=?UTF-8?Q?LEMAIRE_Leslie_=28Charg=C3=A9e_de_mission=29_=2D_SG=2FDNUM=2F?= =?UTF-8?Q?UNI=2FDRC?=
<leslie.lemaire@developpement-durable.gouv.fr>writes:
> I've noticed a rather odd behavior with a PL/pgSQL function whose
> definition includes a cast to regrole on a literal string, such as
> 'a_role'::regrole.
Yeah, the coverage for REG* constants in plan invalidation is pretty
thin --- in fact, I think this *only* works correctly for regclass
constants. AFAIR you're the first to complain, so I'm not sure that
we want to expend the effort to expand that ...
regards, tom lane