This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
RE: [PATCH] Fix MIPS LHX in mips-dsp.md
- From: "Fu, Chao-Ying" <fu at mips dot com>
- To: "Maciej W. Rozycki" <macro at linux-mips dot org>, "David Daney" <ddaney at caviumnetworks dot com>
- Cc: <gcc-patches at gcc dot gnu dot org>, "Lau, David" <davidlau at mips dot com>, <rdsandiford at googlemail dot com>
- Date: Thu, 12 Nov 2009 15:13:08 -0800
- Subject: RE: [PATCH] Fix MIPS LHX in mips-dsp.md
Maciej W. Rozycki wrote:
>
> On Thu, 12 Nov 2009, David Daney wrote:
>
> > > There's a general GCC rule that code patches must be
> regression-tested
> > > against the full GCC testsuite (or at least all the parts
> associated
> > > with the languages that your target supports by default).
> The idea
> > > is to make sure that no existing tests start to fail
> after the patch.
> > > If you're confident they won't (and I agree they're unlikely to in
> > > this case), well, all the better! It should just be a formality.
> >
> > Several times I have created seemingly trivial patches
> where I was certain
> > that there would be no regressions. However, the step of
> running the
> > testsuite for regressions uncovered some problems. So I
> wouldn't classify it
> > a a busy-work type formality, it is an important tool to
> help us improve
> > quality.
>
> And actually even if a change is absolutely correct, the way
> it modifies
> behaviour of a piece of software may uncover a bug elsewhere
> that needs to
> be fixed (not necessarily by the submitter of the original fix, of
> course).
>
> Maciej
>
All your points are great.
My pain is that this bug isn't caught until now.
Should we add more MIPS target configurations to test? ex: -mdsp/-mdspr2
Passing all GCC regression tests for limited targets is still not enough.
And we should have big tests/applications to exercise the compiler.
Thanks!
Regards,
Chao-ying