Merge commit 'e1cc3cbf3d9aca17248b1d9368cf1f972418c5d2' into pipelines/run-20261111-1951-material-cleanup-chat-and-graph-probe-ho-87d5

This commit is contained in:
2026-11-11 20:24:31 +00:00
80 changed files with 616 additions and 743 deletions
@@ -0,0 +1,18 @@
# Chat selector menu unification
## Change
Model and thinking-level controls now use the same `showAppChoiceMenu` root
popup/bottom-sheet primitive and shared `AppChoiceMenuItem` rows as the
permission-mode selector. Their catalog/search, loading, next-turn, server
echo, refusal, retry and dismissal behavior remain in their respective
controls. The model chooser's obsolete Cupertino parameter was removed.
## Checks
- `tool/memwatch/full.sh 'test/features/chat/conversation/chat_mode_control_test.dart test/features/chat/conversation/chat_model_picker_test.dart test/features/chat/conversation/chat_effort_control_test.dart test/features/chat/conversation/chat_effort_header_test.dart'`
— 29 passed, 0 failed, including the current-model dismissal regression.
- `tool/analyze.sh` — clean.
Browser captures were not performed in this implementation stage; UI review
owns the requested full-app captures.
+23
View File
@@ -0,0 +1,23 @@
# Chat terminal shortcut focus
Implemented the macOS Ctrl+Y floating-chat handoff. A chat shortcut now hides
the chat group only when chat is already the active window; from another
window it raises the most recently used chat. The existing window-host
post-frame lifecycle then focuses that chat's composer. Ctrl+Y now joins the
existing Option+Y and Cmd+Y aliases on Apple platforms; plain Y and repeat
events remain untouched.
Evidence:
- `tool/memwatch/full.sh 'test/features/chat/windows/chat_windows_test.dart test/features/chat/workspace/chat_float_shortcut_test.dart test/features/chat/chat_floating_pane_test.dart test/shell/windows/window_manager_test.dart'`
— PASS: 33 passed, 0 failed, 0 skipped; peak 1236 MB; 14s.
- `tool/analyze.sh` — clean; exit 0.
- The floating-pane regression uses a terminal-like `TypingSurface`, presses
macOS Ctrl+Y, verifies the chat composer has primary focus, types
`from terminal` without a click, and verifies the terminal did not receive
Y. The manager regression covers visible-behind-terminal, hidden, multiple
chats, and repeated toggles.
No browser fixture was run in this implementation stage; browser terminal
fixture support and full-app rendered evidence remain for the appropriate
review stage.
@@ -0,0 +1,14 @@
# Material chat Cupertino cleanup
- Removed dead Material-only conversation helpers and the unused Cupertino icon
metadata from the thread header; `ChatBarAction.cupertinoIcon` remains an
optional ignored compatibility parameter for callers in other chat lanes.
- Removed unused `cupertino` arguments from transcript bubbles and copy-button
construction while preserving the existing Material widgets and behavior.
- Focused UI states covered: transcript bubble color rendering, message layout,
and chat thread view/header behavior.
- Checks: `tool/memwatch/full.sh 'test/features/chat/ui/transcript/chat_bubble_color_test.dart test/features/chat/ui/transcript/chat_message_view_test.dart test/features/chat/conversation/chat_thread_view_test.dart'`
— PASS (24 passed); `tool/analyze.sh` — clean.
- No visual redesign or browser/device capture was needed: the edits remove
unused parameters and alternate icon metadata without changing rendered
Material controls, layout, focus, scrolling, or navigation behavior.
@@ -0,0 +1,16 @@
# Material-only feature surface cleanup
- Removed Cupertino imports and alternate glyph metadata from artifact viewer
actions, fullscreen zoom controls, and viewer controls.
- Made search, schedule-toolbar, and artifact fullscreen compatibility flags
optional and ignored while retaining shared adaptive API arguments required
by other lanes. Material icons are used for those required compatibility
glyph parameters.
- Focused checks: `tool/memwatch/full.sh` ran artifact actions/fullscreen and
controls, search bar layout, schedule refresh, and template layout tests: 30
passed, 0 failed.
- `tool/analyze.sh`: clean.
- No rendered behavior was intentionally changed; controls remain the existing
Material widgets and layouts.
- Remaining shared compatibility metadata/flags are outside this focused edit
and should be removed only after all callers migrate during final integration.
@@ -0,0 +1,14 @@
# Material-only terminal and chat shell cleanup
- Removed Cupertino icon imports, alternate icon metadata, and ignored host
flags from terminal and chat-shell widgets while retaining the existing
Material controls and terminal keyboard/focus/resize behavior.
- `TerminalSurface.cupertino` and `TerminalToggleButton.cupertino` remain
optional ignored compatibility parameters for the application lane's
existing `app_terminals.dart` callers; they have no rendering effect and are
ready for removal after those callers migrate.
- Focused check: `tool/memwatch/full.sh` with the nine terminal/chat-shell test
paths, 52 passed, 0 failed, 0 skipped.
- Analysis: `tool/analyze.sh` clean.
- No visual redesign or capture was performed; the changed widgets continue to
use the same Material palette, controls, layout, and semantics.
@@ -0,0 +1,25 @@
# Material startup and shell cleanup
- Consolidated application startup, onboarding, routing pages, connection
states, Search, Templates, Schedules, and the navigation shell onto Material
composition on every platform.
- Kept legacy `cupertino` constructor arguments optional/ignored where callers
still pass them; no Cupertino widgets or platform host selection remain in
the changed application paths.
- Focused watchdog command ran the mirrored application/routing/navigation and
shell tests: 38 passed and 7 failed. The seven failures were teardown
assertions for the pre-existing 15-second `LaunchUpdateChecker` timer in
`PipelinesApp`; no Material cleanup assertion failed.
- `tool/analyze.sh`: clean (exit 0).
- No rendered capture was performed; this cleanup preserves the existing
Material scaffold, bars, connection states, focus, and navigation semantics.
## Manager recovery
The navigation and template-wiring fixtures were inadvertently invoking native
update plugins. They now explicitly disable updates, while dedicated startup
update tests retain and exercise the real update-check behavior.
The focused routing, app, schedules, template-wiring, split-bar, startup-update
check and startup-update-flow tests passed: 39 passed, zero failed. No production
update behavior was changed to hide the teardown failures.