); j ++) + ASSERT_BUG(sp, arm_targets[tgt_ofs] == arm_targets[tgt_ofs+j], "Mismatched target blocks for Value match"); - auto success_block = m_builder.new_bb_unlinked(); - auto val_cmp_gt = push_compare(sp, val.clone(), ::MIR::eBinOp::GT, ::MIR::Constant::make_Float({ branch.first.end, ity })); - m_builder.end_block( ::MIR::Terminator::make_If({ mv$(val_cmp_gt), next_block, success_block }) ); + const auto& r = rules[i][0][ofs]; + ASSERT_BUG(sp, r.is_Value(), "Matching without _Value pattern - " << r.tag_str()); + const auto& re = r.as_Value(); + if(re.is_Const()) + TODO(sp, "Handle Constant::Const in match"); - m_builder.set_cur_block( success_block ); - this->generate_branch(branch.second, and_then); + // IF v < tst : def_blk + { + auto cmp_eq_blk = m_builder.new_bb_unlinked(); + auto cmp_lval_lt = m_builder.lvalue_or_temp(sp, ::HIR::CoreType::Bool, ::MIR::RValue::make_BinOp({ val.clone(), ::MIR::eBinOp::LT, ::MIR::Param(re.clone()) })); + m_builder.end_block( ::MIR::Terminator::make_If({ mv$(cmp_lval_lt), def_blk, cmp_eq_blk }) ); + m_builder.set_cur_block(cmp_eq_blk); + } - m_builder.set_cur_block( next_block ); - } - assert( m_builder.block_active() ); + // IF v == tst : target + { + auto next_cmp_blk = m_builder.new_bb_unlinked(); + auto cmp_lval_eq = m_builder.lvalue_or_temp(sp, ::HIR::CoreType::Bool, ::MIR::RValue::make_BinOp({ val.clone(), ::MIR::eBinOp::EQ, ::MIR::Param(re.clone()) })); + m_builder.end_block( ::MIR::Terminator::make_If({ mv$(cmp_lval_eq), arm_targets[tgt_ofs], next_cmp_blk }) ); + m_builder.set_cur_block(next_cmp_blk); + } - if( default_branch.is_Unset() ) { - ERROR(sp, E0000, "Match over floating point with no `_` arm"); - } - else { - this->generate_branch(default_branch, and_then); + tgt_ofs += rules[i].size(); + } + m_builder.end_block( ::MIR::Terminator::make_Goto(def_blk) ); + break; } } -void DecisionTreeGen::generate_branches_Char( - const Span& sp, - const DecisionTreeNode::Branch& default_branch, - const DecisionTreeNode::Values::Data_Unsigned& branches, - const ::HIR::TypeRef& ty, ::MIR::LValue val, - ::std::function<void(const DecisionTreeNode&)> and_then - ) +void MatchGenGrouped::gen_dispatch__enum(::HIR::TypeRef ty, ::MIR::LValue val, const ::std::vector<t_rules_subset>& rules, size_t ofs, const ::std::vector<::MIR::BasicBlockId>& arm_targets, ::MIR::BasicBlockId def_blk) { - auto default_block = m_builder.new_bb_unlinked(); + TRACE_FUNCTION; + auto& te = ty.m_data.as_Path(); + const auto& pbe = te.binding.as_Enum(); - // TODO: Convert into an integer switch w/ offset instead of chained comparisons + auto decison_arm = m_builder.pause_cur_block(); - for( const auto& branch : branches ) + auto var_count = pbe->m_variants.size(); + ::std::vector< ::MIR::BasicBlockId> arms(var_count, def_blk); + size_t arm_idx = 0; + for(size_t i = 0; i < rules.size(); i ++) { - auto next_block = (&branch == &branches.back() ? default_block : m_builder.new_bb_unlinked()); - - auto cmp_gt_block = m_builder.new_bb_unlinked(); - auto val_cmp_lt = push_compare(sp, val.clone(), ::MIR::eBinOp::LT, ::MIR::Constant::make_Uint({ branch.first.start, ::HIR::CoreType::Char })); - m_builder.end_block( ::MIR::Terminator::make_If({ mv$(val_cmp_lt), default_block, cmp_gt_block }) ); - m_builder.set_cur_block( cmp_gt_block ); - - auto success_block = m_builder.new_bb_unlinked(); - auto val_cmp_gt = push_compare(sp, val.clone(), ::MIR::eBinOp::GT, ::MIR::Constant::make_Uint({ branch.first.end, ::HIR::CoreType::Char })); - m_builder.end_block( ::MIR::Terminator::make_If({ mv$(val_cmp_gt), next_block, success_block }) ); + ASSERT_BUG(sp, rules[i][0][ofs].is_Variant(), "Rule for enum isn't Any or Variant - " << rules[i][0][ofs].tag_str()); + const auto& re = rules[i][0][ofs].as_Variant(); + unsigned int var_idx = re.idx; + DEBUG("Variant " << var_idx); - m_builder.set_cur_block( success_block ); - this->generate_branch(branch.second, and_then); + ASSERT_BUG(sp, re.sub_rules.size() == 0, "Sub-rules in MatchGenGrouped"); - m_builder.set_cur_block( next_block ); - } - assert( m_builder.block_active() ); - - if( default_branch.is_Unset() ) { - // TODO: Error if not exhaustive. - m_builder.end_block( ::MIR::Terminator::make_Diverge({}) ); - } - else { - this->generate_branch(default_branch, and_then); - } -} -void DecisionTreeGen::generate_branches_Bool( - const Span& sp, - const DecisionTreeNode::Branch& default_branch, - const DecisionTreeNode::Values::Data_Bool& branches, - const ::HIR::TypeRef& ty, ::MIR::LValue val, - ::std::function<void(const DecisionTreeNode&)> and_then - ) -{ - //assert( ty.m_data.is_Boolean() ); - - if( default_branch.is_Unset() ) - { - if( branches.false_branch.is_Unset() || branches.true_branch.is_Unset() ) { - // Non-exhaustive match - ERROR - } - } - else - { - if( branches.false_branch.is_Unset() && branches.true_branch.is_Unset() ) { - // Unreachable default (NOTE: Not an error here) + arms[var_idx] = arm_targets[arm_idx]; + for(size_t j = 0; j < rules[i].size(); j ++) + { + assert(arms[var_idx] == arm_targets[arm_idx]); + arm_idx ++; } } - // Emit an if based on the route taken - auto bb_false = m_builder.new_bb_unlinked(); - auto bb_true = m_builder.new_bb_unlinked(); - m_builder.end_block( ::MIR::Terminator::make_If({ mv$(val), bb_true, bb_false }) ); - - // Recurse into sub-patterns - const auto& branch_false = ( !branches.false_branch.is_Unset() ? branches.false_branch : default_branch ); - const auto& branch_true = ( !branches. true_branch.is_Unset() ? branches. true_branch : default_branch ); - - m_builder.set_cur_block(bb_true ); - this->generate_branch(branch_true , and_then); - m_builder.set_cur_block(bb_false); - this->generate_branch(branch_false, and_then); -} - -void DecisionTreeGen::generate_branches_Borrow_str( - const Span& sp, - const DecisionTreeNode::Branch& default_branch, - const DecisionTreeNode::Values::Data_String& branches, - const ::HIR::TypeRef& ty, ::MIR::LValue val, - ::std::function<void(const DecisionTreeNode&)> and_then - ) -{ - // TODO: Chained comparisons with ordering. - // - Would this just emit a eBinOp? That implies deep codegen support for strings. - // - rustc emits calls to PartialEq::eq for this and for slices. mrustc could use PartialOrd and fall back to PartialEq if unavaliable? - // > Requires crate access here! - A memcmp call is probably better, probably via a binop - // NOTE: The below implementation gets the final codegen to call memcmp on the strings by emitting eBinOp::{LT,GT} - - // - Remove the wrapping Deref (which must be there) - ASSERT_BUG(sp, val.is_Deref(), "Match over str without a deref - " << val); - auto tmp = mv$( *val.as_Deref().val ); - val = mv$(tmp); - - auto default_bb = m_builder.new_bb_unlinked(); - - // TODO: Binary search? Perfect-Hash-Function? - assert( !branches.empty() ); - for(const auto& branch : branches) - { - auto next_bb = (&branch == &branches.back() ? default_bb : m_builder.new_bb_unlinked()); - - auto cmp_gt_bb = m_builder.new_bb_unlinked(); - - auto lt_val = push_compare(sp, val.clone(), ::MIR::eBinOp::LT, ::MIR::Constant(branch.first) ); - m_builder.end_block( ::MIR::Terminator::make_If({ mv$(lt_val), default_bb, cmp_gt_bb }) ); - m_builder.set_cur_block(cmp_gt_bb); - - auto eq_bb = m_builder.new_bb_unlinked(); - auto gt_val = push_compare(sp, val.clone(), ::MIR::eBinOp::GT, ::MIR::Constant(branch.first) ); - m_builder.end_block( ::MIR::Terminator::make_If({ mv$(gt_val), next_bb, eq_bb }) ); - m_builder.set_cur_block(eq_bb); - - this->generate_branch(branch.second, and_then); - - m_builder.set_cur_block(next_bb); - } - this->generate_branch(default_branch, and_then); + m_builder.set_cur_block(decison_arm); + m_builder.end_block( ::MIR::Terminator::make_Switch({ mv$(val), mv$(arms) }) ); } -void DecisionTreeGen::generate_branches_Enum( - const Span& sp, - const DecisionTreeNode::Branch& default_branch, - const DecisionTreeNode::Values::Data_Variant& branches, - const field_path_t& field_path, - const ::HIR::TypeRef& ty, ::MIR::LValue val, - ::std::function<void(const DecisionTreeNode&)> and_then - ) +void MatchGenGrouped::gen_dispatch__slice(::HIR::TypeRef ty, ::MIR::LValue val, const ::std::vector<t_rules_subset>& rules, size_t ofs, const ::std::vector<::MIR::BasicBlockId>& arm_targets, ::MIR::BasicBlockId def_blk) { - const auto& enum_ref = *ty.m_data.as_Path().binding.as_Enum(); - const auto& enum_path = ty.m_data.as_Path().path.m_data.as_Generic(); - const auto& variants = enum_ref.m_variants; - auto variant_count = variants.size(); - bool has_any = ! default_branch.is_Unset(); - - if( branches.size() < variant_count && ! has_any ) { - ERROR(sp, E0000, "Non-exhaustive match over " << ty << " - " << branches.size() << " out of " << variant_count << " present"); - } - // DISABLED: Some complex matches don't directly use some defaults - //if( branches.size() == variant_count && has_any ) { - // ERROR(sp, E0000, "Unreachable _ arm - " << branches.size() << " variants in " << enum_path); - //} + auto val_len = m_builder.lvalue_or_temp(sp, ::HIR::CoreType::Usize, ::MIR::RValue::make_DstMeta({ m_builder.get_ptr_to_dst(sp, val).clone() })); - auto any_block = (has_any ? m_builder.new_bb_unlinked() : 0); + // TODO: Re-sort the rules list to interleve Constant::Bytes and Slice - // Emit a switch over the variant - ::std::vector< ::MIR::BasicBlockId> variant_blocks; - variant_blocks.reserve( variant_count ); - for( const auto& branch : branches ) - { - if( variant_blocks.size() != branch.first ) { - assert( variant_blocks.size() < branch.first ); - assert( has_any ); - variant_blocks.resize( branch.first, any_block ); - } - variant_blocks.push_back( m_builder.new_bb_unlinked() ); - } - if( variant_blocks.size() != variant_count ) + // Just needs to check the lengths, then dispatch. + size_t tgt_ofs = 0; + for(size_t i = 0; i < rules.size(); i++) { - ASSERT_BUG(sp, variant_blocks.size() < variant_count, "Branch count (" << variant_blocks.size() << ") > variant count (" << variant_count << ") in match of " << ty); - ASSERT_BUG(sp, has_any, "Non-exhaustive match and no any arm"); - variant_blocks.resize( variant_count, any_block ); - } - bool any_arm_used = ::std::any_of( variant_blocks.begin(), variant_blocks.end(), [any_block](const auto& blk){ return blk == any_block; } ); + const auto& r = rules[i][0][ofs]; + if(const auto* re = r.opt_Slice()) + { + ASSERT_BUG(sp, re->sub_rules.size() == 0, "Sub-rules in MatchGenGrouped"); + auto val_tst = ::MIR::Constant::make_Uint({ re->len, ::HIR::CoreType::Usize }); - m_builder.end_block( ::MIR::Terminator::make_Switch({ - val.clone(), variant_blocks // NOTE: Copies the list, so it can be used lower down - }) ); + for(size_t j = 0; j < rules[i].size(); j ++) + assert(arm_targets[tgt_ofs] == arm_targets[tgt_ofs+j]); - // Emit sub-patterns, looping over variants - for( const auto& branch : branches ) - { - auto bb = variant_blocks[branch.first]; - const auto& var = variants[branch.first]; - DEBUG(branch.first << " " << var.first << " = " << branch.second); + // IF v < tst : target + if( re->len > 0 ) + { + auto cmp_eq_blk = m_builder.new_bb_unlinked(); + auto cmp_lval_lt = this->push_compare( val_len.clone(), ::MIR::eBinOp::LT, val_tst.clone() ); + m_builder.end_block( ::MIR::Terminator::make_If({ mv$(cmp_lval_lt), def_blk, cmp_eq_blk }) ); + m_builder.set_cur_block(cmp_eq_blk); + } - auto var_lval = ::MIR::LValue::make_Downcast({ box$(val.clone()), branch.first }); + // IF v == tst : target + { + auto next_cmp_blk = m_builder.new_bb_unlinked(); + auto cmp_lval_eq = this->push_compare( val_len.clone(), ::MIR::eBinOp::EQ, mv$(val_tst) ); + m_builder.end_block( ::MIR::Terminator::make_If({ mv$(cmp_lval_eq), arm_targets[tgt_ofs], next_cmp_blk }) ); + m_builder.set_cur_block(next_cmp_blk); + } + } + else if(const auto* re = r.opt_Value()) + { + ASSERT_BUG(sp, re->is_Bytes(), "Slice with non-Bytes value - " << *re); + const auto& b = re->as_Bytes(); - ::HIR::TypeRef fake_ty; + auto val_tst = ::MIR::Constant::make_Uint({ b.size(), ::HIR::CoreType::Usize }); + auto cmp_slice_val = m_builder.lvalue_or_temp(sp, + ::HIR::TypeRef::new_borrow( ::HIR::BorrowType::Shared, ::HIR::TypeRef::new_slice(::HIR::CoreType::U8) ), + ::MIR::RValue::make_MakeDst({ ::MIR::Param(re->clone()), val_tst.clone() }) + ); - TU_MATCHA( (var.second), (e), - (Unit, - DEBUG("- Unit"); - ), - (Value, - DEBUG("- Value"); - ), - (Tuple, - // Make a fake tuple - ::std::vector< ::HIR::TypeRef> ents; - for( const auto& fld : e ) + if( b.size() > 0 ) { - ents.push_back( monomorphise_type(sp, enum_ref.m_params, enum_path.m_params, fld.ent) ); + auto cmp_eq_blk = m_builder.new_bb_unlinked(); + auto cmp_lval_lt = this->push_compare( val_len.clone(), ::MIR::eBinOp::LT, val_tst.clone() ); + m_builder.end_block( ::MIR::Terminator::make_If({ mv$(cmp_lval_lt), def_blk, cmp_eq_blk }) ); + m_builder.set_cur_block(cmp_eq_blk); } - fake_ty = ::HIR::TypeRef( mv$(ents) ); - m_builder.resolve().expand_associated_types(sp, fake_ty); - DEBUG("- Tuple - " << fake_ty); - ), - (Struct, - ::std::vector< ::HIR::TypeRef> ents; - for( const auto& fld : e ) + + // IF v == tst : target { - ents.push_back( monomorphise_type(sp, enum_ref.m_params, enum_path.m_params, fld.second.ent) ); - } - fake_ty = ::HIR::TypeRef( mv$(ents) ); - m_builder.resolve().expand_associated_types(sp, fake_ty); - DEBUG("- Struct - " << fake_ty); - ) - ) + auto succ_blk = m_builder.new_bb_unlinked(); + auto next_cmp_blk = m_builder.new_bb_unlinked(); + auto cmp_lval_eq = this->push_compare( val_len.clone(), ::MIR::eBinOp::EQ, mv$(val_tst) ); + m_builder.end_block( ::MIR::Terminator::make_If({ mv$(cmp_lval_eq), succ_blk, next_cmp_blk }) ); + m_builder.set_cur_block(succ_blk); - m_builder.set_cur_block( bb ); - if( fake_ty == ::HIR::TypeRef() || fake_ty.m_data.as_Tuple().size() == 0 ) { - this->generate_branch(branch.second, and_then); - } - else { - this->generate_branch(branch.second, [&](auto& subnode) { - // Call special handler to determine when the enum is over - this->generate_tree_code__enum(sp, subnode, fake_ty, var_lval, field_path, and_then); - }); - } - } + // TODO: What if `val` isn't a Deref? + ASSERT_BUG(sp, val.is_Deref(), "TODO: Handle non-Deref matches of byte strings"); + cmp_lval_eq = this->push_compare( val.as_Deref().val->clone(), ::MIR::eBinOp::EQ, mv$(cmp_slice_val) ); + m_builder.end_block( ::MIR::Terminator::make_If({ mv$(cmp_lval_eq), arm_targets[tgt_ofs], def_blk }) ); - if( any_arm_used ) - { - DEBUG("_ = " << default_branch); - if( !default_branch.is_Unset() ) + m_builder.set_cur_block(next_cmp_blk); + } + } + else { - m_builder.set_cur_block(any_block); - this->generate_branch(default_branch, and_then); + BUG(sp, "Matching without _Slice pattern - " << r.tag_str() << " - " << r); } - } - else - { - DEBUG("_ = UNUSED - " << default_branch); - } -} -void DecisionTreeGen::generate_branches_Slice( - const Span& sp, - const DecisionTreeNode::Branch& default_branch, - const DecisionTreeNode::Values::Data_Slice& branches, - const field_path_t& field_path, - const ::HIR::TypeRef& ty, ::MIR::LValue val, - ::std::function<void(const DecisionTreeNode&)> and_then - ) -{ - if( default_branch.is_Unset() ) { - ERROR(sp, E0000, "Non-exhaustive match over " << ty); + tgt_ofs += rules[i].size(); } + m_builder.end_block( ::MIR::Terminator::make_Goto(def_blk) ); +} - auto val_len = m_builder.lvalue_or_temp(sp, ::HIR::CoreType::Usize, ::MIR::RValue::make_DstMeta({ m_builder.get_ptr_to_dst(sp, val).clone() })); - - // NOTE: Un-deref the slice - ASSERT_BUG(sp, val.is_Deref(), "slice matches must be passed a deref"); - auto tmp = mv$( *val.as_Deref().val ); - val = mv$(tmp); - - auto any_block = m_builder.new_bb_unlinked(); - // TODO: Select one of three ways of picking the arm: - // - Integer switch (unimplemented) - // - Binary search - // - Sequential comparisons +void MatchGenGrouped::gen_dispatch_range(const field_path_t& field_path, const ::MIR::Constant& first, const ::MIR::Constant& last, ::MIR::BasicBlockId def_blk) +{ + TRACE_FUNCTION_F("field_path="<<field_path<<", " << first << " ... " << last); + ::MIR::LValue val; + ::HIR::TypeRef ty; + get_ty_and_val(sp, m_builder.resolve(), m_top_ty, m_top_val, field_path, m_field_path_ofs, ty, val); + DEBUG("ty = " << ty << ", val = " << val); - // TODO: Binary search instead? - for( const auto& branch : branches.fixed_arms ) + if( const auto* tep = ty.m_data.opt_Primitive() ) { - auto val_des = ::MIR::Constant::make_Uint({ static_cast<uint64_t>(branch.first), ::HIR::CoreType::Usize }); - - // Special case - final just does equality - if( &branch == &branches.fixed_arms.back() ) - { - auto val_cmp_eq = push_compare(sp, val_len.clone(), ::MIR::eBinOp::EQ, mv$(val_des)); - - auto success_block = m_builder.new_bb_unlinked(); - m_builder.end_block( ::MIR::Terminator::make_If({ mv$(val_cmp_eq), success_block, any_block }) ); + auto te = *tep; - m_builder.set_cur_block( success_block ); - this->generate_branch(branch.second, and_then); + bool lower_possible = true; + bool upper_possible = true; - m_builder.set_cur_block( any_block ); - } - // Special case for zero (which can't have a LT) - else if( branch.first == 0 ) + switch(te) { - auto next_block = m_builder.new_bb_unlinked(); - auto val_cmp_eq = push_compare(sp, val_len.clone(), ::MIR::eBinOp::EQ, mv$(val_des)); - - auto success_block = m_builder.new_bb_unlinked(); - m_builder.end_block( ::MIR::Terminator::make_If({ mv$(val_cmp_eq), success_block, next_block }) ); + case ::HIR::CoreType::Bool: + BUG(sp, "Range match over Bool"); + break; + case ::HIR::CoreType::Str: + BUG(sp, "Range match over Str - is this valid?"); + break; + case ::HIR::CoreType::U8: + case ::HIR::CoreType::U16: + case ::HIR::CoreType::U32: + case ::HIR::CoreType::U64: + case ::HIR::CoreType::U128: + case ::HIR::CoreType::Usize: + lower_possible = (first.as_Uint().v > 0); + // TODO: Should this also check for the end being the max value of the type? + // - Can just leave that to the optimiser + upper_possible = true; + break; + case ::HIR::CoreType::I8: + case ::HIR::CoreType::I16: + case ::HIR::CoreType::I32: + case ::HIR::CoreType::I64: + case ::HIR::CoreType::I128: + case ::HIR::CoreType::Isize: + lower_possible = true; + upper_possible = true; + break; + case ::HIR::CoreType::Char: + lower_possible = (first.as_Uint().v > 0); + upper_possible = (first.as_Uint().v <= 0x10FFFF); + break; + case ::HIR::CoreType::F32: + case ::HIR::CoreType::F64: + // NOTE: No upper or lower limits + break; + } - m_builder.set_cur_block( success_block ); - this->generate_branch(branch.second, and_then); + if( lower_possible ) + { + auto test_bb_2 = m_builder.new_bb_unlinked(); + // IF `val` < `first` : fail_bb + auto cmp_lt_lval = m_builder.lvalue_or_temp(sp, ::HIR::CoreType::Bool, ::MIR::RValue::make_BinOp({ ::MIR::Param(val.clone()), ::MIR::eBinOp::LT, ::MIR::Param(first.clone()) })); + m_builder.end_block( ::MIR::Terminator::make_If({ mv$(cmp_lt_lval), def_blk, test_bb_2 }) ); - m_builder.set_cur_block( next_block ); + m_builder.set_cur_block(test_bb_2); } - // General case, with two comparisons - else - { - auto next_block = m_builder.new_bb_unlinked(); - auto cmp_gt_block = m_builder.new_bb_unlinked(); - auto val_cmp_lt = push_compare(sp, val_len.clone(), ::MIR::eBinOp::LT, val_des.clone()); - m_builder.end_block( ::MIR::Terminator::make_If({ mv$(val_cmp_lt), any_block, cmp_gt_block }) ); - m_builder.set_cur_block( cmp_gt_block ); - auto success_block = m_builder.new_bb_unlinked(); - auto val_cmp_gt = push_compare(sp, val_len.clone(), ::MIR::eBinOp::GT, mv$(val_des)); - m_builder.end_block( ::MIR::Terminator::make_If({ mv$(val_cmp_gt), next_block, success_block }) ); + if( upper_possible ) + { + auto succ_bb = m_builder.new_bb_unlinked(); - m_builder.set_cur_block( success_block ); - this->generate_branch(branch.second, and_then); + // IF `val` > `last` : fail_bb + auto cmp_gt_lval = m_builder.lvalue_or_temp(sp, ::HIR::CoreType::Bool, ::MIR::RValue::make_BinOp({ ::MIR::Param(val.clone()), ::MIR::eBinOp::GT, ::MIR::Param(last.clone()) })); + m_builder.end_block( ::MIR::Terminator::make_If({ mv$(cmp_gt_lval), def_blk, succ_bb }) ); - m_builder.set_cur_block( next_block ); + m_builder.set_cur_block(succ_bb); } } - assert( m_builder.block_active() ); - - if( default_branch.is_Unset() ) { - // TODO: Emit error if non-exhaustive - m_builder.end_block( ::MIR::Terminator::make_Diverge({}) ); - } - else { - this->generate_branch(default_branch, and_then); + else + { + TODO(sp, "ValueRange on " << ty); } } +void MatchGenGrouped::gen_dispatch_splitslice(const field_path_t& field_path, const PatternRule::Data_SplitSlice& e, ::MIR::BasicBlockId def_blk) +{ + TRACE_FUNCTION_F("field_path="<<field_path<<", [" << e.leading << ", .., " << e.trailing << "]"); + ::MIR::LValue val; + ::HIR::TypeRef ty; + get_ty_and_val(sp, m_builder.resolve(), m_top_ty, m_top_val, field_path, m_field_path_ofs, ty, val); + DEBUG("ty = " << ty << ", val = " << val); -namespace { - bool path_starts_with(const field_path_t& test, const field_path_t& prefix) + ASSERT_BUG(sp, e.leading.size() == 0, "Sub-rules in MatchGenGrouped"); + ASSERT_BUG(sp, e.trailing.size() == 0, "Sub-rules in MatchGenGrouped"); + ASSERT_BUG(sp, ty.m_data.is_Slice(), "SplitSlice pattern on non-slice - " << ty); + + // Obtain slice length + auto val_len = m_builder.lvalue_or_temp(sp, ::HIR::CoreType::Usize, ::MIR::RValue::make_DstMeta({ m_builder.get_ptr_to_dst(sp, val).clone() })); + + // 1. Check that length is sufficient for the pattern to be used + // `IF len < min_len : def_blk, next { - //DEBUG("test="<<test<<", prefix="<<prefix); - if( test.size() < prefix.size() ) - { - return false; - } - else if( ! ::std::equal(prefix.data.begin(), prefix.data.end(), test.data.begin()) ) - { - return false; - } - else - { - return true; - } + auto next = m_builder.new_bb_unlinked(); + auto cmp_val = this->push_compare(val_len.clone(), ::MIR::eBinOp::LT, ::MIR::Constant::make_Uint({ e.min_len, ::HIR::CoreType::Usize })); + m_builder.end_block( ::MIR::Terminator::make_If({ mv$(cmp_val), def_blk, next }) ); + m_builder.set_cur_block(next); } -} -void DecisionTreeGen::generate_tree_code__enum( - const Span& sp, - const DecisionTreeNode& node, const ::HIR::TypeRef& fake_ty, const ::MIR::LValue& val, - const field_path_t& path_prefix, - ::std::function<void(const DecisionTreeNode&)> and_then - ) -{ - if( ! path_starts_with(node.m_field_path, path_prefix) ) + // 2. Recurse into leading patterns. + if( e.min_len > e.trailing_len ) { - and_then(node); + auto next = m_builder.new_bb_unlinked(); + auto inner_set = t_rules_subset { 1, /*is_arm_indexes=*/false }; + inner_set.push_bb( e.leading, next ); + auto inst = MatchGenGrouped { m_builder, sp, ty, val, {}, field_path.size() }; + inst.gen_for_slice(inner_set, 0, def_blk); + + m_builder.set_cur_block(next); } - else + + if( e.trailing_len != 0 ) { - this->generate_tree_code(sp, node, fake_ty, path_prefix.size(), val, - [&](const auto& next_node) { - if( ! path_starts_with(next_node.m_field_path, path_prefix) ) - { - and_then(next_node); - } - else - { - this->generate_tree_code__enum(sp, next_node, fake_ty, val, path_prefix, and_then); - } - }); + TODO(sp, "Handle trailing rules in SplitSlice - " << e.trailing); } } diff --git a/src/mir/helpers.cpp b/src/mir/helpers.cpp index 9c2107ab..9242ccb7 100644 --- a/src/mir/helpers.cpp +++ b/src/mir/helpers.cpp @@ -10,6 +10,7 @@ #include <hir/hir.hpp> #include <hir/type.hpp> #include <mir/mir.hpp> +#include <algorithm> // ::std::find void ::MIR::TypeResolve::fmt_pos(::std::ostream& os) const { @@ -33,6 +34,13 @@ void ::MIR::TypeResolve::print_msg(const char* tag, ::std::function<void(::std:: //throw CheckFailure {}; } +unsigned int ::MIR::TypeResolve::get_cur_stmt_ofs() const +{ + if( this->stmt_idx == STMT_TERM ) + return m_fcn.blocks.at(this->bb_idx).statements.size(); + else + return this->stmt_idx; +} const ::MIR::BasicBlock& ::MIR::TypeResolve::get_block(::MIR::BasicBlockId id) const { MIR_ASSERT(*this, id < m_fcn.blocks.size(), "Block ID " << id << " out of range"); @@ -252,14 +260,17 @@ const ::HIR::TypeRef& ::MIR::TypeResolve::get_lvalue_type(::HIR::TypeRef& tmp, c auto v = m_resolve.get_value(this->sp, e.p, p, /*signature_only=*/true); if( const auto* ve = v.opt_Constant() ) { const auto& ty = (*ve)->m_type; - MIR_TODO(*this, "Monomorphise type " << ty); + if( monomorphise_type_needed(ty) ) + MIR_TODO(*this, "get_const_type - Monomorphise type " << ty); + else + return ty.clone(); } else { - MIR_BUG(*this, ""); + MIR_BUG(*this, "get_const_type - Not a constant"); } ), (ItemAddr, - MIR_TODO(*this, "Get type for constant `" << c << "`"); + MIR_TODO(*this, "get_const_type - Get type for constant `" << c << "`"); ) ) throw ""; @@ -268,3 +279,1287 @@ const ::HIR::TypeRef* ::MIR::TypeResolve::is_type_owned_box(const ::HIR::TypeRef { return m_resolve.is_type_owned_box(ty); } + +using namespace MIR::visit; + +// -------------------------------------------------------------------- +// MIR_Helper_GetLifetimes +// -------------------------------------------------------------------- +namespace MIR { +namespace visit { + bool visit_mir_lvalue(const ::MIR::LValue& lv, ValUsage u, ::std::function<bool(const ::MIR::LValue& , ValUsage)> cb) + { + if( cb(lv, u) ) + return true; + TU_MATCHA( (lv), (e), + (Variable, + ), + (Argument, + ), + (Temporary, + ), + (Static, + ), + (Return, + ), + (Field, + return visit_mir_lvalue(*e.val, u, cb); + ), + (Deref, + return visit_mir_lvalue(*e.val, ValUsage::Read, cb); + ), + (Index, + bool rv = false; + rv |= visit_mir_lvalue(*e.val, u, cb); + rv |= visit_mir_lvalue(*e.idx, ValUsage::Read, cb); + return rv; + ), + (Downcast, + return visit_mir_lvalue(*e.val, u, cb); + ) + ) + return false; + } + + bool visit_mir_lvalue(const ::MIR::Param& p, ValUsage u, ::std::function<bool(const ::MIR::LValue& , ValUsage)> cb) + { + if( const auto* e = p.opt_LValue() ) + { + if(cb(*e, ValUsage::Move)) + return true; + return visit_mir_lvalue(*e, u, cb); + } + else + { + return false; + } + } + + bool visit_mir_lvalues(const ::MIR::RValue& rval, ::std::function<bool(const ::MIR::LValue& , ValUsage)> cb) + { + bool rv = false; + TU_MATCHA( (rval), (se), + (Use, + if(cb(se, ValUsage::Move)) + return true; + rv |= visit_mir_lvalue(se, ValUsage::Read, cb); + ), + (Constant, + ), + (SizedArray, + rv |= visit_mir_lvalue(se.val, ValUsage::Read, cb); + ), + (Borrow, + rv |= visit_mir_lvalue(se.val, ValUsage::Borrow, cb); + ), + (Cast, + rv |= visit_mir_lvalue(se.val, ValUsage::Read, cb); + ), + (BinOp, + rv |= visit_mir_lvalue(se.val_l, ValUsage::Read, cb); + rv |= visit_mir_lvalue(se.val_r, ValUsage::Read, cb); + ), + (UniOp, + rv |= visit_mir_lvalue(se.val, ValUsage::Read, cb); + ), + (DstMeta, + rv |= visit_mir_lvalue(se.val, ValUsage::Read, cb); + ), + (DstPtr, + rv |= visit_mir_lvalue(se.val, ValUsage::Read, cb); + ), + (MakeDst, + rv |= visit_mir_lvalue(se.ptr_val, ValUsage::Read, cb); + rv |= visit_mir_lvalue(se.meta_val, ValUsage::Read, cb); + ), + (Tuple, + for(auto& v : se.vals) + rv |= visit_mir_lvalue(v, ValUsage::Read, cb); + ), + (Array, + for(auto& v : se.vals) + rv |= visit_mir_lvalue(v, ValUsage::Read, cb); + ), + (Variant, + rv |= visit_mir_lvalue(se.val, ValUsage::Read, cb); + ), + (Struct, + for(auto& v : se.vals) + rv |= visit_mir_lvalue(v, ValUsage::Read, cb); + ) + ) + return rv; + } + + bool visit_mir_lvalues(const ::MIR::Statement& stmt, ::std::function<bool(const ::MIR::LValue& , ValUsage)> cb) + { + bool rv = false; + TU_MATCHA( (stmt), (e), + (Assign, + rv |= visit_mir_lvalues(e.src, cb); + rv |= visit_mir_lvalue(e.dst, ValUsage::Write, cb); + ), + (Asm, + for(auto& v : e.inputs) + rv |= visit_mir_lvalue(v.second, ValUsage::Read, cb); + for(auto& v : e.outputs) + rv |= visit_mir_lvalue(v.second, ValUsage::Write, cb); + ), + (SetDropFlag, + ), + (Drop, + rv |= visit_mir_lvalue(e.slot, ValUsage::Move, cb); + ), + (ScopeEnd, + ) + ) + return rv; + } + + bool visit_mir_lvalues(const ::MIR::Terminator& term, ::std::function<bool(const ::MIR::LValue& , ValUsage)> cb) + { + bool rv = false; + TU_MATCHA( (term), (e), + (Incomplete, + ), + (Return, + ), + (Diverge, + ), + (Goto, + ), + (Panic, + ), + (If, + rv |= visit_mir_lvalue(e.cond, ValUsage::Read, cb); + ), + (Switch, + rv |= visit_mir_lvalue(e.val, ValUsage::Read, cb); + ), + (Call, + if( e.fcn.is_Value() ) { + rv |= visit_mir_lvalue(e.fcn.as_Value(), ValUsage::Read, cb); + } + for(auto& v : e.args) + rv |= visit_mir_lvalue(v, ValUsage::Read, cb); + rv |= visit_mir_lvalue(e.ret_val, ValUsage::Write, cb); + ) + ) + return rv; + } + /* + void visit_mir_lvalues_mut(::MIR::TypeResolve& state, ::MIR::Function& fcn, ::std::function<bool(::MIR::LValue& , ValUsage)> cb) + { + for(unsigned int block_idx = 0; block_idx < fcn.blocks.size(); block_idx ++) + { + auto& block = fcn.blocks[block_idx]; + for(auto& stmt : block.statements) + { + state.set_cur_stmt(block_idx, (&stmt - &block.statements.front())); + visit_mir_lvalues_mut(stmt, cb); + } + if( block.terminator.tag() == ::MIR::Terminator::TAGDEAD ) + continue ; + state.set_cur_stmt_term(block_idx); + visit_mir_lvalues_mut(block.terminator, cb); + } + } + void visit_mir_lvalues(::MIR::TypeResolve& state, const ::MIR::Function& fcn, ::std::function<bool(const ::MIR::LValue& , ValUsage)> cb) + { + visit_mir_lvalues_mut(state, const_cast<::MIR::Function&>(fcn), [&](auto& lv, auto im){ return cb(lv, im); }); + } + */ +} // namespace visit +} // namespace MIR +namespace +{ + struct ValueLifetime + { + ::std::vector<bool> stmt_bitmap; + ValueLifetime(size_t stmt_count): + stmt_bitmap(stmt_count) + {} + + void fill(const ::std::vector<size_t>& block_offsets, size_t bb, size_t first_stmt, size_t last_stmt) + { + size_t limit = block_offsets[bb+1] - block_offsets[bb] - 1; + DEBUG("bb" << bb << " : " << first_stmt << "--" << last_stmt); + assert(first_stmt <= limit); + assert(last_stmt <= limit); + for(size_t stmt = first_stmt; stmt <= last_stmt; stmt++) + { + stmt_bitmap[block_offsets[bb] + stmt] = true; + } + } + + void dump_debug(const char* suffix, unsigned i, const ::std::vector<size_t>& block_offsets) + { + ::std::string name = FMT(suffix << "$" << i); + while(name.size() < 3+1+3) + name += " "; + DEBUG(name << " : " << FMT_CB(os, + for(unsigned int j = 0; j < this->stmt_bitmap.size(); j++) + { + if(j != 0 && ::std::find(block_offsets.begin(), block_offsets.end(), j) != block_offsets.end()) + os << "|"; + os << (this->stmt_bitmap[j] ? "X" : " "); + } + )); + } + }; +} +#if 1 +void MIR_Helper_GetLifetimes_DetermineValueLifetime(::MIR::TypeResolve& state, const ::MIR::Function& fcn, size_t bb_idx, size_t stmt_idx, const ::MIR::LValue& lv, const ::std::vector<size_t>& block_offsets, ValueLifetime& vl); + +::MIR::ValueLifetimes MIR_Helper_GetLifetimes(::MIR::TypeResolve& state, const ::MIR::Function& fcn, bool dump_debug) +{ + TRACE_FUNCTION_F(state); + + size_t statement_count = 0; + ::std::vector<size_t> block_offsets; + block_offsets.reserve( fcn.blocks.size() ); + for(const auto& bb : fcn.blocks) + { + block_offsets.push_back(statement_count); + statement_count += bb.statements.size() + 1; // +1 for the terminator + } + block_offsets.push_back(statement_count); // Store the final limit for later code to use. + + ::std::vector<ValueLifetime> temporary_lifetimes( fcn.temporaries.size(), ValueLifetime(statement_count) ); + ::std::vector<ValueLifetime> variable_lifetimes( fcn.named_variables.size(), ValueLifetime(statement_count) ); + + + // Enumerate direct assignments of variables (linear iteration of BB list) + for(size_t bb_idx = 0; bb_idx < fcn.blocks.size(); bb_idx ++) + { + auto assigned_lvalue = [&](size_t bb_idx, size_t stmt_idx, const ::MIR::LValue& lv) { + // NOTE: Fills the first statement after running, just to ensure that any assigned value has _a_ lifetime + if( const auto* de = lv.opt_Variable() ) + { + MIR_Helper_GetLifetimes_DetermineValueLifetime(state, fcn, bb_idx, stmt_idx, lv, block_offsets, variable_lifetimes[*de]); + variable_lifetimes[*de].fill(block_offsets, bb_idx, stmt_idx, stmt_idx); + } + else if( const auto* de = lv.opt_Temporary() ) + { + MIR_Helper_GetLifetimes_DetermineValueLifetime(state, fcn, bb_idx, stmt_idx, lv, block_offsets, temporary_lifetimes[de->idx]); + temporary_lifetimes[de->idx].fill(block_offsets, bb_idx, stmt_idx, stmt_idx); + } + else + { + // Not a direct assignment of a slot + } + }; + + const auto& bb = fcn.blocks[bb_idx]; + for(size_t stmt_idx = 0; stmt_idx < bb.statements.size(); stmt_idx ++) + { + state.set_cur_stmt(bb_idx, stmt_idx); + const auto& stmt = bb.statements[stmt_idx]; + if( const auto* se = stmt.opt_Assign() ) + { + // For assigned variables, determine how long that value will live + assigned_lvalue(bb_idx, stmt_idx+1, se->dst); + } + else if( const auto* se = stmt.opt_Asm() ) + { + for(const auto& e : se->outputs) + { + assigned_lvalue(bb_idx, stmt_idx+1, e.second); + } + } + } + state.set_cur_stmt_term(bb_idx); + + // Only Call can assign a value + TU_IFLET(::MIR::Terminator, bb.terminator, Call, te, + assigned_lvalue(te.ret_block, 0, te.ret_val); + ) + } + + // Dump out variable lifetimes. + if( dump_debug ) + { + for(unsigned int i = 0; i < temporary_lifetimes.size(); i ++) + { + temporary_lifetimes[i].dump_debug("tmp", i, block_offsets); + } + for(unsigned int i = 0; i < variable_lifetimes.size(); i ++) + { + variable_lifetimes[i].dump_debug("var", i, block_offsets); + } + } + + + ::MIR::ValueLifetimes rv; + rv.m_block_offsets = mv$(block_offsets); + rv.m_temporaries.reserve( temporary_lifetimes.size() ); + for(auto& lft : temporary_lifetimes) + rv.m_temporaries.push_back( ::MIR::ValueLifetime(mv$(lft.stmt_bitmap)) ); + rv.m_variables.reserve( variable_lifetimes.size() ); + for(auto& lft : variable_lifetimes) + rv.m_variables.push_back( ::MIR::ValueLifetime(mv$(lft.stmt_bitmap)) ); + return rv; +} +void MIR_Helper_GetLifetimes_DetermineValueLifetime( + ::MIR::TypeResolve& mir_res, const ::MIR::Function& fcn, + size_t bb_idx, size_t stmt_idx, // First statement in which the value is valid (after the assignment) + const ::MIR::LValue& lv, const ::std::vector<size_t>& block_offsets, ValueLifetime& vl + ) +{ + TRACE_FUNCTION_F(mir_res << " " << lv); + // Walk the BB tree until: + // - Loopback + // - Assignment + // - Drop + + struct State + { + const ::std::vector<size_t>& m_block_offsets; + ValueLifetime& m_out_vl; + + ::std::vector<unsigned int> bb_history; + size_t last_read_ofs; // Statement index + bool m_is_borrowed; + + State(const ::std::vector<size_t>& block_offsets, ValueLifetime& vl, size_t init_bb_idx, size_t init_stmt_idx): + m_block_offsets(block_offsets), + m_out_vl(vl), + bb_history(), + last_read_ofs(init_stmt_idx), + m_is_borrowed(false) + { + bb_history.push_back(init_bb_idx); + } + State(State&& x): + m_block_offsets(x.m_block_offsets), + m_out_vl(x.m_out_vl), + bb_history( mv$(x.bb_history) ), + last_read_ofs( x.last_read_ofs ), + m_is_borrowed( x.m_is_borrowed ) + { + } + State& operator=(State&& x) { + this->bb_history = mv$(x.bb_history); + this->last_read_ofs = x.last_read_ofs; + this->m_is_borrowed = x.m_is_borrowed; + return *this; + } + + State clone() const { + State rv { m_block_offsets, m_out_vl, 0, last_read_ofs }; + rv.bb_history = bb_history; + rv.m_is_borrowed = m_is_borrowed; + return rv; + } + + // Returns true if the variable has been borrowed + bool is_borrowed() const { + return this->m_is_borrowed; + } + + void mark_borrowed(size_t stmt_idx) { + if( ! m_is_borrowed ) + { + m_is_borrowed = false; + this->fill_to(stmt_idx); + } + m_is_borrowed = true; + } + void mark_read(size_t stmt_idx) { + if( !m_is_borrowed ) + { + this->fill_to(stmt_idx); + } + else + { + m_is_borrowed = false; + this->fill_to(stmt_idx); + m_is_borrowed = true; + } + } + void fmt(::std::ostream& os) const { + os << "BB" << bb_history.front() << "/" << last_read_ofs << "--"; + os << "[" << bb_history << "]"; + } + void finalise(size_t stmt_idx) + { + if( m_is_borrowed ) + { + m_is_borrowed = false; + this->fill_to(stmt_idx); + m_is_borrowed = true; + } + } + private: + void fill_to(size_t stmt_idx) + { + TRACE_FUNCTION_F(FMT_CB(ss, this->fmt(ss);)); + assert( !m_is_borrowed ); + assert(bb_history.size() > 0); + if( bb_history.size() == 1 ) + { + // only one block + m_out_vl.fill(m_block_offsets, bb_history[0], last_read_ofs, stmt_idx); + } + else + { + // First block. + auto init_bb_idx = bb_history[0]; + auto limit_0 = m_block_offsets[init_bb_idx+1] - m_block_offsets[init_bb_idx] - 1; + m_out_vl.fill(m_block_offsets, init_bb_idx, last_read_ofs, limit_0); + + // Middle blocks + for(size_t i = 1; i < bb_history.size()-1; i++) + { + size_t bb_idx = bb_history[i]; + assert(bb_idx+1 < m_block_offsets.size()); + size_t limit = m_block_offsets[bb_idx+1] - m_block_offsets[bb_idx] - 1; + m_out_vl.fill(m_block_offsets, bb_idx, 0, limit); + } + + // Last block + auto bb_idx = bb_history.back(); + m_out_vl.fill(m_block_offsets, bb_idx, 0, stmt_idx); + } + + last_read_ofs = stmt_idx; + + auto cur = this->bb_history.back(); + this->bb_history.clear(); + this->bb_history.push_back(cur); + } + }; + + struct Runner + { + ::MIR::TypeResolve& m_mir_res; + const ::MIR::Function& m_fcn; + size_t m_init_bb_idx; + size_t m_init_stmt_idx; + const ::MIR::LValue& m_lv; + const ::std::vector<size_t>& m_block_offsets; + ValueLifetime& m_lifetimes; + bool m_is_copy; + + ::std::vector<bool> m_visited_statements; + + ::std::vector<::std::pair<size_t, State>> m_states_to_do; + + Runner(::MIR::TypeResolve& mir_res, const ::MIR::Function& fcn, size_t init_bb_idx, size_t init_stmt_idx, const ::MIR::LValue& lv, const ::std::vector<size_t>& block_offsets, ValueLifetime& vl): + m_mir_res(mir_res), + m_fcn(fcn), + m_init_bb_idx(init_bb_idx), + m_init_stmt_idx(init_stmt_idx), + m_lv(lv), + m_block_offsets(block_offsets), + m_lifetimes(vl), + + m_visited_statements( m_lifetimes.stmt_bitmap.size() ) + { + ::HIR::TypeRef tmp; + m_is_copy = m_mir_res.m_resolve.type_is_copy(mir_res.sp, m_mir_res.get_lvalue_type(tmp, lv)); + } + + void run_block(size_t bb_idx, size_t stmt_idx, State state) + { + const auto& bb = m_fcn.blocks.at(bb_idx); + assert(stmt_idx <= bb.statements.size()); + + bool was_moved = false; + bool was_updated = false; + auto visit_cb = [&](const auto& lv, auto vu) { + if(lv == m_lv) { + if( vu == ValUsage::Read ) { + DEBUG(m_mir_res << "Used"); + state.mark_read(stmt_idx); + was_updated = true; + } + if( vu == ValUsage::Move ) { + DEBUG(m_mir_res << (m_is_copy ? "Read" : "Moved")); + state.mark_read(stmt_idx); + was_moved = ! m_is_copy; + } + if( vu == ValUsage::Borrow ) { + DEBUG(m_mir_res << "Borrowed"); + state.mark_borrowed(stmt_idx); + was_updated = true; + } + return true; + } + return false; + }; + + for( ; stmt_idx < bb.statements.size(); stmt_idx ++) + { + const auto& stmt = bb.statements[stmt_idx]; + m_mir_res.set_cur_stmt(bb_idx, stmt_idx); + m_visited_statements[ m_block_offsets.at(bb_idx) + stmt_idx ] = true; + + // Visit and see if the value is read (setting the read flag or end depending on if the value is Copy) + visit_mir_lvalues(stmt, visit_cb); + + if( was_moved ) + { + // Moved: Update read position and apply + DEBUG(m_mir_res << "Moved, return"); + state.mark_read(stmt_idx); + state.finalise(stmt_idx); + return ; + } + + TU_MATCHA( (stmt), (se), + (Assign, + if( se.dst == m_lv ) + { + DEBUG(m_mir_res << "- Assigned to, return"); + // Value assigned, just apply + state.finalise(stmt_idx); + return ; + } + ), + (Drop, + visit_mir_lvalue(se.slot, ValUsage::Read, visit_cb); + if( se.slot == m_lv ) + { + // Value dropped, update read position and apply + DEBUG(m_mir_res << "- Dropped, return"); + // - If it was borrowed, it can't still be borrowed here. + // TODO: Enable this once it's known to not cause mis-optimisation. It could currently. + //if( state.is_borrowed() ) { + // state.clear_borrowed(); + //} + state.mark_read(stmt_idx); + state.finalise(stmt_idx); + return ; + } + ), + (Asm, + // + for(const auto& e : se.outputs) + { + if(e.second == m_lv) { + // Assigned, just apply + DEBUG(m_mir_res << "- Assigned (asm!), return"); + state.finalise(stmt_idx); + return ; + } + } + ), + (SetDropFlag, + // Ignore + ), + (ScopeEnd, + // Ignore + ) + ) + } + m_mir_res.set_cur_stmt_term(bb_idx); + m_visited_statements[ m_block_offsets.at(bb_idx) + stmt_idx ] = true; + + visit_mir_lvalues(bb.terminator, visit_cb); + + if( was_moved ) + { + // Moved: Update read position and apply + DEBUG(m_mir_res << "- Moved, return"); + state.mark_read(stmt_idx); + state.finalise(stmt_idx); + return ; + } + + // Terminator + TU_MATCHA( (bb.terminator), (te), + (Incomplete, + // TODO: Isn't this a bug? + DEBUG(m_mir_res << "Incomplete"); + state.finalise(stmt_idx); + ), + (Return, + DEBUG(m_mir_res << "Return"); + state.finalise(stmt_idx); + ), + (Diverge, + DEBUG(m_mir_res << "Diverge"); + state.finalise(stmt_idx); + ), + (Goto, + m_states_to_do.push_back( ::std::make_pair(te, mv$(state)) ); + ), + (Panic, + m_states_to_do.push_back( ::std::make_pair(te.dst, mv$(state)) ); + ), + (If, + m_states_to_do.push_back( ::std::make_pair(te.bb0, state.clone()) ); + m_states_to_do.push_back( ::std::make_pair(te.bb1, mv$(state)) ); + ), + (Switch, + for(size_t i = 0; i < te.targets.size(); i ++) + { + auto s = (i == te.targets.size()-1) + ? mv$(state) + : state.clone(); + m_states_to_do.push_back( ::std::make_pair(te.targets[i], mv$(s)) ); + } + ), + (Call, + if( te.ret_val == m_lv ) + { + DEBUG(m_mir_res << "Assigned (Call), return"); + // Value assigned, just apply + state.finalise(stmt_idx); + return ; + } + if( m_fcn.blocks.at(te.panic_block).statements.size() == 0 && m_fcn.blocks.at(te.panic_block).terminator.is_Diverge() ) { + // Shortcut: Don't create a new state if the panic target is Diverge + } + else { + m_states_to_do.push_back( ::std::make_pair(te.panic_block, state.clone()) ); + } + m_states_to_do.push_back( ::std::make_pair(te.ret_block, mv$(state)) ); + ) + ) + } + }; + + ::std::vector<bool> use_bitmap(vl.stmt_bitmap.size()); // Bitmap of locations where this value is used. + { + size_t pos = 0; + for(const auto& bb : fcn.blocks) + { + for(const auto& stmt : bb.statements) + { + use_bitmap[pos] = visit_mir_lvalues(stmt, [&](const ::MIR::LValue& tlv, auto vu){ return tlv == lv && vu != ValUsage::Write; }); + pos ++; + } + use_bitmap[pos] = visit_mir_lvalues(bb.terminator, [&](const ::MIR::LValue& tlv, auto vu){ return tlv == lv && vu != ValUsage::Write; }); + pos ++; + } + } + + Runner runner(mir_res, fcn, bb_idx, stmt_idx, lv, block_offsets, vl); + ::std::vector< ::std::pair<size_t,State>> post_check_list; + + // TODO: Have a bitmap of visited statements. If a visted statement is hit, stop the current state + // - Use the same rules as loopback. + + // Fill the first statement, to ensure that there is at least one bit set. + runner.run_block(bb_idx, stmt_idx, State(block_offsets, vl, bb_idx, stmt_idx)); + + while( ! runner.m_states_to_do.empty() ) + { + auto bb_idx = runner.m_states_to_do.back().first; + auto state = mv$(runner.m_states_to_do.back().second); + runner.m_states_to_do.pop_back(); + + DEBUG("state.bb_history=[" << state.bb_history << "], -> BB" << bb_idx); + state.bb_history.push_back(bb_idx); + + if( runner.m_visited_statements.at( block_offsets.at(bb_idx) + 0 ) ) + { + if( vl.stmt_bitmap.at( block_offsets.at(bb_idx) + 0) ) + { + DEBUG("Looped (to already valid)"); + state.mark_read(0); + state.finalise(0); + continue ; + } + else if( state.is_borrowed() ) + { + DEBUG("Looped (borrowed)"); + state.mark_read(0); + state.finalise(0); + continue ; + } + else + { + // Put this state elsewhere and check if the variable is known valid at that point. + DEBUG("Looped (after last read), push for later"); + post_check_list.push_back( ::std::make_pair(bb_idx, mv$(state)) ); + continue ; + } + } + +#if 0 + // TODO: Have a bitmap of if a BB mentions this value. If there are no unvisited BBs that mention this value, stop early. + // - CATCH: The original BB contains a reference, but might not have been visited (if it was the terminating call that triggered) + // - Also, we don't want to give up early (if we loop back to the start of the first block) + // - A per-statement bitmap would solve this. Return early if `!vl.stmt_bitmap & usage_stmt_bitmap == 0` + // > Requires filling the bitmap as we go (for maximum efficiency) + { + bool found_non_visited = false; + for(size_t i = 0; i < use_bitmap.size(); i ++) + { + // If a place where the value is used is not present in the output bitmap + if( !vl.stmt_bitmap[i] && use_bitmap[i] ) + { + DEBUG("- Still used at +" << i); + found_non_visited = true; + } + } + // If there were no uses of the variable that aren't covered by the lifetime bitmap + if( ! found_non_visited ) + { + // Terminate early + DEBUG("Early terminate - All possible lifetimes covered"); + state.finalise(0); + for(auto& s : runner.m_states_to_do) + { + s.second.bb_history.push_back(bb_idx); + s.second.finalise(0); + } + return ; + } + } +#endif + + // Special case for when doing multiple runs on the same output + if( vl.stmt_bitmap.at( block_offsets.at(bb_idx) + 0) ) + { + DEBUG("Already valid in BB" << bb_idx); + state.mark_read(0); + state.finalise(0); + continue; + } +#if 0 + // TODO: Have a way of knowing if a state will never find a use (the negative of the above) + // - Requires knowing for sure that a BB doesn't end up using the value. + // - IDEA: Store a fork count and counts of Yes/No for each BB. + // > If ForkCount == No, the value isn't used in that branch. + if( runner.m_bb_counts[bb_idx].visit_count > 0 + && runner.m_bb_counts[bb_idx].visit_count == runner.m_bb_counts[bb_idx].val_unused_count ) + { + DEBUG("Target BB known to be not valid"); + runner.apply_state(state, 0); + continue ; + } + runner.m_bb_counts[bb_idx].visit_count ++; +#endif + + runner.run_block(bb_idx, 0, mv$(state)); + } + + // Iterate while there are items in the post_check list + while( !post_check_list.empty() ) + { + bool change = false; + for(auto it = post_check_list.begin(); it != post_check_list.end(); ) + { + auto bb_idx = it->first; + auto& state = it->second; + // If the target of this loopback is valid, then the entire route to the loopback must have been valid + if( vl.stmt_bitmap.at( block_offsets.at(bb_idx) + 0) ) + { + change = true; + DEBUG("Looped (now valid)"); + state.mark_read(0); + state.finalise(0); + + it = post_check_list.erase(it); + } + else + { + ++ it; + } + } + // Keep going while changes happen + if( !change ) + break; + } +} + +#else + +::MIR::ValueLifetimes MIR_Helper_GetLifetimes(::MIR::TypeResolve& state, const ::MIR::Function& fcn, bool dump_debug) +{ + TRACE_FUNCTION_F(state); + // New algorithm notes: + // --- + // The lifetime of a value starts when it is written, and ends the last time it is read + // - When a variable is read, end any existing lifetime and start a new one. + // - When the value is read, update the end of its lifetime. + // --- + // A lifetime is a range in the call graph (with a start and end, including list of blocks) + // - Representation: Bitmap with a bit per statement. + // - Record the current block path in general state, along with known active lifetimes + + // TODO: If a value is borrowed, assume it lives forevermore + // - Ideally there would be borrow tracking to determine its actual required lifetime. + // - NOTE: This doesn't impact the borrows themselves, just the borrowee + + // TODO: Add a statement type StorageDead (or