Re: [PATCH] Revert HJL's G++ generates local references to linkonce

[Mark Mitchell <mark at codesourcery dot com>]

Andrew Pinski wrote:

> > > I figured out a better way of fixing this regression besides reverting the patch.
> > > See <http://gcc.gnu.org/ml/gcc-patches/2004-07/msg00878.html>.
> > >
> > > Here is the updated patch which has been bootstrapped on powerpc-apple-darwin and fixes these
> > > two regressions and also should fix the regressions on the other targets too unlike my previous patch
> > > in the message above.
> > >
> > >      
> > It seems to me like we ought to be able to determine  the linkage of a 
> > symbol before we start fiddling with it, now that we do compilation on a 
> > unit-at-a-time basis.  I can't find HJ's original test case anymore; 
> > would you point me at that?
> >    
>
> He did not post the testcase until after posting the patch and getting the ok:
> http://gcc.gnu.org/ml/gcc-patches/2004-07/msg00637.html
>  
Thanks.

I don't see exactly what is wrong there, but the typeinfo variable 
should not be put into COMDAT in the first place, in that case.  (It's a 
pointer to an incomplete type, so special rules apply to the typeinfo 
object; it should have internal linkage.)

> > And, in the mean time, would you like to revert HJ's patch until we 
> > figure out what we really want to do?
> >    
>
> Ok, I reverted the patch for now.
>  
Thanks!

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