Robert Haas <robertmhaas@gmail.com> writes:
> BTW, has there been any thought to supporting a negative scale for the
> numeric data type? If you can cut off digits after the decimal, why
> not before?
Hm, would there be any real use-case?
An implementation issue is that even in the "long" numeric format,
we cram dscale into a 14-bit unsigned field. You could redefine
the field as signed and pray that nobody has dscales above 8K
stored on disk, but I'm dubious that there's a good argument for
taking that risk.
There might be algorithmic issues as well, haven't really looked.
Any such problems would probably be soluble, if need be by forcing
the scale to be at least 0 for calculation and then rounding
afterwards.
regards, tom lane