chrome-devtools-frontend 1.0.1657855 → 1.0.1660788

package/.agents/skills/migrate-chromium-test/SKILL.md

@@ -1,22 +1,18 @@
 ---
 name: devtools-migrate-chromium-test
-description: Use when migrating chromium `third_party/blink/web_tests/http/tests/devtools/` tests to modern unit tests in `third_party/devtools-frontend/src/front_end/`.
+description: Use when migrating Chromium layout tests to DevTools unit tests
 ---
 
-# Migrating Legacy DevTools Tests
-
-The skill should be used from within a Chromium check out. The agent should make a devtools CL for the changes in `third_party/devtools-frontend`.
-
 ## Workflow
 
 1.  **Identify the Test to Move**:
 
-- Locate the legacy test file (usually `.js`) and its expectation file (`-expected.txt`) in `third_party/blink/web_tests/http/tests/devtools/`.
-- Example: `unit/datagrid-editable-longtext.js`
+- You should have already received the issue ID, the test title, and the exact local paths of the downloaded legacy test file (`.js`) and its expectation file (`-expected.txt`) in your initial prompt.
+- Refer to these local copies as you write the new modern test.
 
 2.  **Identify the Target Location**:
 
-- Find the file being tested in `third_party/devtools-frontend/src/front_end/`.
+- Find the file being tested in `front_end/`.
 - Place the new test file next to the file being tested, following the naming convention `[FileName].test.ts`.
 - Example: If testing `ui/legacy/components/data_grid/DataGrid.ts`, the test should be `ui/legacy/components/data_grid/DataGrid.test.ts`.
 
@@ -35,21 +31,23 @@ The skill should be used from within a Chromium check out. The agent should make
 
 5.  **Verify the Changes**:
 
-- Build the unittests target: `autoninja -C out/Default <target_path>:unittests`
-- Run the test using the DevTools test runner: `npm run test -- <relative_path_to_test>`
+- Use the `verification` skill to correctly build the project and run the new test.
 - Ensure the test passes.
 
-6.  **Clean Up**:
+6. **Evaluate Test Completeness and Utility**:
 
-- Delete the original legacy test file and its expectation file from `third_party/blink/web_tests/...`.
-- _(Optional)_ If instructed by the user, you may leave the files or handle them differently.
+- Critically review the newly created test to ensure it provides meaningful coverage. Cross-reference it with the original Chromium layout test to verify that all original behaviors, edge cases, and assertions have been fully migrated and are actively being tested.
+- If the newly added test does not test any logic from the component but just the getters from the mocked data explain to the user that this test does not add value and it's better to close it without adding the test. Provide a command without executing it to close the issue as "Fixed".
+- If the newly added test uses too extensive mocks of CDP objects and methods explain to the user that the test should be foundation e2e test instead of unit test and we won't migrate it. Provide a command without executing it to close the issue as "Won't Fix (Infeasible)".
+
+7. Upload a CL with your changes using the `version-control` skill. Make sure the CL description has a `Fixed: <issue number>` trailer instead of `Bug: <issue number>`.
 
 ## Example Conversion
 
 ### Legacy Test (`datagrid-editable-longtext.js`)
 
 ```javascript
-import { TestRunner } from 'test_runner';
+import {TestRunner} from 'test_runner';
 import * as DataGrid from 'devtools/ui/legacy/components/data_grid/data_grid.js';
 
 (async function () {
@@ -65,8 +63,8 @@ import * as DataGrid from 'devtools/ui/legacy/components/data_grid/data_grid.js'
 
 ```typescript
 import * as DataGrid from './data_grid.js';
-import { renderElementIntoDOM } from '../../../../testing/DOMHelpers.js';
-import { describeWithEnvironment } from '../../../../testing/EnvironmentHelpers.js';
+import {renderElementIntoDOM} from '../../../../testing/DOMHelpers.js';
+import {describeWithEnvironment} from '../../../../testing/EnvironmentHelpers.js';
 
 describeWithEnvironment('DataGrid', () => {
   it('tests long text in datagrid', () => {
@@ -86,59 +84,6 @@ describeWithEnvironment('DataGrid', () => {
 - **Ignoring Environment Failures**: In some environments, `npm run test` may exit with code 1 due to memory leaks (e.g., `WebFrame LEAKED`) or infrastructure issues (e.g., `gpkg` or `Corp Airlock` logs), even if the tests themselves passed. Always check the end of the test log for `TOTAL: X SUCCESS` to confirm if the test logic was successful.
 - **Rendering issue**: If you encounter an issue where checking values never gives correct information it may be due to async rendering. To resolve that you may trying using `raf()` from `front_end/testing/DOMHelpers.ts` to wait for the next rendering cycle.
 
