ch-ch-changes

[dsimcha]

== Quote from dsimcha (dsimcha at yahoo.com)'s article
> == Quote from Andrei Alexandrescu (SeeWebsiteForEmail at erdani.org)'s
> > Lest a misunderstanding arises, the code:
> > auto r = map!((float a) { return fabs(a); })(a);
> > does *not* involve a delegate; it's still a direct call. This code
> > doesn't currently compile because of a bug in the compiler, but this
> > equivalent code does:
> > float fun(float a) { return fabs(a); }
> > auto r = map!(fun)(a);
> > and again it does not involve a delegate.
> But, for the truly performance obsessed, it's still a non-static nested function
> that doesn't access any local variables in the outer scope.  Isn't it still taking
> a pointer to the outer scope as an implicit argument, kind of like a this pointer,
> or are real-world compilers sufficiently smart to optimize that out if you never
> use any of the variables in the outer scope?
> Yes, I know this is nitpicking, but I'm asking primarily out of curiosity rather
> than out of practical concern.  I also realize that, if you really care that much,
> you can declare fun() as static and get rid of the implicit pointer to the outer
> scope.  This is kind of ugly, but I guess code that's this micro-optimized is
> going to be ugly any way you cut it.

To answer my own question, I benchmarked this using a sorting function, and it
doesn't matter in practice, probably because DMD is smart enough to inline simple
nested functions.  As a positive control (to make sure the benchmark is valid),
when I explicitly made the alias a function _pointer_ instead of a function
_name_, this made a fairly large difference.