Bug 281297 – JavaScriptCore produced non-deterministic results

• Status: RESOLVED
• Resolution: INVALID
• Component: JavaScriptCore

[congli]

Created attachment 472903 [details]
The file to trigger this bug

Running the following `a.js` in JavaScriptCore generates unstable results:

```js
const v0 = [0xAB0110, {}];
function f1(a2, a3) {
  if (a2 == "global") { v0[0] ^= a3;  return; }
  let v7 = v0[1][a2];
  if (v7 == undefined) { v7 = 0; }
  if (v7 < 50) { v0[0] ||= a3;  v0[1][a2] = v7 + 1; }
}
f1("global", 10569);
for (let i = 0; i < 921; i++) {
  try {
    f1("global", 14614);
    (1n)[1n];
    const o263 = { "maxByteLength": 536870912 };  // Reducing "maxByteLength" can mitigate this issue 
    const v265 = new ArrayBuffer(10, o263);
    f1("global", 1047);
    const v273 = new DataView(v265);
    f1("global", 6363);
  } catch(e278) {} finally {}
}
f1("global", 20614);
f1("global", 9736);
print(v0[0]);
```

Below is the results of 5 different runs:

```
[1] root@bb03d964e2f9:/home/fuzzer# ./jsc/jsc a.js
11233217

[2] root@bb03d964e2f9:/home/fuzzer# ./jsc/jsc a.js
11233217

[3] root@bb03d964e2f9:/home/fuzzer# ./jsc/jsc a.js
11238157      <----------- different

[4] root@bb03d964e2f9:/home/fuzzer# ./jsc/jsc a.js
11233217

[5] root@bb03d964e2f9:/home/fuzzer# ./jsc/jsc a.js
11238157      <----------- different
```

After some attempts, I found reducing "maxByteLength" of o263 or reducing the iteration account (921) of its surrounding loop can help mitigate the issue. If this is related to OOM, I think an explicit OOM exception/error should be printed; however, it prints nothing but the checksum value.

[Alexey Proskuryakov]

I can reproduce this on macOS, M3 MacBook Pro. There certainly is an OOM exception that's printed if I remove try-catch.

Could you please clarify what aspect of this you consider to be a bug? I am not super familiar with OOM handling in JS, but this doesn't strike me as obviously wrong.

[congli]

Hi Alexey, thanks for your response. I'm sorry for overlooking the surrounding try-catch. You are right; the OOM is indeed thrown and printed. So this should not be a bug. I marked it as RESOLVED/INVALID.