feat: PK metadata detection, save preserves original query, AI Assistant persistence

CHANGELOG.md

@@ -40,12 +40,16 @@ All notable changes to QueryDen are documented here. This project adheres to [Se
 - **PostgreSQL FK relationships now render in the ER diagram.** The `schemaItems.foreignKeys` list is sometimes null when schema loading fails silently. Added direct `fetchPostgresForeignKeys()` and `fetchMySQLForeignKeys()` fallbacks that query system catalogs directly, plus a `normalizeTableName()` helper to reconcile the `public.` prefix format difference between PostgreSQL's `regclass::text` and the schema items table list.
 - **PostgreSQL columns in `public` schema no longer missing from the ER diagram.** Tables with many columns were showing as header-only because the column map keyed by `public.tablename` didn't match the schema items' bare `tablename` format. Stripping the `public.` prefix in the column lookup fixed it.
 - **ER diagram no longer freezes when opening on databases with many tables.** The table selector dialog now appears first — users pick which tables to include before any schema data is fetched or dagre layout runs. Selected-only data keeps both the query batch and the layout computation scoped to what's visible.
+- **Save no longer replaces custom query results with a full table scan.** After saving edits, `handleSave` re-runs the user's original SELECT query (captured from `lastSelectQueryRef`) instead of a generic `SELECT * FROM table LIMIT 1000`. Column selection, ORDER BY, and LIMIT are preserved — the grid stays on the user's custom result set.
+- **Schema-qualified table PK candidates now use the base table name.** `handleSave` and `handleUpdateRow` derive table-specific PK candidates from the unqualified table name (e.g. `config_id` for `ansible.config`) instead of the full schema-qualified name, fixing PK detection on multi-schema databases.
+- **Fallback UPDATE WHERE clause uses pre-edit values.** When no PK column is found, `handleSave` reads from the `_original` row snapshot (captured at edit time) rather than the current edited values, preventing 0-row-affect UPDATEs.
 
 ### Changed
 - **ER diagram compact view is now the default.** The diagram opens in compact mode (PK/FK columns only) instead of showing all columns, reducing visual clutter on first open. Users can toggle back to full-column view via the toolbar button.
 - **PostgreSQL FK introspection now uses `pg_catalog.pg_constraint` instead of `information_schema`.** The Database Explorer and post-SELECT FK auto-loading both query `pg_constraint` directly, correctly handling cross-schema foreign keys, composite (multi-column) FKs, and environments where `information_schema` permissions are restricted. Cross-schema FK references are schema-qualified in the UI (e.g. `"billing"."invoices"`).
 - **FK metadata is now auto-loaded after SELECT queries.** When the results grid detects a table-backed SELECT, it fetches FK constraints from the relevant engine (PostgreSQL/Supabase/CockroachDB via `information_schema`, MySQL/MariaDB via `information_schema`, SQLite via `PRAGMA foreign_key_list`) and caches them per `schema.table` so subsequent queries on the same table are instant. FK columns then render as link cells without requiring a prior tree-expand.
 - **Webview default right-click context menu re-enabled.** The document-level `contextmenu` `preventDefault()` that suppressed the browser-native Reload/Inspect Element menu on Linux WebKitGTK has been removed. The app's own context menus (tab strip, Monaco editor, grid) stop propagation and remain unaffected.
+- **PK detection now prefers real database metadata over name heuristics.** `loadTableDetails` queries actual PK columns from `information_schema.table_constraints` (PostgreSQL/MySQL) or `PRAGMA table_info` (SQLite). The post-SELECT FK auto-load path also detects PKs in parallel. All CRUD operations use a unified `getPrimaryKey` helper that checks `tableSchema.primaryKeys` first, then falls back to expanded name heuristics (`id`, `uuid`, `uid`, `pk`, `row_id`, `object_id`, `key`, `code`, `_id`, `{table}_id`).
 
