This is the mail archive of the
gdb-patches@sourceware.org
mailing list for the GDB project.
Re: [GCC PR55056] Re: [RFC patch] testsuite: Workaround issues with GCC 4.8.0pre + gdb.trace new KFAIL
- From: Jan Kratochvil <jan dot kratochvil at redhat dot com>
- To: Doug Evans <dje at google dot com>
- Cc: Thomas Schwinge <thomas at codesourcery dot com>, gdb-patches at sourceware dot org
- Date: Wed, 10 Sep 2014 22:06:57 +0200
- Subject: Re: [GCC PR55056] Re: [RFC patch] testsuite: Workaround issues with GCC 4.8.0pre + gdb.trace new KFAIL
- Authentication-results: sourceware.org; auth=none
- References: <20130126202645 dot GA4888 at host2 dot jankratochvil dot net> <5106D7E1 dot 1060705 at codesourcery dot com> <20130129174955 dot GA17823 at host2 dot jankratochvil dot net> <20130202092240 dot GA24458 at host2 dot jankratochvil dot net> <20130203172721 dot GA8611 at host2 dot jankratochvil dot net> <87r4g4vo94 dot fsf at kepler dot schwinge dot homeip dot net> <20130614133849 dot GA12925 at host2 dot jankratochvil dot net> <21519 dot 28248 dot 577451 dot 227482 at ruffy2 dot mtv dot corp dot google dot com>
On Tue, 09 Sep 2014 23:17:12 +0200, Doug Evans wrote:
> Have you thought of how you might do it differently now?
gcc-4.4.7-4.el6.x86_64 at least for restore.exp produces with line 75:
gdb_test "info addr l$var"
only patterns:
Symbol "l\d+" is a variable in $r\d+.
While gcc-4.9.1-7.fc21.x86_64 with the patch of mine we discuss here produces
in half the cases:
Symbol "l1" is a complex DWARF expression: 0: DW_OP_fbreg -44
And therefore the testcase no longer really tests the GDB feature.
Therefore proposing to make it x86_64 arch specific testcase with
gcc-4.4.7-4.el6.x86_64-precompiled .S file.
For the test I tried only restore.exp now, I will check more the others.
Another possibility would be to make the testcase -Og compatible as GCC
hackers in
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=55056
say that the variables then should not be <optimized out>.
I have tested that with -Og the testcase currently FAILs already on various
line-matching issues. But I find the arch-specific .S file safer.
Jan