Bug 298227 – TextTrackCue/VTTCue fails to dispatch enter/exit events (when no track is associated) so 4 WPT tests fail

• Status: RESOLVED
• Resolution: FIXED
• Component: Media

[Rob Smith]

Created attachment 476586 [details]
WPT TextTrackCue results showing 4 onenter/onexit failures

Steps to reproduce:

1. Run Web Platform Tests for TextTrackCue directly or view results at https://wpt.fyi/results/html/semantics/embedded-content/media-elements/interfaces/TextTrackCue?label=master&label=experimental&aligned&q=texttrackcue
2. Note that onenter.html and onexit.html tests fail on WebKit, but pass on Firefox.

Problem description:

Calling dispatchEvent() on a cue derived from TextTrackCue, such as VTTCue or DataCue, should invoke its associated event handler(s) synchronously. This behaviour should be inherited from EventTarget - see https://developer.mozilla.org/en-US/docs/Web/API/EventTarget/dispatchEvent - but the event handler is not invoked by WebKit which causes these four WPT tests to fail.

 * onenter results (2 tests): https://wpt.fyi/results/html/semantics/embedded-content/media-elements/interfaces/TextTrackCue/onenter.html?label=master&label=experimental&aligned&q=texttrackcue
 * onexit results (2 tests): https://wpt.fyi/results/html/semantics/embedded-content/media-elements/interfaces/TextTrackCue/onexit.html?label=master&label=experimental&aligned&q=texttrackcue
 
Additional comments:

1. These four WPT tests pass in Firefox.
2. Issue https://github.com/w3c/webvtt/issues/519 addresses other properties inherited by VTTCue from TextTrackCue and these two issues may be related.
3. A proposal has been raised in WICG DataCue to expose TextTrackCue's constructor that would allow removal of the dependency on VTTCue from TextTrackCue WPT tests which could help to resolve these and future issues: https://github.com/WICG/datacue/blob/main/text-track-cue-constructor.md

[Karl Dubost]

IDL in Safari
The current code has a couple of FIXME 
https://searchfox.org/wubkat/rev/487a3721e3ff1de52e9baa5a4ba9190889bb1204/Source/WebCore/html/track/TextTrackCue.idl

[
    Conditional=VIDEO,
    CustomIsReachable,
    CustomToJSObject,
    JSCustomMarkFunction,
    SkipVTableValidation,
    LegacyFactoryFunctionEnabledBySetting=GenericCueAPIEnabled,
    Exposed=Window
] interface TextTrackCue : EventTarget {
    readonly attribute TextTrack? track;

    attribute [AtomString] DOMString id;
    attribute double startTime;
    // FIXME: Add 'unrestricted' to 'endTime' to align with the specification. See: https://bugs.webkit.org/show_bug.cgi?id=260765
    attribute double endTime;
    attribute boolean pauseOnExit;

    attribute EventHandler onenter;
    attribute EventHandler onexit;

    // FIXME: Add 'unrestricted' to 'endTime' to align with the specification. See: https://bugs.webkit.org/show_bug.cgi?id=260765
    [CallWith=CurrentDocument] constructor(double startTime, double endTime, DocumentFragment cueNode);
    [EnabledBySetting=GenericCueAPIEnabled] DocumentFragment getCueAsHTML();
};


IDL in Firefox.
https://searchfox.org/firefox-main/rev/8a50d94240fb8f312d8b061a425ff0755bcbaac4/dom/webidl/TextTrackCue.webidl

[Exposed=Window]
interface TextTrackCue : EventTarget {
  readonly attribute TextTrack? track;

  attribute DOMString id;
  attribute double startTime;
  attribute double endTime;
  attribute boolean pauseOnExit;

  attribute EventHandler onenter;
  attribute EventHandler onexit;
};


Spec IDL is defined as

[Exposed=Window]
interface TextTrackCue : EventTarget {
  readonly attribute TextTrack? track;

  attribute DOMString id;
  attribute double startTime;
  attribute unrestricted double endTime;
  attribute boolean pauseOnExit;

  attribute EventHandler onenter;
  attribute EventHandler onexit;
};


The first test being failed in 
https://wpt.live/html/semantics/embedded-content/media-elements/interfaces/TextTrackCue/onenter.html


