mypyc-related stubtest crash

[srittau]

Bug Report

In typeshed CI stubtest crashes when run against fpdf2 2.7.5 and the current typeshed stubs. It works fine with fpdf2 2.7.4. Please see python/typeshed#10533, or to be more specific the traceback at https://github.com/python/typeshed/actions/runs/5930009431/job/16078844153?pr=10533:

Traceback (most recent call last):
  File "/opt/hostedtoolcache/Python/3.10.12/x64/lib/python3.10/runpy.py", line 196, in _run_module_as_main
    return _run_code(code, main_globals, None,
  File "/opt/hostedtoolcache/Python/3.10.12/x64/lib/python3.10/runpy.py", line 86, in _run_code
    exec(code, run_globals)
  File "/tmp/tmpo2a_8672/lib/python3.10/site-packages/mypy/stubtest.py", line 1891, in <module>
    sys.exit(main())
  File "/tmp/tmpo2a_8672/lib/python3.10/site-packages/mypy/stubtest.py", line 1887, in main
    return test_stubs(parse_options(sys.argv[1:]))
  File "/tmp/tmpo2a_8672/lib/python3.10/site-packages/mypy/stubtest.py", line 1760, in test_stubs
    for error in test_module(module):
  File "/tmp/tmpo2a_8672/lib/python3.10/site-packages/mypy/stubtest.py", line 223, in test_module
    yield from verify(stub, runtime, [module_name])
  File "/tmp/tmpo2a_8672/lib/python3.10/site-packages/mypy/stubtest.py", line 401, in verify_mypyfile
    yield from verify(stub_entry, runtime_entry, object_path + [entry])
  File "/tmp/tmpo2a_8672/lib/python3.10/site-packages/mypy/stubtest.py", line 543, in verify_typeinfo
    yield from verify(stub_to_verify, runtime_attr, object_path + [entry])
  File "/tmp/tmpo2a_8672/lib/python3.10/site-packages/mypy/stubtest.py", line [100](https://github.com/python/typeshed/actions/runs/5930009431/job/16078844153?pr=10533#step:5:101)6, in verify_funcitem
    for message in _verify_signature(stub_sig, runtime_sig, function_name=stub.name):
  File "/tmp/tmpo2a_8672/lib/python3.10/site-packages/mypy/stubtest.py", line 858, in _verify_signature
    yield from _verify_arg_default_value(stub_arg, runtime_arg)
  File "/tmp/tmpo2a_8672/lib/python3.10/site-packages/mypy/stubtest.py", line 632, in _verify_arg_default_value
    runtime_type = get_mypy_type_of_runtime_value(runtime_arg.default)
  File "/tmp/tmpo2a_8672/lib/python3.10/site-packages/mypy/stubtest.py", line 1563, in get_mypy_type_of_runtime_value
    return mypy.types.LiteralType(value=value, fallback=fallback)
  File "mypy/types.py", line 2584, in __init__
TypeError: union[int, str, bool, float] object expected; got None

This error seems to be generated by mypyc. I can't reproduce it locally, possibly because mypyc is not used?

Your Environment

• Mypy version used: 1.5.1
• Mypy command-line flags:

/tmp/tmpo2a_8672/bin/pip install fpdf2[]==2.7.5 mypy==1.5.1
MYPYPATH=/home/runner/work/typeshed/typeshed/stubs/fpdf2:/home/runner/work/typeshed/typeshed/stubs/Pillow /tmp/tmpo2a_8672/bin/python -m mypy.stubtest --custom-typeshed-dir /home/runner/work/typeshed/typeshed fpdf --allowlist /home/runner/work/typeshed/typeshed/stubs/fpdf2/@tests/stubtest_allowlist.txt

• Python version used: 3.10.12

[AlexWaygood]

I can reproduce this crash locally on Python 3.9 and Python 3.10, but not Python 3.11

[AlexWaygood]

I can reproduce this crash locally on Python 3.9 and Python 3.10, but not Python 3.11

And, as @srittau says, the crash only occurs with a compiled version of mypy, which makes it tricky to debug!

[AlexWaygood]

Looks like the TypeError is raised when mypy is trying to validate the default value typeshed gives for one of the parameters in this __init__ method: https://github.com/py-pdf/fpdf2/blob/ff55d983a0b6d601a78628a45747234fbca59e26/fpdf/encryption.py#L127-L135

[AlexWaygood]

I confirmed locally that applying this diff to typeshed's stubs makes the crash go away:

--- a/stubs/fpdf2/fpdf/encryption.pyi
+++ b/stubs/fpdf2/fpdf/encryption.pyi
@@ -65,15 +65,6 @@ class StandardSecurityHandler:
     k: str
     u: str

-    def __init__(
-        self,
-        fpdf: FPDF,
-        owner_password: str,
-        user_password: str | None = None,
-        permission: Incomplete | None = None,
-        encryption_method: EncryptionMethod | None = None,
-        encrypt_metadata: bool = False,
-    ) -> None: ...
     def generate_passwords(self, file_id) -> None: ...
     def get_encryption_obj(self) -> EncryptionDictionary: ...
     def encrypt(self, text: str | bytes | bytearray, obj_id) -> bytes: ...

