@bettercms-ai/richtext 0.14.0 → 0.17.0

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@bettercms-ai/richtext",
-  "version": "0.14.0",
+  "version": "0.17.0",
   "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.",
@@ -26,7 +26,8 @@
     "test:watch": "vitest"
   },
   "dependencies": {
-    "@bettercms-ai/types": "^1.55.0",
+    "@bettercms-ai/image-url": "^0.4.1",
+    "@bettercms-ai/types": "^1.57.0",
     "@portabletext/to-html": "^6.0.0",
     "@portabletext/html": "^2.0.0",
     "@portabletext/schema": "^3.0.0"

package/src/accept.ts

@@ -30,9 +30,9 @@ export const CORE_ELEMENTS = RICHTEXT_CORE_ELEMENTS
 export const ADVANCED_ELEMENTS = RICHTEXT_ADVANCED_ELEMENTS
 
 /**
- * The V2 blocks' elements. Each is reachable from its feature through `acceptFor`, but none is in
- * the panel, the defaults or the sanitiser until its RT step ships the renderer and the ingest:
- * a toggle with nothing behind it is the fake button the owner rule forbids.
+ * The V2 blocks' elements (RT2–RT8). Each one has its Portable Text type, both renderers, the
+ * ingest and the write check, so each is in the panel, the defaults and the sanitiser like any
+ * other capability.
  */
 export const BLOCK_ELEMENTS = RICHTEXT_BLOCK_ELEMENTS
 
@@ -61,6 +61,8 @@ export const ELEMENT_GROUPS: { id: string; label: string; elements: AcceptElemen
   // `internalLink` stays in ALL_ELEMENTS for stored-config compatibility, but is intentionally
   // absent here: new authoring uses the ordinary URL Link until a real internal-entry contract exists.
   { id: 'insert', label: 'Insert', elements: ['link', 'image', 'table', 'columns', 'embed', 'checklist'] },
+  // The V2 article blocks (RT2–RT8), each with its type, renderers, ingest and write check.
+  { id: 'blocks', label: 'Article blocks', elements: ['button', 'providerEmbed', 'gallery', 'tableOfContents', 'footnotes', 'relatedPosts', 'authorBio'] },
   // `rtl` stays in ALL_ELEMENTS for stored-config compatibility. There is no editor command
   // for it yet, so new schema authoring must not advertise an inert toggle.
 ]
@@ -88,7 +90,7 @@ const ELEMENT_TAGS: Record<AcceptElement, string[]> = {
   checklist: ['ul', 'li', 'input'], table: ['table', 'thead', 'tbody', 'tr', 'th', 'td'],
   columns: ['section', 'div', 'p', 'br'],
   embed: ['figure', 'iframe', 'video', 'source', 'figcaption'],
-  // None until its RT step teaches the ingest the block's markup: a tag the sanitiser lets through
+  // Each V2 block's tags are exactly what its ingest reads back: a tag the sanitiser lets through
   // and the ingest cannot read is content an author loses on save.
   // RT3: the embed's figure and what it draws. Never `script`: X's widget script is written by the
   // renderer for a site that opted in, and no submitted markup may bring one.
@@ -97,7 +99,11 @@ const ELEMENT_TAGS: Record<AcceptElement, string[]> = {
   gallery: ['figure', 'ul', 'li', 'img', 'figcaption'],
   // RT5: the contents' nav, its title and its (nested) list of links.
   tableOfContents: ['nav', 'p', 'ol', 'li', 'a'],
-  relatedPosts: [], authorBio: [], footnotes: [],
+  // RT6: a footnote's marker and link, and the end list its notes make.
+  footnotes: ['sup', 'a', 'ol', 'li'],
+  // RT8: related posts' aside, title and list of links; an author bio's aside, avatar and lines.
+  relatedPosts: ['aside', 'p', 'ul', 'li', 'a'],
+  authorBio: ['aside', 'img', 'p'],
   // RT2: a button is an `<a data-button>`; the ingest reads it back (buttonLinkRule).
   button: ['a'],
 }
@@ -116,7 +122,7 @@ export const HTML_BACKED_UNAVAILABLE_ACCEPT_REASONS: Partial<Record<AcceptElemen
   columns: 'Embedded blocks are available in entry rich text; component rich text stores HTML.',
 }
 
-const NEW_AUTHORING_EXCLUSIONS = new Set<AcceptElement>(['internalLink', 'rtl', ...BLOCK_ELEMENTS])
+const NEW_AUTHORING_EXCLUSIONS = new Set<AcceptElement>(['internalLink', 'rtl'])
 
