"David G. Johnston" <david.g.johnston@gmail.com> writes:
> On Wed, Jan 17, 2018 at 2:58 PM, Tomas Vondra <tomas.vondra@2ndquadrant.com>
> wrote:
>> But maybe we could/should fix it anyway? Most regression tests switched
>> to roles prefixed with regress_* so why not to do the same here?
> The point of the test seems to be to ensure that the special system
> keywords, when quoted, are allowed to be used for role names.
Exactly. Changing the names ruins the point of the test.
> So the
> choice is to make the test conditional (if the role previously exists
> neither create or drop it - and since it existed it doesn't seem like its a
> problem to create it anyway) or to simply not bother testing "user"
> figuring that the other two roles suffice for testing this behavior.
I wouldn't have a big problem with just dropping this whole test stanza.
It's an out-and-out violation of our rule against not creating rolenames
not starting with "regress_", and it's not testing anything that seems
especially likely to break.
regards, tom lane