Re: GCC 3.0 and GLIBC don't work together

[Paolo Carlini <pcarlini at unitus dot it>]

Hi,

thank you very much for you detailed feedback (by the way, good luck for your prelinker!)

IMHO, what are you saying is very important for the *whole* community of Gcc and Glibc2.2.3
users, and should find its way in the official documents, INSTALL, HowTo, or whatever!!!

... at least until 2.2.4 will solve more fundamentally the basic issue...

Thanks again,
Paolo Carlini.


Jakub Jelinek wrote:

> On Wed, Jun 13, 2001 at 10:08:47AM +0200, Paolo Carlini wrote:
> > Hi again...
> >
> > a short follow-up to my previous message. Eventually I found that message by Drepper which
> > is troubling me (in libc-hacker):
> >
> >     http://sources.redhat.com/ml/libc-hacker/2001-03/msg00033.html
> >
> > The problem is still there or not????
>
> It is.
>
> You'd have patch gcc-2.95.3 to add __dso_handle and configure test for
> HAVE_GAS_HIDDEN (plus the atexit patch).
> Something like below (untested, plus you need to rebuild configure).
> Alternatively, gcc-2.96-RH will build glibc just fine, so will gcc-3.0
> snapshots before new unwind code was added, provided you build glibc with
> -static-libgcc.
>
> --- gcc-2.95.x/config/alpha/crtbegin.asm.jj     Wed Dec 16 22:00:53 1998
> +++ gcc-2.95.x/config/alpha/crtbegin.asm        Wed Jun 13 12:41:28 2001
> @@ -97,6 +97,31 @@ __EH_FRAME_BEGIN__:
>   # Support recursive calls to exit.
>  $ptr:  .quad   __DTOR_LIST__
>
> +/* A globally unique widget for c++ local destructors to hang off.
> +
> +   This has a unique value in every dso; in the main program its
> +   value is zero.  The object should be protected.  This means the
> +   instance in any dso or the main program is not used in any other
> +   dso.  The dynamic linker takes care of this.  */
> +
> +       .global __dso_handle
> +       .type __dso_handle,@object
> +       .size __dso_handle,8
> +#ifdef SHARED
> +.section .data
> +       .align 3
> +__dso_handle:
> +       .quad   __dso_handle
> +#else
> +.section .bss
> +       .align 3
> +__dso_handle:
> +       .zero 8
> +#endif
> +#ifdef HAVE_GAS_HIDDEN
> +       .hidden __dso_handle
> +#endif
> +
>  .text
>
>         .align 3
> --- gcc-2.95.x/crtstuff.c.jj    Tue Mar 23 01:43:51 1999
> +++ gcc-2.95.x/crtstuff.c       Wed Jun 13 12:39:14 2001
> @@ -135,6 +135,25 @@ typedef void (*func_ptr) (void);
>
>  #ifdef OBJECT_FORMAT_ELF
>
> +/* Declare the __dso_handle variable.  It should have a unique value
> +   in every shared-object; in a main program its value is zero.  The
> +   object should in any case be protected.  This means the instance
> +   in one DSO or the main program is not used in another object.  The
> +   dynamic linker takes care of this.  */
> +
> +/* XXX Ideally the following should be implemented using
> +   __attribute__ ((__visibility__ ("hidden")))
> +   but the __attribute__ support is not yet there.  */
> +#ifdef HAVE_GAS_HIDDEN
> +asm (".hidden\t__dso_handle");
> +#endif
> +
> +#ifdef CRTSTUFFS_O
> +void *__dso_handle = &__dso_handle;
> +#else
> +void *__dso_handle = 0;
> +#endif
> +
>  /* Run all the global destructors on exit from the program.  */
>
>  /* Some systems place the number of pointers in the first word of the
> --- gcc-2.95.x/configure.in.jj  Wed Oct 13 09:58:02 1999
> +++ gcc-2.95.x/configure.in     Wed Jun 13 12:44:34 2001
> @@ -4053,6 +4053,21 @@ EOF
>  fi
>  AC_MSG_RESULT($gcc_cv_as_subsections)
>
> +AC_MSG_CHECKING(assembler hidden support)
> +gcc_cv_as_hidden=
> +if test x$gcc_cv_as != x; then
> +       # Check if we have .hidden
> +       echo "  .hidden foobar" > conftest.s
> +       echo "foobar:" >> conftest.s
> +       if $gcc_cv_as -o conftest.o conftest.s > /dev/null 2>&1; then
> +               AC_DEFINE(HAVE_GAS_HIDDEN, 1,
> +                       [Define if your assembler supports .hidden.])
> +               gcc_cv_as_hidden="yes"
> +       fi
> +       rm -f conftest.s conftest.o conftest.nm1 conftest.nm2
> +fi
> +AC_MSG_RESULT($gcc_cv_as_hidden)
> +
>  AC_MSG_CHECKING(assembler instructions)
>  gcc_cv_as_instructions=
>  if test x$gcc_cv_as != x; then
> --- gcc-2.95.x/config.in.jj     Mon Oct 25 10:02:08 1999
> +++ gcc-2.95.x/config.in        Wed Jun 13 12:45:56 2001
> @@ -34,6 +34,9 @@
>     emitting at the beginning of your section */
>  #undef HAVE_GAS_SUBSECTION_ORDERING
>
> +/* Define if your assembler supports .hidden. */
> +#undef HAVE_GAS_HIDDEN
> +
>  /* Define if your assembler uses the old HImode fild and fist notation.  */
>  #undef HAVE_GAS_FILDS_FISTS
>
>
>         Jakub