@ai-matrx/records-ui 0.84.4 → 0.84.8

package/CHANGELOG.md

@@ -1,5 +1,76 @@
 # Changelog — @ai-matrx/records-ui
 
+## 0.84.8
+
+**A host's own layout sits beside Grid, Kanban, Calendar and Gallery on the one table page.**
+(lane GRID-PORT `5cb9bcbd99`, published by PUBLISH-SWEEP-4.) `RecordsUiHost.layouts?: HostLayout[]`
+(`{ id, label, render({ tableId, filter }) }`) draws as a button in `ViewSwitcher`'s layout row;
+`?view=<id>` opens it (`openingHostLayout`), and `onViewChanged` gets the id. It is page-level and is
+never written onto the view record. A host id that is one of `PAGE_VIEWS`, or a spelling of one, is
+ignored. **Consumer action:** matrx-frontend passes its ported /data grid as the Sheet layout.
+
+## Unreleased (source only — lane GRID-PORT; the chair's publisher owns the slot)
+
+**A host can draw a layout beside the package's own four, on the one table page.**
+`RecordsUiHost.layouts?: HostLayout[]` — `{ id, label, render({ tableId, filter }) }`. Each
+appears as a button in `ViewSwitcher`'s layout row after Grid / Kanban / Calendar / Gallery, and
+`?view=<id>` opens it (`openingHostLayout`); `TablePage.onViewChanged` is called with the id.
+Page-level, never written onto the view record (the store's layout words stay the four). A host
+id equal to one of `PAGE_VIEWS` (or a spelling of one) is ignored, so `?view=grid` is always the
+package's Grid. Why: the owner's ruling (2026-09-23) — AI Matrx's ported /data grid is the
+"Sheet" layout of `/data-v2/<table>`, not a second page for the same table.
+
+Consumer action: `onViewChanged` now receives `PageView | string`; widen a handler typed
+`(view: PageView) => void`.
+
+## 0.84.7 (unpublished — needs @ai-matrx/records 0.57.3)
+
+* **The suite is green again (386/386), and the class is gone rather than patched.** 0.84.5's
+  publish gate went red in 15 files because records 0.57.2 made every table list ask
+  `custom.table_facts`. Only `TablesHome` sorts tables into sections, so only it asks now
+  (`client.tableFacts()` + `withTableFacts`); no test double had to learn the door. It says
+  `SECTIONS_UNKNOWN` only when the store has the door and it failed.
+
+## 0.84.6 (unpublished)
+
+**A choice column that takes other values lets a person type one.** (lane STORE-RULE-GAPS.)
+The production walk found the "A value that is not one of the choices is added to them" tick
+saved, while the grid cell and the record form drew a closed drop-down — nothing could send an
+off-list word. Now, on a column with `config.allow_other`, `FieldControl` (the one editor the
+grid cell, the record form and the form runner draw) is the relation picker's shape: a search
+over the choices and **Add “<typed>”** at the foot, which sends the word as typed (the store adds
+it to the column's choices and stores its key). A word that already names a choice picks that
+choice. Several-choices columns keep what was picked. Columns that do not take other values are
+unchanged. The field editor's tick is gated on the same code path
+(`CHOICE_EDITOR_ADDS_WHAT_IS_TYPED`). Paste keeps its own "add these options" step.
+
+**One "pick or add" builder.** `PickOrAdd` (search box, list, Add/Create “<typed>” foot) is the
+ONE layout both `RelationPicker` and the open choice editor are built from — same look, one place.
+The choice editor orders choices by the store's own `metadata.option_position` (`inTheStoresOrder`),
+so a choice added by typing is last, where the store put it, not wherever the lookup returned it.
+
+Consumer action: none.
+
+## 0.84.5 (unpublished — needs @ai-matrx/records 0.57.2 published first)
+
+**Mine is what you made, and an archived table's own address says it is archived.** (lane
+HUB-FIX, guide re-walk 2026-09-23 + chair ruling.)
+
+* `TablesHome`'s Mine section is the tables the person signed in MADE (`created_by`, folded in
+  by records 0.57.2's `tableList` from `custom.table_facts`), whatever their visibility — a new
+  internal table is under Mine AND My organization. When the store HAS `custom.table_facts`
+  and it does not answer, the page says so (`SECTIONS_UNKNOWN`) instead of filing everything
+  under My organization with Mine empty; when the store's catalogue does not carry the door at
+  all, nothing is asked and nothing is said (the door-presence pattern). `TablePage`'s archived
+  screen reads only the archive door every table already uses — it does not depend on
+  `table_facts`. `LANE_EMPTY.mine` says what is true under the new meaning.
+* `TablePage` asks the archive (`client.tableArchived`) before saying a table is not here. An
+  archived table reads "This table is archived", who archived it and when, and **Bring it
+  back**; `TABLE_NOT_REACHABLE` no longer says "…or it may have been deleted".
+* New exports `SECTIONS_UNKNOWN`, `TABLE_IS_ARCHIVED`. Full suite 379/379.
+
+**Consumer action:** none beyond installing it with records 0.57.2.
+
 ## 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,