[AlexWaygood]

The issue is this call here:

mypy/mypy/stubtest.py

Line 1563 in 7141d6b

return mypy.types.LiteralType(value=value, fallback=fallback)

When attempting to validate the permission parameter on fpdf.encryption.StandardSecurityHandler.__init__, stubtest ends up trying to call mypy.types.LiteralType(value=None, fallback=fpdf.enums.AccessPermission). That goes against the type annotation for mypy.types.LiteralType.__init__, which says that value should only ever be int, str, bool or float:

mypy/mypy/types.py

Lines 2560 to 2578 in 7141d6b

	class LiteralType(ProperType):
	"""The type of a Literal instance. Literal[Value]
	A Literal always consists of:
	1. A native Python object corresponding to the contained inner value
	2. A fallback for this Literal. The fallback also corresponds to the
	parent type this Literal subtypes.
	For example, 'Literal[42]' is represented as
	'LiteralType(value=42, fallback=instance_of_int)'
	As another example, `Literal[Color.RED]` (where Color is an enum) is
	represented as `LiteralType(value="RED", fallback=instance_of_color)'.
	"""
	__slots__ = ("value", "fallback", "_hash")
	def __init__(

Inserting a reveal_type() call indicates that mypy believes that value can only be an int or a str here, which is why the type violation is only being caught by the runtime checks that mypyc inserts when compiling mypy:

--- a/mypy/stubtest.py
+++ b/mypy/stubtest.py
@@ -1560,6 +1560,7 @@ def get_mypy_type_of_runtime_value(runtime: Any) -> mypy.types.Type | None:
     else:
         return fallback

+    reveal_type(value)
     return mypy.types.LiteralType(value=value, fallback=fallback)

> python runtests.py self
run self: ['C:\\Users\\alexw\\coding\\mypy\\venv\\Scripts\\python.exe', '-m', 'mypy', '--config-file', 'mypy_self_check.ini', '-p', 'mypy', '-p', 'mypyc']
mypy\stubtest.py:1563: note: Revealed type is "Union[builtins.int, builtins.str]"

[AlexWaygood]

Here is the reason why stubtest crashes on Python <=3.10 but not on Python 3.11. On Python 3.10:

Python 3.10.12 | packaged by Anaconda, Inc. | (main, Jul  5 2023, 19:01:18) [MSC v.1916 64 bit (AMD64)] on win32
Type "help", "copyright", "credits" or "license" for more information.
>>> import fpdf.encryption
>>> fpdf.encryption.AccessPermission.all().name is None
True

On Python 3.11:

Python 3.11.4 | packaged by Anaconda, Inc. | (main, Jul  5 2023, 13:47:18) [MSC v.1916 64 bit (AMD64)] on win32
Type "help", "copyright", "credits" or "license" for more information.
>>> import fpdf.encryption
>>> fpdf.encryption.AccessPermission.all().name
'PRINT_LOW_RES|MODIFY|COPY|ANNOTATION|FILL_FORMS|COPY_FOR_ACCESSIBILITY|ASSEMBLE|PRINT_HIGH_RES'

isinstance(fpdf.encryption.AccessPermission.all(), enum.Enum) evaluates to True, which means that value is set to fpdf.encryption.AccessPermission.all().name here:

mypy/mypy/stubtest.py

Lines 1556 to 1557 in 7141d6b

	elif isinstance(runtime, enum.Enum):
	value = runtime.name

But on Python <=3.10, fpdf.encryption.AccessPermission.all().name is None, which triggers the crash.

The fpdf.encryption.AccessPermissions IntFlag is defined here: https://github.com/py-pdf/fpdf2/blob/34ff166de50a757c8cb32d6f047f0901e052d814/fpdf/enums.py#L801-L834

[AlexWaygood]

The reason why mypy doesn't pick up that runtime.name could be None here during the self-check is because enum.Flag violates the Liskov Substitution Principle. The name property can only ever be a str on an instance of an enum.Enum enumeration... unless that enumeration is a subclass of IntFlag, in which case it could also be None: https://github.com/python/typeshed/blob/32b750b6aa2396cdaa757da16da8cd8fb627c09f/stdlib/enum.pyi#L216-L217

[AlexWaygood]

I proposed a fix in #15933. The fix is a one-line change, though finding a way to reproduce the crash in the context of stubtest's test suite was a slight challenge 😄