On 9/20/13 8:54 AM, Joseph Rushton Wakeling wrote:
On 20/09/13 17:28, Andrei Alexandrescu wrote:
Why would there be a performance loss?
Depends on the particular case, but my experience is that _in practice_
stuff based around range interfaces can often be slower than raw iteration.
I don't think anyone's saying a performance loss is inevitable or
unavoidable, but there currently is one.
With dmd that's indeed the case. But LLVM is good at enregistering small
structs, which is how ranges are implemented.
Andrei
