libgcc-compat patch and gcc 2.95.4 on ppclinux

[Franz Sirl]

On Samstag, 13. Juli 2002 09:31, Jakub Jelinek wrote:
> On Fri, Jul 12, 2002 at 10:43:02PM -0700, H. J. Lu wrote:
> > Please try this patch. The problem is gcc 2.95.4 won't make those
> > libgcc functions hidden. We have to help it.
> >
> > --- libc/sysdeps/powerpc/libgcc-compat.c.gcc	Mon May 20 17:19:09 2002
> > +++ libc/sysdeps/powerpc/libgcc-compat.c	Fri Jul 12 22:39:18 2002
> > @@ -23,6 +23,7 @@
> >
> >  #if SHLIB_COMPAT(libc, GLIBC_2_0, GLIBC_2_2_6)
> >
> > +__asm__(".hidden __ashldi3");
> >  extern int64_t __ashldi3 (int64_t, int32_t);
> >  int64_t INTUSE (__ashldi3) (int64_t u, int32_t b)
> >  {
>
> Err, isn't this an exact opposite of what is desirable?
> This would make __ashldi3 in libgcc-compat.o hidden, while it has to
> be visible, so with gcc-3.1+ you'd end up
> with not exporting those functions at all, while with 2.95.x you'd
> be exporting the libgcc.a functions (yes, that's what you want in that
> case).
> IMHO libgcc-compat.c should be only compiled in if the functions are
> .hidden in libgcc.a (ie. gcc 3.1+), otherwise glibc just needs to ensure
> all the needed libgcc.a functions are included in the link.

I think so too. It would be nice if the version script would support a 
runtime: label too besides global: and local:. Or if ld would support 
something like --import-hidden -lgcc --no-import-hidden.

Jakub, one thing I don't understand yet about the symbols from divdi3.c in 
glibc, why are they exported with _linktime_ reference? Wouldn't a runtime 
reference be enough? It would possibly break some broken apps during 
building, but who cares?

And why are these symbols versioned GLIBC2.0 in GCC?

Franz.