fix: port fix discard all message on receiver droped

[driftluo]

fix #1102

I think this fix should be done on the crossbeam channel

[taiki-e]

Thanks for the PR. It seems that Miri and sanitizer reports memory leak:
https://github.com/crossbeam-rs/crossbeam/actions/runs/12221429705/job/34090460906?pr=1121#step:4:168
https://github.com/crossbeam-rs/crossbeam/actions/runs/12221429705/job/34090462020?pr=1121#step:5:186

[driftluo]

Thanks for the PR. It seems that Miri and sanitizer reports memory leak: https://github.com/crossbeam-rs/crossbeam/actions/runs/12221429705/job/34090460906?pr=1121#step:4:168 https://github.com/crossbeam-rs/crossbeam/actions/runs/12221429705/job/34090462020?pr=1121#step:5:186

The memory leak reported by miri is the panic_on_drop test case, which seems to be the expected behavior, because https://github.com/rust-lang/rust/blob/master/library/std/src/sync/mpmc/array.rs#L487-L489 The comment here indicates that if there is a panic, a memory leak will occur

I think std seems to assume that this behavior is expected. Do we need to make some extra efforts for this?

If necessary, should panic::catch_unwind be used to catch the assume_init_drop? can't do this

[ryoqun]

@taiki-e hi, is there some timeline for merging this bug-fix pr? is there some blocker? my pr #1114 is touching same area of code and i'd rather to rework my pr once after this pr landed...

[taiki-e]

@driftluo

The memory leak reported by miri is the panic_on_drop test case, which seems to be the expected behavior, because https://github.com/rust-lang/rust/blob/master/library/std/src/sync/mpmc/array.rs#L487-L489 The comment here indicates that if there is a panic, a memory leak will occur

I think std seems to assume that this behavior is expected. Do we need to make some extra efforts for this?

Assuming this PR is based on the rust-lang/rust PR I reviewed, the implementation should be fine with this. (But if so, please make that clear when submitting a patch based on someone else's work.)

That said, we cannot merge PRs that have not passed CI, so you will need to adjust the failing test. (Perhaps a similar way to 2d3430f 's MIRI_LEAK_CHECK=1 approach is fine.)

[driftluo]

@taiki-e I use -Zmiri-ignore-leaks to disable Miri test on the channel test, but on sanitizer test, I can't find any flag to ignore it

[driftluo]

@taiki-e ok, I ignore panic_on_drop test on sanitizer, now ci is pass