[ty] Add go-to definition for dataclass/TypedDict/NamedTuple fields from keyword arguments by Viicos · Pull Request #24378 · astral-sh/ruff

Viicos · 2026-04-02T21:43:59Z

Also fixes a small unrelated typo.

Summary

Test Plan

…eyword arguments

Viicos · 2026-04-02T21:46:48Z

+    if resolved_definitions.is_empty()
+        && let Some(class_literal) = func_type.as_class_literal()
+        && let Some(static_class_literal) = class_literal.as_static()
+        && let Some(field_policy) = CodeGeneratorKind::from_class(db, class_literal, None)


Note: this is a bit of a leaky abstraction, because it assumes CodeGeneratorKind can only be created for classes that support this goto logic. Maybe we could also guard with:

Suggested change

&& let Some(field_policy) = CodeGeneratorKind::from_class(db, class_literal, None)

&& (static_class_literal.is_dataclass_like(db) || static_class_literal.is_typed_dict(db) || static_class_literal.has_named_tuple_class_in_mro(db))

&& let Some(field_policy) = CodeGeneratorKind::from_class(db, class_literal, None)

astral-sh-bot · 2026-04-02T21:46:54Z

Typing conformance results

No changes detected ✅

Current numbers

The percentage of diagnostics emitted that were expected errors held steady at 87.92%. The percentage of expected errors that received a diagnostic held steady at 83.11%. The number of fully passing files held steady at 78/133.

astral-sh-bot · 2026-04-02T21:47:25Z

Memory usage report

Memory usage unchanged ✅

Viicos · 2026-04-02T21:48:31Z

+    // Go to field definitions if the type is a dataclass-like type, a named tuple or a typed dict. Only do so if no definition was found,
+    // as having existing definition(s) would mean a custom constructor is defined and as such takes priority:


Actually, I think we should explicitly check if a custom constructor is set, because with the following:

@dataclass class A: a: int def __init__(self, b: int) -> None: ... A(a<CURSOR>=1)

It would still go to the a field. Maybe not worth the added complexity?

astral-sh-bot · 2026-04-02T21:50:24Z

`ecosystem-analyzer` results

No diagnostic changes detected ✅

Full report with detailed diff (timing results)

MichaReiser · 2026-04-07T13:02:50Z

Thank you. I'm not sure this is the right approach as it results in needing multiple special casing:

Here, to make go to parameter work
In hover: we want to show the constructor signature when hovering a typed dict constructor, including all its optional arguments
In signature completion: The same as for hover, we want to suggest all optional parameters.

I'm not entirely sure how to integrate this best. Ideally, we'd do something when resolving the constructor call. @oconnor663 I think you worked on TypedDict. Do you think we could synthesize the constructor function somehow during type inference and if so, where would we add this?

AlexWaygood · 2026-04-07T13:16:10Z

Do you think we could synthesize the constructor function somehow during type inference and if so, where would we add this?

We already do this -- here for dataclasses:

ruff/crates/ty_python_semantic/src/types/class/static_literal.rs

Lines 1331 to 1340 in f4c3807

    
           (CodeGeneratorKind::DataclassLike(_), "__init__") => { 
        
               if !self.has_dataclass_param(db, field_policy, DataclassFlags::INIT) { 
        
                   return None; 
        
               } 
        
               let self_parameter = Parameter::positional_or_keyword(Name::new_static("self")) 
        
                   // TODO: could be `Self`. 
        
                   .with_annotated_type(instance_ty); 
        
               signature_from_fields(vec![self_parameter], Type::none(db)) 
        
           }

Here for class-based NamedTuples:

ruff/crates/ty_python_semantic/src/types/class/static_literal.rs

Lines 1341 to 1386 in f4c3807

    
           ( 
        
               CodeGeneratorKind::NamedTuple, 
        
               "__new__" | "__init__" | "_replace" | "__replace__" | "_fields", 
        
           ) if self.namedtuple_base_has_unknown_fields(db) => { 
        
               // When the namedtuple base has unknown fields, fall back to NamedTupleFallback 
        
               // which has generic signatures that accept any arguments. 
        
               KnownClass::NamedTupleFallback 
        
                   .to_class_literal(db) 
        
                   .as_class_literal()? 
        
                   .as_static()? 
        
