Fix serialization and mutation of <textarea> elements taking account the duality that the value can be set in either the child node, or in the value _parameter_ (not attribute)

Author: eoghanmurray

packages/rrweb-snapshot/src/snapshot.ts

@@ -670,6 +670,14 @@ function serializeElementNode(
     const value = (n as HTMLInputElement | HTMLTextAreaElement).value;
     const checked = (n as HTMLInputElement).checked;
     if (
+      tagName === 'textarea' &&
+      value &&
+      n.childNodes.length === 1 &&
+      n.childNodes[0].nodeType === n.TEXT_NODE &&
+      (n.childNodes[0] as Text).data === value
+    ) {
+      // value will be recorded via the childNode instead
+    } else if (
       attributes.type !== 'radio' &&
       attributes.type !== 'checkbox' &&
       attributes.type !== 'submit' &&
@@ -1077,10 +1085,19 @@ export function serializeNodeWithId(
       stylesheetLoadTimeout,
       keepIframeSrcFn,
     };
-    for (const childN of Array.from(n.childNodes)) {
-      const serializedChildNode = serializeNodeWithId(childN, bypassOptions);
-      if (serializedChildNode) {
-        serializedNode.childNodes.push(serializedChildNode);
+
+    if (
+      serializedNode.type === NodeType.Element &&
+      serializedNode.tagName === 'textarea' &&
+      serializedNode.attributes.value !== undefined
+    ) {
+      // value parameter in DOM reflects the correct value, so ignore childNode
+    } else {
+      for (const childN of Array.from(n.childNodes)) {
+        const serializedChildNode = serializeNodeWithId(childN, bypassOptions);
+        if (serializedChildNode) {
+          serializedNode.childNodes.push(serializedChildNode);
+        }
       }
     }
 

packages/rrweb-snapshot/test/__snapshots__/integration.test.ts.snap

@@ -247,6 +247,7 @@ exports[`integration tests [html file]: form-fields.html 1`] = `
       </label>
       <label for=\\"textarea\\">
         <textarea name=\\"\\" id=\\"\\" cols=\\"30\\" rows=\\"10\\">1234</textarea>
+        <textarea name=\\"\\" id=\\"\\" cols=\\"30\\" rows=\\"10\\">1234</textarea>
       </label>
       <label for=\\"select\\">
         <select name=\\"\\" id=\\"\\" value=\\"2\\">

packages/rrweb-snapshot/test/html/form-fields.html

@@ -20,6 +20,7 @@
       </label>
       <label for="textarea">
         <textarea name="" id="" cols="30" rows="10"></textarea>
+        <textarea name="" id="" cols="30" rows="10">-1</textarea>
       </label>
       <label for="select">
         <select name="" id="">
@@ -36,7 +37,8 @@
     document.querySelector('input[type="text"]').value = '1';
     document.querySelector('input[type="radio"]').checked = true;
     document.querySelector('input[type="checkbox"]').checked = true;
-    document.querySelector('textarea').value = '1234';
+    document.querySelector('textarea:empty').value = '1234';
+    document.querySelector('textarea:not(:empty)').value = '1234';
     document.querySelector('select').value = '2';
   </script>
 </html>

packages/rrweb-snapshot/test/snapshot.test.ts

@@ -7,7 +7,7 @@ import {
   serializeNodeWithId,
   _isBlockedElement,
 } from '../src/snapshot';
-import { serializedNodeWithId } from '../src/types';
+import { serializedNodeWithId, elementNode } from '../src/types';
 import { Mirror } from '../src/utils';
 
 describe('absolute url to stylesheet', () => {
@@ -218,3 +218,41 @@ describe('scrollTop/scrollLeft', () => {
     });
   });
 });
+
+describe('form', () => {
+  const serializeNode = (node: Node): serializedNodeWithId | null => {
+    return serializeNodeWithId(node, {
+      doc: document,
+      mirror: new Mirror(),
+      blockClass: 'blockblock',
+      blockSelector: null,
+      maskTextClass: 'maskmask',
+      maskTextSelector: null,
+      skipChild: false,
+      inlineStylesheet: true,
+      maskTextFn: undefined,
+      maskInputFn: undefined,
+      slimDOMOptions: {},
+      newlyAddedElement: false,
+    });
+  };
+
+  const render = (html: string): HTMLTextAreaElement => {
+    document.write(html);
+    return document.querySelector('textarea')!;
+  };
+
+  it('should record textarea values once', () => {
+    const el = render(`<textarea>Lorem ipsum</textarea>`);
+    const sel = serializeNode(el) as elementNode;
+    expect(sel?.attributes).toEqual({}); // shouldn't be stored in .value
+    expect(sel).toMatchObject({
+      childNodes: [
+        {
+          type: 3,
+          textContent: 'Lorem ipsum',
+        },
+      ],
+    });
+  });
+});

packages/rrweb/src/replay/index.ts

@@ -1564,6 +1564,8 @@ export class Replayer {
         const childNodeArray = Array.isArray(parent.childNodes)
           ? parent.childNodes
           : Array.from(parent.childNodes);
+        // This should be redundant now as we are either recording the value or the childNode, and not both
+        // keeping around for backwards compatibility with old bad double data, see
 
         // https://github.com/rrweb-io/rrweb/issues/745
         // parent is textarea, will only keep one child node as the value