Fix false positive on attribute access for TypeVar bounded by type

[yangdanny97]

Summary:
When a TypeVar has bound=type, pyrefly internally represents the bound as
type[Any], which resolves to AttributeBase1::TypeAny during attribute
lookup. The quantified TypeVar attribute resolution only handled
AttributeBase1::ClassInstance, causing any other variant (including TypeAny)
to fall back to object. This produced false positives like
"Object of class object has no attribute __name__" when accessing
attributes that exist on the type class.

The fix extracts the bound-to-ClassType mapping into a quantified_bound_class
helper that handles both ClassInstance (the common case) and TypeAny
(the bound=type case, mapping to builtins.type). This is applied to all
four sites where quantified bounds are resolved: both T and type[T] paths,
for both Bound and Constraints restrictions.

Fixes #2614

Differential Revision: D95419868

[meta-codesync]

@yangdanny97 has exported this pull request. If you are a Meta employee, you can view the originating Diff in D95419868.

[github-actions]

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

kornia (https://github.com/kornia/kornia)
- ERROR kornia/core/check.py:309:37-49: Object of class `object` has no attribute `__name__` [missing-attribute]
- ERROR kornia/core/check.py:309:96-106: Object of class `object` has no attribute `__name__` [missing-attribute]

pydantic (https://github.com/pydantic/pydantic)
- ERROR pydantic/config.py:1290:9-35: Object of class `object` has no attribute `__pydantic_config__` [missing-attribute]
+ ERROR pydantic/config.py:1290:9-35: Object of class `type` has no attribute `__pydantic_config__` [missing-attribute]

strawberry (https://github.com/strawberry-graphql/strawberry)
- ERROR strawberry/types/object_type.py:133:34-46: Object of class `object` has no attribute `__name__` [missing-attribute]

prefect (https://github.com/PrefectHQ/prefect)
- ERROR src/prefect/utilities/dispatch.py:40:25-32: Object of class `object` has no attribute `mro` [missing-attribute]
- ERROR src/prefect/utilities/dispatch.py:129:9-48: Object of class `object` has no attribute `__pydantic_init_subclass_original__` [missing-attribute]
+ ERROR src/prefect/utilities/dispatch.py:129:9-48: Object of class `type` has no attribute `__pydantic_init_subclass_original__` [missing-attribute]
- ERROR src/prefect/utilities/dispatch.py:132:9-39: Object of class `object` has no attribute `__pydantic_init_subclass__` [missing-attribute]
+ ERROR src/prefect/utilities/dispatch.py:132:9-39: Object of class `type` has no attribute `__pydantic_init_subclass__` [missing-attribute]
- ERROR src/prefect/utilities/dispatch.py:134:9-39: Object of class `object` has no attribute `__init_subclass_original__` [missing-attribute]
+ ERROR src/prefect/utilities/dispatch.py:134:9-39: Object of class `type` has no attribute `__init_subclass_original__` [missing-attribute]
- ERROR src/prefect/utilities/dispatch.py:163:45-52: Object of class `object` has no attribute `mro` [missing-attribute]

[github-actions]

Primer Diff Classification

✅ 3 improvement(s) | ➖ 1 neutral | 4 project(s) total

3 improvement(s) across kornia, strawberry, prefect.

Project	Verdict	Changes	Error Kinds	Root Cause
kornia	✅ Improvement	-2	missing-attribute	quantified_bound_class()
pydantic	➖ Neutral	+1, -1	missing-attribute
strawberry	✅ Improvement	-1	missing-attribute	quantified_bound_class()
prefect	✅ Improvement	+3, -5	missing-attribute	quantified_bound_class()

Detailed analysis

✅ Improvement (3)

kornia (-2)

The removed errors were false positives caused by a type inference bug in pyrefly. Looking at line 269, T = TypeVar('T', bound=type) creates a TypeVar bounded by the type class. In the KORNIA_CHECK_TYPE function on line 309, when typ has type T, pyrefly should understand that typ has all attributes of type, including __name__. However, pyrefly was incorrectly resolving TypeVars with bound=type to object instead of type, leading to false missing-attribute errors. The PR description confirms this was a known issue (#2614) where pyrefly's internal representation of bound=type as type[Any] wasn't being handled properly in attribute lookup, causing fallback to object. The fix correctly maps TypeAny variants to builtins.type during quantified bound resolution.
Attribution: The fix in quantified_bound_class() in pyrefly/lib/alt/attr.rs now handles AttributeBase1::TypeAny (which represents type[Any] from bound=type) by mapping it to builtins.type, allowing proper attribute resolution. Previously, only AttributeBase1::ClassInstance was handled, causing TypeVars with bound=type to fall back to object.

strawberry (-1)

This is an improvement. The error was a false positive caused by pyrefly's attribute resolution failing for TypeVars with bound=type. Looking at line 27, we see T = TypeVar('T', bound=builtins.type), and line 133 accesses cls.__name__ where cls: T. Since T is bounded by type, and all type objects have a __name__ attribute, this access is completely valid. The PR description explains that pyrefly was incorrectly representing bound=type as type[Any] which resolved to AttributeBase1::TypeAny, but the quantified TypeVar resolution only handled AttributeBase1::ClassInstance, causing it to fall back to object and produce the false positive. The fix properly maps the TypeAny case to builtins.type so attribute lookup works correctly.
Attribution: The fix in quantified_bound_class() method in pyrefly/lib/alt/attr.rs now handles AttributeBase1::TypeAny (which represents type[Any] from bound=type) by mapping it to builtins.type, allowing proper attribute resolution instead of falling back to object.

prefect (+3, -5)

The removed errors were false positives where pyrefly incorrectly resolved TypeVar('T', bound=type[Any]) to object instead of type, causing it to claim that type-specific attributes like mro() and __init_subclass__ were missing. The PR fixed this by properly handling the bound=type case in attribute resolution. The new errors appear to be a different issue with dynamic attribute tracking, but the core fix (removing false positives about missing type attributes) represents an improvement in type checker accuracy.
Attribution: The change to quantified_bound_class() in pyrefly/lib/alt/attr.rs fixed the TypeVar bound resolution by handling AttributeBase1::TypeAny (which represents type[Any] from bound=type) and mapping it to builtins.type instead of falling back to object. This resolved the false positive missing-attribute errors on the object class.

➖ Neutral (1)

pydantic (+1, -1)

Same errors at same locations with same error kinds — message wording changed, no behavioral impact.

---

Was this helpful? React with 👍 or 👎

_{Classification by primer-classifier (1 heuristic, 3 LLM)}

[grievejia]

Review automatically exported from Phabricator review in Meta.

[meta-codesync]

This pull request has been merged in e311fea.