Re: Exposing constructors of readonly interfaces to web authors

On Jul 9, 2014, at 4:54 AM, Domenic Denicola <domenic@domenicdenicola.com> wrote:

> From: Rik Cabanier <cabanier@gmail.com>
> 
>> This confirms that the current behavior can be implemented using JavaScript.
>> Doesn't this satisfy Domenic's requirement?
> 
> It does, indeed, and gives us a path forward. However, in doing so, it points out that what the spec specifies is nowhere close to what was implemented. Here is a sketch of how we could update the specification to match what is (observably) going on:
> 
> - DOMRectReadOnly contains a [[type]] internal slot, which is either "quadbounds" or "rect"
> - The x getter on DOMRectReadOnly returns either [[x]] if [[type]] is "rect", or the minimum of q.p1.x, q.p2.x, q.p3.x, and q.p4.x if [[type]] is "quadbounds". Similarly for y, width, and height.
> - DOMRectReadOnly's constructor then needs to have two different behaviors (which could be based on arguments, or on e.g. the first parameter): one that sets the [[type]] slot to "quadbounds" and the [[quad]] slot to the given quad, and one that sets the [[type]] to "rect" and sets the [[x]], [[y]], [[width]], and [[height]] slots.
> - DOMRect derives from DOMRectReadOnly.
>  - It has a single constructor, (x, y, width, height), which calls super(x, y, width, height) (or super("rect", x, y, width, height), perhaps).
>  - It has its own getters and setters for manipulating [[x]], [[y]], [[width]], and [[height]].
>  - Thus DOMRects will always have [[type]] "rect", which means that you can successfully apply DOMRectReadOnly's getters to it, as you would expect for an inheritance situation.
> 
> I sketched this up, in a similar fashion to the previous one: https://gist.github.com/domenic/60871738e49a2b8c7a01
> 
> If this looks kind of weird, that's because it is. In general JS programmers would not write code this way. But, we are trying to meet many constraints here that JS programmers would not normally attempt, e.g. readonly views, reusing the same code for both live quad-bounds and one-time readonly snapshots, having a common prototype that can be monkeypatched by user code, and the like.
> 
> This seems to address Roc's concerns, though, and at least it is something that can be implemented (and introspected) in JavaScript.

I looked at your code example. The key for your proposal is the external WeakMap “internalSlots”. Given that you have such a WeakMap that can be modified by the caller, the constructor with the argument DOMQuad seems irrelevant. The spec text for “associated bounding box”[1] already states how to compute x, y, width, height. Given that, I think the main issue is just adding the same constructor from DOMRect to DOMRectReadOnly and let DOMRect call this constructor. We can argue about the exact wording, in general I agree with roc: A spec should describe the behavior and not create a dogma how to implement an interface. I am fine with a normative note that the internal [[x]], [[y]], [[width]] and [[height]] may be modified by the caller of the constructor and am happy to welcome your script as a second implementation of this part of the spec. 

Greetings,
Dirk

[1] http://dev.w3.org/fxtf/geometry/#associated-bounding-rectangle

> 
> All righty then, what do we think this time around?

Received on Thursday, 10 July 2014 06:20:57 UTC