broken nextpow functions

[stevengj]

I've been looking at the prevpow/nextpow functions, and they seem somewhat broken to me:

• nextpow2(n) != nextpow2(2, n): the former returns the power-of-2 value, and the latter returns the exponent. e.g. nextpow2(9) is 16, but nextpow(2, 9) is 4. Similarly for prevpow.
• nextpow2(0) == prevpow2(0) == 0. I can only understand this behavior under the theory that zero is 2^–∞, but that seems questionable to me. nextpow(2, 0) and prevpow(2, 0) raise an InexactError (I would have expected a DomainError, maybe?).
• prevpow(2, 1) correctly returns 0 (1 == 2^0, see first point), but nextpow(2, 1) throws a DomainError, which is certainly wrong.
• nextpow(2, -7) and prevpow(2, -7) raise DomainErrors, but nextpow2(-7) gives -8 and prevpow2(-7) gives -4.
• The documentation for these functions does not clarify any of the above behaviors.

[JeffBezanson]

@timholy can we change the meaning of nextpow(a, x)?

[timholy]

My contribution was nextprod, I don't think I have any special investment in nextpow. Indeed, changing nextpow to return power-of-2 value will make it more consistent with nextprod.

(I get confused about those names myself sometimes.)

[JeffBezanson]

Good. I see the uses in Base look like a.^nextpow(a, x), so the change seems warranted. I checked the ~100 packages I have cloned and I don't see any uses so far.

[stevengj]

(Probably for 0.3, though.)