This is the mail archive of the
libc-alpha@sources.redhat.com
mailing list for the glibc project.
Re: [PATCH] PPC libgcc compatibility
At 11:38 15.05.2002, Jakub Jelinek wrote:
>On Wed, May 15, 2002 at 10:33:19AM +0200, Franz Sirl wrote:
> > At 06:29 15.05.2002, Ulrich Drepper wrote:
> > >On Sun, 2002-05-12 at 14:07, Franz Sirl wrote:
> > >
> > > > the appended patch implements most of the libgcc routines that may
> cause
> > > > broken binaries if a gcc-3.1 compiled glibc is installed. The symbols
> > > fixed
> > > > are (copied from gcc/libgcc2.c):
> > >
> > >The patch does not correspond to what is done for IA-64. The functions
> > >should not be available for linking. The madness must stop. I've made
> > >some changes to your code which are completely untested. Give the CVS
> > >trunk a try.
> >
> > If it's not linkable, that's fine with me, actually I even prefer that.
> > It's just that I started with divdi3.c as a template, which doesn't have
> > this symbol_version() stuff in it, should it?
>
>No, divdi3.c should not IMHO, since it is used by arches where this is a part
>of exported ABI (unlike IA-64 or PPC it has been that way from the
>beginning). I think PPC should use a .c wrapper around
>it and add symbol_version().
Ah, OK. Didn't know that.
>Some more notes about the patch: you include
>longlong.h and all its machinery everywhere, while none of the routines
>does actually use it. I think it would be way better to just use *int*_t
>types everywhere and get rid of the longlong.h stuff which is not needed
>at all. Also, when you're calling say in fixdfdi.c __fixunsdfdi, you should
>do so through INTUSE() and add __fixunsdfdi_internal prototype and
>INTDEF().
I'll check that out.
>Last the comments on top of the files were not changed from divdi3.
Uli fixed that on-the-fly :-).
Dumb question, it just occurred to me that there might be an even simpler
solution here, why don't we just create a
sysdeps/<platform>/libgcc-compat.c that pulls in all needed symbols from
libgcc.a?
Like for example:
#include <stdint.h>
#include <shlib-compat.h>
#if SHLIB_COMPAT(libc, GLIBC_2_0, GLIBC_2_2_6)
extern int64_t __fixdfdi (double);
int64_t INTUSE (__fixdfdi) (double d)
{
return __fixdfdi (d);
}
symbol_version (INTUSE (__fixdfdi), __fixdfdi, GLIBC_2.0);
extern int64_t __fixunsdfdi (double);
int64_t INTUSE (__fixunsdfdi) (double d)
{
return __fixunsdfdi (d);
}
symbol_version (INTUSE (__fixunsdfdi), __fixunsdfdi, GLIBC_2.0);
#endif
After all this is backwards compatibility code, so performance is no issue.
Franz.