Repository navigation
qt + tui: back-to-bottom hides, full-width sidebar replaces the welcome, no send gap, one menu owner per element, native Android picker - #2133
Merged
Conversation
…ner, bus, buses and hints Tristan, 2026-11-18: scrap the welcome screen and replace it with the left bar (channels, threads, search) fully expanded, with the comms logo; on the first tab it becomes the bar, with the theme's extra bar; at the level both frontends inherit. - SidebarLayout owns it: `holds_alone(identity)` is the sole role (the comms sidebar) while the welcome mode is selected, read from the application's selected view (no copy). Such a bar is open whatever its kept fold; SidebarChromeArrangement gives it the whole viewport and the content none; its surface parts are all content (no strip, grip or controls). - Both frontends show the welcome's parts in that sidebar: Toad's banner and the bus or project above the search, the buses (the shared BusListing) and the hints below the rows; Qt also its add-a-machine field. They hide when a tab is shown. - One New Session row: `NewSessionControl.action` starts this launch's session while no tab is shown (as the welcome's button did) and a thread in the selected project beside one. Textual's `n` and ctrl+g answer in the sidebar while it holds the window. - The welcome mode stays the home view and navigation origin and draws nothing. Deleted: the welcome's entries, sections, listings, button, its Qt page content, the shared WelcomeListing family, WelcomeContent/ListedModel/ListedFailure and SidebarChange's apply_welcome. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01RnH6dT7wdh7i4tjJZQdFu3
… hints, controls restored claude-a's review of the 412x915 grab: the full sidebar lists every thread, so buses under the rows were far below the fold (the welcome had capped threads at 8); and a row cut by the tree's edge read as running under the hints. - The buses sit under the bus/project line, above the search and rows, on both frontends: Qt's own `QtSidebarBuses(BusListing, ConversationRows)` sized to its rows (up to six), Textual's `SidebarBuses` moved from the tail to the lead. The comms tree lists only the model's rows again. - A loose gap separates Qt's rows from the hints. - The extent slider and placement actions come back as the reader left them (`SidebarSurface.controls_wanted`) once the bar stops holding the window. - `sole_sidebar` asserts nothing visible intersects the tree's viewport and that a bus line is inside the window unscrolled; the TUI tests assert the same. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01RnH6dT7wdh7i4tjJZQdFu3
… file dialog hears the answer On the phone Attach showed Qt's widget file dialog, a desktop dialog that lists only the app's own directories (Tristan: "it shows the same file picker as if it was on desktop ... bring up the native file picker: is it a photo, is it a camera, or is it a file"). Two owners were in the way. pyqt-reactive's theme turned native dialogs off on every platform (PyQT-reactive#66 keeps them on phones). And python-for-android's PythonActivity overrides onActivityResult without calling super, so Qt's receivers - its native file dialog's - never heard a result: QtActivityResults hands each result to Qt's package-private entry, as QtActivityBase does, and ToadActivity calls it. ToadActivity offers Qt's document request to AttachChooser: ACTION_GET_CONTENT for the same types in Android's chooser (Files, Photos, Drive), with a camera capture into a new MediaStore photo (Pictures/Toad) as its first choice. Either answer reaches Qt's dialog as a content URI, which file_pickers.read_picked already reads. Desktops keep the themed dialog; WindowFilePicker stays the one choice. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01RnH6dT7wdh7i4tjJZQdFu3
…to-bottom bar hides after a jump Tristan, Android, 2026-11-18: "the back to bottom button still doesn't disappear when it brings you to bottom, you need to scroll again." Each frontend's window carried its own `anchor`: Textual's counted the move, anchored, then showed or hid the jump bar; Qt's copy had lost the last step, and the jump's own scroll runs inside layout_scroll, whose offsets the scroll handler skips, so nothing hid the bar until the reader scrolled. `FollowsTail` (history_jump) now owns the step once: count a changed follow intent as a reader move, apply it through the window's `apply_anchor`, then show or hide the bar. Both windows list it before their toolkit's scroll base, so its `anchor` answers ahead of Textual's `Widget.anchor`; each window keeps only its toolkit's move. A subclass that defines `anchor` again is refused when its class is created, so the copies cannot come back. The abstract `anchor` declared on reader_position.ReaderWindow is gone with them. `jump_back` drives it: scroll a screen above the end, tap the bar, touch nothing else. On main the view is at the end with the bar still shown; here the bar is gone. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01RnH6dT7wdh7i4tjJZQdFu3
…state a sole bar shows At start Qt keeps the comms bar's fold from the saved `sidebar.hide` setting while no tab is shown; `set_collapsed` compared against `is_collapsed`, which a bar holding the window answers "open", so an open fold was never stored and the first tab folded the bar to its strip. `kept_collapsed` is the stored fold that `set_collapsed` and `toggle_collapsed` read and write; `set_collapsed` reports whether what the bar shows changed. The hold drive caught it (the sidebar was down where main had it up). The drive's `listed()` no longer reaches `self.rows`, which HistoryScenario names for its own rows. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01RnH6dT7wdh7i4tjJZQdFu3
