Skip to content

Batch 2026-10-01: table quick filter, proxy return, commit extenders, e2e traces, Android menus, demo tooling, docs (11 PRs) - #1947

Merged
joepio merged 71 commits into
developfrom
claude/ui-batch-2026-09-29
Oct 1, 2026
Merged

joepio merged 71 commits into
developfrom
claude/ui-batch-2026-09-29

Conversation

@joepio

@joepio joepio commented Sep 30, 2026

Copy link
Copy Markdown
Member

Requested by Joep

Before: finished PRs by Joep, Polle and Michiel waited one Mancave run each, some behind conflicts with develop or with each other.

After: they sit together on one branch on develop (after #1915's squash, c2f22e6) and are decided by one full run. Squash-merge it like the earlier batches, then close the included PRs.

Every PR here was read before being batched. Big or impactful ones (sync, storage, auth, data formats, removing endpoints, team-wide tooling) are left out and have a review comment.

Included:

Decisions made while resolving:

Left out, with reasons on each PR: #1896/#1898 (calendar data format; #1898 has a timezone bug), #1929 (team-wide agent hook), #1866 (WebSocket parent fan-out and collection semantics), #1663 (reverses the develop-run policy), #1530 endpoint removal, #1524 (theme redesign under a docs title), #1638 (its test relies on the Tags button that develop now hides), #1920 (sync), #1923 (Forms, 40k lines), #1931 (pnpm 11), #1885 (storage), and #1911–#1914, #1918, #1919 (another batch).

Checked locally: data-browser tsc, pnpm run -r lint, vitest for TablePage and Dropdown 458/458, and data-browser vitest otherwise green. The one @tomic/lib failure, genesis-roundtrip.integration, needs https://atomicdata.dev/properties/loroUpdate, which currently 404s from here; it is unrelated to this batch.

claude and others added 30 commits September 17, 2026 21:50
A table row is a resource of its own, so its comment thread is already the
ordinary one — Messages whose `about` points at the row. What was missing was
a way to reach it: the Comments panel could only ever be about the page's own
resource, and a row is never that.

Two pieces:

- The grid takes a `RowHeaderAddonComponent`, rendered in each row's gutter
  left of the row number. It knows rows only by index, so the table page
  supplies the index -> subject mapping (`getRowSubject`) and the component.
  The gutter is widened to hold a bubble beside the row number, and row-scoped
  affordances marked `data-row-affordance` are revealed by hovering anywhere in
  the row rather than only over the gutter.
- The panel gains a target. `commentSubject` says which resource the thread is
  about, `undefined` meaning the page's own — what the NavBar button opens.
  Aiming it at another resource switches threads instead of toggling, so
  clicking row 3's bubble while row 1's thread is up does what it looks like.
  A close request carries no target and so cannot be expressed as a toggle:
  `closePanelState` closes the panel whatever it is aimed at, which is what a
  deleted target, a navigation and the drawer's own dismiss all need.

The bubble stays put, with its count, for a row that has comments (highlighted
while some are unseen) and appears on hover for one that has none.

Not covered: reaching the bubble by keyboard. It shares the gutter's
mouse-only reachability with the existing expand button, because the grid owns
Tab for cell navigation.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SooZ4RQshFY8YncRNq6Qpe
The gutter bubble never appeared for the rows it most needed to: the ones you
had just typed. It asked the collection for every row's subject and treated
`index >= memberCount` as "no resource yet", which is not what that boundary
means. A row added this session keeps its local `_new:` key for its whole life
on the page — materializing persists it but deliberately does NOT turn it into
a collection member, so react-window never remounts it mid-edit. Past the
member count sat real, saved rows, reported as having nothing to point at.

So the context now reports which of the two kinds a row is (`rowSource`),
mirroring what `TablePage/TableRow.tsx` itself renders, and the bubble uses the
matching hook for each: `useMemberFromCollection` for a member,
`useResource` on the `_new:` key for a session row, which the store aliases to
the real subject once it materializes. The two cases genuinely need different
hooks — a session row has no subject at all until then — which is why this
reports the source rather than resolving to a subject.

Caught by `table-row-comments.spec.ts`, which now passes end to end against a
real server.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SooZ4RQshFY8YncRNq6Qpe
scripts/demo-session/demo-session.sh starts an isolated atomic-server and
data-browser for a branch, with a fresh data dir, a dev-only interaction log
and an in-page seed. See scripts/demo-session/README.md.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…n checklist

A seeded note answered a task in the first session, so explanations move to
code comments. Clicks and inputs in the table editor now log their row and
column. The README gets the observer checklist: keep the browser pane
visible, let the tester set phone width by dragging the pane.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A drive-app session has to run against atomic-plugins' pinned host commit,
not a moving branch head, so a SHA gets a detached worktree. --build-only
builds without starting anything, for when another session holds the ports.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ion proxy

--plugin NAME builds atomic-plugins' integrations/NAME/app, starts the mock
proxy (MOCK_SCENARIO=user-testing by default) and points the server and Vite
at it. The issue-tracker seed installs the built app with createApp and
handOverAppKey, as "New app" does. The README records two limits found on
2026-09-25: the desktop browser pane blocks sandboxed frames, and the host's
ProxyConnectReturn fails under StrictMode in the dev build.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Class-extender after_commit handlers run after the commit is checked and
persisted, but their errors propagated out of apply_commit, so the client
saw a failure for a change that was saved and might retry or show an
error. Each extender now runs through Db::run_after_commit_extender; a
failure is logged with tracing::error! (extender, plugin, subject, commit,
error) and the remaining extenders still run. A poisoned extender lock
falls back to the inner list instead of skipping every side effect.

