PlaceBuilder ICE, Part 2

[nikomatsakis]

#88039 fixed #87987 but this issue still fails:

#![feature(capture_disjoint_fields)]

fn foo<MsU>(handler: impl FnOnce() -> MsU + Clone + 'static) {
    Box::new(move |value| {
        (|_| handler.clone()())(value);
        None
    }) as Box<dyn Fn(i32) -> Option<i32>>;
}

fn main() {}

Found trying to migrate seed 0.8.0.

Let me know if you need any more information.

Originally posted by @ehuss in #87987 (comment)

[nikomatsakis]

@rustbot assign @roxelo

[roxelo]

The culprit is https://github.com/rust-lang/rust/blame/master/compiler/rustc_mir_build/src/build/expr/into.rs#L268

[roxelo]

@arora-aman Let me know how I can provide assistance on this one

[roxelo]

@rustbot label +A-closures +A-edition-2021 +T-compiler +I-ICE

[arora-aman]

@rustbot claim

[arora-aman]

This seems like a similar problem as #87814. We end up at a point where we fail to determine type of a variable and end up returning early without possibly capturing.

[nikomatsakis]

@rustbot release-assignment

[nikomatsakis]

@rustbot claim

[Mark-Simulacrum]

Reopening -- I was able to trigger this on rustc_errors:

thread 'rustc' panicked at 'called `Result::unwrap()` on an `Err` value: PlaceBuilder { base: Upvar { var_hir_id: HirId { owner: DefId(0:1099 ~ rustc_errors[2427]::{impl#11}::print_error_count), local_id: 4 }, closure_def_id: DefId(0:1100 ~ rustc_errors[2427]::{impl#11}::print_error_count::{closure#0}), closure_kind: FnMut }, projection: [Deref] }', compiler/rustc_mir_build/src/build/expr/as_place.rs:306:69
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

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/issues/new?labels=C-bug%2C+I-ICE%2C+T-compiler&template=ice.md

note: rustc 1.56.0-beta.1 (e784c962e 2021-09-07) running on x86_64-unknown-linux-gnu

note: compiler flags: -Z macro-backtrace -Z tls-model=initial-exec -Z unstable-options -Z binary-dep-depinfo -Z unstable-options -Z force-unstable-if-unmarked -C opt-level=3 -C embed-bitcode=no -C debuginfo=0 -C link-args=-Wl,-rpath,$ORIGIN/../lib -C prefer-dynamic --crate-type lib

note: some of the compiler flags provided by cargo are hidden

query stack during panic:
#0 [mir_built] building MIR for `<impl at compiler/rustc_errors/src/lib.rs:917:1: 1190:2>::print_error_count::{closure#0}`
#1 [unsafety_check_result] unsafety-checking `<impl at compiler/rustc_errors/src/lib.rs:917:1: 1190:2>::print_error_count::{closure#0}`

Going to try to minify...

[Mark-Simulacrum]

Triggers with the following (rustc --edition=2021), both on beta and nightly. I think this requires if let guards, though.

#![feature(if_let_guard)]

fn print_error_count(registry: &Registry) {
    |x: &Registry| {
        match &x {
            Registry if let _ = registry.try_find_description() => { }
            _ => {}
        }
    };
}

struct Registry;
impl Registry {
    pub fn try_find_description(self) {
        unimplemented!()
    }
}

fn main() {}

[roxelo]

We ICE after attempting to resolve the PlaceBuilder for registry here: https://github.com/rust-lang/rust/blob/master/compiler/rustc_mir_build/src/build/expr/into.rs#L509.
I believe we recently added the option to have if let in pattern matching arms, but we never consumed the IfLet expression when "visiting" the arm leading us to not "read" registry and hence not being able to resolve it.

cc: @arora-aman