chat-ui: reply-timeout backstop gets ref + Retry (CL-6677) - #330
Merged
Conversation
A turn that fails on the cold-wake path (a parked room re-deploying and re-deriving its inference source, PR #327) never emits a single chat.agent event, so the client's own reply-timeout backstop fires instead of the server-side undelivered-turn notice. These tests pin the fix: that backstop must mint a reportError refId and render through the same FailedTurnStrip (ref id quotable, Retry wired) as any other turn that ends with no reply, rather than a bare event line with no ref and no action.
useStreamingReply's PENDING_REPLY_CLEAR_MS backstop (a turn that opens but never gets a token or a terminal event) rendered as a bare quiet event line with no ref id and no Retry -- a weaker, separate backstop sitting beside postUndeliveredNotice's honest one for the exact same failure class. This surfaced starkly on the cold-wake path (PR #327's defer-to-wake single-step restore): a room can wake, redeploy, and re-derive its inference source without ever throwing an error the server side can catch, so every such failure fell through to this weaker client-side notice. useStreamingReply now mints a reportError refId the moment its timer fires (replyTimedOutRefId replaces the bare boolean), and appendReplyTimedOutNotice builds a turnFailed text part carrying it, attributed to the workbench's agent participant the same way mergeStreamingReply already does. That routes it through the same FailedTurnStrip every other undelivered-turn notice uses, so the person gets a quotable ref and a working Retry regardless of which backstop caught the failure. The now-dead chat.reply-timed-out event branch is removed rather than left beside the new path. Also notes, at the sendMail wake call site, that no wake-in-progress signal exists yet to shorten the ~105s of silent typing pulse a real cold wake can take before either backstop fires -- surfacing a "waking up" state needs a signal threaded through sendMail's wake, subscribeToWorkbench's stream, and useStreamingReply, which is follow-up work rather than a client-side fix.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Two independent 120s no-reply backstops existed; only the server-side
postUndeliveredNoticehad the ref+Retry treatment (CL-6308). A cold-wake failure never throws insidedispatchTurn— mail sends fine, the agent just never streams — so it was caught only by an older client-side timer that rendered a bare, ref-less notice.The client timer now reports through
reportError(real refId) and renders via the sameFailedTurnStripas every other failed turn.Known follow-up (documented at the call site, not fixed here): there is no wake-in-progress signal anywhere in the stack, so a ~105s cold wake is indistinguishable from thinking time.
0cfb2cea1a3ce433bun run checkgreen in the worktree.