Some tests sometimes randomly fail

[LoipesMas]

Bevy version

f969c62

What you did

Run cargo test (in crates/bevy_transform)

What went wrong

Sometimes (~10% of the time) test systems::test::correct_transforms_when_no_children or systems::test::did_propagate_command_buffer fail.

Additional information

Output:

---- systems::test::correct_transforms_when_no_children stdout ----
thread 'systems::test::correct_transforms_when_no_children' panicked at 'assertion failed: `(left == r
ight)`
  left: `1v0`,
 right: `2v0`: Malformed hierarchy. This probably means that your hierarchy has been improperly mainta
ined, or contains a cycle', crates/bevy_transform/src/systems.rs:69:9
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace


failures:
    systems::test::correct_transforms_when_no_children

Same assert fails for systems::test::did_propagate_command_buffer.
Considering it happens randomly it's probably a race-condition.

I'm on Linux, but it also happens on GitHub CI (see #4994), so I assume my system information is not relevant.

Also, during my testing it happened ~7 times for correct_transforms_when_no_children and once for systems::test::did_propagate_command_buffer.

[LoipesMas]

Some more testing:

• Running only, for example, correct_transforms_when_no_children hasn't failed.
• Running cargo test --jobs 1 can fail.

[mockersf]

git bisect seems to point to #2250 as causing the issue

[alice-i-cecile]

Ping @james7132 and @hymm; any ideas?

[hymm]

how do globals work in cargo tests? Does each test get it's own instance or are the sharing them?

[bjorn3]

They are shared as tests run in the same process on unique threads.

[hymm]

it looks like the tests don't explicitly order transform_propagate_system and parent_update, but the plugins do. Maybe they would run in the added order normally, but with #2250 the executors get starved and they can run in the reverse order sometimes.