Sergio Benitez

> > Can you tell me which parts you're referring to? I believe the only potential confusion is the existence of both Value::from_any_seq_object() and Value::from_seq_object(). I think this is unfortunate,...

If you're okay with completely redesigning the `Object` system, which this PR assumed to not be the case, then may I propose an single, unifying `Object` trait as follows? ```rust...

> I do think though that it's important for an object to know what shape it has for other operations before interacting with it. So the ValueKind of sorts should...

> That is true in some sense, but some aspects of the engine really depend on this. For instance `|tojson` will serialize into a very different object depending on if...

> While it's probably true that `Enumerate::Range` is likely to be a sequence and not a map, I think it's odd that you need to try to iterate over it...

> More specifically to iterate one needs to hold on to the index and the object somewhere. I suppose you might end up with a helper like this? Exactly. I...

@Aaron1011 > Note that `Span::parent` allows you to write 'completely unhygenic' macros, which can see identifiers in a way that would otherwise be impossible: Not to advocate in either direction,...

> Relatedly, while `triple` gets you a `(Version, Channel, Date)`, there's not much you can do with those values, as the exposed API will `read` fresh ones anyway when you...

The one exception is the `supports_feature()` function, which indeed has no cached equivalent. That's because it's designed for special cases where you want to support a bootstrapped compiler - this...

I agree that we should include more verbiage with respect to how this crate can be used in a less error-prone fashion. On the broader subject, I believe there are...