-## Creating a DevTools CL from a Chromium Checkout
-
-To create a DevTools Change List (CL) from a Chromium checkout, you'll be working within the devtools-frontend repository, which is located inside your Chromium source tree. The depot_tools package (which includes git cl) must be in your PATH.
-
-Here's the typical workflow:
-
-1.  **Navigate to the DevTools Frontend Directory**:
-    The DevTools frontend code resides in `third_party/devtools-frontend/src/` within your Chromium checkout (`chromium/src`).
-
-```bash
-cd path/to/chromium/src/third_party/devtools-frontend/src/
-```
-
-2.  **Create a New Branch**:
-    It's best practice to create a new branch for your changes:
-
-```bash
-git new-branch my-devtools-feature
-```
-
-(Or use `git checkout -b my-devtools-feature`)
-
-3.  **Make Your Code Changes**:
-    Edit the files as needed for your feature or bug fix.
-
-4.  **Commit Changes Locally**:
-    Stage and commit your changes using standard Git commands:
-
-```bash
-git add .
-git commit -m "My DevTools feature description"
-```
-
-_See the Chromium guide on commit messages for best practices on formatting descriptions, including the use of tags like Bug:. _
-
-5.  **Upload the CL for Review**:
-    Use `git cl upload` to send your changes to the Chromium Gerrit code review system:
-
-```bash
-git cl upload
-```
-
-This command will create a CL on `https://chromium-review.googlesource.com`. Follow the prompts to add a description, reviewers, etc. You can also add reviewers and trigger builds through the Gerrit web UI.
-
-6.  **Code Review Process**:
-    Your CL will be reviewed by DevTools owners. You may need to address feedback by amending your commit and uploading new patch sets using `git cl upload` again. Once approved, you can land the change via the Commit Queue (CQ).
-
-### Working on Changes Across DevTools and Chromium Core
-
-- DevTools frontend (`third_party/devtools-frontend/src/`) and the main Chromium codebase (`chromium/src`) are distinct Git repositories.
-- If your work involves changes in both areas, you generally need to create separate CLs: one within the devtools-frontend repo and one within the main `chromium/src` repo.
-- The "Juggling the git submodules" section of the DevTools documentation provides guidance on managing these cross-repository changes.
-
 ## After finishing the migration
 
-Identify any helping tips that you discovered during the migration and add them to this skill.
+If you encountered any novel issues, workarounds, or helpful tips during the migration that are not already documented, please report them to the user so they can be added to the skill documentation.

package/.agents/skills/ui-eng-vision-orchestrator/SKILL.md

@@ -11,7 +11,8 @@ DevTools views into declarative `UI.Widget` components based on lit-html,
 managing a multi-pass sequential CL pipeline to ensure each code modification is
 small, testable, and highly reviewable.
 
-Refer to [ui_engineering.md](../../../docs/ui_engineering.md) for general principles.
+Refer to [ui_engineering.md](../../../docs/ui_engineering.md) for general
+principles.
 
