Robert Haas <robertmhaas@gmail.com> writes:
> Patch along these lines attached.
Frankly, I find this quite ugly, and much prefer the general approach of
your previous patch in <BANLkTim433vF5HWjbJ0FSWm_-xA8DDaGNg@mail.gmail.com>.
However, I don't like where you put the execution-time test there. I'd
put it in ExecOpenScanRelation instead, so that it covers both seqscan
and indexscan accesses.
regards, tom lane