Bug#1117830: release.debian.org: Help postgis migrate to testing

[Christoph Berg <myon@debian.org>]

Re: Paul Gevers
> The point is that I'm also not smart enough to see why britney isn't smart
> enough in this case. I'd like to understand things first before I paper over
> them.

I think I found the problem. pg_buildext (in postgresql-common-dev)
has the problem that in some contexts it has to cover the list of PG
versions supported now (build time) and sometimes the list of PG
versions supported by the current system (test time), and one of the
"wrap this command" modes got that wrong, namely "pg_buildext
run_installed". That explains the "17 missing" and "18 missing"
messages in the mobility test logs.

The fix is there: https://salsa.debian.org/postgresql/postgresql-common/-/commit/56e3f85064e84e5e474ffc4f65b5854c60e5217c , uploaded as postgresql-common 286.

The bad news is that this fix needs to propagate to forky first before
it's used for the mobility and postgis tests. I don't think adding any
"Breaks" would make sense as there's no actual breakage, it's just the
tests being run in the wrong way.

I would propose that we just wait a day until all the extensions that
are currently 1-of-2 days old propagate to forky and then look at the
remaining packages to see if additional fixes are required. At that
point, forcing postgresql-common would likely fix it (or trying
postgresql-common+mobilitydb+postgis together).

Christoph