fix: bibliography not appearing - extract HTML before generateHTML

The renderedhtml attribute can contain > chars (valid in quoted HTML
attributes) which broke the [^>]* regex in postProcess. Now we extract
the bibliography HTML from the TipTap JSON tree before calling
generateHTML, clear the attribute to keep the output clean, and inject
the raw HTML in postProcess using the pre-extracted string.

Made-with: Cursor

backend/src/publisher/html-renderer.ts

@@ -32,8 +32,12 @@ export function renderArticleHTML(
   cssArticle: string
 ): string {
   const extensions = getServerExtensions();
+
+  // Extract bibliography HTML before generateHTML escapes it as an attribute
+  const biblioHtml = extractBibliographyHtml(json);
+
   const bodyHtml = generateHTML(json as any, extensions);
-  const enrichedBody = postProcess(bodyHtml);
+  const enrichedBody = postProcess(bodyHtml, biblioHtml);
 
   const authorNames = meta.authors.map((a) => a.name);
   const authorsStr = authorNames.join(", ");
@@ -262,13 +266,30 @@ dialog.lightbox img {
 </html>`;
 }
 
+/**
+ * Walk the TipTap JSON tree to find a bibliography node, extract its
+ * renderedHtml attribute, and clear it so generateHTML outputs a clean div.
+ */
+function extractBibliographyHtml(json: Record<string, unknown>): string {
+  const content = (json as any)?.content;
+  if (!Array.isArray(content)) return "";
+  for (const node of content) {
+    if (node.type === "bibliography") {
+      const html = node.attrs?.renderedHtml || "";
+      if (node.attrs) node.attrs.renderedHtml = "";
+      return html;
+    }
+  }
+  return "";
+}
+
 /**
  * Post-process the raw generateHTML output:
+ * - Inject bibliography HTML (extracted before generateHTML)
  * - Transform accordion divs into <details><summary>
  * - Transform htmlEmbed into iframes
- * - Add semantic structure where possible
  */
-function postProcess(html: string): string {
+function postProcess(html: string, biblioHtml: string): string {
   let result = html;
 
   // Accordion: div[data-component="accordion"] → <details><summary>
@@ -285,23 +306,17 @@ function postProcess(html: string): string {
     }
   );
 
-  // Bibliography: inject renderedhtml attribute as actual HTML content
+  // Bibliography: replace empty placeholder with the pre-extracted HTML.
+  // renderedHtml was cleared before generateHTML, so the attribute is now
+  // empty and the div is clean (no > chars breaking the regex).
   result = result.replace(
     /<div([^>]*data-type="bibliography"[^>]*)><\/div>/gi,
     (_match, attrs: string) => {
-      const htmlMatch = attrs.match(/renderedhtml="([^"]*)"/i);
       const cleanAttrs = attrs.replace(/\s*renderedhtml="[^"]*"/i, "");
-      if (!htmlMatch || !htmlMatch[1]) {
-        return `<div${cleanAttrs}><p class="bibliography-empty">No citations</p></div>`;
+      if (biblioHtml) {
+        return `<div${cleanAttrs}><h2 class="bibliography-title">References</h2><div class="bibliography-content">${biblioHtml}</div></div>`;
       }
-      const decoded = htmlMatch[1]
-        .replace(/&amp;/g, "&")
-        .replace(/&lt;/g, "<")
-        .replace(/&gt;/g, ">")
-        .replace(/&quot;/g, '"')
-        .replace(/&#039;/g, "'")
-        .replace(/&#x27;/g, "'");
-      return `<div${cleanAttrs}><h2 class="bibliography-title">References</h2><div class="bibliography-content">${decoded}</div></div>`;
+      return `<div${cleanAttrs}><p class="bibliography-empty">No citations</p></div>`;
     }
   );