Fail	
assert_true(false, "dispatching event")
 /html/semantics/embedded-content/media-elements/interfaces/TextTrackCue/onenter.html:20:16


test(function(){
    assert_equals(c1.onenter, null, 'initial value');
    c1.onenter = undefined;
    assert_equals(c1.onenter, null, 'assigning undefined');
    c1.onenter = cb;
    assert_equals(c1.onenter, cb, 'assigning onenter');

// FAILING HERE
    c1.dispatchEvent(ev);
    assert_true(ran, 'dispatching event');

// NEXT TEXT
    c1.onenter = null;
    assert_equals(c1.onenter, null, 'assigning null');
    ran = false;
    c1.dispatchEvent(ev);
    assert_false(ran, 'dispatching event after nulling onenter');
});


https://searchfox.org/wubkat/rev/487a3721e3ff1de52e9baa5a4ba9190889bb1204/Source/WebCore/html/track/TextTrackCue.cpp#338-345
```cpp
void TextTrackCue::dispatchEvent(Event& event)
{
    // When a TextTrack's mode is disabled: no cues are active, no events fired.
    if (!track() || track()->mode() == TextTrack::Mode::Disabled)
        return;

    EventTarget::dispatchEvent(event);
}
```

[Karl Dubost]

in the dispatchEvent, there is a dependency on track being defined and not disabled.
Is the WPT test correct?

The setup function doesn't have an instantiation of a track.


setup(function(){
    window.c1 = new VTTCue(0, 1, 'text1');
    window.ev = new Event('enter');
    window.ran = false;
    window.cb = function() { ran = true; };
});

Would the test PASS if it was initializing a TextTrack first. 

c1.track returns null

The specs says:
cue.track
Returns the TextTrack object to which this text track cue belongs, if any, or null otherwise.


and 

The following are the event handlers (and their corresponding event handler event types) that must be supported, as event handler IDL attributes, by all objects implementing the TextTrackCue interface:

onenter
onexit

So does it make sense to work without a text track in Firefox?

[Rob Smith]

(In reply to Karl Dubost from comment #2)
> in the dispatchEvent, there is a dependency on track being defined and not
> disabled.
> Is the WPT test correct?
The short answer is: yes.

The longer answer is: TextTrackCue inherits dispatchEvent() from EventTarget according to https://html.spec.whatwg.org/multipage/media.html#texttrackcue:

interface TextTrackCue : EventTarget {
  readonly attribute TextTrack? track;

  attribute DOMString id;
  attribute double startTime;
  attribute unrestricted double endTime;
  attribute boolean pauseOnExit;

  attribute EventHandler onenter;
  attribute EventHandler onexit;
};

HTML does not specify any override of dispatchEvent functionality for TextTrackCue, so this function should have no dependency on any attribute of TextTrackCue and the WPT test is correct.

The code you've identified in WebKit's TextTrackCue::dispatchEvent implementation is not compliant with HTML: https://searchfox.org/wubkat/rev/487a3721e3ff1de52e9baa5a4ba9190889bb1204/Source/WebCore/html/track/TextTrackCue.cpp#340-342

> So does it make sense to work without a text track in Firefox?
Yes. HTML imposes no special conditions on how TextTrackCue events are dispatched - which could be either via TextTrack or by a direct call to dispatchEvent. TextTrackCue acts as the container object for timed metadata cues which may not require any real time scheduling at all, such as a video metadata search use case: https://www.w3.org/TR/webvmt/#area-monitoring

Hope this helps.

[Rob Smith]

https://searchfox.org/wubkat/rev/487a3721e3ff1de52e9baa5a4ba9190889bb1204/Source/WebCore/html/track/TextTrackCue.cpp#338

Renaming WebKit's TextTrackCue::dispatchEvent function would retain its existing functionality, avoid overriding EventTarget::dispatchEvent to comply with the HTML specification and allow the WPT tests to pass with no loss of capability.

Hope this helps.

[Radar WebKit Bug Importer]

<rdar://problem/160195643>

[Karl Dubost]

Pull request: https://github.com/WebKit/WebKit/pull/50526

[EWS]

Committed 301836@main (8956674a34a3): <https://commits.webkit.org/301836@main>

Reviewed commits have been landed. Closing PR #50526 and removing active labels.