similar?) that indicates the point where a values scope ends + + // Scan through all possible paths in the graph (with loopback detection using a memory of the path) + // - If a loop is detected, determine if there were changes to the lifetime set during that pass + // > Changes are noticed by recording in the state structure when it triggers a change in the lifetime + // map. + struct Position + { + size_t path_index = 0; // index into the block path. + unsigned int stmt_idx = 0; + + bool operator==(const Position& x) const { + return path_index == x.path_index && stmt_idx == x.stmt_idx; + } + }; + struct ProtoLifetime + { + Position start; + Position end; + + bool is_empty() const { + return start == end; + } + bool is_borrowed() const { + return this->end == Position { ~0u, ~0u }; + } + }; + static unsigned NEXT_INDEX = 0; + struct State + { + unsigned int index = 0; + ::std::vector<unsigned int> block_path; + ::std::vector<unsigned int> block_change_idx; + unsigned int cur_change_idx = 0; + + // if read, update. If set, save and update + ::std::vector<ProtoLifetime> tmp_ends; + ::std::vector<ProtoLifetime> var_ends; + + State(const ::MIR::Function& fcn): + tmp_ends( fcn.temporaries.size(), ProtoLifetime() ), + var_ends( fcn.named_variables.size(), ProtoLifetime() ) + { + } + + State clone() const { + auto rv = *this; + rv.index = ++NEXT_INDEX; + return rv; + } + }; + NEXT_INDEX = 0; + + size_t statement_count = 0; + ::std::vector<size_t> block_offsets; + block_offsets.reserve( fcn.blocks.size() ); + for(const auto& bb : fcn.blocks) + { + block_offsets.push_back(statement_count); + statement_count += bb.statements.size() + 1; // +1 for the terminator + } + + ::std::vector<ValueLifetime> temporary_lifetimes( fcn.temporaries.size(), ValueLifetime(statement_count) ); + ::std::vector<ValueLifetime> variable_lifetimes( fcn.named_variables.size(), ValueLifetime(statement_count) ); + + struct BlockSeenLifetimes { + bool m_has_state = false; + const ::std::vector<size_t>& block_offsets; + ::std::vector< ::std::vector<unsigned int> > tmp; + ::std::vector< ::std::vector<unsigned int> > var; + + BlockSeenLifetimes(const ::std::vector<size_t>& block_offsets, const ::MIR::Function& fcn): + block_offsets( block_offsets ), + tmp( fcn.temporaries.size() ), + var( fcn.named_variables.size() ) + {} + + bool has_state() const + { + return m_has_state; + } + + bool try_merge(const State& val_state) const + { + // TODO: This logic isn't quite correct. Just becase a value's existing end is already marked as valid, + // doesn't mean that we have no new information. + // - Wait, doesn't it? + auto try_merge_lft = [&](const ProtoLifetime& lft, const ::std::vector<unsigned int>& seen)->bool { + if(lft.is_empty()) return false; + // TODO: What should be done for borrow flagged values + if(lft.is_borrowed()) return false; + auto end_idx = block_offsets.at( val_state.block_path.at(lft.end.path_index) ) + lft.end.stmt_idx; + + auto it = ::std::find(seen.begin(), seen.end(), end_idx); + return (it == seen.end()); + }; + for(size_t i = 0; i < val_state.tmp_ends.size(); i++) + { + if( try_merge_lft(val_state.tmp_ends[i], this->tmp[i]) ) + return true; + } + for(size_t i = 0; i < val_state.var_ends.size(); i++) + { + if( try_merge_lft(val_state.var_ends[i], this->var[i]) ) + return true; + } + return false; + } + + bool merge(const State& val_state) + { + bool rv = false; + auto merge_lft = [&](const ProtoLifetime& lft, ::std::vector<unsigned int>& seen)->bool { + if(lft.is_empty()) return false; + // TODO: What should be done for borrow flagged values + if(lft.end == Position { ~0u, ~0u }) return false; + auto end_idx = block_offsets.at( val_state.block_path.at(lft.end.path_index) ) + lft.end.stmt_idx; + + auto it = ::std::find(seen.begin(), seen.end(), end_idx); + if( it == seen.end() ) + { + seen.push_back( end_idx ); + return true; + } + else + { + return false; + } + }; + for(size_t i = 0; i < val_state.tmp_ends.size(); i++) + { + rv |= merge_lft(val_state.tmp_ends[i], this->tmp[i]); + } + for(size_t i = 0; i < val_state.var_ends.size(); i++) + { + rv |= merge_lft(val_state.var_ends[i], this->var[i]); + } + m_has_state = true; + return rv; + } + }; + ::std::vector<BlockSeenLifetimes> block_seen_lifetimes( fcn.blocks.size(), BlockSeenLifetimes(block_offsets, fcn) ); + + State init_state(fcn); + + ::std::vector<::std::pair<unsigned int, State>> todo_queue; + todo_queue.push_back(::std::make_pair( 0, mv$(init_state) )); + + while(!todo_queue.empty()) + { + auto bb_idx = todo_queue.back().first; + auto val_state = mv$(todo_queue.back().second); + todo_queue.pop_back(); + state.set_cur_stmt(bb_idx, 0); + + // Fill alive time in the bitmap + // TODO: Maybe also store the range (as a sequence of {block,start,end}) + auto add_lifetime_s = [&](State& val_state, const ::MIR::LValue& lv, const Position& start, const Position& end) { + assert(start.path_index <= end.path_index); + assert(start.path_index < end.path_index || start.stmt_idx <= end.stmt_idx); + if(start.path_index == end.path_index && start.stmt_idx == end.stmt_idx) + return; + DEBUG("[add_lifetime] " << lv << " (" << start.path_index << "," << start.stmt_idx << ") -- (" << end.path_index << "," << end.stmt_idx << ")"); + ValueLifetime* lft; + if(const auto* e = lv.opt_Temporary()) + { + lft = &temporary_lifetimes[e->idx]; + } + else if(const auto* e = lv.opt_Variable()) + { + lft = &variable_lifetimes[*e]; + } + else + { + MIR_TODO(state, "[add_lifetime] " << lv); + return; + } + + // Fill lifetime map for this temporary in the indicated range + bool did_set = false; + unsigned int j = start.stmt_idx; + unsigned int i = start.path_index; + while( i <= end.path_index && i < val_state.block_path.size() ) + { + auto bb_idx = val_state.block_path.at(i); + const auto& bb = fcn.blocks[bb_idx]; + MIR_ASSERT(state, j <= bb.statements.size(), ""); + MIR_ASSERT(state, bb_idx < block_offsets.size(), ""); + + auto block_base = block_offsets.at(bb_idx); + auto idx = block_base + j; + if( !lft->stmt_bitmap.at(idx) ) + { + lft->stmt_bitmap[idx] = true; + did_set = true; + } + + if( i == end.path_index && j == (end.stmt_idx != ~0u ? end.stmt_idx : bb.statements.size()) ) + break; + + // If the current index is the terminator (one after the size) + if(j == bb.statements.size()) + { + j = 0; + i++; + } + else + { + j ++; + } + } + + // - If the above set a new bit, increment `val_state.cur_change_idx` + if( did_set ) + { + DEBUG("[add_lifetime] " << lv << " (" << start.path_index << "," << start.stmt_idx << ") -- (" << end.path_index << "," << end.stmt_idx << ") - New information"); + val_state.cur_change_idx += 1; + } + }; + auto add_lifetime = [&](const ::MIR::LValue& lv, const Position& start, const Position& end) { + add_lifetime_s(val_state, lv, start, end); + }; + + auto apply_state = [&](State& state) { + // Apply all changes in this state, just in case there was new information + for(unsigned i = 0; i < fcn.temporaries.size(); i++) + add_lifetime_s( state, ::MIR::LValue::make_Temporary({i}), state.tmp_ends[i].start, state.tmp_ends[i].end ); + for(unsigned i = 0; i < fcn.named_variables.size(); i++) + add_lifetime_s( state, ::MIR::LValue::make_Variable({i}), state.var_ends[i].start, state.var_ends[i].end ); + }; + auto add_to_visit = [&](unsigned int new_bb_idx, State new_state) { + auto& bb_memory_ent = block_seen_lifetimes[new_bb_idx]; + if( !bb_memory_ent.has_state() ) + { + // No recorded state, needs to be visited + DEBUG(state << " state" << new_state.index << " -> bb" << new_bb_idx << " (no existing state)"); + } + else if( bb_memory_ent.try_merge(new_state) ) + { + // This state has new information, needs to be visited + DEBUG(state << " state" << new_state.index << " -> bb" << new_bb_idx << " (new info)"); + } + else + { + // Skip + // TODO: Acquire from the target block the actual end of any active lifetimes, then apply them. + DEBUG(state << " state" << new_state.index << " -> bb" << new_bb_idx << " - No new state, no push"); + // - For all variables currently active, check if they're valid in the first statement of the target block. + // - If so, mark as valid at the end of the current block + auto bm_idx = block_offsets[new_bb_idx]; + Position cur_pos; + cur_pos.path_index = val_state.block_path.size() - 1; + cur_pos.stmt_idx = fcn.blocks[bb_idx].statements.size(); + for(unsigned i = 0; i < fcn.temporaries.size(); i++) { + if( ! new_state.tmp_ends[i].is_empty() && temporary_lifetimes[i].stmt_bitmap[bm_idx] ) { + DEBUG("- tmp$" << i << " - Active in target, assume active"); + new_state.tmp_ends[i].end = cur_pos; + } + } + for(unsigned i = 0; i < fcn.named_variables.size(); i++) { + if( ! new_state.var_ends[i].is_empty() && variable_lifetimes[i].stmt_bitmap[bm_idx] ) { + DEBUG("- var$" << i << " - Active in target, assume active"); + new_state.var_ends[i].end = cur_pos; + } + } + // - Apply whatever state was still active + apply_state(new_state); + return ; + } + todo_queue.push_back(::std::make_pair( new_bb_idx, mv$(new_state) )); + }; + + // Compare this state to a composite list of lifetimes seen in this block + // - Just compares the end of each proto lifetime + { + auto& bb_memory_ent = block_seen_lifetimes[bb_idx]; + bool had_state = bb_memory_ent.has_state(); + bool has_new = bb_memory_ent.merge(val_state); + + if( !has_new && had_state ) + { + DEBUG(state << " state" << val_state.index << " - No new entry state"); + apply_state(val_state); + + continue ; + } + } + + // Check if this state has visited this block before, and if anything changed since last time + { + auto it = ::std::find(val_state.block_path.rbegin(), val_state.block_path.rend(), bb_idx); + if( it != val_state.block_path.rend() ) + { + auto idx = &*it - &val_state.block_path.front(); + if( val_state.block_change_idx[idx] == val_state.cur_change_idx ) + { + DEBUG(state << " " << val_state.index << " Loop and no change"); + continue ; + } + else + { + assert( val_state.block_change_idx[idx] < val_state.cur_change_idx ); + DEBUG(state << " " << val_state.index << " --- Loop, " << val_state.cur_change_idx - val_state.block_change_idx[idx] << " changes"); + } + } + else + { + DEBUG(state << " " << val_state.index << " ---"); + } + val_state.block_path.push_back(bb_idx); + val_state.block_change_idx.push_back( val_state.cur_change_idx ); + } + + Position cur_pos; + cur_pos.path_index = val_state.block_path.size() - 1; + cur_pos.stmt_idx = 0; + auto lvalue_read = [&](const ::MIR::LValue& lv) { + ProtoLifetime* slot; + if(const auto* e = lv.opt_Temporary()) { + slot = &val_state.tmp_ends.at(e->idx); + } + else if(const auto* e = lv.opt_Variable()) { + slot = &val_state.var_ends.at(*e); + } + else { + return ; + } + // Update the last read location + //DEBUG("Update END " << lv << " to " << cur_pos); + slot->end = cur_pos; + }; + auto lvalue_set = [&](const ::MIR::LValue& lv) { + ProtoLifetime* slot; + if(const auto* e = lv.opt_Temporary()) { + slot = &val_state.tmp_ends.at(e->idx); + } + else if(const auto* e = lv.opt_Variable()) { + slot = &val_state.var_ends.at(*e); + } + else { + return ; + } + // End whatever value was originally there, and insert this new one + slot->end = cur_pos; + add_lifetime(lv, slot->start, slot->end); + slot->start = cur_pos; + }; + auto lvalue_borrow = [&](const ::MIR::LValue& lv) { + ProtoLifetime* slot; + if(const auto* e = lv.opt_Temporary()) { + slot = &val_state.tmp_ends.at(e->idx); + } + else if(const auto* e = lv.opt_Variable()) { + slot = &val_state.var_ends.at(*e); + } + else { + return ; + } + // TODO: Flag this value as currently being borrowed (a flag that never clears) + slot->end = Position { ~0u, ~0u }; + }; + auto visit_lval_cb = [&](const auto& lv, ValUsage vu)->bool{ + if(vu == ValUsage::Read) + lvalue_read(lv); + if(vu == ValUsage::Borrow) + lvalue_borrow(lv); + if(vu == ValUsage::Write) + lvalue_set(lv); + return false; + }; + + // Run statements + for(const auto& stmt : fcn.blocks[bb_idx].statements) + { + auto stmt_idx = &stmt - &fcn.blocks[bb_idx].statements.front(); + cur_pos.stmt_idx = stmt_idx; + state.set_cur_stmt(bb_idx, stmt_idx); + DEBUG(state << " " << stmt); + + if( const auto* e = stmt.opt_Drop() ) + { + visit_mir_lvalues(stmt, [&](const auto& lv, ValUsage vu)->bool{ + if(vu == ValUsage::Read) + lvalue_read(lv); + return false; + }); + lvalue_read(e->slot); + lvalue_set(e->slot); + } + else + { + visit_mir_lvalues(stmt, visit_lval_cb); + } + } + cur_pos.stmt_idx = fcn.blocks[bb_idx].statements.size(); + + state.set_cur_stmt_term(bb_idx); + DEBUG(state << "TERM " << fcn.blocks[bb_idx].terminator); + TU_MATCH(::MIR::Terminator, (fcn.blocks[bb_idx].terminator), (e), + (Incomplete, + // Should be impossible here. + ), + (Return, + // End all active lifetimes at their previous location. + apply_state(val_state); + ), + (Diverge, + apply_state(val_state); + ), + (Goto, + add_to_visit(e, mv$(val_state)); + ), + (Panic, + // What should be done here? + ), + (If, + visit_mir_lvalue(e.cond, ValUsage::Read, visit_lval_cb); + + // Push blocks + add_to_visit(e.bb0, val_state.clone()); + add_to_visit(e.bb1, mv$(val_state)); + ), + (Switch, + visit_mir_lvalue(e.val, ValUsage::Read, visit_lval_cb); + ::std::set<unsigned int> tgts; + for(const auto& tgt : e.targets) + tgts.insert(tgt); + + for(const auto& tgt : tgts) + { + auto vs = (tgt == *tgts.rbegin() ? mv$(val_state) : val_state.clone()); + add_to_visit(tgt, mv$(vs)); + } + ), + (Call, + if( const auto* f = e.fcn.opt_Value() ) + visit_mir_lvalue(*f, ValUsage::Read, visit_lval_cb); + for(const auto& arg : e.args) + if( const auto* e = arg.opt_LValue() ) + visit_mir_lvalue(*e, ValUsage::Read, visit_lval_cb); + + // Push blocks (with return valid only in one) + add_to_visit(e.panic_block, val_state.clone()); + + // TODO: If the function returns !, don't follow the ret_block + lvalue_set(e.ret_val); + add_to_visit(e.ret_block, mv$(val_state)); + ) + ) + } + + // Dump out variable lifetimes. + if( dump_debug ) + { + for(unsigned int i = 0; i < temporary_lifetimes.size(); i ++) + { + temporary_lifetimes[i].dump_debug("tmp", i, block_offsets); + } + for(unsigned int i = 0; i < variable_lifetimes.size(); i ++) + { + variable_lifetimes[i].dump_debug("var", i, block_offsets); + } + } + + // Move lifetime bitmaps into the variable for the below code + ::MIR::ValueLifetimes rv; + rv.m_block_offsets = mv$(block_offsets); + rv.m_temporaries.reserve( temporary_lifetimes.size() ); + for(auto& lft : temporary_lifetimes) + rv.m_temporaries.push_back( ::MIR::ValueLifetime(mv$(lft.stmt_bitmap)) ); + rv.m_variables.reserve( variable_lifetimes.size() ); + for(auto& lft : variable_lifetimes) + rv.m_variables.push_back( ::MIR::ValueLifetime(mv$(lft.stmt_bitmap)) ); + + return rv; +} +#endif diff --git a/src/mir/helpers.hpp b/src/mir/helpers.hpp index c91e4198..802ce88f 100644 --- a/src/mir/helpers.hpp +++ b/src/mir/helpers.hpp @@ -23,6 +23,10 @@ class Function; class LValue; class Constant; struct BasicBlock; +class Terminator; +class Statement; +class RValue; +class Param; typedef unsigned int BasicBlockId; @@ -78,6 +82,7 @@ public: this->bb_idx = bb_idx; this->stmt_idx = stmt_idx; } + unsigned int get_cur_stmt_ofs() const; void set_cur_stmt_term(unsigned int bb_idx) { this->bb_idx = bb_idx; this->stmt_idx = STMT_TERM; @@ -107,4 +112,78 @@ public: } }; + +// -------------------------------------------------------------------- +// MIR_Helper_GetLifetimes +// -------------------------------------------------------------------- +class ValueLifetime +{ + ::std::vector<bool> statements; + +public: + ValueLifetime(::std::vector<bool> stmts): + statements( mv$(stmts) ) + {} + + bool valid_at(size_t ofs) const { + return statements.at(ofs); + } + + // true if this value is used at any point + bool is_used() const { + for(auto v : statements) + if( v ) + return true; + return false; + } + bool overlaps(const ValueLifetime& x) const { + assert(statements.size() == x.statements.size()); + for(unsigned int i = 0; i < statements.size(); i ++) + { + if( statements[i] && x.statements[i] ) + return true; + } + return false; + } + void unify(const ValueLifetime& x) { + assert(statements.size() == x.statements.size()); + for(unsigned int i = 0; i < statements.size(); i ++) + { + if( x.statements[i] ) + statements[i] = true; + } + } +}; + +struct ValueLifetimes +{ + ::std::vector<size_t> m_block_offsets; + ::std::vector<ValueLifetime> m_temporaries; + ::std::vector<ValueLifetime> m_variables; + + bool var_valid(unsigned var_idx, unsigned bb_idx, unsigned stmt_idx) const { + return m_variables.at(var_idx).valid_at( m_block_offsets[bb_idx] + stmt_idx ); + } + bool tmp_valid(unsigned tmp_idx, unsigned bb_idx, unsigned stmt_idx) const { + return m_temporaries.at(tmp_idx).valid_at( m_block_offsets[bb_idx] + stmt_idx ); + } +}; + +namespace visit { + enum class ValUsage { + Move, + Read, + Write, + Borrow, + }; + + extern bool visit_mir_lvalue(const ::MIR::LValue& lv, ValUsage u, ::std::function<bool(const ::MIR::LValue& , ValUsage)> cb); + extern bool visit_mir_lvalue(const ::MIR::Param& p, ValUsage u, ::std::function<bool(const ::MIR::LValue& , ValUsage)> cb); + extern bool visit_mir_lvalues(const ::MIR::RValue& rval, ::std::function<bool(const ::MIR::LValue& , ValUsage)> cb); + extern bool visit_mir_lvalues(const ::MIR::Statement& stmt, ::std::function<bool(const ::MIR::LValue& , ValUsage)> cb); + extern bool visit_mir_lvalues(const ::MIR::Terminator& term, ::std::function<bool(const ::MIR::LValue& , ValUsage)> cb); +} // namespace visit + } // namespace MIR + +extern ::MIR::ValueLifetimes MIR_Helper_GetLifetimes(::MIR::TypeResolve& state, const ::MIR::Function& fcn, bool dump_debug); diff --git a/src/mir/main_bindings.hpp b/src/mir/main_bindings.hpp index dc9a61a9..0d6074cb 100644 --- a/src/mir/main_bindings.hpp +++ b/src/mir/main_bindings.hpp @@ -15,6 +15,7 @@ class Crate; extern void HIR_GenerateMIR(::HIR::Crate& crate); extern void MIR_Dump(::std::ostream& sink, const ::HIR::Crate& crate); extern void MIR_CheckCrate(/*const*/ ::HIR::Crate& crate); +extern void MIR_CheckCrate_Full(/*const*/ ::HIR::Crate& crate); extern void MIR_CleanupCrate(::HIR::Crate& crate); extern void MIR_OptimiseCrate(::HIR::Crate& crate); diff --git a/src/mir/mir.cpp b/src/mir/mir.cpp index 6ae5808b..9edc925b 100644 --- a/src/mir/mir.cpp +++ b/src/mir/mir.cpp @@ -24,22 +24,22 @@ namespace MIR { os << (e.v ? "true" : "false"); ), (Bytes, - os << "["; + os << "b\""; os << ::std::hex; for(auto v : e) - os << static_cast<unsigned int>(v) << " "; - os << ::std::dec; - os << "]"; - ), - (StaticString, - os << "\""; - for(auto v : e) { + { if( ' ' <= v && v < 0x7F && v != '"' && v != '\\' ) os << v; + else if( v < 16 ) + os << "\\x0" << (unsigned int)v; else - os << "\\u{" << FMT(::std::hex << (unsigned int)v) << "}"; + os << "\\x" << ((unsigned int)v & 0xFF); } os << "\""; + os << ::std::dec; + ), + (StaticString, + os << "\"" << FmtEscaped(e) << "\""; ), (Const, os << e.p; @@ -50,34 +50,40 @@ namespace MIR { ) return os; } - bool Constant::operator==(const Constant& b) const + ::Ordering Constant::ord(const Constant& b) const { if( this->tag() != b.tag() ) - return false; + return ::ord( static_cast<unsigned int>(this->tag()), b.tag() ); TU_MATCHA( (*this,b), (ae,be), (Int, - return ae.v == be.v && ae.t == be.t; + if( ae.v != be.v ) + return ::ord(ae.v, be.v); + return ::ord((unsigned)ae.t, (unsigned)be.t); ), (Uint, - return ae.v == be.v && ae.t == be.t; + if( ae.v != be.v ) + return ::ord(ae.v, be.v); + return ::ord((unsigned)ae.t, (unsigned)be.t); ), (Float, - return ae.v == be.v && ae.t == be.t; + if( ae.v != be.v ) + return ::ord(ae.v, be.v); + return ::ord((unsigned)ae.t, (unsigned)be.t); ), (Bool, - return ae.v == be.v; + return ::ord(ae.v, be.v); ), (Bytes, - return ae == be; + return ::ord(ae, be); ), (StaticString, - return ae == be; + return ::ord(ae, be); ), (Const, - return ae.p == be.p; + return ::ord(ae.p, be.p); ), (ItemAddr, - return ae == be; + return ::ord(ae, be); ) ) throw ""; @@ -444,6 +450,51 @@ namespace MIR { return os; } + ::std::ostream& operator<<(::std::ostream& os, const Statement& x) + { + TU_MATCHA( (x), (e), + (Assign, + os << e.dst << " = " << e.src; + ), + (Asm, + os << "("; + for(const auto& spec : e.outputs) + os << "\"" << spec.first << "\" : " << spec.second << ", "; + os << ") = asm!(\"\", input=( "; + for(const auto& spec : e.inputs) + os << "\"" << spec.first << "\" : " << spec.second << ", "; + os << "), clobbers=[" << e.clobbers << "], flags=[" << e.flags << "])"; + ), + (SetDropFlag, + os << "df$" << e.idx << " = "; + if( e.other == ~0u ) + { + os << e.new_val; + } + else + { + os << (e.new_val ? "!" : "") << "df$" << e.other; + } + ), + (Drop, + os << "drop(" << e.slot; + if(e.kind == ::MIR::eDropKind::SHALLOW) + os << " SHALLOW"; + if(e.flag_idx != ~0u) + os << "IF df$" << e.flag_idx; + os << ")"; + ), + (ScopeEnd, + os << "ScopeEnd("; + for(auto idx : e.vars) + os << "var$" << idx << ","; + for(auto idx : e.tmps) + os << "tmp$" << idx << ","; + os << ")"; + ) + ) + return os; + } } ::MIR::LValue MIR::LValue::clone() const diff --git a/src/mir/mir.hpp b/src/mir/mir.hpp index 75bb29a8..c22f8d5d 100644 --- a/src/mir/mir.hpp +++ b/src/mir/mir.hpp @@ -112,10 +112,13 @@ TAGGED_UNION_EX(Constant, (), Int, ( (ItemAddr, ::HIR::Path) // address of a value ), (), (), ( friend ::std::ostream& operator<<(::std::ostream& os, const Constant& v); - bool operator==(const Constant& b) const; - inline bool operator!=(const Constant& b) const { - return !(*this == b); - } + ::Ordering ord(const Constant& b) const; + inline bool operator==(const Constant& b) const { return ord(b) == ::OrdEqual; } + inline bool operator!