@lokascript/framework 6.0.0 → 6.2.0

package/CHANGELOG.md

@@ -7,6 +7,136 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [6.2.0] - 2026-10-18
+
+The `@lokascript/semantic/core` path weighs about a third less, and its lexicons now register with
+it. MCP `get_diagnostics` gives the engine's verdict for English, and a multi-line BehaviorSpec with
+a line it cannot read is refused instead of compiled without it.
+
+### Fixed
+
+- **`@hyperfixi/mcp-server`: `get_diagnostics` gives the engine's verdict for English, where it
+  stopped** (#1575). Measured on 672 inputs (the agent bench and the docs patterns, in en/ja/es/ar), it
+  flagged 6 of the 27 English inputs the engine rejects, none with a position; gave 9 false errors
+  on English the engine reads (8 of them a possessive `'s` read as an open quote); and warned 467
+  `possible-typo`s, 391 on ja/es/ar words checked against an English command list. Now: 27 of 27,
+  each at the token where the engine stopped; 0 false errors (semantic's doubts about English the
+  engine reads are warnings); and a "did you mean" only for the word the engine stopped on
+  (`toggel`, not `and`), never off English. Style advice (`wait` over `setTimeout`, `then` between
+  commands) stays.
+
+- **`@lokascript/semantic`: `lexicons/<code>` registers its vocabulary with `./core`.** Each
+  lexicon file bundled a private copy of the lexicon registry, so a page that imported
+  `@lokascript/semantic/core`, `languages/ja` and `lexicons/ja` registered the Japanese words in a
+  Map `render` never read, and value words stayed English (`私の valueに` for `私の 値 に`). The root
+  entry and the browser bundles were not affected. `test/core-subpaths.test.ts` holds it on the
+  built files (#1577).
+
+- **`@hyperfixi/mcp-server`: a multi-line BehaviorSpec with a line it cannot read is refused,
+  not compiled without it** (lokascript-domains #54). On `@lokascript/domains` 5.0.1,
+  `compile_behaviorspec` answered `ok: true` for `test "Login"` / `given page /login` /
+  `when I click #submit` / `then #result is visible` with only the `goto`: the two lines it
+  could not read were gone. mcp-server is on domains 5.0.2, whose `compileBehaviorSpec` throws a
+  `BehaviorSpecError`; the multi-line handler answers `ok: false` with `errors` (one per line)
+  and the rest under `partial`, never `code`. `validate_behaviorspec` reports what compile would
+  refuse (a `given page /login when …` URL that takes the rest of its line), and
+  `translate_behaviorspec` refuses such a spec whole (`translated` and `playwright` are `null`)
+  instead of translating what remained (#1576).
+
+- **`@lokascript/compilation-service`: `/health` reports the package's version**, not the `1.4.0`
+  literal it reported through 6.1.0 (`src/version.ts`, rewritten by `set-version.cjs`; #1576).
+
+### Changed
+
+- **`@lokascript/semantic`: the `./core` + `./languages/<code>` path weighs about a third less.**
+  Measured with esbuild (minified, gzip -9) on learn's ten languages with their lexicons:
+  275 KB → 180 KB; `./core` alone 155 KB → 120 KB; core with one language and its lexicon
+  168 KB → 133 KB. Two packaging changes: `KNOWN_PROFILES` is marked pure, so a bundle that does
+  not use it drops the profiles of languages it never imports (35 KB); and the language and
+  lexicon modules share their common code in chunks instead of a copy each (60 KB over ten).
+  Through the packed tarball, the bundled path renders exactly what the root entry does (1,476 of
+  1,476 pairs: 164 docs patterns × 9 languages, there and back), and the root entry is unchanged
+  (3,772 of 3,772 pairs against 6.1.0). `@hyperfixi/vite-plugin`'s debug size table is
+  re-measured (one language ~130–135 KB, `all` ~222 KB) (#1577).
+
+- **`@lokascript/language-server` on `vscode-languageserver` 10** (protocol 3.18; replaces
+  Dependabot #1278), and the LokaScript VS Code extension on `vscode-languageclient` 10 (#1279),
+  which needs VS Code 1.91 (`engines.vscode` was `^1.85.0`). 10.x adds an exports map without
+  `./node.js`, so the server imports `vscode-languageserver/node`; a diagnostic's `message` may be
+  `MarkupContent` in 3.18, so the code actions read its text either way. Over stdio the server
+  answers as 6.1.0 does: the same capabilities, the engine's parse error at the same position,
+  hover and completion (#1579).
+
+### Added
+
+- **`@lokascript/semantic/core` exports `getAllTranslations` and `getAllTranslationsWithStatus`**,
+  over the languages a page registered. lokascript-learn's playground imported the root entry, and
+  with it all 24 languages, for this one function (#1577).
+
+### Deprecated
+
+- **`@hyperfixi/mcp-server`: `parse_multilingual`, removed in 7.0.** It returns only the top node
+  (every handler as `on` and its event) and reported success on 21 of the 27 English inputs the
+  engine rejects; `validate_and_compile` returns the whole IR and the engine's verdict. Its
+  description starts "Deprecated", and each result carries a second text block naming the
+  replacement (#1575).
+
+## [6.1.0] - 2026-10-18
+
+One validator and one translator: MCP `validate_and_compile` accepts the English the engine reads
+and warns `NAME_COLLISION`, and `validate_hyperscript` and `translate_hyperscript` are deprecated for
+7.0. `@hyperfixi/vite-plugin`'s `slavic` and `south-asian` regions load their languages.
+
+### Added
+
+- **`@lokascript/compilation-service`: `validate` warns `NAME_COLLISION`** on foreign input whose
+  variable is spelled like a word of its own language (es `si` is also `if`), with the rename
+  semantic's finder suggests: the one check MCP `validate_hyperscript` made that `validate_and_compile`
+  did not, with its wording and its confidence floor (#1568).
+- **`@lokascript/compilation-service`: a README** (#1570). `files` listed one, but the package had
+  none.
+
+### Deprecated
+
+- **`@hyperfixi/mcp-server`: `validate_hyperscript` and `translate_hyperscript`, removed in 7.0**
+  (#1569).
+  Measured 2026-10-18 on 6.0.0, through the server:
+  - `translate_hyperscript` returns `translate_code`'s text on 471 of 471 pairs, with the same 21
+    refusals; `translate_code` adds a fidelity `verification` and `NAME_COLLISION` warnings.
+  - On 672 inputs `validate_hyperscript` caught nothing `validate_and_compile` missed, and every
+    warning only it gave was a false alarm (a fixed list of 34 event names; an if/end count). Its one
+    useful check, `NAME_COLLISION`, is in `validate_and_compile` now.
+
+  Both still work. Their descriptions start "Deprecated", and each result carries a second text block
+  naming the replacement. The skills, the mcp-server README, MIGRATION.md and the LLM-tool checklist
+  name only `validate_and_compile` and `translate_code`. The lokascript-multilingual skill had called
+  `translate_hyperscript` keyword substitution that does not reorder words, which was not true.
+
+### Fixed
+
+- **`@lokascript/compilation-service`: `validate` (MCP `validate_and_compile`) no longer rejects
+  English the engine reads.** Of the 1,128 command-shape cases `@hyperfixi/engine` reads, it
+  rejected 90:
+  - every behavior definition (a `BehaviorNode` keeps its name as a field, and the schema check read
+    it as a missing role; this also failed behaviors in every other language);
+  - 73 parses under semantic's default confidence bar (`init`, `when`, `def`, `live`, bare `hide`, …);
+  - a quoted event name semantic does not read, and `wait for <event>`.
+
+  For natural-language English the engine's parse now decides `ok`, and semantic's doubts come back
+  as warnings that say the IR may leave part of the code out. A request that sets `confidence`
+  keeps them as errors, as does foreign input, which runs as semantic's translation. On the agent
+  bench every correct rejection was already the engine's (16 of 16), so none moved.
+  testing-framework's `validate-agrees-with-engine.test.ts` holds the 1,128 at 0 rejected (#1566).
+
+- **`@hyperfixi/vite-plugin`: `region: 'slavic'` and `region: 'south-asian'` load their languages**
+  (#1567).
+  Both were in the option's type and the README's language list, but the expansion had a
+  hand-written case per region and none for these two, so they loaded English alone. Regions now
+  expand from `REGIONS` itself, and the README's region table lists both. The debug size table
+  (`SEMANTIC_BUNDLE_SIZES`, January 2025) is re-measured by a new
+  `scripts/measure-semantic-sizes.ts`: semantic's core plus one language is ~163–175 KB gzipped,
+  not the ~14–20 KB it claimed (`all` ~383 KB, not ~61 KB).
+
 ## [6.0.0] - 2026-10-18
 
