Re: Expression indexes and casts

Stephan Szabo <sszabo(at)megazone(dot)bigpanda(dot)com> writes:
> I haven't done any looking around yet (about to head off to work), but it
> looks like in the case where the system decides to cast a to text in order
> to get a working equality, the index isn't used, whereas in the case where
> I explicitly cast it, it can.

I think the problem is that explicit and implicit casts are marked
differently in the cast parse node, causing equal() to consider the two
expressions different.

There is currently a hack involving a "don't care" setting for this
field, but it doesn't help you. I wonder if it would be better to make
equal() explicitly ignore the cast-type field. It seems like that could
break other things though :-(.

A narrower patch would be to change the cast type field to don't-care in
the copy of the parse tree that is made for planner user.

[ thinks some more... ] On the other hand, there are cases where
explicit and implicit casting are actually semantically different (think
varchar() and char() length constraints). Maybe the don't-care business
is itself a bug, and you're just stuck.

regards, tom lane