[Bug driver/112836] New: gcc fails when job control is used

[Bug driver/112836] New: gcc fails when job control is used

[bruno at clisp dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112836

            Bug ID: 112836
           Summary: gcc fails when job control is used
           Product: gcc
           Version: 13.2.1
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: driver
          Assignee: unassigned at gcc dot gnu.org
          Reporter: bruno at clisp dot org
  Target Milestone: ---

A compilation command with gcc failed, when I used bash's job control:

------------------------------------------------------------------------
$ ~/build-32 -C
...
gcc -m32 -DHAVE_CONFIG_H -DEXEEXT=\"\" -DEXEEXT=\"\" -DNO_XMALLOC -DEXEEXT=\"\"
-I. -I../../gllib -I..  -DGNULIB_STRICT_CHECKING=1
-I/home/bruno/prefix32/include -Wall -fvisibility=hidden -g -O2 -MT
git-merge-changelog.o -MD -MP -MF $depbase.Tpo -c -o git-merge-changelog.o
../../gllib/git-merge-changelog.c &&\
mv -f $depbase.Tpo $depbase.Po
^Z
[1]+  Stopped                 ~/build-32 -C
bruno@localhost:~/testdir-all$ free
               total        used        free      shared  buff/cache  
available
Mem:         2056072      119664     1773880         632      224808    
1936408
Swap:         514072           0      514072
bruno@localhost:~/testdir-all$ fg 1
~/build-32 -C
gcc: fatal error: failed to get exit status: Interrupted system call
compilation terminated.
make[4]: *** [Makefile:11320: git-merge-changelog.o] Error 1
make[4]: Leaving directory '/home/bruno/testdir-all/build-32/gllib'
make[3]: *** [Makefile:11366: all-recursive] Error 1
make[3]: Leaving directory '/home/bruno/testdir-all/build-32/gllib'
make[2]: *** [Makefile:8407: all] Error 2
make[2]: Leaving directory '/home/bruno/testdir-all/build-32/gllib'
make[1]: *** [Makefile:3220: all-recursive] Error 1
make[1]: Leaving directory '/home/bruno/testdir-all/build-32'
make: *** [Makefile:3151: all] Error 2
------------------------------------------------------------------------

The expected behaviour that after 'fg 1', stopped jobs continue to run
without failing.

This is on Linux/sparc64. Versions:
  - gcc (T2SDE) 13.2.1 20231124
  - Linux 6.6.3
  - glibc 2.38

[Bug driver/112836] gcc fails when job control is used

[bruno at clisp dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112836

--- Comment #1 from Bruno Haible <bruno at clisp dot org> ---
Created attachment 56779
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=56779&action=edit
proposed fix

Although the error is not easily reproducible, it is easy to analyze
and fix:

The piece of error message string "failed to get exit status" occurs
only in one place: in gcc/gcc.cc line 3462.

The only possible cause for the error message
"failed to get exit status: Interrupted system call" is therefore
that pex_get_status returned 0, with error code EINTR.

pex_get_status is defined in libiberty/pex-common.c, line 547.
It returns 0 only in line 555. We can infer that the function
pex_get_status_and_time had returned 0, with error code EINTR.

pex_get_status_and_time is defined in libiberty/pex-common.c, line 518.
It returns 0 only when ret gets to 0 in line 537. We can infer
that the obj->funcs->wait function must have returned non-zero,
with error code EINTR.

obj->funcs->wait is the method declared in libiberty/pex-common.h
line 128. On this platform, the relevant implementation is in
libiberty/pex-unix.c. The initialization happens in
libiberty/pex-unix.c line 338:
  return pex_init_common (flags, pname, tempbase, &funcs);
The obj->funcs->wait function function is thus pex_unix_wait.
It must have returned non-zero, with error code EINTR.

pex_unix_wait is defined in libiberty/pex-unix.c line 938.
Looking at its code, the function pex_wait must have returned a
negative value, with errno being set to EINTR.

pex_wait has 4 possible implementations, in
  libiberty/pex-unix.c line 128
  libiberty/pex-unix.c line 159
  libiberty/pex-unix.c line 170
  libiberty/pex-unix.c line 219
Since glibc has the 'wait4' function, the implementation that
matters is the one from libiberty/pex-unix.c line 128. But the
other implementations have the same problem: they don't handle
EINTR so far.

Find attached a proposed patch.

[Bug other/112836] gcc fails when job control is used

[sjames at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112836

--- Comment #2 from Sam James <sjames at gcc dot gnu.org> ---
Thanks for chasing this down. This has been driving me nuts for a while but I
hadn't dug into it yet.

[Bug other/112836] gcc fails when job control is used

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112836

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|                            |2023-12-04
             Status|UNCONFIRMED                 |NEW
     Ever confirmed|0                           |1

--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
Can you post it to gcc-patches@gcc.gnu.org please?

[Bug other/112836] gcc fails when job control is used

[glaubitz at physik dot fu-berlin.de]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112836

--- Comment #4 from John Paul Adrian Glaubitz <glaubitz at physik dot fu-berlin.de> ---
I tried this patch but it does not address the issue with posix_spawn that I am
seeing.

Trying to build gcc from git on Linux sparc64 with glibc 2.37 with the
following configuration:

./configure --enable-languages=c,c++ --disable-bootstrap --disable-multilib
--disable-nls --host=sparc64-unknown-linux-gnu

fails with:

xgcc: fatal error: cannot execute '/home/glaubitz/gcc/build/./gcc/cc1plus':
posix_spawn: Bad address
compilation terminated.

This can be worked around when building with one parallel job (make -j1).

[Bug other/112836] gcc fails when job control is used

[bruno at clisp dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112836

--- Comment #5 from Bruno Haible <bruno at clisp dot org> ---
(In reply to John Paul Adrian Glaubitz from comment #4)
> I tried this patch but it does not address the issue with posix_spawn that I
> am seeing.
> 
> Trying to build gcc from git on Linux sparc64 with glibc 2.37 with the
> following configuration:
> 
> ./configure --enable-languages=c,c++ --disable-bootstrap --disable-multilib
> --disable-nls --host=sparc64-unknown-linux-gnu
> 
> fails with:
> 
> xgcc: fatal error: cannot execute '/home/glaubitz/gcc/build/./gcc/cc1plus':
> posix_spawn: Bad address
> compilation terminated.
> 
> This can be worked around when building with one parallel job (make -j1).

This appears to be a different bug, because
  - it appears under different conditions (parallel make, not Ctrl-Z / 'fg'),
  - the message is different ("posix_spawn: Bad address", not "failed to get
exit status: Interrupted system call").

Could you please register it as a separate bug under
https://gcc.gnu.org/bugzilla/ ?