Repository navigation
Conversation
pi emits a successful auto_retry_end only at the retry's first complete assistant message, so the banner said "Retrying" through the whole retried reply. pi's TUI drops its retry indicator when that run starts; clear it on the run's agent_start the same way. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Opening a session's event stream is what resumes a wrapper the idle timeout reaped, so the mount's state read often finds no runtime and leaves the context percentage blank. Since #1058 a run fills it in after its first reply; read it on `connected` too, through the same ordered usage read, so it shows as soon as the runtime is back. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…#877) The start of the tokens-per-second measurement lived in the message view, which a session switch remounts. Back on a running session, the new view took its start at the remount and divided every token streamed so far by half a second, showing thousands of t/s. Keep the start per reply (provider, model and request timestamp) outside the view; a reply joined with no record (a reload) times only the tokens that follow. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The window keeps as many rows as messages are loaded, but a turn's answer and its thinking render as two rows, so a thinking-heavy session hides its oldest loaded rows behind "Load earlier messages". Since #587 the sentinel only fetched from the server, so with no older page there, reaching it did nothing. Widen the window in that case again, renew the observer when it grows, and observe the sentinel whenever it mounts, not only when the history cursor changes. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This branch has not been deployed
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.
Four client-state fixes in
useAgentSession/ChatWindow/MessageView, one commit each (kept in one PR because each adds a note to the same part ofdocs/agents/sessions.md). Each new test fails on main.#988: retry banner stays up through the retried reply
pi 1.1 emits a successful
auto_retry_endonly with the retry's first complete assistant message (agent-session.js~777), so "Retrying (n/max)…" stayed up for the whole retried reply. pi's TUI drops its retry indicator when that run starts. Cleared on the retried run'sagent_start. Test:hooks/retry-banner.test.mjsreplays pi's event order.#1031: context % blank after opening an idle-reaped session
Gap A (busy runs never refreshed it) was already fixed by 9182fdf (#1058). Gap B remained: opening the event stream is what resumes a reaped wrapper, so the mount's state read can come back
running:falseand nothing re-read usage afterconnected. Theconnectedhandler now calls the existingrefreshContextUsage()(one extra GET per stream connect). Test:hooks/context-usage.test.mjs.#877: t/s wrong after switching sessions and back
The tok/s start lived in
MessageView; switching back remounts the chat, so the clock restarted and all tokens streamed so far were divided by ~0.5 s (thousands of t/s). Newlib/stream-token-rate.tskeeps the start per reply (provider, model, request timestamp), capped at 32 replies. After a reload mid-reply only the tokens that follow are counted. Test:lib/stream-token-rate.test.mjs.#896 (part 1): scrolling up to load earlier messages sometimes does nothing
The render window holds as many rows as loaded messages, but a turn with thinking renders more rows than messages, so the oldest loaded rows stay hidden behind "Load earlier messages". Since #587 the sentinel only fetched from the server (
if (!hasEarlierMessages) return), so with nothing older on the server, reaching the top did nothing. Now it widens the window in that case (reusinggetNextVisibleCount()), the observer is renewed when the window grows, and the sentinel is held in state so it is observed whenever it mounts. Test:components/ChatWindow.lazy-load.test.mjs; one #587 source assertion inhooks/useAgentSession.test.mjsupdated.Not covered: the live thinking-block timers still reset on switching back; the extension status bar / widgets / system prompt can miss the same resume race as #1031; #896's loading indicator and minimap (#791).
Checks:
tsc --noEmit, eslint on changed files,npm testpass. Not checked in a browser.Closes #988
Closes #1031
Closes #877
Refs #896
🤖 Generated with Claude Code