Destructors aren't run for detached tasks

[vojtechkral]

I noticed today that when I .detach() a task, its drop() is never run.
Is this to be expected?
If yes, this should probably be documented in Task::detach() docs...

Using smol v1.3.0.

[notgull]

Tasks spawned with smol::spawn and detached are never dropped, this should be noted in that method's documentation. On the other hand, if you spawn tasks with a normal executor they should be dropped when the executor's destructor runs.

[vojtechkral]

It looks like when I don't detach the task, the drop() is only run sometimes. I'm quite confused.

Looks like I have to .cacnel().await, then it runs the dtor for sure.

[notgull]

This is to be expected. The executor in this case is 'static and static types don't always run their destructors.

The intended solution is to create your own executor.

[vojtechkral]

Ah, okay.

I can PR the docs change for .detach() if you'd like...

[notgull]

Yes, that would be appreciated.

[halvors]

It looks like when I don't detach the task, the drop() is only run sometimes. I'm quite confused.

Looks like I have to .cacnel().await, then it runs the dtor for sure.

I have an application starting detached tasks to handle each TCP connection, and this application seems to be leaking memory as a result of that, might this be the reason?

I am a little confused to how i should handle the destruction using a custom executor? Do you have any examples on how to do that properly?

[halvors]

@notgull Is it ok that i ask this here even thought the issue is closed?

[notgull]

@notgull Is it ok that i ask this here even thought the issue is closed?

It's always okay to ask. I've just been distracted for the past week or so.

I have an application starting detached tasks to handle each TCP connection, and this application seems to be leaking memory as a result of that, might this be the reason?

If you have an application like this it's probably already time to consider moving away from smol::spawn.

[vojtechkral]

If you have an application like this it's probably already time to consider moving away from smol::spawn.

Excuse me, what should in such case be used instead?