Smart pointers instead of GC?

[Paulo Pinto]

Am 01.02.2014 05:33, schrieb Frank Bauer:
> Yes, that is what Rust calls an "owning pointer". While we are at it, we
> might as well dream of "non-owning" or "borrowed" pointers, a.k.a.
> references in Rust. They don't have an effect on memory-deallocation
> when they go out of scope, but they prevent the owning pointer they
> "borrow" from (i.e. are assigned from) from being reassigned to a
> different object as long as there are borrowing references in scope.
> There is more to owning and borrowed pointers, but I think that is the
> essence.
>
> @Andrei (who I believe is the go-to (!) guy for all things memeory
> allocation related right now):
> IIRC you mentioned that it was convenient to have the GC around for
> implementing the D language features. Would it just be a minor
> inconvenience to drop that dependency in the generated compiler output
> and replace it with new/delete or something equivalent to owning
> pointers, say over the next one or two years? Or would it be a major
> roadblock that would require too much redesign work? Maybe you could
> test the concept of owning and borrowed pointers internally for some
> compiler components before actually bringing them "to the surface" for
> us to play with, if it turns out useful. But only of course, if we could
> leave the rest of D as good as it is today.
>
> I would really like to see the GC-free requesting crowd be taken a lttle
> more seriously. Without asking them to forego D features or do manual
> C++-style memory management. Javas and .NETs garbage collectors had
> enough time to mature and are still unsatisfactory for many applications.

Just to nitpick, reference counting is also GC from CS point of view. :)