fix Preserve arity for tuple literals with * #2861

[asukaminato0721]

Summary

Fixes #2861

special-casing tuple multiplication so known integer repeats preserve concrete tuple arity instead of falling through to the generic tuple[T, ...] stub.

respects the existing 256-element literal tuple cap by falling back once expansion would get too large.

Test Plan

add test

[Copilot]

Pull request overview

Fixes type inference for tuple multiplication when the repeat count is a known integer literal, preserving concrete tuple arity (up to the existing 256-element expansion cap) instead of degrading to an unbounded tuple[T, ...].

Changes:

• Add tuple-repeat inference for tuple * Literal[int] / Literal[int] * tuple, expanding concrete tuples when the resulting length stays within MAX_TUPLE_LENGTH.
• Re-export the tuple literal expansion cap (MAX_TUPLE_LENGTH) for reuse in operator inference.
• Add regression tests covering tuple repetition (including 0, negative, and overflow-to-fallback cases).

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 3 comments.

File	Description
pyrefly/lib/alt/operators.rs	Adds try_tuple_repeat and wires it into * inference for tuples using literal repeat counts.
pyrefly/lib/alt/expr.rs	Exposes MAX_TUPLE_LENGTH for cross-module use.
pyrefly/lib/test/tuple.rs	Adds test coverage for tuple repetition type inference and the 256-element cap fallback behavior.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

try_tuple_repeat only recognizes Literal[int] as the repeat count. In Python, bool is a subtype of int and tuple * True/False is valid; right now those cases will fall through to the generic __mul__ typing and won't preserve arity. Consider extracting the count via Literal.value.as_index_i64() (covers both Int and Bool) instead of matching only Lit::Int.

Suggested change

	Type::Literal(box Literal {
	value: Lit::Int(n), ..
	}) => n.as_i64(),
	Type::Literal(box Literal { value, .. }) => value.as_index_i64(),

[Copilot]

The tuple-repeat special case is implemented for normal binary * in binop_types, but *= goes through augassign_infer and currently has no analogous handling (unlike +=, which special-cases tuples). This can make x *= 2 produce a less precise type than x = x * 2 for tuples; consider adding a Operator::Mult branch in augassign_infer that reuses try_tuple_repeat.

[Copilot]

MAX_TUPLE_LENGTH is a compile-time constant but is declared as a static and is now being re-exported for cross-module use. Elsewhere in the repo similar limits are typically const (e.g. pyrefly/lib/alt/answers_solver.rs:1979, pyrefly/lib/binding/scope.rs:519). Consider making this pub(crate) const MAX_TUPLE_LENGTH: usize = 256; to match that pattern and avoid introducing an addressable global unnecessarily.

Suggested change

	pub(crate) static MAX_TUPLE_LENGTH: usize = 256;
	pub(crate) const MAX_TUPLE_LENGTH: usize = 256;

[github-actions]

According to mypy_primer, this change doesn't affect type check results on a corpus of open source code. ✅