fix(runtime): only remove active task's Waker by DropHook

[Paraworker]

This PR fixes a rare issue when running Scheduler::clear().

Currently, Scheduler::clear() is to first drain active_tasks and wake their Wakers, and then drop all futures. However, a future might spawn a new task during its drop. In that case, a new Waker gets inserted into active_tasks, but its index might still be used by another task due to active_tasks having been drained and the index being reused.

As a result, the drop hook of the original task could remove a Waker that does not belong to it, which is not the behavior we expect.

[Copilot]

Pull Request Overview

This PR refactors the scheduler's task cleanup mechanism to improve efficiency and correctness. Instead of draining and consuming wakers during cleanup, the code now uses non-consuming iterations and relies on drop hooks to properly clean up waker references.

• Changed the clear() method to use wake_by_ref() with iteration instead of wake() with drain
• Replaced try_remove() with remove() in the drop hook
• Improved documentation comments for clarity

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Berrysoft]

Look familiar? Things like this (TOCTOU) are happening all over the galaxy (really), right now!

[Berrysoft]

Could you add a test for this fix?

[Paraworker]

Could you add a test for this fix?

tbh, I don’t know how to write this test case, because the previous situation did not cause a memory leak or a panic, It was just a sudden realization that the logic wasn’t entirely correct.

[Berrysoft]

Almost LGTM. @George-Miao to review it too.

[Berrysoft]

I don’t know how to write this test case

How about this: add two tasks to the runtime, and the second one panics after being woke. If there's a bug, the second one might be removed and never panics.

[Paraworker]

How about this: add two tasks to the runtime, and the second one panics after being woke. If there's a bug, the second one might be removed and never panics.

Not fully understand, in Scheduler::clear() the futures are not polled, only dropped. waking a Waker only pushes it to the queue, so it should never panic.

[Berrysoft]

You're right. Let's wait for another reviewer:)

[George-Miao]

Exact problem with slab I described in the first PR lol

[Berrysoft]

Could you bump the version so that I can publish a fix release?

[Paraworker]

Could you bump the version so that I can publish a fix release?

Done, version bumped.

[Berrysoft]

LGTM