Rewrite event handlers

[TimothyGu]

Tests: web-platform-tests/wpt#12201

Fixes: #3836

---

/infrastructure.html ( diff )
/obsolete.html ( diff )
/sections.html ( diff )
/webappapis.html ( diff )

[TimothyGu]

This is now ready to be reviewed.

[annevk]

This looks really great. I hope to have time next week for a more detailed review. What would help with that is a small write-up (that we can eventually use in the commit message) of what is being changed and why. As this seemingly changes a bit more than just the lack of removal of event listeners added through event handlers.

[annevk]

body should be in <code> here, no?

[TimothyGu]

It applies to frameset elements too, and I was trying to allude to the body element.

[annevk]

s/, and if/ and the/

(you could keep the additional comma I suppose, but I usually try to omit it if there's only two items)

[annevk]

It's a little surprising this step comes this late. This is only applicable to body and frameset elements?

[TimothyGu]

Yes that's correct. The key here is that we want to make sure the event handler name check is before "active document" check.

This is tested in https://github.com/web-platform-tests/wpt/blob/7cdedf945221ce000c977008769a09b034e87716/html/webappapis/scripting/events/event-handler-attributes-windowless-body.html.

[TimothyGu]

What would help with that is a small write-up (that we can eventually use in the commit message) of what is being changed and why.

Of course.

• Move the "When an event handler H of an EventTarget object T is first set to a non-null value" steps into a dedicated activate an event handler algorithm. Create a corresponding deactivate an event handler algorithm, called in the circumstances outlined in Event handlers-created event listener need to be removed sometimes #3836, thereby fixing that issue.

• Define "When an event handler content attribute is set" and "When an event handler content attribute is removed" steps rigorously using the attribute change steps hook in DOM. Care is taken to clarify certain edge cases, such as calling removeAttribute() on a nonexistent event handler content attribute but with a corresponding event handler IDL attribute defined, and calling setAttribute() an attribute with the same value as it currently possesses but with the event handler deactivated (exhibit B in Event handlers-created event listener need to be removed sometimes #3836).

• Formalize multiple definitions:

  • event handler as a struct containing a value and an event listener
  • location of event handlers and the concept of event handler target
  • rigorous set of steps on how to obtain the target of an event handler through determine the target of an event handler algorithm. The latter algorithm supersedes the handwavy and inconsistently applied "If an event handler IDL attribute exposes an event handler of an object that doesn't exist" steps
  • storage of event handlers in the event handler map

• More descriptive names for H (eventHandler), T (eventTarget</var), and E (event) in various algorithms

[domenic]

Pushed some minor tweaks, but LGTM. Would love @annevk's review too.

I made a half-hearted attempt at sectioning out this mega-section into subsections, but it didn't work. I'll file a follow-up issue explaining and brainstorming for the future.

[domenic]

Oh, shoot, I forgot. I don't think this works after all:

If eventTarget's node document is not an active document, then return null.

In particular this will not bail out for const eventTarget = document.createElement("body") inside an active document.

I think this might be correct:

If eventTarget is not the body element of its node document.

[TimothyGu]

@domenic Unfortunately, what you described doesn't seem to be what browsers implement:

const b = document.createElement("body");
const f = () => {};
b.onblur = f;
console.assert(b.onblur === f);
console.assert(window.onblur === f);

On Chrome, Safari, and Firefox this code executes fine.

I guess I should add another test case then.

[domenic]

That seems quite broken, but if everyone agrees, then it's probably not worth fixing. Amazing...

Edit: we may even want to add a spec note/example about this unusual fact.