Normative: iterator helpers close receiver on argument validation failure

[bakkot]

The general rule when working with iterators / iterables is that once you get an iterator it is your responsibility to close it when you're done with it, unless it errors out or completes. We're consistent about this when we get the iterator from an iterable, and in some of the places where we get the iterator directly (e.g. in Set.prototype.isDisjointFrom step 5.c.ii.1.a), but as pointed out by @zloirock, iterator helpers don't close their receiver when one of their argument is invalid. Since iterator helpers are conceptually "take ownership" of their receiver (either by consuming it fully, or by closing the receiver when the iterator helper itself is closed), and since an argument failing validation is not an error of the underlying iterator, helpers should close their receiver in this case to be consistent with the general rule above.

Unfortunately IfAbruptCloseIterator and IteratorClose both take an Iterator Record, not the iterator object itself, and we don't construct the Iterator Record until after argument validation. So we have to make a "partial" Iterator Record for use by these methods. (It's still a well-typed Iterator Record.) This is a spec fiction to simplify the spec text and doesn't need to correspond to anything in an actual implementation.

[michaelficarra]

I would inline these ThrowCompletions into the IteratorClose calls.

[syg]

lgtm

As a follow up, as we discussed, would weakly prefer a IteratorCloseDirect instead of making these administrative Iterator Records just so we can close them.

[bakkot]

would weakly prefer a IteratorCloseDirect instead of making these administrative Iterator Records just so we can close them.

Started doing this but it turns out we'd also need an IfAbruptCloseIteratorDirect, or some additional bookkeeping to avoid that, so I'm now less convinced this would be worth it.