Adopt hand-written liquid-html-parser for performance

.changeset/hand-written-liquid-html-parser.md

@@ -0,0 +1,16 @@
+---
+'@shopify/liquid-html-parser': minor
+'@shopify/theme-language-server-common': minor
+'@shopify/theme-check-common': minor
+'@shopify/prettier-plugin-liquid': minor
+'@shopify/theme-graph': minor
+---
+
+Adopt the hand-written `liquid-html-parser` for performance
+
+Replace the parser-combinator-based Liquid/HTML parser with a hand-written
+recursive-descent parser. The new parser is significantly faster, adds
+resilient parsing (it recovers from malformed input instead of bailing), and is
+adapted to the theme-tools source model. `theme-language-server-common`,
+`theme-check-common`, `prettier-plugin-liquid`, and `theme-graph` are updated to
+consume the new parser.

.changeset/port-theme-check-checks.md

@@ -0,0 +1,26 @@
+---
+'@shopify/theme-check-common': minor
+---
+
+Port additional theme-check checks
+
+Add the following checks:
+  - `LiquidComplexity` -- Reports Liquid files with high cyclomatic complexity (default threshold 120).
+  - `LiquidNestingDepth` -- Reports Liquid files with deeply nested control-flow structures (default max depth 10).
+  - `LiquidSyntaxError` -- Reports Liquid syntax errors.
+  - `MaxFileSize` -- Reports theme files that exceed Shopify's maximum per-file size.
+  - `ExcessiveSettingsCount` -- Reports section/block schemas declaring more top-level settings than the max (default 40).
+  - `BlockArgumentSettingCollision` -- Reports a plain block-tag argument whose name matches a setting id in the block's schema.
+  - `DuplicateBlockArguments` -- Reports duplicate argument names in a block tag.
+  - `MissingBlockArguments` -- Reports required `{% doc %}` arguments not provided on a block tag.
+  - `UnknownBlockSetting` -- Reports a `block.settings.<name>` argument where `<name>` is not in the block's schema.
+  - `UnrecognizedBlockArguments` -- Reports block-tag arguments not declared in the block's `{% doc %}` tag.
+  - `ValidBlockArgumentTypes` -- Reports type mismatches between block-tag arguments and their `{% doc %}` declarations.
+  - `SchemaOncePerFile` -- Reports when `{% schema %}` appears more than once in a file.
+  - `SchemaSectionOrBlockOnly` -- Reports `{% schema %}` used outside section or block files.
+  - `StylesheetOncePerFile` -- Reports when `{% stylesheet %}` appears more than once in a file.
+  - `StylesheetTagInWrongFile` -- Reports `{% stylesheet %}` used outside section, block, or snippet files.
+  - `JavascriptOncePerFile` -- Reports when `{% javascript %}` appears more than once in a file.
+  - `JavascriptTagInWrongFile` -- Reports `{% javascript %}` used outside section, block, or snippet files.
+
+`RequiredLayoutThemeObject` now flags any `layout/*.liquid` file missing the required theme objects, not just `layout/theme.liquid`.

packages/liquid-html-parser/.gitignore

@@ -4,9 +4,12 @@ pnpm-debug.log*
 .DS_Store
 dawn
 TODO
-grammar/liquid-html.ohm.js
-standalone.js
-standalone.js.LICENSE.txt
 **/actual.liquid
 coverage
 .nyc_output
+
+# Generated parser fixtures (goldens + downloaded themes) — regenerated by scripts/setup-fixtures.ts
+fixtures/golden-html-ast/
+fixtures/golden-liquid-ast/
+fixtures/theme/
+fixtures/theme-bundle.ts