 /** New defaults contain every capability whose complete consumer pipeline is currently live. */
 export const DEFAULT_ACCEPT: AcceptElement[] = ALL_ELEMENTS.filter(

package/src/columns.ts

@@ -0,0 +1,117 @@
+import { PT_TYPES, type PortableText } from "./types";
+
+/**
+ * RT7: columns (`PortableTextColumns`, types.ts), the first block that holds Portable Text of its
+ * own: `{ gap, columns: [{ _key, content }] }`. A V2 column is an editable run of text, so a
+ * column holds text blocks only, and never another block (R5: depth 1). The column count is
+ * derived (`data-cols`), never stored.
+ *
+ * A stored `bcmsBlock` under `builtin:columns` (plain text per column, `gap: small|medium|large`)
+ * is read the same way, so it renders instead of the empty marker it used to; nothing writes it.
+ * A legacy block in any other shape stays a marker.
+ */
+export const PT_COLUMNS_LIMITS = { min: 2, max: 4 } as const;
+export const COLUMNS_GAPS = ["s", "m", "l"] as const;
+export type ColumnsGap = (typeof COLUMNS_GAPS)[number];
+export const BUILTIN_COLUMNS_KEY = "builtin:columns" as const;
+
+const isRecord = (v: unknown): v is Record<string, unknown> =>
+  !!v && typeof v === "object" && !Array.isArray(v);
+
+/** The inline layout both renderers write (a hosted page has no stylesheet for it): legacy columns' flex row. */
+export const COLUMNS_STYLES: Record<ColumnsGap, { row: string; col: string }> = {
+  s: { row: "display:flex;flex-wrap:wrap;gap:0.75rem", col: "flex:1 1 14rem;min-width:0" },
+  m: { row: "display:flex;flex-wrap:wrap;gap:1.25rem", col: "flex:1 1 14rem;min-width:0" },
+  l: { row: "display:flex;flex-wrap:wrap;gap:2rem", col: "flex:1 1 14rem;min-width:0" },
+};
+
+/** A legacy column's plain text as text blocks: a paragraph per blank-line-separated run. */
+const legacyContent = (text: string, key: string): PortableText =>
+  text
+    .split(/\n{2,}/)
+    .filter((p) => p.trim() !== "")
+    .map((p, i) => ({ _type: "block", _key: `${key}p${i}`, style: "normal", markDefs: [], children: [{ _type: "span", _key: `${key}s${i}`, text: p, marks: [] }] }) as never);
+
+/**
+ * What a reader sees of a columns block, ours or a legacy `builtin:columns`: its gap and each
+ * column's content. Null for anything else. Read defensively: a column without content renders empty.
+ */
+export function columnsView(node: unknown): { gap: ColumnsGap; columns: { key: string; content: PortableText }[] } | null {
+  if (!isRecord(node)) return null;
+  if (node._type === PT_TYPES.columns) {
+    const gap: ColumnsGap = node.gap === "s" || node.gap === "l" ? node.gap : "m";
+    const columns = (Array.isArray(node.columns) ? node.columns : []).map((c, i) => ({
+      key: isRecord(c) && typeof c._key === "string" && c._key ? c._key : `c${i}`,
+      content: (isRecord(c) && Array.isArray(c.content) ? c.content : []) as PortableText,
+    }));
+    return { gap, columns };
+  }
+  if (node._type === PT_TYPES.block && node.schemaKey === BUILTIN_COLUMNS_KEY) {
+    const f = isRecord(node.fields) ? node.fields : {};
+    // Only the shape the legacy renderer wrote (plain text per column) reads as columns; any other
+    // stays the marker, which keeps "something is here" visible rather than drawing it empty.
+    if (!Array.isArray(f.columns) || !f.columns.every((c) => typeof c === "string")) return null;
+    const gap: ColumnsGap = f.gap === "small" ? "s" : f.gap === "large" ? "l" : "m";
