Nested method calls with `&mut` receivers result in borrowck errors

[nikomatsakis]

This is challenging to resolve. I thought I could fit into the new borrow checker, but it is worth addressing separately since there are a number of subtle issues. What this really amounts to is that you need to support borrows where the lifetime of the borrow pointer does not encompass the borrow itself. To handle this safely, we wind up with two loans, the primary one and then a secondary, aliasing loan that tracks the fact that a pointer has been created, even if the memory it points at cannot be dereferenced yet.

Part of #5074

[nikomatsakis]

There are various FIXMEs in the code with commented out bits of code related to this issue.

[Dretch]

Here is a test case (at least I think it relates to this issue):

struct MyStruct {x: int}

impl MyStruct {

    fn takes_mut_self(&mut self, _x:int) {}

    fn takes_self(&self) -> int { 42 }

    fn also_takes_mut_self(&mut self) {
        self.takes_mut_self(self.takes_self());
    }

}

fn main() {}

The error rustc produces is:

mut-self-issues.rs:12:28: 12:33 error: cannot borrow `*self` as immutable because it is also borrowed as mutable
mut-self-issues.rs:12         self.takes_mut_self(self.takes_self());
                                                  ^~~~~
mut-self-issues.rs:12:8: 12:13 note: second borrow of `*self` occurs here
mut-self-issues.rs:12         self.takes_mut_self(self.takes_self());
                              ^~~~~
error: aborting due to previous error

[nikomatsakis]

Nominated for feature complete (not backwards compatible since it generates strictly more errors than it should)

[catamorphism]

accepted for feature-complete

[pnkfelix]

This can be added backwards-compatibly. Its a nice enhancement but not known to be a major problem to work-around.

P-low, not 1.0.

[pnkfelix]

By the way, the reason that this does not type check under our current model is provided in the internal docs for the type checker; see: typeck/infer/region_inference/doc.rs lines 171 through 262.

[cmr]

Triage bump; still an issue.

[lilyball]

cc me, I still hit this with surprising regularity.

[nikomatsakis]

Closing in favor of rust-lang/rfcs#811

[dcsommer]

Seems to be a blocker for some async IO patterns needed for lowest-latency processing. cc me.