> > The patch looks unduly invasive to me, but I think that it might be
> > right that we should go back to a macro-based implementation, because
> > otherwise we don't have a good way to be certain that the function
> > parameter won't get evaluated first.
>
> I'd first like to see some actual evidence of this being a problem,
> rather than just the order of the checks.
There seem to be enough evidence of this being the problem. We are
better off going back to the macro-based implementation. I polished
Keisuke Kuroda's patch commenting about the performance issue, removed
the check_float*_val() functions completely, and added unlikely() as
Tom Lane suggested. It is attached. I confirmed with different
compilers that the macro, and unlikely() makes this noticeably faster.