Bug 304181 – Incorrect handling of invalid UTF-8 in streaming decoder

• Component: Platform
• Status: RESOLVED FIXED

[Nikita Skovoroda]

```js
> const x = Uint8Array.of(0xf0, 0xc3, 0x80, 42)
> new TextDecoder().decode(x) // valid
'�À*'
> const d = new TextDecoder();
> [d.decode(x.subarray(0, 1), { stream: true }), d.decode(x.subarray(1), { stream: true }), d.decode()].join('') // invalid
'�À�'
```

See https://issues.chromium.org/issues/468458744, WebKit is also affected

This is already public but has security implications

utf8 decoder is affected by the structure of underlying memory chunks
Anything checking signatures / computing hashes etc is not affected by that

Responses with the exact same bytes are decoded differently depending on network timing and chunking, and could potentially be affected by a MitM to trigger decoding to different data, without affecting TLS

See a live demo at https://tmp-demo.rray.org/utf-8

[Radar WebKit Bug Importer]

<rdar://problem/166583808>

[Darin Adler]

Pull request: https://github.com/WebKit/WebKit/pull/55452

[Nikita Skovoroda]

To clarify in addition to the title change: this does not only affect `TextDecoder`

This also affects `await res.text()` in fetch, and it also affects page/resource loads (as seen in the live demo - that's a plain html decoding differently)

[EWS]

Committed 304496@main (ab37a057cd38): <https://commits.webkit.org/304496@main>

Reviewed commits have been landed. Closing PR #55452 and removing active labels.

[Tim Nguyen (:ntim)]

https://github.com/web-platform-tests/wpt/pull/56799