OpSQLiteDriver: executeAsync result format not handled by extractRowsFromStatementResult, causing silent data loss on SELECT queries

[vitorcamachoo]

• I've validated the bug against the latest version of DB packages

Describe the bug

OpSQLiteDriver in @tanstack/react-native-db-sqlite-persistence silently returns empty arrays for SELECT queries when op-sqlite's executeAsync method is used. This causes the persistence layer to think collections don't exist in collection_registry on app restart, leading to UNIQUE constraint violations.

Root cause

resolveExecuteMethod picks the first available method from [executeAsync, execute, executeRaw, execAsync]. On op-sqlite v14, executeAsync is available and gets selected.

The problem is that op-sqlite's executeAsync returns a different result format than execute:

• execute returns: { rows: Array<Record<string, Scalar>> } (object rows)
• executeAsync returns: { rowsAffected: number, rawRows: unknown[][], columnNames: string[] } (raw columnar format)

extractRowsFromStatementResult only handles the { rows } and { resultRows } shapes. When it receives an executeAsync result:

1. toRowArray(value.rows) -> value.rows is undefined -> returns null
2. toRowArray(value.resultRows) -> value.resultRows is undefined -> returns null
3. hasWriteResultMarker(value) -> "rowsAffected" in value -> true -> returns []

The SELECT result is silently treated as a write result with zero rows, even though the data is present in rawRows.

Impact

This causes a cascade of failures on app restart when the database already has data:

1. ensureCollectionReadyInternal SELECTs from collection_registry -> gets [] instead of the existing row
2. Code takes the INSERT branch -> fails with UNIQUE constraint failed: collection_registry.tombstone_table_name
3. getStreamPosition rejects -> the persistence runtime's ensureStartupMetadataLoaded fails
4. The Electric sync function is never called -> the collection stays in loading state forever with no data

The ALTER TABLE ADD COLUMN errors in ensureInitialized are a separate but related symptom — executeAsync is used for those DDL statements too, and the existing error handling (isDuplicateColumnAddError) works but still logs errors.

To Reproduce

1. Create a collection with persistedCollectionOptions and createReactNativeSQLitePersistence using op-sqlite v14+
2. Let it sync data from an Electric shape
3. Kill the app (full process kill, not hot reload)
4. Relaunch the app
5. The persistence layer crashes on startup — the collection never reaches ready status

Expected behavior

extractRowsFromStatementResult should handle the { rawRows, columnNames } format returned by executeAsync, converting it into the expected Array<Record<string, unknown>> shape. Alternatively, resolveExecuteMethod should prefer execute over executeAsync.

Smartphone (please complete the following information):

• Device: iOS Simulator & physical Android device
• OS: iOS 18, Android 14
• Version: N/A (React Native app, not browser)

Additional context

Package versions:

• @tanstack/db-sqlite-persistence-core: 0.1.9
• @tanstack/react-native-db-sqlite-persistence: 0.1.9
• @op-engineering/op-sqlite: 14.1.4
• React Native (Expo)

Current workaround:

Remove executeAsync from the database handle before passing it to createReactNativeSQLitePersistence, forcing the driver to fall back to execute:

const database = open({ name: 'my-db.sqlite', location: 'default' });
delete (database as any).executeAsync;

const persistence = createReactNativeSQLitePersistence({ database });

Suggested fix:

Either:

1. Add rawRows + columnNames handling to extractRowsFromStatementResult:

function extractRowsFromStatementResult(value) {
  // Handle op-sqlite executeAsync format: { rawRows, columnNames }
  if (Array.isArray(value.rawRows) && Array.isArray(value.columnNames)) {
    return value.rawRows.map((row) =>
      Object.fromEntries(value.columnNames.map((col, i) => [col, row[i]]))
    );
  }
  // ... existing logic
}

2. Or change resolveExecuteMethod to prefer execute over executeAsync, since execute returns the { rows } format the driver already handles.