[asukaminato0721]

Summary

Fixes #1886
fixes #2838

The dict-key matcher now treats Ruff’s error-recovery {""} shape as a placeholder key position, then recovers the contextual TypedDict from the enclosing assignment/call when the literal itself only infers as set[str].

also added a specialized dict(...) kwargs path so contextual TypedDict fields show up as name= / age= completions in the existing kwargs flow.

Test Plan

add test

[Copilot]

Pull request overview

Improves LSP autocompletion for TypedDict construction in Python code, especially when parsing incomplete dict literals that Ruff represents as sets (e.g. {""}), and when constructing dicts via dict(...) kwargs.

Changes:

• Treat Ruff’s {""} error-recovery shape (ExprSet) as a dict-key placeholder to enable key completion.
• Recover contextual TypedDict type information from surrounding assignment/call sites when the literal itself doesn’t infer as a TypedDict.
• Add dict(...)-constructor kwargs completions for contextual TypedDict fields, plus tests covering these cases.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 2 comments.

File	Description
pyrefly/lib/state/lsp/dict_completions.rs	Extends dict-key completion to handle ExprSet placeholders and adds contextual type recovery + dict(...) kwargs completion.
pyrefly/lib/lsp/wasm/completion.rs	Wires in the new dict(...)-constructor kwargs completion and exposes expected_call_argument_type for reuse.
pyrefly/lib/test/lsp/completion.rs	Adds regression tests for incomplete literal {""}, call-argument {""}, and dict( kwargs completion.

---

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

[Copilot]

contextual_typed_dict_type is now used for both dict literals and key-access contexts. For key access (u[""] / d[""]), falling back to the enclosing call/assignment can produce incorrect TypedDict key suggestions even when the base expression is not a TypedDict (e.g. cfg: Config = data[""] could start suggesting Config fields for the key into data). Consider restricting the call/assignment fallback to the dict-literal construction path only (or gating it behind context.base_expr().is_none()), and keep key-access completion typed keys derived solely from the base expression’s own type trace.

[Copilot]

add_typed_dict_constructor_kwargs_completions triggers solely on Expr::Name("dict"). This will also fire if dict is shadowed (e.g., a local function/variable named dict), producing TypedDict-field completions in calls that aren’t the built-in dict constructor. Consider resolving the callee via bindings/exports (or otherwise confirming it refers to builtins.dict) before adding these completions.

[stroxler]

This PR needs a rebase before it could be merged.

In addition, there are a number of correctness issues we need to fix:

1. Dictionary completion becomes quadratic — [lines 715–716](

   pyrefly/pyrefly/lib/state/lsp/dict_completions.rs

   Lines 715 to 716 in 23015d2

   	if should_update {
   	best = Some((priority, dist, dict.clone(), literal.clone()));

   ).
   Each improved candidate clones the entire dictionary AST. Completing the last key in a 1,000-entry ordinary dictionary took ~375 ms versus ~1.4 ms on the parent, in the same local development build. Retain references while selecting the winner and clone once afterward.

2. Real sets lose DataFrame column completions — [lines 651–658](

   pyrefly/pyrefly/lib/state/lsp/dict_completions.rs

   Lines 651 to 658 in 23015d2

   	AnyNodeRef::ExprSet(set) => Self::best_string_literal_in_set(set, position).map(
   	|(priority, dist, range, literal)| {
   	(
   	priority,
   	dist,
   	DictKeyLiteralContext::RecoveredSetLiteral { range, literal },
   	)
   	},

   ).
   df.select({"|"}) and pandas df.filter(items={"|"}) previously offered column names; now they offer none. The set matcher takes precedence over the DataFrame matcher even without a TypedDict context. Preserve that fallback unless TypedDict recovery actually applies.

3. Expected types are applied to the wrong expression — [lines 948–953](

   pyrefly/pyrefly/lib/state/lsp/dict_completions.rs

   Lines 948 to 953 in 23015d2

   	self.get_type_trace(handle, range)
   	.filter(Self::type_contains_typed_dict)
   	.or_else(|| {
   	self.expected_call_argument_type(handle, position)
   	.filter(Self::type_contains_typed_dict)
   	})

   ).
   With lookup: dict[str, Config], take(lookup["|"]) incorrectly offers Config’s fields as lookup keys. Nested literals also pick up outer TypedDict fields. Expected-type recovery needs to belong to the constructed expression, not merely the cursor’s enclosing call.

4. Constructor kwargs target unrelated calls — [lines 1180–1186](

   pyrefly/pyrefly/lib/state/lsp/dict_completions.rs

   Lines 1180 to 1186 in 23015d2

   	.find_map(|node| match node {
   	AnyNodeRef::ExprCall(call)
   	if call.arguments.range.contains_inclusive(position)
   	&& matches!(
   	call.func.as_ref(),
   	Expr::Name(name) if name.id.as_str() == "dict"
   	) =>

   ).
   cfg: Config = dict(name=str(|)) offers name= and age= inside str(). A user-defined function named dict also receives those suggestions despite not accepting them. Require the innermost owning call and resolve its callee to builtin dict.

5. Invalid keyword names are emitted — [lines 1203–1210](

   pyrefly/pyrefly/lib/state/lsp/dict_completions.rs

   Lines 1203 to 1210 in 23015d2

   	for (label, ty) in typed_keys {
   	completions.push(RankedCompletion::new(CompletionItem {
   	label: format!("{label}="),
   	detail: Some(ty.to_string()),
   	kind: Some(CompletionItemKind::VARIABLE),
   	..Default::default()
   	}));
   	}

   ).
   Functional TypedDict fields such as "class" and "two words" produce class= and two words=, neither valid Python. Route these through the existing push_kwarg_completion validation.

[github-actions]

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

[stroxler]

Review automatically exported from Phabricator review in Meta.

[meta-codesync]

This pull request has been merged in cf94755.