Robert Haas <robertmhaas@gmail.com> writes:
> Yeah, I feel like it was a mistake to allow the list of permutations
> to be unspecified. It encourages people to just run them all, which is
> almost never a thoughtful decision. Maybe there's something to be said
> for running these tests in one successful permutation and one failing
> permutation -- or maybe even that is overkill -- but running them all
> seems like a poor idea.
Yeah, I considered letting the no-error permutation survive. But
I didn't really see what coverage it was adding at all, let alone
coverage that'd justify doubling the test runtime.
Also ... while doing further research I was reminded that a couple
years ago we were seriously discussing nuking old_snapshot_threshold
altogether, on the grounds that it was so buggy as to be unsafe
to use, and nobody was stepping up to fix it [1][2]. It doesn't
appear to me that the situation has got any better, so I wonder if
we're prepared to pull that trigger yet.
regards, tom lane
[1] https://www.postgresql.org/message-id/flat/20200401064008.qob7bfnnbu4w5cw4%40alap3.anarazel.de
[2]
https://www.postgresql.org/message-id/flat/CA%2BTgmoY%3Daqf0zjTD%2B3dUWYkgMiNDegDLFjo%2B6ze%3DWtpik%2B3XqA%40mail.gmail.com