On 01/09/2014 06:48 PM, Dean Rasheed wrote:
> On 8 January 2014 10:19, Dean Rasheed <dean.a.rasheed@gmail.com> wrote:
>> The assertion failure with inheritance and sublinks is a separate
>> issue --- adjust_appendrel_attrs() is not expecting to find any
>> unplanned sublinks in the query tree when it is invoked, since they
>> would normally have all been planned by that point. However, the
>> addition of the new security barrier subqueries after inheritance
>> expansion can now insert new sublinks which need to be planned. I'll
>> look into how best to make that happen.
>
> The attached patch does that, which fixes the case you reported.
Dean, any objections to adding this to the current CF, or to my doing so?
I want to adjust the RLS patch to build on top of this patch, splitting
the RLS patch up into a series that can be considered separately. To
have any hope of doing that, I'm going to need to be able to base it on
this patch.
Even if -hackers collectively decides the approach you've posted for
updatable s.b. views isn't the best way at some future point, we can
replace it with a better one then without upsetting users. RLS only
needs quite a high level interface over this, so it should adapt well to
anything that lets it wrap a table into a s.b. qualified subquery.
If there's no better approach forthcoming, then I think this proposal
should be committed. I'll do further testing to see if I can find
anything that breaks it, of course.
I've been bashing my head against this for weeks without great
inspiration - everything I try when doing this in the rewriter creates
three problems for every one it fixes. That's not to say it can't be
done, just that I haven't been able to do it while trying to learn the
basics of the rewriter, planner and executor at the same time ;-)
I've been consistently stuck on how to expand the tlist and inject ctid
(and oid, where required) Vars back *up* the chain of expanded
subqueries after views are expanded in rewrite. We only know the
required tlist and can only access the ctid and oid attrs once we expand
the inner-most view, but we've got no way to walk back up the tree of
subqueries (Query*) to inject Var tlis entries pointing to the
next-inner-most subquery. Need some way to walk back up the nested tree
of queries injecting this info. (I've had another idea about this that I
need to explore tonight, but every previous approach I've tried has
circled back to this same problem).
-- Craig Ringer http://www.2ndQuadrant.com/PostgreSQL Development, 24x7 Support, Training & Services