+    const columns = (Array.isArray(f.columns) ? f.columns : []).slice(0, PT_COLUMNS_LIMITS.max).map((c, i) => ({
+      key: `c${i}`,
+      content: legacyContent(typeof c === "string" ? c : "", `c${i}`),
+    }));
+    return { gap, columns };
+  }
+  return null;
+}
+
+/** One reason a stored columns block is refused, at its own path inside the block. */
+export type PortableTextColumnsProblem = { path: Array<string | number>; code: string; message: string };
+
+const KEYS = new Set(["_type", "_key", "gap", "columns"]);
+
+/**
+ * Every reason `node` cannot be stored: `columns_invalid` (a shape problem, a column holding
+ * anything but text blocks, which includes another columns block (R5) and a footnote) or
+ * `columns_limit_exceeded` (outside V2's 2 to 4 columns). A derived `cols` is refused by R8, by name.
+ */
+export function columnsProblems(node: unknown): PortableTextColumnsProblem[] {
+  const out: PortableTextColumnsProblem[] = [];
+  if (!isRecord(node) || node._type !== PT_TYPES.columns) return out;
+  const bad = (path: Array<string | number>, message: string, code = "columns_invalid") =>
+    out.push({ path, code, message: `${code}: ${message}` });
+  for (const k of Object.keys(node)) if (!KEYS.has(k) && k !== "cols") bad([k], `unknown key "${k}" (gap, columns)`);
+  if (!(COLUMNS_GAPS as readonly unknown[]).includes(node.gap)) bad(["gap"], 'gap must be "s", "m" or "l"');
+  if (!Array.isArray(node.columns)) {
+    bad(["columns"], "columns must be an array");
+    return out;
+  }
+  const n = node.columns.length;
+  if (n < PT_COLUMNS_LIMITS.min || n > PT_COLUMNS_LIMITS.max) {
+    bad(["columns"], `${n} columns; ${PT_COLUMNS_LIMITS.min} to ${PT_COLUMNS_LIMITS.max}`, "columns_limit_exceeded");
+  }
+  node.columns.forEach((col, i) => {
+    const at = ["columns", i];
+    if (!isRecord(col)) return bad(at, `columns[${i}] must be an object`);
+    if (typeof col._key !== "string" || !col._key) bad([...at, "_key"], `columns[${i}] needs a _key`);
+    for (const k of Object.keys(col)) if (k !== "_key" && k !== "content") bad([...at, k], `unknown key "${k}" (content)`);
+    if (!Array.isArray(col.content)) return bad([...at, "content"], "content must be an array of text blocks");
+    col.content.forEach((b, j) => {
+      const bp = [...at, "content", j];
+      if (!isRecord(b) || b._type !== "block") {
+        return bad(bp, `a column holds text blocks only, not ${isRecord(b) && typeof b._type === "string" ? `"${b._type}"` : "this"}`);
+      }
+      (Array.isArray(b.children) ? b.children : []).forEach((c, k) => {
+        if (isRecord(c) && c._type === PT_TYPES.footnote) bad([...bp, "children", k], "a footnote belongs in the body text, not a column");
+      });
+    });
+  });
+  return out;
+}
+
+/** `columnsProblems` for every columns block at the top level of a value, with its block key. */
+export function portableTextColumnsProblems(value: unknown): Array<PortableTextColumnsProblem & { block: string }> {
+  const out: Array<PortableTextColumnsProblem & { block: string }> = [];
+  for (const node of Array.isArray(value) ? value : []) {
+    if (!isRecord(node) || node._type !== PT_TYPES.columns) continue;
+    const block = typeof node._key === "string" ? node._key : "";
+    for (const p of columnsProblems(node)) out.push({ ...p, block });
+  }
+  return out;
+}

