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

On Wed, Dec 15, 2021 at 10:47:24AM +0900, Michael Paquier wrote:
> Missed that, thanks!  I'll think about all that a bit more before
> sending a long-overdue rebased version.

Okay, here is finally a rebase of this patch, where I have fixed a
couple of existing issues, and I have extended the patch to the point
where the support range is what I expect should be.  In short:
- Added support for MSVC for the TAP test.  I have considered making
upgradecheck silent, but after thinking about it I have just filtered
pg_upgrade from bincheck, and simplified upgradecheck to launch the
new test.  It is simple to switch from one approach to another.  This
shaves some code in vcregress.pl.
- Fixed a set of issues with various chdir commands in the previous
patches.  The command of pg_regress has been tweaked so as all results
are part of src/bin/pg_upgrade/.  Any logs generated by pg_upgrade
stay in this location, same way as HEAD.
- Adapted to the new modules of src/test/perl/.
- Support for cross-upgrades now uses upgrade_adapt.sql (I have sent a
patch for the buildfarm client about that yesterday actually), same
way as test.sh on HEAD.  Like HEAD, attempting to use the
cross-version HEAD causes diffs between the old and the new dumps.
But there is nothing new here.  This could be improved more but the
attached does already a lot.
- Like the previous versions, this supports two modes when setting up
the to-be-upgraded cluster: setup things from an old dump or use
pg_regress.  The buildfarm does the former for upgrades down to 9.2.
The core code does the latter.

I may have missed one thing or two, but I think that's pretty much
what we should be looking for to do the switch to TAP in terms of
coverage.
--
Michael