Re: Bogus collation version recording in recordMultipleDependencies

On Thu, Apr 15, 2021 at 10:06:24AM -0400, Tom Lane wrote:
> Julien Rouhaud <rjuju123(at)gmail(dot)com> writes:
> > On Wed, Apr 14, 2021 at 01:18:07PM -0400, Tom Lane wrote:
> >> (To be clear: 0002 passes check-world as-is, while 0001 is not
> >> committable without some regression-test fiddling.)
>
> > I'm probably missing something obvious but both 0001 and 0002 pass check-world
> > for me, on a glibc box and --with-icu.
>
> 0001 fails for me :-(. I think that requires default collation to be C.

Oh right, adding --no-locale to the regress opts I see that create_index is
failing, and that's not the one I was expecting.

We could change create_index test to create c2 with a C collation, in order to
test that we don't track dependency on unversioned locales, and add an extra
test in collate.linux.utf8 to check that we do track a dependency on the
default collation as this test isn't run in the --no-locale case. The only
case not tested would be default unversioned collation, but I'm not sure where
to properly test that. Maybe a short leading test in collate.linux.utf8 that
would be run on linux in that case (when getdatabaseencoding() != 'UTF8')? It
would require an extra alternate file but it wouldn't cause too much
maintenance problem as there should be only one test.