@bettercms-ai/richtext 0.12.0 → 0.12.1

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@bettercms-ai/richtext",
-  "version": "0.12.0",
+  "version": "0.12.1",
   "type": "module",
   "sideEffects": false,
   "description": "The one Portable Text contract for BetterCMS: schema, converters, and rendering. Shared by the backend, the dashboard and every delivery SDK.",

package/src/from-lexical.ts

@@ -95,7 +95,8 @@ function spansFrom(
   if (!isRecord(value)) return [];
   const node = value as LexNode;
 
-  if (node.type === "link") {
+  // An `autolink` is Lexical's link for a URL typed as text: the same annotation, so the same mark.
+  if (node.type === "link" || node.type === "autolink") {
     const href = safeHref(node.url ?? "");
     // An unsafe href does not become an unsafe link and does not silently delete the label —
     // the text survives, unlinked.

package/src/ingest.ts

@@ -795,11 +795,39 @@ const allowsBlockObject = (schema: PtSchema | undefined, name: string): boolean
 const isElement = (n: unknown): n is Element =>
   !!n && typeof n === "object" && (n as Node).nodeType === 1;
 
+const BLOCK_TAGS = new Set([
+  "ADDRESS", "ARTICLE", "ASIDE", "BLOCKQUOTE", "DD", "DETAILS", "DIV", "DL", "DT", "FIGCAPTION", "FIGURE", "FOOTER",
+  "H1", "H2", "H3", "H4", "H5", "H6", "HEADER", "HR", "LI", "MAIN", "NAV", "OL", "P", "PRE", "SECTION", "TABLE", "UL",
+]);
+
+/**
+ * The parser, minus the indentation between block elements. The converter skips that whitespace
+ * at the top level but turns it into an empty block inside a wrapper, so `<div><p>a</p>\n<p>b</p></div>`
+ * came back as three blocks. Only ASCII whitespace beside a block element goes: a space between two
+ * inline elements is a word break, and `&nbsp;` or a `<br>` is an author's blank line.
+ */
+const withoutBlockWhitespace =
+  (parse: (html: string) => Document) =>
+  (html: string): Document => {
+    const doc = parse(html);
+    const walker = doc.createTreeWalker(doc.body ?? doc, 4 /* NodeFilter.SHOW_TEXT */);
+    const blank: Node[] = [];
+    for (let n = walker.nextNode(); n; n = walker.nextNode()) {
+      const beside = [n.previousSibling, n.nextSibling].some((s) => !!s && BLOCK_TAGS.has(s.nodeName));
+      if (beside && /^[ \t\n\r]*$/.test(n.nodeValue ?? "") && n.parentNode?.nodeName !== "PRE") blank.push(n);
+    }
+    for (const n of blank) n.parentNode?.removeChild(n);
+    return doc;
+  };
+
 export function htmlToPortableText(html: string, opts: IngestOptions = {}): PortableText {
   if (typeof html !== "string" || html.trim() === "") return [];
   const key = opts.keyGenerator ?? randomKey;
+  const parse =
+    opts.parseHtml ?? (typeof DOMParser !== "undefined" ? (markup: string) => new DOMParser().parseFromString(markup, "text/html") : undefined);
   const converted = ptHtmlToPortableText(html, {
-    ...(opts.parseHtml ? { parseHtml: opts.parseHtml } : {}),
+    // The browser's own parser when none is passed (Node must pass one), so both drop the same whitespace.
+    ...(parse ? { parseHtml: withoutBlockWhitespace(parse) } : {}),
     // `ptSchemaFor` returns a schema DEFINITION — the same object `EditorProvider` takes, so
     // the editor and this filter are literally configured from one value. The deserialiser
     // wants the COMPILED form, so the compile happens here rather than at every call site.