Re: pgsql: Trial fix for old cross-version upgrades.

I wrote:
> ... I suspect that the problem is that
> pg_dump is interpreting old-version stats in some way that doesn't
> match up with what we get from restoring the dump.

Hmm. I forced my local BF installation to run a v17-to-HEAD upgrade
test, and it still failed, though seemingly with fewer diffs than
the buildfarm is reporting for older branches. (Diffs attached for
amusement's sake.) I don't believe we've made any definitional
changes in the contents of pg_statistic since v17, so whatever's
going on here seems a little subtler than I was hoping.

I wonder if it'd be a good idea to rearrange TestUpgradeXversion.pm
so that instead of testing upgrades from oldest prior version to
newest, it tested from newest to oldest? My thought here is that
the oldest cases are most likely to fail, and when they do, it'd
be valuable information to know which branches still work.

regards, tom lane

PS: this is with the no-autovacuum hack I posted before.