Re: [COMMITTERS] pgsql: Use a bitmask to represent role attributes

José Luis Tallón wrote:
> On 12/23/2014 04:46 PM, Andres Freund wrote:

> >I personally would prefer a 'custom' type to represent the
> >permissions. Internally that could very well be current bitmask, but the
> >external representation could be more complex (i.e. some textual
> >representation). That'd make it easy to make the representation wider/more
> >complex if needed.
>
> Indeed, though this would imply adding a new "bitstring?" type to core
> Postgres.

We already have varlena bitstrings, in the guise of types bit and
varbit.

> Do you have any further input on what this type would look like ? Any
> operators that might be useful? ISTM that this would actually be the
> greatest strength of a type proper (vs. "hardcoded" bit-wise operations in
> core)

I imagine something like the "reg*" types (regclass, regtype etc): on
input you can pass them an OID, or an possibly-qualified object name;
internally they store the OID. You can cast them to OID to obtain the
numerical value, or just print them out to get the possibly-qualified
name.

In the case at hand, on output you would get the equivalent of the
text[] you get from pg_role_all_attributes(), and you can input it in
the same way or you can input the bitmask; and the underlying storage is
the bitmask.

This doesn't solve the client compatibility break, or the issue that
querying pg_roles is expensive.

--
Álvaro Herrera http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services