[python] Improve UX for extend_enumeration_values with non-Arrow types (#3939)

Author: johnkerl

apis/python/src/tiledbsoma/_dataframe.py

@@ -416,7 +416,12 @@ def extend_enumeration_values(
         # These assertions could be done in C++. However, it's easier here
         # to do the exception-type multiplexing, raising ValueError for one
         # thing, TileDBSOMAError for another.
-        for column_name in values.keys():
+        for column_name, values_for_column in values.items():
+            if not isinstance(values_for_column, pa.Array):
+                raise ValueError(
+                    f"value for column name '{column_name}' must be pyarrow.Array: got '{type(values_for_column)}'"
+                )
+
             # As with get_enumeration_values: we are trusting pyarrow to raise
             # KeyError, and raise it with a sufficiently clear error message,
             # when the column name is not present within the schema.
@@ -425,7 +430,7 @@ def extend_enumeration_values(
                 raise KeyError(
                     f"schema column name '{column_name}' is not of dictionary type"
                 )
-            if pa.types.is_dictionary(values[column_name].type):
+            if pa.types.is_dictionary(values_for_column.type):
                 raise ValueError(
                     f"value column name '{column_name}' is of dictionary type: pass its dictionary array instead"
                 )

apis/python/tests/test_dataframe.py

@@ -14,6 +14,7 @@
 import somacore
 from numpy.testing import assert_array_equal
 from pandas.api.types import union_categoricals
+from typeguard import suppress_type_checks
 
 import tiledbsoma as soma
 
@@ -577,6 +578,13 @@ def test_extend_enumeration_values(tmp_path, extend_not_write, ordered):
         ):
             sdf.extend_enumeration_values(xvalues)
 
+        # The values provided must be Arrow arrays. Our unit tests run with typeguard,
+        # but our end users nominally do not -- so we have to ask typeguard to take
+        # a breather here so we can test the UX our users will have.
+        with suppress_type_checks():
+            with pytest.raises(ValueError):
+                sdf.extend_enumeration_values({"string_enum1": ["plain", "strings"]})
+
         # The values provided must all be non-null
         for nvalues in [
             {"string_enum1": pa.array(["greetings", None])},