fix Autocomplete parameters of class pattern in match-case situation #2830

[asukaminato0721]

Summary

Fixes #2830

adds class-pattern keyword completions from the matched class’s instance attributes, so case A(...): now suggests a= / b= for dataclass fields.

Test Plan

add test

[Copilot]

Pull request overview

Fixes LSP completion behavior for Python match/case class patterns by suggesting keyword names derived from the matched class’s instance attributes (e.g. dataclass fields in case A(...)).

Changes:

• Add completion logic to suggest attr= items when completing inside case <Class>(...).
• Wire the new completion logic into the main completion flow (pattern-match keyword context + generic non-identifier context).
• Add a regression test ensuring dataclass fields (a=, b=) are suggested in case A( ).

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

File	Description
pyrefly/lib/lsp/wasm/completion.rs	Adds match-class keyword completion generation and hooks it into completion dispatch.
pyrefly/lib/test/lsp/completion.rs	Adds a regression test covering dataclass-field keyword completions in class patterns.

---

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

[Copilot]

add_match_class_keyword_completions is invoked in the generic identifier_at == None completion path, which means these attr= suggestions will appear anywhere the cursor is inside a PatternMatchClass node (including the pattern value position after =). This can surface invalid/noisy suggestions while the user is typing the subpattern (e.g. case A(a=|)), where attr= is not syntactically valid without inserting a comma. Consider gating this call to positions where a new keyword can start (e.g. when the previous non-whitespace character is ( or ,, or by checking AST/token context so the cursor is in the keyword-name slot, not inside the value subpattern).

Suggested change

	self.add_match_class_keyword_completions(handle, &nodes, &mut result);
	// Only offer match class keyword completions when the innermost
	// node at the cursor is a PatternMatchClass. This avoids
	// suggesting `attr=` in value positions such as `case A(a=|)`,
	// where a new keyword cannot legally start.
	if let Some(first) = nodes.first()
	&& matches!(first, AnyNodeRef::PatternMatchClass(_))
	{
	self.add_match_class_keyword_completions(handle, &nodes, &mut result);
	}

[Copilot]

The completion detail here uses Type::to_string(), while other completion paths in this file format types with as_lsp_string(LspDisplayMode::Hover) for consistency and readability. Consider using the same formatting helper so match-class keyword completions display types consistently with attribute completions.

Suggested change

	detail: attr.ty.map(|ty| ty.to_string()),
	detail: attr
	.ty
	.map(|ty| ty.as_lsp_string(LspDisplayMode::Hover)),

[github-actions]

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