fix(io): deserialize parametrized Decimal dtypes from yaml/json (#2434)

* fix(io): deserialize parametrized Decimal dtypes from yaml/json

Signed-off-by: Sebastian Cao <cycsmail@gmail.com>

* match Decimal pattern before the engine fallback

Signed-off-by: Sebastian Cao <cycsmail@gmail.com>

---------

Signed-off-by: Sebastian Cao <cycsmail@gmail.com>

Author: cycsmail

pandera/engines/pandas_engine.py

@@ -577,8 +577,10 @@ class Decimal(DataType, dtypes.Decimal):
     supported by the Python :py:class:`decimal.Decimal` class.
     """
 
-    _exp: decimal.Decimal = dataclasses.field(init=False)
-    _ctx: decimal.Context = dataclasses.field(init=False)
+    # attributes derived from precision, scale, and rounding, so they are
+    # excluded from comparisons: decimal.Context doesn't implement __eq__.
+    _exp: decimal.Decimal = dataclasses.field(init=False, compare=False)
+    _ctx: decimal.Context = dataclasses.field(init=False, compare=False)
 
     def __init__(
         self,

pandera/io/pandas_io.py

@@ -4,6 +4,7 @@
 
 import enum
 import json
+import re
 import warnings
 from collections.abc import Mapping
 from functools import partial
@@ -311,13 +312,37 @@ def handle_stat_dtype(stat):
     return check_instance
 
 
+_DECIMAL_DTYPE_PATTERN = re.compile(
+    r"Decimal\(\s*(\d+)\s*,\s*(\d+)\s*\)", re.IGNORECASE
+)
+
+
+def _deserialize_dtype(serialized_dtype):
+    """Deserialize a dtype, supporting the string representation of
+    parametrized dtypes that the engine doesn't recognize as aliases,
+    e.g. ``"Decimal(28, 0)"``. GH#1165
+
+    Parametrized dtype strings are handled before falling back to
+    :meth:`Engine.dtype`, since the exception raised for unrecognized
+    strings varies across numpy/pandas versions (e.g. numpy raises
+    ``ValueError`` for strings containing commas).
+    """
+    if isinstance(serialized_dtype, str):
+        match = _DECIMAL_DTYPE_PATTERN.fullmatch(serialized_dtype)
+        if match:
+            return pandas_engine.Decimal(
+                precision=int(match.group(1)), scale=int(match.group(2))
+            )
+    return pandas_engine.Engine.dtype(serialized_dtype)
+
+
 def _deserialize_component_stats(serialized_component_stats):
     serialized_component_stats = dict(serialized_component_stats)
     unflatten_component_checks_dict(serialized_component_stats)
 
     dtype = serialized_component_stats.get("dtype")
     if dtype:
-        dtype = pandas_engine.Engine.dtype(dtype)
+        dtype = _deserialize_dtype(dtype)
 
     description = serialized_component_stats.get("description")
     title = serialized_component_stats.get("title")

tests/io/test_pandas_io.py

@@ -1410,6 +1410,28 @@ def test_io_yaml_roundtrip_preserves_drop_invalid_rows():
     assert restored == schema
 
 
+@pytest.mark.skipif(
+    SKIP_YAML_TESTS,
+    reason="pyyaml >= 5.1.0 required",
+)
+def test_io_roundtrip_decimal_dtype():
+    """Test yaml/json roundtrip of parametrized Decimal dtypes, GH#1165."""
+    schema = pandera.DataFrameSchema(
+        columns={
+            "default": pandera.Column(pandera_base.dtypes.Decimal),
+            "parametrized": pandera.Column(pandas_engine.Decimal(10, 2)),
+        }
+    )
+
+    restored = schema.from_yaml(schema.to_yaml())
+    assert restored.columns["parametrized"].dtype == pandas_engine.Decimal(
+        10, 2
+    )
+    assert restored == schema
+
+    assert schema.from_json(schema.to_json()) == schema
+
+
 def test_io_json_with_multiindex_column_labels():
     """Test JSON serialization for schemas with tuple column labels."""
     import json