In both cases we'd return a double but we use the fast ipow if it's
possible (which can be 20x faster), so at the cost of an extra cast if
you need an int, we'd have a consistent API. Would this be acceptable?
It seems OK to me.
Computing as an int, casting to double and back to int8 can generate a
loss of precision. However for powers of 2 it works exactly, so eg
computing a mask it would be ok.
This proposal does not exactly match SQL behavior, but I do not see this
as a problem, which is why I was happy with the previous proposal.
--
Fabien.
