Andres Freund <andres@anarazel.de> writes:
> On 2022-09-09 12:48:38 -0400, Tom Lane wrote:
>> Pushed with those changes and doc updates. I did not push the
>> variant expected-file. I think the entire point here is that
>> we are *not* deeming the new NetBSD implementation acceptable,
>> so allowing it to pass regression tests is the wrong thing.
> What do we gain from the regression test failing exactly this way, given that
> we know it's a problem?
It tells people not to use --with-uuid=bsd on those NetBSD versions.
They can either do without uuid-ossp, or install ossp or e2fs.
("Do without" is not much of a hardship, now that we have
gen_random_uuid() in core.)
IMV a substantial part of the point of the regression tests is to
let end users and/or packagers verify that they have a non-broken
installation. Hiding a problem by making the tests not fail
basically breaks that use-case.
If we had, say, a known openssl security bug that was exposed by our
test cases, would you advocate dumbing down the tests to not expose
the bug?
> It just makes it harder to run tests.
Harder for who? AFAICT there is nobody but me routinely running
full tests on NetBSD, else we'd have found this problem much earlier.
I've got my animals configured not to use --with-uuid (not much of
a lift considering that's the buildfarm's default). End of problem.
regards, tom lane
