On May 12, 11:15 am, Arnaud Delobelle <[EMAIL PROTECTED]> wrote: > But exp(y*log(x)) -> 1 as (x, y) -> (0, 0) along any analytic curve > which is not the x=0 axis (I think at least - it seems easy to prove > that given f and g analytic over R, f(x)*ln g(x) -> 0 as x -> 0 if > f(0)=g(0)=0 and g(x)>0 in the neighbourhood of 0).
Agreed. And this makes an excellent argument that if you're going to choose a number for 0.0**0.0 then it's got to be 1. But I still don't find it completely convincing as an argument that 0.0**0.0 should be defined at all. > This should cover most practical uses? Maybe. But if you're evaluating x**y in a situation where x and y represent physical quantities, or otherwise have some degree of error, then you probably want to be warned if x and y both turn out to be zero. I seem to be digging myself into a hole here. I'm personally firmly in the "0**0 should be 1" camp, and always have been--- there are just too many practical benefits to defining 0**0==1 to ignore, and in the case where you're interested in integer exponents anything else is just plain wrong. The lack of continuity of the power function at (0,0) seems a small price to pay. Mark -- http://mail.python.org/mailman/listinfo/python-list