 --------------------------------------------------------------------------------
 
@@ -28,9 +29,12 @@ Refer to [ui_engineering.md](../../../docs/ui_engineering.md) for general princi
     file matches older patterns or generic examples. Always perform a structure
     analysis of the active file first and adapt the plan to the actual, current
     code.
-4.  **Hybrid Files**: Files may contain multiple classes in different states of
-    modernization (e.g., one partially migrated class and one fully legacy
-    class). Propose modifications only for classes with actual violations.
+4.  **Hybrid Files/Multiple Classes**: Files may contain multiple classes in
+    different states of modernization (e.g., one partially migrated class and
+    one fully legacy class). If the file being migrated contains several
+    classes, the migration needs to be done separately for each class, starting
+    with the leaves and going up the dependency chain. Propose modifications
+    only for classes with actual violations.
 5.  **Coupling Analysis**: Always check the integration files (e.g., sidebar
     files, parent panel files) where the target class is instantiated to see if
     there are any direct `.element` or `.contentElement` couplings. If found,
@@ -43,12 +47,21 @@ Refer to [ui_engineering.md](../../../docs/ui_engineering.md) for general princi
     single responsibility per CL. Do not combine technology migrations (e.g.,
     converting imperative DOM to Lit-html) with base class refactoring in the
     same diff.
-8.  **Never Break Compilation**: Run a compiler build and execute the DevTools
-    test suite between every pass.
+8.  **Strict Intermediate Stability**: The code must compile and work correctly,
+    and all tests must pass after each step. We should not split the class API
+    update from its usage update.
 9.  **Establish Comprehensive Guardrails First**: Ensure both **logic tests**
-    (verifying presenter interactions and state updates) and **screenshot tests**
-    (verifying layout/styling) exist before rewriting code. If coverage is
-    missing, scaffold it first.
+    (verifying presenter interactions and state updates) and **screenshot
+    tests** (verifying layout/styling) exist before rewriting code. If coverage
+    is missing, scaffold it first.
+10. **Pre-flight Clarification**: Before starting, verify if the task is
+    underspecified or ambiguous. If so, point this out and ask for
+    clarifications before proceeding.
+11. **Step Sizing & Complexity Assessment**: Assess the complexity of each step
+    beforehand. Aim for intermediate steps that involve changing or adding
+    roughly 300 lines of code. If a step is predicted to be too complex, break
+    it down into smaller substeps. The agent must stop and report back after
+    each step or substep to allow verification.
 
 --------------------------------------------------------------------------------
 
@@ -57,17 +70,22 @@ Refer to [ui_engineering.md](../../../docs/ui_engineering.md) for general princi
 ### Plan Generation (Gated)
 
 *   **Actions**:
-    1.  Scan the target file to determine size, complexity, legacy elements
+    1.  Verify if the task is underspecified or ambiguous. If so, ask for
+        clarifications first.
+    2.  Scan the target file to determine size, complexity, legacy elements
         (e.g., `ReportView`, `ToolbarButton`, `DataGrid`, `TreeOutline`), and
         actual class inheritance. Deconstruct the file mentally (or in the plan)
         into **Business Logic** (state, event handlers) and **Rendering Logic**
         (DOM creation, CSS classes). Reference code elements by **Name** (not
         line numbers) to ensure the plan remains valid as the codebase evolves.
-        Check if the class dependencies have been migrated to the ui eng vision or not.
-    2.  Check the files depending on this file, to understand how the code in
+        Check if the class dependencies have been migrated to the ui eng vision
+        or not. If the file contains multiple classes, plan the migration
+        separately for each class, starting with the leaves and going up the
+        dependency chain.
+    3.  Check the files depending on this file, to understand how the code in
         this file is being used.
-    3.  **Dynamic Ordering Decision**: Determine whether to run
-        Logic Consolidation first or Local lit-html rendering first:
+    4.  **Dynamic Ordering Decision**: Determine whether to run Logic
+        Consolidation first or Local lit-html rendering first:
         *   **Option A (Logic First)**: Choose this if the component has a
             complex, nested state model or highly coupled event-handlers.
             Consolidating the state variables and defining the `ViewInput`
@@ -78,9 +96,9 @@ Refer to [ui_engineering.md](../../../docs/ui_engineering.md) for general princi
             DOM construction. Converting the layout to declarative Lit templates
             first (using temporary inline or local state) makes it much easier
             to isolate and group the remaining state updates afterward.
-    4.  Generate a `migration_plan.md` reflecting this ordering and outlining
+    5.  Generate a `migration_plan.md` reflecting this ordering and outlining
         the target steps.
-    5.  Halt execution and print the formatted plan in Markdown to the chat.
+    6.  Halt execution and print the formatted plan in Markdown to the chat.
         Wait for user approval.
 
