Add a TreeMap.remove_with method.

[chris-morgan]

TreeMap has methods find, find_mut and remove, which take &K and
return Option of &V, &mut V and V; and find_with and
find_with_mut which take |&K| -> Ordering, but remove_with was
missing from this symmetry.

[chris-morgan]

Of course, it rotted twenty minutes before I pushed… updated from pop_with to remove_with.

[Gankra]

I believe this is what the Borrow stuff in collections reform is supposed to deprecate. CC @aturon

[alexcrichton]

I'm ok landing this for now as it's a nice-to-have, but if @aturon is close to landing the new Borrow trait then we may wish to hold off to not add a function and immediately deprecate it.

[chris-morgan]

I just want something that can achieve this general goal for my JSON Pointer implementation so that I can avoid allocations entirely; I don’t mind what it is, and until it comes I can just do it a less efficient way (produce the actual String and call pop).

[aturon]

@alexcrichton

I'm ok landing this for now as it's a nice-to-have, but if @aturon is close to landing the new Borrow trait then we may wish to hold off to not add a function and immediately deprecate it.

I plan to do this early next week.

[alexcrichton]

Ok! @chris-morgan would you be ok holding out on this for a week? We definitely want to support this regardless!

[chris-morgan]

Sure, I can hold on with no difficulty. My own use case only gains efficiency by doing this.

[Gankra]

@alexcrichton This trivially falls out of collection reform's design, if I recall correctly. The K taken by get and remove is of type Borrow. As such I think we can close this PR altogether.

Although honestly I'm a bit fuzzy on exactly which of the few different forms of Borrow @aturon ultimately settled on, it's 2am, and I don't have the energy to pore over the whole RFC again. Regardless, if the proposed design handles find_with, remove_with should fall out trivially.

[Gankra]

Looks like current design is punting on supporting the full _with usecase at the moment. There's some ideas to handle it, but the rest of the libraries/compiler aren't there yet. I suppose this should be merged, as an #[experimental] shim, then.

[chris-morgan]

Yes, it’s not doing quite everything this can, but the main reason one would want this is taken care of. I don’t care about this any more.

[alexcrichton]

Ok, for now I think the Q: BorrowFrom bound will satisfy this particular request. We can look into investigating *_with later.