AIX regressions

[Mark Mitchell]

> 	Unlike the other failures, this is a second-order failure of
> gcc-3.1 mis-compiling itself.  gcc-3.1 compiled by gcc-3.0 works
> correctly, but gcc-3.1 compiled by gcc-3.1 after Richard's change now has
> an internal compiler error for this testcase.  Richard's patch does not
> seem to be broken, but it will be hard to debug.

Yes.  This seems to be the same debate we had a while back; this is
probably a "latent bug".  Someone will have to do the effort to fix
the problem.  Our policy says that we can ask Richard to do that work
and revert the patch if he does not.  Of course, as the AIX maintainer,
you could also try to solve the problem yourself.

You can play the usual binary search games with .o's to try to find
which .o in the new compiler has been miscompiled.

--
Mark Mitchell                   mark@codesourcery.com
CodeSourcery, LLC               http://www.codesourcery.com