[exKAZUu]

Why

Variadic type names such as Ts in class C[*Ts] and P in def f[**P]() receive constructor or constant colors from naming heuristics instead of type colors.

Requirements

• required: Improve our forked tree-sitter repositories and address eligible items in dedicated PRs after checking for duplicates; backward compatibility is not required.

• chosen: Highlight variadic names in type positions as types, including declaration parameters, unpacked generic arguments, and star-parameter annotations, matching their syntactic role.

• chosen: Capture identifiers beneath splat_type directly: every production and alias of that node denotes a type position. Restricting its parent to type_parameter misses typed_parameter annotations such as *args: *Ts; ordinary variables and value unpacking use different nodes.

• chosen: Render all highlight fixtures through the existing testCommand runner, which terminates the entire child process group and derives the test timeout from the command timeout; the synchronous helper bounded only the direct CLI child and duplicated timeout settings.

• chosen: Include unpacked generic arguments in direct class-base subscripts, where the grammar emits list_splat rather than splat_type; PEP 646 explicitly uses Generic[*Shape] as a base. Keep the capture scoped to class bases so ordinary value unpacking retains its non-type highlighting.

• chosen: Negative highlight controls assert that ordinary assignments and value unpacking are not types, without pinning unrelated constructor/constant naming heuristics.

• chosen: Keep this PR focused on bare variadic-name captures. Composite unpacked operands already have the same missing type captures on published 2.3.1 (tuple/int/str and typing.Ts controls); their grammar/query repair is a separate upstream-audit item, not a regression introduced here.

• chosen: Handle qualified annotation subscripts in a separate follow-up: typing.Tuple[*Ts], typing.Tuple[int, *Us], and collections.abc.Sequence[*Ts] produce identical missing type captures on published 2.3.1 and this head. This PR repairs the existing splat_type and direct class-base captures.

• chosen: Retain explicit highlight CLI arguments in the three fixture tests. Process-group handling and timeout derivation are shared in testCommand; extracting only the identical argument array and timeout changes no behavior, contract, or verification result and addresses no failing control.

Customer Summary

Variadic type parameters, unpacked type arguments, and star-parameter annotations receive type highlighting.

Technical Summary

The shipped highlight query captures identifiers beneath splat_type, which only denotes type positions. A scoped class-base pattern also covers the list_splat nodes used by unpacked generic bases. A CLI-rendered HTML check covers class, function, type-alias, generic-argument and star-parameter annotation positions alongside ordinary variables.

Testing

The variadic, lazy-import, and default-type highlight fixtures use the existing process-group-aware CLI runner, with the test timeout derived from the command timeout. Eight tests across six files pass, including all rendered-highlight assertions, type-parameter expression controls, and 207 native and Wasm corpus cases after integrating nested generic calls.

The original rendered-highlight regression fails on the baseline and passes with the query change. Generic and qualified generic base-class regressions also fail before their query repair; assignments and value unpacking remain non-type controls without pinning naming-heuristic colors. Class-base variadics follow PEP 646. Canonical generation and bun run verify pass; no code-gauge violations occur in changed code. Full platform checks run in CI.

Notes

Composite unpacked annotations such as *args: *tuple[int, str] and *args: *typing.Ts retain their pre-existing missing type highlights and are queued for a separate fix.

Qualified annotation subscripts such as typing.Tuple[*Ts] retain their pre-existing missing type capture and are queued separately.