Clarify that "this" is not available while in parallel #11129
Conversation
| This seems reasonable, but we should fix at least the case I introduced as part of this I think. And maybe also the cases @domenic knows about. People typically copypasta their way into a specification after all. |
| I scanned all references of "this's relevant" in the spec and fixed any that were operating incorrectly from parallel steps. That resulted in the offscreen canvas fix, the #parallelism editorial "example" fix, as well as one more instance I could find, in |
domenic left a comment
domenic left a comment There was a problem hiding this comment.
Nice work! A couple minor suggestions, but I'm happy to see this cleaned up.
source Outdated
| data-x="global object">global</span>, or <span>environment settings object</span>. By extension, | ||
| you cannot access Web IDL's <span>this</span> value from steps running <span>in parallel</span>, | ||
| even if those steps were activated by an algorithm that <em>does</em> have access to the | ||
| <span>this</span> value. (Stated in more familiar terms, you must not directly access main-thread |
There was a problem hiding this comment.
I think the new sentence would probably be best as a new paragraph. The parenthetical and the sentence about data races seem a bit awkward after the new interruption.
There was a problem hiding this comment.
Sounds good. I guess one interesting thing is that this paragraph says we cannot use "global" objects while in parallel, while we clearly agree that we actually make an exception for globals, since that's the only way we can schedule things back on the main thread. In other words, we are assuming #6475 (comment) to be true—that most accesses to a global, especially when scheduling tasks, are atomic. That's the right assumption make IMO. I'm just trying to think if we should state this assumption/exception clearly in this paragraph too.
There was a problem hiding this comment.
I think there's a case to be made for porting (some much shorter version of) w3c/ServiceWorker#1755 (comment) into the spec, to acknowledge the messy reality and carve out some clearer exceptions. But that's a more ambitious followup, I think.
This fixes #11110, by clarifying that Web IDL's this value is not available from steps running in parallel, even if those steps were kicked off my a non-parallel algorithm that can access "this".
This PR clarifies this in the parallelism example, prose about how to operate from while parallel, and fixes all examples of "this's relevant foo" being used improperly from parallel steps, which is a good start to setting the right precedent.
/canvas.html ( diff )
/infrastructure.html ( diff )
/webappapis.html ( diff )
/worklets.html ( diff )