[llvm-commits] [cfe-commits] Historical ARM-linux regression test failures are now fixed

[David Blaikie]

On Fri, Oct 26, 2012 at 6:12 AM, David Tweed <david.tweed at arm.com> wrote:
> Hi all,
>
> Just to point out something that some LLVM/clang contributors may not have
> noticed. Historically the regression tests "on" ARM (and hence the
> buildbots) have had quite a few longstanding failures, which made seeing any
> new regressions difficult. Over the past couple of months various people --
> including Wei-Ren Chen, Joey Gouly, Amara Emerson, Eli Friedman and probably
> others whose names I've unfortunately forgotten -- have been making patches
> to fix those issues properly, and all the historical regression failures are
> now gone on the current ubuntu gnueabihf platfrom. (We're not green just now
> due to a couple of failing tests introduced just yesterday.)
>
> I'm just highlighting this to suggest that it's now worth people's time
> looking at ARM failures (eg, looking at the output of the ARM buildbots
>
> http://lab.llvm.org:8011/builders/llvm-arm-linux
> http://lab.llvm.org:8011/builders/clang-native-arm-cortex-a9
>
> listed on
>
> http://lab.llvm.org:8011/builders/
>
> ) after patches have been committed in order to keep LLVM's quality high.
> (For historical bugs it's often the case failures that "on" ARM are actually
> due to using ARM parameters/definitons and can be recreated on x86 with the
> appropriate triple, etc.) I'll be keeping half an eye on their status and
> notify people when I notice things, but as the saying goes "with more
> eyeballs..."

Thanks for all your work, David (& others).

Galina - is there any way we can classify the "good" bots that have
current maintainers & are reliably green (and/or where failures are
reliably actionable, etc) as important & have the buildmaster send
email for those failures (which builders currently send email on
failure? I certainly see some failure mail - not sure if that's from
the lab or from smooshlab)?

If so, could we get these builders classified as "good". (It might
also be helpful to move the current "bad" builders off to the side in
the dashboard, or just remove them entirely - just as a machine with
regular failures makes it hard to see the real ones, a dashboard with
half the builders in a constant red state makes it hard to tell which
builders are important)

- David