[mir-opt] Fix Inline pass to handle inlining into box expressions

[wesleywiser]

r? @oli-obk

Before, the test case just ICE'd here:

rust/src/librustc_mir/transform/inline.rs

Line 668 in a605441

ref place => bug!("Return place is {:?}, not local", place),

[oli-obk]

Would x.field = foo(); have hit the same ICE before?

[wesleywiser]

Would x.field = foo(); have hit the same ICE before?

I tried this

#[inline]
fn bar() -> i32 {
    3
}

fn main() {
    let mut foo = (1, 2);
    foo.1 = bar();
}

and it doesn't trigger the issue because we assign the return of bar() to a temporary and then assign that temporary to foo.1 after the call. This ICE occurs specifically when the destination contains a projection.

[oli-obk]

This ICE occurs specifically when the destination contains a projection.

So... let x = (42, foo());? Or how is box special in this case?

[wesleywiser]

box is special because it causes dest_needs_borrow() to return true. This triggers the reborrow and means that when the Integrator runs, destination has a projection. For let x = (42, foo());, destination is a plain Local.

[oli-obk]

@bors r+

[bors]

📌 Commit b50cee3 has been approved by oli-obk

[bors]

🌲 The tree is currently closed for pull requests below priority 1000, this pull request will be tested once the tree is reopened

[Centril]

Failed in #67448 (comment), @bors r-

[wesleywiser]

Rebased and modified new test to skip wasm which aborts instead of unwinding.

@bors r=oli-obk

[bors]

📌 Commit f1325a7 has been approved by oli-obk

[bors]

🌲 The tree is currently closed for pull requests below priority 100, this pull request will be tested once the tree is reopened