bugfix/1607: coercion error should correctly report relevant failure …

…cases

Signed-off-by: cosmicBboy <niels.bantilan@gmail.com>

pandera/api/polars/model.py

@@ -57,7 +57,7 @@ def _build_columns(  # pylint:disable=too-many-locals
             try:
                 engine_dtype = pe.Engine.dtype(annotation.raw_annotation)
                 dtype = engine_dtype.type
-            except TypeError as exc:
+            except (TypeError, ValueError) as exc:
                 if annotation.metadata:
                     if field.dtype_kwargs:
                         raise TypeError(

pandera/backends/polars/base.py

@@ -158,7 +158,7 @@ def failure_cases_metadata(
                         failure_case=pl.Series(
                             err.failure_cases.rows(named=True)
                         )
-                    ).select(pl.col.failure_case)
+                    ).select(pl.col.failure_case.struct.json_encode())
                 else:
                     failure_cases_df = err.failure_cases.rename(
                         {err.failure_cases.columns[0]: "failure_case"}

pandera/engines/polars_engine.py

@@ -156,6 +156,8 @@ def try_coerce(self, data_container: PolarsDataContainer) -> pl.LazyFrame:
             is_coercible, failure_cases = polars_coerce_failure_cases(
                 data_container=data_container, type_=self.type
             )
+            if data_container.key:
+                failure_cases = failure_cases.select(data_container.key)
             raise errors.ParserError(
                 f"Could not coerce {_key} LazyFrame with schema "
                 f"{data_container.lazyframe.schema} "

tests/polars/test_polars_config.py

@@ -171,6 +171,4 @@ def test_coerce_validation_depth_none(validation_depth_none, schema):
         try:
             schema.validate(data)
         except pa.errors.SchemaError as exc:
-            assert exc.failure_cases.rows(named=True) == [
-                {"a": "foo", "b": "c"}
-            ]
+            assert exc.failure_cases.rows(named=True) == [{"a": "foo"}]

tests/polars/test_polars_container.py

@@ -564,3 +564,28 @@ def test_dataframe_schema_with_tz_agnostic_dates(time_zone, data):
     if time_zone:
         with pytest.raises(pa.errors.SchemaError):
             schema_tz_sensitive.validate(lf)
+
+
+def test_dataframe_coerce_col_with_null_in_other_column():
+    class Model(DataFrameModel):
+        col1: int = pa.Field(nullable=False, coerce=True)
+        col2: float = pa.Field(nullable=True, coerce=True)
+
+    invalid_lf = pl.DataFrame(
+        {
+            "col1": ["1", "2", "abc"],
+            "col2": [1.0, 2.0, None],
+        }
+    )
+
+    try:
+        print(Model.validate(invalid_lf, lazy=True))
+    except pa.errors.SchemaErrors as exc:
+        failures = exc.failure_cases.select("failure_case").rows(named=True)
+        # two failures should occur:
+        # - Coercing "abc" to int
+        # - Validating that col1 is an integer
+        assert failures == [
+            {"failure_case": "abc"},
+            {"failure_case": "String"},
+        ]