`retain` improvements

[eira-fransham]

Firstly, I believe it would be sound for the function passed to retain to take a mutable reference instead of an immutable one. There is no real reason to restrict this. Additionally, I've found it useful to return the removed array as an &mut [T] - since all the elements are at the end anyway this is essentially free. In my case I only allow this when the elements are Copy (as a hack around not being able to specify !Drop), but it would be sound to allow Drop types. The issue is that it would unconditionally leak, which would be solved by an additional type that derefences to &mut [T] but that drops its elements on drop. It would have to take a lifetime parameter to disallow mutating the vector after returning the elements.

[ExpHP]

The deficiency of retain supplying &T was discussed a while back, but nothing solid came of it.

[CryZe]

There's drain_filter on nightly which is very similar (except the condition is flipped).

[eira-fransham]

I think drain_filter basically does exactly what I'm talking about - returning removed elements and taking &mut T. I think it might be better to return a slice-esque type that prevents mutating the vec while it's still alive though, since it's strictly more expressive. To get the behaviour of drain_filter you can then just do .into_iter(). I think it might be useful to have a slice type that owns its elements but not their storage, anyway.

[Centril]

Now that https://doc.rust-lang.org/nightly/std/vec/struct.Vec.html#method.drain_filter is stable it seems that this is sufficiently fulfilled to close.

[RReverser]

Now that doc.rust-lang.org/nightly/std/vec/struct.Vec.html#method.drain_filter is stable it seems that this is sufficiently fulfilled to close.

Maybe I misunderstand what "stable" meant in this context, but ~1.5 years later it seems to be still unstable / nightly-only?

[kennytm]

Even if drain_filter is still unstable, discussion should be in rust-lang/rust#43244 rather than this issue.