Closes #1848

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
In dev, StrictMode ran ProxyConnectReturn's effect twice. The first run
consumed the pending handoff and was cancelled before redirecting; the
second found no handoff and showed "This connection return is not one
this browser started." Share one redemption per integration_state so the
second run awaits the first and redirects.

Closes #1883

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Adds planning/durable-after-commit.md: a queued, at-least-once afterCommit
for extension-world plugins on tables showing their view, reading #1850's
change list with a host-kept cursor per (installation, table) and writing
through #1788's row grant. Points plugin-runtime-convergence.md step 4 at it.

Refs #1851

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Editing inside the settings section swapped the list for a long form and
put the Save/Cancel buttons below it, which felt cramped. The lists stay
on the AI settings page; creating or editing an agent, skill or MCP
server now opens a shared ConfigEditorDialog with the form and
Cancel/Save actions. Escape, the close button and an outside click
cancel; Save is disabled until the required fields are filled in.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01P648McUYrscNsheXY5Ya81
The App panel at the bottom of the sidebar listed eight entries. It is now
a single row: the current user opens a dropdown with Profile,
Notifications (with the unread count), Integrations, Sync, Feedback,
About and, when available, Install App. Settings sits beside the row as a
gear icon button. Signed out, the row reads "Login / New User" and its
first item leads to the sign-in page.

The row reuses DropdownMenu, so arrow keys, Enter and Escape work and
focus returns to the trigger. An unread-notifications dot shows on the
row while the menu is closed.

To serve both the floating Feedback button and the menu item, the
feedback dialog moved into useFeedbackDialog (FeedbackMenuItem is now
FeedbackButton), and SyncMenuItem became a useSyncStatus hook.

E2E specs that clicked the old panel entries go through the new
openAccountMenu / clickAccountMenuItem helpers, and signIn waits for the
account row instead of a link to /app/agent.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01P648McUYrscNsheXY5Ya81
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01P648McUYrscNsheXY5Ya81
The email invite field focused its input on every click inside it,
including clicks on the "Can write / Can read" select. Moving focus to
the input blurred the native select right after it opened its option
list, so the list closed again. The field now only focuses the input
when the click lands outside its controls.

The input also asks password managers not to fill it (data-bwignore,
data-1p-ignore, data-lpignore, data-form-type) and uses
autocomplete="off", since it holds other people's addresses rather than
the user's own. It stays type="email".

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01P648McUYrscNsheXY5Ya81
A "Find rows…" field next to Filter and Columns keeps the rows where any
visible column shows the text, case-insensitive: text and markdown as
plain text, numbers as formatted (and as typed), dates and timestamps as
displayed, a ticked checkbox as its column label, references and select
tags by title, multi-value cells by any item.

It narrows the view's own column-filtered, sorted collection on the
client, so it combines with both. While active it walks every page of
that collection, not just the 30 rows loaded. The same filtered rows
feed the table, timer, board, issue list and calendar views; dashboards
and app views don't show it. Session state only: it resets when you
switch views or leave the table. Clear with the x or Escape; below 40rem
it folds into a search icon.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
An image uploaded into a document keeps only its `downloadURL`
(`<server>/download/files/<hash>`) in the node's `src`. In a browser-only
drive (the default for the hosted distribution and for drives created
there), neither the File's commit nor its bytes ever reach a server, so
that URL answers 404 and the editor's node view showed "Failed to load
image". File previews did not have this problem because they resolve
the bytes locally through `useFileObjectUrl`; the editor image never did.

- `@tomic/lib`: `blobSubjectFromDownloadUrl` maps a content-addressed
  download URL back to its `atomic:blob:<hash>`.
- `@tomic/react`: `useBlobObjectUrl`, the blob-reference half of
  `useFileObjectUrl`, for callers without a File resource.
- `ImagePicker`: render the local copy when there is one, fall back to
  the URL otherwise, wait for the lookup instead of racing it, and key
  the error state by src so a new src gets a fresh attempt. The document
  still stores the portable URL.
- Wrap two label texts in spans: the i18n transform turned them into
  keyless arrays, which logged a React key warning whenever the picker
  opened.