 ### Added
 - **FK-aware inline navigation in the results grid.** FK columns render as styled link cells with an external-link icon. Clicking the icon fires a `SELECT * FROM refTable WHERE refCol = value` query, navigating directly to the referenced row — enabling quick parent-row lookup from any query result.
@@ -63,7 +67,10 @@ All notable changes to QueryDen are documented here. This project adheres to [Se
 - **PostgreSQL operators enabled by default in schema tree.** The `showOperators` setting now defaults to `true`, so operator nodes appear in the schema tree without having to toggle the setting first.
 - **ER Diagram visualization for PostgreSQL, MySQL/MariaDB, and SQLite.** A new ER Diagram dialog (toolbar <Table /> button) renders the current database as an interactive entity-relationship diagram powered by `@xyflow/react` v12 and `dagre` auto-layout. Tables are shown as cards with PK (key icon) and FK (link icon) column markers; relationships are drawn as smooth-step edges with cardinality labels (`*` at the FK end, `1` at the PK end). The dialog opens with a table selector so users pick which tables to include — preventing the freeze that would occur when auto-introspecting many tables. A toolbar provides schema multi-select (PostgreSQL), search/filter with related-table expansion, compact/PK-only column mode, refresh, and SVG export (Tauri `save()` dialog + `writeTextFile`). Schema data is cached with a 30-second TTL so re-opening the same tables is instant. Provider-specific introspection: PostgreSQL queries `pg_attribute`/`pg_index`/`pg_constraint` in bulk, MySQL uses `information_schema`, SQLite uses batched `PRAGMA table_info`/`PRAGMA foreign_key_list` via `Promise.all`.
 - **ER Diagram now opens on the correct database.** Previously the diagram always read the global `activeConnection` — if the active editor tab targeted a different database, the diagram showed the wrong schema. The toolbar button now switches the connection to match the tab's target before opening.
+- **Grid scrolls to the saved record after save.** After saving changes (via `handleSave`), the grid dispatches `grid-scroll-to-row` to scroll to the modified row matched by PK value, or `grid-scroll-to-bottom` for newly inserted rows — keeping the user's viewport focused on the record they just edited.
 
+### Removed
+- **Roadmap items removed as addressed.** "Vault crypto hardening" (#14, #15), "Schema-aware row editor" (#50), and "ER Diagram" (#51) removed from the website roadmap — vault hardening, schema-aware FK dropdowns, NOT NULL validation, and ER diagram support in this release subsume those feature requests.
 
 ## [1.0.23] - 2026-05-26
 

src/components/explorer/DatabaseExplorer.tsx

@@ -38,6 +38,7 @@ interface TableDetails {
   foreignKeys: { columns: string[]; refTable: string; refColumns: string[] }[];
   indexes: { name: string; columns: string[]; unique: boolean }[];
   triggers: string[];
+  primaryKeys: string[];
 }
 
 interface DatabaseExplorerProps {
@@ -590,7 +591,7 @@ export function DatabaseExplorer({ isAddConnectionDialogOpen = false }: Database
     setLoadingTableDetails(prev => new Set(prev).add(key));
 
     try {
-      const details: TableDetails = { columns: [], constraints: [], foreignKeys: [], indexes: [], triggers: [] };
+      const details: TableDetails = { columns: [], constraints: [], foreignKeys: [], indexes: [], triggers: [], primaryKeys: [] };
 
       if (["postgres", "supabase", "cockroach"].includes(activeConnection.type)) {
         // Load columns
@@ -607,6 +608,20 @@ export function DatabaseExplorer({ isAddConnectionDialogOpen = false }: Database
           default: c.column_default
         }));
 
+        // Load primary keys
+        const pks = await currentDb.select(`
+          SELECT kcu.column_name
+          FROM information_schema.table_constraints tc
+          JOIN information_schema.key_column_usage kcu
+            ON tc.constraint_name = kcu.constraint_name
+            AND tc.table_schema = kcu.table_schema
+          WHERE tc.table_schema = $1
+            AND tc.table_name = $2
+            AND tc.constraint_type = 'PRIMARY KEY'
+          ORDER BY kcu.ordinal_position
+        `, [schemaName, tableName]);
+        details.primaryKeys = pks.map((p: any) => p.column_name);
+
         // Load indexes
         const idxs = await currentDb.select(`
           SELECT indexname, indexdef
@@ -688,6 +703,16 @@ export function DatabaseExplorer({ isAddConnectionDialogOpen = false }: Database
           default: c.column_default,
         }));
 
+        const pks = await currentDb.select(`
+          SELECT column_name
+          FROM information_schema.key_column_usage
+          WHERE table_schema = DATABASE()
+            AND table_name = ?
+            AND constraint_name = 'PRIMARY'
+          ORDER BY ordinal_position
+        `, [tableName]);
+        details.primaryKeys = pks.map((p: any) => p.column_name);
+
         const idxs = await currentDb.select(`SHOW INDEX FROM ${quoteIdentifier(tableName, activeConnection.type)}`);
         const idxMap: Record<string, { columns: string[]; unique: boolean }> = {};
         for (const idx of idxs) {
@@ -747,6 +772,11 @@ export function DatabaseExplorer({ isAddConnectionDialogOpen = false }: Database
           nullable: c.notnull === 0,
           default: c.dflt_value,
         }));
+        // PRAGMA table_info returns `pk` as the 1-based PK ordinal (0 = not PK)
+        details.primaryKeys = cols
+          .filter((c: any) => c.pk > 0)
+          .sort((a: any, b: any) => a.pk - b.pk)
+          .map((c: any) => c.name);
 
         const idxs = await currentDb.select(`PRAGMA index_list(${quotedTable})`);
         for (const idx of idxs) {