The following comment was made by an LLM, it may be inaccurate:

Summary

No duplicate PRs found. However, PR #7043 is part of a related series:

Related PRs (Not Duplicates):

1. fix(core): add dispose functions to prevent subscription memory leaks #7032 - fix(core): add dispose functions to prevent subscription memory leaks

   • Adds cleanup methods to Share, ShareNext, Format, Plugin bus subscriptions
   • Referenced in the PR description as foundational work

2. fix: add cleanup mechanisms to prevent memory leaks #7039 - fix: add cleanup mechanisms to prevent memory leaks

   • Adds session/state cleanup methods to ACPSessionManager, Rpc timeout, FileTime, OAuth TTL
   • Also referenced as foundational work

Relationship:

PR #7043 complements #7032 and #7039 by actually wiring up the cleanup to be called when ACP connections end, rather than just adding the cleanup methods. The three PRs form a complete solution to the memory leak issues (#4315, #5363).

Conclusion: PR #7043 is the integration point that makes the previous cleanup work actually functional—not a duplicate.

Addressing Review Comments

Re: break statements in finally blocks (lines 168, 372)
Fixed - moved the break statements outside of the finally blocks. The break is still needed to prevent switch fallthrough, but placing it in finally was causing early loop termination. Also added proper catch blocks for error handling.

Re: AbortController signal not passed to subscribe()
Good point. The SDK's event.subscribe() doesn't currently support an AbortSignal parameter. However, I've added a check for both controller.signal.aborted and this.disposed at the top of the event loop, which will break out on the next iteration after abort/dispose. While this doesn't immediately close the underlying stream, it prevents further processing and allows graceful exit.

Re: disposed checks in async operations
Fixed - added a this.disposed check at the top of the event processing loop alongside the abort signal check. This ensures that after dispose() is called, no further events will be processed.

Additional fix for disposed checks:

Added more comprehensive this.disposed checks:

1. At the start of each event case (permission.asked, message.part.updated) to prevent starting new operations
2. After key await calls to prevent follow-up operations if dispose() was called during the await

This ensures that in-flight async operations will exit gracefully rather than continuing to use potentially invalid resources.

Fixed: AbortController signal now passed to subscribe()

After investigating, the SDK's event.subscribe() does support a signal option. The SSE client uses it to:

1. Check signal.aborted at the start of each retry loop
2. Register an abort handler that calls reader.cancel() to properly close the stream

Updated the code to pass { signal: controller.signal } to subscribe(), which ensures the stream is properly closed when dispose() or cleanupSession() is called.

Closing as superseded by #7032 which now includes the ACP agent cleanup functionality (dispose method, sessionAbortControllers, cleanupSession, and connection.closed handler). The session manager utility methods (remove, size, sessionIds, clear) can be added in a follow-up PR if needed.