Tests occasionally fail (Apple M2 Max)

[SergioBenitez]

On an M2 Max MacBook Pro, I am experiencing unexpected causality issues tied to using ThreadLocal. I ran the library's tests to see if I could exhibit similar behavior, given that the iter() test in particular looks like a decent model of my real-world situation, and indeed I get occasional test failures within seconds of repeatedly running the test suite. Here are a few:

running 6 tests
test thread_id::test_thread ... ok
test tests::same_thread ... ok
test tests::is_sync ... ok
test tests::different_thread ... ok
test tests::test_drop ... FAILED
test tests::iter ... ok

failures:

---- tests::test_drop stdout ----
thread 'tests::test_drop' panicked at 'assertion failed: `(left == right)`
  left: `0`,
 right: `1`', src/lib.rs:641:9
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

running 6 tests
test tests::test_drop ... ok
test thread_id::test_thread ... ok
test tests::is_sync ... ok
test tests::same_thread ... ok
test tests::different_thread ... ok
test tests::iter ... FAILED

failures:

---- tests::iter stdout ----
thread 'tests::iter' panicked at 'assertion failed: `(left == right)`
  left: `[1, 2, 3]`,
 right: `[1]`', src/lib.rs:616:9
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

running 6 tests
test tests::is_sync ... ok
test thread_id::test_thread ... ok
test tests::test_drop ... ok
test tests::same_thread ... ok
test tests::different_thread ... ok
test tests::iter ... FAILED

failures:

---- tests::iter stdout ----
thread 'tests::iter' panicked at 'assertion failed: `(left == right)`
  left: `[1, 2, 3]`,
 right: `[1, 3]`', src/lib.rs:616:9
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

running 6 tests
test tests::is_sync ... ok
test thread_id::test_thread ... ok
test tests::test_drop ... ok
test tests::same_thread ... ok
test tests::different_thread ... ok
test tests::iter ... FAILED

failures:

---- tests::iter stdout ----
thread 'tests::iter' panicked at 'assertion failed: `(left == right)`
  left: `[1, 2, 3]`,
 right: `[2, 3]`', src/lib.rs:616:9
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

Note that exhibiting failures is much harder if you don't run the complete test suite, so I suspect the issue has something to do with threads being reused.

[Amanieu]

I can't reproduce this and it's very difficult for me to figure out what is going on without being able to see it myself. It could even be something specific to macOS.

[SergioBenitez]

I can't reproduce this

Where (on which platform) are you trying to reproduce this? I can reproduce this within seconds on my MacBook with an M2 Max chip running macOS Ventura 13.X with the following:

while cargo test --all-features --release --lib; do sleep :; done

I also tried on an M2 MacBook Air running macOS Monterrey with the same results, so this is certainly not specific to the version of macOS, though it could be macOS, or it could be the CPU (I don't have an Intel Mac to check on).

Nevertheless, this seems like a fairly serious bug.

[SergioBenitez]

One approach you could take is to use the GitHub CI to run this on macOS, if you don't have access to a Mac yourself and can't or don't want to set-up a VM. Given the above, I would imagine running the loop for something like 30s to a minute should suffice.

[xosmig]

Cannot reproduce on M2 MacBook Pro with Ventura 13.4.1.
Could this already have been fixed?

[Swatinem]

I also hit this just now, but only with a high cargo test -- --test-threads count.

I opened #68 as I believe this regression was introduced recently, and included a regression test as well.