fix: preserve _marimo_row_id when clamping table columns (#8029)

marimo/_plugins/ui/_impl/table.py

@@ -1353,7 +1353,13 @@ def clamp_rows_and_columns(manager: TableManager[Any]) -> str:
 
             # Do not clamp if max_columns is None
             if max_columns is not None and len(column_names) > max_columns:
-                data = data.select_columns(column_names[:max_columns])
+                columns_to_select = column_names[:max_columns]
+                # Always include _marimo_row_id so the frontend can use
+                # stable row IDs for selection, even when columns are clamped.
+                # Without this, filtering + selecting returns wrong rows.
+                if self._has_stable_row_id:
+                    columns_to_select = [INDEX_COLUMN_NAME] + columns_to_select
+                data = data.select_columns(columns_to_select)
 
             try:
                 return data.to_json_str(self._format_mapping)

tests/_plugins/ui/_impl/test_table.py

@@ -1582,7 +1582,9 @@ def test_column_clamping_with_dataframes(df: Any):
     assert table._component_args["max-columns"] == DEFAULT_MAX_COLUMNS
     json_data = json.loads(table._component_args["data"])
     headers = json_data[0].keys()
-    assert len(headers) == DEFAULT_MAX_COLUMNS  # 50 columns
+    assert (
+        len(headers) == DEFAULT_MAX_COLUMNS + 1
+    )  # 50 columns + _marimo_row_id
     # Field types are not clamped
     assert len(table._component_args["field-types"]) == 60
 
@@ -1594,7 +1596,7 @@ def test_column_clamping_with_dataframes(df: Any):
     assert table._component_args["max-columns"] == 40
     json_data = json.loads(table._component_args["data"])
     headers = json_data[0].keys()
-    assert len(headers) == 40  # 40 columns
+    assert len(headers) == 41  # 40 columns + _marimo_row_id
     # Field types aren't clamped
     assert len(table._component_args["field-types"]) == 60
 
@@ -1611,6 +1613,57 @@ def test_column_clamping_with_dataframes(df: Any):
     assert len(table._component_args["field-types"]) == 60
 
 
+@pytest.mark.parametrize(
+    "df",
+    create_dataframes(
+        {
+            "person": ["Alice", "Bob", "Charlie"],
+            "age": [20, 30, 40],
+            **{f"col{i}": [1, 2, 3] for i in range(49)},
+        },
+        exclude=NON_EAGER_LIBS,
+    ),
+)
+def test_selection_with_clamped_columns_and_filter(df: Any):
+    """Regression test for #8029: selection returns wrong row when table
+    has more columns than max_columns and is filtered."""
+    import narwhals as nw
+
+    table = ui.table(df, max_columns=50)
+
+    # The data sent to frontend should include _marimo_row_id
+    # even when columns are clamped
+    json_data = json.loads(table._component_args["data"])
+    assert INDEX_COLUMN_NAME in json_data[0]
+
+    # Apply a filter to show only rows where age >= 30
+    # (should return Bob and Charlie, with _marimo_row_id 1 and 2)
+    search_args = SearchTableArgs(
+        page_size=10,
+        page_number=0,
+        filters=[Condition(column_id="age", operator=">=", value=30)],
+    )
+    response = table._search(search_args)
+    result_data = json.loads(response.data)
+
+    # Filtered data should still include _marimo_row_id
+    assert INDEX_COLUMN_NAME in result_data[0]
+    assert len(result_data) == 2  # Bob and Charlie
+
+    # Select row with _marimo_row_id=2 (Charlie)
+    value = table._convert_value(["2"])
+    assert not isinstance(value, nw.DataFrame)
+    nw_value = nw.from_native(value)
+    assert nw_value["person"][0] == "Charlie"
+    assert nw_value["age"][0] == 40
+
+    # Select row with _marimo_row_id=1 (Bob)
+    value = table._convert_value(["1"])
+    nw_value = nw.from_native(value)
+    assert nw_value["person"][0] == "Bob"
+    assert nw_value["age"][0] == 30
+
+
 @pytest.mark.skipif(
     not DependencyManager.pandas.has(), reason="Pandas not installed"
 )