feat(native): port Objective-C extractor to Rust (#1106)

* feat(native): port Objective-C extractor to Rust

Mirrors `src/extractors/objc.ts` in `crates/codegraph-core/src/extractors/objc.rs`.
Adds the `tree-sitter-objc` dependency, wires `LanguageKind::ObjC` (`.m`) in
the Rust `parser_registry` and `file_collector`, adds `.m` to
`NATIVE_SUPPORTED_EXTENSIONS` on the JS side, and registers
`OBJC_AST_TYPES` / `OBJC_AST_CONFIG` so the native and WASM engines extract
identical `ast_nodes` for Objective-C source.

Handles class interfaces / implementations (with `: Superclass`), protocols,
instance and class method declarations/definitions with multi-part selectors
assembled from leading identifiers and `method_parameter` children, C-level
function declarations/definitions, `#import`/`@import` imports, and message
expression call extraction.

* fix(extractors): address ObjC review feedback

- Use `module` field (not `path`) for `@import` in Rust to mirror the JS
  extractor and match the tree-sitter-objc `module_import` grammar field.
- Drop the unreachable `implementation_definition` branch from
  `collect_class_members` — it is only invoked from `handle_class_interface`
  and `class_interface` nodes do not contain `implementation_definition`
  children.
- Qualify category methods with `(Category)` in the JS extractor so its
  output matches Rust for `@interface Foo (Cat)` / `@implementation Foo (Cat)`
  when the grammar emits `class_interface`/`class_implementation` rather
  than dedicated `category_interface` nodes. Two categories can declare
  same-named methods, so the qualified parent disambiguates the symbols.
- Document the `.m` extension collision with MATLAB/Octave in the file
  collector since `.m` files are unconditionally routed to the ObjC parser.

* fix: correct expected count after merge dropped two entries (#1106)

The native-drop-classification test asserts the count of WASM-only
languages. The Clojure PR (#1097) removed src/c.clj and decremented
the count from 11 to 10. This PR removes src/k.m (now natively
supported via ObjC). After the merge both entries are removed, so
the assertion needs to be 9, not 10.

* fix(extractors): align JS ObjC engine with native for @import and calls

Three engine-parity gaps Greptile flagged on the WASM side that broke the
"identical output" goal:

- @import statements: tree-sitter-objc v3 emits `module_import` not
  `import_declaration`, so the JS dispatch arm never matched and every
  `@import Foundation;` was silently dropped. Accept both node types.
- C-style calls (printf, CGContextFillRect, …): the grammar lacks a
  `function` field on `call_expression`, so the named-field lookup always
  misses. Rust falls back to the first identifier child; JS did not, so
  every C call was dropped. Add the same fallback.
- Message expressions: the grammar tags each keyword identifier with the
  `method` field rather than exposing a `selector` field, so the JS
  selector lookup misfired for multi-keyword selectors. Assemble the
  selector from `method` children with `:` joining, matching
  `build_message_selector` in the Rust extractor.

Also expose `fieldNameForChild` on the `TreeSitterNode` type and add three
JS extractor tests covering the new parity behaviour.

* fix: bump EXPECTED_LEN after merging Solidity LanguageKind (#1106)

* fix(extractors): align JS ObjC engine with native for v3 grammar (#1106)

The v3 tree-sitter-objc grammar emits flat 'identifier' + 'method_parameter'
children directly under method nodes (no 'keyword_selector' wrapper) and
nests property names under 'struct_declaration > struct_declarator >
[pointer_declarator >] identifier' rather than exposing a 'name' field. The
JS extractor was still following the old grammar shape, which silently
dropped multi-keyword method definitions, never populated parameter
children, and omitted all '@Property' members from class children.

- buildSelector: assemble selector from flat 'identifier'+'method_parameter'
  children directly under the method node, matching build_selector in
  crates/codegraph-core/src/extractors/objc.rs.
- extractMethodParams: iterate 'method_parameter' children directly and use
  the last 'identifier' child as the parameter name (mirrors
  extract_method_params in the Rust extractor).
- collectClassMembers: extract '@Property' names via a deep identifier walk
  under 'struct_declaration > struct_declarator' (mirrors
  extract_property_name in the Rust extractor).

Added three regression tests covering keyword method definitions with
parameters, and pointer + non-pointer '@Property' member names.

* fix(extractors): handle parameterized_arguments for ObjC protocol adoption (#1106)

The JS extractor was looking up 'protocol_qualifiers' which doesn't exist in
tree-sitter-objc v3 (the grammar wraps adopted protocols in
'parameterized_arguments' instead). As a result every '@interface Foo : NSObject
<Bar, Baz>' silently dropped its 'implements' relations on the JS side, while
the Rust extractor correctly extracted them. Mirror handle_class_interface in
crates/codegraph-core/src/extractors/objc.rs (parameterized_arguments + the
type_name > type_identifier nesting) and add a regression test.

Author: carlos-alm

Cargo.lock

@@ -36,6 +36,7 @@ tree-sitter-dart = "0.0.4"
 tree-sitter-zig = "1"
 tree-sitter-haskell = "0.23"
 tree-sitter-ocaml = "0.24"
+tree-sitter-objc = "3"
 tree-sitter-gleam = "1"
 tree-sitter-julia = "0.23"
 tree-sitter-clojure-orchard = "0.2"

crates/codegraph-core/Cargo.toml

@@ -374,6 +374,19 @@ pub const OCAML_AST_CONFIG: LangAstConfig = LangAstConfig {
     string_prefixes: &[],
 };
 
+/// Objective-C string literals use the `@"..."` prefix. The shared
+/// `build_string_node` strips a leading `@` before applying prefixes, so we
+/// don't need to list it explicitly here.
+pub const OBJC_AST_CONFIG: LangAstConfig = LangAstConfig {
+    new_types: &[],
+    throw_types: &["throw_statement"],
+    await_types: &[],
+    string_types: &["string_literal"],
+    regex_types: &[],
+    quote_chars: &['"'],
+    string_prefixes: &[],
+};
+
 pub const GLEAM_AST_CONFIG: LangAstConfig = LangAstConfig {
     new_types: &[],
     throw_types: &[],

crates/codegraph-core/src/extractors/helpers.rs

@@ -18,6 +18,7 @@ pub mod javascript;
 pub mod julia;
 pub mod kotlin;
 pub mod lua;
+pub mod objc;
 pub mod ocaml;
 pub mod php;
 pub mod python;
@@ -137,6 +138,9 @@ pub fn extract_symbols_with_opts(
         LanguageKind::Ocaml | LanguageKind::OcamlInterface => {
             ocaml::OcamlExtractor.extract_with_opts(tree, source, file_path, include_ast_nodes)
         }
+        LanguageKind::ObjC => {
+            objc::ObjCExtractor.extract_with_opts(tree, source, file_path, include_ast_nodes)
+        }
         LanguageKind::Gleam => {
             gleam::GleamExtractor.extract_with_opts(tree, source, file_path, include_ast_nodes)
         }