PR 8146

[Greg Schafer]

On Tue, Nov 05, 2002 at 05:38:57PM -0800, Zack Weinberg wrote:
> On Tue, Nov 05, 2002 at 05:17:06PM -0800, Joe Buck wrote:
> > HJ writes:
> > > I am using 2 patches:
> > > 
> > > http://gcc.gnu.org/ml/gcc-patches/2002-08/msg01401.html
> > 
> > Are you using those two patches, but otherwise using the current
> > gcc 3.2.1 branch?
> > 
> > If you can show that:
> > 
> > a) gcc 3.2.1-pre as of now miscompiles gcc 2.95.3;
> > b) you have a patch that fixes it;
> > c) the patch doesn't cause more regressions
> > 
> > I would push hard to include it, provided that other platforms aren't
> > broken.
> 
> A small selfcontained test case miscompiled without the patch and not
> with the patch would also be a good idea.  The earlier discussion
> references c-torture/execute/2002307-1.c (now known as
> execute/20020412-1.c) which is fairly complicated and it's not obvious
> to me what about it provokes the bug you see.

GNATS PR 7591 contains a small test case. The problem was fixed on the
mainline by rth. See here:-

http://gcc.gnu.org/ml/gcc-patches/2002-08/msg01715.html

That patch is not easily back-portable to the 3.2 branch as it depends on
some earlier cleanups.


HJ's patch was not agreeable to rth:-

http://gcc.gnu.org/ml/gcc/2002-08/msg01921.html


But in my experience, HJ's patch fixes PR 7591 and 2002307-1.c on the branch
for -march=i686 (and it looks like it fixes PR 8404 too). At least 2 "distros"
I know of are using HJ's patch (Gentoo and LinuxFromScratch) so some wider
testing has been achieved. And I can confirm no testsuite regressions.

None of this has likely anything to do with subject PR 8146 but it would
be nice if HJ's patch was applied on the branch to stop the blatant miscompiles.

Greg