Compilation regression on nightly with `i686-unknown-linux-gnu`

[carllerche]

An existing Tokio test started to fail with a stack overflow when updating nightly on target i686-unknown-linux-gnu

https://github.com/tokio-rs/tokio/actions/runs/5029230797/jobs/9020678779?pr=5706

Code

This test started failing on the latest nightly:

https://github.com/tokio-rs/tokio/blob/master/tokio/src/sync/mpsc/block.rs#L433-L448

To run the test, from the tokio directory in the Tokio repo:

cross +nightly test --all-features --target i686-unknown-linux-gnu --lib mpsc::block -- --nocapture

Version it worked on

Version 1.69

Version with regression

rustc 1.71.0-nightly (d3f416d 2023-05-18)

[carllerche]

The overflow happens here: https://github.com/tokio-rs/tokio/blob/master/tokio/src/sync/mpsc/block.rs#L415

if_loom! expands to:

const LOOM: bool = false;

if LOOM {
    $($t)*
}

[Noratrieb]

probably because of const prop, @cjgillot @oli-obk
#109900

[apiraino]

WG-prioritization assigning priority (Zulip discussion).

@rustbot label -I-prioritize +P-high +T-compiler

[oli-obk]

I'm assuming this only happens in debug mode. As mentioned by Nils, #109900 caused some functions to use more stack memory in debug mode due to not eliminating some dead code anymore.