Bug 313230 – [JSC] `new DataView(buffer, byteOffset, byteLength)` should compliant the step 9-b of `25.3.2.1 DataView` abstract operation in the spec

• Status: RESOLVED
• Resolution: FIXED
• Component: JavaScriptCore

[Tetsuharu Ohzeki [UTC+9]]

Created attachment 479282 [details]
testcase as jsc stress test

According to the spec (April 24, 2026),  the following code (the attached diff contains ) should works, but not at https://commits.webkit.org/309786@main.
I attaches the jsc stress test format.


```js
function assert(ok, message = '') {
    if (!ok)
        throw new Error(`Assertion!: ${message}`); 
}

function sameValue(a, b, testname) {
    if (a !== b)
        throw new Error(`${testname}: Expected ${b} but got ${a}`);
}

function shouldThrow(caseName, fn, expectedErrorCtor, expectedErrorMessage) {
    if (!caseName)
        throw new Error(`must specify test case name`);

    const expected = `${expectedErrorCtor.name}(${expectedErrorMessage})`;
    try {
        fn();
        throw new Error(`${caseName}: Expected to throw ${expected}, but succeeded`);
    } catch (e) {
        const actual = `${e.name}(${e.message})`;
        if (!(e instanceof expectedErrorCtor) || e.message !== expectedErrorMessage)
            throw new Error(`${caseName}: Expected ${expected} but got ${actual}`);
    }
}

const TEST_TARGET = [
    DataView,
];

for (const targetCtor of TEST_TARGET) {
    const name = targetCtor.name;

    const buffer = new ArrayBuffer(4096);
    const byteOffset = 2048;
    const actualByteLength = 2048 + 1;
    // https://tc39.es/ecma262/#sec-dataview-buffer-byteoffset-bytelength
    assert((byteOffset + actualByteLength ) > buffer.byteLength, `25.3.2.1, step 9-b's condition not satisfied`);

    const byteLength = {
        valueOf: function () {
            $.detachArrayBuffer(buffer);
            $.gc();
            return actualByteLength;
        }
    };

    // By the spec (April 24, 2026),
    // https://tc39.es/ecma262/#sec-dataview-buffer-byteoffset-bytelength defines:
    //
    //  1. The step 3 get _offset_ by ToIndex(byteOffset).
    //      - We fix that in bug 311903 if the byteOffset is weird and this steps detachs the buffer.
    //  2. Check IsDetachedBuffer(buffer), but ok.
    //  3. The step 9-a get _viewByteLength_ by ToIndex(byteLength).
    //      - This detach the buffer.
    //  4. The step 9-b check whether `offset + viewByteLength > bufferByteLength` and throw RangeError.
    shouldThrow(`${name}: should throw as the expected`, () => {
        new targetCtor(buffer, byteOffset, byteLength);
    }, RangeError, 'TODO: this should be update the JSC actual error message');
    sameValue(buffer.detached, true, `${label}: arrayBuffer is detached as expectedly`);

    // The detached ArrayBuffer.byteLength should be set to 0.
    //
    //  - https://tc39.es/ecma262/#sec-detacharraybuffer
    //  - https://tc39.es/ecma262/#sec-get-arraybuffer.prototype.bytelength
    sameValue(buffer.byteLength, 0, `${label}: arrayBuffer.byteLength is 0`);
}
```

---

The similar pattern of this does not applied to other %TypedArray%.
In https://tc39.es/ecma262/#sec-initializetypedarrayfromarraybuffer, which is invoked by the step 7-c of https://tc39.es/ecma262/#sec-typedarray:

1. The step 3 get _offset_ by ToIndex(byteOffset).
    - The weird `byteOffset ` detach the buffer here.
2. The step 7 check IsDetachedBuffer(buffer) and throw TypeError.
3. The step 9-b-2 checks whether `offset + newByteLength > bufferByteLength` and throw RangeError if it's true.

[Tetsuharu Ohzeki [UTC+9]]

- Firefox 152.0a1 https://hg-edge.mozilla.org/mozilla-central/rev/2a5398c41a2cbc8221cfebda2e46759afd1fa482
- Google Chrome 149.0.7810.2

works fine with the attached testcase (we can test it by replacing `$.detachArrayBuffer(buffer)` to `buffer.transfer()`).

[Tetsuharu Ohzeki [UTC+9]]

Pull request: https://github.com/WebKit/WebKit/pull/63535

[Radar WebKit Bug Importer]

<rdar://problem/176016346>

[EWS]

Committed 312964@main (bc5968157109): <https://commits.webkit.org/312964@main>

Reviewed commits have been landed. Closing PR #63535 and removing active labels.