Re: [GENERAL] Question about partial functional indexes and the query planner

Robert Haas <robertmhaas(at)gmail(dot)com> writes:
> On Tue, Jun 10, 2014 at 7:19 PM, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
>> Given the lack of previous complaints, I'm not sure this amounts to
>> a back-patchable bug, but it does seem like something worth fixing
>> going forward.

> Agreed, although I'd be willing to see us slip it into 9.4. It's
> doubtful that anyone will get upset if their query plans change
> between beta1 and beta2, but the same cannot be said for released
> branches.

After further thought about this I realized that there's another category
of proof possibilities that is pretty simple to add while we're at it.
Namely, once we've found that both subexpressions of the two operator
clauses are equal(), we can use btree opclass relationships to prove that,
say, "x < y implies x <= y" or "x < y refutes x > y", independently of
just what x and y are. (Well, they have to be immutable expressions, but
we'd not get this far if they're not.) We already had pretty nearly all
of the machinery for that, but again it was only used for proving cases
involving comparisons to constants.

A little bit of refactoring later, I offer the attached draft patch.
I'm thinking this is probably more than we want to slip into 9.4
at this point, but I'll commit it to HEAD soon if there are not
objections.

regards, tom lane