Difference between input range and forward range

[via Digitalmars-d]

>     auto osmData = datastream.take(size).array;
>     datastream.popFrontN(size);
>     auto header = BlobHeader(osmData);
>
> http://he-the-great.livejournal.com/49636.html
>
> How do I know if popFrontN is needed? If I was given a value 
> base range then it is. If I was given a reference range (in its 
> many forms) the extra call to popFrontN will result in an 
> unneeded data jump. I could require that a forward range is 
> passed in, then I can save() before calling .array and thus 
> always require popFrontN.
>
> The best option is probably to use the RefRange wrapper, but it 
> does create an annoying element of surprise.

Yes. I agree. It's good example of problem. I just had problem 
like this with *take* function, *startsWith* and others. So we 
have two kinds of ranges: with reference and value semantics. But 
both of them could be of reference or value *types* (struct or 
class). So how would we determine at CT (by generic algorithms) 
wheter current range has ref or value semantics (we couldn't just 
rely on testing if it's class or struct)!? I think it's 
important, because it can make influence on programme logic. And 
again I want to say that we must explicitly say (in doc) that we 
have two logical range categories, so new users would not make 
stupid mistakes. Also unittests would be good for ref and value 
ranges to illustrate our intentions.