Call channel initializer when using async methods on pipe bootstrap #3309
Conversation
Motivation: For our async initializers on bootstraps, it's imortant that _both_ channel initializers are called, not only the one that comes as an argument on the bind function. Modifications: Fix the takeOwnershipOfDescriptor async functions to call the initializer from the bootstrap, and add regression tests. Result: Better behaved code.
…pple#3309) Motivation: For our async initializers on bootstraps, it's imortant that _both_ channel initializers are called, not only the one that comes as an argument on the bind function. Modifications: Fix the takeOwnershipOfDescriptor async functions to call the initializer from the bootstrap, and add regression tests. Result: Better behaved code.
|
@Lukasa I'm hitting a precondition failure after bumping from nio 2.84.0 to 2.85.0 and I believe it is due to this change. I'm adding a I'm happy to submit a reproducer but curious on your thoughts first and what the fix would be. I might be misusing the API to begin with. I'm able to work around it by initializing the channel after taking ownership: Broken (hits precondition): Work around: |
|
Ah, yeah, I see the issue. This new API surface is frustratingly under-tested. |
Motivation: In apple#3309 I fixed an issue where the channel initializer would not be called in some PipeBootstrap paths. Unfortunately, that introduced a new bug where the initializer was now called twice on some other paths. Modifications: Get the number of calls back to 1 on the non-async paths. Add some tests Result: Better pipe bootstrap behaviour.
|
See #3330. |
Motivation: In #3309 I fixed an issue where the channel initializer would not be called in some PipeBootstrap paths. Unfortunately, that introduced a new bug where the initializer was now called twice on some other paths. Modifications: Get the number of calls back to 1 on the non-async paths. Add some tests Result: Better pipe bootstrap behaviour.
Motivation:
For our async initializers on bootstraps, it's imortant that both channel initializers are called, not only the one that comes as an argument on the bind function.
Modifications:
Fix the takeOwnershipOfDescriptor async functions to call the initializer from the bootstrap, and add regression tests.
Result:
Better behaved code.