On Linux/ia32, revision 180194: http://gcc.gnu.org/ml/gcc-cvs/2011-10/msg00790.html caused: FAIL: gcc.dg/guality/pr43177.c -O2 -flto line 15 x == 7 FAIL: gcc.dg/guality/pr43177.c -O2 -flto line 24 x == 7 FAIL: gcc.dg/guality/pr43177.c -O2 -flto -flto-partition=none line 15 x == 7 FAIL: gcc.dg/guality/pr43177.c -O2 -flto -flto-partition=none line 24 x == 7
On it. FWIW, I don't get these failures with rev 180136, with or without the patch, and I can't trigger them at 180194 either. Can you get more info on what the failure is (say, debug dumps, asm output, whatever) so I can try to blind-debug it? TIA,
I saw them on Fedora 15 with gdb-7.3-43.fc15.x86_64.
I'm on BLAG 140k/x86_64 (=~ Fedora 14). The fact that the same compiler code works on a similar platform suggests to me that the problem is not in the compiler, but rather in the linker or the debugger. I'm confused as to your report. You wrote Linux/ia32, but it's a GDB for x86_64, which suggests a 64-x-32 build or a 32-bit bootstrap starting with a 64-bit -m32 compiler, or maybe a 32-bit bootstrap compiler on a x86_64 platform. Anyhow, I tried all but the last of these options, to no avail, so I'll leave this alone unless I get more evidence that this is indeed a problem in the compiler, rather than some bug elsewhere that the patch happened to expose. For the record, I've used: gcc-4.5.1-4.fc14.x86_64 (with -m32) glibc-2.14-4.i686 binutils-2.20.51.0.7-8.fc14.x86_64 gdb-7.2-51.fc14.x86_64 and ld.bfd is the linker used by the bootstrap and the built compilers.
Just tried with ld.gold instead, still no failure.
You should try gdb 7.3 to see if it makes a difference.
What kind of error are you getting from gdb 7.3? Since 7.2 is getting the correct info, that's the bug report that ought to be submitted to GDB.
(In reply to comment #6) > What kind of error are you getting from gdb 7.3? Since 7.2 is getting the > correct info, that's the bug report that ought to be submitted to GDB. My understanding is most of guality tests are skipped with older GDB, like # make check-gcc RUNTESTFLAGS="guality.exp=pr43177.c" ... # of expected passes 20 # of unsupported tests 28 Is gcc.dg/guality/pr43177.c tested with GDB 7.2?
Err, that doesn't seem to tell what errors you're getting from gdb. I know older GDB might cause tests to be skipped, but that's not the case here. I'm getting the expected 48 passes and nothing else from make check-gcc guality.exp=pr43177.c
(In reply to comment #6) > What kind of error are you getting from gdb 7.3? Since 7.2 is getting the > correct info, that's the bug report that ought to be submitted to GDB. I got Breakpoint 1, baz (x=<optimized out>) at /export/gnu/import/git/gcc/gcc/testsuite/gcc.dg/guality/pr43177.c:24^M 24 bar (l);/* { dg-final { gdb-test 24 "x" "7" } } */^M $1 = <optimized out>^M $2 = 7^M <optimized out> != 7 It could be a GDB regression.
Yep. That, or some other difference in the compiler's assembly output, that I asked for in comment 1.