Martijn van Oosterhout <kleptog(at)svana(dot)org> writes:
> I can't see why this wouldn't work.
Doubtless you could do it. The problem with the idea is that those
inference tests are pretty expensive. I think that any such thing would
waste significant numbers of cycles on ordinary queries while only being
a win on a few poorly-written queries.
We do have to make a tradeoff between planning time and execution time,
and I fear that this idea is not going to be a win in those terms.
If you feel like experimenting, though, go for it ...
regards, tom lane