Make it explicit that "this" is available in parallel

[annevk]

What is the issue with the HTML Standard?

Various other operations (such as method steps) can go in parallel at some point. In various places we rely on "this" being available (and thus kept alive). We should make this assumption explicit in the in parallel section.

[domfarolino]

Yikes! Do we want this to be able to be relied on in parallel? Whenever I catch this being used in parallel, I find that this blurs the line between Window event loop and parallel contexts to the point of adding more confusion—people think they can do everything else that they can from a Window event loop, like resolve promises, or grab state that actually has to be synchronized correctly without thinking through it. I've also found that folks consider the lifetime of global-bound objects from parallel contexts even less than normal, which I think is bad, since when you go in parallel the lifetimes of things are now unbound.

I usually encourage people to make this explicit by closing over what they actually need from this (sometimes grabbing that object itself if necessary) and explicitly checking in on or its global before using it later. I'd love for us to converge on that model for global event loop 🤝 parallelism.

[annevk]

I mean that's what I did, but @domenic suggested I do the opposite.

I would personally also prefer that eventually we get more explicit about what kind of state can be shared.

[domfarolino]

Ah, that surprises me. I don't think we should promote using this in parallel, but if @domenic has a compelling rationale I am all ears.

[domenic]

I don't feel too strongly. My rationale is that this cannot change what it points to, so saving it in a variable doesn't feel very useful.

But, your point that people should not generally access Web IDL objects in parallel is a good one. Similarly, your point about how bringing this in might cause people to start thinking other stuff is OK is compelling.

My main remaining complaint is that I find the pattern

1. Let global be this's relevant global object.
2. In parallel:
   1. ...
   2. Queue a global task on X task source using global to...

to be a bit annoyingly-verbose compared to

1. In parallel:
   1. ...
   2. Queue a global task on X task source using this's relevant global object to...

but I guess the better solution to that is to work on whatwg/webidl#135 (and in particular the sub-thread around in parallel behavior. (Where I tend to agree more with @jyasskin these days than I do with my 2017 self. I guess my 2021 proposal at whatwg/webidl#135 (comment) is the best current idea.)

So, I'm OK with saying that this must not be accessed in parallel.