@ai-matrx/content-ir 0.19.15 → 0.19.17

package/CHANGELOG.md

@@ -1,5 +1,16 @@
 # Changelog
 
+## 0.19.17
+### Unreleased
+
+- **Release note:** 0.19.16 was tagged but never reached npm — its tag points at a commit whose workspace lockfile (`apps/pnpm-lock.yaml`) was out of date, so every publish retry of that immutable tag fails at `pnpm install --frozen-lockfile`. The round-9 changes it carried ship in the next version.
+
+## 0.19.16
+
+Automatic changed-only republish (docs/metadata drift since the last tag — see
+`git diff npm/content-ir-core/v0.19.14..npm/content-ir-core/v0.19.16 -- apps/shared/content-ir-core`).
+No source changes intended and no consumer action required.
+
 ## 0.19.15
 
 Automatic changed-only republish (docs/metadata drift since the last tag — see
@@ -39,6 +50,7 @@ No source changes intended and no consumer action required.
 ## 0.19.9
 ### Unreleased
 
+- `/source` **Round 9 verifier findings (verify-RC-B4 R9-1, R9-4, R9-5):** a footnote definition (`[^1]: …`) ends a table (Python twin moved); a table never opens INSIDE an HTML block that opened above it with no blank line (`<details>` then `| a | b |` is raw HTML — types 1–5 end at their marker, 6–7 at the blank line, 7 cannot interrupt a paragraph); new THE link-reference rule — `collectLinkDefinitions(text)`, `findLinkDefinitionLines(text)`, `resolveReferenceLinks(fragment, defs)`, `normalizeLinkLabel` — so a cell or split block resolves `[Apple Support][1]` against the document's definitions, as GFM does (a definition cannot interrupt a paragraph or a row run). Vectors +2 ends, +11 starts, +1 document.
 - `/source` **`isHtmlBlockTagName(name)`** (verify-RC-B4 round 9): the tag names CommonMark reads as an HTML BLOCK when they open a line (conditions 1 and 6: `div`, `details`, `section`, `script`, …; case-insensitive) — so a splitter can leave such a line to the markdown renderer, ending where GFM ends it, instead of opening a custom XML container that owns what follows.
 - `/source` **A table ends where GFM ends it (verify-RC-B4 round 9):** `continuesTable` / `findTableEnd` end a table at an HTML block start (all seven CommonMark 4.6 conditions — `<script`/`<pre`/`<style`/`<textarea`, `<!--`, `<?`, `<!X`, `<![CDATA[`, a block tag name, or any other complete tag alone on its line such as `</artifact>`) and at indented code: a line 4+ columns past the table's container. New `startsHtmlBlock`, `lineIndent`, `tableContainerIndent` (the content column of the list item a table sits in); `continuesTable(line, containerIndent = 0)`. A lone `|` is a row (one empty cell) — `rowCells("|")` is `[""]`, no longer `[]`. Since 0.19.4 an HTML line or an indented line after a table was read as a table row. Shared vectors gain `tableEnds` (30 cases, each judged against remark-gfm in matrx-frontend); the Python twin moved with them.
 - `/source` **A table never opens on a lazy continuation line (verify-RC-B4 round 9):** `opensTable` / `tableStartsAt` / `findTableStart` refuse a header that lazily continues a list item's or a block quote's paragraph (no blank line above, not indented to the item's content column, or missing the quote's `>`) — GFM reads it as paragraph text (`- item` then `| a | b |` is no table). Vectors gain `tableStarts` (15 cases, judged against remark-gfm).

package/dist/index.cjs

@@ -7672,20 +7672,36 @@ function normalizeLinkLabel(label) {
 var DEFINITION = /^ {0,3}\[((?:[^\]\\]|\\.){1,999})\]:[ \t]*(?:<([^<>\n]*)>|(\S+))(?:[ \t]+(?:"((?:[^"\\]|\\.)*)"|'((?:[^'\\]|\\.)*)'|\(((?:[^()\\]|\\.)*)\)))?[ \t]*$/;
 function collectLinkDefinitions(text) {
   const defs = /* @__PURE__ */ new Map();
-  if (!text.includes("]:")) return defs;
+  for (const { label, definition } of scanLinkDefinitions(text)) {
+    const key = normalizeLinkLabel(label);
+    if (key && !defs.has(key)) defs.set(key, definition);
+  }
+  return defs;
+}
+function findLinkDefinitionLines(text) {
+  return scanLinkDefinitions(text).map(({ line }) => line);
+}
+function scanLinkDefinitions(text) {
+  const found = [];
+  if (!text.includes("]:")) return found;
   const lines = text.split("\n");
   const kinds = fenceLineKinds(text);
+  let previousWasDefinition = false;
   lines.forEach((line, i) => {
+    const wasDefinition = previousWasDefinition;
+    previousWasDefinition = false;
     if (kinds[i] !== "prose") return;
     const m = DEFINITION.exec(line);
     if (!m) return;
+    const above = i > 0 ? lines[i - 1] ?? "" : "";
+    const opensBlock = i === 0 || !above.trim() || wasDefinition || kinds[i - 1] === "close" || /^ {0,3}(?:#{1,6}(?:\s|$)|(?:[-*_][ \t]*){3,}$)/.test(above);
+    if (!opensBlock) return;
     const label = m[1] ?? "";
     if (label.startsWith("^")) return;
-    const key = normalizeLinkLabel(label);
-    if (!key || defs.has(key)) return;
-    defs.set(key, { url: m[2] ?? m[3] ?? "", title: m[4] ?? m[5] ?? m[6] ?? null });
+    previousWasDefinition = true;
+    found.push({ line: i, label, definition: { url: m[2] ?? m[3] ?? "", title: m[4] ?? m[5] ?? m[6] ?? null } });
   });
-  return defs;
+  return found;
 }
 var REFERENCE = /(!?)\[((?:[^\]\\]|\\.)*)\](?:\[((?:[^\]\\]|\\.)*)\])?(?![(:[])/g;
 var escapeTitle = (title) => title.replace(/["\\]/g, "\\$&");
@@ -7984,6 +8000,7 @@ exports.fenceParts = fenceParts;
 exports.fieldsToDbPayload = fieldsToDbPayload;
 exports.findCodeRanges = findCodeRanges;
 exports.findInlineCodeSpans = findInlineCodeSpans;
+exports.findLinkDefinitionLines = findLinkDefinitionLines;
 exports.findTableEnd = findTableEnd;
 exports.findTableStart = findTableStart;
 exports.fingerprintText = fingerprintText;