Martijn van Oosterhout <kleptog@svana.org> writes:
> I considered a patch to make the parser treat all numbers as unknown, but
> that would break any code that relies on numbers-are-int4 (think function
> and operator type resolution).
Right ... if there were a simple solution we'd have done it already.
The line of thought that's been discussed in the past is to add an
UNKNOWNNUMERIC placeholder type, with similar behavior to UNKNOWN except
that it's used for something that was a numeric-looking literal rather
than a string-looking literal. I don't recall that anyone worked out
a complete proposal --- check the archives if you want to work on this.
regards, tom lane