From dc20cd45cc63058325784444af6bd32ed2cace48 Mon Sep 17 00:00:00 2001
From: Eoghan Murray <eoghan@getthere.ie>
Date: Fri, 10 Jan 2025 18:27:13 +0000
Subject: [PATCH] Fix performance of splitCssText  (#1615)

- Fix bug where the right split point was not being picked for the 3rd section onwards
- Fix that it wasn't able to find a split when both halves were identical
- Add test to put splitCssText through it's paces with a large file
- Introduce a limit on the iteration which causes the 'efficiently' test to fail
- Fix poor 'crawling' performance in the 'matching' algorithm for large css texts - e.g. for a (doubled) benchmark.css, we were running `normalizeCssText` 9480 times before `k` got to the right place
- Further algorithm efficiency: need to take larger jumps; use the scaling factor to make better guess at how big a jump to make
---
 .changeset/efficiently-splitCssText-1603.md   |  6 ++
 packages/rrweb-snapshot/src/utils.ts          | 65 +++++++++++++++----
 packages/rrweb-snapshot/test/css.test.ts      | 58 +++++++++++++++++
 .../test/__snapshots__/replayer.test.ts.snap  |  4 +-
 .../test/events/style-sheet-rule-events.ts    |  2 +-
 5 files changed, 121 insertions(+), 14 deletions(-)
 create mode 100644 .changeset/efficiently-splitCssText-1603.md

diff --git a/.changeset/efficiently-splitCssText-1603.md b/.changeset/efficiently-splitCssText-1603.md
new file mode 100644
index 0000000000..57c6d5e6c4
--- /dev/null
+++ b/.changeset/efficiently-splitCssText-1603.md
@@ -0,0 +1,6 @@
+---
+"rrweb-snapshot": patch
+"rrweb": patch
+---
+
+Improve performance of splitCssText for <style> elements with large css content - see #1603
diff --git a/packages/rrweb-snapshot/src/utils.ts b/packages/rrweb-snapshot/src/utils.ts
index 42aba4d649..e1ae7a3ad4 100644
--- a/packages/rrweb-snapshot/src/utils.ts
+++ b/packages/rrweb-snapshot/src/utils.ts
@@ -456,8 +456,9 @@ export function normalizeCssString(cssText: string): string {
 
