Boris Folgmann <misc(at)folgmann(dot)com> writes:
> Look closely: postmaster now thinks that the first argument 1.0 is NUMERIC,
Yup. This is not a bug, it's intentional (and per SQL spec, AFAICT).
Your problem is that 123::real/25 yields a double precision result,
which is not implicitly castable to real anymore. I'd suggest
switching the function arguments to double precision.
regards, tom lane