[Regression]: APIRequestContext JSON-encodes string data for "application/json;charset=UTF-8" since 1.65 (not in release notes)

Version

1.65.0 (same in 1.66.0-alpha-2026-11-05). Different in 1.64.1.

Steps to reproduce

// repro.mjs — "type": "module", @playwright/test installed
import http from 'node:http';
import { request } from '@playwright/test';

const srv = http.createServer((req, res) => { let b = ''; req.on('data', c => b += c); req.on('end', () => res.end(b)); });
await new Promise(r => srv.listen(0, r));
const url = `http://127.0.0.1:${srv.address().port}/`;
const ctx = await request.newContext();

for (const ct of ['application/json', 'application/json;charset=UTF-8', 'application/json; charset=utf-8', 'Application/JSON', 'application/problem+json']) {
  const r = await ctx.post(url, { data: 'plain text', headers: { 'content-type': ct } });
  console.log(ct, '->', JSON.stringify(await r.text()));
}
await ctx.dispose(); srv.close();

Expected behavior

Either the 1.64 behavior, or the change to be listed in the release notes. In 1.64.1 only an exact application/json header made Playwright JSON-encode a string data that is not valid JSON:

application/json                -> "\"plain text\""
application/json;charset=UTF-8  -> "plain text"
application/json; charset=utf-8 -> "plain text"
Application/JSON                -> "plain text"
application/problem+json        -> "plain text"

Actual behavior

In 1.65.0 the JSON content type is matched with parameters and case-insensitively, so the string is now JSON-encoded for these headers too:

application/json                -> "\"plain text\""
application/json;charset=UTF-8  -> "\"plain text\""
application/json; charset=utf-8 -> "\"plain text\""
Application/JSON                -> "\"plain text\""
application/problem+json        -> "plain text"

A string that parses as JSON ('{"a":1}') and a Buffer are sent as is in both versions.

Additional context

The new matching looks more correct, but it silently changes the request body. It is not mentioned in the 1.65 release notes.

What broke for us:

Negative API tests that send a deliberately malformed JSON body as a string (data: '{"a": ' with content-type: application/json;charset=UTF-8). The server now receives a valid JSON string literal instead of a malformed document, so the test checks something else (400 "Syntax error" becomes a different response).
Requests signed with HMAC over the raw body string: the signature is computed over {"a": , but "{\"a\": " is sent, so the server answers 401.

Our workaround is data: Buffer.from(body).

Requests:

Please add this to the 1.65 breaking changes.
Please document in APIRequestContext data that a string with a JSON content type is sent as is only when it parses as JSON, and is JSON-encoded otherwise. The docs currently read as if strings are always sent as is. Buffer is the way to send exact bytes.

Environment

OS: Linux (mcr.microsoft.com/playwright:v1.65.0-noble)
Node.js: 26.13.0
@playwright/test: 1.65.0

Hi, I'm the Playwright bot. I reproduced this and prepared a documentation fix.

Confirmed in 1.65.0 and current main. The matching change was intentional in #43395, but the release notes and data documentation did not explain its effect on strings. The proposal adds the 1.65 breaking-change entry and documents when strings are JSON-encoded and how to send exact bytes with Buffer. It does not change request behavior.

Reproduction and checks

Using the repository server fixture:

const request = await playwright.request.newContext();
const [req] = await Promise.all([
  server.waitForRequest('/empty.html'),
  request.post(server.EMPTY_PAGE, {
    headers: { 'content-type': 'application/json;charset=UTF-8' },
    data: '{"a": ',
  }),
]);
console.log((await req.postBody).toString());
await request.dispose();

In 1.64.1 the body is {"a": . In 1.65.0 and main it is "{\"a\": ". Passing Buffer.from('{"a": ', 'utf8') preserves the original body in every version tested.

I checked application/json, application/json;charset=UTF-8, application/json; charset=utf-8, Application/JSON and application/problem+json, each with malformed JSON, a valid JSON string and a Buffer:

Version	Result
1.64.1	Only exact `application/json` JSON-encodes the malformed string.
1.65.0	All four `application/json` variants JSON-encode it.
1.66.0-next, main `17ec699e4f4760bab5ad95f6ef9b954d5f1ce7eb`	Same as 1.65.0 on Chromium, Firefox and WebKit test projects.

application/problem+json stays raw; valid JSON strings and Buffers are unchanged. These standalone API requests do not use a browser engine.

The change is commit 39ce04859ac1759611967ae76b5c590bf995149d, first released in 1.65.0. Existing request tests and the temporary reproduction passed across all three browser projects: 616 passed, 2 skipped. The build and npm run flint passed.

Draft PR: #43945

_{Triaged on Linux by the Playwright bot - agent run}

Navigation Menu

Repository navigation

[Regression]: APIRequestContext JSON-encodes string data for "application/json;charset=UTF-8" since 1.65 (not in release notes) #43937

Version

Steps to reproduce

Expected behavior

Actual behavior

Additional context

Environment

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

[Regression]: APIRequestContext JSON-encodes string data for "application/json;charset=UTF-8" since 1.65 (not in release notes) #43937

Description

Version

Steps to reproduce

Expected behavior

Actual behavior

Additional context

Environment

Activity

github-actions commented on Nov 5, 2026

microsoft-playwright-automation commented on Nov 5, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions