* Separate out SDKContext
...from SDKContextClass, as the comments hopefully explain.
Also make the captitalisation consistent.
* Adding the new file usually helps too
* Fix import
* More comment
* Update the various imports of SDKContextClass
* Also fix case of TestSDKContext
* Update the mock
* fix(room-list): instantiate message previewers lazily
Removing the unused SettingsStore import from MessagePreviewStore (when
the
feature_new_room_list flag is dropped) changed module load order and
exposed a latent circular dependency: ReactionEventPreview imports
MessagePreviewStore, which eagerly did `new ReactionEventPreview()` at
module-eval — so importing ReactionEventPreview first (as its unit test
does) hit "ReactionEventPreview is not a constructor".
Construct the previewers lazily on first use (cached) instead of at
module
load, so nothing dereferences a mid-evaluation module. Fixes
ReactionEventPreview-test.
* refactor: pass MessagePreviewStore to ReactEventPreview
* Modify LinkPreview to export shared atomics
* Implement MessageComposerUrlPreviewView
* Create UrlPreviewFetcher utility function
* Modify view models
* Implement in composer
* Support running tests in dom-less vitest environment
* Add a playwright test
* hide another one
* fmt
* cleanup
* test rte too
* fixup
* Add back docstring
* cleanup
* off by one
* remove description check
* Cleanup hacks
* Remove another hack
* cleanup
* one more
* fixup window here too
* whoops type
* Rename to be cleaeer
* Trim URLs first
* fix bug
* Pluralise the multi-session remove button and confirmation dialog
When several sessions are selected in Settings > Sessions, the bulk remove button and its confirmation dialog used the singular "Remove this device" string. The singular "this" misleadingly implied the current device would be removed. Use the existing count-aware settings|sessions|sign_out_n_sessions key, matching the other-sessions menu, so the label reads "Remove N sessions".
Fixes#33812
* Update device-management e2e test for the pluralised remove button
The multi-select remove button now reads "Remove N sessions" instead of "Remove this device", so update the Playwright assertion to match.
* Put "Cancel" before "Remove N sessions" in the bulk session action bar
In the "Other sessions" multi-select action bar, "Remove N sessions" was
rendered before "Cancel", which placed the destructive action on the left.
Swap the two buttons so "Cancel" is on the left and "Remove N sessions" on the
right, per review feedback on #33983.
* Fix more menu focus issue when focused with no hover.
* Use a js based solution for focus
* Return focus to the menu trigger when closing the menu with Escape
Keep the keyboard-focus marker set while a row/section-header menu is open
(focus is then in the portaled popover, outside the element). This keeps the
trigger revealed so the menu's own focus restoration lands on it when closed
with Escape, instead of dropping to <body>.
* test: cover keyboard-focus reveal of the room list hover menus
Adds unit tests that focus a room row / section header via the keyboard
(:focus-visible on mount) and assert the hover menu is revealed, then cleared
when focus leaves. Brings diff coverage of the focus handlers to 100%.
* Add a pinned sticky group header to GroupedVirtualizedList
Section-header rows pin natively with `position: sticky`, applied via a
`components.Item` wrapper that detects header entries. Consecutive stuck
headers stack in DOM order, giving compositor-driven cover/reveal: the
incoming header covers the current one on scroll-down and slides off to
reveal the previous on scroll-up.
Because the flat virtualizer unmounts off-screen rows, a sticky header
row vanishes once a tall section scrolls past the overscan window. An
always-mounted overlay outside the item stream backstops this, keeping
the current section's header pinned; a mounted real header row sits over
it for an invisible hand-off. The current section is tracked from the
rendered items' offsets and live scrollTop on Virtuoso's onScroll.
`scrollPaddingTop` on the useVirtualizedList hook lands keyboard-focused
items below the pinned header via Virtuoso's calculateViewLocation.
Signed-off-by: David Langley <langley.dave@gmail.com>
* Show a sticky section header in the room list
VirtualizedRoomListView renders the current section's header into the
pinned overlay via RoomListStickySectionHeaderView, which reuses
RoomListSectionHeaderContent so the chevron, title, notification badge
and section menu stay identical to the real header. The overlay is
mouse-interactive but `aria-hidden` and out of the tab order; the real
header rows keep all keyboard, screen-reader and drag-and-drop duties.
Headers use a uniform 44px height so a covering header fully covers the
one behind it, and an opaque background so stacked stuck headers don't
show through each other. Snapshot regenerated.
Signed-off-by: David Langley <langley.dave@gmail.com>
* Stop enforcing 44px height
* Clarify sticky-header z-index layering
Only the relative order of the room-list sticky stack matters (room rows <
overlay backstop < real sticky header rows), so use 1/2 instead of 10/11 and
spell out the ordering in comments. Explain why the overlay's opaque fill needs
z-index: -1 (absolutely-positioned, so it would otherwise paint over the header
content). No behavioural change.
---------
Signed-off-by: David Langley <langley.dave@gmail.com>
2026-06-29 19:28:50 +00:00
Will HuntGitHubCopilot Autofix powered by AI <223894421+github-code-quality[bot]@users.noreply.github.com>
* Move ToastContext and utilities to shared components
* lint
* fix type
* cleanup
* fix broken test
* fix lint
* Add more tests for ToastContext
* Potential fix for pull request finding 'Unused variable, import, function or class'
Co-authored-by: Copilot Autofix powered by AI <223894421+github-code-quality[bot]@users.noreply.github.com>
---------
Co-authored-by: Copilot Autofix powered by AI <223894421+github-code-quality[bot]@users.noreply.github.com>
* Mock out window.addEventListener globally
As some code uses it
* Move to setupGlobals
* clean up import
* Only needed if we don't have happy-dom
* thank you!
Co-authored-by: Michael Telatynski <7t3chguy@gmail.com>
---------
Co-authored-by: Michael Telatynski <7t3chguy@gmail.com>
* feat:add missing methods to rls V3
* feat: use rls v3 methods instead of old alrgorithms
* feat: move stabel function to own utils
* feat: use this new moved functions
* test: cleaner test
* Don't re-export MainSplitContentType enum
I don't see why this was necessary, importing RoomView pulls in all sort
of stuff, whereas RoomContext is much smaller.
* ts, not tsx
* unintentional change
* Forward the scroller element and scroll handle through VirtualizedList
The room list needs two things the generic list did not expose: the underlying
scroll container (to observe which items are genuinely visible) and the imperative
scroll handle (to scroll an item into view). Forward an optional scrollerRef from
useVirtualizedList and pass scrollHandleRef through FlatVirtualizedList, mirroring
what GroupedVirtualizedList already exposes. Both are additive and optional, so other
consumers are unaffected.
Signed-off-by: David Langley <langley.dave@gmail.com>
* Add an unread-activity toast to the room list
Show a clickable "You have unread activity" pill when there are unread rooms
scrolled below the visible fold, including unreads hidden inside collapsed sections
whose header is below the fold. Clicking it scrolls the next such unread into view.
The visible fold is tracked with an IntersectionObserver over the rendered item
elements, so the toast appears as soon as a room crosses the fold rather than only
once it leaves Virtuoso's overscan buffer. The toast click is wired through an
imperative scroll handle the view registers with the view model.
Signed-off-by: David Langley <langley.dave@gmail.com>
* Change toast to target notifications only, not unread activity.
* Fix formatting and refresh Toast snapshots for compound-web 9.7.0
- Apply oxfmt to the room list view model and virtualized view.
- Refresh the RoomListToast/RoomListView story snapshots: compound-web
9.7.0 moves the typography classes onto the toast .content element and
rebuilds the Toast CSS module hashes.
- Add the missing UnreadActivityToast render snapshot baseline.
* Add e2e tests for the room list unread activity toast
Cover the unread-activity toast end to end:
- it appears for a notifying room scrolled below the fold and clicking it
scrolls that room into view (then the toast clears);
- a room with only an unread-activity dot (no notification count) does not
raise it;
- a collapsed section hiding a notifying room raises it, and clicking
scrolls the section header into view.
* Add visual snapshot baselines for the unread-activity toast stories
* Converge room-list toast codepaths into a single state-driven toast
Reviewer feedback: the unread-activity toast and the transient event toasts
(section_created / chat_moved) should share one mechanism, with toast
lifecycle and precedence owned by the view model rather than split across a
separate snapshot flag and a view-level ternary.
- RoomListViewModel now reconciles the transient event toast and the derived
unread-activity state into a single snapshot.toast via recomputeToast(); the
event toast still takes precedence and auto-dismisses, and the unread toast
reappears once it clears. Drops hasUnreadActivityBelow from the snapshot.
- RoomListView renders a single RoomListToast driven by snapshot.toast; the
ToastType union gains 'unread_activity'. The standalone UnreadActivityToast
component is folded into RoomListToast (clickable arrow-down variant).
- Adds VM tests for the unread-activity toast and the event/unread precedence.
* Pre-bundle the room-list dnd-kit + virtuoso graph in vitest browser mode
The room-list suites (RoomListView, VirtualizedRoomListView,
RoomListItemMoreOptionsMenu) import a heavy @dnd-kit + react-virtuoso graph.
Only @dnd-kit/abstract was pinned in optimizeDeps.include; @dnd-kit/dom,
@dnd-kit/react, @dnd-kit/abstract/modifiers and react-virtuoso were left to
runtime discovery. Under CI load the browser-mode dep optimizer can discover
them late, re-bundle and reload the page, which fails the in-flight
setupTests.ts import for those suites. Pinning the whole graph forces it into
the initial optimize pass so no re-run happens while tests load.
---------
Signed-off-by: David Langley <langley.dave@gmail.com>
TimoandGitHub
Michael TelatynskiandGitHub
David BakerandGitHub
Florian DurosandGitHub
Will HuntandGitHub
Huy HoangandGitHub
David LangleyandGitHub
Richard van der HoffandGitHub