[pkg-gnupg-maint] Bug#902316: gnupg failing completely in dgit autopkgtests

[Ian Jackson]

Package: gnupg
Version: 2.2.8-3

(See also #868550 which discusses a race with a much lower occurrence
probability and and different symptoms: the race in #868550 can be
worked around by simply retrying the failed gnupg2 invocation, whereas
the race I am reported in this bug apparent cannot.)

Paul Gevers writes ("dgit in unstable"):
> Just spotted the state of dgit in autopkgtest on unstable. I assume that
> you are aware. (No regression spotted yet in testing).

This is almost certianly a bug in gnupg2.  It works with my up to date
sid chroot when run with stretch's adt-run on dgit_5.1.dsc. [1]

So I think it must be a race or some other stochastic bug or maybe
some weird kernel interaction (my testbed is an schroot and I'm
running stretch's amd64 kernel).

ci.debian.net is losing this race every time in unstable but not for
testing.  None of the package version changes look relevant (see
below).

As I write in #868550 I have sometimes seen similar symptoms, which
are generally resolved by saying `killall gpg-agent'.


gnupg2 maintainers: What should we do to debug this ?

It is very disappointing that I first reported problems of this kind
in October 2016 in #841143, and although that was closed in January
2017, I found in July 2017 that there were still problems and opened
#868550.  That was a year ago.  If gnupg2 cannot be made to work
reliably, can we switch back to gnupg1 as default ?


Paul, is it possible that the test runner somehow has a leftover
gpg-agent ?  What, if anything, is shared with previous test runs ?

Incidentally, I don't understand why it has suddenly started failing
much more often with so little change to any of the packages.

I looked at this
  https://ci.debian.net/data/packages/unstable/amd64/d/dgit/495400.log
which is the debci log of the first failing run in unstable and the
updates are:

-linux-libc-dev 4.16.16-1
+linux-libc-dev 4.16.16-2
-linux-libc-dev 4.16.16-1
+linux-libc-dev 4.16.16-2
-binutils       2.30-21
-binutils-common        2.30-21
-binutils-x86-64-linux-gnu      2.30-21
+binutils       2.30-22
+binutils-common        2.30-22
+binutils-x86-64-linux-gnu      2.30-22
-libbinutils    2.30-21
+libbinutils    2.30-22
-libpam-systemd 238-5
+libpam-systemd 239-1
-libsystemd0    238-5
+libsystemd0    239-1
-libudev1       238-5
+libudev1       239-1
-systemd        238-5
-systemd-sysv   238-5
+systemd        239-1
+systemd-sysv   239-1

Has something else changed ?

Ian.

[1] My complete adt-run has not yet finished but it's obviously not
behaving like the one on ci.debian.net.

-- 
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.