Re: Rewriting the test of pg_upgrade as a TAP test - take three - remastered set

On Tue, May 18, 2021 at 10:49:39AM +0900, Michael Paquier wrote:
> Makes sense. For now, I'll update this patch set so as it is possible
> to use custom dumps, as an option in parallel of pg_regress when
> specifying a different source code path. I'll also decouple the
> business with probin updates and stick with the approach used by the
> buildfarm code.

This has proved to not be that difficult. With the updated version
attached, pg_upgrade has two modes to set up the old instance used for
the upgrade with older binaries:
- With oldsrc and oldinstall set, pg_regress gets used, same way as
HEAD.
- With olddump and oldinstall set, an old dump is loaded instead in
the old instance before launching the upgrade.

oldsrc and olddump are exclusive options. Similarly to HEAD, the
dumps taken from the old and new instances generate diffs that can be
inspected manually. The updates of probin are done without any
dependencies to the source path of the old instance, copying from the
buildfarm.

While on it, I have fixed a couple of things that exist in test.sh but
were not reflected in this new script:
- Handling of postfix operations with ~13 clusters.
- Handling oldstyle_length for ~9.6 clusters.
- Handling of EXTRA_REGRESS_OPT.

This stuff still needs to be expanded depending on how PostgresNode is
made backward-compatible, but I'll wait for that to happen before
going further down here. I have also spent some time testing all that
with MSVC, and the installation paths used for pg_regress&co make the
script a tad more confusing, so I have dropped this part for now.

Thanks,
--
Michael