Reject non-strict types in Pydantic models (#13502)

2 files changed, 426 insertions, 0 deletions

diff --git a/scripts-dev/check_pydantic_models.py b/scripts-dev/check_pydantic_models.py
new file mode 100755
index 0000000000..d0fb811bdb
--- /dev/null
+++ b/scripts-dev/check_pydantic_models.py
@@ -0,0 +1,425 @@
+#! /usr/bin/env python
+# Copyright 2022 The Matrix.org Foundation C.I.C.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+A script which enforces that Synapse always uses strict types when defining a Pydantic
+model.
+
+Pydantic does not yet offer a strict mode, but it is planned for pydantic v2. See
+
+    https://github.com/pydantic/pydantic/issues/1098
+    https://pydantic-docs.helpmanual.io/blog/pydantic-v2/#strict-mode
+
+until then, this script is a best effort to stop us from introducing type coersion bugs
+(like the infamous stringy power levels fixed in room version 10).
+"""
+import argparse
+import contextlib
+import functools
+import importlib
+import logging
+import os
+import pkgutil
+import sys
+import textwrap
+import traceback
+import unittest.mock
+from contextlib import contextmanager
+from typing import Any, Callable, Dict, Generator, List, Set, Type, TypeVar
+
+from parameterized import parameterized
+from pydantic import BaseModel as PydanticBaseModel, conbytes, confloat, conint, constr
+from pydantic.typing import get_args
+from typing_extensions import ParamSpec
+
+logger = logging.getLogger(__name__)
+
+CONSTRAINED_TYPE_FACTORIES_WITH_STRICT_FLAG: List[Callable] = [
+    constr,
+    conbytes,
+    conint,
+    confloat,
+]
+
+TYPES_THAT_PYDANTIC_WILL_COERCE_TO = [
+    str,
+    bytes,
+    int,
+    float,
+    bool,
+]
+
+
+P = ParamSpec("P")
+R = TypeVar("R")
+
+
+class ModelCheckerException(Exception):
+    """Dummy exception. Allows us to detect unwanted types during a module import."""
+
+
+class MissingStrictInConstrainedTypeException(ModelCheckerException):
+    factory_name: str
+
+    def __init__(self, factory_name: str):
+        self.factory_name = factory_name
+
+
+class FieldHasUnwantedTypeException(ModelCheckerException):
+    message: str
+
+    def __init__(self, message: str):
+        self.message = message
+
+
+def make_wrapper(factory: Callable[P, R]) -> Callable[P, R]:
+    """We patch `constr` and friends with wrappers that enforce strict=True."""
+
+    @functools.wraps(factory)
+    def wrapper(*args: P.args, **kwargs: P.kwargs) -> R:
+        # type-ignore: should be redundant once we can use https://github.com/python/mypy/pull/12668
+        if "strict" not in kwargs:  # type: ignore[attr-defined]
+            raise MissingStrictInConstrainedTypeException(factory.__name__)
+        if not kwargs["strict"]:  # type: ignore[index]
+            raise MissingStrictInConstrainedTypeException(factory.__name__)
+        return factory(*args, **kwargs)
+
+    return wrapper
+
+
+def field_type_unwanted(type_: Any) -> bool:
+    """Very rough attempt to detect if a type is unwanted as a Pydantic annotation.
+
+    At present, we exclude types which will coerce, or any generic type involving types
+    which will coerce."""
+    logger.debug("Is %s unwanted?")
+    if type_ in TYPES_THAT_PYDANTIC_WILL_COERCE_TO:
+        logger.debug("yes")
+        return True
+    logger.debug("Maybe. Subargs are %s", get_args(type_))
+    rv = any(field_type_unwanted(t) for t in get_args(type_))
+    logger.debug("Conclusion: %s %s unwanted", type_, "is" if rv else "is not")
+    return rv
+
+
+class PatchedBaseModel(PydanticBaseModel):
+    """A patched version of BaseModel that inspects fields after models are defined.
+
+    We complain loudly if we see an unwanted type.
+
+    Beware: ModelField.type_ is presumably private; this is likely to be very brittle.
+    """
+
+    @classmethod
+    def __init_subclass__(cls: Type[PydanticBaseModel], **kwargs: object):
+        for field in cls.__fields__.values():
+            # Note that field.type_ and field.outer_type are computed based on the
+            # annotation type, see pydantic.fields.ModelField._type_analysis
+            if field_type_unwanted(field.outer_type_):
+                # TODO: this only reports the first bad field. Can we find all bad ones
+                #  and report them all?
+                raise FieldHasUnwantedTypeException(
+                    f"{cls.__module__}.{cls.__qualname__} has field '{field.name}' "
+                    f"with unwanted type `{field.outer_type_}`"
+                )
+
+
+@contextmanager
+def monkeypatch_pydantic() -> Generator[None, None, None]:
+    """Patch pydantic with our snooping versions of BaseModel and the con* functions.
+
+    If the snooping functions see something they don't like, they'll raise a
+    ModelCheckingException instance.
