Pyrefly v1.2.0-dev.2

Release date: July 10, 2026

About dev releases
Dev releases (versions like X.Y.Z-dev.N) are non-stable snapshots cut periodically from trunk. They give early adopters a chance to try in-progress features and surface issues before the next stable release, but they don't carry the same stability or compatibility guarantees as a stable release — don't pin production projects to a dev version.

Pyrefly v1.2.0-dev.2 bundles 158 commits from 23 contributors.

---

✨ New & Improved

Type Checking

• Functions decorated with functools.singledispatch now type-check calls using the signature of the function you decorated, catching argument errors like f("bad") when that function expects int. Registered implementations are also checked against that function's first parameter, so their dispatch types must be compatible.
• Generic singledispatch functions now infer their generic types from call arguments instead of reporting Unknown.
• Pyrefly can now warn when a function declared to return a concrete type returns Any, with separate diagnostics for explicit and inferred Any returns. These warnings are off by default for backward compatibility; you can enable them explicitly, or let Pyrefly migrate the setting from mypy when warn_return_any is enabled.
• isinstance(x, type) now preserves type arguments when narrowing unions, so type[int] | str narrows to type[int] instead of bare type.
• Class decorators annotated as returning type[Any] now make the decorated class behave dynamically, matching decorators that can substantially change the class at runtime.
• Class methods can now access generic class attributes through cls, fixing false missing-attribute errors on abstract base classes.
• Membership checks with Any/unknown items now invoke the container's __contains__ method and return bool instead of Unknown.
• copy.replace is now type-checked like dataclass replacement, so Pyrefly validates immutable updates more accurately.

Language Server

• Hover tooltips for union methods now look at each union member, so x: A | B; x.foo() correctly shows foo as a method when both A.foo and B.foo are methods.
• Hovering over augmented assignment operators like += now shows the __iadd__ definition and type, or the right-hand side literal type when that is the best available information.
• Attribute hovers like c.x now show where the attribute's type was narrowed instead of unrelated first-use info.
• Hover and type lookups now work correctly in notebook cells after the first.
• Inlay hints, document symbols, references, and diagnostic grouping now work correctly in notebook code cells that follow markdown cells.
• Auto-import completions now respect the python.analysis.autoImportCompletions setting, skipping auto-imports when disabled.
• Value completions are now skipped when typing a keyword-argument name (e.g., func(foo=1, ba|)), and statement keywords like try/while are no longer offered in expression contexts.
• Deprecated stdlib typing aliases (e.g., typing.List) are now ranked below their collections.abc equivalents in auto-import completions.

Bazel Integration

• A new, experimental bazel-check command adds Bazel-focused type-checking for projects that provide source information through JSON manifests. It is intended to collaborate with an in-progress rules_pyrefly Bazel rule implementation.

Tensor Shape Types

• Shape-aware stubs can now express methods whose types follow another method on the same object. This lets nn.Module.__call__ type-check like the module's forward method.
• SizeTuple is now the public way to write shape parameters, with support for compact syntax like ndarray[[2, 3], dtype] and unpacking with *Elements[SizeTuple].
• Torch and NumPy shape stubs now use SizeTuple, making it easier to model APIs involving multiple tensors with different shapes.
• Symbolic shape variables now use SymVar, written as N: SymVar in PEP 695 syntax. Shape arithmetic is now limited to these symbolic shape variables.

---

🐛 Bug fixes

We closed 26 bug reports this release 👏

• #897: Fixed an issue where attribute narrowing with optional values failed to preserve the narrowed type across conditional branches, causing false positives when accessing attributes after a null check.
• #3585: Fixed isinstance(value, type) narrowing for type[int] | str unions, which previously collapsed to bare type instead of preserving type[int].
• #3584: Fixed isinstance(value, type) narrowing for TypeForm[object], which previously produced type & TypeForm[object] instead of the correct type[int] after an issubclass check.
• #2689: Fixed subsequent isinstance(cls, type) and issubclass(cls, upper_bound) checks erasing TypeVar information, so type[F] is preserved through assertion chains instead of widening to the bound.
• #3806: Added no-any-return error kinds to detect returning Any from functions declared to return concrete types, with explicit and implicit variants and support for migrating mypy's warn_return_any setting.
• #4024: Fixed pyrefly coverage check --strict incorrectly labeling fully-Any symbols (e.g., def f(x: Any) -> Any) as coverage-partial instead of coverage-missing.
• #4062: Fixed pytorch-efficiency-lints not recognizing Tensor in common stub layouts.
• #4083: Fixed tuple(xs) losing track of tuple length when xs is already a tuple, now preserving known shapes like tuple[int, int] instead of widening to tuple[int, ...].
• #4093: Fixed an internal error when accessing __class__.__setattr__ on a class object, caused by missing metaclass handling.
• #4073: Fixed unpacked TypeVarTuple arguments like tuple[int, *Ts] not being assignable to varargs type tuple[int, *Ts] due to an internal mismatch in how Pyrefly represented the tuple types.
• And more! #4025, #4019, #4020, #4021, #3730, #3732, #3997, #4018, #4023, #3949, #4022, #3920, #4092, #3078, #3960, #4064

Thank you to all our contributors who found these bugs and reported them! Did you know this is one of the most helpful contributions you can make to an open-source project? If you find any bugs in Pyrefly, we want to know about them! Please open a bug report issue here.

---

📦 Upgrade

pip install --upgrade pyrefly==1.2.0-dev.2

How to safely upgrade your codebase

Upgrading the version of Pyrefly you're using or a third-party library you depend on can reveal new type errors in your code. Fixing them all at once is often unrealistic. We've written scripts to help you temporarily silence them. After upgrading, follow these steps:

1. pyrefly check --suppress-errors
2. Run your code formatter of choice
3. pyrefly check --remove-unused-ignores
4. Repeat until you achieve a clean formatting run and a clean type check.

This will add # pyrefly: ignore comments to your code, enabling you to silence errors and return to them later. This can make the process of upgrading a large codebase much more manageable.

Read more about error suppressions in the Pyrefly documentation.

---

🖊️ Contributors this release

@stroxler, @grievejia, @shobhitmehro, @asukaminato0721, @jorenham, @kinto0, @yangdanny97, David Tolnay, @ericzhonghou, @connernilsen, @qnox, @mikeleppane, @dillydill123, @arthaud, @alexander-beedie, @NathanTempest, @fhoehle, @WilliamK112, @xaskii, @durvesh1992, @lennardwalter, @kz357, @nitishagar

---

Please note: These release notes summarize major updates and features. For brevity, not all individual commits are listed.