RFC to remove some special treatment of Box by borrow checker

[nikomatsakis]

Rendered

[anasazi]

I don't understand the motivation for doing this (but then I've never understood the obsession everyone else has with treating unique pointers just like non-unique pointers) especially when the proposed design still special cases Box so that moves out are allowed.

I really just don't understand why we would want to make our type system more restrictive just because we can exploit information about Box that we can't about smart pointers. Seems like it would be going backward.

[zwarich]

As discussed in this week's Rust meeting, I went ahead and implemented this change to see what the fallout would be like. There are only two changes required in the Rust distribution (I'm not including tests, since they are designed to test all of the edge cases):

https://gist.github.com/zwarich/348d5726d1ef42f109ee

I'll finish going through tests and double-check that I implemented everything correctly, but this looks like minimal impact on any real-world code.

[zwarich]

@anasazi We can remove the special-casing for moves out of Box by doing something like #178. It doesn't appear to be possible to do the same thing in a safe fashion for the special cases discussed in this RFC.

[zwarich]

I went through and wrote a bunch of new test cases, and found one place I missed, which resulted in a few more changes: https://gist.github.com/zwarich/348d5726d1ef42f109ee

[brson]

Discussed in https://github.com/rust-lang/meeting-minutes/blob/master/weekly-meetings/2014-07-29.md. Accepted as RFC 43. Tracking: rust-lang/rust#16094.

[pnkfelix]

At some point we should document the fact that NLL has gone back to adding (more) special treatment of Box<T>. One big reason we were able to justify this is that we discovered several more instances where AST-borrowck was also special casing its treatment of Box<T>, so it was more consistent overall to just special case Box<T> consistently within NLL.

[pnkfelix]

See e.g. this snippet of conversation regarding the special treatment of Box (which was originally a concern blocking NLL from moving forward): rust-lang/rust#43234 (comment)