@ai-matrx/records-ui 0.83.1 → 0.84.4

package/CHANGELOG.md

@@ -1,5 +1,158 @@
 # Changelog — @ai-matrx/records-ui
 
+## 0.84.4 (unpublished — needs @ai-matrx/records 0.57.1 published first)
+
+**The guide re-walk's small wrong things, each fixed as a class.** (lane RECORDS-UI-FIX,
+2026-09-23.)
+
+* **Two records with one name are told apart in the linked-record picker.** Rincon's
+  Customers really holds two Patricia Marshes; the picker drew only the name, so the list
+  read as duplicates. New `tellApart(rows, nameOf)`: a colliding name carries the first of
+  its own columns that differs (an address), and search reads it too.
+* **Charts no longer print their timing ("34.5 ms").** The number stays on the block as
+  `data-store-ms` for whoever measures dashboards.
+* **The Bookings suggestion is English** ("a 30-minute appointment … put each one in Jobs",
+  never "a 30-minute appointments slot"). `bookingSuggestion` is exported.
+* **Table Settings' Proposed box names no door by its code** and shows no baseless "0"; it
+  says proposals wait in the table's Inbox.
+* **A record's history follows the writes on its page.** `HistoryPanel` re-reads on
+  `useRecordChangeRevision` (records 0.57.1). This was not the no-op-version class: the
+  store had the row, the panel never asked again.
+* A refused value no longer lights the app's error pill — see records 0.57.1.
+* Tests (red on 0.84.3): `two-customers-with-one-name-are-told-apart`,
+  `the-history-count-follows-the-writes`, `guide-rewalk-words`, and the reversed ms clause in
+  `demo/dashboard-canvas.test.tsx` (it had enshrined printing the timing).
+
+**Also, in matrx-frontend (`features/sharing`):** the Share dialog's Access tab no longer
+reads "Internal — readable inside the owning organization" over "Private — only you".
+
+## 0.84.3 (unpublished — rides on top of 0.84.2)
+
+**After an import writes, the wizard is a report — never the same import again — and every
+line it names is the line in the person's file.** (lane RECORDS-UI-FIX, VERIFIER-16 H1 + M8,
+VERIFIER-15 M5.)
+
+* **No second write of the same file.** A run that landed rows used to leave the pre-write
+  notice ("…Nothing has been written yet", now false) and a live "Import N rows" beside the
+  result. Once a file has been offered to the store — finished, stopped part-way, or turned
+  away as already imported — every pre-write step (what each record is called, the mapping,
+  the options, the rule notice, the Import button) is replaced by one sentence saying the
+  file was sent, and the report of what landed and what was refused stays. Another run
+  means choosing a file again.
+* **Line numbers are the file's.** The store and the rule check number rows from 1 under the
+  header; the file has its header on line 1. New export `lineInYourFile(dataRow)` is used on
+  the column's refusal, the run's refused rows and past runs' refused rows — the `-3` on
+  line 3 is called line 3.
+* **Nothing sits past the rail.** The mapping is one block per column (name, where it goes,
+  what it is, what would be refused, first values) instead of a 52rem row in a 26rem rail.
+* **No machine word.** A column the table already has is named from its Field
+  (`fieldTypeLabel`), so a number reads "Number", never "range"; an unknown token is written
+  out, never shown raw.
+* **Whole sentences.** `refusalLineForAPerson` ends the store's sentence before the remedy
+  ("…at least 0. Send this column…"). The report's "The 1 row that was refused is kept".
+* Tests: new `src/the-import-after-a-write-is-a-report.test.tsx` (4, red on 0.84.1 bytes);
+  `import-asks-the-rules-first` now expects line 4 for the fourth line (it had enshrined the
+  off-by-one); `import-wizard` expects "is kept"; the F9 source guard now forbids a row wider
+  than the rail and keeps the 14rem "Goes to" floor.
+
+**Consumer action:** none.
+
+## 0.84.2 (unpublished — rides on top of 0.84.1)
+
+**A booking link opens that booking page, and the table list's empty sections stop
+describing a product that does not exist.** (lane HUB-FIX, VERIFIER-16 M4 + M6.)
+
+* `BookingSlots` gains `activeBookingId`: the page a link named is marked (`data-linked`) and
+  scrolled to; a page the rail does not list says `BOOKING_NOT_HERE_LINE`, never a different
+  page. `TablePage` hands it `?item=` when `?rail=bookings`.
+* `LANE_EMPTY.mine` said "New table makes one that only you can see" and
+  `LANE_EMPTY.organization` said "a table you make is yours until you change its visibility
+  to internal" — New table makes an internal table, and no screen here changes visibility.
+  Both now say what is true.
+* New export `BOOKING_NOT_HERE_LINE`. Tests: `rail-deep-link.test.tsx` +1,
+  `visibility-lanes.test.tsx` +1.
+
+**Consumer action (matrx-frontend):** the hub's Bookings row links
+`/data-v2/<table>?rail=bookings&item=<form_id>` (it already can; the item is marked from
+this version).
+
+## 0.84.1 (unpublished — lands with the 2026-09-24 window)
+
+**A blank the store can explain is never drawn as a dash.** (lane OLD-TABLES-4, VERIFIER-16.)
+`renderValue` — the one value renderer for the grid, the peek and the forms — now says two
