Re: [HACKERS] RI status report #1

wieck(at)debis(dot)com (Jan Wieck) writes:
> just to give anyone a chance to complain, I'd like to
> describe what I plan to implement for v6.6 in the referential
> integrity (RI) corner.

Jan, I have no comments about the RI features, but I am a little worried
about not creating big headaches in merging different changes. Can we
work out a schedule that will minimize tromping on each others' toes?

I am in the middle of some fairly extensive revisions in
rewriteManip.c, rewriteHandler.c, and ruleutils.c --- basically,
fixing all the routines that recurse through expression trees to use
expression_tree_walker and expression_tree_mutator, for a big space
savings and elimination of a bunch of routine-X-doesn't-handle-node-
type-Y bugs. Also I'm going to fix the rule deparser to use a
stringinfo buffer so it doesn't have any hardwired limits on the textual
length of a rule. And I think I know how to fix some of the problems
with aggregates in subselects, like this one:

create table t1 (name text, value float8);
CREATE
select name from t1 where name IN
(select name from t1 group by name having 2 = count(*));
ERROR: SELECT/HAVING requires aggregates to be valid

(It looks to me like some of the routines recurse into subselects when
they shouldn't. It's a lot easier to see that sort of issue when
there's no code left except the actual Var manipulation and the
nonstandard recursion decisions ;-).)

I intended to finish these up in the next few days and commit them,
but if you've already started major hacking in these files then maybe
we should find another way.

Also, I believe Thomas is in the middle of wide-ranging revisions in
the parser, so you'd better coordinate with him on touching that area.

regards, tom lane