Arithmetic makes borrow checker overly restrictive

[dotdash]

UPDATE: This was fixed by #47167 but still needs a regression test to be added.

---

fn main() {
    let mut i = [1,2,3];
    i[i[0]] = 0; // Works
    i[i[0] - 1] = 0; // Is rejected
}

I had expect the - 1 not to make any difference, but the borrow checker complains with the following error:

<anon>:4:7: 4:11 error: cannot use `i[..]` because it was mutably borrowed
<anon>:4     i[i[0] - 1] = 0;
               ^~~~
<anon>:4:5: 4:6 note: borrow of `i` occurs here
<anon>:4     i[i[0] - 1] = 0;

[steveklabnik]

@rust-lang/lang, is this behavior expected? I wonder if there's something weird going on here

[nikomatsakis]

Does seem odd. I suspect that the borrow checker is (incorrectly) still locked into the days when indices were passed by reference, rather than by value.

[nikomatsakis]

triage: P-medium

[nikomatsakis]

Likely to be fixed by MIR, though.

[sanxiyn]

Still reproduces.

[nikomatsakis]

(Note: would be fixed by a MIR-based borrowck is what I meant, which is not yet done.)

[cyplo]

Still seems to be a problem on rustc 1.23.0-nightly (8b22e70 2017-10-31) - not sure if expected ?

error[E0503]: cannot use `i[..]` because it was mutably borrowed
 --> main.rs:4:7
  |
4 |     i[i[0] - 1] = 0; // Is rejected
  |     - ^^^^ use of borrowed `i`
  |     |
  |     borrow of `i` occurs here

error: aborting due to previous error

[pnkfelix]

Hmm MIR-borrowck does not seem to address this case, at least not out-of-the-box

[nikomatsakis]

The first question is to decide, I suppose, what the expected behavior is =)

[nikomatsakis]

In other words, I think I would expect consistent behavior, but I could see how this might wind up as an error in both cases.

[nikomatsakis]

Well, based on this play test case, we do not get consistent behavior.

[nikomatsakis]

I wonder if #47167 will affect this, actually.

[nikomatsakis]

Seems like #47167 fixes it.

[nikomatsakis]

It now works on nightly. Marking as E-needstest. Removing from WG-compiler-nll, since this was not specific to NLL in any way.