 ### Baseline & Safety Scaffolding
@@ -90,8 +108,9 @@ Refer to [ui_engineering.md](../../../docs/ui_engineering.md) for general princi
 *   **Subskill to Import**: `ui-eng-vision-test-scaffolder`
 *   **Actions**:
     1.  Verify existing test coverage. Propose and add missing **logic tests**
-        for interactions and **screenshot tests** for visual validation to avoid regressions.
-    2. Pay extra attention not to skip this step.
+        for interactions and **screenshot tests** for visual validation to avoid
+        regressions.
+    2.  Pay extra attention not to skip this step.
 
 ### Logic Consolidation
 

package/.agents/skills/version-control/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: devtools-version-control
-description: Use when starting a new task, creating a branch, switching branches, managing branches, creating and uploading CLs, or handling stacked changes in the DevTools Gerrit-based workflow. ALWAYS use this instead of running standard git checkout/switch commands for branch creation.
+description: Use when starting a new task, creating a branch, switching branches, managing branches, creating and uploading CLs, handling stacked changes, or checking release and roll status in the DevTools Gerrit-based workflow. ALWAYS use this instead of running standard git checkout/switch commands for branch creation.
 ---
 
 # DevTools Version Control
@@ -73,6 +73,12 @@ To upload an updated CL:
 git cl upload -d -t "<one sentence patch set description>"
 ```
 
+## Release and Roll Status
+
+To check whether a DevTools commit (`devtools/devtools-frontend`) has rolled into `chromium/src` (`Roll status`) and what version or channel it is deployed to (`Release status`), look it up via the Chromium Dash API:
+
+`https://chromiumdash.appspot.com/fetch_commit?commit=<sha>`
+
 ## Quick Reference
 
 | Action | Command |
@@ -83,6 +89,7 @@ git cl upload -d -t "<one sentence patch set description>"
 | Upload to Gerrit | `git cl upload` |
 | Change branch parent | `git reparent-branch <new-parent>` |
 | Sync all branches | `git rebase-update` |
+| Check release & roll status | Query `https://chromiumdash.appspot.com/fetch_commit?commit=<sha>` |
 
 ## Common Mistakes
 - **Multiple commits on one branch:** Gerrit expects one commit per CL. Always `commit --amend`.

package/AUTHORS

@@ -82,6 +82,7 @@ Nourhan Hasan <nourhan.m.hasan@gmail.com>
 Paras Awasthi <awasthiparas6@gmail.com>
 Paul Fisher <paul@pfish.zone>
 Peng Zhou <zhoupeng.1996@bytedance.com>
+Percy Wang <a945970854@gmail.com>
 Per-Olov Jernberg <possan@possan.se>
 PhistucK <phistuck@gmail.com>
 Pradhuman Singh Baid <pradhumansinghbaid@gmail.com>

package/SECURITY.md

@@ -54,3 +54,6 @@ Chrome DevTools is a privileged web app running in a sandboxed renderer process,
 
 ### Side-Effect Free Evaluation
 * Side-effect free JavaScript evaluation is a best-effort developer heuristic to prevent accidental state changes, not a security boundary. Bypassing it to cause side effects on the debugged page is not a vulnerability.
+
+### Self-XSS Paste Gate
+* The self-XSS paste gate is not a security boundary. It is supposed to reduce the chance for abuse, not to fully prevent it. Issues around bypassing it are social engineering, and outside of our threat model.

package/docs/get_the_code.md

@@ -499,3 +499,9 @@ ln -s src/third_party/devtools-frontend/src devtools-frontend
 If you did run `gclient sync` first, remove the devtools-frontend directory and start over.
 
 Run `gclient sync` after creating the link to fetch the dependencies for the standalone checkout.
+
+### Known issues
+
+- GN/Siso wrappers in `depot_tools` don't assume the side-by-side solutions, and
+  always use the binaries in Chromium. Please make sure updating Chromium when
+  you get a build error.

package/front_end/Tests.js

@@ -496,8 +496,8 @@
 
