Stdlib: add many missing __hash__ and __eq__ methods

[AlexWaygood]

Pyright has recently been trying to add some sophisticated understanding of how hashability works at runtime. Namely, if a class overrides __eq__ at runtime but doesn't override __hash__, the runtime views instances of the class as unhashable: if you override __eq__, you also have to explicitly override __hash__ as well if you want instances to be hashable. However, pyright is currently unable to implement its new feature, because of the fact that many classes in the stdlib override __hash__ at runtime, but don't currently include __hash__ in typeshed's stubs: microsoft/pyright#5513. Let's fix that.

I also added many missing __eq__ methods. As well as their relevance to hashability, these are also used as a heuristic for mypy's overlapping-equality check. Like with __hash__, therefore, it's important to include these in the stub even if the signature on a subclass is the same as on a superclass.

I found all these with the following stubtest patch:

diff --git a/mypy/stubtest.py b/mypy/stubtest.py
index f06faa962..5921bd49d 100644
--- a/mypy/stubtest.py
+++ b/mypy/stubtest.py
@@ -512,6 +512,8 @@ def verify_typeinfo(
     if is_runtime_typeddict:
         to_check.discard("__new__")

+    errored_entries: set[str] = set()
+
     for entry in sorted(to_check):
         mangled_entry = entry
         if entry.startswith("__") and not entry.endswith("__"):
@@ -536,7 +538,43 @@ def verify_typeinfo(
             and isinstance(runtime_attr, types.WrapperDescriptorType)
             and is_dunder(mangled_entry, exclude_special=True)
         ):
-            yield from verify(stub_to_verify, runtime_attr, object_path + [entry])
+            for error in verify(stub_to_verify, runtime_attr, object_path + [entry]):
+                errored_entries.add(entry)
+                yield error
+
+    if "__hash__" not in errored_entries:
+        if (
+            runtime.__dict__.get("__hash__") is not None
+            and "__hash__" not in stub.names
+        ):
+            yield Error(
+                object_path + ["__hash__"],
+                "overrides __hash__ at runtime but not in the stub",
+                stub_object=MISSING,
+                runtime_object=runtime.__hash__
+            )
+        if "__hash__" not in runtime.__dict__ and "__hash__" in stub.names:
+            yield Error(
+                object_path + ["__hash__"],
+                "overrides __hash__ in the stub but not at runtime",
+                stub_object=stub.names["__hash__"].node,
+                runtime_object=MISSING
+            )
+    if "__eq__" not in errored_entries:
+        if "__eq__" in runtime.__dict__ and "__eq__" not in stub.names:
+            yield Error(
+                object_path + ["__eq__"],
+                "overrides __eq__ at runtime but not in the stub",
+                stub_object=MISSING,
+                runtime_object=runtime.__eq__
+            )
+        if "__eq__" not in runtime.__dict__ and "__eq__" in stub.names:
+            yield Error(
+                object_path + ["__eq__"],
+                "overrides __eq__ in the stub but not at runtime",
+                stub_object=stub.names["__eq__"].node,
+                runtime_object=MISSING
+            )

Cc. @TheTripleV

[github-actions]

Diff from mypy_primer, showing the effect of this PR on open source code:

streamlit (https://github.com/streamlit/streamlit)
- lib/tests/streamlit/elements/date_input_test.py: note: In member "test_range_session_state" of class "DateInputTest":
- lib/tests/streamlit/elements/date_input_test.py:174:16: error: Non-overlapping equality check (left operand type: "Union[date, Tuple[], Tuple[date], Tuple[date, date]]", right operand type: "List[datetime]")  [comparison-overlap]

[srittau]

Thanks!