Bruce Momjian <pgman(at)candle(dot)pha(dot)pa(dot)us> writes:
> The reason I was being cautious is to handle cases where people are
> poking in pg_shadow directly.
If they're poking pg_shadow directly, I think it's up to them to avoid
or cope with sysid conflicts (the unique indexes on the table will
prevent the worst errors). ISTM it is sufficient for CREATE USER
to not cause conflicts when used as documented.
regards, tom lane