@ai-matrx/content-ir 0.19.18 → 0.19.20

package/CHANGELOG.md

@@ -1,5 +1,15 @@
 # Changelog
 
+## 0.19.20
+### Unreleased
+
+- `/source` **The tokenizer's raw-HTML allow-list is the renderer's list** (verify-RC-B4 round 16): `ALLOWED_RAW_HTML_TAGS` (and `HTML_BLOCK_LEVEL_TAGS`) gain `details` and `summary`, which the screen's `rehypeSafeRawHtml` renders — a line-leading `<details>` is now an `html_block` island like `<table>`, not an inline `xml_tag`. Pinned by a shared vector: the list must equal the frontend-generated `rawHtmlTags` in `gfm-table-vectors.json` (content-ir and both Python twins). Island vectors: 1 of 896 changed; Python fixture copied.
+
+## 0.19.19
+### Unreleased
+
+- `/source` **THE ONE DELIBERATE GFM DEVIATION** (verify-RC-B4 round 16 ruling: content never silently disappears): a line opening an HTML block whose element every sanitizer strips — `STRIPPED_HTML_BLOCK_NAMES` (`link`, `nav`, `header`, `footer`, `article`, `aside`, `main`, `form`, `iframe`, `title`, … the type-6 names neither the screen's raw-HTML allow-list nor print's GitHub schema keeps) — ends that block at its own line, so a table or list under it is read as markdown (`insideHtmlBlock`; new `opensStrippedHtmlBlock(trimmed)`). GFM keeps such lines as raw HTML the sanitizer then empties; a stored answer quoting `<link rel="canonical" …/>` above its content lost every list that way. Python twin moved; `tableStarts` vectors +4 (the frontend's GFM judge applies the deviation). Documented in FEATURE.md § `source/`.
+
 ## 0.19.18
 ### Unreleased
 

package/dist/index.cjs

@@ -5193,7 +5193,9 @@ var ALLOWED_RAW_HTML_TAGS = /* @__PURE__ */ new Set([
   "colgroup",
   "col",
   "figure",
-  "figcaption"
+  "figcaption",
+  "details",
+  "summary"
 ]);
 var HTML_BLOCK_LEVEL_TAGS = /* @__PURE__ */ new Set([
   "p",
@@ -5219,7 +5221,9 @@ var HTML_BLOCK_LEVEL_TAGS = /* @__PURE__ */ new Set([
   "colgroup",
   "col",
   "figure",
-  "figcaption"
+  "figcaption",
+  "details",
+  "summary"
 ]);
 var SPECIAL_CODE_LANGUAGES = /* @__PURE__ */ new Set([
   "transcript",
@@ -6095,6 +6099,11 @@ var HTML_BLOCK_NAMES = "address|article|aside|base|basefont|blockquote|body|capt
 var HTML_BLOCK_TAG = new RegExp(`^</?(?:${HTML_BLOCK_NAMES})(?:[\\s>]|/>|$)`, "i");
 var ATTRIBUTE = `\\s+[A-Za-z_:][A-Za-z0-9_.:-]*(?:\\s*=\\s*(?:[^\\s"'=<>\`]+|'[^']*'|"[^"]*"))?`;
 var HTML_COMPLETE_TAG = new RegExp(`^(?:<[A-Za-z][A-Za-z0-9-]*(?:${ATTRIBUTE})*\\s*/?>|</[A-Za-z][A-Za-z0-9-]*\\s*>)\\s*$`);
+var STRIPPED_HTML_BLOCK_NAMES = "address|article|aside|base|basefont|body|center|dialog|dir|fieldset|footer|form|frame|frameset|head|header|html|iframe|legend|link|main|menu|menuitem|nav|noframes|optgroup|option|param|search|title|track";
+var STRIPPED_HTML_BLOCK_TAG = new RegExp(`^</?(?:${STRIPPED_HTML_BLOCK_NAMES})(?:[\\s>]|/>|$)`, "i");
+function opensStrippedHtmlBlock(trimmed) {
+  return STRIPPED_HTML_BLOCK_TAG.test(trimmed);
+}
 var HTML_BLOCK_NAME_SET = /* @__PURE__ */ new Set([
   // Condition 1 (raw text) and condition 6 (block) tag names — CommonMark 0.31.
   "script",
@@ -6212,6 +6221,7 @@ function insideHtmlBlock(lines, index) {
       continue;
     }
     paragraph = false;
+    if (opensStrippedHtmlBlock(trimmed)) continue;
     open = end && end.test(trimmed.slice(2)) ? void 0 : end;
   }
   return open !== void 0;
@@ -8060,6 +8070,7 @@ exports.nounFamily = nounFamily;
 exports.nounLabel = nounLabel;
 exports.nounTitleColumn = nounTitleColumn;
 exports.openParseSession = openParseSession;
+exports.opensStrippedHtmlBlock = opensStrippedHtmlBlock;
 exports.opensTable = opensTable;
 exports.pairDisplayMath = pairDisplayMath;
 exports.parseDirectiveSlug = parseDirectiveSlug;