On Sat, Oct 02, 2021 at 11:34:38PM -0400, Tom Lane wrote:
> Maybe we could leave test.sh in place for awhile? I'd rather
> not cause a flag day for buildfarm owners. (Also, how do we
> see this working in the back branches?)
I would be fine with test.sh staying around for now.
If we do that, though, I think that we had better remove the support
for upgrades across different major versions in test.sh, and keep this
capability in the new script. I am not sure that a lot of people use
that to begin with, but it would be weird to support that with a
different configuration layer for both at the same time (test.sh uses
a combination of bin/ and lib/ paths, while TAP uses just installation
path to accomodate with what PostgresNode.pm is able to do). The
patch of this thread also adds support for the load of an old dump
instead of an installcheck run of the old instance, which is something
the buildfarm could use.
I also looked two days ago at a proposal to move all the
pg_upgrade-specific SQLs into a new, separate, file that makes use of
psql's \if to do the job encoded now in test.sh. I think that it
would be strange to duplicate this logic in a the pg_upgrade TAP test
and test.sh if we finish by keeping both around for now. So that's a
second item we had better deal with first, in my opinion:
https://www.postgresql.org/message-id/YVa/se5gxr1PsXDy@paquier.xyz
Thoughts?
--
Michael