feat(datagrid): show whether a materialized view can be refreshed concurrently, and gate its structure edits by kind

[datlechin]

Summary

A materialized view's Indexes tab now says whether Refresh Materialized View can run concurrently, and the Structure tab stops offering edits a materialized view refuses on the routes that still offered them.

Asks 1 and 2 of #2522 (list a matview's indexes, add and drop them) already worked on PostgreSQL when the view was opened from the sidebar, since #2772 and #2778. What was left:

• Ask 3 had a data source and no consumer. concurrentRefreshAvailability (since feat(plugins): refresh materialized views, show view DDL and edit object comments (#2726) #2753) was read by the Refresh prompt only.
• Open Quickly dropped the object's kind. It maps a materialized view to its .view row kind for the icon and opened the tab with isView: true and no objectType, so the Structure tab gated it as a plain view and refused Add Index with "A view cannot have indexes."
• PGlite runs the PostgreSQL driver and PostgreSQL 17, and was left on the tables-only edit matrix, so a PGlite matview's index add and drop were refused the same way.
• Three surfaces beside the grid skipped the edit gate: the inspector let you change a matview column's Type, Nullable and Default (the server refuses all three), the row menu offered Delete and Duplicate where they did nothing, and the Triggers tab offered New Trigger on a matview.

What changed

• StructureEditingSession holds the answer as a MetadataLoadState, asked of the server after every index fetch (first load, Cmd+R, after Save, after Refresh Materialized View) and on remount. It is never derived from the index grid: an index left INVALID by a failed CREATE UNIQUE INDEX CONCURRENTLY lists exactly like a valid one. A failed recheck shows "Couldn't check…" rather than the previous answer, because the index list beside it has already been refetched, and only the latest request may settle the state, so a remount's check that finishes after a post-save check cannot overwrite it.
• ConcurrentRefreshNoteView shows it as one secondary line in a bottom safeAreaInset of the Indexes tab, mirroring the Filter field's top inset. Nothing shows where the engine has no concurrent refresh.
• QuickSwitcherItem.tableType carries the TableInfo.TableType the row was built from, through both same-window arms and the cross-window route (LaunchIntent.openTable(objectType:) into EditorTabPayload).
• PGlite gets structureEdits: .postgreSQL.
• StructureEditGate.locksField(at:on:orderedFields:) is the one lock rule. The grid's per-heading lock, the inspector (new InspectorRowField.isEditable) and the commit path all read it, so no surface takes a keystroke another refuses. A list whose add the object or engine refuses (Indexes on a view, Foreign Keys on a matview, Indexes on an engine without CREATE INDEX) is locked whole instead of taking input and dropping it.
• Row menu: canDuplicate / canDelete from the gate replace one engine-wide Bool.
• TriggerEditEligibility withholds triggers only from the kinds measured to refuse them. It is deliberately not a cell of the edit matrix, which offers an uncurated engine nothing but tables: SQLite, SQL Server and Oracle take INSTEAD OF triggers on views.
• "Available once the view holds rows" was the wrong condition in the Refresh prompt too. PostgreSQL checks relispopulated, and a view refreshed over an empty result is populated with zero rows. Both strings now say "populated".

No PluginKit change, so no kit bump.

Measured on PostgreSQL 17.11

• The driver's indexList returns a matview's indexes, expression keys included (no relkind filter).
• CREATE TRIGGER: accepted on a table, partitioned table, view (INSTEAD OF) and foreign table; refused on a matview (relation "mv" cannot have triggers) and a system catalog.
• ALTER COLUMN … TYPE, SET DEFAULT, SET NOT NULL on a matview: all refused.
• REFRESH MATERIALIZED VIEW CONCURRENTLY on a matview created WITH NO DATA over an empty query: refused until one plain refresh, then accepted with relispopulated = t and 0 rows.
• End to end in a sandboxed Debug build: daily_totals (unique index on day, region) shows "This view can be refreshed concurrently."; region_totals (non-unique index only) shows the unique-index requirement; after CREATE UNIQUE INDEX on the server and a refresh, the line switched to available; the matview's Triggers tab showed no action bar.

Before / After

Before: the same tab with no line under the list, which is what 0.75.0 shows. After:

Tests

• New: MaterializedViewConcurrentRefreshNoteTests, MaterializedViewConcurrentRefreshCheckTests (name, schema and scope reach the driver; other kinds never ask; recheck replaces; a failed recheck reports the failure; an older request finishing last cannot overwrite a newer answer), QuickSwitcherObjectKindTests.
• Extended: StructureEditGateTests (PGlite matrix, trigger rule, the per-tab lock), StructureGridDelegateInspectorTests (a matview's locked fields refuse inspector commits, unlocked ones take them, row menu flags), StructureRowMenuRouteTests, MultiRowEditStateTests.
• 434 cases across the 31 suites that own the changed types pass locally.
• No UI test: the flow needs a live PostgreSQL server, and the UI test sandbox has none it can reach deterministically. It was checked by hand as above.

Deliberately not fixed here

A tab opened by name alone (a deeplink, AppleScript, the MCP open_table_tab tool, or a tab file from before tabs stored their kind) still guesses its kind from isView, as it did before this change. A first version resolved it from SchemaService's loaded object list, and two review rounds found five defects in that path: the per-schema cache has no database identity, the $states subscription can fire before loadedScope is committed, adopting a kind left isView/isEditable on the guess, dropping a guessed session could lose a half-typed inline edit, and the lookup used the raw tab schema instead of the resolved one. Doing it properly needs a scope-atomic catalog revision first, so it was cut to what the Open Quickly route can carry today. For the same reason, choosing a materialized view in Open Quickly while a kind-less tab on it is already open reuses that tab as it is.

Open Quickly still opens a materialized view with isView: true, which keeps its rows read-only, as before this change. That is the right answer on PostgreSQL, where UPDATE on a materialized view fails with cannot change materialized view; the sidebar route, which makes those rows editable, is the one that is wrong, and it is reported separately rather than copied here.

Fixes #2522

[mintlify]

Preview deployment for your docs. Learn more about Mintlify Previews.

Project	Status	Preview	Updated
TablePro	🟢 Ready	View Preview	Sep 23, 2026, 7:18 PM

💡 Tip: Enable Automations to automatically generate PRs for you.