"TypeError: out-of-bound numeric index" on typed arrays in strict mode

[TooTallNate]

QuickJS has an unexpected behavior (at least when compared with other JS engines), where it throws when a typed array is attempting to be set at a index that is out of bounds:

'use strict';
const a = new Uint8Array(1);
console.log(a[0] = 1);
console.log(a[1] = 2);
console.log(a);

$ qjs o.mjs 
1
TypeError: out-of-bound numeric index
    at <anonymous> (o.mjs:4:20)

Compared to other runtimes:

$ node o.mjs 
1
2
Uint8Array(1) [ 1 ]

$ bun o.mjs 
1
2
Uint8Array(1) [ 1 ]

$ deno run o.mjs 
1
2
Uint8Array(1) [ 1 ]

I believe the relevant part of the spec is here: TypedArraySetElement

It's not super clear to be if this is technically incorrect behavior or not (that is, I see that the function may result in a throw completion, but I don't see where that would happen based on the steps described).

I also found this 10 year old mailing list thread which seems relevant (but never seemed to reach a conclusion): https://esdiscuss.org/topic/should-assigning-to-out-of-bounds-elements-of-a-typed-array-throw-in-strict-mode-code

In practice, this is making the __toBinary() function defined by esbuild throw this out of bounds error in certain cases,
For reference:

var __toBinary = /* @__PURE__ */ (() => {
  var table = new Uint8Array(128);
  for (var i = 0; i < 64; i++)
    table[i < 26 ? i + 65 : i < 52 ? i + 71 : i < 62 ? i - 4 : i * 4 - 205] = i;
  return (base64) => {
    var n = base64.length, bytes = new Uint8Array((n - (base64[n - 1] == "=") - (base64[n - 2] == "=")) * 3 / 4 | 0);
    for (var i2 = 0, j = 0; i2 < n; ) {
      var c0 = table[base64.charCodeAt(i2++)], c1 = table[base64.charCodeAt(i2++)];
      var c2 = table[base64.charCodeAt(i2++)], c3 = table[base64.charCodeAt(i2++)];
      bytes[j++] = c0 << 2 | c1 >> 4;
      bytes[j++] = c1 << 4 | c2 >> 2;
      bytes[j++] = c2 << 6 | c3;
    }
    return bytes;
  };
})();

My vote would be to remove this behavior from QuickJS and make the behavior match other JS engines.

[saghul]

I'll dive in the spec, but aligning with other engines would likely be good here.

[saghul]

PR: #647

It's interesting test262 doesn't seem to cover this. It does, however, cover the case of setting a numeric property on a TA where the backing array buffer is detached.

[saghul]

This is fixed as of #646

[saghul]

Sorry, my bad, strict mode!