Re: Clause accidentally pushed down ( Possible bug in Making Vars outer-join aware)

I wrote:
> I have a WIP patch that does it like this, and it fixes the presented
> case; but it's not complete so it's still failing some existing
> regression tests. More later.

Here's said patch. Although this fixes the described problem and
passes check-world, I'm not totally happy with it yet: it feels
like the new add_outer_joins_to_relids() function is too expensive
to be doing every time we construct a join relation. I wonder if
there is a way we can add more info to the SpecialJoinInfo data
structure to make it cheaper. An obvious improvement is to store
commute_below_l explicitly instead of recomputing it over and over,
but that isn't going to move the needle all that far. Is there a
way to not have to scan all the SpecialJoinInfos?

I may be worrying over nothing --- it's likely that the path
construction work that will happen after we make the join relation
RelOptInfo will swamp this cost anyway. But it feels ugly.

Another thing I'm wondering about is whether this'd let us get
rid of RestrictInfo.is_pushed_down and RINFO_IS_PUSHED_DOWN.
I tried really hard to remove those in favor of checking to see
whether a qual clause's required_relids include the outer join
being formed, which seems like a much more principled way of
deciding whether it's a filter or join clause. I couldn't get
that to work, but now I wonder if what I was running into was
really bugs associated with not understanding that we haven't
fully formed the lower outer join when we apply identity 3.

regards, tom lane