On Mon, Apr 12, 2021 at 10:36:10PM -0700, Noah Misch wrote:
> Christoph Berg's first message on this thread reported doing that. If
> supporting server_user!=pg_regress_user is unwarranted and Christoph Berg
> should stop, then already-committed code suffices.
Not sure that we have ever claimed that. It is unfortunate that this
has broken a case that used to work, perhaps accidentally. At the
same time, Christoph has a workaround for the Debian suite, so it does
not seem like there is anything to do now, anyway.
> The race that commit 6c788d9 fixed is not inherent to Makefile rules. For
> example, you could have instead caused test.sh to issue 'make
> installcheck-parallel TABLESPACEDIR="$outputdir"/testtablespace' and have the
> makefiles consult $(TABLESPACEDIR) rather than hard-code ./testtablespace.
> (That said, I like how 6c788d9 consolidated Windows and non-Windows paths.)
FWIW, I don't really want to split again this code path across
platforms. Better to have one to rule them all.
> I don't know. I never considered server_user!=pg_regress_user before this
> thread, and I don't plan to use it myself. Your latest patch originated to
> make that case work, and my last message was reporting that the patch works
> for a rather-narrow interpretation of server_user!=pg_regress_user, failing on
> variations thereof. That might be fine.
Okay. So.. As I am not sure if there is anything that needs to be
acted on here for the moment, I would just close the case. If there
are more voices in favor of the SQL function using mkdir(), I would
not object to use that, as that looks to work for all the cases where
pg_regress is used.
--
Michael
