On Tuesday, September 12, 2017 13:47:47 Azi Hassan via Digitalmars-d-learn wrote: > On Tuesday, 12 September 2017 at 01:13:29 UTC, Hasen Judy wrote: > > Now, a lot of library functions seem to expect ranges as inputs > > and return ranges as output. > > Unless I'm mistaken, it was done on purpose to reduce the amount > of memory allocations in the standard library so that it becomes > @nogc-friendly. But yes, you can use std.array.array to turn it > into an array, but keep in mind that it does allocate so you > might need to watch out if you're dealing with large CSV files.
Ranges were a big thing in Phobos years before @nogc came along. They're used because they're an extremely powerful idiom. I don't even recall much in the way of discussions related to their efficiency vs arrays early on. That did come later, and ranges have significantly helped the memory efficiency of D programs regardless of @nogc, but that really wasn't the motivating factor. And even now, a lot of the basic uses of ranges _do_ allocate, because they use lambdas that end up with closures being allocated for them (which uses far less memory than allocating a dynamic array but is not @nogc). That can be avoided by doing stuff like using static nested functions or functors, but your average rangd-based code doesn't tend to jump through those hoops. Ranges do take some getting used to, but we have them precisely because they're so flexible and powerful. - Jonathan M Davis
