Thomas Lockhart <lockhart@fourpalms.org> writes:
> Yes. Mixing the float with the int seems to invite math troubles though
True, any multi-member struct will be a huge PITA for arithmetic
operations.
Using int8 if available or float8 if not might work pretty nicely
now that I think about it, at least from a coding point of view.
You could hide a lot of the differences in the PG_GET and PG_RETURN
macros.
But do we want to have such a radical difference in accuracy and
range across platforms?
regards, tom lane