Bug 163083 – Overwriting an attribute event listener can lead to wrong event listener firing order

• Component: DOM
• Status: RESOLVED FIXED

[Chris Dumez]

Overwriting an attribute event listener can lead to wrong event listener firing order. This is because we remove the old event listener and then append the new one in this case instead of actually replacing the new one.

e.g.
element.onclick = function() { console.log('NOT RUN'); }
element.addEventListener('click', function() { console.log('SECOND'); });
element.onclick = function() { console.log('FIRST'); }

Prints FIRST then SECOND on Gecko, but prints SECOND then FIRST on WebKit.

[Chris Dumez]

Created attachment 290865 [details]
Patch

[Darin Adler]

Comment on attachment 290865 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=290865&action=review

> Source/WebCore/dom/EventListenerMap.cpp:120
> +    listeners->at(index)->markAsRemoved();
> +    listeners->at(index) = RegisteredEventListener::create(WTFMove(newListener), options);

Should we use a reference to avoid calling at() twice?

> Source/WebCore/dom/EventListenerMap.h:58
> +    void replace(const AtomicString& eventType, EventListener& oldListener, Ref<EventListener>&& newListener, const RegisteredEventListener::Options&);

This interface is intrinsically inefficient. Callers always have to call find before you call replace, so will always have to search the map twice!

In the future for better efficiency we should consider adding an operation that does exactly the right thing so that setAttributeEventListener can be efficient. This would be analogous to HashMap::add function that either adds if there is no existing item, or returns the location of the existing item if there is one.

[Chris Dumez]

Created attachment 290871 [details]
Patch

[WebKit Commit Bot]

Comment on attachment 290871 [details]
Patch

Clearing flags on attachment: 290871

Committed r206889: <http://trac.webkit.org/changeset/206889>

[WebKit Commit Bot]

All reviewed patches have been landed.  Closing bug.