|
provided by Thomas Klausner.
Diffs from previous version:
2002-03-20 Michael Meeks <michael@ximian.com>
* configure.in: copy the autoconf macro into here so we
can rely on it not changing across autoconf versions.
2002-03-11 Michael Meeks <michael@ximian.com>
* src/IIOP/giop-msg-buffer.c
kill #define writev mess.
(giop_send_buffer_write): handle EINTR, handle EAGAIN
correctly, handle short writev's correctly, test
thorougly.
* src/ORBitutil/compat.c (g_writev): kill this
abomination.
2002-02-11 Sebastian Wilhelmi <wilhelmi@ira.uka.de>
* src/orb/allocators.h, src/orb/allocators.c (ORBit_alloc_2,
ORBit_free): Make the free marker really always sit directly in
front of the memory block, not at some random lower position.
2002-02-09 Elliot Lee <sopwith@redhat.com>
* src/ORBitutil/os-feature-alloca.h: Don't redefine alloca if it is already defined.
2002-01-27 Darin Adler <darin@bentspoon.com>
* test/everything/Makefile.am: Added -lm -- can't build without it.
2002-01-25 Mark McLoughlin <mark@skynet.ie>
* configure.in: (AC_CHECK_ALIGNOF): error out if an alignment
is found to be zero.
Thomas thought that the modification to src/IIOP/giop-msg-buffer.c to
"kill #define writev mess" might mean that we could drop the
NetBSD-specific wrapper for writev(2), but sadly this isn't so -
configure undefines HAVE_LIMITED_WRITEV on NetBSD, and rightly so,
since IOV_MAX (1024) iovecs can be written, not 50. However, we still
have the basic "attempt to send more than IOV_MAX iovecs and you get
EINVAL" problem with our writev(2) implementation.
|
|
calls writev(2). Some of the applications which use ORBit, such as
oaf, can send 1214 iovecs, which is slightly more than IOV_MAX. Add a
wrapper for writev(2), to check the number of iovecs passed to writev,
and loop, sending MIN(IOV_MAX, count) until the iovecs have all been
written.
|