+    """
+    with contextlib.ExitStack() as patches:
+        # Most Synapse code ought to import the patched objects directly from
+        # `pydantic`. But we also patch their containing modules `pydantic.main` and
+        # `pydantic.types` for completeness.
+        patch_basemodel1 = unittest.mock.patch(
+            "pydantic.BaseModel", new=PatchedBaseModel
+        )
+        patch_basemodel2 = unittest.mock.patch(
+            "pydantic.main.BaseModel", new=PatchedBaseModel
+        )
+        patches.enter_context(patch_basemodel1)
+        patches.enter_context(patch_basemodel2)
+        for factory in CONSTRAINED_TYPE_FACTORIES_WITH_STRICT_FLAG:
+            wrapper: Callable = make_wrapper(factory)
+            patch1 = unittest.mock.patch(f"pydantic.{factory.__name__}", new=wrapper)
+            patch2 = unittest.mock.patch(
+                f"pydantic.types.{factory.__name__}", new=wrapper
+            )
+            patches.enter_context(patch1)
+            patches.enter_context(patch2)
+        yield
+
+
+def format_model_checker_exception(e: ModelCheckerException) -> str:
+    """Work out which line of code caused e. Format the line in a human-friendly way."""
+    # TODO. FieldHasUnwantedTypeException gives better error messages. Can we ditch the
+    #   patches of constr() etc, and instead inspect fields to look for ConstrainedStr
+    #   with strict=False? There is some difficulty with the inheritance hierarchy
+    #   because StrictStr < ConstrainedStr < str.
+    if isinstance(e, FieldHasUnwantedTypeException):
+        return e.message
+    elif isinstance(e, MissingStrictInConstrainedTypeException):
+        frame_summary = traceback.extract_tb(e.__traceback__)[-2]
+        return (
+            f"Missing `strict=True` from {e.factory_name}() call \n"
+            + traceback.format_list([frame_summary])[0].lstrip()
+        )
+    else:
+        raise ValueError(f"Unknown exception {e}") from e
+
+
+def lint() -> int:
+    """Try to import all of Synapse and see if we spot any Pydantic type coercions.
+
+    Print any problems, then return a status code suitable for sys.exit."""
+    failures = do_lint()
+    if failures:
+        print(f"Found {len(failures)} problem(s)")
+    for failure in sorted(failures):
+        print(failure)
+    return os.EX_DATAERR if failures else os.EX_OK
+
+
+def do_lint() -> Set[str]:
+    """Try to import all of Synapse and see if we spot any Pydantic type coercions."""
+    failures = set()
+
+    with monkeypatch_pydantic():
+        logger.debug("Importing synapse")
+        try:
+            # TODO: make "synapse" an argument so we can target this script at
+            # a subpackage
+            module = importlib.import_module("synapse")
+        except ModelCheckerException as e:
+            logger.warning("Bad annotation found when importing synapse")
+            failures.add(format_model_checker_exception(e))
+            return failures
+
+        try:
+            logger.debug("Fetching subpackages")
+            module_infos = list(
+                pkgutil.walk_packages(module.__path__, f"{module.__name__}.")
+            )
+        except ModelCheckerException as e:
+            logger.warning("Bad annotation found when looking for modules to import")
+            failures.add(format_model_checker_exception(e))
+            return failures
+
+        for module_info in module_infos:
+            logger.debug("Importing %s", module_info.name)
+            try:
+                importlib.import_module(module_info.name)
+            except ModelCheckerException as e:
+                logger.warning(
+                    f"Bad annotation found when importing {module_info.name}"
+                )
+                failures.add(format_model_checker_exception(e))
+
+    return failures
+
+
+def run_test_snippet(source: str) -> None:
+    """Exec a snippet of source code in an isolated environment."""
+    # To emulate `source` being called at the top level of the module,
+    # the globals and locals we provide apparently have to be the same mapping.
+    #
+    # > Remember that at the module level, globals and locals are the same dictionary.
+    # > If exec gets two separate objects as globals and locals, the code will be
+    # > executed as if it were embedded in a class definition.
