[Debichem-devel] Bug#1118669: gnome-chemistry-utils: autopkgtests intermittently time out, especially on s390x, and don't seem to really test the package

[Simon McVittie]

Source: gnome-chemistry-utils
Version: 0.14.17-6.3
Severity: serious
X-Debbugs-Cc: debian-ci at lists.debian.org
User: debian-s390 at lists.debian.org
Usertags: s390x
User: debian-ci at lists.debian.org
Usertags: flaky

The autopkgtests for src:gnome-chemistry-utils usually take seconds, but 
sometimes one of them blocks until the ~ 3 hour timeout is hit. For 
whatever reason this seems to be more common on s390x.

When this happens, it blocks migrations to testing: for example, at the 
moment it's preventing glib2.0 from migrating.

Having looked into this, I'm not at all sure that these tests are 
actually doing anything useful.  The test command for each one is 
something like:

    xvfb-run -a /usr/bin/gchemcalc &

but if gchemcalc crashed (perhaps a segfault or a missing library 
dependency), would that test actually even fail?

As an experiment I tried changing the test to:

Test-Command: xvfb-run -a sh -c 'echo Pretending to crash >&2; kill -ABRT $$' &
Depends: xauth, xvfb, gcu-bin
Features: test-name=run-sigabrt
Restrictions: allow-stderr

and it trivially "passes":

>autopkgtest [19:05:04]: test run-sigabrt: xvfb-run -a sh -c 'echo Pretending to crash >&2; kill -ABRT $$' &
>autopkgtest [19:05:04]: test run-sigabrt: [-----------------------
>/tmp/autopkgtest-lxc.nigxlbus/downtmp/wrapper.sh: Killing leaked background processes: 2205
>    PID TTY      STAT   TIME COMMAND
>   2205 ?        S      0:00 /bin/sh /usr/bin/xvfb-run -a sh -c echo Pretending to crash >&2; kill -ABRT $$
>autopkgtest [19:05:04]: test run-sigabrt: -----------------------]
>autopkgtest [19:05:04]: test run-sigabrt:  - - - - - - - - - - results - - - - - - - - - -
>run-sigabrt          PASS
>autopkgtest [19:05:04]: @@@@@@@@@@@@@@@@@@@@ summary
>run-sigabrt          PASS

which I think means these tests do not meet 
https://release.debian.org/testing/rc_policy.txt §6a.

I think the timeouts might have more to do with autopkgtest not always 
being able to kill the leaked background xvfb-run process; possibly 
xvfb-run is blocking signals?

    smcv