   TestSuite.prototype.testConsoleOnNavigateBack = function() {
     function filteredMessages() {
-      return SDK.ConsoleModel.ConsoleModel.allMessagesUnordered().filter(
-          a => a.source !== Protocol.Log.LogEntrySource.Violation);
+      return SDK.ConsoleModel.ConsoleModel.allMessagesUnordered(SDK.TargetManager.TargetManager.instance())
+          .filter(a => a.source !== Protocol.Log.LogEntrySource.Violation);
     }
 
     if (filteredMessages().length === 1) {
@@ -710,7 +710,7 @@
     SDK.TargetManager.TargetManager.instance().addModelListener(
         SDK.ConsoleModel.ConsoleModel, SDK.ConsoleModel.Events.MessageAdded, onConsoleMessage, this);
 
-    const messages = SDK.ConsoleModel.ConsoleModel.allMessagesUnordered();
+    const messages = SDK.ConsoleModel.ConsoleModel.allMessagesUnordered(SDK.TargetManager.TargetManager.instance());
     if (messages.length) {
       const text = messages[0].messageText;
       this.assertEquals('ready', text);
@@ -976,7 +976,7 @@
   TestSuite.prototype.testWindowInitializedOnNavigateBack = function() {
     const test = this;
     test.takeControl();
-    const messages = SDK.ConsoleModel.ConsoleModel.allMessagesUnordered();
+    const messages = SDK.ConsoleModel.ConsoleModel.allMessagesUnordered(SDK.TargetManager.TargetManager.instance());
     if (messages.length === 1) {
       checkMessages();
     } else {
@@ -985,7 +985,7 @@
     }
 
     function checkMessages() {
-      const messages = SDK.ConsoleModel.ConsoleModel.allMessagesUnordered();
+      const messages = SDK.ConsoleModel.ConsoleModel.allMessagesUnordered(SDK.TargetManager.TargetManager.instance());
       test.assertEquals(1, messages.length);
       test.assertTrue(messages[0].messageText.indexOf('Uncaught') === -1);
       test.releaseControl();
@@ -1062,7 +1062,7 @@
   };
 
   TestSuite.prototype.waitForTestResultsInConsole = function() {
-    const messages = SDK.ConsoleModel.ConsoleModel.allMessagesUnordered();
+    const messages = SDK.ConsoleModel.ConsoleModel.allMessagesUnordered(SDK.TargetManager.TargetManager.instance());
     for (let i = 0; i < messages.length; ++i) {
       const text = messages[i].messageText;
       if (text === 'PASS') {

package/front_end/core/common/Gzip.ts

@@ -38,24 +38,31 @@ export async function fileToString(file: File): Promise<string> {
  * Consider using `arrayBufferToString` instead, which can handle both gzipped and plain text buffers.
  */
 export async function decompress(gzippedBuffer: ArrayBufferLike, charset = 'utf-8'): Promise<string> {
-  const buffer = await gzipCodec(gzippedBuffer, new DecompressionStream('gzip'));
+  const buffer = await decompressToBuffer(gzippedBuffer);
   const str = new TextDecoder(charset).decode(buffer);
   return str;
 }
 
+export async function decompressToBuffer(gzippedBuffer: ArrayBufferLike): Promise<ArrayBuffer> {
+  return await gzipCodec(gzippedBuffer, new DecompressionStream('gzip'));
+}
+
 /**
  * Decompress a deflate-encoded ArrayBuffer to a string.
  * Tries 'deflate' (zlib wrapper) first, then falls back to 'deflate-raw'.
  */
 export async function decompressDeflate(buffer: ArrayBufferLike, charset = 'utf-8'): Promise<string> {
-  let decompressedBuffer: ArrayBuffer;
+  const decompressedBuffer = await decompressDeflateToBuffer(buffer);
+  return new TextDecoder(charset).decode(decompressedBuffer);
+}
+
+export async function decompressDeflateToBuffer(buffer: ArrayBufferLike): Promise<ArrayBuffer> {
   try {
-    decompressedBuffer = await gzipCodec(buffer, new DecompressionStream('deflate'));
+    return await gzipCodec(buffer, new DecompressionStream('deflate'));
   } catch {
     // Try deflate-raw format if zlib-wrapped deflate fails.
-    decompressedBuffer = await gzipCodec(buffer, new DecompressionStream('deflate-raw'));