                   .own_class_member(db, inherited_generic_context, None, name) 
        
                   .ignore_possibly_undefined() 
        
                   .map(|ty| { 
        
                       ty.apply_type_mapping( 
        
                           db, 
        
                           &TypeMapping::ReplaceSelf { 
        
                               new_upper_bound: instance_ty, 
        
                           }, 
        
                           TypeContext::default(), 
        
                       ) 
        
                   }) 
        
           } 
        
           ( 
        
               CodeGeneratorKind::NamedTuple, 
        
               "__new__" | "_replace" | "__replace__" | "_fields" | "__slots__", 
        
           ) => { 
        
               let fields = self.fields(db, specialization, field_policy); 
        
               let fields_iter = fields.iter().map(|(name, field)| { 
        
                   let default_ty = match &field.kind { 
        
                       FieldKind::NamedTuple { default_ty } => *default_ty, 
        
                       _ => None, 
        
                   }; 
        
                   NamedTupleField { 
        
                       name: name.clone(), 
        
                       ty: field.declared_ty, 
        
                       default: default_ty, 
        
                   } 
        
               }); 
        
               synthesize_namedtuple_class_member( 
        
                   db, 
        
                   name, 
        
                   instance_ty, 
        
                   fields_iter, 
        
                   specialization.map(|s| s.generic_context(db)), 
        
               ) 
        
           }

Here for functional namedtuples:

ruff/crates/ty_python_semantic/src/types/class/named_tuple.rs

Lines 386 to 412 in f4c3807

    
               let result = synthesize_namedtuple_class_member( 
        
                   db, 
        
                   name, 
        
                   instance_ty, 
        
                   self.fields(db).iter().cloned(), 
        
                   None, 
        
               ); 
        
               // For fallback members from NamedTupleFallback, apply type mapping to handle 
        
               // `Self` types. The explicitly synthesized members (__new__, _fields, _replace, 
        
               // __replace__) don't need this mapping. 
        
               if matches!( 
        
                   name, 
        
                   "__new__" | "_fields" | "_replace" | "__replace__" | "__slots__" 
        
               ) { 
        
                   result 
        
               } else { 
        
                   result.map(|ty| { 
        
                       ty.apply_type_mapping( 
        
                           db, 
        
                           &TypeMapping::ReplaceSelf { 
        
                               new_upper_bound: instance_ty, 
        
                           }, 
        
                           TypeContext::default(), 
        
                       ) 
        
                   }) 
        
               } 
        
           }

I'm not sure where TypedDict constructors get synthesized off the top of my head, but there must be a similar hook somewhere in type inference

Viicos · 2026-04-12T11:40:07Z

Yes to clarify, this doesn't require the synthesized signatures to be available at all. The definitions_for_keyword_argument() function I extended here already did this for non-synthesized (i.e. explicitly defined in code) constructors:

class CustomClass:
    def __init__(self, a: int) -> None: ...

# Trigger go-to on a keyworg argument will go to `a` parameter in `CustomClass.__init__()`:
CustomClass(a=1)

Playground

For synthesized methods, this condition in definitions_for_keyword_argument():

ruff/crates/ty_python_semantic/src/types/ide_support.rs

Lines 528 to 530 in d85aac2

    
           if let Some(parameter_range) = 
        
               find_parameter_range(&function_node.parameters, keyword_name_str) 
        
           {

is never reached because synthesized constructors parameters don't have any range. We end up with an empty resolved_definitions vec, and so my added code path gets triggered. It only relies on the field definitions, not the constructor.

MichaReiser · 2026-04-14T07:07:17Z

That makes sense. I still think it would be nice to avoid special casing typed dict in the IDE code and instead expose the necessary information in synthesized constructors.

For example, what if we add a origin_definition: Option<Definition> to Parameter and set it when synthesizing the constructor? We could even consider doing this for non-synthesized constructors too (could it then even be required). @carljm does this sound reasonable?

carljm · 2026-04-14T17:50:39Z

Tracking an optional definition with Parameter seems reasonable to me. I suspect we can't make it required, because e.g. a Callable[...] type annotation defines parameters, but there is no associated Definition available.

