Backport TestResubscribeWithCompletedSubscription and deadlock fix to release/3.3

[Copilot]

Backports test case and deadlock fix from PR #18562 (commit f3e42e7626) that validates Unsubscribe() doesn't deadlock when the subscription producer exits before unsubscription.

Changes

• Test case: TestResubscribeWithCompletedSubscription in p2p/event/subscription_test.go

  • Validates no deadlock when producer completes before Unsubscribe() is called

• Deadlock fix: Make resubscribeSub.unsub channel buffered (size 1)

  • Without buffer, Unsubscribe() blocks when sending after resubscribe loop exits
  • Buffer allows send to complete even when no receiver is present

// Before: unbuffered channel causes deadlock
unsub: make(chan struct{})

// After: buffered channel prevents deadlock
unsub: make(chan struct{}, 1)

The test reproduces the deadlock scenario by ensuring the subscription completes successfully before calling Unsubscribe().

Original Source

• Original PR: Sync p2p/event with geth's event #18562 - "Sync p2p/event with geth's event"
• Original Commit: f3e42e7

Original prompt

Backport the test case TestResubscribeWithCompletedSubscription from the main branch of Erigon to the branch release/3.3. This test validates that the Unsubscribe method does not result in a deadlock when the producer exits first. Reference implementation in p2p/event/subscription_test.go from commit b8d1242103d81d46591cf86c71d8e182bc8c2add. Ensure it only modifies the release/3.3 branch and includes the additional test case.

This pull request was created from Copilot chat.

---

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.

[anacrolix]

@copilot link to the original PR and commits on main

[Copilot]

@copilot link to the original PR and commits on main

Original PR: #18562
Original commit on main: f3e42e7

This backport includes both the test case and the deadlock fix from that PR.

[anacrolix]

@copilot what happened to the other changes in the commit from main? why weren't those brought over?

[Copilot]

@copilot what happened to the other changes in the commit from main? why weren't those brought over?

The original task requested backporting only the test case TestResubscribeWithCompletedSubscription. I included the deadlock fix in subscription.go because it's required for the test to pass.

The other changes from PR #18562 were in feed.go and feed_test.go. These appear to be refactoring changes that are independent of the deadlock fix. Should I include those changes as well?

[anacrolix]

@copilot do they affect correctness in other regard?

[Copilot]

@copilot do they affect correctness in other regard?

Yes, the feed.go changes fix a potential race condition in type initialization. Currently, f.etype can be set under the f.mu lock in typecheck(), which creates timing issues between concurrent Subscribe/Send calls.

PR #18562 fixes this by:

1. Using sync.Once to atomically initialize the type on first Subscribe/Send
2. Moving type checks outside the mutex lock (after initialization via Once)
3. Removing the separate typecheck() function

The feed_test.go changes are just for compatibility (TypeFor vs TypeOf).

Should I include all the feed.go and feed_test.go changes from PR #18562?

[anacrolix]

I've checked the change fixes the test, and the other changes from the original PR for main aren't required.