Types: add new type array-like objects

[arthurvr]

Fixes #708.

[markelog]

We could a more strict here, not only length property should be in such object, but also their other properties names (but not necessary all properties) should be numeric too and they should corresponds with length.

Personally, i like Axel Rauschmayer definition - http://www.2ality.com/2013/05/quirk-array-like-objects.html

[dmethvin]

We don't enforce those extra rules though. A few days ago I read that ES6 defined "arraylike" as having a non-negative length value, which would be easy to enforce. However, I can't find any reference to that in the spec.

[markelog]

We can't iterate on non-numeric properties, whereas merge docs says:

Merge the contents of two arrays together into the first array.

if there would be array-like there, it should, according to this definition of array-like do this:

var a = { test: 1 };
a.length = 1;

console.log( $.merge( [], a ).test ); // 1

since "test" property is also "content" of the array-like object.

[gibson042]

Our de facto definition, which I observe agrees very well with Axel Rauschmayer's, is a nonnegative numeric length and a property corresponding to the highest index: https://github.com/jquery/jquery/blob/2380028ec4a6a77401b867a51de26a3cb8e8d311/src/core.js#L434-L448

[dmethvin]

Oh, I remembered it backwards. Our definition checks for non-negative but I couldn't find that in the spec.

[arthurvr]

So, what's the consensus in here?

Our de facto definition, which I observe agrees very well with Axel Rauschmayer's, is a nonnegative numeric length and a property corresponding to the highest index:

Using exactly that definition would suggest something like this works, isn't it?

var obj = { test: 3, 1: 4 };
obj.length = 2;

console.log( isArraylike(obj) ); // true
console.log( $.merge( [], obj ).test ); // undefined

[dmethvin]

Yes I think by quoting @gibson042 you can't go wrong. 😄

[gibson042]

I'd like this to be something like "an Object that contains a nonnegative integer length property and index properties from 0 up to length - 1". It describes the density that we assume (but are unwilling to fully verify at this point), and hopefully is clear enough to unambiguously include { length: 0 }.

Yes I think by quoting @gibson042 you can't go wrong.

I may hold you to that 😉

[markelog]

And no one notice the guy who initially point that out :-)

[arthurvr]

That suggestion seems like reasonable @gibson042! Thanks y'all! I'll update the PR.

[arthurvr]

Well, what do we think about this version?

[gibson042]

👍

Also, ❤️ @markelog for identifying both the problem and the solution. 😄

[dmethvin]

Just coming in to say that I ❤️ @markelog too. Oh and the PR looks good too 😸

[arthurvr]

❤️ 💓 @markelog 😻 💟