protocol for using InputRanges

[monarch_dodra]

On Saturday, 29 March 2014 at 00:14:40 UTC, Andrei Alexandrescu 
wrote:
> It increasingly seems to me we need to do this. -- Andrei

Question: Should this new restriction only be applied to 
*Input*Ranges? If so, then it might make more sense.

Ranges that also satisfy the *Forward* interface (which has 
save), or specifically, the *Output* interface (that doesn't even 
have empty) should not have this obligation. In particular, 
output ranges *write*, so usually can't even buffer data.

*Input*-only Ranges are the ones that are usually implemented 
over streams, and that are most subject to the "test empty" issue 
too, so they are the ones that need this restriction.

Could this maybe be a good solution for everyone?