Editorial: Fix realmless ArrayBuffer creation

[bakkot]

Fixes #1675, maybe.

@annevk in that issue also points out that the extract a body [...] algorithm also creates a Uint8Array in an ad-hoc way, which I've also tried to fix. Unfortunately I don't know what realm to use there - unlike the original issue, these aren't method steps. I am guessing that when the realm is unspecified (which it very often isn't - for example, earlier in the same algorithm there is "set stream to a new ReadableStream object" without a realm argument), it's implicitly the current realm? But maybe we need to thread a realm argument into this algorithm (and quite a few others), or possibly just explicitly pass "the current realm".

(If this right I'll follow up with a PR for #1732.)

---

Preview | Diff

[annevk]

This looks correct, but I wonder if we have test coverage for this.

It also needs some wrapping adjustment, but happy to push a fix for that myself before landing.

The longer term plan is to fix this more generally through Web IDL by making an ambient realm available to be used for creating (most) new objects.

[bakkot]

This looks correct, but I wonder if we have test coverage for this.

There are no tests for the realm of created objects at all, that I can find. I could add some but I'd be inventing the conventions.

The longer term plan is to fix this more generally through Web IDL by making an ambient realm available to be used for creating (most) new objects.

Presumably that would apply to the first hunk (the one in extract a body with type), but not the second (the one in arrayBuffer()), right? Since the ambient realm is not the one used in the second case. (Specifically, if you do await firstRealm.Response.prototype.arrayBuffer.call(new secondRealm.Response('asdf')), the resulting ArrayBuffer comes from the second realm, not the first, i.e. it is derived from the realm of this rather than the realm of the arrayBuffer function.)

[annevk]

Well, "ambient" was doing a lot of work there. The idea would be for Web IDL to handle these kind of cases correctly so nobody has to think about them again.

If you could add a minimal window.js test I think inventing the conventions is fine?

[bakkot]

web-platform-tests/wpt#46104