| Age | Commit message (Collapse) | Author |
|---|
|
|
|
collect2: ld terminated with signal 11 [Segmentation fault]
|
|
(Modified as suggested by millert@)
|
|
|
|
|
|
Thanks to Marco for pointing that out.
|
|
|
|
This fixes an important bug: libc now uses weak symbols a lot.
Without this patch, dynamic linking with libraries that reference libc
would mean those symbols would not be resolved, as weak symbols are good
enough for linking, but 2nd hand libraries symbols are not.
Not 100% sure this won't trigger problems later (nothing yet), but this
fixes the immediate problem.
Approved by niklas@ and millert@
|
|
- Makefile.bsd-wrapper: man pages, disable NLS for now.
- doc/Makefile.in: install man pages manually, remove buggy targets that
would break `make clean'.
- makeinfo/Makefile.in: shell failure ??? rework problematic line.
- util/texindex.c: let maketempname create the file, remove race condition.
|
|
A bit more grumpy about bad texinfo, though.
|
|
|
|
o Don't try and look up an empty element in the hints file
o Convert "" to "." when adding to the list of search dirs
Perviously, an LD_LIBRARY_PATH that ended in ':' was ignored.
|
|
if you are building a shared lib from a .a full of pic objects. This
is not how perl builds shared libs and it hoses tk.
Also sink with the hints file in perl5.005_63 (cosmetic)
|
|
* Rely on BYTE_ORDER instead of specific values of XHOST to determine
cross-endianness,
* Link necessary cross-includes to the right location,
* Add some necessary stubs for cross-endian sparc and cross-endian m68k.
This does let i386->m68k work, and probably brings i386->sparc most of the
way there...
Approved by mickey@, niklas@ is to busy to answer email as usual...
|
|
some help from him.
Let gcc generate more sensible code for stack adjustments.
Specifically, gcc is a bit lame in that area, as it does emit stack
instructions `specially', without using all its optimizer machinery for it.
So, at times, you will get
subl %esp,12
subl %esp,8
and other such stupid sequences out of it.
This peephole matches at least some of those innane sequences and optimizes
them.
Shrinks code a little bit.
|
|
|
|
|
|
|
|
|
|
Tested by espie@
|
|
loclibpth to be /usr/local/include and /usr/local/lib respectively.
This allows third party modules to grab stuff from /usr/local/{lib,include}
but perl itself doesn't search for things there during a build.
|
|
handle reported_undefineds correctly.
Probably unseen before now because this warning only occurred for symbols
that also triggered other warnings ? or maybe no-one cares.
|
|
the pic register).
FreeBSD PR 3441... Thanks to David O'Brien for letting me know,
and Alexander N. Kabaev for the actual fix.
|
|
was not what autoconf would have produced starting from configure.in + our
local changes...
Regen'ed for correct configure.in line numbers.
|
|
- patch to function.c was badly indented.
- invoke.texi no longer need to special-case OpenBSD, as 2.95.2 turns
strict-aliasing off everywhere.
|
|
|
|
in /usr/local/lib.
|
|
|
|
and major !=-1 anywhere).
Simplify test logic: it's enough to check path != NULL to know whether
we found something. cmpdewey() code is enough to ensure we get the best
one (libfoo.a doesn't change n, any appropriate libfoo.so will take
precedence).
|
|
we found something correct or not.
|
|
|
|
|
|
|
|
be looked for in the libary search path. If a name has a '/' in it,
treat it as a pathname, even if it starts with "lib"; gsar@ActiveState.com
|
|
|
|
|
|
ld (and ld.so) should take the shared library of the highest version
in the first directory that it founds. this is the way ld/ld.so has
traditionally worked since SunOS 4 and this is the way it should work.
before changing this back - please make sure you understand the
semantics of this and that you are not just hiding some other bug but
toggling this change. also verify the example in the PR.
see PR/972
|
|
|
|
|
|
if -Wtraditional, warn about ANSI cpp string concatenation.
|
|
Amazed, this bug has been in the gcc main source for almost a year.
Is this file compiled somewhere ?
(reported to gcc folks, thanks to Andreas Gunnarsson for finding the
bug)
|
|
|
|
Fix prototypes, propagate const where applicable.
|
|
Also checked other calls to
error
warning
fatal
error_with_file_and_line
warning_with_file_and_line
fatal_with_file_and_line
lex_error
lex_warning
c_error
|
|
|
|
(No need to wait for a rebuild, as the bug only manifests itself as
a warning anyway).
|
|
|
|
|
|
|
|
instead of explicitly in the top-level Makefile.
|
