Re: [RFC PATCH] Don't use ELF weak for entities with vague linkagewhen COMDAT groups are available

[Mark Mitchell <mark at codesourcery dot com>]

Daniel Jacobowitz wrote:
> On Thu, Apr 14, 2005 at 01:27:57PM -0700, Mark Mitchell wrote:
>
> > Julian Brown wrote:
> >
> >
> > > I have tried the building a gcc on x86-64 with the vanilla binutils from 
> > > SuSE 9.2 built from source, and it doesn't exhibit this problem. With 
> > > SuSE patches applied, the build fails for for me (on a non-SuSE system). 
> > > So in this case, the problem would appear to be one of the SuSE patches.
> >
> > It sounds like we need an autoconf test.  That's going to be a bit of a 
> > pain, in that I guess we need to both assemble and link something.
> >
> > We could do:
> >
> >  as -o conftest.o conftest.s
> >  ld -r -o /dev/null conftest.o conftest.o
> >
> > to make sure that COMDAT works.  But, the linker warnings Diego is 
> > seeing probably wouldn't show up, as that only happens when symbols have 
> > different sizes in the two object files.
>
>
> A multi-file autoconf test is not particularly hard to arrange.  I
> think we should make a solid effort to have this work by default...

I just don't know how to write assembly code that has the right 
complexities (and therefore exposes the bugs in the tools people have), 
but remains portable.  If you do, though, then that would be terrific. 
Maybe you could give Julian a hint, and he could work through the actual 
autoconf-y bits?

--
Mark Mitchell
CodeSourcery, LLC
mark@codesourcery.com
(916) 791-8304