Andres Freund <andres@anarazel.de> writes:
> On 2021-06-15 19:26:25 -0400, Tom Lane wrote:
>> Going forward it wouldn't be a problem, but back-patching isolation
>> test cases might find it annoying. On the other hand, my nearby
>> patch to improve isolation test stability is already going to create
>> issues of that sort. (Unless, dare I say it, we back-patch that.)
> It might be worth to back-patch - aren't there some back branch cases of
> test instability? And perhaps more importantly, I'm sure we'll encounter
> cases of writing new isolation tests in the course of fixing bugs that
> we'd want to backpatch that are hard to make reliable without the new
> features?
Yeah, there absolutely is a case to back-patch things like this. Whether
it's a strong enough case, I dunno. I'm probably too close to the patch
to have an unbiased opinion about that.
However, a quick look through the commit history finds several places
where we complained about not being able to back-patch isolation tests to
before 9.6 because we hadn't back-patched that version's isolationtester
improvements. I found 6b802cfc7, 790026972, c88411995, 8b21b416e without
looking too hard. So that history certainly suggests that not
back-patching such test infrastructure is the Wrong Thing.
(And yeah, the failures we complained of in the other thread are
certainly there in the back branches. I think the only reason there
seem to be fewer is that the back branches see fewer test runs.)
regards, tom lane