Map Purity

[via Digitalmars-d-learn]

On Sunday, 28 June 2015 at 15:55:51 UTC, jmh530 wrote:
> My understanding of pure is that a function labeled pure can 
> only include pure functions. I've been confused by the fact 
> that a function calling map (like below) can be labeled pure 
> without any problems. The only way I can rationalize it is that 
> map really isn't a function, it's (if I'm understanding it 
> correctly) a template that creates a template function that 
> calls a struct template.
>
> auto test_map(T)(T x) pure
> {
> 	return x.map!(a => a + a);
> }
>
> This is related to
> http://forum.dlang.org/thread/ppmokalxdgtszzllzorb@forum.dlang.org?page=1
> but maybe my question is more basic.

There are two aspects to that. First, purity in D is less strict 
than the concept in functional programming languages. It allows 
access (even mutation) to anything that is reachable through the 
parameters. David Nadlinger has written a nice article about the 
concept and its implications:

http://klickverbot.at/blog/2012/05/purity-in-d/

Secondly, `map` is indeed a template function, as you write. For 
templates functions, the compiler infers many properties, 
including purity. Neither the `map` function nor the constructor 
of the struct it returns do anything impure, therefore they are 
treated as pure. And indeed, the return value of `test_map` will 
only depend on its arguments, which is the first requirement for 
purity, and of course it doesn't change global state either, 
which fulfills the second requirement.