Re: API change advice: Passing plan invalidation info from the rewriter into the planner?

On Tue, Jun 24, 2014 at 10:30 AM, Alvaro Herrera
<alvherre@2ndquadrant.com> wrote:
> Robert Haas wrote:
>> > Right, if we were to support multiple policies on a given table then we
>> > would have to support adding and removing them individually, as well as
>> > specify when they are to be applied- and what if that "when" overlaps?
>> > Do we apply both and only a row which passed them all gets sent to the
>> > user?  Essentially we'd be defining the RLS policies to be AND'd
>> > together, right?  Would we want to support both AND-based and OR-based,
>> > and allow users to pick what set of conditionals they want applied to
>> > their various overlapping RLS policies?
>>
>> AND is not a sensible policy; it would need to be OR.  If you grant
>> someone access to two different subsets of the rows in a table, it
>> stands to reason that they will expect to have access to all of the
>> rows that are in at least one of those subsets.
>
> I haven't been following this thread, but this bit caught my attention.
> I'm not sure I agree that OR is always the right policy either.
> There is a case for a policy that says "forbid these rows to these guys,
> even if they have read permissions from elsewhere".  If OR is the only
> way to mix multiple policies there might not be a way to implement this.
> So ISTM each policy must be able to indicate what to do -- sort of how
> PAM config files allow you to specify "required", "optional" and so
> forth for each module.

Hmm.  Well, that could be useful, but I'm not sure I'd view it as
something we absolutely have to have...

-- 
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company