PostGIS 3.6.0beta1 released
Regina Obe
lr at pcorp.us
Mon Jul 21 19:13:00 PDT 2025
> >> gdt=> NOTICE: Updating extension postgis 3.4.3rc1
> >> gdt-> ERROR: extension "postgis" has no update path from version
> >> "3.4.3rc1" to version "ANY"
> >
> > We remove alphas, betas, rcs on final release cause they take up space
> > and no one would be upgrading a production system from those.
>
> I did, but ok that's a fair point.
>
> > Is this as a super user? I recall we allow super users to upgrade
> > regardless if there is an extension path cause you can always upgrade
> > from ANY and we just update the system catalogs directly to ANY if you
> > are a super user.
> >
> > @strk is that true or did we take that logic out?
>
> No, I decided to make myself not a super user.
>
> I can see it needs some kind of --force, but gating on super user seems wrong.
> That's about permissions, and this is about "do you really want
> to whack it, or do you want to think". Being root shoudln't get you
> out of consistency checks.
>
In this case I just consider it a deficiency in the postgres system that we have to ship stupid
0-byte scripts to change a system catalog. Honestly it is the only way to fix upgrade that is running
with missing 0-byte files. I hardly think this is something that should require root in the first place.
But don't get me started about this. Until strk's patch or some remnant of it is supported upstream, this is easier than
Explaining to folks running rcs and betas how to upgrade their systems.
Those who run rcs and betas and alphas are probably super users anyway.
This path of updating the system catalogs is never used on super user systems if the 0-byte files exist.
> >> Then tests get pretty far and error out on permissions to set
> >> lc_message. Do I need to run tests as a pgsql superuser, or the owner
> >> of the postgis extension?
> >>
> >> (I have recently tightened up pgsql permissions substantially:
> >> - I'm no longer a superuser.
> >> - pgsql listens only on unix-domain socket.
> >> - pgsql has a uid map to pgsql users
> >> )
> >>
> >> If the tests need privs, it would be nice to check and give a better
> >> message. I realize this is likely not new in 3.6 :-)
> >
> > Sounds like you are running into this issue
> >
> > https://trac.osgeo.org/postgis/ticket/5212
> >
> > I think the workaround is to set environment variable:
> > POSTGIS_REGRESS_DB_OWNER
>
> > As we do in our tests which also run sandboxed
> >
> > https://git.osgeo.org/gitea/postgis/postgis/src/branch/master/.woodpec
> > ker/regress.yml
>
> My take is:
>
> one should be able to test without being superuser, if one can install
> postgis in a db without being superuser
>
> if there are extra tests one can do with superuser, then they should
> be skipped if not, reporting pass/fail/skip like most test harnesses.
>
> if there is a way to tell the tests to make them work, then make test
> should just do that. putting test magic in some other place where a
> user that just runs tests per the docs is a workaround in the wrong
> place.
>
> > Yah agree we should outline this in the docs. This was a change we had
> > to make because of PG15 changes.
> >
> > Ticketed here - https://trac.osgeo.org/postgis/ticket/5950
>
> I don't understand why this can't just be part of "make test".
>
All tests can be run as non-superuser and we do that in the woodie runs.
I don't think @Sandro Santilli could make it work without adding that extra variable
And it's more important for us to test all tests than skip tests cause we expect all tests to work as non-superuser
so we know it works in a sandboxed environment.
The reason the whole locale comes up is because PG15 the output of some tests changed if you don't set the locale to c.
If you have a better way to test this without introducing a PostGIS _REGRESS_DB_OWNER, feel free to state it.
> If it really can't, it should be explained in the README where users are told to
> just run tests
>
>
> (Thanks for all your replies and tickets; now I get what's going on sort of, and
> am confident it isn't my system being odd.)
>
> Greg
Yah we can add to the README. I didn't realize people read READMEs 😊
More information about the postgis-devel
mailing list