Re: [4.1] Patch ping

[Roger Sayle <roger at eyesopen dot com>]

On Wed, 3 May 2006, Jakub Jelinek wrote:
> PR tree-optimization/27285
>
> http://gcc.gnu.org/ml/gcc-patches/2006-04/msg00961.html
>
> Backport that fixes a regression from 4.1.0 on gcc-4_1-branch.
> Bootstrapped/regtested several times on 7 linux platforms.

Given the seriousness of this 4.1 regression and the level of
testing of your patch, this is OK for the gcc-4_1-branch.  The
new testcase is also OK for mainline.

Please keep a watch for possible issues.  The tree-ssa loop
ivopts code is volatile, and although this patch has been
tested on mainline, that's been in the presence of other
changes.  Much like Paul's backport for PR18527 has required
the backport of this change, it'd be good to know whether
this change will require any further follow-ups/fallout.

Zdenek, do you know of any potential issues with these backports?


Does anyone know how easy it is to test the 4.1 branch against
mainline's testsuite?  Checking for new regessions against
mainline's testsuite, would at least warn that backporting a
patch doesn't introduce a known regression to a release branch
that has already been fixed on mainline.

It might also make sense to have some mechanism for maintaining/evaluating
testcases for newly identified 4.2 regressions, so that we can avoid
backporting the patches that caused them to release branches.  This
goes against our current policy of not adding a testcase to the testsuite
until it's been resolved.  But maybe it would be useful for maintainers/
developers to have access to a set of tests, that weren't part of the
default "make check", i.e. that wouldn't be visible to distribution
end-users who might not understand the meaning of large numbers of
"pending" failures, PFAILs?

Thoughts?

Roger
--