protocol for using InputRanges

[monarch_dodra]

On Sunday, 23 March 2014 at 11:52:53 UTC, Timon Gehr wrote:
> On 03/23/2014 01:50 AM, Walter Bright wrote:
>>
>> 2. Can r.front be called n times in a row? I.e. is calling 
>> front()
>> destructive?
>>
>> If true, this means that r.front will have to cache a copy in 
>> many cases.
>
> An analogous problem exists and is more severe for 
> RandomAccessRanges.
>
> import std.algorithm, std.range;
> class Obj{ this(int){} }
> void main(){
>     auto x=[1,2,3].map!(a=>new Obj(a));
>     auto a=x.front;
>     auto b=x.front;
>     auto c=x[0];
>     auto d=x[0];
>     assert(a is b); // fail
>     assert(b is c); // fail
>     assert(c is d); // fail
> }

I have an open proposal for a "cache" range adaptor that somewhat 
eliviates the problem:
https://github.com/D-Programming-Language/phobos/pull/1364

But it restricts the range to bidirectional (for the reasons 
above).

Actually, it's mimited to Bidir, but if you *do* use it as 
bidirectional, then 1 element will be evaluated twice. So I'm 
wondering if it might not be better to simply restrict it to 
forward...