feat (ui): reconnect desktop ACP sessions after sleep and connection loss - #10411
Conversation
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 9ecd99c9c9
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| } catch (error) { | ||
| if (generation !== connectionGeneration) { | ||
| throw error; | ||
| } | ||
| } |
There was a problem hiding this comment.
When the goose serve backend has exited, the get-acp-url IPC path calls GooseServeLeaseRegistry.getAcpUrl, which throws GOOSE_SERVE_EXITED_USER_MESSAGE instead of returning a URL. This catch treats that permanent error like a transient socket failure and retries forever, so after an unexpected backend exit the renderer stays in recovering and every getAcpClient() caller waits on a never-settling pendingConnection instead of surfacing the close-window error.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: c2609cb4a2
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| if (isAcpRecovering()) { | ||
| return; |
There was a problem hiding this comment.
When a session with an initialMessage/shouldStartAgent is mounted while ACP recovery is in progress, useAutoSubmit marks the prompt as submitted and clears the initial message immediately after calling handleSubmit; this new guard then silently returns before queuing or sending anything. After reconnect, the auto-submit effect will not retry because hasAutoSubmittedRef is already true, so deep-link/fork/continue flows can lose the user's prompt instead of sending it once the connection is restored.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 4924819ec6
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| if (isAcpRecovering()) { | ||
| return; |
There was a problem hiding this comment.
Fresh evidence beyond the earlier auto-submit case is that BaseChat still wires recipe activities and message-list append actions directly to handleSubmit, and those controls are not disabled by queueProcessingBlocked. When a user clicks one of those append controls while ACP recovery is active, this guard returns without adding a local message or retrying after recovery, so the prompt is silently lost; please either disable those append entry points while reconnecting or keep the submitted input pending instead of dropping it here.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
this is a valid edge case. Normal submissions are blocked and auto-submit is delayed during recovery. Recipe/MCP append actions are rare during the brief reconnect window, so we’re leaving them out of scope to avoid additional queueing complexity.
lifeizhou-ap
changed the title
lifeizhou-ap
changed the title
3 participants
Summary
Added desktop ACP recovery after laptop sleep or connection loss.
Testing
Unit testing and manual
Related issue
Fix #10368