libcore: add num::Int::pow() and deprecate num::pow().

[nodakai]

[breaking-change]

Deprecates core::num::pow in favor of Int::pow.

[Gankra]

Should probably be #[deprecated]?

[alexcrichton]

I think that the design of std::num is deliberately quite conservative in an attempt to cut down on the number of extraneous traits to ensure we can extend the design into the future. That would currently mean that the sq method would actually be part of the Int trait instead of having a new separate trait (and perhaps also defined on Float) if we were to add it. We don't currently use the traits for many convenience methods, however, so we may not want to add it at this time.

The pow method seems fine though to add to the Int trait! I would second @gankro's suggestion for leaving the old #[deprecated] function instead for now.

cc @aturon

[aturon]

I agree with @alexcrichton; please see the numerics reform RFC for details.

I had been thinking we could/should move pow to be a method on Int -- can you cut down the PR to just that aspect?

[aturon]

cc @bjz

[nodakai]

@gankro Yeah, I shouldn't have been so brutal! I deprecated the free function core::num::pow() and removed the doctest (what's the official name for it?) from it because make check yields a deny(deprecated) error there otherwise.

@alexcrichton@aturon Now this PR is only about the Int::pow() method.

[brendanzab]

The Int::pow method should be fine :)

[brendanzab]

Why did the implementation change?

[nodakai]

I simplified the code by removing the special case for exp==1 because we don't have the move semantics with primitive integers. exp /= 2 is more readable than exp = exp >> 1 and equivalent to it:

• http://is.gd/cYcfFU (hit the [ir] button)

[brendanzab]

I'm guessing LLVM would optimise to a bit shift anyway.

What do you mean by 'don't have the move semantics with primitive integers'?

[brendanzab]

cc. @huonw

[huonw]

Seems fine to me.

[nodakai]

@bjz The original code could move the base parameter to the caller as its return value when exp is 1. It could perhaps achieve better performance only with bigint objects whose initialization cost was not negligible.

[brendanzab]

Updated the issue title and description.