fix improve enum error suggestion #3128

[asukaminato0721]

Summary

Fixes #3128

The change teaches call-argument diagnostics to suggest an enum member when the provided raw literal exactly matches a single expected enum value.

Test Plan

add test

[github-actions]

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

[Copilot]

Pull request overview

Adds enum-member “did you mean …” suggestions to call-argument type mismatch diagnostics when a provided literal exactly matches an enum’s underlying value (fixing #3128).

Changes:

• Add enum-value → enum-member suggestion helper in enum handling code.
• Extend call-argument type-check error reporting to append an enum-member suggestion line when applicable.
• Add a regression test covering StrEnum argument mismatch suggestions.

Reviewed changes

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

File	Description
pyrefly/lib/test/enums.rs	Adds regression test asserting the new call-argument suggestion text for StrEnum.
pyrefly/lib/alt/class/enums.rs	Implements logic to map a raw literal type to a unique enum member suggestion.
pyrefly/lib/alt/answers_solver.rs	Injects enum-member suggestion lines into call-argument diagnostics during check_type failures.

---

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

[Copilot]

Type::Union handling here short-circuits to None when any union member doesn't produce an enum-member suggestion (due to the ?). That means common expected types like T | None (or T | Other) will never suggest T.A even if the provided literal exactly matches a single enum value. Consider collecting suggestions from union members, ignoring None results, and returning a suggestion only when there is exactly one distinct candidate across the union.

Suggested change

	let candidate = self.suggest_enum_member_for_value(member, got)?;
	if suggestion
	.as_ref()
	.is_some_and(|existing: &String| existing != &candidate)
	{
	return None;
	}
	suggestion = Some(candidate);
	if let Some(candidate) = self.suggest_enum_member_for_value(member, got) {
	if suggestion
	.as_ref()
	.is_some_and(|existing: &String| existing != &candidate)
	{
	return None;
	}
	suggestion = Some(candidate);
	}

[Copilot]

enum_metadata is cloned here but only is_django is used. To avoid unnecessary cloning (which includes a ClassType), consider storing just the needed boolean (e.g., let is_django = ...enum_metadata()?.is_django;) and passing that into enum_literal_to_value_type.

Suggested change

	let enum_metadata = self.get_metadata_for_class(cls).enum_metadata()?.clone();
	let mut suggestion = None;
	for lit in self.get_enum_members(cls) {
	let Lit::Enum(lit_enum) = lit else {
	unreachable!("enum members must be represented as enum literals");
	};
	let value_ty =
	self.enum_literal_to_value_type((*lit_enum).clone(), enum_metadata.is_django);
	let is_django = self.get_metadata_for_class(cls).enum_metadata()?.is_django;
	let mut suggestion = None;
	for lit in self.get_enum_members(cls) {
	let Lit::Enum(lit_enum) = lit else {
	unreachable!("enum members must be represented as enum literals");
	};
	let value_ty = self.enum_literal_to_value_type((*lit_enum).clone(), is_django);

[Copilot]

check_type now duplicates the error-message construction logic from solver::Solver::error in order to inject the enum suggestion. This creates a maintenance risk (future changes to formatting/annotations in the solver can diverge from this path). Consider refactoring to share the common message-building logic (e.g., a helper that returns Vec1<String> given got/want/tcc/subset_error, with an option to append extra lines).

[rchen152]

The Copilot feedback looks legitimate here - specifically, the concerns about union handling and code duplication.