Re: Michael Paquier > So you basically mimicked the makefile rule that this commit removed > into your own test suite. Reverting the change does not really help, > because we'd be back to square one where there would be problems in > parallel runs for developers. Saying that, I would not mind adding an > option to pg_regress to control if this cleanup code is triggered or > not, say something like --no-tablespace-cleanup? Then, you could just > pass down the option by yourself before creating your tablespace path > as you wish.
I don't think adding more snowflake code just for this use case makes sense, so I can stick to my workaround. I just wanted to point out that the only thing preventing the core testsuite from being run as a true client app is this tablespace thing, which might be a worthwhile fix on its own. Maybe creating the tablespace directory from within the testsuite would suffice? CREATE TABLE foo (t text); COPY foo FROM PROGRAM 'mkdir @testtablespace@'; CREATE TABLESPACE regress_tblspace LOCATION '@testtablespace@'; Christoph
