> The only alternative that came to mind when I wrote it was using a numeric > instead of float.
No, I'm wondering what's the justification for smashing it to a single number at all, when the inputs are three-field values. Interval divided by float doesn't produce just a float, for example.
I think I see what you're getting at here. '1 month' / '1 day' could return a number of reasonable values depending on how many days are in the month (28 to 31) and on how many hours are in a day (generally 24, but can be 23 or 25 for DST adjustments). The definition above simply assumes that EXTRACT(epoch...) does the Right Thing. Hmmm. I'm at a loss for the right way to solve this. It seems very reasonable to want to divide intervals by intervals (how many nanocenturies in a fortnight?), but I'm at a loss for how to do that correctly. I'll read the code from EXTRACT(epoch...) and see what happening there.
