[Piuparts-devel] Bug#850917: Bug#850917: Bug#854317: piuparts: adequate reports obsolete conffile for piuparts

[Simon McVittie]

On Sun, 26 Feb 2017 at 15:53:19 +0000, Holger Levsen wrote:
> somewhat contrary to what I've said before, I've now merged your patch into
> the develop branch, aimed for 0.76 and stretch, as I realized it's a no-op
> in all current and default configurations. As such, I've also already deployed
> it on piuparts.debian.org.

This seems to have caused piuparts regressions for what I believe are
otherwise correct packages:
https://piuparts.debian.org/sid/fail/libgtk-3-doc_3.22.9-3.log
Since the release team use piuparts regressions as a way to gate packages
entering testing, this is likely to be creating extra work for the
release team and for maintainers/NMUers who need changes unblocked.

I can't see anything in that log that looks like a failure, other than:

0m13.6s DUMP:
  Removing 'local diversion of /usr/bin/update-alternatives to /usr/bin/update-alternatives.orig'
  cat: /tmp/log-alternatives.log: No such file or directory
  rm: cannot remove '/tmp/log-alternatives.log': No such file or directory
0m13.6s ERROR: Command failed (status=1): ['chroot', '/srv/piuparts.debian.org/tmp/tmpMU3r5N', 'tmp/scripts/post_install_log_alternatives']
  Removing 'local diversion of /usr/bin/update-alternatives to /usr/bin/update-alternatives.orig'
  cat: /tmp/log-alternatives.log: No such file or directory
  rm: cannot remove '/tmp/log-alternatives.log': No such file or directory

I assume this is because libgtk-3-doc is such a simple package that it
doesn't depend on anything that would cause any alternatives to be
updated, so the log was never created.

To be robust against the log either never being created, or being cleaned up
by tmpreaper or similar, it would seem better to do:

-cat /tmp/log-alternatives.log
-rm /tmp/log-alternatives.log
+cat /tmp/log-alternatives.log || :
+rm -f /tmp/log-alternatives.log

The scripts added by
https://github.com/stapelberg/piuparts/commit/4867f7251189bd9f48fc229804a6ae44623546e4
also don't use set -e, unless that was added later. They probably should.

    S