Adds `document-image.spec.ts`, which fails with "Failed to load image."
before this change and passes after, including across a reload.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01P648McUYrscNsheXY5Ya81
The status icon's SVG title ("Connected", "Offline") became part of the
menu item's accessible name. Hide the icon from assistive tech; the status
is still in the item's tooltip.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01P648McUYrscNsheXY5Ya81
With text in "Find rows…", the totals under the table (sum, average, min,
max, count, and the per-group breakdown) now cover the matching rows — the
rows on screen and the ones the footer counts — instead of every row the
column filters let through. The store can't be asked about the quick
filter's rows, so they are computed client-side by `aggregateRows`, which
mirrors the store's rules (`aggregate.rs`). With the field empty, the
store's totals pass through unchanged.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Choosing Calendar from the menu of a table's only view tab turned that
view into a calendar, and the table layout was gone (#1806). The user
should always be able to switch between all the applicable views.

- A tab's menu now has an "Add a view" section (built-in kinds and
  apps), the same as the + tab: it adds a view named after the type and
  switches to it, leaving the tab's view intact.
- "Change this view to" is offered only while another view of the
  current type remains, so the last table view (or the only view of an
  app) can never be converted away. setViewKind enforces the same rule
  and adds a view instead.
- Delete is disabled on the last table view while other views exist.
- A converted view whose name was a default ("Default View", a type
  label, a template view name, or an "All <rows>" table name) takes the
  new type's name; a name a person chose is kept.

Closes #1806

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
useDriveApps published nothing until its whole mount chain had finished:
findSchema, the local app query, then an awaited getResource per app, one
after the other. Live events that landed meanwhile were held back too. After
a reload, an app whose sync delta could not apply to the page's copy
("incomplete Loro import ... missing base state") sits as a placeholder under
repair, and getResource waits up to 10 s on it, which is as long as the
money e2e waits for the "New app" menu item.

- Attach the listeners before anything else and fold every answer in as it
  arrives, reading the live resource each time, so no answer overrules a
  newer one and no slow read holds back the others.
- Read the query's members in parallel.
- Expose refresh(), and call it when "+ Add view" and a view tab's menu
  open, so opening a menu re-runs the query and re-reads what it knows.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
# Conflicts:
#	browser/CHANGELOG.md
#	browser/data-browser/src/components/SideBar/FeedbackMenuItem.tsx
#	browser/data-browser/src/locales/de.po
#	browser/data-browser/src/locales/en.po
#	browser/data-browser/src/locales/es.po
#	browser/data-browser/src/locales/fr.po
# Conflicts:
#	browser/CHANGELOG.md
#	browser/data-browser/src/locales/de.po
#	browser/data-browser/src/locales/en.po
#	browser/data-browser/src/locales/es.po
#	browser/data-browser/src/locales/fr.po
…al bootstrap

# Conflicts:
#	browser/data-browser/src/locales/de.po
#	browser/data-browser/src/locales/en.po
#	browser/data-browser/src/locales/es.po
#	browser/data-browser/src/locales/fr.po
Split out of #1530, which also removes the /version endpoints and waits for
review. This part is a safe fix on its own: useResource('') resolved against
the page origin, so every page fetched / and got a 404.
…leViewTabs mock for useDriveApps' {apps, refresh} (#1904 x #1902), settle catalogs
joepio added 4 commits October 1, 2026 06:59
This reverts commit 41921bb, reversing
changes made to d55e7b1.
…u opens"

This reverts commit d55e7b1, reversing
changes made to 7017f58.
…r replace the last one"

This reverts commit 20b8947, reversing
changes made to 4f164ed.
@joepio joepio changed the title Batch 2026-09-30 (2): table quick filter, view tabs, row comments, Android menus, commit extenders, e2e traces, and more (15 PRs) Batch 2026-10-01: table quick filter, proxy return, commit extenders, e2e traces, Android menus, demo tooling, docs (11 PRs) Oct 1, 2026

joepio commented Oct 1, 2026

Copy link
Copy Markdown
Member Author

Rebuilt on develop ef46810 (after #1946 and #1866 landed). The first run of this batch failed three e2e tests, all from members that do not fit current develop, so #1902 and #1904 (tab menu) and #1523 (row comments) are out of it. Their reasons are on those PRs. #1945 landed already through #1946. What stays: #1894, #1884, #1930, #1940, #1888, #1933, #1882, #1834, #1631, #1932 and the ForkBar part of #1530. New run dispatched on 59b9651.


Generated by Claude Code

@joepio
joepio merged commit 1f892aa into develop Oct 1, 2026
8 checks passed
@joepio
joepio deleted the claude/ui-batch-2026-09-29 branch October 1, 2026 16:56
joepio pushed a commit that referenced this pull request Oct 3, 2026
develop carries the fork-bar fix now (#1947), so this branch's copy of it
auto-merged to a no-op and the diff is just the endpoint removal.

browser/CHANGELOG.md was the only conflict: both sides appended to
UNRELEASED. Kept both, develop's table-row-comments entry and this
branch's two. develop has no fork-bar entry of its own, so the one here
is still the only record of that fix.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01YUL3EPFVmaDAok34JPev3F
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants