fix: Kepler table-layer dataset identity handling for DuckDB-backed maps - #703

packages/kepler/src/KeplerSlice.ts (1)

606-645: ⚠️ Potential issue | 🟠 Major | ⚡ Quick win

Separate dataset ID from SQL table reference in addTableToMap.

tableName is currently used as both Kepler dataset id (info.id) and SQL query target. With custom tableSelection.getDatasetIdForTable / findTableForDatasetId, dataset ids can be aliases that are not queryable SQL identifiers, so restore/add flows can fail even after successful table resolution.

💡 Proposed fix

         addTableToMap: async (mapId, tableName, options = {}, config = {}) => {
+          const resolvedTable = findKeplerTableForDatasetId(
+            get().db.tables,
+            tableName,
+            get().kepler.tableSelection,
+          );
+          const sqlTableName = resolvedTable?.table.toString() ?? tableName;
           const connector = await get().db.getConnector();
           const duckDbColumns = await getDuckDBColumnTypes(
             {
               query: (query: string) => connector.query(query).result,
             } as DatabaseConnection,
-            tableName,
+            sqlTableName,
           );
           const tableDuckDBTypes = getDuckDBColumnTypesMap(duckDbColumns);
           const adjustedQuery = castDuckDBTypesForKepler(
-            tableName,
+            sqlTableName,
             duckDbColumns,
           );
           const arrowResult = await connector.query(adjustedQuery).result;
@@
             const datasets: AddDataToMapPayload['datasets'] = {
               data: {fields, cols, rows: [], arrowTable: arrowResult},
               info: {label, id: tableName},
-              metadata: {tableName},
+              metadata: {tableName: sqlTableName},
             };

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@packages/kepler/src/KeplerSlice.ts` around lines 606 - 645, addTableToMap
currently reuses the incoming tableName as the Kepler dataset id (info.id) which
breaks when dataset ids are aliases; fix by separating SQL reference from
dataset id: keep the original tableName as the SQL source in metadata.tableName,
then derive datasetId via the table selection API (e.g. if a table object is
found via findKeplerTableForDatasetId(...) use
get().kepler.tableSelection.getDatasetIdForTable(table) or otherwise fall back
to getUnqualifiedSqlIdentifier(String(tableName)) || tableName), and set
datasets.info.id = datasetId; ensure getKeplerTableLabel(...) still uses the
resolved table object for label logic and do not replace metadata.tableName with
the dataset id.