summaryrefslogtreecommitdiff
path: root/src/cmd/godoc/spec.go
diff options
context:
space:
mode:
Diffstat (limited to 'src/cmd/godoc/spec.go')
-rw-r--r--src/cmd/godoc/spec.go122
1 files changed, 61 insertions, 61 deletions
diff --git a/src/cmd/godoc/spec.go b/src/cmd/godoc/spec.go
index 15ce2fb9e..15f3cba20 100644
--- a/src/cmd/godoc/spec.go
+++ b/src/cmd/godoc/spec.go
@@ -11,35 +11,35 @@
package main
import (
- "bytes";
- "fmt";
- "go/scanner";
- "go/token";
- "io";
- "strings";
+ "bytes"
+ "fmt"
+ "go/scanner"
+ "go/token"
+ "io"
+ "strings"
)
type ebnfParser struct {
- out io.Writer; // parser output
- src []byte; // parser source
- scanner scanner.Scanner;
- prev int; // offset of previous token
- pos token.Position; // token position
- tok token.Token; // one token look-ahead
- lit []byte; // token literal
+ out io.Writer // parser output
+ src []byte // parser source
+ scanner scanner.Scanner
+ prev int // offset of previous token
+ pos token.Position // token position
+ tok token.Token // one token look-ahead
+ lit []byte // token literal
}
func (p *ebnfParser) flush() {
- p.out.Write(p.src[p.prev:p.pos.Offset]);
- p.prev = p.pos.Offset;
+ p.out.Write(p.src[p.prev:p.pos.Offset])
+ p.prev = p.pos.Offset
}
func (p *ebnfParser) next() {
- p.flush();
- p.pos, p.tok, p.lit = p.scanner.Scan();
+ p.flush()
+ p.pos, p.tok, p.lit = p.scanner.Scan()
if p.tok.IsKeyword() {
// TODO Should keyword mapping always happen outside scanner?
// Or should there be a flag to scanner to enable keyword mapping?
@@ -54,38 +54,38 @@ func (p *ebnfParser) Error(pos token.Position, msg string) {
func (p *ebnfParser) errorExpected(pos token.Position, msg string) {
- msg = "expected " + msg;
+ msg = "expected " + msg
if pos.Offset == p.pos.Offset {
// the error happened at the current position;
// make the error message more specific
- msg += ", found '" + p.tok.String() + "'";
+ msg += ", found '" + p.tok.String() + "'"
if p.tok.IsLiteral() {
msg += " " + string(p.lit)
}
}
- p.Error(pos, msg);
+ p.Error(pos, msg)
}
func (p *ebnfParser) expect(tok token.Token) token.Position {
- pos := p.pos;
+ pos := p.pos
if p.tok != tok {
p.errorExpected(pos, "'"+tok.String()+"'")
}
- p.next(); // make progress in any case
- return pos;
+ p.next() // make progress in any case
+ return pos
}
func (p *ebnfParser) parseIdentifier(def bool) {
- name := string(p.lit);
- p.expect(token.IDENT);
+ name := string(p.lit)
+ p.expect(token.IDENT)
if def {
fmt.Fprintf(p.out, `<a id="%s">%s</a>`, name, name)
} else {
fmt.Fprintf(p.out, `<a href="#%s" class="noline">%s</a>`, name, name)
}
- p.prev += len(name); // skip identifier when calling flush
+ p.prev += len(name) // skip identifier when calling flush
}
@@ -95,32 +95,32 @@ func (p *ebnfParser) parseTerm() bool {
p.parseIdentifier(false)
case token.STRING:
- p.next();
+ p.next()
if p.tok == token.ELLIPSIS {
- p.next();
- p.expect(token.STRING);
+ p.next()
+ p.expect(token.STRING)
}
case token.LPAREN:
- p.next();
- p.parseExpression();
- p.expect(token.RPAREN);
+ p.next()
+ p.parseExpression()
+ p.expect(token.RPAREN)
case token.LBRACK:
- p.next();
- p.parseExpression();
- p.expect(token.RBRACK);
+ p.next()
+ p.parseExpression()
+ p.expect(token.RBRACK)
case token.LBRACE:
- p.next();
- p.parseExpression();
- p.expect(token.RBRACE);
+ p.next()
+ p.parseExpression()
+ p.expect(token.RBRACE)
default:
return false
}
- return true;
+ return true
}
@@ -132,70 +132,70 @@ func (p *ebnfParser) parseSequence() {
func (p *ebnfParser) parseExpression() {
for {
- p.parseSequence();
+ p.parseSequence()
if p.tok != token.OR {
break
}
- p.next();
+ p.next()
}
}
func (p *ebnfParser) parseProduction() {
- p.parseIdentifier(true);
- p.expect(token.ASSIGN);
- p.parseExpression();
- p.expect(token.PERIOD);
+ p.parseIdentifier(true)
+ p.expect(token.ASSIGN)
+ p.parseExpression()
+ p.expect(token.PERIOD)
}
func (p *ebnfParser) parse(out io.Writer, src []byte) {
// initialize ebnfParser
- p.out = out;
- p.src = src;
- p.scanner.Init("", src, p, 0);
- p.next(); // initializes pos, tok, lit
+ p.out = out
+ p.src = src
+ p.scanner.Init("", src, p, 0)
+ p.next() // initializes pos, tok, lit
// process source
for p.tok != token.EOF {
p.parseProduction()
}
- p.flush();
+ p.flush()
}
// Markers around EBNF sections
var (
- openTag = strings.Bytes(`<pre class="ebnf">`);
- closeTag = strings.Bytes(`</pre>`);
+ openTag = strings.Bytes(`<pre class="ebnf">`)
+ closeTag = strings.Bytes(`</pre>`)
)
func linkify(out io.Writer, src []byte) {
for len(src) > 0 {
- n := len(src);
+ n := len(src)
// i: beginning of EBNF text (or end of source)
- i := bytes.Index(src, openTag);
+ i := bytes.Index(src, openTag)
if i < 0 {
i = n - len(openTag)
}
- i += len(openTag);
+ i += len(openTag)
// j: end of EBNF text (or end of source)
- j := bytes.Index(src[i:n], closeTag); // close marker
+ j := bytes.Index(src[i:n], closeTag) // close marker
if j < 0 {
j = n - i
}
- j += i;
+ j += i
// write text before EBNF
- out.Write(src[0:i]);
+ out.Write(src[0:i])
// parse and write EBNF
- var p ebnfParser;
- p.parse(out, src[i:j]);
+ var p ebnfParser
+ p.parse(out, src[i:j])
// advance
- src = src[j:n];
+ src = src[j:n]
}
}