mirror of
https://github.com/DeusData/codebase-memory-mcp.git
synced 2026-11-04 10:48:20 +00:00
Martin Vogel
0df990aa60
Routing .ets through the plain TypeScript grammar produced no definition node for any @Component struct (0/19 across the OpenHarmony sample corpus), parsed struct members as statement labels and build() as a call expression, and registered ArkUI builtins (Column, Row, ListItem, ...) as user function definitions with 28% fabricated cross-file CALLS edges. This adds a dedicated ArkTS grammar that keeps everything TypeScript already handles and models the ArkTS syntax TS cannot parse. Grammar (tools/tree-sitter-arkts/, generated parser vendored to internal/cbm/vendored/grammars/arkts/, ABI 15 via tree-sitter-cli 0.25.10): first-party derivative of tree-sitter/tree-sitter-typescript v0.23.2 @ 75b3874edb2d (the commit our vendored typescript/tsx come from) on the tree-sitter-javascript v0.23.1 base @ 3a837b6f3658 (the version upstream's package-lock pins). LICENSE carries the upstream MIT text verbatim plus a derivative notice; the external scanner is a symbol-rename trampoline over a byte-identical copy of the already-vendored typescript scanner header. ArkTS additions: decorated struct component declarations with class-shaped bodies; UI-DSL trailing-closure component calls with attribute chains after the closing brace; decorators on function declarations (@Builder, @Extend(T), @Styles); @Extend/@Styles leading-dot attribute-chain bodies; anonymous stateStyles style blocks; import lazy named imports. Wiring: CBM_LANG_ARKTS enum + lang spec (struct_declaration as a class kind labeled Struct, public_field_definition fields), .ets extension, LANG_NAMES/userconfig entries, grammar shim, and TS-family joins across the extractors and the name-resolution guards (parse_es_imports, JS keywords/test-file/func-kind/scope tables, TS heritage/param/receiver/ namespace/type-ref/channel/usage paths, js_ts_family and the weak-method- match suppression). Field definitions now also carry their decorators, so @State/@Prop/@Link members are findable with their roles. Tests: grammar corpus (35 cases), per-language regression + label golden + import net + extension cases, seven ArkTS extraction tests including the phantom-builtin regression (ArkUI builtins must never be definitions), and capability-ledger / call-node-manifest / call-argument-matrix rows. On the 14-file OpenHarmony sample corpus the TS-grammar baseline had 12/14 files with parse errors (231 ERROR nodes); the arkts grammar parses 14/14 clean, and an end-to-end index yields all 19 components as Struct nodes, 68 methods, 119 fields, and zero phantom builtin definitions. Signed-off-by: Martin Vogel <martin.vogel.tech@gmail.com>
17 lines
343 B
JSON
17 lines
343 B
JSON
{
|
|
"grammars": [
|
|
{
|
|
"name": "arkts",
|
|
"camelcase": "Arkts",
|
|
"scope": "source.arkts",
|
|
"path": ".",
|
|
"file-types": ["ets"]
|
|
}
|
|
],
|
|
"metadata": {
|
|
"version": "0.1.0",
|
|
"license": "MIT",
|
|
"description": "ArkTS (HarmonyOS .ets) grammar for tree-sitter — fork of tree-sitter-typescript v0.23.2"
|
|
}
|
|
}
|