Josh Berkus <josh@agliodbs.com> writes:
> If #19 is missing it's because Oleg & I could not get it to complete. That
> was also the query which we are most interested in testing.
Q19 doesn't seem to be particularly slow in either the 7.4 or 7.5 tests
--- there are many others with longer runtimes. I speculate that what
is actually being run here is a modified Q19 query with the merge join
condition pulled out by hand. The CVS-tip planner should be able to do
that for itself, though, and obtain essentially this same performance
with the per-spec query.
regards, tom lane