package/src/footnote.ts

@@ -0,0 +1,70 @@
+import { PT_TYPES } from "./types";
+import { isPortableTextTable } from "./table";
+
+/**
+ * RT6: the footnote (`PortableTextFootnote`, types.ts), an inline object in body text that stores
+ * only its `note` (plain text, as V2 keeps it on the marker). Everything else is derived on every
+ * render: its number (its place in text order across the value's top-level text blocks), the
+ * `fnref-N` / `fn-N` ids, the end list and its back links.
+ *
+ * Not the table's footnote (`PortableTextTableFootnote`), which belongs to a table and stays apart.
+ */
+const isRecord = (v: unknown): v is Record<string, unknown> =>
+  !!v && typeof v === "object" && !Array.isArray(v);
+
+/**
+ * Each footnote in the value's top-level text blocks, numbered in text order (V2
+ * `renumberFootnotes`), and the notes in that order. `nodes` maps each marker to its number, which
+ * is how a renderer finds the number of the marker it is drawing.
+ */
+export function footnoteOrder(value: unknown): { notes: string[]; nodes: WeakMap<object, number> } {
+  const notes: string[] = [];
+  const nodes = new WeakMap<object, number>();
+  for (const node of Array.isArray(value) ? value : []) {
+    if (!isRecord(node) || node._type !== "block" || !Array.isArray(node.children)) continue;
+    for (const child of node.children) {
+      if (!isRecord(child) || child._type !== PT_TYPES.footnote) continue;
+      nodes.set(child, notes.push(typeof child.note === "string" ? child.note : ""));
+    }
+  }
+  return { notes, nodes };
+}
+
+/** One reason a stored footnote is refused, with the block it sits in. */
+export type PortableTextFootnoteProblem = { block: string; path: Array<string | number>; code: string; message: string };
+
+const KEYS = new Set(["_type", "_key", "note"]);
+
+/**
+ * Every reason a value's footnotes cannot be stored, by block and path: `footnote_invalid` for a
+ * shape problem, and for a footnote in a table cell (V2 refuses one inside a block: a cell has no
+ * place in the body's numbering). A derived `number` is refused by the R8 check, by name.
+ */
+export function portableTextFootnoteProblems(value: unknown): PortableTextFootnoteProblem[] {
+  const out: PortableTextFootnoteProblem[] = [];
+  const bad = (block: string, path: Array<string | number>, message: string) =>
+    out.push({ block, path, code: "footnote_invalid", message: `footnote_invalid: ${message}` });
+  const blockKey = (n: Record<string, unknown>) => (typeof n._key === "string" ? n._key : "");
+  const each = (blocks: unknown, inCell: string | null) => {
+    for (const node of Array.isArray(blocks) ? blocks : []) {
+      if (isPortableTextTable(node)) {
+        for (const row of node.rows as unknown[]) {
+          if (!isRecord(row) || !Array.isArray(row.cells)) continue;
+          for (const cell of row.cells) if (isRecord(cell)) each(cell.content, blockKey(node as Record<string, unknown>));
+        }
+        continue;
+      }
+      if (!isRecord(node) || node._type !== "block" || !Array.isArray(node.children)) continue;
+      node.children.forEach((child, i) => {
+        if (!isRecord(child) || child._type !== PT_TYPES.footnote) return;
+        const block = inCell ?? blockKey(node);
+        const path = inCell === null ? ["children", i] : [blockKey(node), "children", i];
+        if (inCell !== null) bad(block, path, "a footnote belongs in the body text, not a table cell");
+        for (const k of Object.keys(child)) if (!KEYS.has(k) && k !== "number") bad(block, [...path, k], `unknown key "${k}" (note)`);
+        if (typeof child.note !== "string") bad(block, [...path, "note"], "note must be a string");
+      });
+    }
+  };
+  each(value, null);
+  return out;
+}

package/src/gallery.ts

@@ -1,6 +1,7 @@
 import { PT_TYPES } from "./types";
 import { isPortableTextTable } from "./table";
 import { safeHref } from "./href";
+import { richTextImageFrame } from "./image-layout";
 
 /**
  * RT4: the gallery (`PortableTextGallery`, types.ts): the limits, the write check, and the read
@@ -27,7 +28,8 @@ export type PortableTextGalleryProblem = { path: Array<string | number>; code: s
 
 const KEYS = new Set(["_type", "_key", "layout", "images"]);
 const ITEM_KEYS = new Set(["_key", "image", "caption"]);
-const IMAGE_KEYS = new Set(["src", "alt", "mediaId"]);
+const IMAGE_KEYS = new Set(["src", "alt", "mediaId", "crop", "focal"]);
+const fraction = (v: unknown): v is number => typeof v === "number" && Number.isFinite(v) && v >= 0 && v <= 1;
 