 /**
  * Maps the output of stringifyStylesheet to individual text nodes of a <style> element
- * performance is not considered as this is anticipated to be very much an edge case
- * (javascript is needed to add extra text nodes to a <style>)
+ * which occurs when javascript is used to append to the style element
+ * and may also occur when browsers opt to break up large text nodes
+ * performance needs to be considered, see e.g. #1603
  */
 export function splitCssText(
   cssText: string,
@@ -465,27 +466,69 @@ export function splitCssText(
 ): string[] {
   const childNodes = Array.from(style.childNodes);
   const splits: string[] = [];
+  let iterLimit = 0;
   if (childNodes.length > 1 && cssText && typeof cssText === 'string') {
-    const cssTextNorm = normalizeCssString(cssText);
+    let cssTextNorm = normalizeCssString(cssText);
+    const normFactor = cssTextNorm.length / cssText.length;
     for (let i = 1; i < childNodes.length; i++) {
       if (
         childNodes[i].textContent &&
         typeof childNodes[i].textContent === 'string'
       ) {
         const textContentNorm = normalizeCssString(childNodes[i].textContent!);
-        for (let j = 3; j < textContentNorm.length; j++) {
-          // find a  substring that appears only once
+        let j = 3;
+        for (; j < textContentNorm.length; j++) {
+          if (
+            // keep consuming css identifiers (to get a decent chunk more quickly)
+            textContentNorm[j].match(/[a-zA-Z0-9]/) ||
+            // substring needs to be unique to this section
+            textContentNorm.indexOf(textContentNorm.substring(0, j), 1) !== -1
+          ) {
+            continue;
+          }
+          break;
+        }
+        for (; j < textContentNorm.length; j++) {
           const bit = textContentNorm.substring(0, j);
-          if (cssTextNorm.split(bit).length === 2) {
-            const splitNorm = cssTextNorm.indexOf(bit);
+          // this substring should appears only once in overall text too
+          const bits = cssTextNorm.split(bit);
+          let splitNorm = -1;
+          if (bits.length === 2) {
+            splitNorm = cssTextNorm.indexOf(bit);
+          } else if (
+            bits.length > 2 &&
+            bits[0] === '' &&
+            childNodes[i - 1].textContent !== ''
+          ) {
+            // this childNode has same starting content as previous
+            splitNorm = cssTextNorm.indexOf(bit, 1);
+          }
+          if (splitNorm !== -1) {
             // find the split point in the original text
-            for (let k = splitNorm; k < cssText.length; k++) {
-              if (
-                normalizeCssString(cssText.substring(0, k)).length === splitNorm
-              ) {
+            let k = Math.floor(splitNorm / normFactor);
+            for (; k > 0 && k < cssText.length; ) {
+              iterLimit += 1;
+              if (iterLimit > 50 * childNodes.length) {
+                // quit for performance purposes
+                splits.push(cssText);
+                return splits;
+              }
+              const normPart = normalizeCssString(cssText.substring(0, k));
+              if (normPart.length === splitNorm) {
                 splits.push(cssText.substring(0, k));
                 cssText = cssText.substring(k);
+                cssTextNorm = cssTextNorm.substring(splitNorm);
                 break;
+              } else if (normPart.length < splitNorm) {
+                k += Math.max(
+                  1,
+                  Math.floor((splitNorm - normPart.length) / normFactor),
+                );
+              } else {
+                k -= Math.max(
+                  1,
+                  Math.floor((normPart.length - splitNorm) * normFactor),
+                );
               }
             }
             break;
diff --git a/packages/rrweb-snapshot/test/css.test.ts b/packages/rrweb-snapshot/test/css.test.ts
index 99a5b362e9..a26e0005fd 100644
--- a/packages/rrweb-snapshot/test/css.test.ts
+++ b/packages/rrweb-snapshot/test/css.test.ts
@@ -7,6 +7,8 @@ import postcss, { type AcceptedPlugin } from 'postcss';
 import { JSDOM } from 'jsdom';
 import { splitCssText, stringifyStylesheet } from './../src/utils';
 import { applyCssSplits } from './../src/rebuild';
+import * as fs from 'fs';
+import * as path from 'path';
 import type {
   serializedElementNodeWithId,
   BuildCache,
@@ -105,10 +107,16 @@ describe('css splitter', () => {
       // as authored, e.g. no spaces
       style.append('.a{background-color:black;}');
 
+      // test how normalization finds the right sections
+      style.append('.b      {background-color:black;}');
+      style.append('.c{      background-color:                     black}');
+
       // how it is currently stringified (spaces present)
       const expected = [
         '.a { background-color: red; }',
         '.a { background-color: black; }',
+        '.b { background-color: black; }',
+        '.c { background-color: black; }',
       ];
       const browserSheet = expected.join('');
       expect(stringifyStylesheet(style.sheet!)).toEqual(browserSheet);
@@ -137,6 +145,28 @@ describe('css splitter', () => {
     }
   });
 
+  it('finds css textElement splits correctly with two identical text nodes', () => {
+    const window = new Window({ url: 'https://localhost:8080' });
+    const document = window.document;
+    // as authored, with comment, missing semicolons
+    const textContent = '.a { color:red; } .b { color:blue; }';
+    document.head.innerHTML = '<style></style>';
+    const style = document.querySelector('style');
+    if (style) {
+      style.append(textContent);
+      style.append(textContent);
+
+      const expected = [textContent, textContent];
+      const browserSheet = expected.join('');
+      expect(splitCssText(browserSheet, style)).toEqual(expected);
+
+      style.append(textContent);
+      const expected3 = [textContent, textContent, textContent];
+      const browserSheet3 = expected3.join('');
+      expect(splitCssText(browserSheet3, style)).toEqual(expected3);
+    }
+  });
+
   it('finds css textElement splits correctly when vendor prefixed rules have been removed', () => {
     const style = JSDOM.fragment(`<style></style>`).querySelector('style');
     if (style) {
@@ -169,6 +199,34 @@ describe('css splitter', () => {
       expect(splitCssText(browserSheet, style)).toEqual(expected);
     }
   });
+
+  it('efficiently finds split points in large files', () => {
+    const cssText = fs.readFileSync(
+      path.resolve(__dirname, './css/benchmark.css'),
+      'utf8',
+    );
+
+    const parts = cssText.split('}');
+    const sections = [];
+    for (let i = 0; i < parts.length - 1; i++) {
+      if (i % 100 === 0) {
+        sections.push(parts[i] + '}');
+      } else {
+        sections[sections.length - 1] += parts[i] + '}';
+      }
+    }
+    sections[sections.length - 1] += parts[parts.length - 1];
+
+    expect(cssText.length).toEqual(sections.join('').length);
+
+    const style = JSDOM.fragment(`<style></style>`).querySelector('style');
+    if (style) {
+      sections.forEach((section) => {
+        style.appendChild(JSDOM.fragment(section));
+      });
+    }
+    expect(splitCssText(cssText, style)).toEqual(sections);
+  });
 });
 
