Robert Haas <robertmhaas@gmail.com> writes:
> ...
> 2. Modify pg_amop by adding a new column amopcategory, probably either
> int2 or maybe even just char.
> ...
> I'm not prepared to endorse doing #3 in core for 9.0, but I wonder if
> it would be feasible to think about doing #1 and #2 and putting
> something into contrib for #3.
No, we are not touching the system catalogs for this in 9.0, much less
fooling with any planner logic. If it had been submitted a few months
earlier that could have happened, but we are barely 48 hours away from
alpha freeze. The only part of this that I think can even be considered
at this point is to do a backwards-compatible source code upgrade that
will decouple the various call sites from knowledge of exactly how many
key columns the syscaches allow. And even that is not for the benefit
of this feature; it's mainly to minimize breakage of other patches
that will be developed between now and whenever knngist does land.
(IOW, I agree with your point that making the call syntax change now
is the least painful time to do that.)
regards, tom lane