On Sat, May 06, 2017 at 03:44:42PM +0200, Michael Stapelberg wrote: > Tomasz Buchert <tom...@debian.org> writes: > > On 04/05/17 04:47, Adam Borowski wrote: > >> [...] > > > > I cannot reproduce these failures. I've built in my stretch sbuild > > around 15 times, and succedeed every time.
I found just one of my machines that doesn't reproduce the failure (arm64, also in armhf chroot). It succeeds 100% of the time, unlike the rest which fail 100% -- no randomness that's mentioned in this report. > > I use: > > gbp buildpackage --git-builder='sbuild --source-only-changes -v -As > > --build-dep-resolver=apt --dist=stretch -j4' "$@" > > The issue is reproducible as soon as something starts listening on port > 9002 (e.g. nc -l -p 9002). Hmm... that seems kind of strange, as my setup on most machines is identical except for host[non-autotool meaning] architecture. The sbuild config is identical, so are per-chroot settings, and I tried even physically the very same chroot (nbd from a third server). Listening on a port sounds like something not host arch independent. > I pushed a commit adding a patch which fixes the test by picking an > unused port. The mechanism is not atomic (i.e., the port is picked by > the test process, as opposed to the listening process itself and > communicated back to the test process), but it’s an improvement and > should fix the issue at hand. Lemme try your patch. -- Don't be racist. White, amber or black, all beers should be judged based solely on their merits. Heck, even if occasionally a cider applies for a beer's job, why not? On the other hand, corpo lager is not a race.