 /**
  * Every reason `node` cannot be stored, each named with a code: `gallery_invalid` (a shape
@@ -63,6 +65,18 @@ export function galleryProblems(node: unknown): PortableTextGalleryProblem[] {
     if (typeof image.src !== "string") bad([...at, "image", "src"], "src must be a string");
     if (typeof image.alt !== "string") bad([...at, "image", "alt"], "alt must be a string");
     if (image.mediaId !== undefined && typeof image.mediaId !== "string") bad([...at, "image", "mediaId"], "mediaId must be a string");
+    // RT9: the rich-text image's framing, image-url's shape: edges cut as 0–1 fractions, a 0–1 point.
+    const crop = image.crop;
+    if (crop !== undefined && crop !== null) {
+      const c = isRecord(crop) ? crop : {};
+      if (![c.top, c.right, c.bottom, c.left].every(fraction) || (c.top as number) + (c.bottom as number) >= 1 || (c.left as number) + (c.right as number) >= 1) {
+        bad([...at, "image", "crop"], "crop must be { top, right, bottom, left }, 0–1 fractions that leave some picture");
+      }
+    }
+    const focal = image.focal;
+    if (focal !== undefined && focal !== null && !(isRecord(focal) && fraction(focal.x) && fraction(focal.y))) {
+      bad([...at, "image", "focal"], "focal must be { x, y }, 0–1 fractions");
+    }
   });
   if (node.images.length > PT_GALLERY_LIMITS.images) {
     bad(["images"], `${node.images.length} images; at most ${PT_GALLERY_LIMITS.images}`, "gallery_limit_exceeded");
@@ -128,17 +142,23 @@ export const GALLERY_STYLES: Record<GalleryLayout, { ul: string; li: string; fig
  * `renderArticle("gallery")` leaves it out. Read defensively: a value that never passed the write
  * check renders what it can.
  */
-export function galleryView(value: unknown): { layout: GalleryLayout; images: { key: string; src: string; alt: string; caption: string }[] } {
+export type GalleryViewImage = { key: string; src: string; alt: string; caption: string; position: string | null; data: [string, string][] };
+
+export function galleryView(value: unknown): { layout: GalleryLayout; images: GalleryViewImage[] } {
   const v = isRecord(value) ? value : {};
   const layout: GalleryLayout = v.layout === "carousel" ? "carousel" : "grid";
-  const images: { key: string; src: string; alt: string; caption: string }[] = [];
+  const images: GalleryViewImage[] = [];
   (Array.isArray(v.images) ? v.images : []).forEach((item, i) => {
     if (!isRecord(item) || !isRecord(item.image)) return;
     const src = galleryMediaSrc(typeof item.image.src === "string" ? item.image.src.trim() : "");
     if (!src) return;
+    // RT9: the crop and focal point, as a rich-text image draws them (image-layout.ts).
+    const frame = richTextImageFrame(src, item.image.crop, item.image.focal);
     images.push({
       key: typeof item._key === "string" && item._key ? item._key : `i${i}`,
-      src,
+      src: frame.src,
+      position: frame.position,
+      data: frame.data,
       alt: typeof item.image.alt === "string" ? item.image.alt : "",
       caption: typeof item.caption === "string" ? item.caption.trim() : "",
     });

package/src/href.ts

@@ -25,3 +25,42 @@ export function safeHref(raw: string): string | null {
     return null;
   }
 }
+
+/**
+ * A MEDIA url: `safeHref`, plus a plain relative path.
+ *
+ * An imported page is full of `images/hero.png` and `../media/clip.mp4`, and `safeHref` refuses
+ * a path with no leading slash — so every one of those images would render as nothing. Widened
+ * only for `<img>`, `<video>` and a poster: they load a picture, while an `<iframe>` frames a
+ * whole page and keeps the strict check. A colon or whitespace is refused, so no scheme
+ * (`javascript:`) can ride this path, and a PROTOCOL-RELATIVE `//host/clip.mp4` is refused with
+ * it — it is off-origin, not relative, and `safeHref` rejects the same shape for the same reason.
+ */
+export const safeMediaSrc = (raw: string): string | null =>
+  safeHref(raw) ?? (raw !== "" && !raw.startsWith("//") && !/[:\s]/.test(raw) ? raw : null);
+
+/** What a new tab always carries on render, so it is never the author's own `rel`. */
+const NEW_TAB_REL = ["noopener", "noreferrer"];
+
+/**
+ * A link's `rel` is author data (V2's link form writes `nofollow`, `sponsored`, `ugc`; the
+ * importer keeps `prev`/`next`): plain link-type tokens only, lowercased, once each. V2's
+ * `follow` is its "no rel" and is no token.
+ */
+const relTokens = (raw: unknown): string[] =>
+  typeof raw === "string"
+    ? [...new Set(raw.toLowerCase().split(/\s+/).filter((t) => /^[a-z][a-z-]*$/.test(t) && t !== "follow"))]
+    : [];
+
+/** The `rel` to store from a link's markup: a new tab's noopener/noreferrer are the renderer's, not kept. */
+export function readLinkRel(raw: unknown, blank: boolean): string | undefined {
+  const tokens = relTokens(raw).filter((t) => !blank || !NEW_TAB_REL.includes(t));
+  return tokens.length ? tokens.join(" ") : undefined;
+}
+
+/** The `rel` both renderers write: the stored tokens, and a new tab always adds noopener + noreferrer. */
+export function writeLinkRel(raw: unknown, blank: boolean): string | undefined {
+  const tokens = new Set(relTokens(raw));
+  if (blank) for (const t of NEW_TAB_REL) tokens.add(t);
+  return tokens.size ? [...tokens].join(" ") : undefined;
+}