[phobos] phobos commit, revision 1767

[Shin Fujishiro]

Andrei Alexandrescu <andrei at erdani.com> wrote:
> 
> Would it mess your plans if we hold off (at least for now) with getNext? 

No. :-)  It should be easy to make them to use front etc.

> I feel the consensus is really weak and I'm not sure getNext is a big 
> improvement. If we do commit to it, that would probably mean changing 
> the core language as well, (foreach) std.algorithm etc.

Agreed.  I'll replace getNext with empty/front/popFront.

Actually, I thought I would eventually use the current range primitives
anyway.  getNext was not very comfortable to use.  (It fairly simplified
streaming range implementation though!)

Here's my feeling about getNext:

- It can't represent infinite ranges.

- Type inference like the following does not work:

auto e = r.front;

- The result is strange:  E* getNext(ref E store).

Using pointer is ok for me, but it's strange that returned address may
be either &store or &some_internal_item.  store is a sheer temporary
and can't be used intuitively like this:

if (r.getNext(store))
    writeln(store); // The front item might not be assigned to store!


Shin