=(const Constant& b) const { return ord(b) != ::OrdEqual; } + inline bool operator<(const Constant& b) const { return ord(b) == ::OrdLess; } + inline bool operator<=(const Constant& b) const { return ord(b) != ::OrdGreater; } + inline bool operator>(const Constant& b) const { return ord(b) == ::OrdGreater; } + inline bool operator>=(const Constant& b) const { return ord(b) != ::OrdLess; } Constant clone() const; ) ); @@ -269,8 +272,13 @@ TAGGED_UNION(Statement, Assign, eDropKind kind; // NOTE: For the `box` primitive LValue slot; unsigned int flag_idx; // Valid if != ~0u + }), + (ScopeEnd, struct { + ::std::vector<unsigned> vars; + ::std::vector<unsigned> tmps; }) ); +extern ::std::ostream& operator<<(::std::ostream& os, const Statement& x); struct BasicBlock { @@ -282,6 +290,7 @@ struct BasicBlock class Function { public: + // TODO: Unify Variables, Temporaries, and Arguments ::std::vector< ::HIR::TypeRef> named_variables; ::std::vector< ::HIR::TypeRef> temporaries; ::std::vector<bool> drop_flags; diff --git a/src/mir/mir_builder.cpp b/src/mir/mir_builder.cpp index 296dfa91..eb50e9f7 100644 --- a/src/mir/mir_builder.cpp +++ b/src/mir/mir_builder.cpp @@ -32,8 +32,14 @@ MirBuilder::MirBuilder(const Span& sp, const StaticTraitResolve& resolve, const m_scopes.push_back( ScopeDef { sp, ScopeType::make_Temporaries({}) } ); m_scope_stack.push_back( 1 ); + + m_if_cond_lval = this->new_temporary(::HIR::CoreType::Bool); + + m_arg_states.reserve( args.size() ); + for(size_t i = 0; i < args.size(); i ++ ) + m_arg_states.push_back( VarState::make_Valid({}) ); m_variable_states.reserve( output.named_variables.size() ); - for(unsigned int i = 0; i < output.named_variables.size(); i ++ ) + for(size_t i = 0; i < output.named_variables.size(); i ++ ) m_variable_states.push_back( VarState::make_Invalid(InvalidType::Uninit) ); } MirBuilder::~MirBuilder() @@ -116,6 +122,22 @@ void MirBuilder::define_variable(unsigned int idx) top_scope = &m_scopes.at(idx); break ; } + else if( m_scopes.at(idx).data.is_Loop() ) + { + // Newly created temporary within a loop, if there is a saved + // state this temp needs a drop flag. + // TODO: ^ + } + else if( m_scopes.at(idx).data.is_Split() ) + { + // Newly created temporary within a split, if there is a saved + // state this temp needs a drop flag. + // TODO: ^ + } + else + { + // Nothign. + } } assert( top_scope ); auto& tmp_scope = top_scope->data.as_Temporaries(); @@ -190,7 +212,7 @@ void MirBuilder::define_variable(unsigned int idx) { auto temp = new_temporary(ty); push_stmt_assign( sp, ::MIR::LValue(temp.clone()), mv$(rv) ); - return ::MIR::Param( mv$(temp) ); + return temp; } } void MirBuilder::set_result(const Span& sp, ::MIR::RValue val) @@ -263,7 +285,7 @@ void MirBuilder::push_stmt_assign(const Span& sp, ::MIR::LValue dst, ::MIR::RVal // Doesn't move ), (MakeDst, - // Doesn't move ptr_val + moved_param(e.ptr_val); moved_param(e.meta_val); ), (Tuple, @@ -285,7 +307,7 @@ void MirBuilder::push_stmt_assign(const Span& sp, ::MIR::LValue dst, ::MIR::RVal // Drop target if populated mark_value_assigned(sp, dst); - m_output.blocks.at(m_current_block).statements.push_back( ::MIR::Statement::make_Assign({ mv$(dst), mv$(val) }) ); + this->push_stmt( sp, ::MIR::Statement::make_Assign({ mv$(dst), mv$(val) }) ); } void MirBuilder::push_stmt_drop(const Span& sp, ::MIR::LValue val, unsigned int flag/*=~0u*/) { @@ -297,11 +319,13 @@ void MirBuilder::push_stmt_drop(const Span& sp, ::MIR::LValue val, unsigned int return ; } - DEBUG("DROP " << val); - - auto stmt = ::MIR::Statement::make_Drop({ ::MIR::eDropKind::DEEP, mv$(val), flag }); + this->push_stmt(sp, ::MIR::Statement::make_Drop({ ::MIR::eDropKind::DEEP, mv$(val), flag })); - m_output.blocks.at(m_current_block).statements.push_back( mv$(stmt) ); + if( flag != ~0u ) + { + // Reset flag value back to default. + push_stmt_set_dropflag_val(sp, flag, m_output.drop_flags.at(flag)); + } } void MirBuilder::push_stmt_drop_shallow(const Span& sp, ::MIR::LValue val, unsigned int flag/*=~0u*/) { @@ -310,8 +334,13 @@ void MirBuilder::push_stmt_drop_shallow(const Span& sp, ::MIR::LValue val, unsig // TODO: Ensure that the type is a Box? - DEBUG("DROP shallow " << val); - m_output.blocks.at(m_current_block).statements.push_back( ::MIR::Statement::make_Drop({ ::MIR::eDropKind::SHALLOW, mv$(val), flag }) ); + this->push_stmt(sp, ::MIR::Statement::make_Drop({ ::MIR::eDropKind::SHALLOW, mv$(val), flag })); + + if( flag != ~0u ) + { + // Reset flag value back to default. + push_stmt_set_dropflag_val(sp, flag, m_output.drop_flags.at(flag)); + } } void MirBuilder::push_stmt_asm(const Span& sp, ::MIR::Statement::Data_Asm data) { @@ -322,17 +351,25 @@ void MirBuilder::push_stmt_asm(const Span& sp, ::MIR::Statement::Data_Asm data) mark_value_assigned(sp, v.second); // 2. Push - m_output.blocks.at(m_current_block).statements.push_back( ::MIR::Statement::make_Asm( mv$(data) ) ); + this->push_stmt(sp, ::MIR::Statement::make_Asm( mv$(data) )); } void MirBuilder::push_stmt_set_dropflag_val(const Span& sp, unsigned int idx, bool value) { - ASSERT_BUG(sp, m_block_active, "Pushing statement with no active block"); - m_output.blocks.at(m_current_block).statements.push_back( ::MIR::Statement::make_SetDropFlag({ idx, value }) ); + this->push_stmt(sp, ::MIR::Statement::make_SetDropFlag({ idx, value })); } void MirBuilder::push_stmt_set_dropflag_other(const Span& sp, unsigned int idx, unsigned int other) { + this->push_stmt(sp, ::MIR::Statement::make_SetDropFlag({ idx, false, other })); +} +void MirBuilder::push_stmt_set_dropflag_default(const Span& sp, unsigned int idx) +{ + this->push_stmt(sp, ::MIR::Statement::make_SetDropFlag({ idx, this->get_drop_flag_default(sp, idx) })); +} +void MirBuilder::push_stmt(const Span& sp, ::MIR::Statement stmt) +{ ASSERT_BUG(sp, m_block_active, "Pushing statement with no active block"); - m_output.blocks.at(m_current_block).statements.push_back( ::MIR::Statement::make_SetDropFlag({ idx, false, other }) ); + DEBUG(stmt); + m_output.blocks.at(m_current_block).statements.push_back( mv$(stmt) ); } void MirBuilder::mark_value_assigned(const Span& sp, const ::MIR::LValue& dst) @@ -374,32 +411,56 @@ void MirBuilder::mark_value_assigned(const Span& sp, const ::MIR::LValue& dst) } } -void MirBuilder::raise_variables(const Span& sp, const ::MIR::LValue& val, const ScopeHandle& scope) +void MirBuilder::raise_variables(const Span& sp, const ::MIR::LValue& val, const ScopeHandle& scope, bool to_above/*=false*/) { TRACE_FUNCTION_F(val); TU_MATCH_DEF(::MIR::LValue, (val), (e), ( + // No raising of these source values? + return ; ), // TODO: This may not be correct, because it can change the drop points and ordering // HACK: Working around cases where values are dropped while the result is not yet used. + (Index, + raise_variables(sp, *e.val, scope, to_above); + raise_variables(sp, *e.idx, scope, to_above); + return ; + ), (Deref, - raise_variables(sp, *e.val, scope); + raise_variables(sp, *e.val, scope, to_above); + return ; ), (Field, - raise_variables(sp, *e.val, scope); + raise_variables(sp, *e.val, scope, to_above); + return ; ), (Downcast, - raise_variables(sp, *e.val, scope); + raise_variables(sp, *e.val, scope, to_above); + return ; ), // Actual value types (Variable, - auto idx = e; - auto scope_it = m_scope_stack.rbegin(); - while( scope_it != m_scope_stack.rend() ) + ), + (Temporary, + ) + ) + ASSERT_BUG(sp, val.is_Variable() || val.is_Temporary(), "Hit value raising code with non-variable value - " << val); + + // Find controlling scope + auto scope_it = m_scope_stack.rbegin(); + while( scope_it != m_scope_stack.rend() ) + { + auto& scope_def = m_scopes.at(*scope_it); + + if( *scope_it == scope.idx && !to_above ) { - auto& scope_def = m_scopes.at(*scope_it); + DEBUG(val << " defined in or above target (scope " << scope << ")"); + } - TU_IFLET( ScopeType, scope_def.data, Variables, e, + TU_IFLET( ScopeType, scope_def.data, Variables, e, + if( const auto* ve = val.opt_Variable() ) + { + auto idx = *ve; auto tmp_it = ::std::find( e.vars.begin(), e.vars.end(), idx ); if( tmp_it != e.vars.end() ) { @@ -407,41 +468,12 @@ void MirBuilder::raise_variables(const Span& sp, const ::MIR::LValue& val, const DEBUG("Raise variable " << idx << " from " << *scope_it); break ; } - ) - // If the variable was defined above the desired scope (i.e. this didn't find it), return - if( *scope_it == scope.idx ) - return ; - ++scope_it; - } - if( scope_it == m_scope_stack.rend() ) - { - // Temporary wasn't defined in a visible scope? - return ; - } - ++scope_it; - - while( scope_it != m_scope_stack.rend() ) - { - auto& scope_def = m_scopes.at(*scope_it); - - TU_IFLET( ScopeType, scope_def.data, Variables, e, - e.vars.push_back( idx ); - DEBUG("- to " << *scope_it); - return ; - ) - ++scope_it; - } - - DEBUG("- top"); - ), - (Temporary, - auto idx = e.idx; - auto scope_it = m_scope_stack.rbegin(); - while( scope_it != m_scope_stack.rend() ) - { - auto& scope_def = m_scopes.at(*scope_it); - - TU_IFLET( ScopeType, scope_def.data, Temporaries, e, + } + ) + else TU_IFLET( ScopeType, scope_def.data, Temporaries, e, + if( const auto* ve = val.opt_Temporary() ) + { + auto idx = ve->idx; auto tmp_it = ::std::find( e.temporaries.begin(), e.temporaries.end(), idx ); if( tmp_it != e.temporaries.end() ) { @@ -449,45 +481,169 @@ void MirBuilder::raise_variables(const Span& sp, const ::MIR::LValue& val, const DEBUG("Raise temporary " << idx << " from " << *scope_it); break ; } - ) - - // If the temporary was defined above the desired scope (i.e. this didn't find it), return - if( *scope_it == scope.idx ) - return ; - ++scope_it; + } + ) + else + { + // TODO: Does this need to handle this value being set in the + // split scopes? } - if( scope_it == m_scope_stack.rend() ) + // If the variable was defined above the desired scope (i.e. this didn't find it), return + if( *scope_it == scope.idx ) { - // Temporary wasn't defined in a visible scope? + DEBUG("Value " << val << " is defined above the target (scope " << scope << ")"); return ; } ++scope_it; + } + if( scope_it == m_scope_stack.rend() ) + { + // Temporary wasn't defined in a visible scope? + BUG(sp, val << " wasn't defined in a visible scope"); + return ; + } - while( scope_it != m_scope_stack.rend() ) - { - auto& scope_def = m_scopes.at(*scope_it); + // If the definition scope was the target scope + bool target_seen = false; + if( *scope_it == scope.idx ) + { + if( to_above ) { + // Want to shift to any above (but not including) it + ++ scope_it; + } + else { + // Want to shift to it or above. + } - TU_IFLET( ScopeType, scope_def.data, Temporaries, e, - e.temporaries.push_back( idx ); - DEBUG("- to " << *scope_it); - return ; - ) - ++scope_it; + target_seen = true; + } + else + { + // Don't bother searching the original definition scope + ++scope_it; + } + + // Iterate stack until: + // - The target scope is seen + // - AND a scope was found for it + for( ; scope_it != m_scope_stack.rend(); ++ scope_it ) + { + auto& scope_def = m_scopes.at(*scope_it); + DEBUG("> Cross " << *scope_it << " - " << scope_def.data.tag_str()); + + if( *scope_it == scope.idx ) + { + target_seen = true; } - DEBUG("- top"); + TU_IFLET( ScopeType, scope_def.data, Variables, e, + if( target_seen ) + { + if( const auto* ve = val.opt_Variable() ) + { + e.vars.push_back( *ve ); + DEBUG("- to " << *scope_it); + return ; + } + } ) - ) + else TU_IFLET( ScopeType, scope_def.data, Temporaries, e, + if( target_seen ) + { + if( const auto* ve = val.opt_Temporary() ) + { + e.temporaries.push_back( ve->idx ); + DEBUG("- to " << *scope_it); + return ; + } + } + ) + else if( auto* sd_loop = scope_def.data.opt_Loop() ) + { + // If there is an exit state present, ensure that this variable is + // present in that state (as invalid, as it can't have been valid + // externally) + if( sd_loop->exit_state_valid ) + { + DEBUG("Adding " << val << " as unset to loop exit state"); + if( const auto* ve = val.opt_Variable() ) + { + auto v = sd_loop->exit_state.var_states.insert( ::std::make_pair(*ve, VarState(InvalidType::Uninit)) ); + ASSERT_BUG(sp, v.second, "Raising " << val << " which already had a state entry"); + } + else if( const auto* ve = val.opt_Temporary() ) + { + auto v = sd_loop->exit_state.tmp_states.insert( ::std::make_pair(ve->idx, VarState(InvalidType::Uninit)) ); + ASSERT_BUG(sp, v.second, "Raising " << val << " which already had a state entry"); + } + else { + BUG(sp, "Impossible raise value"); + } + } + else + { + DEBUG("Crossing loop with no existing exit state"); + } + } + else if( auto* sd_split = scope_def.data.opt_Split() ) + { + // If the split has already registered an exit state, ensure that + // this variable is present in it. (as invalid) + if( sd_split->end_state_valid ) + { + DEBUG("Adding " << val << " as unset to loop exit state"); + if( const auto* ve = val.opt_Variable() ) + { + auto v = sd_split->end_state.var_states.insert( ::std::make_pair(*ve, VarState(InvalidType::Uninit)) ); + ASSERT_BUG(sp, v.second, "Raising " << val << " which already had a state entry"); + } + else if( const auto* ve = val.opt_Temporary() ) + { + auto v = sd_split->end_state.tmp_states.insert( ::std::make_pair(ve->idx, VarState(InvalidType::Uninit)) ); + ASSERT_BUG(sp, v.second, "Raising " << val << " which already had a state entry"); + } + else { + BUG(sp, "Impossible raise value"); + } + } + else + { + DEBUG("Crossing split with no existing end state"); + } + + // TODO: This should update the outer state to unset. + auto& arm = sd_split->arms.back(); + if( const auto* ve = val.opt_Variable() ) + { + arm.var_states.insert(::std::make_pair( *ve, get_variable_state(sp, *ve).clone() )); + m_variable_states.at(*ve) = VarState(InvalidType::Uninit); + } + else if( const auto* ve = val.opt_Temporary() ) + { + arm.tmp_states.insert(::std::make_pair( ve->idx, get_temp_state(sp, ve->idx).clone() )); + m_temporary_states.at(ve->idx) = VarState(InvalidType::Uninit); + } + else + { + BUG(sp, "Impossible raise value"); + } + } + else + { + BUG(sp, "Crossing unknown scope type - " << scope_def.data.tag_str()); + } + } + BUG(sp, "Couldn't find a scope to raise " << val << " into"); } -void MirBuilder::raise_variables(const Span& sp, const ::MIR::RValue& rval, const ScopeHandle& scope) +void MirBuilder::raise_variables(const Span& sp, const ::MIR::RValue& rval, const ScopeHandle& scope, bool to_above/*=false*/) { auto raise_vars = [&](const ::MIR::Param& p) { if( const auto* e = p.opt_LValue() ) - this->raise_variables(sp, *e, scope); + this->raise_variables(sp, *e, scope, to_above); }; TU_MATCHA( (rval), (e), (Use, - this->raise_variables(sp, e, scope); + this->raise_variables(sp, e, scope, to_above); ), (Constant, ), @@ -496,23 +652,23 @@ void MirBuilder::raise_variables(const Span& sp, const ::MIR::RValue& rval, cons ), (Borrow, // TODO: Wait, is this valid? - this->raise_variables(sp, e.val, scope); + this->raise_variables(sp, e.val, scope, to_above); ), (Cast, - this->raise_variables(sp, e.val, scope); + this->raise_variables(sp, e.val, scope, to_above); ), (BinOp, raise_vars(e.val_l); raise_vars(e.val_r); ), (UniOp, - this->raise_variables(sp, e.val, scope); + this->raise_variables(sp, e.val, scope, to_above); ), (DstMeta, - this->raise_variables(sp, e.val, scope); + this->raise_variables(sp, e.val, scope, to_above); ), (DstPtr, - this->raise_variables(sp, e.val, scope); + this->raise_variables(sp, e.val, scope, to_above); ), (MakeDst, raise_vars(e.ptr_val); @@ -656,6 +812,21 @@ void MirBuilder::terminate_scope(const Span& sp, ScopeHandle scope, bool emit_cl { // 2. Emit drops for all non-moved variables (share with below) drop_scope_values(scope_def); + + // Emit ScopeEnd for all controlled values + ::MIR::Statement::Data_ScopeEnd se; + if(const auto* e = scope_def.data.opt_Variables() ) { + se.vars = e->vars; + } + else if(const auto* e = scope_def.data.opt_Temporaries()) { + se.tmps = e->temporaries; + } + else { + } + // Only push the ScopeEnd if there were variables to end + if( !se.vars.empty() || !se.tmps.empty() ) { + this->push_stmt(sp, ::MIR::Statement( mv$(se) )); + } } // 3. Pop scope (last because `drop_scope_values` uses the stack) @@ -663,6 +834,107 @@ void MirBuilder::terminate_scope(const Span& sp, ScopeHandle scope, bool emit_cl complete_scope(scope_def); } +void MirBuilder::raise_all(const Span& sp, ScopeHandle source, const ScopeHandle& target) +{ + TRACE_FUNCTION_F("scope " << source.idx << " => " << target.idx); + + // 1. Check that this is the current scope (at the top of the stack) + if( m_scope_stack.empty() || m_scope_stack.back() != source.idx ) + { + DEBUG("- m_scope_stack = [" << m_scope_stack << "]"); + auto it = ::std::find( m_scope_stack.begin(), m_scope_stack.end(), source.idx ); + if( it == m_scope_stack.end() ) + BUG(sp, "Terminating scope not on the stack - scope " << source.idx); + BUG(sp, "Terminating scope " << source.idx << " when not at top of stack, " << (m_scope_stack.end() - it - 1) << " scopes in the way"); + } + auto& src_scope_def = m_scopes.at(source.idx); + +#if 1 + ASSERT_BUG(sp, src_scope_def.data.is_Temporaries(), "Rasising scopes can only be done on temporaries (source)"); + auto& src_list = src_scope_def.data.as_Temporaries().temporaries; + for(auto idx : src_list) + { + DEBUG("> Raising " << ::MIR::LValue::make_Temporary({ idx })); + } + + // Seek up stack until the target scope is seen + auto it = m_scope_stack.rbegin() + 1; + for( ; it != m_scope_stack.rend() && *it != target.idx; ++it) + { + auto& scope_def = m_scopes.at(*it); + + if(auto* sd_loop = scope_def.data.opt_Loop()) + { + if(sd_loop->exit_state_valid) + { + DEBUG("Crossing loop with existing end state"); + // Insert these values as Invalid, both in the existing exit state, and in the changed list + for(auto idx : src_list) + { + auto v = sd_loop->exit_state.tmp_states.insert(::std::make_pair( idx, VarState(InvalidType::Uninit) )); + ASSERT_BUG(sp, v.second, ""); + } + } + else + { + DEBUG("Crossing loop with no end state"); + } + + for(auto idx : src_list) + { + auto v2 = sd_loop->changed_tmps.insert(::std::make_pair( idx, VarState(InvalidType::Uninit) )); + ASSERT_BUG(sp, v2.second, ""); + } + } + else if(auto* sd_split = scope_def.data.opt_Split()) + { + if(sd_split->end_state_valid) + { + DEBUG("Crossing split with existing end state"); + // Insert these indexes as Invalid + for(auto idx : src_list) + { + auto v = sd_split->end_state.tmp_states.insert(::std::make_pair( idx, VarState(InvalidType::Uninit) )); + ASSERT_BUG(sp, v.second, ""); + } + } + else + { + DEBUG("Crossing split with no end state"); + } + + // TODO: Insert current state in the current arm + assert(!sd_split->arms.empty()); + auto& arm = sd_split->arms.back(); + for(auto idx : src_list) + { + arm.tmp_states.insert(::std::make_pair( idx, mv$(m_temporary_states.at(idx)) )); + m_temporary_states.at(idx) = VarState(InvalidType::Uninit); + } + } + } + if(it == m_scope_stack.rend()) + { + BUG(sp, "Moving values to a scope not on the stack - scope " << target.idx); + } + auto& tgt_scope_def = m_scopes.at(target.idx); + ASSERT_BUG(sp, tgt_scope_def.data.is_Temporaries(), "Rasising scopes can only be done on temporaries (target)"); + + // Move all defined variables from one to the other + auto& tgt_list = tgt_scope_def.data.as_Temporaries().temporaries; + tgt_list.insert( tgt_list.end(), src_list.begin(), src_list.end() ); +#else + auto list = src_scope_def.data.as_Temporaries().temporaries; + for(auto idx : list) + { + this->raise_variables(sp, ::MIR::LValue::make_Temporary({ idx }), target); + } +#endif + + // Scope completed + m_scope_stack.pop_back(); + src_scope_def.complete = true; +} void MirBuilder::terminate_scope_early(const Span& sp, const ScopeHandle& scope, bool loop_exit/*=false*/) { @@ -717,7 +989,7 @@ namespace { static void merge_state(const Span& sp, MirBuilder& builder, const ::MIR::LValue& lv, VarState& old_state, const VarState& new_state) { - DEBUG(lv << " : " << old_state << " <= " << new_state); + TRACE_FUNCTION_FR(lv << " : " << old_state << " <= " << new_state, lv << " : " << old_state); switch(old_state.tag()) { case VarState::TAGDEAD: throw ""; @@ -735,61 +1007,104 @@ namespace case VarState::TAG_Optional: { // Was invalid, now optional. auto flag_idx = new_state.as_Optional(); - if( builder.get_drop_flag_default(sp, flag_idx) != false ) { + if( true || builder.get_drop_flag_default(sp, flag_idx) != false ) { #if 1 auto new_flag = builder.new_drop_flag(false); builder.push_stmt_set_dropflag_other(sp, new_flag, flag_idx); + builder.push_stmt_set_dropflag_default(sp, flag_idx); old_state = VarState::make_Optional( new_flag ); #else // TODO: Rewrite history. I.e. visit all previous branches and set this drop flag to `false` in all of them TODO(sp, "Drop flag default not false when going Invalid->Optional"); #endif } - old_state = VarState::make_Optional( flag_idx ); + else { + old_state = VarState::make_Optional( flag_idx ); + } return ; } - case VarState::TAG_Partial: { - const auto& nse = new_state.as_Partial(); + case VarState::TAG_MovedOut: { + const auto& nse = new_state.as_MovedOut(); + + // Create a new state that is internally valid and uses the same drop flag + old_state = VarState::make_MovedOut({ box$(old_state.clone()), nse.outer_flag }); + auto& ose = old_state.as_MovedOut(); + if( ose.outer_flag != ~0u ) + { + // If the flag's default isn't false, then create a new flag that does have such a default + // - Other arm (old_state) uses default, this arm (new_state) can be manipulated + if( builder.get_drop_flag_default(sp, ose.outer_flag) != false ) + { + auto new_flag = builder.new_drop_flag(false); + builder.push_stmt_set_dropflag_other(sp, new_flag, nse.outer_flag); + builder.push_stmt_set_dropflag_default(sp, nse.outer_flag); + ose.outer_flag = new_flag; + } + } + else + { + // In the old arm, the container isn't valid. Create a drop flag with a default of false and set it to true + ose.outer_flag = builder.new_drop_flag(false); + builder.push_stmt_set_dropflag_val(sp, ose.outer_flag, true); + } + bool is_box = false; - builder.with_val_type(sp, lv, [&](const auto& ty){ is_box = builder.is_type_owned_box(ty); }); - if( is_box ) { - ASSERT_BUG(sp, nse.inner_states.size() == 1, ""); + builder.with_val_type(sp, lv, [&](const auto& ty){ + is_box = builder.is_type_owned_box(ty); + }); + if( is_box ) + { + merge_state(sp, builder, ::MIR::LValue::make_Deref({ box$(lv.clone()) }), *ose.inner_state, *nse.inner_state); } - if( nse.outer_flag != ~0u ) { - // Set the outer flag to `true` if its default isn't true - if( builder.get_drop_flag_default(sp, nse.outer_flag) != false ) { - builder.push_stmt_set_dropflag_val(sp, nse.outer_flag, false); - } + else + { + BUG(sp, "Handle MovedOut on non-Box"); + } + return ; } + case VarState::TAG_Partial: { + const auto& nse = new_state.as_Partial(); + bool is_enum = false; + builder.with_val_type(sp, lv, [&](const auto& ty){ + is_enum = ty.m_data.is_Path() && ty.m_data.as_Path().binding.is_Enum(); + }); - auto out = new_state.clone(); - auto& ose = out.as_Partial(); - if( ose.outer_flag == ~0u ) + // Create a partial filled with Invalid { - ose.outer_flag = builder.new_drop_flag_and_set(sp, true); // Only in this arm is the container valid + ::std::vector<VarState> inner; inner.reserve( nse.inner_states.size() ); + for(size_t i = 0; i < nse.inner_states.size(); i++) + inner.push_back( old_state.clone() ); + old_state = VarState::make_Partial({ mv$(inner) }); } - if( is_box ) { - merge_state(sp, builder, ::MIR::LValue::make_Deref({ box$(lv.clone()) }), ose.inner_states[0], old_state); + auto& ose = old_state.as_Partial(); + if( is_enum ) { + for(size_t i = 0; i < ose.inner_states.size(); i ++) + { + merge_state(sp, builder, ::MIR::LValue::make_Downcast({ box$(lv.clone()), static_cast<unsigned int>(i) }), ose.inner_states[i], nse.inner_states[i]); + } } else { for(unsigned int i = 0; i < ose.inner_states.size(); i ++ ) { - merge_state(sp, builder, ::MIR::LValue::make_Field({ box$(lv.clone()), i }), ose.inner_states[i], old_state); + merge_state(sp, builder, ::MIR::LValue::make_Field({ box$(lv.clone()), i }), ose.inner_states[i], nse.inner_states[i]); } } - old_state = mv$(out); } return; } break; + // Valid <= ... case VarState::TAG_Valid: switch( new_state.tag() ) { case VarState::TAGDEAD: throw ""; + // Valid <= Invalid case VarState::TAG_Invalid: old_state = VarState::make_Optional( builder.new_drop_flag_and_set(sp, false) ); return ; + // Valid <= Valid case VarState::TAG_Valid: return ; + // Valid <= Optional case VarState::TAG_Optional: { auto flag_idx = new_state.as_Optional(); // Was valid, now optional. @@ -798,6 +1113,7 @@ namespace #if 1 auto new_flag = builder.new_drop_flag(true); builder.push_stmt_set_dropflag_other(sp, new_flag, flag_idx); + builder.push_stmt_set_dropflag_default(sp, flag_idx); old_state = VarState::make_Optional( new_flag ); #else // OR: Push an assign of this flag to every other completed arm @@ -816,39 +1132,79 @@ namespace } return ; } - case VarState::TAG_Partial: { - const auto& nse = new_state.as_Partial(); + // Valid <= MovedOut + case VarState::TAG_MovedOut: { + const auto& nse = new_state.as_MovedOut(); + + // Create a new staet that is internally valid and uses the same drop flag + old_state = VarState::make_MovedOut({ box$(VarState::make_Valid({})), nse.outer_flag }); + auto& ose = old_state.as_MovedOut(); + if( ose.outer_flag != ~0u ) + { + // If the flag's default isn't true, then create a new flag that does have such a default + // - Other arm (old_state) uses default, this arm (new_state) can be manipulated + if( builder.get_drop_flag_default(sp, ose.outer_flag) != true ) + { + auto new_flag = builder.new_drop_flag(true); + builder.push_stmt_set_dropflag_other(sp, new_flag, nse.outer_flag); + builder.push_stmt_set_dropflag_default(sp, nse.outer_flag); + ose.outer_flag = new_flag; + } + } + else + { + // In both arms, the container is valid. No need for a drop flag + } + bool is_box = false; - builder.with_val_type(sp, lv, [&](const auto& ty){ is_box = builder.is_type_owned_box(ty); }); + builder.with_val_type(sp, lv, [&](const auto& ty){ + is_box = builder.is_type_owned_box(ty); + }); + if( is_box ) { - ASSERT_BUG(sp, nse.inner_states.size() == 1, ""); + merge_state(sp, builder, ::MIR::LValue::make_Deref({ box$(lv.clone()) }), *ose.inner_state, *nse.inner_state); } - if( nse.outer_flag != ~0u ) { - // Set the outer flag to `true` if its default isn't true - if( builder.get_drop_flag_default(sp, nse.outer_flag) != true ) { - builder.push_stmt_set_dropflag_val(sp, nse.outer_flag, true); - } + else { + BUG(sp, "MovedOut on non-Box"); + } + return; } + // Valid <= Partial + case VarState::TAG_Partial: { + const auto& nse = new_state.as_Partial(); + bool is_enum = false; + builder.with_val_type(sp, lv, [&](const auto& ty){ + is_enum = ty.m_data.is_Path() && ty.m_data.as_Path().binding.is_Enum(); + }); - auto out = new_state.clone(); - auto& ose = out.as_Partial(); - if( ose.outer_flag == ~0u ) + // Create a partial filled with Valid { - ose.outer_flag = builder.new_drop_flag(true); // In both arms, the container is valid + ::std::vector<VarState> inner; inner.reserve( nse.inner_states.size() ); + for(size_t i = 0; i < nse.inner_states.size(); i++) + inner.push_back( VarState::make_Valid({}) ); + old_state = VarState::make_Partial({ mv$(inner) }); } - if( is_box ) { - merge_state(sp, builder, ::MIR::LValue::make_Deref({ box$(lv.clone()) }), ose.inner_states[0], old_state); + auto& ose = old_state.as_Partial(); + if( is_enum ) { + auto ilv = ::MIR::LValue::make_Downcast({ box$(lv.clone()), 0 }); + for(size_t i = 0; i < ose.inner_states.size(); i ++) + { + merge_state(sp, builder, ilv, ose.inner_states[i], nse.inner_states[i]); + ilv.as_Downcast().variant_index ++; + } } else { + auto ilv = ::MIR::LValue::make_Field({ box$(lv.clone()), 0 }); for(unsigned int i = 0; i < ose.inner_states.size(); i ++ ) { - merge_state(sp, builder, ::MIR::LValue::make_Field({ box$(lv.clone()), i }), ose.inner_states[i], old_state); + merge_state(sp, builder, ilv, ose.inner_states[i], nse.inner_states[i]); + ilv.as_Field().field_index ++; } } - old_state = mv$(out); } return; } break; + // Optional <= ... case VarState::TAG_Optional: switch( new_state.tag() ) { @@ -863,46 +1219,134 @@ namespace if( old_state.as_Optional() != new_state.as_Optional() ) { #if 1 builder.push_stmt_set_dropflag_other(sp, old_state.as_Optional(), new_state.as_Optional()); + builder.push_stmt_set_dropflag_default(sp, new_state.as_Optional()); #else // TODO: Rewrite history replacing one flag with another (if they have the same default) #endif } return ; - case VarState::TAG_Partial: - TODO(sp, "Handle Optional->Partial in split scope"); + case VarState::TAG_MovedOut: + TODO(sp, "Handle Optional->MovedOut in split scope"); + case VarState::TAG_Partial: { + const auto& nse = new_state.as_Partial(); + bool is_enum = false; + builder.with_val_type(sp, lv, [&](const auto& ty){ + assert( !builder.is_type_owned_box(ty) ); + is_enum = ty.m_data.is_Path() && ty.m_data.as_Path().binding.is_Enum(); + }); + // Create a Partial filled with copies of the Optional + { + ::std::vector<VarState> inner; + inner.reserve( nse.inner_states.size() ); + for(size_t i = 0; i < nse.inner_states.size(); i ++) + inner.push_back(old_state.clone()); + old_state = VarState::make_Partial({ mv$(inner) }); + } + auto& ose = old_state.as_Partial(); + // Propagate to inners + if( is_enum ) { + for(size_t i = 0; i < ose.inner_states.size(); i ++) + { + merge_state(sp, builder, ::MIR::LValue::make_Downcast({ box$(lv.clone()), static_cast<unsigned int>(i) }), ose.inner_states[i], nse.inner_states[i]); + } + } + else { + for(unsigned int i = 0; i < ose.inner_states.size(); i ++ ) + { + merge_state(sp, builder, ::MIR::LValue::make_Field({ box$(lv.clone()), i }), ose.inner_states[i], nse.inner_states[i]); + } + } + return; } } break; - case VarState::TAG_Partial: { - auto& ose = old_state.as_Partial(); + case VarState::TAG_MovedOut: { + auto& ose = old_state.as_MovedOut(); bool is_box = false; - builder.with_val_type(sp, lv, [&](const auto& ty){ is_box = builder.is_type_owned_box(ty); }); - if( is_box ) { - ASSERT_BUG(sp, ose.inner_states.size() == 1, ""); + builder.with_val_type(sp, lv, [&](const auto& ty){ + is_box = builder.is_type_owned_box(ty); + }); + if( !is_box ) { + BUG(sp, "MovedOut on non-Box"); } - // Need to tag for conditional shallow drop? Or just do that at the end of the split? - // - End of the split means that the only optional state is outer drop. switch( new_state.tag() ) { case VarState::TAGDEAD: throw ""; case VarState::TAG_Invalid: - if( ose.outer_flag != ~0u ) { - // Set the outer flag to `false` if its default isn't false - if( builder.get_drop_flag_default(sp, ose.outer_flag) != false ) { + case VarState::TAG_Valid: { + bool is_valid = new_state.is_Valid(); + if( ose.outer_flag == ~0u ) + { + // If not valid in new arm, then the outer state is conditional + if( !is_valid ) + { + ose.outer_flag = builder.new_drop_flag(true); builder.push_stmt_set_dropflag_val(sp, ose.outer_flag, false); } } - if( 0 ) - // - Fall through - case VarState::TAG_Valid: - if( ose.outer_flag != ~0u ) { - // Set the outer flag to `true` if its default isn't true - if( builder.get_drop_flag_default(sp, ose.outer_flag) != true ) { - builder.push_stmt_set_dropflag_val(sp, ose.outer_flag, true); + else + { + builder.push_stmt_set_dropflag_val(sp, ose.outer_flag, is_valid); + } + + merge_state(sp, builder, ::MIR::LValue::make_Deref({ box$(lv.clone()) }), *ose.inner_state, new_state); + return ; } + case VarState::TAG_Optional: { + const auto& nse = new_state.as_Optional(); + if( ose.outer_flag == ~0u ) + { + if( ! builder.get_drop_flag_default(sp, nse) ) + { + // Default wasn't true, need to make a new flag that does have a default of true + auto new_flag = builder.new_drop_flag(true); + builder.push_stmt_set_dropflag_other(sp, new_flag, nse); + builder.push_stmt_set_dropflag_default(sp, nse); + ose.outer_flag = new_flag; + } + else + { + ose.outer_flag = nse; } } - - if( is_box ) { - merge_state(sp, builder, ::MIR::LValue::make_Deref({ box$(lv.clone()) }), ose.inner_states[0], new_state); + else + { + // In this arm, assign the outer state to this drop flag + builder.push_stmt_set_dropflag_other(sp, ose.outer_flag, nse); + builder.push_stmt_set_dropflag_default(sp, nse); + } + merge_state(sp, builder, ::MIR::LValue::make_Deref({ box$(lv.clone()) }), *ose.inner_state, new_state); + return; } + case VarState::TAG_MovedOut: { + const auto& nse = new_state.as_MovedOut(); + if( ose.outer_flag != nse.outer_flag ) + { + TODO(sp, "Handle mismatched flags in MovedOut"); + } + merge_state(sp, builder, ::MIR::LValue::make_Deref({ box$(lv.clone()) }), *ose.inner_state, *nse.inner_state); + return; } + case VarState::TAG_Partial: + BUG(sp, "MovedOut->Partial not valid"); + } + break; } + case VarState::TAG_Partial: { + auto& ose = old_state.as_Partial(); + bool is_enum = false; + builder.with_val_type(sp, lv, [&](const auto& ty){ + assert( !builder.is_type_owned_box(ty) ); + is_enum = ty.m_data.is_Path() && ty.m_data.as_Path().binding.is_Enum(); + }); + // Need to tag for conditional shallow drop? Or just do that at the end of the split? + // - End of the split means that the only optional state is outer drop. + switch( new_state.tag() ) + { + case VarState::TAGDEAD: throw ""; + case VarState::TAG_Invalid: + case VarState::TAG_Valid: + case VarState::TAG_Optional: + if( is_enum ) { + for(size_t i = 0; i < ose.inner_states.size(); i ++) + { + merge_state(sp, builder, ::MIR::LValue::make_Downcast({ box$(lv.clone()), static_cast<unsigned int>(i) }), ose.inner_states[i], new_state); + } } else { for(unsigned int i = 0; i < ose.inner_states.size(); i ++ ) @@ -911,17 +1355,16 @@ namespace } } return ; - case VarState::TAG_Optional: { - //auto flag_idx = new_state.as_Optional(); - TODO(sp, "Handle Partial->Optional in split scope"); - } return; + case VarState::TAG_MovedOut: + BUG(sp, "Partial->MovedOut not valid"); case VarState::TAG_Partial: { const auto& nse = new_state.as_Partial(); ASSERT_BUG(sp, ose.inner_states.size() == nse.inner_states.size(), "Partial->Partial with mismatched sizes - " << old_state << " <= " << new_state); - ASSERT_BUG(sp, ose.outer_flag == nse.outer_flag, "Partial->Partial with mismatched drop flags - " << old_state << " <= " << new_state); - if( is_box ) { - ASSERT_BUG(sp, nse.inner_states.size() == 1, ""); - merge_state(sp, builder, ::MIR::LValue::make_Deref({ box$(lv.clone()) }), ose.inner_states[0], nse.inner_states[0]); + if( is_enum ) { + for(size_t i = 0; i < ose.inner_states.size(); i ++) + { + merge_state(sp, builder, ::MIR::LValue::make_Downcast({ box$(lv.clone()), static_cast<unsigned int>(i) }), ose.inner_states[i], nse.inner_states[i]); + } } else { for(unsigned int i = 0; i < ose.inner_states.size(); i ++ ) @@ -983,11 +1426,11 @@ void MirBuilder::terminate_loop_early(const Span& sp, ScopeType::Data_Loop& sd_l void MirBuilder::end_split_arm(const Span& sp, const ScopeHandle& handle, bool reachable) { - ASSERT_BUG(sp, handle.idx < m_scopes.size(), ""); + ASSERT_BUG(sp, handle.idx < m_scopes.size(), "Handle passed to end_split_arm is invalid"); auto& sd = m_scopes.at( handle.idx ); - ASSERT_BUG(sp, sd.data.is_Split(), ""); + ASSERT_BUG(sp, sd.data.is_Split(), "Ending split arm on non-Split arm - " << sd.data.tag_str()); auto& sd_split = sd.data.as_Split(); - ASSERT_BUG(sp, !sd_split.arms.empty(), ""); + ASSERT_BUG(sp, !sd_split.arms.empty(), "Split arm list is empty (impossible)"); TRACE_FUNCTION_F("end split scope " << handle.idx << " arm " << (sd_split.arms.size()-1)); if( reachable ) @@ -1092,10 +1535,10 @@ void MirBuilder::complete_scope(ScopeDef& sd) TU_MATCHA( (sd.data), (e), (Temporaries, - DEBUG("Temporaries " << e.temporaries); + DEBUG("Temporaries - " << e.temporaries); ), (Variables, - DEBUG("Variables " << e.vars); + DEBUG("Variables - " << e.vars); ), (Loop, DEBUG("Loop"); @@ -1282,37 +1725,62 @@ void MirBuilder::with_val_type(const Span& sp, const ::MIR::LValue& val, ::std:: BUG(sp, "Downcast on unexpected type - " << ty); ), (Path, - //ASSERT_BUG(sp, !te.binding.is_Unbound(), "Unbound path " << ty << " encountered"); - ASSERT_BUG(sp, te.binding.is_Enum(), "Downcast on non-Enum - " << ty << " for " << val); - const auto& enm = *te.binding.as_Enum(); - const auto& variants = enm.m_variants; - ASSERT_BUG(sp, e.variant_index < variants.size(), "Variant index out of range"); - const auto& variant = variants[e.variant_index]; - // TODO: Make data variants refer to associated types (unify enum and struct handling) - TU_MATCHA( (variant.second), (ve), - (Value, - ), - (Unit, - ), - (Tuple, - // HACK! Create tuple. - ::std::vector< ::HIR::TypeRef> tys; - for(const auto& fld : ve) - tys.push_back( monomorphise_type(sp, enm.m_params, te.path.m_data.as_Generic().m_params, fld.ent) ); - ::HIR::TypeRef tup( mv$(tys) ); - m_resolve.expand_associated_types(sp, tup); - cb(tup); - ), - (Struct, - // HACK! Create tuple. - ::std::vector< ::HIR::TypeRef> tys; - for(const auto& fld : ve) - tys.push_back( monomorphise_type(sp, enm.m_params, te.path.m_data.as_Generic().m_params, fld.second.ent) ); - ::HIR::TypeRef tup( mv$(tys) ); - m_resolve.expand_associated_types(sp, tup); - cb(tup); + // TODO: Union? + if( const auto* pbe = te.binding.opt_Enum() ) + { + const auto& enm = **pbe; + const auto& variants = enm.m_variants; + ASSERT_BUG(sp, e.variant_index < variants.size(), "Variant index out of range"); + const auto& variant = variants[e.variant_index]; + // TODO: Make data variants refer to associated types (unify enum and struct handling) + TU_MATCHA( (variant.second), (ve), + (Value, + DEBUG(""); + cb(::HIR::TypeRef::new_unit()); + ), + (Unit, + cb(::HIR::TypeRef::new_unit()); + ), + (Tuple, + // HACK! Create tuple. + ::std::vector< ::HIR::TypeRef> tys; + for(const auto& fld : ve) + tys.push_back( monomorphise_type(sp, enm.m_params, te.path.m_data.as_Generic().m_params, fld.ent) ); + ::HIR::TypeRef tup( mv$(tys) ); + m_resolve.expand_associated_types(sp, tup); + cb(tup); + ), + (Struct, + // HACK! Create tuple. + ::std::vector< ::HIR::TypeRef> tys; + for(const auto& fld : ve) + tys.push_back( monomorphise_type(sp, enm.m_params, te.path.m_data.as_Generic().m_params, fld.second.ent) ); + ::HIR::TypeRef tup( mv$(tys) ); + m_resolve.expand_associated_types(sp, tup); + cb(tup); + ) ) - ) + } + else if( const auto* pbe = te.binding.opt_Union() ) + { + const auto& unm = **pbe; + ASSERT_BUG(sp, e.variant_index < unm.m_variants.size(), "Variant index out of range"); + const auto& variant = unm.m_variants.at(e.variant_index); + const auto& fld = variant.second; + + if( monomorphise_type_needed(fld.ent) ) { + auto sty = monomorphise_type(sp, unm.m_params, te.path.m_data.as_Generic().m_params, fld.ent); + m_resolve.expand_associated_types(sp, sty); + cb(sty); + } + else { + cb(fld.ent); + } + } + else + { + BUG(sp, "Downcast on non-Enum/Union - " << ty << " for " << val); + } ) ) }); @@ -1327,185 +1795,380 @@ bool MirBuilder::lvalue_is_copy(const Span& sp, const ::MIR::LValue& val) const DEBUG("[lvalue_is_copy] ty="<<ty); rv = (m_resolve.type_is_copy(sp, ty) ? 2 : 1); }); - assert(rv != 0); + ASSERT_BUG(sp, rv != 0, "Type for " << val << " can't be determined"); return rv == 2; } -const VarState& MirBuilder::get_variable_state(const Span& sp, unsigned int idx, unsigned int skip_count) const +const VarState& MirBuilder::get_slot_state(const Span& sp, VarGroup ty, unsigned int idx, unsigned int skip_count/*=0*/) const { + // 1. Find an applicable Split scope for( auto scope_idx : ::reverse(m_scope_stack) ) { const auto& scope_def = m_scopes.at(scope_idx); TU_MATCH_DEF( ScopeType, (scope_def.data), (e), ( ), + (Temporaries, + if( ty == VarGroup::Temporary ) + { + auto it = ::std::find(e.temporaries.begin(), e.temporaries.end(), idx); + if( it != e.temporaries.end() ) { + break ; + } + } + ), (Variables, - auto it = ::std::find(e.vars.begin(), e.vars.end(), idx); - if( it != e.vars.end() ) { - // If controlled by this block, exit early (won't find it elsewhere) - break ; + if( ty == VarGroup::Variable ) + { + auto it = ::std::find(e.vars.begin(), e.vars.end(), idx); + if( it != e.vars.end() ) { + // If controlled by this block, exit early (won't find it elsewhere) + break ; + } } ), (Split, const auto& cur_arm = e.arms.back(); - auto it = cur_arm.var_states.find(idx); - if( it != cur_arm.var_states.end() ) + if( ty == VarGroup::Variable ) { - if( ! skip_count -- ) + auto it = cur_arm.var_states.find(idx); + if( it != cur_arm.var_states.end() ) { - return it->second; + if( ! skip_count -- ) + { + return it->second; + } + } + } + else if( ty == VarGroup::Temporary ) + { + auto it = cur_arm.tmp_states.find(idx); + if( it != cur_arm.tmp_states.end() ) + { + if( ! skip_count -- ) + { + return it->second; + } } } ) ) } - - ASSERT_BUG(sp, idx < m_variable_states.size(), "Variable " << idx << " out of range for state table"); - return m_variable_states[idx]; + switch(ty) + { + case VarGroup::Return: + return m_return_state; + case VarGroup::Argument: + ASSERT_BUG(sp, idx < m_arg_states.size(), "Argument " << idx << " out of range for state table"); + return m_arg_states.at(idx); + case VarGroup::Variable: + ASSERT_BUG(sp, idx < m_variable_states.size(), "Variable " << idx << " out of range for state table"); + return m_variable_states[idx]; + case VarGroup::Temporary: + ASSERT_BUG(sp, idx < m_temporary_states.size(), "Temporary " << idx << " out of range for state table"); + return m_temporary_states[idx]; + } + BUG(sp, "Fell off the end of get_slot_state"); } -VarState& MirBuilder::get_variable_state_mut(const Span& sp, unsigned int idx) +VarState& MirBuilder::get_slot_state_mut(const Span& sp, VarGroup ty, unsigned int idx) { VarState* ret = nullptr; for( auto scope_idx : ::reverse(m_scope_stack) ) { auto& scope_def = m_scopes.at(scope_idx); - if( scope_def.data.is_Variables() ) + if( const auto* e = scope_def.data.opt_Variables() ) { - const auto& e = scope_def.data.as_Variables(); - auto it = ::std::find(e.vars.begin(), e.vars.end(), idx); - if( it != e.vars.end() ) { - break ; + if( ty == VarGroup::Variable ) + { + auto it = ::std::find(e->vars.begin(), e->vars.end(), idx); + if( it != e->vars.end() ) { + break ; + } + } + } + else if( const auto* e = scope_def.data.opt_Temporaries() ) + { + if( ty == VarGroup::Temporary ) + { + auto it = ::std::find(e->temporaries.begin(), e->temporaries.end(), idx); + if( it != e->temporaries.end() ) { + break ; + } } } else if( scope_def.data.is_Split() ) { + auto& e = scope_def.data.as_Split(); + auto& cur_arm = e.arms.back(); if( ! ret ) { - auto& e = scope_def.data.as_Split(); - auto& cur_arm = e.arms.back(); - auto it = cur_arm.var_states.find(idx); - if( it == cur_arm.var_states.end() ) + ::std::map<unsigned int, VarState>* states; + switch(ty) { - DEBUG("Split new (scope " << scope_idx << ")"); - ret = &(cur_arm.var_states[idx] = get_variable_state(sp, idx).clone()); + case VarGroup::Return: states = nullptr; break; + case VarGroup::Argument: BUG(sp, "Mutating state of argument"); break; + case VarGroup::Variable: states = &cur_arm.var_states; break; + case VarGroup::Temporary: states = &cur_arm.tmp_states; break; } - else + + if( states ) { - DEBUG("Split existing (scope " << scope_idx << ")"); - ret = &it->second; + auto it = states->find(idx); + if( it == states->end() ) + { + DEBUG("Split new (scope " << scope_idx << ")"); + ret = &( (*states)[idx] = get_slot_state(sp, ty, idx).clone() ); + } + else + { + DEBUG("Split existing (scope " << scope_idx << ")"); + ret = &it->second; + } } } } else if( scope_def.data.is_Loop() ) { auto& e = scope_def.data.as_Loop(); - if( e.changed_vars.count(idx) == 0 ) + ::std::map<unsigned int, VarState>* states = nullptr; + switch(ty) { - auto state = e.exit_state_valid ? get_variable_state(sp, idx).clone() : VarState::make_Valid({}); - e.changed_vars.insert(::std::make_pair( idx, mv$(state) )); + case VarGroup::Return: states = nullptr; break; + case VarGroup::Argument: BUG(sp, "Mutating state of argument"); break; + case VarGroup::Variable: states = &e.changed_vars; break; + case VarGroup::Temporary: states = &e.changed_tmps; break; + } + + if( states ) + { + if( states->count(idx) == 0 ) + { + auto state = e.exit_state_valid ? get_slot_state(sp, ty, idx).clone() : VarState::make_Valid({}); + states->insert(::std::make_pair( idx, mv$(state) )); + } } } else { } } - - if( !ret ) + if( ret ) { - DEBUG("Outer"); - ASSERT_BUG(sp, idx < m_variable_states.size(), "Variable " << idx << " out of range for state table"); - return m_variable_states[idx]; + return *ret; } else { - return *ret; + switch(ty) + { + case VarGroup::Return: + return m_return_state; + case VarGroup::Argument: + ASSERT_BUG(sp, idx < m_arg_states.size(), "Argument " << idx << " out of range for state table"); + return m_arg_states.at(idx); + case VarGroup::Variable: + ASSERT_BUG(sp, idx < m_variable_states.size(), "Variable " << idx << " out of range for state table"); + return m_variable_states[idx]; + case VarGroup::Temporary: + ASSERT_BUG(sp, idx < m_temporary_states.size(), "Temporary " << idx << " out of range for state table"); + return m_temporary_states[idx]; + } + BUG(sp, "Fell off the end of get_slot_state_mut"); } } +const VarState& MirBuilder::get_variable_state(const Span& sp, unsigned int idx, unsigned int skip_count) const +{ + return get_slot_state(sp, VarGroup::Variable, idx, skip_count); +} +VarState& MirBuilder::get_variable_state_mut(const Span& sp, unsigned int idx) +{ + return get_slot_state_mut(sp, VarGroup::Variable, idx); +} const VarState& MirBuilder::get_temp_state(const Span& sp, unsigned int idx, unsigned int skip_count) const { - for( auto scope_idx : ::reverse(m_scope_stack) ) - { - const auto& scope_def = m_scopes.at(scope_idx); - if( scope_def.data.is_Temporaries() ) + return get_slot_state(sp, VarGroup::Temporary, idx, skip_count); +} +VarState& MirBuilder::get_temp_state_mut(const Span& sp, unsigned int idx) +{ + return get_slot_state_mut(sp, VarGroup::Temporary, idx); +} + +const VarState& MirBuilder::get_val_state(const Span& sp, const ::MIR::LValue& lv, unsigned int skip_count) +{ + TODO(sp, ""); +} +VarState& MirBuilder::get_val_state_mut(const Span& sp, const ::MIR::LValue& lv) +{ + TRACE_FUNCTION_F(lv); + TU_MATCHA( (lv), (e), + (Variable, + return get_slot_state_mut(sp, VarGroup::Variable, e); + ), + (Temporary, + return get_slot_state_mut(sp, VarGroup::Temporary, e.idx); + ), + (Argument, + return get_slot_state_mut(sp, VarGroup::Argument, e.idx); + ), + (Static, + BUG(sp, "Attempting to mutate state of a static"); + ), + (Return, + BUG(sp, "Move of return value"); + return get_slot_state_mut(sp, VarGroup::Return, 0); + ), + (Field, + auto& ivs = get_val_state_mut(sp, *e.val); + VarState tpl; + TU_MATCHA( (ivs), (ivse), + (Invalid, + //BUG(sp, "Mutating inner state of an invalidated composite - " << lv); + tpl = VarState::make_Valid({}); + ), + (MovedOut, + BUG(sp, "Field on value with MovedOut state - " << lv); + ), + (Partial, + ), + (Optional, + tpl = ivs.clone(); + ), + (Valid, + tpl = VarState::make_Valid({}); + ) + ) + if( !ivs.is_Partial() ) { - const auto& e = scope_def.data.as_Temporaries(); - auto it = ::std::find(e.temporaries.begin(), e.temporaries.end(), idx); - if( it != e.temporaries.end() ) { - break ; - } + size_t n_flds = 0; + with_val_type(sp, *e.val, [&](const auto& ty) { + DEBUG("ty = " << ty); + if(const auto* e = ty.m_data.opt_Path()) { + ASSERT_BUG(sp, e->binding.is_Struct(), ""); + const auto& str = *e->binding.as_Struct(); + TU_MATCHA( (str.m_data), (se), + (Unit, + BUG(sp, "Field access of unit-like struct"); + ), + (Tuple, + n_flds = se.size(); + ), + (Named, + n_flds = se.size(); + ) + ) + } + else if(const auto* e = ty.m_data.opt_Tuple()) { + n_flds = e->size(); + } + else if(const auto* e = ty.m_data.opt_Array()) { + n_flds = e->size_val; + } + else { + TODO(sp, "Determine field count for " << ty); + } + }); + ::std::vector<VarState> inner_vs; inner_vs.reserve(n_flds); + for(size_t i = 0; i < n_flds; i++) + inner_vs.push_back( tpl.clone() ); + ivs = VarState::make_Partial({ mv$(inner_vs) }); } - else if( scope_def.data.is_Split() ) + return ivs.as_Partial().inner_states.at(e.field_index); + ), + (Deref, + // HACK: If the dereferenced type is a Box ("owned_box") then hack in move and shallow drop + bool is_box = false; + if( this->m_lang_Box ) { - const auto& e = scope_def.data.as_Split(); - const auto& cur_arm = e.arms.back(); - auto it = cur_arm.tmp_states.find(idx); - if( it != cur_arm.tmp_states.end() ) - { - if( ! skip_count -- ) - { - return it->second; - } - } + with_val_type(sp, *e.val, [&](const auto& ty){ + DEBUG("ty = " << ty); + is_box = this->is_type_owned_box(ty); + }); } - } - ASSERT_BUG(sp, idx < m_temporary_states.size(), "Temporary " << idx << " out of range for state table"); - return m_temporary_states[idx]; -} -VarState& MirBuilder::get_temp_state_mut(const Span& sp, unsigned int idx) -{ - VarState* ret = nullptr; - for( auto scope_idx : ::reverse(m_scope_stack) ) - { - auto& scope_def = m_scopes.at(scope_idx); - if( scope_def.data.is_Temporaries() ) + + if( is_box ) { - const auto& e = scope_def.data.as_Temporaries(); - auto it = ::std::find(e.temporaries.begin(), e.temporaries.end(), idx); - if( it != e.temporaries.end() ) { - break ; + ::MIR::LValue inner_lv; + // 1. If the inner lvalue isn't a slot with move information, move out of the lvalue into