pyrefly-shape-extensions

Runtime helpers for Pyrefly tensor shape annotations.

This package provides the lightweight shape_extensions module used by Pyrefly's tensor shape stubs. It defines runtime no-op versions of the shape typing primitives so annotations such as Tensor[B, T], IntVar("B"), and assert_shape(x.shape, (2, 3)) can be evaluated by Python while Pyrefly uses the corresponding stubs for static shape checking.

NamedInts and CaptureNamedInts carry named axis lengths from **kwargs into shape rules for einops-like APIs.

RegularNestedList[Shape, Domain] represents regular (non-jagged) nested list literals; for example, [[1, 2], [3, 4]] binds Shape to [2, 2]. Unsupported containers and irregular literals use ordinary typing and any fallback overload supplied by the consumer.

IntTupleOrList[Values] is a stub-authoring parameter type for APIs that accept integer tuples and lists. A direct, unstarred list literal such as [2, 3] binds Values to IntTuple[2, 3]; an existing or starred list remains gradual, while a direct literal containing a non-integer is rejected.

The package is versioned in lockstep with Pyrefly.

Portable shape annotations

Shaped[T, "..."] is an alias for typing.Annotated, so other type checkers read T and ignore the shape string. Inside a @shape_vars function or class, Pyrefly also reads the string in annotations, casts, type aliases, and class bases. Use @shape_vars("") for a literal shape with no declared dimensions.

Legacy type aliases cannot capture dimensions declared on an enclosing @shape_vars function or class. For example, an Alias: TypeAlias = Shaped[Array, "[N]"] inside a @shape_vars("N") definition reports that N is not in scope for the alias. This is the same restriction that applies when legacy type aliases capture ordinary enclosing type parameters. Use the Shaped annotation directly in that scope; literal-only aliases are supported.

A defaulted @shape_vars("N") dimension after a *Ts class parameter is not supported: Pyrefly reports the declaration and can mistake a trailing ordinary type argument for the dimension. Pyrefly accepts @shape_vars("N", required=True) class Required[*Ts] with an explicit shape, as in Required[int, str, 3]; Required[int, str] still treats str as a dimension. This explicit specialization is specific to Pyrefly, not a portable workaround for other type checkers.

At runtime, Shaped[Base, "..."] can appear as a class base because Annotated resolves to Base. Pyrefly reads its shape. In Pyright 1.1.414, the base is rejected and inherited attributes and methods are inferred as Unknown downstream. Do not rely on this base spelling when Pyright users need inherited signatures.