partition(range, leftsubrange) or partition(range, rightsubrange)

[Andrei Alexandrescu]

superdan wrote:
> got a question on this range stuff. in stl partition is
> partition(begin, mid, end). neat. in std.algorithm partition is
> partition(range, mid). so-so. never like it mucho. in the new stuff
> with ranges n all there are two choices partition(range,
> leftsubrange) or partition(range, rightsubrange). question is, is
> there a better choice between the two or are they just the same.
> would be cool to have a clear rule with some advantage. and that's
> easy to remember too.
> 
> like steve i think ranges are cool n all but fraid iterators are
> still good at sumthin'. either-way choices ain't a good sign.

I think I have an answer. All algorithms supposed to take begin, middle, 
and end should take range(begin, end) and range(middle, end) and not any 
other combination. Moreover, whenever a collection can choose to returns 
a subrange or another, it should return the range to the right.

This is because right-open ranges are the most general, e.g. 
singly-linked lists with range==Node* (or other similar 
sentinel-terminated collections).

Imposing a range of the kind range(begin, middle) would make that 
algorithm not work for singly-linked list unless they implement a "fat" 
range containing two Node*.

So rotate should be:

R rotate(R)(R range, R subrange);

Description: moves subrange to the front of range, and whatever was in 
front of the subrange right after it. Preconditions: subrange is a 
subrange of range (duh). Returns the range after the moved subrange. (In 
fact I already mentioned I plan to give rotate the more descriptive name 
moveToFront.)

Makes sense?


Andrei