[Bug bootstrap/87252] gcc-4.4 cross-builds broken, apparently in self-tests

[Bug bootstrap/87252] gcc-4.4 cross-builds broken, apparently in self-tests

[abrahm.scully at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87252

--- Comment #7 from Abrahm Scully <abrahm.scully at gmail dot com> ---
I hit an ICE today that looks like this bug. I attempted to build
gcc-10-20200329 on 32-bit CentOS 6 using g++ 4.4.7-23.el6. I don't see this bug
building gcc-6.3, gcc-7.3, gcc-8.3, or gcc-9-20200118 on the same platform.

I'll dig into the problem more soon...

...
/build/Sonar/Alofts/tools/.build1/gcc-10-20200329/build/./gcc/xgcc
-B/build/Sonar/Alofts/tools/.build1/gcc-10-20200329/build/./gcc/ -xc -nostdinc
/dev/null -S -o /dev/null -fself-test=../../gcc/testsuite/selftests
pure virtual method called
terminate called without an active exception
cc1: internal compiler error: Aborted
0x8a9d967 crash_signal
        ../../gcc/toplev.c:328
0x9815c35 print_any_labels
        ../../gcc/diagnostic-show-locus.c:1691
0x9815c35 print_line
        ../../gcc/diagnostic-show-locus.c:2512
0x9819400 diagnostic_show_locus(diagnostic_context*, rich_location*,
diagnostic_t)
        ../../gcc/diagnostic-show-locus.c:2595
0x981e9c2 test_one_liner_labels_utf8
        ../../gcc/diagnostic-show-locus.c:3816
0x981fc0b test_diagnostic_show_locus_one_liner_utf8
        ../../gcc/diagnostic-show-locus.c:3902
0x9831637 selftest::for_each_line_table_case(void (*)(selftest::line_table_case
const&))
        ../../gcc/input.c:3571
0x9819d05 selftest::diagnostic_show_locus_c_tests()
        ../../gcc/diagnostic-show-locus.c:5021
0x9754969 selftest::run_tests()
        ../../gcc/selftest-run-tests.c:95
0x8a9e0e9 toplev::run_self_tests()
        ../../gcc/toplev.c:2337
Please submit a full bug report,
with preprocessed source if appropriate.
Please include the complete backtrace with any bug report.
See <https://gcc.gnu.org/bugs/> for instructions.
make[2]: *** [s-selftest-c] Error 1
make[2]: Leaving directory
`/build/Sonar/Alofts/tools/.build1/gcc-10-20200329/build/gcc'
make[1]: *** [all-gcc] Error 2
make[1]: Leaving directory
`/build/Sonar/Alofts/tools/.build1/gcc-10-20200329/build'
make: *** [all] Error 2

[Bug bootstrap/87252] gcc-4.4 cross-builds broken, apparently in self-tests

[abrahm.scully at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87252

--- Comment #8 from Abrahm Scully <abrahm.scully at gmail dot com> ---
I don't think anything is wrong with gcc-10-20200329. The code looks fine.

I realized later that the versions of gcc I mentioned where I don't see this
problem are all from release branches. As described in
https://gcc.gnu.org/install/configure.html, the --enable-checking default
behavior is different in release branches. They just don't run the test during
the build.

Configuring the gcc-10-20200329 snapshot with --enable-checking=release skips
the test and the build finishes fine.

Sorry for the noise, but thanks for listening.

[Bug bootstrap/87252] gcc-4.4 cross-builds broken, apparently in self-tests

[abrahm.scully at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87252

--- Comment #9 from Abrahm Scully <abrahm.scully at gmail dot com> ---
I'm no longer convinced that I didn't see the problem previously because I just
wasn't running the tests. Stage 1 has checking enabled... so I don't know why
this problem showed up for others in gcc 9 but not for me until gcc 10.

Either way, building gcc-4.7.4 first and then building gcc-10 with that
produces a compiler without the problem.

Again, sorry for the noise.

[Bug bootstrap/87252] gcc-4.4 cross-builds broken, apparently in self-tests

[hp at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87252

--- Comment #10 from Hans-Peter Nilsson <hp at gcc dot gnu.org> ---
(In reply to Abrahm Scully from comment #9)
> Either way, building gcc-4.7.4 first and then building gcc-10 with that
> produces a compiler without the problem.
> 
> Again, sorry for the noise.

Please, do *not* apologize.  Even if you'd just repeat other observations, your
observations would serve as a confirmation.  But they're also new AFAICT.

So, your recent observation is that gcc-4.7.4 apparently works here.  Looking
deeper into what construct is used in the self-tests, that started working with
that version would likely help, if someone is so inclined.

Or, maybe it'd be ok to just disable the self-testing code before that version.
 Alternatively, plain raising the base version of the gcc being supported for
gcc *development* to 4.7.4 would probably be acceptable too.  To newcomers:
remember, this is just the cross-gcc case.

[Bug bootstrap/87252] gcc-4.4 cross-builds broken, apparently in self-tests

[hp at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87252

--- Comment #11 from Hans-Peter Nilsson <hp at gcc dot gnu.org> ---
(In reply to Hans-Peter Nilsson from comment #0)
> (Note: no cross-binutils or anything needed, just gcc sources and a native
> gcc-4.4. This is an *old* installation which identifies itself as "gcc
> version 4.4.4 20100630 (Red Hat 4.4.4-10) (GCC)".)

Distribution: "Fedora 12".