Implement deref coercions.

[eddyb]

Coercions will now attempt to autoderef as needed before reborrowing.
This includes overloaded Deref, e.g. &Rc<T> coerces to &T, and
DerefMut, e.g. &mut Vec<T> coerces to &mut [T] (in addition to &[T]).
Closes #21432.

[rust-highfive]

r? @pnkfelix

(rust_highfive has picked a reviewer for you, use r? to override)

[jdm]

💃

[pnkfelix]

@nikomatsakis I'm not sure I'm the best person to review this. Maybe @nick29581 ?

[nrc]

Do you know what change made all the coercion cruft in commit three obsolete? Do we have tests to ensure nothing has regressed here?

[nrc]

LGTM. I'd like to see a few more tests - in particular some more compile fail tests to check we don't allow anything we shouldn't. The rpass tests look OK, I think we should probably have some more to check more extreme edge cases - coercions equivalent to &**********foo, that kind of thing.

And of course we need to wait for the RFC to land

[eddyb]

The removed trait object coercions were dead code, the cases are handled by coerce_unsized.
As for the other simplifications, they're also DST fallout that never got properly unified with the common paths.

[aturon]

FYI: The RFC has been merged.

[nikomatsakis]

I'd like to see some negative tests for borrow checker failures.

For example, passing the same &mut Box<T> value for two distinct &mut T arguments.

[nikomatsakis]

Also, passing to one &T and one &mut T argument.

[nikomatsakis]

And one like this:

fn copy<T>(x: &mut T) -> &mut T { x }

fn foo<T>(x: &mut Box<T>) {
    let y = copy(x);
    let z = copy(x); // ERROR (I hope)
}

[nikomatsakis]

And one like this:

fn copy<T>(x: &T) -> &T { x }

fn foo(x: &mut Box<int>) {
    let y = copy(x); // OK
    let z = copy(x); // OK
    **x += 1; // ERROR, borrowed
}

[nikomatsakis]

@bors r+ 46696c84fd2d2459186252657bcf1e76428c5043

[nikomatsakis]

The coercion logic around unsafe pointers in particular looks strange to me, but that seems to be a pre-existing problem. e.g. this line here https://github.com/eddyb/rust/blob/67718004499d1e446eef4def35aeafc8bf4c0858/src/librustc_typeck/check/coercion.rs#L235 is strange, since it seems to only permit coercions from &T to *Unsized and not *T to *Unsized? There is also a question of transitivity. (cc @nick29581, this relates to our discussion from the other day)

[bors]

⌛ Testing commit 46696c8 with merge 40c00b4...

[bors]

💔 Test failed - auto-mac-32-opt

[eddyb]

@bors retry q_q

[bors]

☔ Merge conflict

[eddyb]

@bors r=nikomatsakis ae076e1

[bors]

⌛ Testing commit ae076e1 with merge 398ac8f...

[bors]

💔 Test failed - auto-win-32-nopt-t

[eddyb]

@bors retry

[bors]

⌛ Testing commit ae076e1 with merge e0f5980...

[bors]

☀️ Test successful - auto-linux-32-nopt-t, auto-linux-32-opt, auto-linux-64-nopt-t, auto-linux-64-opt, auto-linux-64-x-android-t, auto-mac-32-opt, auto-mac-64-nopt-t, auto-mac-64-opt, auto-win-32-nopt-t, auto-win-32-opt, auto-win-64-nopt-t, auto-win-64-opt