Tom Lane wrote:
> That does look weird. Do we have a self-contained test case?
I've been tinkering with this and I now have a self-contained test
case (SQL statements and run results attached). I've debugged through
it and things don't seem right in set_append_rel_pathlist, since
childrel->rows seems to contain the total rows in each table rather
than the number which meet the join conditions. That is reflected in
the output from OPTIMIZER_DEBUG logging, but not in the query plan
from ANALYZE?
I'm afraid I'm a bit stuck on getting farther. Hopefully this much
will be of use to someone. If you could point out where I should
have looked next, I'd be grateful. :-)
-Kevin