I wrote:
> I'm a bit disturbed by that, because AFAICS from the plans, these queries
> did not involve any semi or anti joins, which should mean that the patch
> would not have changed the planner's behavior.
After contemplating my navel for awhile, it occurred to me that maybe the
patch's changes in add_path_precheck could have caused behavioral changes
at the margins. Could you try a run without that (without the last two
hunks in pathnode.c)?
regards, tom lane