Michael Paquier <michael@paquier.xyz> writes:
> Move tablespace path re-creation from the makefiles to pg_regress
So this didn't seem like a problem at the time, but while building
beta1 tarballs I discovered that it leaves behind "testtablespace"
subdirectories in various places where they aren't cleaned by
"make distclean", resulting in scary noise in my diff against the
tarballs:
Only in /home/postgres/pgsql/contrib/dblink: testtablespace
Only in /home/postgres/pgsql/contrib/file_fdw: testtablespace
Only in /home/postgres/pgsql/src/pl/plpgsql/src: testtablespace
This appears to be because pg_regress.c will now create the
tablespace directory in any directory that has an "input"
subdirectory (and that randomness is because somebody saw
fit to drop the code into convert_sourcefiles_in(), where
it surely has no business being, not least because that
means it's run twice).
(BTW, the reason we don't see git complaining about this seems
to be that it doesn't complain about empty subdirectories.)
I think what we want to do is have this code invoked only in
test directories that explicitly ask for it, say with a new
"--make-testtablespace" switch for pg_regress.
regards, tom lane