[exKAZUu]

Why

This repository was imported with upstream's setup: npm with ESLint, the C/Go/Node.js/Python/Swift bindings, and upstream's CI, fuzz, and lint workflows, with no release pipeline. WillBooster/code-gauge needs a Rust crate of this grammar that it can pin to a release, as it does with WillBooster/tree-sitter-kotlin. That requires the same tests and release pipeline as that fork.

Requirements

• required: fork the grammars that need fixes as WillBooster/tree-sitter-kotlin is forked, following its format.
• required: do not be impolite to upstream or cause it trouble.
• required: keep grammar.js in the npm package, since WillBooster/tree-sitter-typescript will build on it.
• required: fix defects that the stricter checks find at their root cause in this PR instead of exempting them from the checks.
• required: leave the JSX text and attribute strings containing a bare & and the tagged templates with invalid escapes (e.g. String.raw) unfixed here; they get their own issues.
• required: leave the misparse of a line starting with [ after an arrow function with a block body (examples/cli/lib/base-cmd.js in script/known-failures.txt) unfixed here; the requester files its issue.
• chosen: end declarations in a for_statement initializer only with ;, because ECMAScript inserts no semicolon inside a for header and accepting one there made for (let x\n of y) ambiguous, which broke incremental reparsing.
• chosen: credit upstream in the README (one sentence, "not an official release") and keep the upstream copyright line in LICENSE, as tree-sitter-kotlin does, without naming the upstream authors as authors of this fork's packages.
• chosen: publish only under WillBooster names (@willbooster/tree-sitter-javascript, the crate willbooster-tree-sitter-javascript), so nothing competes with upstream's tree-sitter-javascript packages.
• chosen: ship only the Wasm build and the Rust binding (exporting upstream's LANGUAGE and NODE_TYPES), the two forms the Kotlin fork ships and its consumers use.
• chosen: parse the repository upstream's CI cloned (npm/cli), pinned to a commit, plus the checked-in examples/, against script/known-failures.txt, so the list of failures changes only with the grammar.
• chosen: keep upstream's queries/ in the repository, untested and not referenced from tree-sitter.json, because the test layout allows only corpus and unit tests and neither package ships it, so the shipped tree-sitter.json must not name query files absent from the npm package and the crate (the alternative, shipping untested queries, was rejected); the Kotlin fork's metadata names no queries either.

Customer Summary

Nothing is released yet. After this merges, the first feat/fix commit releases @willbooster/tree-sitter-javascript to npm (with tree-sitter-javascript.wasm and grammar.js) and the willbooster-tree-sitter-javascript crate to crates.io, as version 1.0.0.

Technical Summary

1. Copied from WillBooster/tree-sitter-kotlin and renamed: agent instructions, issue and PR templates, lefthook, bun, mise, oxlint/oxfmt, Renovate, .releaserc.json, the test/test-rust/robustness/release/semantic-pr workflows, script/, test/unit/, bindings/rust/, Cargo.toml, Cargo.lock, and package.json.
2. Removed: package-lock.json, eslint.config.mjs, the non-Rust bindings and their build files, upstream's FUNDING.yml, issue templates, dependabot config, and workflows, and the highlight and tag tests in test/highlight and test/tags.
3. tree-sitter.json keeps upstream's grammar entry without the query paths, since neither package ships queries/, names WillBooster Inc. as the author, and enables only the Rust binding.
4. grammar.js is reformatted by oxfmt and made lint-clean: () => for unused $, \u0000-\u001F instead of \x00-\x1F, an unnecessary \- removed, and the u flag on the regexes using \p{…}. At the first commit, the regenerated src/parser.c is byte-identical to the one generated from upstream's grammar.js, and src/grammar.json differs only in those regex spellings and flags. What to check hardest: the \[ escapes in the regex-pattern and identifier character classes stay with lint suppressions, because the autofix removed them and tree-sitter's regex parser then rejects the pattern.
5. The Robustness test job fails when the committed src/ differs from what grammar.js generates, since both packages ship the committed src/. test/unit/performance.test.ts also rejects a Wasm build compiled from a src/parser.c older than grammar.js. Renovate updates the fuzz job's tree-sitter-version together with tree-sitter-cli through a regex custom manager in renovate.jsonc. oxlint and oxfmt ignore examples/, whose JavaScript is test input: the autofix had rewritten it.
6. The second commit fixes an incremental-parsing defect that the fuzz check found in the corpus case "For statements". After let j inside for (, both a declaration ended by an automatic semicolon and a for-in header ended by an optional automatic semicolon were live, so for (let j\n of k) stayed ambiguous until of. After an edit that turned of k into something else and its undo, the reparse reused the lexical_declaration built when the for-in path had died, and yielded an ERROR. ECMAScript never inserts a semicolon inside a for header, so the declarations in a for_statement initializer now end only with ; (for_lexical_declaration/for_variable_declaration rules used only through aliases to the public node names, so the node types are unchanged; they are visible rules because a hidden one would make for_statement inherit their kind field), and _for_header no longer accepts an automatic semicolon. The ambiguity and its GLR conflict are gone, and for (var j\n of k) now parses, where it used to fail; the corpus case covers it. What to check hardest: that the 1,640 example files parse to the same trees as before the fix (checked), and that rejecting for (let i = 0\n i < n; i++), which is not valid JavaScript, loses nothing real-world code needs.
7. src/ is regenerated with tree-sitter-cli 0.27 (upstream used 0.25).

Testing

• bun run verify, bun wb test (corpus 116/116 with the native and the Wasm build, incremental-parsing fuzz, performance, and examples), and cargo test --locked pass locally.
• tree-sitter fuzz passes on all corpus cases with seeds 1, 2, 3, 4, 5, 42, and 1234 (1,000 iterations, 10 edits) and with seeds 7 and 99 (20 edits).
• The parse trees of all 1,640 example files are identical before and after the fix.
• script/parse-examples: 1 of 1,640 files fails, examples/cli/lib/base-cmd.js, which is valid JavaScript: a line starting with [ after an arrow function with a block body is parsed as a subscript of that arrow function instead of a new statement.
• Not run: the release workflow (it runs only on main), the macOS Intel and Linux arm64 jobs, and libFuzzer (Apple clang has no libFuzzer).

Notes

• The highlight and tag tests were removed because the test layout allows only corpus and unit tests; queries/ is kept but no longer tested (neither package ships it).
• semantic-pr / semantic-pr runs on pull_request_target, which uses the workflow files on main, so it cannot run for this PR.
• Publishing uses trusted publishing on npm and crates.io, which has to be configured for the new package names before the first release. crates.io lets a trusted publisher be configured only for an existing crate, so the first crate version must be published manually (as the Kotlin fork's was) before the first automated release; otherwise that release publishes the npm package and fails at the crate.