[buildd-tools-devel] Bug#590911: Bug#590911: sbuild: broken return code

[Wouter Verhelst]

Hi Josch,

On Thu, Dec 24, 2015 at 01:06:32AM +0100, Johannes Schauer wrote:
> On Sat, 20 Nov 2010 17:10:11 +0000 Roger Leigh <rleigh at codelibre.net> wrote:
> > On Thu, Jul 29, 2010 at 10:25:22PM -0400, Wouter Verhelst wrote:
> > > cyrus-sasl2-heimdal_2.1.23.dfsg1-5.1 failed on voltaire. Sbuild sent me
> > > a mail with 'Log for attempted build' as the subject, but daemon.log
> > > contains this:
> > > 
> > > Jul 29 12:49:09 buildd[2705]: sbuild of cyrus-sasl2-heimdal_2.1.23.dfsg1-5.1 failed with status 1 (local problem) -- giving back
> > > 
> > > ... which is clearly wrong.
> > > 
> > > I've been thinking that, as sbuild needs to interpret exit states which
> > > aren't exactly reliable, it might be better to have it communicate over
> > > stdout instead?
> > > 
> > > Anyway, regardless, this shouldn't happen :)
> > 
> > Just to keep this bug report in the loop, we have a patch to implement
> > just this, which is currently a work in progress (attached).  I've
> > been discussing this with Philipp Kern (CCd).
> > 
> > With the patch the status of each job is passed back on stdout to
> > buildd as RFC-822-style fields, together with statistics for the
> > build and other metadata about the job.  This removes all ambiguity
> > about what the failure was (and what caused it).
> > 
> > The patch should apply to the current master branch in git.
> > 
> > If you have any comments or suggestions, that would be great.
> 
> what should be done about this five year old bug? Is the original issue still a
> problem?  Does the patch by Roger fix it? If the answer is yes, I shall rebase
> it onto current sbuild master.

The original issue was never a real problem, it was more of a "this
seems incorrect" kind of thing. The resulting behaviour by buildd when
sbuild exits with nonzero exit state was never something that would
break builds or whole architectures or anything of the sorts.

I guess that patch will probably fix it, and applying it would seem to
result in more robust signalling between sbuild and buildd.

Whether it's a good idea isn't something I think should be my call,
however :-)

-- 
It is easy to love a country that is famous for chocolate and beer

  -- Barack Obama, speaking in Brussels, Belgium, 2014-03-26