We get substantial wins on all of fx, fx3, fx4. fx2 is the case that gets inlined and never reaches functions.c, so the lack of change there is expected. What I found odd is that I saw a small speedup (~6%) on fx5 and fx6; those functions are in plpgsql so they really shouldn't change either. The only thing I can think of is that I made the hash key computation a tiny bit faster by omitting unused argtypes[] entries. That does avoid hashing several hundred bytes typically, but it's hard to believe it'd amount to any visible savings overall.
Anyway, PFA v10.
I can confirm so all tests passed without problems