 Removes `@lokascript/compilation-service`'s test and component generation, deprecated in 5.1.0. For
@@ -1943,7 +2073,9 @@ _Synchronized version release. See git history for details._
 - npm access token stored in GitHub Secrets
 - 2FA recommended for npm organization
 
-[Unreleased]: https://github.com/codetalcott/hyperfixi/compare/v6.0.0...HEAD
+[Unreleased]: https://github.com/codetalcott/hyperfixi/compare/v6.2.0...HEAD
+[6.2.0]: https://github.com/codetalcott/hyperfixi/compare/v6.1.0...v6.2.0
+[6.1.0]: https://github.com/codetalcott/hyperfixi/compare/v6.0.0...v6.1.0
 [6.0.0]: https://github.com/codetalcott/hyperfixi/compare/v5.1.0...v6.0.0
 [5.1.0]: https://github.com/codetalcott/hyperfixi/compare/v5.0.0...v5.1.0
 [5.0.0]: https://github.com/codetalcott/hyperfixi/compare/v4.8.0...v5.0.0

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@lokascript/framework",
-  "version": "6.0.0",
+  "version": "6.2.0",
   "description": "Generic framework for building multilingual DSLs with semantic parsing and grammar transformation",
   "type": "module",
   "main": "dist/index.cjs",
@@ -95,13 +95,13 @@
   "author": "LokaScript Contributors",
   "license": "MIT",
   "dependencies": {
-    "@lokascript/intent": "^6.0.0"
+    "@lokascript/intent": "^6.2.0"
   },
   "devDependencies": {
     "@lokascript/domains": "^5.0.0",
     "@types/node": "^26.6.3",
     "@vitest/coverage-v8": "^5.0.0",
-    "tsup": "^8.0.0",
+    "tsup": "^9.0.0",
     "typescript": "^5.0.0",
     "vitest": "^5.0.2"
   },