Re: vismain failure

["H . J . Lu" <hjl at lucon dot org>]

On Tue, Jul 25, 2000 at 08:12:45PM +0200, Mark Kettenis wrote:
>    From: Andreas Jaeger <aj@suse.de>
>    Date: 25 Jul 2000 18:27:37 +0200
> 
>    >>>>> Bruno Haible writes:
> 
>     > Andreas Jaeger writes:
>    >> But 2.9.5.0.24 has the features but is buggy and we can't (so far at
>    >> least) check for this.
> 
>     > But you can add an AC_CHECK_PROG_VER test for it in configure.in and
>     > refuse to configure glibc if this particular ld version is used.
> 
>    We could try.  The problem is that glibc currently works also with
>    older binutils, AFAIK only some versions in between are broken.
> 
> Please keep in mind that glibc itself doesn't use the visibility
> stuff.  It's just the test that fails.

Glibc doesn't use the ELF visibility directly. But it uses it
indirectly via __dso_handle in crtbegin.o/crtend.o. That is if gcc
is built by the binutils with the ELF visibility support, glibc will
 use it:

# find -type f | xargs grep __dso_handle
./linuxthreads/ChangeLog:       * pthread.c (pthread_initialize): Test for address of __dso_handle
./linuxthreads/pthread.c:extern void *__dso_handle __attribute__ ((weak));
./linuxthreads/pthread.c:  if (&__dso_handle != NULL)
./linuxthreads/pthread.c:    __cxa_atexit((void (*) (void *)) pthread_exit_process, NULL, __dso_handle);


H.J.