 describe('applyCssSplits css rejoiner', function () {
diff --git a/packages/rrweb/test/__snapshots__/replayer.test.ts.snap b/packages/rrweb/test/__snapshots__/replayer.test.ts.snap
index c4edd6d540..4d4711ba63 100644
--- a/packages/rrweb/test/__snapshots__/replayer.test.ts.snap
+++ b/packages/rrweb/test/__snapshots__/replayer.test.ts.snap
@@ -72,7 +72,7 @@ file-cid-3
 
 .css-added-at-200 { position: fixed; top: 0px; right: 0px; left: 4rem; z-index: 15; flex-shrink: 0; height: 0.25rem; overflow: hidden; background-color: rgb(17, 171, 209); }
 
-.css-added-at-200.alt { height: 0.25rem; background-color: rgb(190, 232, 242); opacity: 0; transition: opacity 0.5s ease 0s; }
+.css-added-at-200.alt { height: 0.25rem; background-color: rgb(190, 232, 242); opacity: 0; transition: opacity 0.5s ease-in 0.1s; }
 
 .css-added-at-1000-deleted-at-2500 { display: flex; flex-direction: column; min-width: 60rem; min-height: 100vh; color: blue; }
 
@@ -152,7 +152,7 @@ file-cid-3
 
 .css-added-at-200 { position: fixed; top: 0px; right: 0px; left: 4rem; z-index: 15; flex-shrink: 0; height: 0.25rem; overflow: hidden; background-color: rgb(17, 171, 209); }
 
-.css-added-at-200.alt { height: 0.25rem; background-color: rgb(190, 232, 242); opacity: 0; transition: opacity 0.5s ease 0s; }
+.css-added-at-200.alt { height: 0.25rem; background-color: rgb(190, 232, 242); opacity: 0; transition: opacity 0.5s ease-in 0.1s; }
 
 .css-added-at-200.alt2 { padding-left: 4rem; }
 "
diff --git a/packages/rrweb/test/events/style-sheet-rule-events.ts b/packages/rrweb/test/events/style-sheet-rule-events.ts
index 577c0caf41..4835961d32 100644
--- a/packages/rrweb/test/events/style-sheet-rule-events.ts
+++ b/packages/rrweb/test/events/style-sheet-rule-events.ts
@@ -65,7 +65,7 @@ const events: eventWithTime[] = [
                     tagName: 'style',
                     attributes: {
                       _cssText:
-                        '.css-added-at-200 { position: fixed; top: 0px; right: 0px; left: 4rem; z-index: 15; flex-shrink: 0; height: 0.25rem; overflow: hidden; background-color: rgb(17, 171, 209); }.css-added-at-200.alt { height: 0.25rem; background-color: rgb(190, 232, 242); opacity: 0; transition: opacity 0.5s ease 0s; }.css-added-at-200.alt2 { padding-left: 4rem; }',
+                        '.css-added-at-200 { position: fixed; top: 0px; right: 0px; left: 4rem; z-index: 15; flex-shrink: 0; height: 0.25rem; overflow: hidden; background-color: rgb(17, 171, 209); }.css-added-at-200.alt { height: 0.25rem; background-color: rgb(190, 232, 242); opacity: 0; transition: opacity 0.5s ease-in 0.1s; }.css-added-at-200.alt2 { padding-left: 4rem; }',
                       'data-emotion': 'css',
                     },
                     childNodes: [