+    return await gzipCodec(buffer, new DecompressionStream('deflate-raw'));
   }
-  return new TextDecoder(charset).decode(decompressedBuffer);
 }
 export async function compress(str: string): Promise<ArrayBuffer> {
   const encoded = new TextEncoder().encode(str);

package/front_end/core/host/AidaClient.ts

@@ -63,6 +63,7 @@ const AidaLanguageToMarkdown: Record<AidaInferenceLanguage, string> = {
 
 export class AidaAbortError extends Error {}
 export class AidaBlockError extends Error {}
+export class AidaQuotaError extends Error {}
 
 interface AiStream {
   write: (data: string) => Promise<void>;
@@ -185,12 +186,23 @@ export class AidaClient {
           debugLog('doConversation failed with error:', JSON.stringify(err));
           if (err instanceof DispatchHttpRequestClient.DispatchHttpRequestError && err.response) {
             const result = err.response;
+            if (result.statusCode === 429) {
+              stream.fail(new AidaQuotaError('Server responded: quota exceeded'));
+              return;
+            }
             if (result.statusCode === 403) {
               stream.fail(new Error('Server responded: permission denied'));
               return;
             }
             if ('error' in result && result.error) {
-              stream.fail(new Error(`Cannot send request: ${result.error} ${result.detail || ''}`));
+              const errorStr = typeof result.error === 'string' ? result.error : '';
+              const detailStr = typeof result.detail === 'string' ? result.detail : '';
+              if (errorStr.toLowerCase().includes('quota') || detailStr.toLowerCase().includes('quota')) {
+                stream.fail(new AidaQuotaError(
+                    `Cannot send request: ${result.error}${result.detail ? ` ${result.detail}` : ''}`));
+                return;
+              }
+              stream.fail(new Error(`Cannot send request: ${result.error}${result.detail ? ` ${result.detail}` : ''}`));
               return;
             }
             if ('netErrorName' in result && result.netErrorName === 'net::ERR_TIMED_OUT') {
@@ -249,6 +261,9 @@ export class AidaClient {
             thoughtSignature: result.functionCallChunk.functionCall.thoughtSignature,
           });
         } else if ('error' in result) {
+          if (typeof result.error === 'string' && result.error.toLowerCase().includes('quota')) {
+            throw new AidaQuotaError(`Server responded: ${JSON.stringify(result)}`);
+          }
           throw new Error(`Server responded: ${JSON.stringify(result)}`);
         } else {
           throw new Error(`Unknown chunk result ${JSON.stringify(result)}`);
@@ -435,21 +450,32 @@ export function getClientFeatureName(feature: ClientFeature): string {
   return name;
 }
 
-let hostConfigTrackerInstance: HostConfigTracker|undefined;
-
 export class HostConfigTracker extends Common.ObjectWrapper.ObjectWrapper<EventTypes> {
   #pollTimer?: ReturnType<typeof setTimeout>;
   #aidaAvailability?: AidaAccessPreconditions;
 
-  private constructor() {
-    super();
+  static instance({forceNew}: {
+    forceNew: boolean,
+  } = {forceNew: false}): HostConfigTracker {
+    if (!Root.DevToolsContext.globalInstance().has(HostConfigTracker) || forceNew) {
+      Root.DevToolsContext.globalInstance().set(
+          HostConfigTracker,
+          new HostConfigTracker(),
+      );
+    }
+    return Root.DevToolsContext.globalInstance().get(HostConfigTracker);
+  }
+
+  dispose(): void {
+    clearTimeout(this.#pollTimer);
+    this.listeners = undefined;
   }
 
-  static instance(): HostConfigTracker {
-    if (!hostConfigTrackerInstance) {
-      hostConfigTrackerInstance = new HostConfigTracker();
+  static removeInstance(): void {
+    if (Root.DevToolsContext.globalInstance().has(HostConfigTracker)) {
+      Root.DevToolsContext.globalInstance().get(HostConfigTracker).dispose();
