This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [PATCH] Fix {sparc64,x86_64}-linux ABI problem (take 2)


On Tue, May 07, 2002 at 06:41:34PM -0400, Jakub Jelinek wrote:
> > 	cris-linux	GLIBC_2.2
> > 	sh-linux	GLIBC_2.2
> > 	hppa-*		GLIBC_2.2
> 
> None of these arches actually reexport any of the libgcc symbols from glibc,
> so they should probably just use GCC_3.0 symver for everything.

cris didn't exist in 3.0, so that one's a no-brainer.  sh-linux and
pa-linux did exist, but since they didn't export division routines,
that puts them in the same boat as ia64.

> For IA-64 (where __divdi3, __moddi3, __udivdi3, __umoddi3 are
> exported at GLIBC_2.0 while glibc has no such version), ...

Hum.  The difference between these symbols and __register_frame_info
and friends is that glibc never exported __divdi3 etc on ia64.  So
we're more likely to work properly as-is.

Do we know of a positive failure on ia64 wrt division routines?

Another thought:  What do we do with losers who configure glibc
with --enable-oldest-abi=2.3?  It'll all break again.  I guess
we have to trust that no Real Distribution is going to be that
stupid.

ARG!  I hate this.

How hard would it be to write a script that generated a link
script for glibc compatibility by looking at the installed...
No, I guess that's pretty hard for multilibs.

I guess the only permanent solution is to coordinate with glibc
and move these (default) symbols out of GLIBC_* into GCC_*, and
install the backward compatibility symbols you suggested.  Ug.


r~


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]