On Mon, Jun 08, 2020 at 11:09:42AM +0200, Daniel Gustafsson wrote:
> Nice catch, testing the attached patch does indeed make pg_regress run without
> errors if outputdir doesn't exist.
No objections to that. That seems more useful than having to recreate
them before running any custom script specifying --outputdir.
> While looking at it though, I noticed that we nearby use stat() in a single
> place while the rest use check_directory(). Is there a reason to not use
> check_directory consistently as per the attached diff?
Indeed. And two extra things are that you can just simplify
src/bin/pg_upgrade/test.sh and src/tools/msvc/vcregress.pl so as they
don't create manually those repositories. This report is not really
annoying for the average user in my opinion, so there may be little
point in backpatching this stuff.
And this leads me to the attached.
--
Michael