bugfix: column-level coercion is properly implemented

pandera/backends/polars/container.py

@@ -431,6 +431,10 @@ def _coerce_dtype_helper(
         error_handler = ErrorHandler(lazy=True)
 
         config_ctx = get_config_context(validation_depth_default=None)
+
+        # If validation depth involves validating data, use try_coerce since we
+        # want to check actual data values. Otherwise, coerce simply detects
+        # datatype mismatches.
         coerce_fn: str = (
             "try_coerce"
             if config_ctx.validation_depth
@@ -446,9 +450,10 @@ def _coerce_dtype_helper(
                 obj = getattr(schema.dtype, coerce_fn)(obj)
             else:
                 for col_schema in schema.columns.values():
-                    obj = getattr(col_schema.dtype, coerce_fn)(
-                        PolarsData(obj, col_schema.selector)
-                    )
+                    if schema.coerce or col_schema.coerce:
+                        obj = getattr(col_schema.dtype, coerce_fn)(
+                            PolarsData(obj, col_schema.selector)
+                        )
         except ParserError as exc:
             error_handler.collect_error(
                 validation_type(SchemaErrorReason.DATATYPE_COERCION),

tests/polars/test_polars_container.py

@@ -7,6 +7,7 @@
 import pytest
 from hypothesis import given
 from hypothesis import strategies as st
+from polars.testing import assert_frame_equal
 from polars.testing.parametric import column, dataframes
 
 import pandera as pa
@@ -588,3 +589,23 @@ class Model(DataFrameModel):
             {"failure_case": "abc"},
             {"failure_case": "String"},
         ]
+
+
+def test_dataframe_column_level_coerce():
+
+    schema = DataFrameSchema(
+        {
+            "a": Column(int, coerce=True),
+            "b": Column(float, coerce=False),
+        }
+    )
+
+    df = pl.DataFrame({"a": [1.5, 2.2, 3.1], "b": ["1.0", "2.8", "3"]})
+    with pytest.raises(
+        pa.errors.SchemaError,
+        match="expected column 'b' to have type Float64, got String",
+    ):
+        schema.validate(df)
+
+    schema = schema.update_column("b", coerce=True)
+    assert_frame_equal(schema.validate(df), df.cast({"a": int, "b": float}))