Re: Binary search in fmgr_isbuiltin() is a bottleneck.

I wrote:
> [ we should use an index array ]

Just to prove the point, I threw together the attached trivial test case,
which time-trials the existing fmgr_isbuiltin implementation against both
the proposed hash implementation and a simple index array.  On my machine,
with a repeat count of 10000, I get

NOTICE:  bsearch runtime 4234.087 ms
NOTICE:  hash runtime 2542.636 ms
NOTICE:  index runtime 165.184 ms

(These numbers are repeatable within 1% or so.)

It could be argued that trialling OIDs sequentially gives a bit of an
unfair advantage to the bsearch and index methods over the hash method,
because the former are going to suffer fewer cache misses that way.
But I don't see a randomized lookup order changing the conclusion much.

			regards, tom lane