Stephen Frost <sfrost@snowman.net> writes:
> If that's the only consideration for this, well, that's certainly quite
> straight-forward to change in the other direction too. The new function
> suggested by Andres actually makes it really easy to get a textual list
> of all the role attributes which a role has from the bitmask too.
We could have that regardless of the representation, if the function is
defined along the lines of "given a user OID, give me a text string
representing the user's attributes". However, that only helps for
pg_dumpall and any other clients whose requirement is exactly satisfied
by a string that fits into CREATE/ALTER USER. The current formatting
of psql's \du, for example, absolutely requires adding more client-side
code every time we add a property; whether the catalog representation is
bools or a bitmask really isn't going to change the pain level much there.
regards, tom lane