I tried the following SQL under PostgreSQL 6.3 and 6.4:
create table contents ( id int not null, ix int not null,
volid int not null, level int not null, bdate datetime
not null );
create view freecount asselect c1.id as id, c1.ix as ix, count(c2.ix) as freenessfrom contents c1, contents c2where
c1.volid= c2.volid and c1.bdate <= c2.bdate and c1.level >= c2.levelgroup by c1.id, c1.ix;
Under 6.3, doing the view creation as an ordinary users I got ERROR: pg_rewrite: Permission denied.
which, if I recall, means postgres view support wasn't quite up to
letting everyone creates views. Doing the view creation as the
postgres superuser succeeded but doing select * from freecount;
then crashed the backend.
So I installed the recently announced postsgres 6.4 RPM for Linux and
tried again. This time, I could create the view as a normal user and
it worked fine for that simple select. However, what I actually want
to do on top of that view is
create view freetapes asselect id, min(freeness) - 1from freecountgroup by id;
(i.e. do the nested aggregation that SQL syntax won't let me do
directly.) That view creates successfully but doing a select * from freetapes
produces the error message
ERROR: ExecAgg: Bad Agg->Target for Agg 0
and doing the explicit query
select id, min(freeness) - 1 from freecount group by id;
gives the same message. I'm not familiar with postgres internals but
it looks as though the internal handling of views is still having
trouble with those two levels of aggregations despite the underlying
queries being OK. As a data point, the view creation and queries work
fine under Informix IDS 7.3 (though there's an entirely separate
buglet if you try to subtract one at count() time instead of min()
time :-). Is this problem with postgres something which is a fixable
bug, a missing feature request that is planned to arrive soon (maybe
it's in 6.5?) or a missing feature which isn't going to happen any
time soon?
(I normally only subscribe to pgsql-admin and pgsql-interfaces but
I've now subscribed to pgsql-sql since that seemed to be the most
appropriate choice for this query: if not, I apologise.)
--Malcolm
--
Malcolm Beattie <mbeattie@sable.ox.ac.uk>
Unix Systems Programmer
Oxford University Computing Services