MichaReiser · 2026-04-28T10:37:02Z

I went ahead and implemented my described approach in #24897

Add go-to definition for dataclass/TypedDict/NamedTuple fields from k…

f74fc43

…eyword arguments

Viicos requested review from AlexWaygood, BurntSushi, Gankra, MichaReiser, carljm, dcreager, ibraheemdev and sharkdp as code owners April 2, 2026 21:44

astral-sh-bot Bot added the ty Multi-file analysis & type inference label Apr 2, 2026

astral-sh-bot Bot assigned BurntSushi Apr 2, 2026

fix typo

fa7784d

Viicos commented Apr 2, 2026

View reviewed changes

carljm removed their request for review April 3, 2026 17:17

MichaReiser assigned MichaReiser and unassigned BurntSushi Apr 3, 2026

sharkdp changed the title ~~Add go-to definition for dataclass/TypedDict/NamedTuple fields from keyword arguments~~ [ty] Add go-to definition for dataclass/TypedDict/NamedTuple fields from keyword arguments Apr 7, 2026

MichaReiser mentioned this pull request Apr 9, 2026

[ty] Synthesize __init__ for TypedDict #24476

Merged

Viicos force-pushed the goto-dc-td-nt-kw-args branch from 83a6ef7 to fa7784d Compare April 12, 2026 11:43

MichaReiser mentioned this pull request Apr 28, 2026

Find references for synthesized fields astral-sh/ty#3369

Open

MichaReiser closed this Apr 28, 2026

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[ty] Add go-to definition for dataclass/TypedDict/NamedTuple fields from keyword arguments#24378

[ty] Add go-to definition for dataclass/TypedDict/NamedTuple fields from keyword arguments#24378
Viicos wants to merge 2 commits intoastral-sh:mainfrom
Viicos:goto-dc-td-nt-kw-args

Viicos commented Apr 2, 2026

Uh oh!

Viicos Apr 2, 2026

Uh oh!

astral-sh-bot Bot commented Apr 2, 2026 •

edited

Loading

Uh oh!

astral-sh-bot Bot commented Apr 2, 2026 •

edited

Loading

Uh oh!

Viicos Apr 2, 2026

Uh oh!

astral-sh-bot Bot commented Apr 2, 2026 •

edited

Loading

Uh oh!

MichaReiser commented Apr 7, 2026

Uh oh!

AlexWaygood commented Apr 7, 2026

Uh oh!

Viicos commented Apr 12, 2026

Uh oh!

MichaReiser commented Apr 14, 2026

Uh oh!

carljm commented Apr 14, 2026

Uh oh!

MichaReiser commented Apr 28, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

	&& let Some(field_policy) = CodeGeneratorKind::from_class(db, class_literal, None)
	&& (static_class_literal.is_dataclass_like(db) \|\| static_class_literal.is_typed_dict(db) \|\| static_class_literal.has_named_tuple_class_in_mro(db))
	&& let Some(field_policy) = CodeGeneratorKind::from_class(db, class_literal, None)

		// Go to field definitions if the type is a dataclass-like type, a named tuple or a typed dict. Only do so if no definition was found,
		// as having existing definition(s) would mean a custom constructor is defined and as such takes priority:

Conversation

Viicos commented Apr 2, 2026

Summary

Test Plan

Uh oh!

Viicos Apr 2, 2026

Choose a reason for hiding this comment

Uh oh!

astral-sh-bot Bot commented Apr 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Typing conformance results

No changes detected ✅

Uh oh!

astral-sh-bot Bot commented Apr 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Memory usage report

Uh oh!

Viicos Apr 2, 2026

Choose a reason for hiding this comment

Uh oh!

astral-sh-bot Bot commented Apr 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

ecosystem-analyzer results

Uh oh!

MichaReiser commented Apr 7, 2026

Uh oh!

AlexWaygood commented Apr 7, 2026

Uh oh!

Viicos commented Apr 12, 2026

Uh oh!

MichaReiser commented Apr 14, 2026

Uh oh!

carljm commented Apr 14, 2026

Uh oh!

MichaReiser commented Apr 28, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

astral-sh-bot Bot commented Apr 2, 2026 •

edited

Loading

astral-sh-bot Bot commented Apr 2, 2026 •

edited

Loading

astral-sh-bot Bot commented Apr 2, 2026 •

edited

Loading

`ecosystem-analyzer` results