MIR-borrowck: ICE during diagnostic search for assignment (unwrap of None)

[pnkfelix]

PR #44806 added some code to search for the lexically first assignment in order to report an illegal assignment.

However, oddly enough, for some tests this search, which ends in an unwrap, is causing the compiler to ICE.

For example:

%  build/x86_64-apple-darwin/stage1/bin/rustc  -Z borrowck-mir \
     ../src/test/compile-fail/borrowck/borrowck-for-loop-head-linkage.rs
[...]
error[E0384]: re-assignment of immutable variable `val` (Mir)
  --> ../src/test/compile-fail/borrowck/borrowck-for-loop-head-linkage.rs:15:5
   |
15 |        for &x in &vector {
   |   _____^
   |  |_____|
   | ||
16 | ||         let cap = vector.capacity();
17 | ||         vector.extend(repeat(0));      //~ ERROR cannot borrow
18 | ||         vector[1] = 5;   //~ ERROR cannot borrow
19 | ||     }
   | ||     ^
   | ||_____|
   | |______re-assignment of immutable variable
   |        first assignment to `val`

error[E0384]: re-assignment of immutable variable `x` (Mir)
  --> ../src/test/compile-fail/borrowck/borrowck-for-loop-head-linkage.rs:15:10
   |
15 |     for &x in &vector {
   |          ^
   |          |
   |          re-assignment of immutable variable
   |          first assignment to `x`

error: internal compiler error: unexpected panic

note: the compiler unexpectedly panicked. this is a bug.

note: we would appreciate a bug report: https://github.com/rust-lang/rust/blob/master/CONTRIBUTING.md#bug-reports

note: rustc 1.22.0-dev running on x86_64-apple-darwin

thread 'rustc' panicked at 'called `Option::unwrap()` on a `None` value', /Users/fklock/Dev/Mozilla/rust.git/src/libcore/option.rs:335:20

[pnkfelix]

Its worth noting that the diagnostic reported by AST-borrowck is very different from that for MIR-borrowck. This is almost certainly because the MIR-borrowck is seeing a desugared for-loop (and thus we're seeing reports of an assignment to val that does not actually occur in the surface level syntax.

Nonetheless, part of resolving this bug might include trying figure out how to make the MIR-borrowck diagnostic more understandable to the end user in this scenario.

[pnkfelix]

(another approach that I did not consider earlier: Change the MIR construction to accommodate for-loops. E.g. maybe the MIR assignment statement could take a flag indicating that it is an assignment arising from for-desugaring, and thus it is "safe" even when it is reassigning a variable declared without mut...)

[matthewjasper]

Working on this (and this error message in general)

[nikomatsakis]

@matthewjasper nice! please stop in on gitter with any questions, if you've not done so already.

[matthewjasper]

Minified case (that still ICEs after #45936)

fn main() {
    let x = String::new();
    x = String::new();
}