[tannevaled]

The parser shipped v0.2.0 on 2026-09-21 with fourteen grammar fixes. This site did not mention it — nor any parser version, nor any release date. Worse, all three entries under "Known limitations" are now false, and one of them has been wrong for three months.

What I verified

I built a probe against github.com/go-ruby-parser/parser v0.2.0 and ran each documented limitation through parser.Parse, with ruby -c (MRI 4.0.5) as the oracle:

paren-less cmd kwargs  foo a: 1    PARSES
paren-less cmd splat   foo *a      PARSES
paren-less cmd block   foo a: 1 do end   PARSES
splat block param      { |*a| }    PARSES
default block param    { |a = 1| } PARSES
empty block params     { || }      PARSES
find-pattern Class(a)              PARSES
find-pattern Class[a]              PARSES

Every one. The splat-block-param entry in particular had been wrong since 2026-06-22 — upstream corrected its own README about eleven hours after this page was last edited, and the page never caught up.

What is actually left

A 41-construct differential sweep against MRI 4.0.5 (parser.Parse vs ruby -c, same sources): one disagreement, 0 over-permissive.

• BEGIN { } / END { } blocks do not parse. MRI accepts them.

Everything else probed parses: __END__/DATA, flip-flop, redo, method-level rescue/ensure, heredoc with a method call, two heredocs on one line, named-capture assignment, $-0, magic encoding comments, p(x:), def a(&) = b(&), def a(...) = b(...), one-line =>/in, a&.b = 1, defined?(yield), **nil, alias $new $old, 2r/3i, keywords as method names, and the rest.

I have also noted the four language/ spec files that parse but stop later in go-embedded-ruby's own compiler (for_spec, block_spec, defined_spec, variables_spec) — consumer-side work, not a parser gap, and the page now says so rather than leaving a reader to infer a parser limitation.

docs/grammar.md

The limitations section is replaced by "What v0.2.0 added" — the fourteen gaps, with v0.2.0's own measure: every one of the 68 files in the pinned ruby/spec language/ corpus now parses, thirteen of which had not parsed at all before. That last point is the one worth having on a docs page: those files were contributing nothing to either column of any conformance measurement — not failing, invisible.

It also keeps the two silent mis-parses among the fourteen, because they are worse than refusals — nothing diagnoses them:

o.s (:a){ 1 }          # the brace block bound to :a, not to the call
case [0,1,2,3]
in [0, 1, ]            # the trailing comma was dropped, so a partial pattern
  :partial             # became an exact one and this fell through to else
else
  :exact
end

docs/index.md

• names v0.2.0 (2026-09-21) as the current release and versions the go get line (it had no version suffix);
• the coverage claim said "100% coverage enforced in CI across the six 64-bit targets". Per the parser's own ci.yml the gate is on the three OS lanes (ubuntu, macOS, windows); the six 64-bit targets — amd64/arm64 natively, riscv64/loong64/ppc64le/s390x under QEMU — plus js/wasm and wasip1/wasm run the suite without the coverage gate. Both facts are now stated, correctly attributed.

.github/workflows/docs.yml

mike deploy --push --update-aliases **0.1** latest. With v0.2.0 tagged, the live version dropdown was labelling post-v0.2.0 docs as 0.1 docs. Bumped to 0.2, and the comment above it with it.

Worth knowing, not fixed here

The repo has zero GitHub Releases — gh release list --repo go-ruby-parser/parser is empty and gh api …/releases returns []. v0.2.0 exists only as a git tag. So there are no release notes for a docs page to link to, and I have linked the tag tree (/tree/v0.2.0) rather than /releases/tag/v0.2.0, which renders an empty page. Publishing the releases, and wiring mike's tag-driven versioning (its own comment says "tag-driven versions can be added later", never done), are both worth doing — but they are repo-process changes, not doc corrections, so they are not in this PR.

Left alone deliberately

The upstream Performance section (615 harvested snippets, 1.5×/2.1× vs RubyVM::AbstractSyntaxTree.parse, ~6× vs Ripper.sexp, 24–30× on the lexer) is absent from this site entirely. That is a gap rather than a wrong claim, and I could not reproduce those benchmarks in this pass, so I did not import numbers I have not measured.

Also left: the Ruby 4.0 / MRI 4.0.5 / CGO=0 / Go 1.26.4+ claims, which I checked against the parser repo and which still hold. And requirements.txt's pins (mkdocs-material==9.7.7, mike==2.2.0) — dependency hygiene, not doc drift.

No merge. No screenshots.

🤖 Generated with Claude Code