Marc L. Allen wrote, On 26/04/12 15:30:
> I suppose using a temporary table is out of the question?
It's much simpler (and probably quicker though I didn't test it) to read
the three rows, sort them by date, pick the first row and use that. No
need to generate random table names, create and drop the table.
Python time.time() gives 0.001sec for all that (to 3 dec places) which
is fast enough for me ;) (and faster than PostgreSQL at 0.023 using the
> again, that only solves the specific issue. I guess the more general
> question is how views with unions interact with aggregates and order
> What happens if you don't use the view, but perform the query using
> the actual tables?
Makes no difference - the planner is still using the scans instead of