Change range_rev to be inclusive on the low end

[sanxiyn]

Currently range_rev(hi, lo) iterates [hi..lo), but it probably should iterate (hi..lo].

For example, consider iterating indexes of vector of length 10 from the end. Currently that is range_rev(10-1, -1), but since hi and lo should be the same type, this can't be uint, and to be used as an index it needs to be cast. range_rev(10, 0) would be better.

This is from the mailing list discussion.

[yjh0502]

How about range_rev(lo, hi) which iterates over (hi..lo]? For example, range_rev(0, 10) => [9, ..., 0]

[pnkfelix]

Not sure if I understand what "I-wrong" is supposed to denote. Anyway, this seems easy to fix (apart from bikeshed'ing the argument order).

[pnkfelix]

Well, there is actually the issue that both range and range_rev are implemented in terms of range_step, which is easy right now because they both traverse a range of the form [x, y). If we make this change, it might be easier to make things work if they are implemented atop a function that traverses a range of the form [x,y], rather than having to handle such different cases of [x,y) and (x,y] within a single routine. (I.e. generate the inclusive limit value y from the original stop by subtracting step from stop.) Or maybe this is a bad idea overall.

[pnkfelix]

I think I have a pretty clean way to put this in. I'll have a patch up for review shortly. Assigning bug to self.