+things a dash used to hide:
+
+* **A relation whose record no longer exists** reads *"The record this pointed at no longer
+  exists"* (the id it held is in the tooltip). The move of an older table lands such a cell
+  empty and keeps the id on the record's provenance entry, `_sources.<ptr>.unresolved.<key>`;
+  `unresolvedFor(document, key)` is exported for any other reader of that fact.
+* **A worked-out column with no formula in it yet** reads *"Worked out once its formula is in
+  place"* instead of an answer of nothing.
+
+Consumer action: none.
+
+**The field editor sets the three column settings the older grid carried.** (lane
+STORE-RULE-GAPS.) Existing controls only, no new primitive:
+
+* **A pattern's example** — a box beside a "Looks like" check (`rule.example`). A refused value
+  is told "Enter it like 949-555-0142.": `refusalForAPerson` reads the example from the store's
+  `detail.example` as data, so an example shaped like an identifier ("SKU-12") is never dropped
+  by the machine-identity filter.
+* **A length's shortest** — the "Number of characters" check (was "No longer than") has a
+  *fewest* box beside its *most* (`rule.min`).
+* **A choice list that takes other values** — one tick, "A value that is not one of the choices
+  is added to them" (`allow_other`), on select / multi-select columns only; unticking sends false.
+
+Consumer action: none. Needs `@ai-matrx/records` with `FieldRule.example` / `.min` and
+`allow_other` (aidream `290411764c`) and, for the store to honour them, matrx-frontend
+`storerulegaps_an_example_a_shortest_length_and_other_choices.sql` on the database.
+
+## 0.84.0
+
+**A refusal is said once, with the doors that fit it, and a change a person made is kept or
+said — never dropped.** (lane RECORDS-UI-FIX, VERIFIER-15 H2, H3, M3, M4, M6, M7.)
+
+* **The grid's cell refusal is the one notice.** A rule refusal (`-3` in a column "at least
+  0") used to draw a small red line with *Try again* (which re-sent the refused value) and
+  *Keep theirs* (conflict wording when nobody else wrote). It is now `RefusalNotice` with
+  **Keep editing** — the cell reopens holding `-3`, the reason still beside it — and
+  **Discard**. *Try again* / *Keep theirs* appear only on a real version conflict, beside
+  the other person's value. A tick box's refusal carries Discard.
+* **The record form says a refusal once, with Keep editing / Discard, and before Save.** The
+  mirror's per-column notice is the store's rule, so the store's copy under the form is no
+  longer drawn beside it; the mirror speaks the moment a changed column would be refused.
+  What was typed survives the (slower) version read landing late. An edit sends only the
+  columns the person changed, and a Save with nothing changed writes nothing.
+* **A board's layout or grouping change is kept, or said.** Every view save carried a version
+  read from the view's history, which can stop short of the record, so every Kanban / Group
+  by press was refused `stale_write` and dropped in silence. New `saveViewPatch` (the one
+  view-save door) writes on the version the refusal names; a save still refused is said
+  through `RefusalNotice` with Try again / Discard.
+* **"Someone else wrote first" only when somebody did.** The store's `contested_fields`
+  carries each patched column's current value; when every one is still what the screen
+  started from, the write goes once more on the winning version (the grid cell session, the
+  record form and saved views apply the same rule — internal `updateOnTheWinningVersion`).
+* **The next digest summary is never in the past.** New export `nextSummaryAt(subscription,
+  now)` carries a passed due time forward by whole periods to the first slot after now,
+  keeping the store's alignment (tested at a fixed clock).
+* **Table Settings never cuts a field name to two letters** — the name wraps with a readable
+  minimum and the row wraps its controls first.
+* **On a phone the table's rail is a sheet over the table** — below the design system's
+  breakpoint (`useIsMobile`, 768px) it renders through the canonical `BottomSheet`, the way
+  `Dialog` and `CommandDialog` already do; dragging it down closes the rail.
+* Tests (each shown failing on the prior bytes): `src/a-refused-cell-keeps-what-you-typed`,
+  `src/a-form-refusal-is-said-once-with-its-doors`, `src/a-board-layout-change-is-kept`,
+  `src/the-next-summary-is-never-in-the-past`, `src/a-rail-on-a-phone-is-a-sheet`, and a
+  new clause in `src/TableSettings.hooks.test`.
+
+**Store-side, reported to the chair (not in this package):** `custom.record_update` of an
+unchanged value moves `version` and writes no history row (reproduced on the dev clone), and
+the digest runner computes the next slot from the last summary SENT, so an empty window
+freezes it.
+
+**Consumer action (matrx-frontend):** none required — install picks it up. A host that
+shows a digest's next time outside `SubscriptionsPanel` should use `nextSummaryAt`.
+
 ## 0.83.1
 
 **A table shared by link is in the World lane, not Community.** (lane HUB-FIX, chair ruling