Role incompatibilities

Trying to work in the new role features into the information schema, I
noticed that there might be a few incompatibilities between the
implementation and what the SQL standard would like to see.

The way I understand this is that, according to the SQL standard, there
should be a current user and optionally a current role. A current role
can be set by running SET ROLE, and that is only permissible if that
role has been granted to the current user. (It seems that this must
have been a direct grant, but that is less important.) The set of
applicable privileges (used for permission checking) is now the
privileges held by the current user, the current role, and all roles
that have been granted to the current role.

It seems that the "inherit" functionality was invented to simulate
something like this but it doesn't quite do it. What we'd really need
is a system where roles granted to the current user are not
automatically activated but roles granted to the current role are. The
inherit functionality is then only to simulate traditional groups that
activate all their privileges automatically depending on who is the
current user.

The other problem is that using SET ROLE activates the privileges of a
role but loses the privileges of the current user. In practice this
may mean that it reduces your privileges while you might want to use it
to augment your privileges.

What both of these observations come down to is that in my estimation
current user and current role should be separated.

It's quite possible that I'm reading this wrong in a hurry or can't
quite simulate it right, so please enlighten me.

--
Peter Eisentraut
http://developer.postgresql.org/~petere/