packages/liquid-html-parser/fixtures/README.md

@@ -0,0 +1,52 @@
+# Fixtures
+
+Golden fixtures and theme files are gitignored (too large for PR diffs).
+
+## Directory layout
+
+- `theme/` -- raw `.liquid` files organized by theme (e.g. `theme/dawn/`, `theme/horizon/`)
+- `theme-bundle.ts` -- auto-generated TypeScript module exporting all theme sources as an array
+- `golden-html-ast/` -- JSON snapshots from `toLiquidHtmlAST` per theme
+- `golden-liquid-ast/` -- JSON snapshots from `toLiquidAST` per theme
+
+## Scripts
+
+All scripts live in `../scripts/` and are run with `pnpx tsx`.
+
+### Quick setup (all-in-one)
+
+```
+pnpx tsx scripts/setup-fixtures.ts
+```
+
+Downloads themes, generates the theme bundle, builds the parser, and generates golden fixtures. Idempotent — safe to re-run.
+
+### 1. Download theme files
+
+```
+pnpx tsx scripts/download-themes.ts
+```
+
+Downloads Dawn (`Shopify/dawn`), Horizon (`Shopify/horizon`), and base-theme (`shopify-playground/ose-next-theme`) from GitHub using `gh api`. Extracts `.liquid` files into `fixtures/theme/<theme-name>/`. Requires `gh` CLI authenticated with repo access.
+
+### 2. Generate theme bundle
+
+```
+pnpx tsx scripts/generate-theme-bundle.ts
+```
+
+Reads every `.liquid` file under `fixtures/theme/`, sorts them, and writes `fixtures/theme-bundle.ts` -- a TypeScript module that exports `THEME_FILES` (an array of `{ path, source }` objects). This lets tests import theme sources without filesystem access.
+
+Requires `fixtures/theme/` to exist and contain at least one `.liquid` file.
+
+### 3. Generate golden AST fixtures
+
+```
+pnpx tsx scripts/generate-golden.ts [path-to-parser-dist]
+```
+
+Parses every `.liquid` file under `fixtures/theme/` with both `toLiquidHtmlAST` and `toLiquidAST`, strips non-essential properties (positions, linked-list pointers, source), and writes one JSON file per template into `fixtures/golden-html-ast/<theme>/` and `fixtures/golden-liquid-ast/<theme>/`.
+
+The optional argument overrides the parser module path (defaults to `dist/index.js`). Build the parser first (`pnpm run build`) when using the default.
+
+Prints a per-theme success/failure summary on completion.

packages/liquid-html-parser/fixtures/error-corpus.ts

@@ -0,0 +1,75 @@
+/*
+ * Adversarial error corpus for the tolerant-parser overhead benchmark.
+ *
+ * Every `source` here is *invalid* Liquid/HTML: the strict `toLiquidHtmlAST`
+ * throws on each one, so these sources exist only to exercise the tolerant
+ * path (`toTolerantLiquidHtmlAST`), which recovers instead of throwing and
+ * surfaces one `LiquidErrorNode` per region it gives up on.
+ *
+ * The bench arm that consumes this corpus measures resync/recovery cost, not
+ * throughput on clean input.
+ *
+ * Shape is identical to `THEME_FILES` in `theme-bundle.ts`
+ * (`Array<{ path; source }>`) so the bench loop stays uniform across corpora.
+ *
+ * Contract (verified in C2): each `source` passed to
+ * `toTolerantLiquidHtmlAST` returns a `DocumentNode` without throwing and
+ * contains at least one `LiquidErrorNode`.
+ */
+
+/*
+ * A single orphan close tag followed by a valid variable output. The close
+ * has no matching open, so the tolerant parser emits one error node, then
+ * resynchronizes on the next construct-open boundary and recovers the output.
+ *
+ * Repeating this unit forces the resync loop to fire once per unit — the
+ * "error every few tokens" density stress.
+ */
+const FREQUENT_ERROR_UNIT = "{% endfor %}{{ x }}{% endif %}{{ y }}";
+
+/*
+ * A well-formed Liquid+HTML fragment the strict parser accepts as-is.
+ *
+ * Repeated many times it builds a large clean body; a single orphan close
+ * tag appended near EOF then costs exactly one recovery, proving tail
+ * recovery does not rescan the whole document.
+ */
+const CLEAN_UNIT =
+  '<div class="card">{{ product.title }}' +
+  "{% if product.available %}<span>{{ product.price }}</span>{% endif %}" +
+  "</div>\n";
+
+export const ERROR_FILES: Array<{ path: string; source: string }> = [
+  {
+    /* Seeded verbatim from tolerant.test.ts:150 — one orphan close tag. */
+    path: "error-corpus/single-error.liquid",
+    source: "{% endfor %}",
+  },
+  {
+    /*
+     * Seeded verbatim from tolerant.test.ts:178 — two orphan closes with a
+     * valid output recovered between them (interleaved resync).
+     */
+    path: "error-corpus/interleaved-resync.liquid",
+    source: "{% endfor %}{{ good }}{% endif %}",
+  },
+  {
+    /*
+     * Net-new: an orphan close roughly every few tokens across a moderately
+     * long source, bounding worst-case resync frequency.
+     *
+     * 60 units yields 120 error nodes interleaved with 120 recovered
+     * outputs.
+     */
+    path: "error-corpus/pathological-frequent.liquid",
+    source: FREQUENT_ERROR_UNIT.repeat(60),
+  },
+  {
+    /*
+     * Net-new: a large valid body (150 clean units) with a single orphan
+     * close tag just before EOF — "parse a lot, then recover once".
+     */
+    path: "error-corpus/large-clean-error-near-eof.liquid",
+    source: CLEAN_UNIT.repeat(150) + "{% endfor %}",
+  },
+];