[Bug]: WebKit decodes HTML from route.fulfill() as windows-1252, ignoring charset=utf-8 #43649

Activity

1. Ayush-S-inha commented on Oct 26, 2026

   

   Ayush Sinha (Ayush-S-inha)

   on Oct 26, 2026

   More actions

   I was able to reproduce this on Windows with playwright-core@1.64.1 and WebKit 27.2.
   I get: [ 'windows-1252', '₹' ]

   I'd like to try working on a fix. Before I start digging into the code, could a maintainer confirm whether this issue is open for community contributions and whether the fix would likely belong in Playwright's TypeScript code or the patched WebKit engine?

   I haven't made any source changes yet.
2. Pavel Feldman (pavelfeldman)

   added

   needs-triage-win

   on Oct 26, 2026
3. github-actions commented on Oct 26, 2026

   

   github-actionsBot

   on Oct 26, 2026 – with GitHub Actions

   Contributor

   More actions

   Hi, I'm the Playwright bot and I reproduced this on Windows.

   Reproduced on current main, only in WebKit. The browser's interception handler discards the supplied charset: it replaces Content-Type with the bare MIME type and leaves the response's text encoding empty. This needs a WebKit engine change, not just a TypeScript change. Adding <meta charset="utf-8"> works around it.

   Reproduction and results

   Tested Playwright 1.65.0-next, commit e478109de5542aa6bd69f817d6cb054523af9ce4, on Windows:

   Case	Chromium 157.0.8103.15	Firefox 158.0	WebKit 27.2 (r2380)
   Fulfilled HTML, string body, UTF-8 contentType, headless	correct	correct	windows-1252, corrupted text
   Same, headed	correct	correct	windows-1252, corrupted text
   Buffer body, explicit Content-Type: text/html;charset=UTF-8, headless	correct	correct	windows-1252, corrupted text
   Real HTTP response with UTF-8 header, headless	correct	correct	correct
   Fulfilled HTML with UTF-8 meta tag, headless	correct	correct	correct

   The reported releases were not separately tested; this reproduces on main. Linux and macOS were not tested in this Windows run.

   The regression fits the existing RO-096 should work test in tests/page/page-request-fulfill.spec.ts:

   it('RO-096 should work', {
     annotation: { type: 'issue', description: 'https://github.com/microsoft/playwright/issues/43649' },
   }, async ({ page, server }) => {
     await page.route('**/*', route => route.fulfill({
       status: 201,
       headers: { foo: 'bar' },
       contentType: 'text/html; charset=utf-8',
       body: '<!doctype html><p>\u20b9</p>',
     }));
     const response = await page.goto(server.EMPTY_PAGE);
     expect(response.status()).toBe(201);
     expect(response.headers().foo).toBe('bar');
     expect(await page.evaluate(() => [
       document.characterSet, document.querySelector('p').textContent
     ])).toEqual(['UTF-8', '\u20b9']);
   });

   WebKit returns ['windows-1252', '\u00e2\u201a\u00b9']. Its received response header is text/html; Chromium and Firefox retain text/html;charset=UTF-8. Protocol logging confirms Playwright sent the full charset-bearing header to WebKit.

   Commands:

   npm run test -- page-request-fulfill.spec.ts --grep RO-096 --workers=3 --reporter=line
   npm run test -- page-request-fulfill.spec.ts --grep RO-096 --headed --retries=0 --workers=3 --reporter=line
   npm run test -- page-request-fulfill.spec.ts --grep RO-096 --retries=0 --workers=3 --reporter=line
   

   The last command included the HTTP, Buffer/header and meta-tag controls. A further WebKit-only run of those controls used DEBUG=pw:protocol with npm run wtest -- page-request-fulfill.spec.ts --grep RO-096 --retries=0 --reporter=line. Both faulty fulfillment cases failed; the HTTP and meta-tag controls passed. The initial headless regression failed on all four WebKit attempts, including retries.

   Engine fix location

   InspectorNetworkAgent::interceptRequestWithResponse constructs ResourceResponse with an empty encoding, copies the supplied headers, then overwrites Content-Type with response.mimeType(). Playwright's browser_patches/webkit/patches/bootstrap.diff keeps those lines.

   The suggested engine fix is to preserve the explicit Content-Type (using the MIME type only as a fallback) and populate ResourceResponse::textEncodingName from its charset, using WebKit's extractCharsetFromMediaType helper. Changing only the TypeScript mimeType value would not populate the encoding field.

   No engine build or engine-fix verification was attempted in this workflow.

   _{Triaged on Windows by the Playwright bot - agent run}
4. dgozman commented on Oct 28, 2026

   

   Dmitry Gozman (dgozman)

   on Oct 28, 2026

   Collaborator

   More actions

   🛑 Blocked on upstream issue https://github.com/microsoft/playwright-browsers/issues/2623
5. Dmitry Gozman (dgozman)

   added

   browser-webkitBlocked on upstream/fork webkit issueBlocked on upstream/fork webkit issue

   and removed

   needs-triage-win

   on Oct 28, 2026