On Wed, Jun 26, 2013 at 3:29 AM, Szymon Guz <mabewlun@gmail.com> wrote: > OK, so I think this patch can be committed, I will change the status.
We have a convention that roles created by the regression tests needs to have "regress" or something of the sort in the name, and that they need to be dropped by the regression tests. The idea is that if someone runs "make installcheck" against an installed server, it should pass - even if you run it twice in succession. And also, it shouldn't be likely to try to create (and then drop!) a role name that already exists.
Setting this to "Waiting on Author".
Hi Robert,
Attached is an updated patch that prepends 'regress' before role names.
As for dropping ROLEs is concerned, all the roles created in the previous patch were within transactions. So didn't have to explicitly drop any ROLEs at the end of the script.