🐛 Make generics work with SQLModel as base class

sqlmodel/main.py

@@ -538,6 +538,12 @@ def __new__(
         config_kwargs = {
             key: kwargs[key] for key in kwargs.keys() & allowed_config_kwargs
         }
+        # Also include pydantic's internal kwargs
+        config_kwargs.update(
+            (key, value)
+            for key, value in kwargs.items()
+            if key.startswith("__pydantic_")
+        )
         new_cls = super().__new__(cls, name, bases, dict_used, **config_kwargs)
         new_cls.__annotations__ = {
             **relationship_annotations,

tests/test_pydantic/test_generic.py

@@ -0,0 +1,83 @@
+from typing import Generic, List, Optional, TypeVar
+
+import pydantic
+import pytest
+from sqlmodel import SQLModel
+
+from tests.conftest import needs_pydanticv2
+
+# Example adapted from
+# https://docs.pydantic.dev/2.10/concepts/models/#generic-models
+DataT = TypeVar("DataT")
+
+
+class DataModel(SQLModel):
+    numbers: List[int]
+    people: List[str]
+
+
+class Response(SQLModel, Generic[DataT]):
+    data: Optional[DataT] = None
+
+
+@needs_pydanticv2
+@pytest.mark.parametrize(
+    ["data_type", "data_value"],
+    [
+        (int, 1),
+        (str, "value"),
+        (DataModel, DataModel(numbers=[1, 2, 3], people=[])),
+        (DataModel, {"numbers": [1, 2, 3], "people": []}),
+    ],
+)
+def test_valid_generics(data_type, data_value):
+    # Should be able to create a model without an error.
+    response = Response[data_type](data=data_value)
+    assert Response[data_type](**response.model_dump()) == response
+
+
+@needs_pydanticv2
+@pytest.mark.parametrize(
+    ["data_type", "data_value", "error_loc", "error_type"],
+    [
+        (
+            str,
+            1,
+            ("data",),
+            "string_type",
+        ),
+        (
+            int,
+            "some-string",
+            ("data",),
+            "int_parsing",
+        ),
+        (
+            DataModel,
+            "some-string",
+            ("data",),
+            "model_attributes_type",
+        ),
+        (
+            DataModel,
+            {"numbers": [1, 2, "unexpected string"], "people": []},
+            ("data", "numbers", 2),
+            "int_parsing",
+        ),
+    ],
+)
+def test_invalid_generics(data_type, data_value, error_loc, error_type):
+    with pytest.raises(pydantic.ValidationError) as raised:
+        Response[data_type](data=data_value)
+    [error_dict] = raised.value.errors()
+    assert error_dict["loc"] == error_loc
+    assert error_dict["type"] == error_type
+
+
+@needs_pydanticv2
+def test_generic_json_schema():
+    schema = Response[DataModel].model_json_schema()
+    # Should have referenced the schema in $defs
+    assert schema["properties"]["data"]["anyOf"][0] == {"$ref": "#/$defs/DataModel"}
+    # Schema in $defs should match DataModel's schema.
+    assert schema["$defs"]["DataModel"] == DataModel.model_json_schema()