Re: Buildbot failure in Wildebeest Builder on whole buildset

[Mark Wielaard <mark@klomp.org>]

Hi Thomas,

On Thu, 2022-02-03 at 21:55 +0100, Thomas Schwinge wrote:
> On 2022-01-29T21:20:45+0100, Mark Wielaard <mark@klomp.org> wrote:
> > On Tue, Jan 25, 2022 at 11:42:41PM +0100, Mark Wielaard wrote:
> > > I added a filesIsImportant filter to the buildbot gccrs
> > > scheduler:
> > > 
> > > gccrs_files = ["gcc/rust/", "gcc/testsuite/rust/",
> > > "gcc/config/.*/*-rust.c"]
> 
> Is that last one correct, or should that be 'gcc/config/*/*-rust.c'
> (glob) or 'gcc/config/.*/.*-rust\.c' (regexp)?
> 
> After my recent GCC/Rust commit
> 5691503f11fb6bc5acd8be1e43faa0c5898c4b14
> 'GCC/Rust pieces of GCC upstream "Mass rename of C++ .c files to .cc
> suffix"',
> please change the last one to '.cc' suffix.

Well spotted. Thanks. Changed it to:

gccrs_files = ["gcc/rust/", "gcc/testsuite/rust/", "gcc/config/.*/.*-rust.cc"]

I really should clean the buildbot cfg file up (it contains everything,
including passwords and tokens in one file) and check it into git so
others can more easily contribute to it.

> Maybe include 'gcc/DATESTAMP' in 'gccrs_files', so that we also build
> after merges from upstream (assuming that they span more than a
> day/do change that file)?

That won't work. At least not with the current setup/filter. The
buildbot doesn't know what the correct "side" is for the commits that
are being merged. So the buildbot would see the commits containing
gcc/DATESTAMP and think those commits should be build. Even if those
commits come from the "side" where there are no gccrs files yet.

We could however add a daily or weekly builder/scheduler. Which would
be especially helpful when we add container image builders, where the
image itself might be updated once a week. But at the moment we seem to
have an average of at least one commit a day, so in practice we would
notice pretty soon after a merge.

Cheers,

Mark