+    globals_: Dict[str, object]
+    locals_: Dict[str, object]
+    globals_ = locals_ = {}
+    exec(textwrap.dedent(source), globals_, locals_)
+
+
+class TestConstrainedTypesPatch(unittest.TestCase):
+    def test_expression_without_strict_raises(self) -> None:
+        with monkeypatch_pydantic(), self.assertRaises(ModelCheckerException):
+            run_test_snippet(
+                """
+                from pydantic import constr
+                constr()
+                """
+            )
+
+    def test_called_as_module_attribute_raises(self) -> None:
+        with monkeypatch_pydantic(), self.assertRaises(ModelCheckerException):
+            run_test_snippet(
+                """
+                import pydantic
+                pydantic.constr()
+                """
+            )
+
+    def test_wildcard_import_raises(self) -> None:
+        with monkeypatch_pydantic(), self.assertRaises(ModelCheckerException):
+            run_test_snippet(
+                """
+                from pydantic import *
+                constr()
+                """
+            )
+
+    def test_alternative_import_raises(self) -> None:
+        with monkeypatch_pydantic(), self.assertRaises(ModelCheckerException):
+            run_test_snippet(
+                """
+                from pydantic.types import constr
+                constr()
+                """
+            )
+
+    def test_alternative_import_attribute_raises(self) -> None:
+        with monkeypatch_pydantic(), self.assertRaises(ModelCheckerException):
+            run_test_snippet(
+                """
+                import pydantic.types
+                pydantic.types.constr()
+                """
+            )
+
+    def test_kwarg_but_no_strict_raises(self) -> None:
+        with monkeypatch_pydantic(), self.assertRaises(ModelCheckerException):
+            run_test_snippet(
+                """
+                from pydantic import constr
+                constr(min_length=10)
+                """
+            )
+
+    def test_kwarg_strict_False_raises(self) -> None:
+        with monkeypatch_pydantic(), self.assertRaises(ModelCheckerException):
+            run_test_snippet(
+                """
+                from pydantic import constr
+                constr(strict=False)
+                """
+            )
+
+    def test_kwarg_strict_True_doesnt_raise(self) -> None:
+        with monkeypatch_pydantic():
+            run_test_snippet(
+                """
+                from pydantic import constr
+                constr(strict=True)
+                """
+            )
+
+    def test_annotation_without_strict_raises(self) -> None:
+        with monkeypatch_pydantic(), self.assertRaises(ModelCheckerException):
+            run_test_snippet(
+                """
+                from pydantic import constr
+                x: constr()
+                """
+            )
+
+    def test_field_annotation_without_strict_raises(self) -> None:
+        with monkeypatch_pydantic(), self.assertRaises(ModelCheckerException):
+            run_test_snippet(
+                """
+                from pydantic import BaseModel, conint
+                class C:
+                    x: conint()
+                """
+            )
+
+
+class TestFieldTypeInspection(unittest.TestCase):
+    @parameterized.expand(
+        [
+            ("str",),
+            ("bytes"),
+            ("int",),
+            ("float",),
+            ("bool"),
+            ("Optional[str]",),
+            ("Union[None, str]",),
+            ("List[str]",),
+            ("List[List[str]]",),
+            ("Dict[StrictStr, str]",),
+            ("Dict[str, StrictStr]",),
+            ("TypedDict('D', x=int)",),
+        ]
+    )
+    def test_field_holding_unwanted_type_raises(self, annotation: str) -> None:
+        with monkeypatch_pydantic(), self.assertRaises(ModelCheckerException):
+            run_test_snippet(
+                f"""
+                from typing import *
+                from pydantic import *
+                class C(BaseModel):
+                    f: {annotation}
+                """
+            )
+
+    @parameterized.expand(
+        [
+            ("StrictStr",),
+            ("StrictBytes"),
+            ("StrictInt",),
+            ("StrictFloat",),
+            ("StrictBool"),
+            ("constr(strict=True, min_length=10)",),
+            ("Optional[StrictStr]",),
+            ("Union[None, StrictStr]",),
+            ("List[StrictStr]",),
+            ("List[List[StrictStr]]",),
+            ("Dict[StrictStr, StrictStr]",),
+            ("TypedDict('D', x=StrictInt)",),
+        ]
+    )
+    def test_field_holding_accepted_type_doesnt_raise(self, annotation: str) -> None:
+        with monkeypatch_pydantic():
+            run_test_snippet(
+                f"""
+                from typing import *
+                from pydantic import *
+                class C(BaseModel):
+                    f: {annotation}
+                """
+            )
+
+    def test_field_holding_str_raises_with_alternative_import(self) -> None:
+        with monkeypatch_pydantic(), self.assertRaises(ModelCheckerException):
+            run_test_snippet(
+                """
+                from pydantic.main import BaseModel
+                class C(BaseModel):
+                    f: str
+                """
+            )
+
+
+parser = argparse.ArgumentParser()
+parser.add_argument("mode", choices=["lint", "test"], default="lint", nargs="?")
+parser.add_argument("-v", "--verbose", action="store_true")
+
+
+if __name__ == "__main__":
+    args = parser.parse_args(sys.argv[1:])
+    logging.basicConfig(
+        format="%(asctime)s %(name)s:%(lineno)d %(levelname)s %(message)s",
+        level=logging.DEBUG if args.verbose else logging.INFO,
+    )
+    # suppress logs we don't care about
+    logging.getLogger("xmlschema").setLevel(logging.WARNING)
+    if args.mode == "lint":
+        sys.exit(lint())
+    elif args.mode == "test":
+        unittest.main(argv=sys.argv[:1])
diff --git a/scripts-dev/lint.sh b/scripts-dev/lint.sh
index 377348b107..bf900645b1 100755
--- a/scripts-dev/lint.sh
+++ b/scripts-dev/lint.sh
@@ -106,4 +106,5 @@ isort "${files[@]}"
 python3 -m black "${files[@]}"
 ./scripts-dev/config-lint.sh
 flake8 "${files[@]}"
+./scripts-dev/check_pydantic_models.py lint
 mypy