@ai-react-markdown/core 1.4.0 → 1.4.2

package/dist/index.d.cts

@@ -217,10 +217,18 @@ interface AIMarkdownRenderState<TConfig extends AIMarkdownRenderConfig = AIMarkd
     documentId: string;
     /**
      * Per-document URI-safe id prefix used by all clobberable attributes
-     * (`id="…"` / `href="#…"`). Derived once by the provider as
-     * `${encodeURIComponent(documentId)}-user-content-` and exposed here so
-     * downstream consumers (placeholder components, cross-chunk anchor logic)
-     * read a single canonical source instead of recomputing locally.
+     * (`id="…"` / `href="#…"`). Derived once by the provider and exposed here
+     * so downstream consumers (placeholder components, cross-chunk anchor
+     * logic) read a single canonical source instead of recomputing locally.
+     *
+     * Derivation is `${encodeURIComponent(shortenDocumentId(documentId))}-user-content-`:
+     * `documentId` is first run through a MurmurHash3 → Base62 shortener
+     * (no-op for ids ≤16 chars, ≤6-char hash otherwise) to keep the rendered
+     * HTML compact when consumers pass long UUIDs/nanoids. The shortening
+     * is purely a render-output concern — `state.documentId` itself is left
+     * raw, and registry keying (`useDocumentRegistry`) reads that raw value.
+     * Do not reconstruct this prefix from `documentId` at call sites; always
+     * read it through this field.
      */
     clobberPrefix: string;
     /** Active render configuration. */

package/dist/index.d.ts

@@ -217,10 +217,18 @@ interface AIMarkdownRenderState<TConfig extends AIMarkdownRenderConfig = AIMarkd
     documentId: string;
     /**
      * Per-document URI-safe id prefix used by all clobberable attributes
-     * (`id="…"` / `href="#…"`). Derived once by the provider as
-     * `${encodeURIComponent(documentId)}-user-content-` and exposed here so
-     * downstream consumers (placeholder components, cross-chunk anchor logic)
-     * read a single canonical source instead of recomputing locally.
+     * (`id="…"` / `href="#…"`). Derived once by the provider and exposed here
+     * so downstream consumers (placeholder components, cross-chunk anchor
+     * logic) read a single canonical source instead of recomputing locally.
+     *
+     * Derivation is `${encodeURIComponent(shortenDocumentId(documentId))}-user-content-`:
+     * `documentId` is first run through a MurmurHash3 → Base62 shortener
+     * (no-op for ids ≤16 chars, ≤6-char hash otherwise) to keep the rendered
+     * HTML compact when consumers pass long UUIDs/nanoids. The shortening
+     * is purely a render-output concern — `state.documentId` itself is left
+     * raw, and registry keying (`useDocumentRegistry`) reads that raw value.
+     * Do not reconstruct this prefix from `documentId` at call sites; always
+     * read it through this field.
      */
     clobberPrefix: string;
     /** Active render configuration. */

package/dist/index.js

@@ -1071,6 +1071,64 @@ var defaultAIMarkdownRenderConfig = Object.freeze({
   preserveOrphanReferences: true
 });
 
+// src/components/shortenDocumentId.ts
+var BASE62_ALPHABET = "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789";
+function toBase62(n) {
+  if (n === 0) return BASE62_ALPHABET[0];
+  let v = n >>> 0;
+  let out = "";
+  while (v > 0) {
+    out = BASE62_ALPHABET[v % 62] + out;
+    v = Math.floor(v / 62);
+  }
+  return out;
+}
+function rotl32(x, r) {
+  return (x << r | x >>> 32 - r) >>> 0;
+}
+var C1 = 3432918353;
+var C2 = 461845907;
+var UTF8_ENCODER = /* @__PURE__ */ new TextEncoder();
+function murmur3_32(s, seed = 0) {
+  const bytes = UTF8_ENCODER.encode(s);
+  const len = bytes.length;
+  let h1 = seed >>> 0;
+  const nblocks = len >>> 2;
+  for (let i = 0; i < nblocks; i++) {
+    const b = i * 4;
+    let k1 = bytes[b] | bytes[b + 1] << 8 | bytes[b + 2] << 16 | bytes[b + 3] << 24;
+    k1 = Math.imul(k1, C1);
+    k1 = rotl32(k1, 15);
+    k1 = Math.imul(k1, C2);
+    h1 ^= k1;
+    h1 = rotl32(h1, 13);
+    h1 = Math.imul(h1, 5) + 3864292196 >>> 0;
+  }
+  const tail = nblocks * 4;
+  const rem = len & 3;
+  if (rem > 0) {
+    let k1 = 0;
+    if (rem === 3) k1 ^= bytes[tail + 2] << 16;
+    if (rem >= 2) k1 ^= bytes[tail + 1] << 8;
+    k1 ^= bytes[tail];
+    k1 = Math.imul(k1, C1);
+    k1 = rotl32(k1, 15);
+    k1 = Math.imul(k1, C2);
+    h1 ^= k1;
+  }
+  h1 ^= len;
+  h1 ^= h1 >>> 16;
+  h1 = Math.imul(h1, 2246822507);
+  h1 ^= h1 >>> 13;
+  h1 = Math.imul(h1, 3266489909);
+  h1 ^= h1 >>> 16;
+  return h1 >>> 0;
+}
+function shortenDocumentId(id, threshold = 16) {
+  if (id.length <= threshold) return id;
+  return toBase62(murmur3_32(id));
+}
+
 // src/context.tsx
 import { jsx } from "react/jsx-runtime";
 var AIMarkdownRenderStateContext = createContext(null);
@@ -1126,7 +1184,15 @@ var AIMarkdownRenderStateProvider = ({
       // construction site, not at the documentId storage site, so consumers
       // accessing `documentId` directly still see the raw React-native value
       // (e.g. `useId()`'s `_r_0_`) while id="..."/href="#..." bytes are safe.
-      clobberPrefix: `${encodeURIComponent(resolvedDocumentId)}-user-content-`,
+      //
+      // `shortenDocumentId` is applied here (NOT at the documentId storage
+      // site) for the same reason: consumer-supplied UUIDs and nanoids
+      // shouldn't bloat every rendered `id="…"`. Registry keying — which
+      // reads `state.documentId` directly — stays on the raw value, so the
+      // shortening is a pure HTML-output concern and the `useDocumentRegistry`
+      // API surface is unaffected. Pure function ⇒ all chunks sharing one
+      // logical documentId still produce identical prefixes.
+      clobberPrefix: `${encodeURIComponent(shortenDocumentId(resolvedDocumentId))}-user-content-`,
       config: mergedConfig
     }),
     [streaming, fontSize, variant, colorScheme, resolvedDocumentId, mergedConfig]