Avoid error if origin has a buggy __eq__

[JelleZijlstra]

#419

[AlexWaygood]

This is a good solution. Something we could also do is to check that frame.f_globals.get("__name__") == "typing" before we check the value of origin. We only specifically care about an origin local variable if it's being set in a frame from typing.py, not if it's being set in a frame from site-packages/dirty_equals/_base.py

[JelleZijlstra]

Not sure that applies here, since in the pydantic stacktrace we're still being invoked from typing.py, it's just that origin is some funny object that has a broken __eq__. However, it's definitely possible that someone would invoke typing_extensions. _collect_type_vars directly without going through typing, so I'll add that check.

[AlexWaygood]

Yes, I don't think it applies here, I'm just inclined at this point to err on the side of absolute caution when it comes to this hack :-)

Have you been able to figure out exactly why the NameError is occurring in the pydantic tests? _length_repr looks like it's defined here. My guess would be that something to do with the _get_frame hack means that the __eq__ method can't find it in the globals for some reason, but I haven't been able to quickly find a minimised repro (and probably shouldn't spend too much time on it right now :/ )

[AlexWaygood]

(To be clear I don't think we need to necessarily have a test for this one, since it's pretty obvious that this would fix it. But I'd still be curious to know exactly what the cause of this is, if we can figure it out. We don't need to figure it out before merging this.)

[JelleZijlstra]

Yeah, I don't understand the NameError either. I was able to reproduce the failure in a test:

+    def test_generic_with_broken_eq(self):
+        # See https://github.com/python/typing_extensions/pull/422 for context
+        class BrokenEq(abc.ABCMeta):
+            def __eq__(self, other):
+                if other is typing_extensions.Protocol:
+                    raise TypeError("I'm broken")
+                return False
+
+        class G(Generic[T], metaclass=BrokenEq):
+            pass
+
+        alias = G[int]
+        self.assertIs(get_origin(alias), G)
+        self.assertEqual(get_args(alias), (int,))
+

Traceback (most recent call last):
  File "test_typing_extensions.py", line 6631, in test_generic_with_broken_eq
    alias = G[int]
  File "/Users/jelle/.pyenv/versions/3.8.16/lib/python3.8/typing.py", line 261, in inner
    return func(*args, **kwds)
  File "/Users/jelle/.pyenv/versions/3.8.16/lib/python3.8/typing.py", line 898, in __class_getitem__
    return _GenericAlias(cls, params)
  File "/Users/jelle/.pyenv/versions/3.8.16/lib/python3.8/typing.py", line 672, in __init__
    self.__parameters__ = _collect_type_vars(params)
  File "/Users/jelle/py/typing_extensions/src/typing_extensions.py", line 2994, in _collect_type_vars
    enforce_default_ordering = _has_generic_or_protocol_as_origin()
  File "/Users/jelle/py/typing_extensions/src/typing_extensions.py", line 2961, in _has_generic_or_protocol_as_origin
    return frame.f_locals.get("origin") in (
  File "test_typing_extensions.py", line 6625, in __eq__
    raise TypeError("I'm broken")
TypeError: I'm broken

Though it only reproduces on 3.8 for some reason.

[AlexWaygood]

That test looks good enough to me; it's raising the TypeError in the right place in the stack trace

Though it only reproduces on 3.8 for some reason.

Yeah, and the same with the pydantic test failure, right? Maybe something about frames where it wouldn't lookup globals properly under tracing got fixed in Python 3.9. Who knows.

[JelleZijlstra]

Actually the NameError appears to be because the error happens while dirty_equals is being imported. The code that triggers the error is on line 215 of _sequence.py, and _length_repr is on line 257, so it doesn't exist yet.

[JelleZijlstra]

I guess I found a CPython bug too.

[AlexWaygood]

If we want to be maximally paranoid, we could also make this change:

--- a/src/typing_extensions.py
+++ b/src/typing_extensions.py
@@ -2954,8 +2954,10 @@ if not _PEP_696_IMPLEMENTED:
 def _has_generic_or_protocol_as_origin() -> bool:
     try:
         frame = sys._getframe(2)
-    # not all platforms have sys._getframe()
-    except AttributeError:
+    # - Catch AttributeError: not all Python implementations have sys._getframe()
+    # - Catch ValueError: maybe we're called from an unexpected module
+    #   and the call stack isn't deep enough
+    except (AttributeError, ValueError):

[JelleZijlstra]

% python -c 'from typing_extensions import _has_generic_or_protocol_as_origin; _has_generic_or_protocol_as_origin()'
Traceback (most recent call last):
  File "<string>", line 1, in <module>
  File "/Users/jelle/py/typing_extensions/src/typing_extensions.py", line 2956, in _has_generic_or_protocol_as_origin
    frame = sys._getframe(2)
            ^^^^^^^^^^^^^^^^
ValueError: call stack is not deep enough

I guess I get to fix another critical bug.

[AlexWaygood]

LGTM