… draws it Tristan, 2026-11-18: "the queue shows up and then the message gets submitted and the queue disappears, but then there's a gap where I don't see the queue and I don't see the message on the screen". Three frames went missing, all in the shared submission owner: - On Core's answer the delivery preview dropped the message (`DeliveryOutcome.submitting` is empty), though the execution stays until its view draws it. `SubmittingPhase.previewed` is what the preview lists; a `SentMessage` keeps its text there until it is drawn. Send now still reads `delivering`, so a message Core has is not offered again. - "Drawn" meant paged into range: the preview let go in the frame before a following window scrolled to the new row. `DrawsBusMessages.draws` now asks the window where the message's first row is (`message_top`, from the burst stack) and counts it shown above the bottom edge; one scrolled past the top still counts. The shared `repin_header` (every scroll offset and committed layout) tells its histories, and a `ConversationPager` settles its view's submissions then. - Send cleared the editor and the message was listed a worker turn later. `ConversationSubmissions.received` takes it in the composer's call: a message is begun and published there (`AgentInputSubmission.started`); text that may name a command still waits for the catalog (`CommandCatalog.may_name_command`). Qt's composer draws that preview row in the same frame as the cleared editor. Measured: Qt drive `send_gap` (#scratch, 10 sends, CPU held to 45%) had 5-13 sampled frames per send in neither place before, 0 at every 16 ms sample and every window repaint after; the TUI test samples every Textual frame: 3-6 per send on main, 0 after. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01RnH6dT7wdh7i4tjJZQdFu3
…ent; toolkits only draw it Tristan: "The hold to right click is parallel slop to the toad TUI, the options aren't the same" and "they shouldn't EVEN BE ABLE TO DRIFT APART". Qt's sidebar line offered a hand-rolled Open / Copy name while Textual's offers its target's commands; message menus were assembled twice; a tab had a menu only in Qt. - TargetLine.menu_entries (client/sidebar_lines.py): a channel or thread line's menu, its target's commands (CommandCatalog.pointer_menu) in its tree's context. Textual CommsRow and Qt ConversationRow inherit it; Qt's Open / Copy name override is deleted. - RelationshipLine.menu_entries: a Comms relationship's Open / Copy name. Textual RelationshipRow and the new Qt QtRelationshipItem inherit it (Qt now copies the same entry_target name Textual does). - TabElement.menu_entries (session_navigation.py): SessionNavigation.tab_menu. Textual SessionLabel answers a right-click with it; Qt's QtTabAt draws it. - BlockMenu.block_commands: one assembly for a block's menu (Menu and QtMenu); Qt's _block_menu_commands is deleted; MaximizeBlockCommand reads a declared allow_maximize instead of probing getattr. - Qt builds Textual's own context: QtWorkspaceViews.navigation_owner derives from the selected page (conversation pages are NavigationOwners, home is the shared HomeNavigation that WelcomeScreen also inherits), FindsOpenViews is shared by both sidebars, Qt's Sidebar answers `app`. - ActionMenu.offer: the one Textual renderer of declared entries. - tests/model/test_menu_owners.py fails when a toolkit calls a menu's ingredients itself, defines `menu_entries`, spells a declared entry, or its element stops inheriting the declaring parent; on main it fails on exactly the parallel lists this removes. - qt_drive menu_parity: each element's drawn Qt menu equals its declared entries; tests/textual/test_menu_parity.py checks the same in Textual. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01RnH6dT7wdh7i4tjJZQdFu3
…tab's label is found by its mode The menu_parity drive connected blockMenuRequested(fragment, pressed) to list.append, and the TUI test looked the selected tab's label up by a DOM id SessionLabel does not carry. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01RnH6dT7wdh7i4tjJZQdFu3
…urrent screen's tab strip The home view is no tab once the sidebar holds the window, and a session's tab strip is on its own screen, so the test opens the thread and reads the selected tab where test_channel_tab_unread does. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01RnH6dT7wdh7i4tjJZQdFu3
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
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.
Five branches, integrated and gated together so that the gated head is the merged head. It supersedes #2132.
What's in it
follows-tail): one sharedFollowsTail.anchorowns joining and leaving the tail. It counts the move, applies it through each window'sapply_anchor, then shows or hides the jump bar. Qt's copy had lost the last step, so the bar stayed up after a jump until the reader scrolled. A subclass that redefinesanchoris refused when its class is created, so the copies can't come back.welcome-sidebar): with no tab open,SidebarLayoutgives the comms sidebar the whole window: banner, bus line, buses on the first screen, search, channels and threads, hints. On the first tab it returns to its placement beside the conversation. The welcome page's own content is deleted in both front ends; the WELCOME mode stays as the navigation origin.send-gap): a sent message is on screen at every frame from Send until the history draws it. There were three gaps in the shared submission owner: the preview dropped the message at Core's answer, "drawn" meant paged rather than scrolled into view, and listing waited for a worker turn. Measured blank frames per send: Qt 5–13 → 0, TUI 3–6 → 0.menu-parity): each element's context menu is declared once on its shared parent, and both toolkits only draw it. Qt's sidebar lines now offer the target's commands (Archive, Copy name, pin, Mark view read…) rather than a hand-written Open / Copy name. TUI tabs answer a right-click with the shared tab menu. Message menus are assembled in one place.tests/model/test_menu_owners.pyfails on any parallel list; it fails on main.QtActivityResultsforwards activity results to Qt, which python-for-android swallowed. PyQT-reactive#66 (merged) keeps native dialogs on Android.Integration gate (benchmarks, not suites)
Runtime core-11a019b19, Pixel 8a profile, live bus, each drive on its own config and state, every heavy step through a 2-slot lock.
check.sh: exit 0, 0 fatal or traceback lines. The realtoad.jsonwas untouched.🤖 Generated with Claude Code
https://claude.ai/code/session_01RnH6dT7wdh7i4tjJZQdFu3