Fix range sugar

[globin]

Range sugar was using the trait incorrectly plus the trait implementation was wrong.

Both had the arguments swapped which actually caused the tests to run but threw the debug_assert
in debug builds, causing normal builds of anything using the sugar to fail.

[rust-highfive]

r? @pcwalton

(rust_highfive has picked a reviewer for you, use r? to override)

[huonw]

cc @nick29581, what's the intention of the Option return value of this trait? Is it intended to signal errors in this manner?

[huonw]

Thanks @globin for fixing this!

[nrc]

No, this is not right - the Option is None if for the case where it is impractical or impossible to calculate the size hint, e.g., for open ranges, or ranges over data types where calculating the difference is expensive or impossible.

So, this patch should just switch the params (the history there is the method used to be called difference or something, where it made sense to have the larger thing first) and continue with the debug_assert.

[Thiez]

Perhaps a and b should be given more informative names, such as start and finish or begin and end, or first_index and last_index. Having them named a and b was probably partially responsible for this buggy code.

And surely the step_impl_no_between! macro could be more intelligent, right now, on 32 bits, Step::steps_between(1i64, 5i64) will return None even though the result fits in a uint perfectly. Speaking of which, why does steps_between return the steps as a uint in the first place, rather than Self (which would always avoid overflow)?

[globin]

I've just updated with new param names and the debug_assert! reintroduced.

[globin]

Is steps_between supposed to allow start == end?

[nrc]

Yeah, I think start == end is fine