feat: add ArkTS (.ets) language support with tree-sitter parser

[Happy-26]

Summary

Add ArkTS (HarmonyOS) language support to codebase-memory-mcp, enabling full AST-based indexing of .ets files via the tree-sitter-arkts parser.

Resolves #760

Background

ArkTS is TypeScript-like language used for HarmonyOS application development (file extension .ets). It extends TypeScript with decorators (@Component, @Entry, @Prop, etc.), component declarations, and UI builder methods (build()).

Changes

Language Registration

• Add CBM_LANG_ARKTS enum in cbm.h
• Add .ets → CBM_LANG_ARKTS mapping in language.c
• Add ArkTS display name configuration

Parser Integration

• Add grammar_arkts.c wrapping the vendored tree-sitter-arkts parser
• Add vendored/grammars/arkts/parser.c and tree_sitter/parser.h

Node Type Configuration (lang_specs.c)

Define ArkTS-specific node types matching the actual tree-sitter-arkts parser output:

• Function types: function_declaration, function_expression, arrow_function, method_declaration, constructor_declaration, build_method, decorated_function_declaration, ui_builder_arrow_function
• Class types: class_declaration, enum_declaration, interface_declaration, type_declaration, component_declaration, decorated_export_declaration

Name Extraction Fallback (extract_defs.c)

The tree-sitter-arkts parser does not define a name field in its AST node field map, causing ts_node_child_by_field_name(node, "name") to return null. To handle this:

• Add cbm_find_child_by_kind fallback in cbm_resolve_func_name for ArkTS function_declaration and decorated_function_declaration
• Synthesize "build" name for build_method nodes (these have no name child at all)
• Add ArkTS identifier fallback in extract_class_def and compute_class_qn
• Add component_body to class body traversal (find_class_body, push_class_body_children)
• Add ArkTS method name fallback in resolve_method_name
• Add build_method synthesis in extract_class_methods
• Include CBM_LANG_ARKTS in walk_defs descend_into_func check
• Add Component label for component_declaration and decorated_export_declaration in class_label_for_kind

Verification

Indexed a HarmonyOS project (~40 .ets files) with results:

• 888 nodes, 2253 edges (previously 0 Function/Class/Method nodes without the fallback logic)
• Correctly extracts Components, classes, functions, methods, enums, interfaces
• build_method inside @Component structs is recognized as a Method with synthesized name "build"

Also verified no regression on the codebase-memory-mcp project itself: 12,389 nodes, 58,847 edges.

Note on tree-sitter-arkts

The root cause of the fallback logic is that tree-sitter-arkts lacks field('name', ...) mappings in its grammar. I've planned to submit a PR to Million-mo/tree-sitter-arkts to add proper field definitions. If that PR is merged, the fallback logic in this PR will remain as a harmless safety net for older parser versions.

Vendored Grammar Provenance

Per the vendored-grammar review requirements, the following details are provided for audit:

Item	Detail
Upstream source	Million-mo/tree-sitter-arkts — community grammar, not tracked by nvim-treesitter or Helix registries (verdict: COMMUNITY)
Pinned commit	2fd0ad75e2d8 (upstream master HEAD as of 2025-10-27)
License	MIT — declared in grammar.js header (@license MIT, @author million) and package.json ("license": "MIT", "author": {"name": "million"}). Upstream ships no standalone LICENSE file; the vendored LICENSE is reconstructed from this declaration, copyright (c) 2024 million
Generated parser	parser.c generated by tree-sitter CLI v0.25.3 (per file header /* Automatically generated by tree-sitter v0.25.3 */), ABI 15
Vendored files	parser.c (149,394 lines), tree_sitter/parser.h — no scanner.c (grammar has EXTERNAL_TOKEN_COUNT 0)
Static security	Reviewed the vendored C surface only (parser.c + tree_sitter/parser.h); no system()/popen()/fork()/network calls; no package manager hooks, workflow files, prompt/agent instruction files, or generated lockfiles were vendored
Registry status	Not in nvim-treesitter or Helix — this is a COMMUNITY grammar. The pinned commit is upstream master HEAD (not a registry-pinned SHA) because no registry tracks this grammar

[DeusData]

Thanks for adding ArkTS support. Triage: language-support PR for #760, but this needs the full vendored-grammar review path before it can move forward.

Because this vendors a parser and changes provenance tooling, review must verify upstream source, exact pinned commit, license compatibility, generated parser provenance, and static security of vendored sources. Please make sure the PR description includes those details and that the vendored grammar can be audited without guessing.

[Happy-26]

Thanks for adding ArkTS support. Triage: language-support PR for #760, but this needs the full vendored-grammar review path before it can move forward.

Because this vendors a parser and changes provenance tooling, review must verify upstream source, exact pinned commit, license compatibility, generated parser provenance, and static security of vendored sources. Please make sure the PR description includes those details and that the vendored grammar can be audited without guessing.

Thanks for the triage. I've pushed a fix (commit 35d521b) addressing the vendored-grammar review requirements:

1. Upstream source: Million-mo/tree-sitter-arkts — community grammar (not in nvim-treesitter/Helix registries, verdict: COMMUNITY)

2. Pinned commit: 2fd0ad75e2d8 (upstream master HEAD, 2025-10-27)

3. License: MIT — declared in grammar.js header (@license MIT, @author million) and package.json ("license": "MIT", "author": {"name": "million"}). Upstream ships no standalone LICENSE file; the vendored LICENSE is reconstructed from this declaration. Fixed: the copyright holder was incorrectly listed as "aspect-ux" — corrected to "million" per grammar.js @author million.

4. Generated parser provenance: parser.c generated by tree-sitter CLI v0.25.3, ABI 15. No scanner.c (grammar has EXTERNAL_TOKEN_COUNT 0).

5. Static security: Vendored C surface reviewed (parser.c + tree_sitter/parser.h); no system()/popen()/fork()/network calls; no package manager hooks or workflow files vendored. Security review statement added to MANIFEST.md.

All details are documented in the updated PR description (Vendored Grammar Provenance section) and the MANIFEST.md entry. Also fixed an incorrect URL in lang_specs.c that pointed to a non-existent aspect-ux/tree-sitter-arkts repo.