This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: 1008 segfaults in genattr
> > 4) Make the test case call _exit() instead of exit(), but just
> > let code in the field learn this the hard way.
>
> I think I'm going to go with #4.
OK. There's an implied "and put back have_atexit".
Please apply these patches:
Fri Nov 21 07:43:04 1997 Robert Lipe <robertl@dgii.com>
* sco5.h: revert previous change. The cost of picking up
more code from libgcc2 exceeds the cost of p9732b.C having
static destructors that call exit calling abort.
*** config/i386/sco5.h__ Fri Nov 21 07:42:03 1997
--- config/i386/sco5.h Fri Nov 21 07:26:46 1997
***************
*** 930,933 ****
--- 930,934 ----
# if defined (_SCO_ELF)
# define OBJECT_FORMAT_ELF
+ # define HAVE_ATEXIT 1
# define INIT_SECTION_ASM_OP INIT_SECTION_ASM_OP_ELF
# define FINI_SECTION_ASM_OP FINI_SECTION_ASM_OP_ELF
Fri Nov 21 07:43:04 1997 Robert Lipe <robertl@dgii.com>
* p9732b.C: Don't call exit from static destructors. Since
we may be calling them via atexit, we want to avoid cleanup
recursion.
*** testsuite/g++.old-deja/g++.mike/p9732b.C_ Tue Aug 19 02:34:53 1997
--- testsuite/g++.old-deja/g++.mike/p9732b.C Fri Nov 21 07:47:13 1997
***************
*** 21,28 ****
else
{
if (count != 0)
! exit (1);
! exit (0);
}
}
} d;
--- 21,28 ----
else
{
if (count != 0)
! _exit (1);
! _exit (0);
}
}
} d;
> Shouldn't OpenServer COFF also define HAVE_ATEXIT?
Morally, yes, it should.
Even though it is, at that point, a COFF target, it still
has a _fini section that gets registered to execute via
atexit().
$ file crtendS.o ; dis crtendS.o | tail
crtendS.o: iAPX 386 COFF object file not stripped - version 30821
2b: 90 nop
section .fini
.fini
__DTOR_END__
2c: 00 00 addb %al,(%eax)
2e: 00 00 addb %al,(%eax)
__FRAME_END__
30: 00 00 addb %al,(%eax)
32: 00 00 addb %al,(%eax)
As you can see, we are going to try to execute this (after we've done
all our destructors which is done first) we're headed for a SIGILL.
Can you help me untangle the right combination to get a useful
binary? Between crtstuff.c and gbl-ctors.h and all the preprocessor
voodoo going on, I can't grok it. There is much voodoo at the end
of sco5.h to try to outsmart all the voodoo in the other two files.
I spent many hours on this when I first did it and just decide to
punt this issue and let COFF not use ATEXIT.