Ashutosh Bapat <ashutosh.bapat.oss@gmail.com> writes:
> partition-wise join should be willing to fallback to non-partitionwise
> join in such a case. After spending a few minutes with the code, I
> think generate_partitionwise_join_paths() should not call
> set_cheapest() is the pathlist of the child is NULL and should just
> wind up and avoid adding any path.
We clearly need to not call set_cheapest(), but that's not sufficient;
we still fail at higher levels, as you'll see if you try the example
Richard found. I ended up making fe12f2f8f to fix this.
I don't especially like "rel->nparts = 0" as a way of disabling
partitionwise join; ISTM it'd be clearer and more flexible to reset
consider_partitionwise_join instead of destroying the data structure.
But that's the way it's being done elsewhere, and I didn't want to
tamper with it in a bug fix. I see various assertions about parent
and child consider_partitionwise_join flags being equal, which we
might have to revisit if we try to make it work that way.
regards, tom lane