a temporary (with standard temp scope) + TU_MATCH_DEF( ::MIR::LValue, (*e.val), (ei), + ( + with_val_type(sp, *e.val, [&](const auto& ty){ inner_lv = this->new_temporary(ty); }); + this->push_stmt_assign(sp, inner_lv.clone(), ::MIR::RValue( mv$(*e.val) )); + *e.val = inner_lv.clone(); + ), + (Variable, + inner_lv = ::MIR::LValue(ei); + ), + (Temporary, + inner_lv = ::MIR::LValue(ei); + ), + (Argument, + inner_lv = ::MIR::LValue(ei); + ) + ) + // 2. Mark the slot as requiring only a shallow drop + ::std::vector<VarState> inner; + inner.push_back(VarState::make_Valid({})); + auto& ivs = get_val_state_mut(sp, inner_lv); + if( ! ivs.is_MovedOut() ) + { + unsigned int drop_flag = (ivs.is_Optional() ? ivs.as_Optional() : ~0u); + ivs = VarState::make_MovedOut({ box$(VarState::make_Valid({})), drop_flag }); } + return *ivs.as_MovedOut().inner_state; } - else if( scope_def.data.is_Split() ) + else { - if( ! ret ) - { - auto& e = scope_def.data.as_Split(); - auto& cur_arm = e.arms.back(); - auto it = cur_arm.tmp_states.find(idx); - if(it == cur_arm.tmp_states.end()) + BUG(sp, "Move out of deref with non-Copy values - &move? - " << lv << " : " << FMT_CB(ss, this->with_val_type(sp, lv, [&](const auto& ty){ss<<ty;});) ); + } + ), + (Index, + BUG(sp, "Move out of index with non-Copy values - Partial move?"); + ), + (Downcast, + // TODO: What if the inner is Copy? What if the inner is a hidden pointer? + auto& ivs = get_val_state_mut(sp, *e.val); + //static VarState ivs; ivs = VarState::make_Valid({}); + + if( !ivs.is_Partial() ) + { + ASSERT_BUG(sp, !ivs.is_MovedOut(), "Downcast of a MovedOut value"); + + size_t var_count = 0; + with_val_type(sp, *e.val, [&](const auto& ty){ + DEBUG("ty = " << ty); + ASSERT_BUG(sp, ty.m_data.is_Path(), "Downcast on non-Path type - " << ty); + const auto& pb = ty.m_data.as_Path().binding; + // TODO: What about unions? + // - Iirc, you can't move out of them so they will never have state mutated + if( pb.is_Enum() ) { - ret = &(cur_arm.tmp_states[idx] = get_temp_state(sp, idx).clone()); + const auto& enm = *pb.as_Enum(); + var_count = enm.m_variants.size(); + } + else if( const auto* pbe = pb.opt_Union() ) + { + const auto& unm = **pbe; + var_count = unm.m_variants.size(); } else { - ret = &it->second; + BUG(sp, "Downcast on non-Enum/Union - " << ty); } - } - } - else if( scope_def.data.is_Loop() ) - { - auto& e = scope_def.data.as_Loop(); - if( e.changed_tmps.count(idx) == 0 ) + }); + + ::std::vector<VarState> inner; + for(size_t i = 0; i < var_count; i ++) { - auto state = e.exit_state_valid ? get_temp_state(sp, idx).clone() : VarState::make_Valid({}); - e.changed_tmps.insert(::std::make_pair( idx, mv$(state) )); + inner.push_back( VarState::make_Invalid(InvalidType::Uninit) ); } + inner[e.variant_index] = mv$(ivs); + ivs = VarState::make_Partial({ mv$(inner) }); } - else - { - } - } - if( !ret ) - { - ASSERT_BUG(sp, idx < m_temporary_states.size(), "Temporary " << idx << " out of range for state table"); - return m_temporary_states[idx]; - } - else - { - return *ret; - } + return ivs.as_Partial().inner_states.at(e.variant_index); + ) + ) + BUG(sp, "Fell off send of get_val_state_mut"); } void MirBuilder::drop_value_from_state(const Span& sp, const VarState& vs, ::MIR::LValue lv) @@ -1516,12 +2179,41 @@ void MirBuilder::drop_value_from_state(const Span& sp, const VarState& vs, ::MIR (Valid, push_stmt_drop(sp, mv$(lv)); ), + (MovedOut, + bool is_box = false; + with_val_type(sp, lv, [&](const auto& ty){ + is_box = this->is_type_owned_box(ty); + }); + if( is_box ) + { + drop_value_from_state(sp, *vse.inner_state, ::MIR::LValue::make_Deref({ box$(lv.clone()) })); + push_stmt_drop_shallow(sp, mv$(lv), vse.outer_flag); + } + else + { + TODO(sp, ""); + } + ), (Partial, - // TODO: Actual destructuring based on the type - with_val_type(sp, lv, [&](const auto& ty){ ASSERT_BUG(sp, this->is_type_owned_box(ty), "TODO: Partial on non-Box"); }); - assert( vse.inner_states.size() == 1 ); - drop_value_from_state(sp, vse.inner_states[0], ::MIR::LValue::make_Deref({ box$(lv.clone()) })); - push_stmt_drop_shallow(sp, mv$(lv), vse.outer_flag); + bool is_enum = false; + with_val_type(sp, lv, [&](const auto& ty){ + is_enum = ty.m_data.is_Path() && ty.m_data.as_Path().binding.is_Enum(); + }); + if(is_enum) + { + DEBUG("TODO: Switch based on enum value"); + //for(size_t i = 0; i < vse.inner_states.size(); i ++) + //{ + // drop_value_from_state(sp, vse.inner_states[i], ::MIR::LValue::make_Downcast({ box$(lv.clone()), static_cast<unsigned int>(i) })); + //} + } + else + { + for(size_t i = 0; i < vse.inner_states.size(); i ++) + { + drop_value_from_state(sp, vse.inner_states[i], ::MIR::LValue::make_Field({ box$(lv.clone()), static_cast<unsigned int>(i) })); + } + } ), (Optional, push_stmt_drop(sp, mv$(lv), vse); @@ -1536,6 +2228,7 @@ void MirBuilder::drop_scope_values(const ScopeDef& sd) for(auto tmp_idx : ::reverse(e.temporaries)) { const auto& vs = get_temp_state(sd.span, tmp_idx); + DEBUG("tmp" << tmp_idx << " - " << vs); drop_value_from_state( sd.span, vs, ::MIR::LValue::make_Temporary({ tmp_idx }) ); } ), @@ -1543,6 +2236,7 @@ void MirBuilder::drop_scope_values(const ScopeDef& sd) for(auto var_idx : ::reverse(e.vars)) { const auto& vs = get_variable_state(sd.span, var_idx); + DEBUG("var" << var_idx << " - " << vs); drop_value_from_state( sd.span, vs, ::MIR::LValue::make_Variable(var_idx) ); } ), @@ -1555,109 +2249,13 @@ void MirBuilder::drop_scope_values(const ScopeDef& sd) ) } + void MirBuilder::moved_lvalue(const Span& sp, const ::MIR::LValue& lv) { - TRACE_FUNCTION_F(lv); - TU_MATCHA( (lv), (e), - (Variable, - if( !lvalue_is_copy(sp, lv) ) { - get_variable_state_mut(sp, e) = VarState::make_Invalid(InvalidType::Moved); - } - ), - (Temporary, - if( !lvalue_is_copy(sp, lv) ) { - get_temp_state_mut(sp, e.idx) = VarState::make_Invalid(InvalidType::Moved); - } - ), - (Argument, - //TODO(sp, "Mark argument as moved"); - ), - (Static, - //TODO(sp, "Static - Assert that type is Copy"); - ), - (Return, - BUG(sp, "Read of return value"); - ), - (Field, - if( lvalue_is_copy(sp, lv) ) { - } - else { - // TODO: Partial moves. - moved_lvalue(sp, *e.val); - } - ), - (Deref, - if( lvalue_is_copy(sp, lv) ) { - } - else { - // HACK: If the dereferenced type is a Box ("owned_box") then hack in move and shallow drop - if( this->m_lang_Box ) - { - bool is_box = false; - with_val_type(sp, *e.val, [&](const auto& ty){ - DEBUG("ty = " << ty); - is_box = this->is_type_owned_box(ty); - }); - if( is_box ) - { - ::MIR::LValue inner_lv; - // 1. If the inner lvalue isn't a slot with move information, move out of the lvalue into a temporary (with standard temp scope) - TU_MATCH_DEF( ::MIR::LValue, (*e.val), (ei), - ( - with_val_type(sp, *e.val, [&](const auto& ty){ inner_lv = this->new_temporary(ty); }); - this->push_stmt_assign(sp, inner_lv.clone(), ::MIR::RValue( mv$(*e.val) )); - *e.val = inner_lv.clone(); - ), - (Variable, - inner_lv = ::MIR::LValue(ei); - ), - (Temporary, - inner_lv = ::MIR::LValue(ei); - ), - (Argument, - inner_lv = ::MIR::LValue(ei); - ) - ) - // 2. Mark the slot as requiring only a shallow drop - // - TODO: Have a drop flag attached to the - ::std::vector<VarState> ivs; - ivs.push_back(VarState::make_Invalid(InvalidType::Moved)); - TU_MATCH_DEF( ::MIR::LValue, (inner_lv), (ei), - ( - BUG(sp, "Box move out of invalid LValue " << inner_lv << " - should have been moved"); - ), - (Variable, - get_variable_state_mut(sp, ei) = VarState::make_Partial({ mv$(ivs), ~0u }); - ), - (Temporary, - get_temp_state_mut(sp, ei.idx) = VarState::make_Partial({ mv$(ivs), ~0u }); - ), - (Argument, - TODO(sp, "Mark arg " << ei.idx << " for shallow drop"); - ) - ) - // Early return! - return ; - } - } - BUG(sp, "Move out of deref with non-Copy values - &move? - " << lv << " : " << FMT_CB(ss, this->with_val_type(sp, lv, [&](const auto& ty){ss<<ty;});) ); - moved_lvalue(sp, *e.val); - } - ), - (Index, - if( lvalue_is_copy(sp, lv) ) { - } - else { - BUG(sp, "Move out of index with non-Copy values - Partial move?"); - moved_lvalue(sp, *e.val); - } - moved_lvalue(sp, *e.idx); - ), - (Downcast, - // TODO: What if the inner is Copy? What if the inner is a hidden pointer? - moved_lvalue(sp, *e.val); - ) - ) + if( !lvalue_is_copy(sp, lv) ) { + auto& vs = get_val_state_mut(sp, lv); + vs = VarState::make_Invalid(InvalidType::Moved); + } } const ::MIR::LValue& MirBuilder::get_ptr_to_dst(const Span& sp, const ::MIR::LValue& lv) const @@ -1702,12 +2300,15 @@ VarState VarState::clone() const (Optional, return VarState(e); ), + (MovedOut, + return VarState::make_MovedOut({ box$(e.inner_state->clone()), e.outer_flag }); + ), (Partial, ::std::vector<VarState> n; n.reserve(e.inner_states.size()); for(const auto& a : e.inner_states) n.push_back( a.clone() ); - return VarState::make_Partial({ mv$(n), e.outer_flag }); + return VarState::make_Partial({ mv$(n) }); ) ) throw ""; @@ -1726,9 +2327,12 @@ bool VarState::operator==(VarState& x) const (Optional, return te == xe; ), - (Partial, + (MovedOut, if( te.outer_flag != xe.outer_flag ) return false; + return *te.inner_state == *xe.inner_state; + ), + (Partial, if( te.inner_states.size() != xe.inner_states.size() ) return false; for(unsigned int i = 0; i < te.inner_states.size(); i ++) @@ -1758,12 +2362,16 @@ bool VarState::operator==(VarState& x) const (Optional, os << "Optional(" << e << ")"; ), - (Partial, - os << "Partial("; + (MovedOut, + os << "MovedOut("; if( e.outer_flag == ~0u ) os << "-"; else os << "df" << e.outer_flag; + os << " " << *e.inner_state <<")"; + ), + (Partial, + os << "Partial("; os << ", [" << e.inner_states << "])"; ) ) diff --git a/src/mir/operations.hpp b/src/mir/operations.hpp index cdc9c00b..1c06bc8c 100644 --- a/src/mir/operations.hpp +++ b/src/mir/operations.hpp @@ -10,6 +10,8 @@ // Check that the MIR is well-formed extern void MIR_Validate(const StaticTraitResolve& resolve, const ::HIR::ItemPath& path, const ::MIR::Function& fcn, const ::HIR::Function::args_t& args, const ::HIR::TypeRef& ret_type); +// - +extern void MIR_Validate_Full(const StaticTraitResolve& resolve, const ::HIR::ItemPath& path, const ::MIR::Function& fcn, const ::HIR::Function::args_t& args, const ::HIR::TypeRef& ret_type); // Perform needed changes to the generated MIR (virtualisation, Unsize/CoerceUnsize, ...) extern void MIR_Cleanup(const StaticTraitResolve& resolve, const ::HIR::ItemPath& path, ::MIR::Function& fcn, const ::HIR::Function::args_t& args, const ::HIR::TypeRef& ret_type); // Optimise the MIR diff --git a/src/mir/optimise.cpp b/src/mir/optimise.cpp index cb96cba0..d7379526 100644 --- a/src/mir/optimise.cpp +++ b/src/mir/optimise.cpp @@ -13,6 +13,16 @@ #include <mir/helpers.hpp> #include <mir/operations.hpp> #include <mir/visit_crate_mir.hpp> +#include <algorithm> +#include <iomanip> +#include <trans/target.hpp> + +#define DUMP_BEFORE_ALL 0 +#define DUMP_BEFORE_CONSTPROPAGATE 0 +#define CHECK_AFTER_PASS 0 + +#define DUMP_AFTER_DONE 0 +#define CHECK_AFTER_DONE 1 namespace { ::MIR::BasicBlockId get_new_target(const ::MIR::TypeResolve& state, ::MIR::BasicBlockId bb) @@ -182,6 +192,8 @@ namespace { (Drop, // Well, it mutates... rv |= visit_mir_lvalue_mut(e.slot, ValUsage::Write, cb); + ), + (ScopeEnd, ) ) return rv; @@ -269,7 +281,7 @@ namespace { ), (UfcsKnown, TRACE_FUNCTION_F(path); - + // Obtain trait pointer (for default impl and to know what the item type is) const auto& trait_ref = state.m_resolve.m_crate.get_trait_by_path(state.sp, pe.trait.m_path); auto trait_vi_it = trait_ref.m_values.find(pe.item); @@ -316,7 +328,7 @@ namespace { } return false; }); - + if( bound_found ) { return nullptr; } @@ -330,6 +342,7 @@ namespace { params.self_ty = &*pe.type; params.fcn_params = &pe.params; + // Search for the method in the impl auto fit = impl.m_methods.find(pe.item); if( fit != impl.m_methods.end() ) { @@ -424,35 +437,46 @@ void MIR_Optimise(const StaticTraitResolve& resolve, const ::HIR::ItemPath& path while( MIR_Optimise_PropagateSingleAssignments(state, fcn) ) change_happened = true; + change_happened |= MIR_Optimise_UnifyBlocks(state, fcn); + // >> Unify duplicate temporaries // If two temporaries don't overlap in lifetime (blocks in which they're valid), unify the two change_happened |= MIR_Optimise_UnifyTemporaries(state, fcn); // >> Combine Duplicate Blocks change_happened |= MIR_Optimise_UnifyBlocks(state, fcn); - #if 0 if( change_happened ) { - //MIR_Dump_Fcn(::std::cout, fcn); + #if DUMP_AFTER_PASS + if( debug_enabled() ) { + MIR_Dump_Fcn(::std::cout, fcn); + } + #endif + #if CHECK_AFTER_PASS MIR_Validate(resolve, path, fcn, args, ret_type); + #endif } - #endif MIR_Optimise_GarbageCollect_Partial(state, fcn); pass_num += 1; } while( change_happened ); - #if 1 + #if DUMP_AFTER_DONE if( debug_enabled() ) { MIR_Dump_Fcn(::std::cout, fcn); } #endif + #if CHECK_AFTER_DONE // DEFENCE: Run validation _before_ GC (so validation errors refer to the pre-gc numbers) MIR_Validate(resolve, path, fcn, args, ret_type); + #endif // GC pass on blocks and variables // - Find unused blocks, then delete and rewrite all references. MIR_Optimise_GarbageCollect(state, fcn); + + //MIR_Validate(resolve, path, fcn, args, ret_type); + //MIR_Validate_Full(resolve, path, fcn, args, ret_type); } // -------------------------------------------------------------------- @@ -463,6 +487,31 @@ bool MIR_Optimise_BlockSimplify(::MIR::TypeResolve& state, ::MIR::Function& fcn) // >> Replace targets that point to a block that is just a goto for(auto& block : fcn.blocks) { + // Unify sequential ScopeEnd statements + if( block.statements.size() > 1 ) + { + for(auto it = block.statements.begin() + 1; it != block.statements.end(); ) + { + if( (it-1)->is_ScopeEnd() && it->is_ScopeEnd() ) + { + auto& dst = (it-1)->as_ScopeEnd(); + const auto& src = it->as_ScopeEnd(); + DEBUG("Unify " << *(it-1) << " and " << *it); + for(auto v : src.vars) + dst.vars.push_back(v); + for(auto v : src.tmps) + dst.tmps.push_back(v); + ::std::sort(dst.vars.begin(), dst.vars.end()); + ::std::sort(dst.tmps.begin(), dst.tmps.end()); + it = block.statements.erase(it); + } + else + { + ++ it; + } + } + } + TU_MATCHA( (block.terminator), (e), (Incomplete, ), @@ -583,7 +632,7 @@ bool MIR_Optimise_BlockSimplify(::MIR::TypeResolve& state, ::MIR::Function& fcn) bool MIR_Optimise_Inlining(::MIR::TypeResolve& state, ::MIR::Function& fcn) { TRACE_FUNCTION; - + struct H { static bool can_inline(const ::HIR::Path& path, const ::MIR::Function& fcn) @@ -680,22 +729,21 @@ bool MIR_Optimise_Inlining(::MIR::TypeResolve& state, ::MIR::Function& fcn) return rv; } - ::MIR::BasicBlock clone_bb(const ::MIR::BasicBlock& src) const + ::MIR::BasicBlock clone_bb(const ::MIR::BasicBlock& src, unsigned src_idx, unsigned new_idx) const { ::MIR::BasicBlock rv; rv.statements.reserve( src.statements.size() ); for(const auto& stmt : src.statements) { + DEBUG("BB" << src_idx << "->BB" << new_idx << "/" << rv.statements.size() << ": " << stmt); TU_MATCHA( (stmt), (se), (Assign, - DEBUG(se.dst << " = " << se.src); rv.statements.push_back( ::MIR::Statement::make_Assign({ this->clone_lval(se.dst), this->clone_rval(se.src) }) ); ), (Asm, - DEBUG("asm!"); rv.statements.push_back( ::MIR::Statement::make_Asm({ se.tpl, this->clone_name_lval_vec(se.outputs), @@ -705,7 +753,6 @@ bool MIR_Optimise_Inlining(::MIR::TypeResolve& state, ::MIR::Function& fcn) }) ); ), (SetDropFlag, - DEBUG("df" << se.idx << " = "); rv.statements.push_back( ::MIR::Statement::make_SetDropFlag({ this->df_base + se.idx, se.new_val, @@ -713,17 +760,27 @@ bool MIR_Optimise_Inlining(::MIR::TypeResolve& state, ::MIR::Function& fcn) }) ); ), (Drop, - DEBUG("drop " << se.slot); rv.statements.push_back( ::MIR::Statement::make_Drop({ se.kind, this->clone_lval(se.slot), se.flag_idx == ~0u ? ~0u : this->df_base + se.flag_idx }) ); + ), + (ScopeEnd, + ::MIR::Statement::Data_ScopeEnd new_se; + new_se.vars.reserve(se.vars.size()); + for(auto idx : se.vars) + new_se.vars.push_back(this->var_base + idx); + new_se.tmps.reserve(se.tmps.size()); + for(auto idx : se.tmps) + new_se.tmps.push_back(this->tmp_base + idx); + rv.statements.push_back(::MIR::Statement( mv$(new_se) )); ) ) } - DEBUG(src.terminator); + DEBUG("BB" << src_idx << "->BB" << new_idx << "/" << rv.statements.size() << ": " << src.terminator); rv.terminator = this->clone_term(src.terminator); + DEBUG("-> " << rv.terminator); return rv; } ::MIR::Terminator clone_term(const ::MIR::Terminator& src) const @@ -975,7 +1032,7 @@ bool MIR_Optimise_Inlining(::MIR::TypeResolve& state, ::MIR::Function& fcn) new_blocks.reserve( called_mir->blocks.size() ); for(const auto& bb : called_mir->blocks) { - new_blocks.push_back( cloner.clone_bb(bb) ); + new_blocks.push_back( cloner.clone_bb(bb, (&bb - called_mir->blocks.data()), fcn.blocks.size() + new_blocks.size()) ); } // > Append new temporaries for(auto& val : cloner.const_assignments) @@ -1000,6 +1057,7 @@ bool MIR_Optimise_Inlining(::MIR::TypeResolve& state, ::MIR::Function& fcn) return inline_happened; } + // -------------------------------------------------------------------- // If two temporaries don't overlap in lifetime (blocks in which they're valid), unify the two // -------------------------------------------------------------------- @@ -1030,249 +1088,9 @@ bool MIR_Optimise_UnifyTemporaries(::MIR::TypeResolve& state, ::MIR::Function& f return false; } - struct VarLifetime { - ::std::vector<bool> blocks; - - VarLifetime(const ::MIR::Function& fcn): - blocks(fcn.blocks.size()) - { - } - - bool is_valid() const { - for(auto v : blocks) - if( v ) - return true; - return false; - } - bool overlaps(const VarLifetime& x) const { - assert(blocks.size() == x.blocks.size()); - for(unsigned int i = 0; i < blocks.size(); i ++) - { - if( blocks[i] && x.blocks[i] ) - return true; - } - return false; - } - void unify(const VarLifetime& x) { - assert(blocks.size() == x.blocks.size()); - for(unsigned int i = 0; i < blocks.size(); i ++) - { - if( x.blocks[i] ) - blocks[i] = true; - } - } - }; - //::std::vector<VarLifetime> var_lifetimes; - ::std::vector<VarLifetime> tmp_lifetimes( fcn.temporaries.size(), VarLifetime(fcn) ); - - // 1. Calculate lifetimes of all variables/temporaries that are eligable to be merged - // - Lifetime is from first write to last read. Borrows lead to the value being assumed to live forever - // - > BUT: Since this is lazy, it's taken as only being the lifetime of non-Copy items (as determined by the drop call or a move) - { - auto mark_borrowed = [&](const ::MIR::LValue& lv) { - if( const auto* ve = lv.opt_Temporary() ) { - replacable[ve->idx] = false; - } - // TODO: Recurse! - }; - - struct State { - //::std::vector<bool> vars; - ::std::vector<bool> tmps; - - State() {} - State(const ::MIR::Function& fcn): - tmps(fcn.temporaries.size()) - { - } - - bool merge(const State& other) { - if( tmps.size() == 0 ) - { - assert(other.tmps.size() != 0); - tmps = other.tmps; - return true; - } - else - { - assert(tmps.size() == other.tmps.size()); - bool rv = false; - for(unsigned int i = 0; i < tmps.size(); i ++) - { - if( tmps[i] != other.tmps[i] && other.tmps[i] ) { - tmps[i] = true; - rv = true; - } - } - return rv; - } - } - - void mark_validity(const ::MIR::TypeResolve& mir_res, const ::MIR::LValue& lv, bool val) { - if( const auto& ve = lv.opt_Temporary() ) { - tmps[ve->idx] = val; - } - else { - } - } - void move_val(const ::MIR::TypeResolve& mir_res, const ::MIR::LValue& lv) { - ::HIR::TypeRef tmp; - if( mir_res.m_resolve.type_is_copy( mir_res.sp, mir_res.get_lvalue_type(tmp, lv) ) ) { - } - else { - mark_validity(mir_res, lv, false); - } - } - void move_val(const ::MIR::TypeResolve& mir_res, const ::MIR::Param& p) { - if(const auto* e = p.opt_LValue()) - { - move_val(mir_res, *e); - } - } - }; - ::std::vector<State> block_states( fcn.blocks.size() ); - ::std::vector< ::std::pair<unsigned int, State> > to_visit; - auto add_to_visit = [&to_visit](unsigned int bb, State state) { - to_visit.push_back( ::std::make_pair(bb, mv$(state)) ); - }; - to_visit.push_back( ::std::make_pair(0, State(fcn)) ); - while( !to_visit.empty() ) - { - auto bb_idx = to_visit.back().first; - auto val_state = mv$(to_visit.back().second); - to_visit.pop_back(); - - // 1. Merge with block state - if( ! block_states[bb_idx].merge(val_state) ) - continue ; - //DEBUG("BB" << bb_idx); - - // 2. Run block - const auto& bb = fcn.blocks[bb_idx]; - for(unsigned int stmt_idx = 0; stmt_idx < bb.statements.size(); stmt_idx ++) - { - const auto& stmt = bb.statements[stmt_idx]; - state.set_cur_stmt(bb_idx, stmt_idx); - - switch( stmt.tag() ) - { - case ::MIR::Statement::TAGDEAD: - throw ""; - case ::MIR::Statement::TAG_SetDropFlag: - break; - case ::MIR::Statement::TAG_Drop: - val_state.mark_validity( state, stmt.as_Drop().slot, false ); - break; - case ::MIR::Statement::TAG_Asm: - for(const auto& v : stmt.as_Asm().outputs) - val_state.mark_validity( state, v.second, true ); - break; - case ::MIR::Statement::TAG_Assign: - // Check source (and invalidate sources) - TU_MATCH( ::MIR::RValue, (stmt.as_Assign().src), (se), - (Use, - val_state.move_val(state, se); - ), - (Constant, - ), - (SizedArray, - val_state.move_val(state, se.val); - ), - (Borrow, - mark_borrowed(se.val); - ), - (Cast, - ), - (BinOp, - ), - (UniOp, - ), - (DstMeta, - ), - (DstPtr, - ), - (MakeDst, - val_state.move_val(state, se.meta_val); - ), - (Tuple, - for(const auto& v : se.vals) - val_state.move_val(state, v); - ), - (Array, - for(const auto& v : se.vals) - val_state.move_val(state, v); - ), - (Variant, - val_state.move_val(state, se.val); - ), - (Struct, - for(const auto& v : se.vals) - val_state.move_val(state, v); - ) - ) - // Mark destination as valid - val_state.mark_validity( state, stmt.as_Assign().dst, true ); - break; - } - block_states[bb_idx].merge(val_state); - } - - // 3. During terminator, merge again - state.set_cur_stmt_term(bb_idx); - //DEBUG("- " << bb.terminator); - TU_MATCH(::MIR::Terminator, (bb.terminator), (e), - (Incomplete, - // Should be impossible here. - ), - (Return, - block_states[bb_idx].merge(val_state); - ), - (Diverge, - ), - (Goto, - block_states[bb_idx].merge(val_state); - // Push block with the new state - add_to_visit( e, mv$(val_state) ); - ), - (Panic, - // What should be done here? - ), - (If, - // Push blocks - block_states[bb_idx].merge(val_state); - add_to_visit( e.bb0, val_state ); - add_to_visit( e.bb1, mv$(val_state) ); - ), - (Switch, - block_states[bb_idx].merge(val_state); - for(const auto& tgt : e.targets) - { - add_to_visit( tgt, val_state ); - } - ), - (Call, - for(const auto& arg : e.args) - val_state.move_val( state, arg ); - block_states[bb_idx].merge(val_state); - // Push blocks (with return valid only in one) - add_to_visit(e.panic_block, val_state); - - // TODO: If the function returns !, don't follow the ret_block - val_state.mark_validity( state, e.ret_val, true ); - add_to_visit(e.ret_block, mv$(val_state)); - ) - ) - } - - // Convert block states into temp states - for(unsigned int bb_idx = 0; bb_idx < block_states.size(); bb_idx ++) - { - for(unsigned int tmp_idx = 0; tmp_idx < block_states[bb_idx].tmps.size(); tmp_idx ++) - { - tmp_lifetimes[tmp_idx].blocks[bb_idx] = block_states[bb_idx].tmps[tmp_idx]; - } - } - } + auto lifetimes = MIR_Helper_GetLifetimes(state, fcn, /*dump_debug=*/true); + //::std::vector<::MIR::ValueLifetime> var_lifetimes = mv$(lifetimes.m_variables); + ::std::vector<::MIR::ValueLifetime> tmp_lifetimes = mv$(lifetimes.m_temporaries); // 2. Unify variables of the same type with distinct non-overlapping lifetimes ::std::map<unsigned int, unsigned int> replacements; @@ -1282,7 +1100,7 @@ bool MIR_Optimise_UnifyTemporaries(::MIR::TypeResolve& state, ::MIR::Function& f { if( ! replacable[tmpidx] ) continue ; if( visited[tmpidx] ) continue ; - if( ! tmp_lifetimes[tmpidx].is_valid() ) continue ; + if( ! tmp_lifetimes[tmpidx].is_used() ) continue ; visited[tmpidx] = true; for(unsigned int i = tmpidx+1; i < fcn.temporaries.size(); i ++) @@ -1291,11 +1109,12 @@ bool MIR_Optimise_UnifyTemporaries(::MIR::TypeResolve& state, ::MIR::Function& f continue ; if( fcn.temporaries[i] != fcn.temporaries[tmpidx] ) continue ; - if( ! tmp_lifetimes[i].is_valid() ) continue ; + if( ! tmp_lifetimes[i].is_used() ) + continue ; // Variables are of the same type, check if they overlap if( tmp_lifetimes[tmpidx].overlaps( tmp_lifetimes[i] ) ) continue ; - // They overlap, unify + // They don't overlap, unify tmp_lifetimes[tmpidx].unify( tmp_lifetimes[i] ); replacements[i] = tmpidx; replacement_needed = true; @@ -1318,6 +1137,8 @@ bool MIR_Optimise_UnifyTemporaries(::MIR::TypeResolve& state, ::MIR::Function& f } return false; }); + + // TODO: Replace in ScopeEnd too? } return replacement_needed; @@ -1370,6 +1191,12 @@ bool MIR_Optimise_UnifyBlocks(::MIR::TypeResolve& state, ::MIR::Function& fcn) return false; if( ae.slot != be.slot ) return false; + ), + (ScopeEnd, + if( ae.vars != be.vars ) + return false; + if( ae.tmps == be.tmps ) + return false; ) ) } @@ -1462,7 +1289,7 @@ bool MIR_Optimise_UnifyBlocks(::MIR::TypeResolve& state, ::MIR::Function& fcn) if( ! replacements.empty() ) { //MIR_TODO(state, "Unify blocks - " << replacements); - DEBUG("Unify blocks - " << replacements); + DEBUG("Unify blocks (old: new) - " << replacements); auto patch_tgt = [&replacements](::MIR::BasicBlockId& tgt) { auto it = replacements.find(tgt); if( it != replacements.end() ) @@ -1524,6 +1351,9 @@ bool MIR_Optimise_UnifyBlocks(::MIR::TypeResolve& state, ::MIR::Function& fcn) // -------------------------------------------------------------------- bool MIR_Optimise_ConstPropagte(::MIR::TypeResolve& state, ::MIR::Function& fcn) { +#if DUMP_BEFORE_ALL || DUMP_BEFORE_CONSTPROPAGATE + if( debug_enabled() ) MIR_Dump_Fcn(::std::cout, fcn); +#endif bool changed = false; TRACE_FUNCTION_FR("", changed); @@ -1538,26 +1368,28 @@ bool MIR_Optimise_ConstPropagte(::MIR::TypeResolve& state, ::MIR::Function& fcn) const auto& tef = te.fcn.as_Intrinsic(); if( tef.name == "size_of" ) { - //size_t size_val = 0; - //if( Target_GetSizeOf(tef.params.m_types.at(0), size_val) ) - //{ - // bb.statements.push_back(::MIR::Statement::make_Assign({ mv$(te.ret_val), ::MIR::Constant::make_Uint(size_val) })); - // bb.terminator = ::MIR::Terminator::make_Goto(te.ret_block); - // changed = true; - //} + size_t size_val = 0; + if( Target_GetSizeOf(state.sp, tef.params.m_types.at(0), size_val) ) + { + auto val = ::MIR::Constant::make_Uint({ size_val, ::HIR::CoreType::Usize }); + bb.statements.push_back(::MIR::Statement::make_Assign({ mv$(te.ret_val), mv$(val) })); + bb.terminator = ::MIR::Terminator::make_Goto(te.ret_block); + changed = true; + } } else if( tef.name == "align_of" ) { - //size_t size_val = 0; - //if( Target_GetAlignOf(tef.params.m_types.at(0), size_val) ) - //{ - // bb.statements.push_back(::MIR::Statement::make_Assign({ mv$(te.ret_val), ::MIR::Constant::make_Uint(size_val) })); - // bb.terminator = ::MIR::Terminator::make_Goto(te.ret_block); - // changed = true; - //} + size_t align_val = 0; + if( Target_GetAlignOf(state.sp, tef.params.m_types.at(0), align_val) ) + { + auto val = ::MIR::Constant::make_Uint({ align_val, ::HIR::CoreType::Usize }); + bb.statements.push_back(::MIR::Statement::make_Assign({ mv$(te.ret_val), mv$(val) })); + bb.terminator = ::MIR::Terminator::make_Goto(te.ret_block); + changed = true; + } } - // NOTE: Quick special-case for bswap<u8> (a no-op) - else if( tef.name == "bswap" && tef.params.m_types.at(0) == ::HIR::CoreType::U8 ) + // NOTE: Quick special-case for bswap<u8/i8> (a no-op) + else if( tef.name == "bswap" && (tef.params.m_types.at(0) == ::HIR::CoreType::U8 || tef.params.m_types.at(0) == ::HIR::CoreType::I8) ) { DEBUG("bswap<u8> is a no-op"); if( auto* e = te.args.at(0).opt_LValue() ) @@ -1567,6 +1399,14 @@ bool MIR_Optimise_ConstPropagte(::MIR::TypeResolve& state, ::MIR::Function& fcn) bb.terminator = ::MIR::Terminator::make_Goto(te.ret_block); changed = true; } + else if( tef.name == "mrustc_slice_len" ) + { + MIR_ASSERT(state, te.args.at(0).is_LValue(), "Argument to `get_dst_meta` must be a lvalue"); + auto& e = te.args.at(0).as_LValue(); + bb.statements.push_back(::MIR::Statement::make_Assign({ mv$(te.ret_val), ::MIR::RValue::make_DstMeta({ mv$(e) }) })); + bb.terminator = ::MIR::Terminator::make_Goto(te.ret_block); + changed = true; + } else { // Ignore any other intrinsics @@ -1579,8 +1419,267 @@ bool MIR_Optimise_ConstPropagte(::MIR::TypeResolve& state, ::MIR::Function& fcn) // > NOTE: No need to locally stitch blocks, next pass will do that // TODO: Use ValState to do full constant propagation across blocks - // 1. Remove based on known booleans within a single block - // - Eliminates `if false`/`if true` branches + // Remove redundant temporaries and evaluate known binops + for(auto& bb : fcn.blocks) + { + auto bbidx = &bb - &fcn.blocks.front(); + + ::std::map< ::MIR::LValue, ::MIR::Constant > known_values; + ::std::map< unsigned, bool > known_drop_flags; + + auto check_param = [&](::MIR::Param& p) { + if(const auto* pe = p.opt_LValue()) { + auto it = known_values.find(*pe); + if( it != known_values.end() ) + { + DEBUG(state << "Value " << *pe << " known to be " << it->second); + p = it->second.clone(); + } + } + }; + + for(auto& stmt : bb.statements) + { + auto stmtidx = &stmt - &bb.statements.front(); + state.set_cur_stmt(bbidx, stmtidx); + // Scan statements forwards: + // - If a known temporary is used as Param::LValue, replace LValue with the value + // - If a UniOp has its input known, evaluate + // - If a BinOp has both values known, evaluate + if( auto* e = stmt.opt_Assign() ) + { + TU_MATCHA( (e->src), (se), + (Use, + auto it = known_values.find(se); + if( it != known_values.end() ) + { + DEBUG(state << "Value " << se << " known to be" << it->second); + e->src = it->second.clone(); + } + ), + (Constant, + // Ignore (knowledge done below) + ), + (SizedArray, + check_param(se.val); + ), + (Borrow, + ), + (Cast, + ), + (BinOp, + check_param(se.val_l); + check_param(se.val_r); + + if( se.val_l.is_Constant() && se.val_r.is_Constant() ) + { + const auto& val_l = se.val_l.as_Constant(); + const auto& val_r = se.val_r.as_Constant(); + + ::MIR::Constant new_value; + bool replace = false; + switch(se.op) + { + case ::MIR::eBinOp::EQ: + if( val_l.is_Const() ) + ; + else + { + replace = true; + new_value = ::MIR::Constant::make_Bool({val_l == val_r}); + } + break; + case ::MIR::eBinOp::NE: + if( val_l.is_Const() ) + ; + else + { + replace = true; + new_value = ::MIR::Constant::make_Bool({val_l != val_r}); + } + break; + // TODO: Other binary operations + default: + break; + } + + if( replace ) + { + DEBUG(state << " " << e->src << " = " << new_value); + e->src = mv$(new_value); + } + } + ), + (UniOp, + auto it = known_values.find(se.val); + if( it != known_values.end() ) + { + const auto& val = it->second; + ::MIR::Constant new_value; + bool replace = false; + // TODO: Evaluate UniOp + switch( se.op ) + { + case ::MIR::eUniOp::INV: + TU_MATCHA( (val), (ve), + (Uint, + auto val = ve.v; + switch(ve.t) + { + case ::HIR::CoreType::U8: val = (~val) & 0xFF; break; + case ::HIR::CoreType::U16: val = (~val) & 0xFFFF; break; + case ::HIR::CoreType::U32: val = (~val) & 0xFFFFFFFF; break; + case ::HIR::CoreType::Usize: + case ::HIR::CoreType::U64: + val = ~val; + break; + case ::HIR::CoreType::U128: + val = ~val; + break; + case ::HIR::CoreType::Char: + MIR_BUG(state, "Invalid use of ! on char"); + break; + default: + // Invalid type for Uint literal + break; + } + new_value = ::MIR::Constant::make_Uint({ val, ve.t }); + replace = true; + ), + (Int, + // Is ! valid on Int? + ), + (Float, + // Not valid? + ), + (Bool, + new_value = ::MIR::Constant::make_Bool({ !ve.v }); + replace = true; + ), + (Bytes, ), + (StaticString, ), + (Const, + // TODO: + ), + (ItemAddr, + ) + ) + break; + case ::MIR::eUniOp::NEG: + TU_MATCHA( (val), (ve), + (Uint, + // Not valid? + ), + (Int, + new_value = ::MIR::Constant::make_Int({ -ve.v, ve.t }); + replace = true; + ), + (Float, + new_value = ::MIR::Constant::make_Float({ -ve.v, ve.t }); + replace = true; + ), + (Bool, + // Not valid? + ), + (Bytes, ), + (StaticString, ), + (Const, + // TODO: + ), + (ItemAddr, + ) + ) + break; + } + if( replace ) + { + DEBUG(state << " " << e->src << " = " << new_value); + e->src = mv$(new_value); + } + } + ), + (DstMeta, + ), + (DstPtr, + ), + (MakeDst, + check_param(se.ptr_val); + check_param(se.meta_val); + ), + (Tuple, + for(auto& p : se.vals) + check_param(p); + ), + (Array, + for(auto& p : se.vals) + check_param(p); + ), + (Variant, + check_param(se.val); + ), + (Struct, + for(auto& p : se.vals) + check_param(p); + ) + ) + } + else if( const auto* se = stmt.opt_SetDropFlag() ) + { + if( se->other == ~0u ) + { + known_drop_flags.insert(::std::make_pair( se->idx, se->new_val )); + } + else + { + auto it = known_drop_flags.find(se->other); + if( it != known_drop_flags.end() ) + { + known_drop_flags.insert(::std::make_pair( se->idx, se->new_val ^ it->second )); + } + } + } + else if( auto* se = stmt.opt_Drop() ) + { + if( se->flag_idx != ~0u ) + { + auto it = known_drop_flags.find(se->flag_idx); + if( it != known_drop_flags.end() ) + { + if( it->second ) { + se->flag_idx = ~0u; + } + else { + // TODO: Delete drop + stmt = ::MIR::Statement::make_ScopeEnd({ }); + } + } + } + } + // - If a known temporary is borrowed mutably or mutated somehow, clear its knowledge + visit_mir_lvalues(stmt, [&known_values](const ::MIR::LValue& lv, ValUsage vu)->bool { + if( vu == ValUsage::Write ) { + known_values.erase(lv); + } + return false; + }); + // - Locate `temp = SOME_CONST` and record value + if( const auto* e = stmt.opt_Assign() ) + { + if( e->dst.is_Temporary() || e->dst.is_Variable() ) + { + if( const auto* ce = e->src.opt_Constant() ) + { + known_values.insert(::std::make_pair( e->dst.clone(), ce->clone() )); + } + } + } + } + + state.set_cur_stmt_term(bbidx); + } + + // - Remove based on known booleans within a single block + // > Eliminates `if false`/`if true` branches for(auto& bb : fcn.blocks) { auto bbidx = &bb - &fcn.blocks.front(); @@ -1612,12 +1711,15 @@ bool MIR_Optimise_ConstPropagte(::MIR::TypeResolve& state, ::MIR::Function& fcn) if( se.dst != te.cond ) continue; - if( !se.src.is_Constant() ) - continue; - if( !se.src.as_Constant().is_Bool() ) - continue; - val_known = true; - known_val = se.src.as_Constant().as_Bool().v; + if( se.src.is_Constant() && se.src.as_Constant().is_Bool() ) + { + val_known = true; + known_val = se.src.as_Constant().as_Bool().v; + } + else + { + val_known = false; + } break; } else @@ -1759,6 +1861,12 @@ bool MIR_Optimise_PropagateSingleAssignments(::MIR::TypeResolve& state, ::MIR::F srcp = &*srcp->as_Field().val; if( !( srcp->is_Temporary() || srcp->is_Variable() || srcp->is_Argument() ) ) continue ; + + if( replacements.find(*srcp) != replacements.end() ) + { + DEBUG("> Can't replace, source has pending replacement"); + continue; + } } // TODO: Allow any rvalue, but that currently breaks due to chaining //else if( e.src.is_Borrow() ) @@ -1785,6 +1893,7 @@ bool MIR_Optimise_PropagateSingleAssignments(::MIR::TypeResolve& state, ::MIR::F for(unsigned int si2 = stmt_idx+1; si2 < block.statements.size(); si2 ++) { const auto& stmt2 = block.statements[si2]; + DEBUG("[find usage] " << stmt2); // Usage found. if( visit_mir_lvalues(stmt2, is_lvalue_usage) ) @@ -1808,7 +1917,7 @@ bool MIR_Optimise_PropagateSingleAssignments(::MIR::TypeResolve& state, ::MIR::F // Determine if source is mutated. // > Assume that any mutating access of the root value counts (over-cautious) - if( visit_mir_lvalues(block.statements[si2], [&](const auto& lv, auto vu){ return /*vu == ValUsage::Write &&*/ is_lvalue_in_val(lv); }) ) + if( visit_mir_lvalues(stmt2, [&](const auto& lv, auto vu){ return /*vu == ValUsage::Write &&*/ is_lvalue_in_val(lv); }) ) { stop = true; break; @@ -1816,7 +1925,7 @@ bool MIR_Optimise_PropagateSingleAssignments(::MIR::TypeResolve& state, ::MIR::F } if( !stop ) { - DEBUG(block.terminator); + DEBUG("[find usage] " << block.terminator); TU_MATCHA( (block.terminator), (e), (Incomplete, ), @@ -2212,11 +2321,11 @@ bool MIR_Optimise_GarbageCollect(::MIR::TypeResolve& state, ::MIR::Function& fcn TU_IFLET( ::MIR::Statement, stmt, Assign, e, assigned_lval(e.dst); ) - else if( const auto* e = stmt.opt_Drop() ) - { - if( e->flag_idx != ~0u ) - used_dfs.at(e->flag_idx) = true; - } + //else if( const auto* e = stmt.opt_Drop() ) + //{ + // //if( e->flag_idx != ~0u ) + // // used_dfs.at(e->flag_idx) = true; + //} else if( const auto* e = stmt.opt_Asm() ) { for(const auto& val : e->outputs) @@ -2226,6 +2335,7 @@ bool MIR_Optimise_GarbageCollect(::MIR::TypeResolve& state, ::MIR::Function& fcn { if( e->other != ~0u ) used_dfs.at(e->other) = true; + used_dfs.at(e->idx) = true; } } @@ -2275,11 +2385,22 @@ bool MIR_Optimise_GarbageCollect(::MIR::TypeResolve& state, ::MIR::Function& fcn { if( !used_temps[i] ) { - DEBUG("GC Temporary(" << i << ")"); fcn.temporaries.erase(fcn.temporaries.begin() + j); } + else { + DEBUG("tmp$" << i << " => tmp$" << j); + } temp_rewrite_table.push_back( used_temps[i] ? j ++ : ~0u ); } + DEBUG("Deleted Temporaries:" << FMT_CB(ss, + for(auto run : runs(used_temps)) + if( !used_temps[run.first] ) + { + ss << " " << run.first; + if(run.second != run.first) + ss << "-" << run.second; + } + )); ::std::vector<unsigned int> var_rewrite_table; unsigned int n_var = fcn.named_variables.size(); for(unsigned int i = 0, j = 0; i < n_var; i ++) @@ -2289,6 +2410,9 @@ bool MIR_Optimise_GarbageCollect(::MIR::TypeResolve& state, ::MIR::Function& fcn DEBUG("GC Variable(" << i << ")"); fcn.named_variables.erase(fcn.named_variables.begin() + j); } + else { + DEBUG("var$" << i << " => var$" << j); + } var_rewrite_table.push_back( used_vars[i] ? j ++ : ~0u ); } ::std::vector<unsigned int> df_rewrite_table; @@ -2298,7 +2422,7 @@ bool MIR_Optimise_GarbageCollect(::MIR::TypeResolve& state, ::MIR::Function& fcn if( !used_dfs[i] ) { DEBUG("GC df" << i); - fcn.drop_flags.erase(fcn.drop_flags.begin() + j); + // NOTE: Not erased until after rewriting } df_rewrite_table.push_back( used_dfs[i] ? j ++ : ~0u ); } @@ -2329,27 +2453,75 @@ bool MIR_Optimise_GarbageCollect(::MIR::TypeResolve& state, ::MIR::Function& fcn } return false; }; - for(auto stmt_it = it->statements.begin(); stmt_it != it->statements.end(); ++ stmt_it) + ::std::vector<bool> to_remove_statements(it->statements.size()); + for(auto& stmt : it->statements) { - state.set_cur_stmt(i, stmt_it - it->statements.begin()); - visit_mir_lvalues_mut(*stmt_it, lvalue_cb); - if( auto* se = stmt_it->opt_Drop() ) + auto stmt_idx = &stmt - &it->statements.front(); + state.set_cur_stmt(i, stmt_idx); + if( auto* se = stmt.opt_Drop() ) + { + // If the drop flag was unset, either remove the drop or remove the drop flag reference + if( se->flag_idx != ~0u && df_rewrite_table[se->flag_idx] == ~0u) + { + if( fcn.drop_flags.at(se->flag_idx) ) { + DEBUG(state << "Remove flag from " << stmt); + se->flag_idx = ~0u; + } + else { + DEBUG(state << "Remove " << stmt); + to_remove_statements[stmt_idx] = true; + continue ; + } + } + } + + visit_mir_lvalues_mut(stmt, lvalue_cb); + if( auto* se = stmt.opt_Drop() ) { // Rewrite drop flag indexes if( se->flag_idx != ~0u ) se->flag_idx = df_rewrite_table[se->flag_idx]; } - else if( auto* se = stmt_it->opt_SetDropFlag() ) + else if( auto* se = stmt.opt_SetDropFlag() ) { // Rewrite drop flag indexes OR delete if( df_rewrite_table[se->idx] == ~0u ) { - stmt_it = it->statements.erase(stmt_it)-1; + to_remove_statements[stmt_idx] = true; continue ; } se->idx = df_rewrite_table[se->idx]; if( se->other != ~0u ) se->other = df_rewrite_table[se->other]; } + else if( auto* se = stmt.opt_ScopeEnd() ) + { + for(auto it = se->vars.begin(); it != se->vars.end(); ) + { + if( var_rewrite_table.at(*it) == ~0u ) { + it = se->vars.erase(it); + } + else { + *it = var_rewrite_table.at(*it); + ++ it; + } + } + for(auto it = se->tmps.begin(); it != se->tmps.end(); ) + { + if( temp_rewrite_table.at(*it) == ~0u ) { + it = se->tmps.erase(it); + } + else { + *it = temp_rewrite_table.at(*it); + ++ it; + } + } + + if( se->vars.empty() && se->tmps.empty() ) { + DEBUG(state << "Delete ScopeEnd (now empty)"); + to_remove_statements[stmt_idx] = true; + continue ; + } + } } state.set_cur_stmt_term(i); // Rewrite and advance @@ -2387,10 +2559,34 @@ bool MIR_Optimise_GarbageCollect(::MIR::TypeResolve& state, ::MIR::Function& fcn ) ) + // Delete all statements flagged in a bitmap for deletion + auto stmt_it = it->statements.begin(); + for(auto flag : to_remove_statements) + { + if(flag) { + stmt_it = it->statements.erase(stmt_it); + } + else { + ++ stmt_it; + } + } + ++it; } } + for(unsigned int i = 0, j = 0; i < n_df; i ++) + { + if( !used_dfs[i] ) + { + fcn.drop_flags.erase(fcn.drop_flags.begin() + j); + } + else + { + j ++; + } + } + // TODO: Detect if any optimisations happened, and return true in that case return false; } diff --git a/src/parse/expr.cpp b/src/parse/expr.cpp index 6227dca4..6dd9bd75 100644 --- a/src/parse/expr.cpp +++ b/src/parse/expr.cpp @@ -119,6 +119,16 @@ ExprNodeP Parse_ExprBlockLine_WithItems(TokenStream& lex, ::std::shared_ptr<AST: Parse_Mod_Item(lex, *local_mod, mv$(item_attrs)); return ExprNodeP(); } + + if( tok.type() == TOK_MACRO && tok.str() == "macro_rules" ) + { + // Special case - create a local module if macro_rules! is seen + // - Allows correct scoping of defined macros + if( !local_mod ) { + local_mod = lex.parse_state().get_current_mod().add_anon(); + } + } + switch(tok.type()) { // Items: @@ -669,6 +679,9 @@ bool Parse_IsTokValue(eTokenType tok_type) case TOK_PAREN_OPEN: case TOK_SQUARE_OPEN: + case TOK_INTERPOLATED_PATH: + case TOK_INTERPOLATED_EXPR: + case TOK_MACRO: case TOK_PIPE: diff --git a/src/parse/lex.cpp b/src/parse/lex.cpp index caa640f9..3c4bd795 100644 --- a/src/parse/lex.cpp +++ b/src/parse/lex.cpp @@ -860,9 +860,13 @@ uint32_t Lexer::parseEscape(char enclosing) case '\n': while( ch.isspace() ) ch = this->getc(); - this->ungetc(); - if( ch == enclosing ) + if(ch == '\\' ) + return parseEscape(enclosing); + else if( ch == enclosing ) + { + this->ungetc(); return ~0; + } else return ch.v; default: @@ -1009,12 +1013,12 @@ bool Codepoint::isxdigit() const { s += (char)(0xC0 | ((cp.v >> 6) & 0x1F)); s += (char)(0x80 | ((cp.v >> 0) & 0x3F)); } - else if( cp.v <= (0x0F+1)<<(2*6) ) { + else if( cp.v < (0x0F+1)<<(2*6) ) { s += (char)(0xE0 | ((cp.v >> 12) & 0x0F)); s += (char)(0x80 | ((cp.v >> 6) & 0x3F)); s += (char)(0x80 | ((cp.v >> 0) & 0x3F)); } - else if( cp.v <= (0x07+1)<<(3*6) ) { + else if( cp.v < (0x07+1)<<(3*6) ) { s += (char)(0xF0 | ((cp.v >> 18) & 0x07)); s += (char)(0x80 | ((cp.v >> 12) & 0x3F)); s += (char)(0x80 | ((cp.v >> 6) & 0x3F)); @@ -1034,12 +1038,12 @@ bool Codepoint::isxdigit() const { os << (char)(0xC0 | ((cp.v >> 6) & 0x1F)); os << (char)(0x80 | ((cp.v >> 0) & 0x3F)); } - else if( cp.v <= (0x0F+1)<<(2*6) ) { + else if( cp.v < (0x0F+1)<<(2*6) ) { os << (char)(0xE0 | ((cp.v >> 12) & 0x0F)); os << (char)(0x80 | ((cp.v >> 6) & 0x3F)); os << (char)(0x80 | ((cp.v >> 0) & 0x3F)); } - else if( cp.v <= (0x07+1)<<(2*6) ) { + else if( cp.v < (0x07+1)<<(2*6) ) { os << (char)(0xF0 | ((cp.v >> 18) & 0x07)); os << (char)(0x80 | ((cp.v >> 12) & 0x3F)); os << (char)(0x80 | ((cp.v >> 6) & 0x3F)); diff --git a/src/parse/root.cpp b/src/parse/root.cpp index aefd5a13..d40e1f95 100644 --- a/src/parse/root.cpp +++ b/src/parse/root.cpp @@ -44,13 +44,13 @@ AST::MetaItem Parse_MetaItem(TokenStream& lex); void Parse_ModRoot(TokenStream& lex, AST::Module& mod, AST::MetaItems& mod_attrs); //::AST::Path Parse_Publicity(TokenStream& lex) -bool Parse_Publicity(TokenStream& lex) +bool Parse_Publicity(TokenStream& lex, bool allow_restricted=true) { Token tok; if( LOOK_AHEAD(lex) == TOK_RWORD_PUB ) { GET_TOK(tok, lex); - if( LOOK_AHEAD(lex) == TOK_PAREN_OPEN ) + if( allow_restricted && LOOK_AHEAD(lex) == TOK_PAREN_OPEN ) { auto path = AST::Path("", {}); // Restricted publicity. @@ -541,7 +541,7 @@ AST::Struct Parse_Struct(TokenStream& lex, const AST::MetaItems& meta_items) SET_ATTRS(lex, item_attrs); PUTBACK(tok, lex); - bool is_pub = Parse_Publicity(lex); + bool is_pub = Parse_Publicity(lex, /*allow_restricted=*/false); // HACK: Disable `pub(restricted)` syntax in tuple structs, due to ambiguity refs.push_back( AST::TupleItem( mv$(item_attrs), is_pub, Parse_Type(lex) ) ); if( GET_TOK(tok, lex) != TOK_COMMA ) @@ -1336,7 +1336,6 @@ void Parse_Use(TokenStream& lex, ::std::function<void(AST::UseStmt, ::std::strin { GET_CHECK_TOK(tok, lex, TOK_STRING); path = ::AST::Path(tok.str(), {}); - GET_CHECK_TOK(tok, lex, TOK_DOUBLE_COLON); } else { PUTBACK(tok, lex); @@ -1387,7 +1386,7 @@ void Parse_Use(TokenStream& lex, ::std::function<void(AST::UseStmt, ::std::strin else { PUTBACK(tok, lex); - assert(path.nodes().size() > 0); + ASSERT_BUG(lex.getPosition(), path.nodes().size() > 0, "`use` with no path"); name = path.nodes().back().name(); } diff --git a/src/parse/token.cpp b/src/parse/token.cpp index c7d11d03..05c4dbe6 100644 --- a/src/parse/token.cpp +++ b/src/parse/token.cpp @@ -289,16 +289,16 @@ struct EscapedString { case TOK_NEWLINE: return "\n"; case TOK_WHITESPACE: return " "; case TOK_COMMENT: return "/*" + m_data.as_String() + "*/"; - case TOK_INTERPOLATED_TYPE: return "/*:ty*/"; - case TOK_INTERPOLATED_PATH: return "/*:path*/"; - case TOK_INTERPOLATED_PATTERN: return "/*:pat*/"; + case TOK_INTERPOLATED_TYPE: return FMT( *reinterpret_cast<const ::TypeRef*>(m_data.as_Fragment()) ); + case TOK_INTERPOLATED_PATH: return FMT( *reinterpret_cast<const ::AST::Path*>(m_data.as_Fragment()) ); + case TOK_INTERPOLATED_PATTERN: return FMT( *reinterpret_cast<const ::AST::Pattern*>(m_data.as_Fragment()) ); + case TOK_INTERPOLATED_STMT: + case TOK_INTERPOLATED_BLOCK: case TOK_INTERPOLATED_EXPR: { ::std::stringstream ss; reinterpret_cast<const ::AST::ExprNode*>(m_data.as_Fragment())->print(ss); return ss.str(); } - case TOK_INTERPOLATED_STMT: return "/*:stmt*/"; - case TOK_INTERPOLATED_BLOCK: return "/*:block*/"; case TOK_INTERPOLATED_META: return "/*:meta*/"; case TOK_INTERPOLATED_ITEM: return "/*:item*/"; case TOK_INTERPOLATED_IDENT: return "/*:ident*/"; diff --git a/src/trans/codegen.cpp b/src/trans/codegen.cpp index 6a581d93..9d536181 100644 --- a/src/trans/codegen.cpp +++ b/src/trans/codegen.cpp @@ -160,8 +160,7 @@ void Trans_Codegen(const ::std::string& outfile, const TransOptions& opt, const auto mir = Trans_Monomorphise(resolve, pp, fcn.m_code.m_mir); MIR_Validate(resolve, ::HIR::ItemPath(""), *mir, args, ret_type); MIR_Cleanup(resolve, ::HIR::ItemPath(""), *mir, args, ret_type); - // TODO: MIR Optimisation - //MIR_Optimise(resolve, ::HIR::ItemPath(""), *mir, args, ret_type); + MIR_Optimise(resolve, ::HIR::ItemPath(""), *mir, args, ret_type); MIR_Validate(resolve, ::HIR::ItemPath(""), *mir, args, ret_type); codegen->emit_function_code(path, fcn, ent.second->pp, mir); } diff --git a/src/trans/codegen_c.cpp b/src/trans/codegen_c.cpp index 5e365ceb..2d7a521d 100644 --- a/src/trans/codegen_c.cpp +++ b/src/trans/codegen_c.cpp @@ -65,7 +65,7 @@ namespace { << "typedef struct { } tTYPEID;\n" << "typedef struct { void* PTR; size_t META; } SLICE_PTR;\n" << "typedef struct { void* PTR; void* META; } TRAITOBJ_PTR;\n" - << "typedef struct { size_t size; size_t align; } VTABLE_HDR;\n" + << "typedef struct { size_t size; size_t align; void (*drop)(void*); } VTABLE_HDR;\n" << "\n" << "extern void _Unwind_Resume(void) __attribute__((noreturn));\n" << "\n" @@ -84,7 +84,7 @@ namespace { << "\treturn (v >> 32 != 0 ? __builtin_clz(v>>32) : 32 + __builtin_clz(v));\n" << "}\n" << "static inline uint64_t __builtin_ctz64(uint64_t v) {\n" - << "\treturn (v&0xFFFFFFFF == 0 ? __builtin_ctz(v>>32) + 32 : __builtin_ctz(v));\n" + << "\treturn ((v&0xFFFFFFFF) == 0 ? __builtin_ctz(v>>32) + 32 : __builtin_ctz(v));\n" << "}\n" << "static inline unsigned __int128 __builtin_bswap128(unsigned __int128 v) {\n" << "\tuint64_t lo = __builtin_bswap64((uint64_t)v);\n" @@ -95,9 +95,11 @@ namespace { << "\treturn (v >> 64 != 0 ? __builtin_clz64(v>>64) : 64 + __builtin_clz64(v));\n" << "}\n" << "static inline unsigned __int128 __builtin_ctz128(unsigned __int128 v) {\n" - << "\treturn (v&0xFFFFFFFFFFFFFFFF == 0 ? __builtin_ctz64(v>>64) + 64 : __builtin_ctz64(v));\n" + << "\treturn ((v&0xFFFFFFFFFFFFFFFF) == 0 ? __builtin_ctz64(v>>64) + 64 : __builtin_ctz64(v));\n" << "}\n" << "\n" + << "static inline void noop_drop(void *p) {}\n" + << "\n" ; } @@ -134,6 +136,7 @@ namespace { ::std::vector<::std::string> tmp; ::std::vector<const char*> args; args.push_back( getenv("CC") ? getenv("CC") : "gcc" ); + args.push_back("-ffunction-sections"); args.push_back("-pthread"); switch(opt.opt_level) { @@ -189,6 +192,7 @@ namespace { ::std::stringstream cmd_ss; for(const auto& arg : args) { + // TODO: use a formatter specific to shell escaping cmd_ss << "\"" << FmtEscaped(arg) << "\" "; } DEBUG("- " << cmd_ss.str()); @@ -310,6 +314,21 @@ namespace { } m_of << "} "; emit_ctype(ty); m_of << ";\n"; } + + auto drop_glue_path = ::HIR::Path(ty.clone(), "#drop_glue"); + auto args = ::std::vector< ::std::pair<::HIR::Pattern,::HIR::TypeRef> >(); + auto ty_ptr = ::HIR::TypeRef::new_pointer(::HIR::BorrowType::Owned, ty.clone()); + ::MIR::TypeResolve mir_res { sp, m_resolve, FMT_CB(ss, ss << drop_glue_path;), ty_ptr, args, *(::MIR::Function*)nullptr }; + m_mir_res = &mir_res; + m_of << "void " << Trans_Mangle(drop_glue_path) << "("; emit_ctype(ty); m_of << "* rv) {"; + auto self = ::MIR::LValue::make_Deref({ box$(::MIR::LValue::make_Return({})) }); + auto fld_lv = ::MIR::LValue::make_Field({ box$(self), 0 }); + for(const auto& ity : te) + { + emit_destructor_call(fld_lv, ity, /*unsized_valid=*/false); + fld_lv.as_Field().field_index ++; + } + m_of << "}\n"; ) else TU_IFLET( ::HIR::TypeRef::Data, ty.m_data, Function, te, emit_type_fn(ty); @@ -406,7 +425,7 @@ namespace { ::std::vector< ::std::pair<::HIR::Pattern,::HIR::TypeRef> > args; if( item.m_markings.has_drop_impl ) { - m_of << "tUNIT " << Trans_Mangle( ::HIR::Path(struct_ty.clone(), m_resolve.m_lang_Drop, "drop") ) << "(struct s_" << Trans_Mangle(p) << "*rv);\n"; + m_of << "tUNIT " << Trans_Mangle( ::HIR::Path(struct_ty.clone(), m_resolve.m_lang_Drop, "drop") ) << "("; emit_ctype(struct_ty_ptr, FMT_CB(ss, ss << "rv";)); m_of << ");\n"; } else if( m_resolve.is_type_owned_box(struct_ty) ) { @@ -968,10 +987,10 @@ namespace { m_of << "}"; } } - else if( enm.m_repr != ::HIR::Enum::Repr::Rust || ::std::all_of(enm.m_variants.begin(), enm.m_variants.end(), [](const auto& x){return x.second.is_Unit() || x.second.is_Value();}) ) + else if( enm.is_value() ) { MIR_ASSERT(*m_mir_res, e.vals.empty(), "Value-only enum with fields"); - m_of << "{" << e.idx << "}"; + m_of << "{" << enm.get_value(e.idx) << "}"; } else { @@ -1178,7 +1197,14 @@ namespace { m_of << "\t{ "; m_of << "sizeof("; emit_ctype(type); m_of << "),"; m_of << "__alignof__("; emit_ctype(type); m_of << "),"; - // TODO: Drop glue pointer + if( type.m_data.is_Borrow() || m_resolve.type_is_copy(sp, type) ) + { + m_of << "noop_drop,"; + } + else + { + m_of << "(void*)" << Trans_Mangle(::HIR::Path(type.clone(), "#drop_glue")) << ","; + } m_of << "}"; // No newline, added below for(unsigned int i = 0; i < trait.m_value_indexes.size(); i ++ ) @@ -1288,6 +1314,9 @@ namespace { switch( stmt.tag() ) { case ::MIR::Statement::TAGDEAD: throw ""; + case ::MIR::Statement::TAG_ScopeEnd: + m_of << "// " << stmt << "\n"; + break; case ::MIR::Statement::TAG_SetDropFlag: { const auto& e = stmt.as_SetDropFlag(); m_of << "\tdf" << e.idx << " = "; @@ -1303,7 +1332,7 @@ namespace { const auto& ty = mir_res.get_lvalue_type(tmp, e.slot); if( e.flag_idx != ~0u ) - m_of << "if( df" << e.flag_idx << " ) {\n"; + m_of << "\tif( df" << e.flag_idx << " ) {\n"; switch( e.kind ) { @@ -1326,7 +1355,7 @@ namespace { break; } if( e.flag_idx != ~0u ) - m_of << "}\n"; + m_of << "\t}\n"; break; } case ::MIR::Statement::TAG_Asm: { const auto& e = stmt.as_Asm(); @@ -1350,6 +1379,7 @@ namespace { m_of << "\t__asm__ "; if(is_volatile) m_of << "__volatile__"; // TODO: Convert format string? + // TODO: Use a C-specific escaper here. m_of << "(\"" << (is_intel ? ".syntax intel; " : "") << FmtEscaped(e.tpl) << (is_intel ? ".syntax att; " : "") << "\""; m_of << ": "; for(unsigned int i = 0; i < e.outputs.size(); i ++ ) @@ -1498,9 +1528,8 @@ namespace { (BinOp, emit_lvalue(e.dst); m_of << " = "; - MIR_ASSERT(mir_res, ve.val_l.is_LValue() || ve.val_r.is_LValue(), ""); ::HIR::TypeRef tmp; - const auto& ty = mir_res.get_lvalue_type(tmp, ve.val_l.is_LValue() ? ve.val_l.as_LValue() : ve.val_r.as_LValue()); + const auto& ty = ve.val_l.is_LValue() ? mir_res.get_lvalue_type(tmp, ve.val_l.as_LValue()) : tmp = mir_res.get_const_type(ve.val_l.as_Constant()); if( ty.m_data.is_Borrow() ) { m_of << "(slice_cmp("; emit_param(ve.val_l); m_of << ", "; emit_param(ve.val_r); m_of << ")"; switch(ve.op) @@ -1646,43 +1675,65 @@ namespace { } ), (Variant, - if( m_crate.get_typeitem_by_path(sp, ve.path.m_path).is_Union() ) + const auto& tyi = m_crate.get_typeitem_by_path(sp, ve.path.m_path); + if( tyi.is_Union() ) { emit_lvalue(e.dst); + m_of << ".var_" << ve.index << " = "; emit_param(ve.val); } - else + else if( const auto* enm_p = tyi.opt_Enum() ) { MIR_TODO(mir_res, "Construct enum with RValue::Variant"); - emit_lvalue(e.dst); m_of << ".TAG = " << ve.index << ";\n\t"; - emit_lvalue(e.dst); m_of << ".DATA"; + if( enm_p->is_value() ) + { + emit_lvalue(e.dst); m_of << ".TAG = " << enm_p->get_value(ve.index) << ""; + } + else + { + emit_lvalue(e.dst); m_of << ".TAG = " << ve.index << ";\n\t"; + emit_lvalue(e.dst); m_of << ".DATA"; + m_of << ".var_" << ve.index << " = "; emit_param(ve.val); + } + } + else + { + BUG(mir_res.sp, "Unexpected type in Variant"); } - m_of << ".var_" << ve.index << " = "; emit_param(ve.val); ), (Struct, - if(ve.variant_idx != ~0u) { + if(ve.variant_idx != ~0u) + { ::HIR::TypeRef tmp; const auto& ty = mir_res.get_lvalue_type(tmp, e.dst); - if( ty.m_data.as_Path().binding.is_Enum() ) { - auto it = m_enum_repr_cache.find(ty.m_data.as_Path().path.m_data.as_Generic()); - if( it != m_enum_repr_cache.end() ) - { - if( ve.variant_idx == 0 ) { - // TODO: Use nonzero_path - m_of << "memset(&"; emit_lvalue(e.dst); m_of << ", 0, sizeof("; emit_ctype(ty); m_of << "))"; - } - else if( ve.variant_idx == 1 ) { - emit_lvalue(e.dst); - m_of << "._0 = "; - emit_param(ve.vals[0]); - } - else { - } - break; + const auto* enm_p = ty.m_data.as_Path().binding.as_Enum(); + + auto it = m_enum_repr_cache.find(ty.m_data.as_Path().path.m_data.as_Generic()); + if( it != m_enum_repr_cache.end() ) + { + if( ve.variant_idx == 0 ) { + // TODO: Use nonzero_path + m_of << "memset(&"; emit_lvalue(e.dst); m_of << ", 0, sizeof("; emit_ctype(ty); m_of << "))"; + } + else if( ve.variant_idx == 1 ) { + emit_lvalue(e.dst); + m_of << "._0 = "; + emit_param(ve.vals[0]); + } + else { } + break; + } + else if( enm_p->is_value() ) + { + emit_lvalue(e.dst); + m_of << ".TAG = " << enm_p->get_value(ve.variant_idx); + assert(ve.vals.size() == 0); + } + else + { + emit_lvalue(e.dst); + m_of << ".TAG = " << ve.variant_idx; } - - emit_lvalue(e.dst); - m_of << ".TAG = " << ve.variant_idx; if(ve.vals.size() > 0) m_of << ";\n\t"; } @@ -1736,6 +1787,7 @@ namespace { const auto& ty = mir_res.get_lvalue_type(tmp, e.val); MIR_ASSERT(mir_res, ty.m_data.is_Path(), ""); MIR_ASSERT(mir_res, ty.m_data.as_Path().binding.is_Enum(), ""); + const auto* enm = ty.m_data.as_Path().binding.as_Enum(); auto it = m_enum_repr_cache.find( ty.m_data.as_Path().path.m_data.as_Generic() ); if( it != m_enum_repr_cache.end() ) { @@ -1745,6 +1797,16 @@ namespace { m_of << "\telse\n"; m_of << "\t\tgoto bb" << e.targets[0] << ";\n"; } + else if( enm->is_value() ) + { + m_of << "\tswitch("; emit_lvalue(e.val); m_of << ".TAG) {\n"; + for(unsigned int j = 0; j < e.targets.size(); j ++) + { + m_of << "\t\tcase " << enm->get_value(j) << ": goto bb" << e.targets[j] << ";\n"; + } + m_of << "\t\tdefault: abort();\n"; + m_of << "\t}\n"; + } else { m_of << "\tswitch("; emit_lvalue(e.val); m_of << ".TAG) {\n"; @@ -2046,7 +2108,16 @@ namespace { emit_lvalue(e.ret_val); m_of << ".META = " << s.size() << ""; } else if( name == "transmute" ) { - m_of << "memcpy( &"; emit_lvalue(e.ret_val); m_of << ", &"; emit_param(e.args.at(0)); m_of << ", sizeof("; emit_ctype(params.m_types.at(0)); m_of << "))"; + if( e.args.at(0).is_Constant() ) + { + m_of << "{ "; emit_ctype(params.m_types.at(1), FMT_CB(s, s << "v";)); m_of << " = "; emit_param(e.args.at(0)); m_of << ";"; + m_of << "memcpy( &"; emit_lvalue(e.ret_val); m_of << ", &v, sizeof("; emit_ctype(params.m_types.at(0)); m_of << ")); "; + m_of << "}"; + } + else + { + m_of << "memcpy( &"; emit_lvalue(e.ret_val); m_of << ", &"; emit_param(e.args.at(0)); m_of << ", sizeof("; emit_ctype(params.m_types.at(0)); m_of << "))"; + } } else if( name == "copy_nonoverlapping" || name == "copy" ) { if( name == "copy" ) { @@ -2077,15 +2148,15 @@ namespace { else if( name == "needs_drop" ) { // Returns `true` if the actual type given as `T` requires drop glue; // returns `false` if the actual type provided for `T` implements `Copy`. (Either otherwise) + // NOTE: libarena assumes that this returns `true` iff T doesn't require drop glue. const auto& ty = params.m_types.at(0); emit_lvalue(e.ret_val); m_of << " = "; - if( m_resolve.type_is_copy(Span(), ty) ) { - m_of << "false"; + if( m_resolve.type_needs_drop_glue(mir_res.sp, ty) ) { + m_of << "true"; } - // If T: !Copy, return true else { - m_of << "true"; + m_of << "false"; } } else if( name == "uninit" ) { @@ -2299,6 +2370,13 @@ namespace { else if( name == "fmaf32" || name == "fmaf64" ) { emit_lvalue(e.ret_val); m_of << " = fma" << (name.back()=='2'?"f":"") << "("; emit_param(e.args.at(0)); m_of << ", "; emit_param(e.args.at(1)); m_of << ", "; emit_param(e.args.at(1)); m_of << ")"; } + // --- Volatile Load/Store + else if( name == "volatile_load" ) { + emit_lvalue(e.ret_val); m_of << " = *(volatile "; emit_ctype(params.m_types.at(0)); m_of << "*)"; emit_param(e.args.at(0)); + } + else if( name == "volatile_store" ) { + m_of << "*(volatile "; emit_ctype(params.m_types.at(0)); m_of << "*)"; emit_param(e.args.at(0)); m_of << " = "; emit_param(e.args.at(1)); + } // --- Atomics! // > Single-ordering atomics else if( name == "atomic_xadd" || name.compare(0, 7+4+1, "atomic_xadd_") == 0 ) { @@ -2362,9 +2440,20 @@ namespace { auto fail_ordering = H::get_atomic_ordering(mir_res, name, 7+10+4); emit_atomic_cxchg(e, "memory_order_seq_cst", fail_ordering, true); } - else if( name == "atomic_cxchgweak" || name.compare(0, 7+9+1, "atomic_cxchgweak_") == 0 ) { - auto ordering = H::get_atomic_ordering(mir_res, name, 7+9+1); - emit_atomic_cxchg(e, ordering, ordering, true); + else if( name == "atomic_cxchgweak" ) { + emit_atomic_cxchg(e, "memory_order_seq_cst", "memory_order_seq_cst", true); + } + else if( name == "atomic_cxchgweak_acq" ) { + emit_atomic_cxchg(e, "memory_order_acquire", "memory_order_acquire", true); + } + else if( name == "atomic_cxchgweak_rel" ) { + emit_atomic_cxchg(e, "memory_order_release", "memory_order_relaxed", true); + } + else if( name == "atomic_cxchgweak_acqrel" ) { + emit_atomic_cxchg(e, "memory_order_acq_rel", "memory_order_acquire", true); + } + else if( name == "atomic_cxchgweak_relaxed" ) { + emit_atomic_cxchg(e, "memory_order_relaxed", "memory_order_relaxed", true); } else if( name == "atomic_xchg" || name.compare(0, 7+5, "atomic_xchg_") == 0 ) { auto ordering = H::get_atomic_ordering(mir_res, name, 7+5); @@ -2419,7 +2508,17 @@ namespace { make_fcn = "make_sliceptr"; if(0) case MetadataType::TraitObject: make_fcn = "make_traitobjptr"; - m_of << "\t" << Trans_Mangle(p) << "( " << make_fcn << "(&"; emit_lvalue(slot); m_of << ", "; + m_of << "\t" << Trans_Mangle(p) << "( " << make_fcn << "("; + if( slot.is_Deref() ) + { + emit_lvalue(*slot.as_Deref().val); + m_of << ".PTR"; + } + else + { + m_of << "&"; emit_lvalue(slot); + } + m_of << ", "; const auto* lvp = &slot; while(const auto* le = lvp->opt_Field()) lvp = &*le->val; MIR_ASSERT(*m_mir_res, lvp->is_Deref(), "Access to unized type without a deref - " << *lvp << " (part of " << slot << ")"); @@ -2454,13 +2553,31 @@ namespace { ), (TraitObject, MIR_ASSERT(*m_mir_res, unsized_valid, "Dropping TraitObject without a pointer"); - //MIR_ASSERT(*m_mir_res, slot.is_Deref(), "Dropping a TraitObject through a non-Deref"); // Call destructor in vtable + const auto* lvp = &slot; + while(const auto* le = lvp->opt_Field()) lvp = &*le->val; + MIR_ASSERT(*m_mir_res, lvp->is_Deref(), "Access to unized type without a deref - " << *lvp << " (part of " << slot << ")"); + m_of << "((VTABLE_HDR*)"; emit_lvalue(*lvp->as_Deref().val); m_of << ".META)->drop("; + if( const auto* ve = slot.opt_Deref() ) + { + emit_lvalue(*ve->val); m_of << ".PTR"; + } + else + { + m_of << "&"; emit_lvalue(slot); + } + m_of << ");"; ), (Slice, MIR_ASSERT(*m_mir_res, unsized_valid, "Dropping Slice without a pointer"); - //MIR_ASSERT(*m_mir_res, slot.is_Deref(), "Dropping a slice through a non-Deref"); + const auto* lvp = &slot; + while(const auto* le = lvp->opt_Field()) lvp = &*le->val; + MIR_ASSERT(*m_mir_res, lvp->is_Deref(), "Access to unized type without a deref - " << *lvp << " (part of " << slot << ")"); // Call destructor on all entries + m_of << "for(unsigned i = 0; i < "; emit_lvalue(*lvp->as_Deref().val); m_of << ".META; i++) {"; + m_of << "\t\t"; + emit_destructor_call(::MIR::LValue::make_Index({ box$(slot.clone()), box$(::MIR::LValue::make_Temporary({~0u})) }), *te.inner, false); + m_of << "\n\t}"; ) ) } @@ -2582,10 +2699,10 @@ namespace { assign_from_literal([&](){ emit_dst(); m_of << "._0"; }, get_inner_type(e.idx, 0), e.vals[0]); } } - else if( enm.m_repr != ::HIR::Enum::Repr::Rust || ::std::all_of(enm.m_variants.begin(), enm.m_variants.end(), [](const auto& x){return x.second.is_Unit() || x.second.is_Value();}) ) + else if( enm.is_value() ) { MIR_ASSERT(*m_mir_res, e.vals.empty(), "Value-only enum with fields"); - emit_dst(); m_of << ".TAG = " << e.idx; + emit_dst(); m_of << ".TAG = " << enm.get_value(e.idx); } else { @@ -2645,7 +2762,11 @@ namespace { if( ' ' <= v && v < 0x7F && v != '"' && v != '\\' ) m_of << v; else - m_of << "\\" << (unsigned int)v; + { + m_of << "\\" << ((unsigned int)v & 0xFF); + if( isdigit( *(&v+1) ) ) + m_of << "\"\""; + } } m_of << "\"" << ::std::dec; m_of << ";\n\t"; @@ -2660,7 +2781,10 @@ namespace { m_of << "var" << e; ), (Temporary, - m_of << "tmp" << e.idx; + if( e.idx == ~0u ) + m_of << "i"; + else + m_of << "tmp" << e.idx; ), (Argument, m_of << "arg" << e.idx; @@ -2713,11 +2837,11 @@ namespace { ::HIR::TypeRef tmp; const auto& ty = m_mir_res->get_lvalue_type(tmp, val); auto dst_type = metadata_type(ty); - if( dst_type != MetadataType:: None ) + if( dst_type != MetadataType::None ) { m_of << "(*("; emit_ctype(ty); m_of << "*)"; emit_lvalue(*e.val); - m_of << ")"; + m_of << ".PTR)"; } else { @@ -2783,7 +2907,18 @@ namespace { if( c.v == INT64_MIN ) m_of << "INT64_MIN"; else + { m_of << c.v; + switch(c.t) + { + case ::HIR::CoreType::I64: + case ::HIR::CoreType::I128: + case ::HIR::CoreType::Isize: + m_of << "ll"; + default: + break; + } + } ), (Uint, switch(c.t) @@ -2800,7 +2935,7 @@ namespace { case ::HIR::CoreType::U64: case ::HIR::CoreType::U128: case ::HIR::CoreType::Usize: - m_of << ::std::hex << "0x" << c.v << ::std::dec; + m_of << ::std::hex << "0x" << c.v << "ull" << ::std::dec; break; case ::HIR::CoreType::Char: assert(0 <= c.v && c.v <= 0x10FFFF); @@ -2837,7 +2972,7 @@ namespace { if( ' ' <= v && v < 0x7F && v != '"' && v != '\\' ) m_of << v; else - m_of << "\\" << (unsigned int)v; + m_of << "\\" << ((unsigned int)v & 0xFF); } m_of << "\"" << ::std::dec; ), @@ -2847,7 +2982,7 @@ namespace { if( ' ' <= v && v < 0x7F && v != '"' && v != '\\' ) m_of << v; else - m_of << "\\" << (unsigned int)v; + m_of << "\\" << ((unsigned int)v & 0xFF); } m_of << "\", " << ::std::dec << c.size() << ")"; ), diff --git a/src/trans/enumerate.cpp b/src/trans/enumerate.cpp index 89846344..02116d18 100644 --- a/src/trans/enumerate.cpp +++ b/src/trans/enumerate.cpp @@ -515,6 +515,7 @@ namespace { // Ensure that the data trait's vtable is present const auto& trait = *te.m_trait.m_trait_ptr; + ASSERT_BUG(Span(), ! te.m_trait.m_path.m_path.m_components.empty(), "TODO: Data trait is empty, what can be done?"); auto vtable_ty_spath = te.m_trait.m_path.m_path; vtable_ty_spath.m_components.back() += "#vtable"; const auto& vtable_ref = m_crate.get_struct_by_path(sp, vtable_ty_spath); @@ -874,6 +875,8 @@ void Trans_Enumerate_Types(EnumState& state) for(const auto& v : se.inputs) H::visit_lvalue(tv,pp,fcn, v.second); ), + (ScopeEnd, + ), (Assign, H::visit_lvalue(tv,pp,fcn, se.dst); TU_MATCHA( (se.src), (re), @@ -989,6 +992,7 @@ void Trans_Enumerate_Types(EnumState& state) tv.m_resolve.expand_associated_types( sp, vtable_params.m_types[idx] ); } + tv.visit_type( *ent.first.m_data.as_UfcsKnown().type ); tv.visit_type( ::HIR::TypeRef( ::HIR::GenericPath(vtable_ty_spath, mv$(vtable_params)), &vtable_ref ) ); } @@ -1150,7 +1154,7 @@ namespace { ::std::vector<::HIR::TypeRef> best_impl_params; const ::HIR::TraitImpl* best_impl = nullptr; resolve.find_impl(sp, e.trait.m_path, e.trait.m_params, *e.type, [&](auto impl_ref, auto is_fuzz) { - DEBUG("Found " << impl_ref); + DEBUG("[get_ent_fullpath] Found " << impl_ref); //ASSERT_BUG(sp, !is_fuzz, "Fuzzy match not allowed here"); if( ! impl_ref.m_data.is_TraitImpl() ) { DEBUG("Trans impl search found an invalid impl type"); @@ -1201,6 +1205,8 @@ namespace { else BUG(sp, "Parameter " << i << " unset"); } + if( is_spec ) + DEBUG("- Specialisable"); return !is_spec; } return false; @@ -1457,6 +1463,8 @@ void Trans_Enumerate_FillFrom_MIR(EnumState& state, const ::MIR::Function& code, ), (SetDropFlag, ), + (ScopeEnd, + ), (Drop, DEBUG("- DROP " << se.slot); Trans_Enumerate_FillFrom_MIR_LValue(state, se.slot, pp); diff --git a/src/trans/main_bindings.hpp b/src/trans/main_bindings.hpp index 2878cc66..59933863 100644 --- a/src/trans/main_bindings.hpp +++ b/src/trans/main_bindings.hpp @@ -23,6 +23,7 @@ struct TransOptions }; extern TransList Trans_Enumerate_Main(const ::HIR::Crate& crate); +extern TransList Trans_Enumerate_Test(const ::HIR::Crate& crate); // NOTE: This also sets the saveout flags extern TransList Trans_Enumerate_Public(::HIR::Crate& crate); diff --git a/src/trans/monomorphise.cpp b/src/trans/monomorphise.cpp index 1688a8f5..3ac2ee09 100644 --- a/src/trans/monomorphise.cpp +++ b/src/trans/monomorphise.cpp @@ -150,6 +150,9 @@ namespace { case ::MIR::Statement::TAG_SetDropFlag: statements.push_back( ::MIR::Statement( stmt.as_SetDropFlag() ) ); break; + case ::MIR::Statement::TAG_ScopeEnd: + statements.push_back( ::MIR::Statement( stmt.as_ScopeEnd() ) ); + break; case ::MIR::Statement::TAG_Drop: { const auto& e = stmt.as_Drop(); DEBUG("- DROP " << e.slot); @@ -249,9 +252,21 @@ namespace { mv$(rval) }) ); } break; - case ::MIR::Statement::TAG_Asm: - TODO(params.sp, "Monormorphise asm!"); - break; + case ::MIR::Statement::TAG_Asm: { + const auto& e = stmt.as_Asm(); + DEBUG("- asm! \"" << e.tpl << "\""); + ::std::vector< ::std::pair<::std::string, ::MIR::LValue>> new_out, new_in; + new_out.reserve( e.outputs.size() ); + for(auto& ent : e.outputs) + new_out.push_back(::std::make_pair( ent.first, monomorph_LValue(resolve, params, ent.second) )); + new_in.reserve( e.inputs.size() ); + for(auto& ent : e.inputs) + new_in.push_back(::std::make_pair( ent.first, monomorph_LValue(resolve, params, ent.second) )); + + statements.push_back( ::MIR::Statement::make_Asm({ + e.tpl, mv$(new_out), mv$(new_in), e.clobbers, e.flags + }) ); + } break; } } diff --git a/src/trans/target.cpp b/src/trans/target.cpp new file mode 100644 index 00000000..21555101 --- /dev/null +++ b/src/trans/target.cpp @@ -0,0 +1,134 @@ +/* + * MRustC - Rust Compiler + * - By John Hodge (Mutabah/thePowersGang) + * + * trans/target.cpp + * - Target-specific information + */ +#include "target.hpp" + +// TODO: Replace with target selection +#define POINTER_SIZE_BYTES 8 + +bool Target_GetSizeAndAlignOf(const Span& sp, const ::HIR::TypeRef& ty, size_t& out_size, size_t& out_align) +{ + TU_MATCHA( (ty.m_data), (te), + (Infer, + BUG(sp, "sizeof on _ type"); + ), + (Diverge, + out_size = 0; + out_align = 0; + return true; + ), + (Primitive, + switch(te) + { + case ::HIR::CoreType::Bool: + case ::HIR::CoreType::U8: + case ::HIR::CoreType::I8: + out_size = 1; + out_align = 1; + return true; + case ::HIR::CoreType::U16: + case ::HIR::CoreType::I16: + out_size = 2; + out_align = 2; + return true; + case ::HIR::CoreType::U32: + case ::HIR::CoreType::I32: + case ::HIR::CoreType::Char: + out_size = 4; + out_align = 4; + return true; + case ::HIR::CoreType::U64: + case ::HIR::CoreType::I64: + out_size = 8; + out_align = 8; + return true; + case ::HIR::CoreType::U128: + case ::HIR::CoreType::I128: + out_size = 16; + // TODO: If i128 is emulated, this can be 8 + out_align = 16; + return true; + case ::HIR::CoreType::Usize: + case ::HIR::CoreType::Isize: + out_size = POINTER_SIZE_BYTES; + out_align = POINTER_SIZE_BYTES; + return true; + case ::HIR::CoreType::F32: + out_size = 4; + out_align = 4; + return true; + case ::HIR::CoreType::F64: + out_size = 8; + out_align = 8; + return true; + case ::HIR::CoreType::Str: + BUG(sp, "sizeof on a `str` - unsized"); + } + ), + (Path, + // TODO: + return false; + ), + (Generic, + // Unknown - return false + return false; + ), + (TraitObject, + BUG(sp, "sizeof on a trait object - unsized"); + ), + (ErasedType, + BUG(sp, "sizeof on an erased type - shouldn't exist"); + ), + (Array, + // TODO: + size_t size; + if( !Target_GetSizeAndAlignOf(sp, *te.inner, size,out_align) ) + return false; + size *= te.size_val; + ), + (Slice, + BUG(sp, "sizeof on a slice - unsized"); + ), + (Tuple, + out_size = 0; + out_align = 0; + + // TODO: Struct reordering + for(const auto& t : te) + { + size_t size, align; + if( !Target_GetSizeAndAlignOf(sp, t, size,align) ) + return false; + out_size += size; + out_align = ::std::max(out_align, align); + } + ), + (Borrow, + // TODO + ), + (Pointer, + // TODO + ), + (Function, + // Pointer size + ), + (Closure, + // TODO. + ) + ) + return false; +} +bool Target_GetSizeOf(const Span& sp, const ::HIR::TypeRef& ty, size_t& out_size) +{ + size_t ignore_align; + return Target_GetSizeAndAlignOf(sp, ty, out_size, ignore_align); +} +bool Target_GetAlignOf(const Span& sp, const ::HIR::TypeRef& ty, size_t& out_align) +{ + size_t ignore_size; + return Target_GetSizeAndAlignOf(sp, ty, ignore_size, out_align); +} diff --git a/src/trans/target.hpp b/src/trans/target.hpp new file mode 100644 index 00000000..1c081b54 --- /dev/null +++ b/src/trans/target.hpp @@ -0,0 +1,15 @@ +/* + * MRustC - Rust Compiler + * - By John Hodge (Mutabah/thePowersGang) + * + * trans/target.hpp + * - Target-specific information + */ +#pragma once + +#include <cstddef> +#include <hir/type.hpp> + +extern bool Target_GetSizeOf(const Span& sp, const ::HIR::TypeRef& ty, size_t& out_size); +extern bool Target_GetAlignOf(const Span& sp, const ::HIR::TypeRef& ty, size_t& out_align); + |