[pkg-gnupg-maint] Bug#868550: Bug#868550: reprepro seems to provide a repro

[Ian Jackson]

Hi, Sean, and anyone else reading dgit at p.d.o.

I thought I should give some information about how to run the dgit
test suite with the in-tree makefile-based parallelism without
becoming blocked by the gnupg race bugs.

Firstly, dgit developers will see that chiark/master and sid (4.2)
contain some more workarounds which seem to me to help.

Secondly, I have been using a modified version of gnupg2.  My gnupg2
branch is here.  I was using ecbc72bafb3767aaa77b4a2c41ce52edadabfa85.
  https://www.chiark.greenend.org.uk/ucgi/~ian/git?p=gnupg2.git;a=shortlog;h=refs/heads/racetest-stretch
  git://git.chiark.greenend.org.uk/~ian/gnupg2.git
The main thing that seems to make a difference is to change the
timeouts.  I have left them at an intermediate value, which IME allows
the tests to proceed relatively soon (due to my test suite retry log)
even if the retry loop within gpg itself is doomed.

Although I haven't done any formal testing of the effects of each
individual change I found that with all of these changes together
   tests/using-intree tests/run-all
(in the dgit tree) worked (?almost) every time.  Certainly I was able
to test each commit in several patch series without it bombing out, so
the overall failure rate of the whole test suite is tolerably low.

I think it is still necessary to occasionally kill gpg-agent.

Running the tests using adt-run and adt-virt-schroot seems to work for
me too.  Of course that is really slow (and certainly not practical
for per-commit testing).

I hope this information is useful.

Regards,
Ian.

-- 
Ian Jackson <ijackson at chiark.greenend.org.uk>   These opinions are my own.

If I emailed you from an address @fyvzl.net or @evade.org.uk, that is
a private address which bypasses my fierce spamfilter.