PEP 849 Draft Proposes More Expressive Type Expressions in Python 3.16

PEP 849 proposes a new annotation format that allows arbitrary Python expressions within type annotations. This format instructs annotation functions to return the AST and namespaces of the annotation, enabling runtime consumers to evaluate them into typing objects. This fundamentally expands the kinds of expressions that can be assigned meaning within the typing specification.

The core mechanism involves modifying `__annotate__` functions to return `AnnotationAST` objects when called with a new `Format.AST` option, containing the AST and namespace. A new `typing.get_type_annotations` function would internally handle this, evaluating the ASTs to construct familiar type objects, thus decoupling type expression evaluation from standard Python semantics for improved runtime introspection.

Currently, Python's type annotations are restricted to a small subset of expressions due to runtime introspection limitations. Examples include the inability to use `1 | 2 | 3` for literal unions, or ternary expressions for conditional types. The PEP provides groundwork to support such intuitive type expressions, which are currently not feasible because their runtime evaluation semantics are incompatible with reflection needs.

The proposal suggests `__annotate__` could populate namespaces and load compact binary AST data, then use an intrinsic to build `AnnotationAST` objects. Additionally, a `create_annotate_function` in the `annotationlib` module is proposed to help users synthesize annotate functions for this more complex format, building on existing functionality in the `dataclasses` module.

The PEP acknowledges the need for careful consideration of performance impact, especially since type annotations are optional. It aims to analyze the effects on three user groups: those who don't use type annotations, those who use them for type checkers/linters without runtime introspection, and those who utilize runtime introspection.