Re: [PATCH] Avoid creating TREE_OVERFLOW constants in convert_to_integer (PR middle-end/26729)

[Jakub Jelinek <jakub at redhat dot com>]

On Thu, Apr 20, 2006 at 08:47:25AM +0200, Eric Botcazou wrote:
> > If you want to say "look, i've tested it against our internal tests,
> > and it causes a regression there, here's a testcase", that's fine.
> 
> I've exactly done that for Kazu's patch.  No reaction.  I've suggested a 
> possible fix.  No reaction.  I've posted a fix.  No reaction.
> 
> > But we shouldn't be blocking people from putting things on release
> > branches that fix bugs if they've gone through the required testing.
> > At least, that's not the policy we have documented.
> 
> Sure, but I claim that the policy has already been violated for Kazu's patch 
> so I now consider that I've no other solution than violating it myself. :-)

This is a P1 regression on many primary platforms, so I'd say that (of
course, if approved) the patch shouldn't be held because it might cause
regressions on some uncontributed Ada tests.

That said, I can't test the patch reliably with Ada at all, since I'm
consistently getting about 250 FAILs in acats on all platforms I'm building
Ada on (e.g. i686, x86-64, ppc, ia64), but the exact failed tests differ
greatly each time, so I really can't trust Ada regression testing at all.

	Jakub