Repository navigation
Tags
Tags
fix: open /model during a turn, drop side borders, no duplicate model (… …#1528) - /model typed while a turn runs opens the picker at once; the model picked is queued as /model <name> and switches after the turn. - Boxes keep their top and bottom rules only. Tabs are expanded before a line is measured, so read_file output no longer wraps a row and breaks the feed's border. - The status box no longer repeats the model the header shows. Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
fix: stop the full-screen view freezing on scroll (#1527) A live session pinned a core: every mouse wheel event re-rendered every feed entry (feed.truncate cut lines one rune at a time, quadratic) and re-wrapped the whole main pane, several times per event and per frame. One wheel event with a busy session took 112s; it now takes 3.9ms. - feed: rendered lines cached per width; only new entries are rendered; truncate is one ansi.Truncate pass; the feed is locked (the agent adds entries while the view reads). - screen: the main pane's wrapped rows are cached per width until its lines change. - An input filter folds wheel notches and thins motion to one sample per ~16ms before they reach the update loop; the drag release sets the final feed width. - @ completion (files and memory ids read the disk) runs in a tea.Cmd; matches for an edited prompt are dropped. Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
ci: make the Windows test jobs faster (#1526) Each Windows job spent ~4 minutes building from cold. setup-go keys its cache on go.sum alone with no fallback, so any go.sum change missed every cache, and main never saved the new key because its pushes skip the test jobs. The Windows jobs now restore the Go cache with a fallback key, and a new windows-go-cache job on main saves it once per go.sum change (or once a week). The e2e and integration jobs no longer run go build ./... (e2e builds only the binary; go test compiles what integration needs). test_chat_cmd_repl.sh and test_features_kdeps_api.sh move out of Windows e2e shard 1, which ran 460s against 196s for shard 2. Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
PreviousNext