Scara Maccai <m_lists@yahoo.it> writes:
> I ran ANALYZE on both tables, but the plan shows rows=1 in the inner
> table results.
> If I change the "left outer join" into an "inner join" the bitmap index
> scan of the inner table of the nested loop shows a correct number of
> rows (not the exact same thing as "actual rows", but very close).
Yeah, this is a bug: it's effectively double-counting the selectivity of
the index clause. The error isn't enormously critical, since the join
size estimate is right; but it could perhaps lead to choosing a plain
indexscan when a bitmap scan would be better. I've applied a patch.
regards, tom lane