+      Root.DevToolsContext.globalInstance().delete(HostConfigTracker);
     }
-    return hostConfigTrackerInstance;
   }
 
   override addEventListener(eventType: Events, listener: Common.EventTarget.EventListener<EventTypes, Events>):

package/front_end/core/host/GdpClient.ts

@@ -98,21 +98,24 @@ async function makeHttpRequest<R>(request: DispatchHttpRequestRequest): Promise<
 }
 
 const SERVICE_NAME = 'gdpService';
-let gdpClientInstance: GdpClient|null = null;
 export class GdpClient {
   #cachedProfilePromise?: Promise<Profile>;
   #cachedEligibilityPromise?: Promise<CheckElibigilityResponse>;
 
-  private constructor() {
-  }
-
   static instance({forceNew}: {
     forceNew: boolean,
   } = {forceNew: false}): GdpClient {
-    if (!gdpClientInstance || forceNew) {
-      gdpClientInstance = new GdpClient();
+    if (!Root.DevToolsContext.globalInstance().has(GdpClient) || forceNew) {
+      Root.DevToolsContext.globalInstance().set(
+          GdpClient,
+          new GdpClient(),
+      );
     }
-    return gdpClientInstance;
+    return Root.DevToolsContext.globalInstance().get(GdpClient);
+  }
+
+  static removeInstance(): void {
+    Root.DevToolsContext.globalInstance().delete(GdpClient);
   }
 
   /**

package/front_end/core/platform/StringUtilities.ts

@@ -4,6 +4,8 @@
 
 import type {Brand} from './Brand.js';
 
+let graphemeSegmenter: Intl.Segmenter|null = null;
+
 export const escapeCharacters = (inputString: string, charsToEscape: string): string => {
   let foundChar = false;
   for (let i = 0; i < charsToEscape.length; ++i) {
@@ -447,6 +449,39 @@ export const trimEndWithMaxLength = (str: string, maxLength: number): string =>
   return str.slice(0, lastSegmentIndex) + ellipsis;
 };
 
+/**
+ * Truncates a string to not exceed a maximum number of UTF-16 code units (as measured by JS `string.length`).
+ *
+ * Unlike simple character limiters, this helper is grapheme-aware and uses `Intl.Segmenter` to prevent
+ * slicing inside surrogate pairs (e.g. Plane 1+ characters or emojis like '𠜎' / '🥳') or combining characters
+ * (e.g. 'é' represented in NFD as 'e' + combining acute accent). If the limit falls in the middle of a
+ * grapheme cluster, the function backs off to drop the entire cluster, ensuring the result is always a valid
+ * Unicode string.
+ *
+ * @param str The string to truncate.
+ * @param maxCodeUnits The maximum allowed code unit length (must be >= 0).
+ * @returns The truncated string, guaranteed to be <= maxCodeUnits in length and grapheme-safe.
+ */
+export const truncateToCodeUnitLength = (str: string, maxCodeUnits: number): string => {
+  if (isNaN(maxCodeUnits) || maxCodeUnits <= 0) {
+    return '';
+  }
+  if (str.length <= maxCodeUnits) {
+    return str;
+  }
+  if (!graphemeSegmenter) {
+    graphemeSegmenter = new Intl.Segmenter(undefined, {granularity: 'grapheme'});
+  }
+  let lastSafeIndex = 0;
+  for (const {index, segment} of graphemeSegmenter.segment(str)) {
+    if (index + segment.length > maxCodeUnits) {
+      break;
+    }
+    lastSafeIndex = index + segment.length;
+  }
+  return str.slice(0, lastSafeIndex);
+};
+
 export const escapeForRegExp = (str: string): string => {
   return escapeCharacters(str, SPECIAL_REGEX_CHARACTERS);
 };

package/front_end/core/sdk/AccessibilityModel.snapshot.txt

@@ -0,0 +1,17 @@
+Title: AccessibilityModel axNodeToText correctly formats an accessibility node tree to text
+Content:
+dialog "Confirmation"
+  button "Submit" focusable: true
+=== end content
+
+Title: AccessibilityModel axNodeToText correctly formats when the root node is ignored
+Content:
+Ignored
+  button "Submit" focusable: true
+=== end content
+
+Title: AccessibilityModel axNodeToText correctly formats when a nested node is ignored
+Content:
+dialog "Confirmation"
+  button "Submit" focusable: true
+=== end content