Re: Virtual generated columns

On 09.09.24 08:02, Peter Eisentraut wrote:
> On 04.09.24 12:33, Dean Rasheed wrote:
>>> I left the 0001 patch alone for now and put the new rewriting
>>> implementation into 0002.  (Unfortunately, the diff is kind of useless
>>> for visual inspection.)  Let me know if this matches what you had in
>>> mind, please.  Also, is this the right place in fireRIRrules()?
>> Yes, that's what I had in mind except that it has to be called from
>> the second loop in fireRIRrules(), after any RLS policies have been
>> added, because it's possible for a RLS policy expression to refer to
>> virtual generated columns. It's OK to do it in the same loop that
>> expands RLS policies, because such policies can only refer to columns
>> of the same relation, so once the RLS policies have been expanded for
>> a given relation, nothing else should get added to the query that can
>> refer to columns of that relation, at that query level, so at that
>> point it should be safe to expand virtual generated columns.
>
> If I move the code like that, then the postgres_fdw test fails.  So
> there is some additional interaction there that I need to study.

This was actually a trivial issue. The RLS loop skips relation kinds
that can't have RLS policies, which includes foreign tables. So I did
this slightly differently and added another loop below the RLS loop for
the virtual columns. Now this all works.

I'm attaching a consolidated patch here, so we have something up to date
on the record. I haven't worked through all the other recent feedback
from Jian He yet; I'll do that next.