191242 - Event propagation quirky with changed EventTarget nodes

Status: RESOLVED FIXED

(Core :: DOM: Events, defect)

Description

[Alex Vincent [:WeirdAl]]

I've noticed a rather interesting behavior in Mozilla, one which the DOM 2
Events Recommendation and the DOM 3 Events Last Call Working Draft from W3C do
not cover.

Basically, when I replace the actual event target (or likely any ancestor node
of the actual target) with another node, the DOM specs are not clear on correct
behavior.

I'm attaching two testcases, an XUL testcase and an XHTML testcase.  The XUL
testcase is more descriptive, but the XHTML testcase will render in non-Gecko
based browsers as well.

I believe there is at least one bug in the behaviors of the browser here; I'm
just not sure what exactly is "wrong" and what is "correct".

Comment 1

[Alex Vincent [:WeirdAl]]

Attachment: XUL testcase

Comment 2

[Alex Vincent [:WeirdAl]]

Attachment: XHTML testcase

Comment 3

[Alex Vincent [:WeirdAl]]

jst:  thought you'd like to see this, as a member of the DOM WG.

Comment 4

[Alex Vincent [:WeirdAl]]

Running the XUL testcase, the third step, it shows the event propagating through
foo_menu, not bar_menu.  This is clearly a bug.

Comment 5

[Johnny Stenback (:jst)]

Alex, could you send a request to the DOM WG ([e-mail]) to clarify the
intended behavior in the DOM Events specification?

Comment 6

[Alex Vincent [:WeirdAl]]

http://lists.w3.org/Archives/Public/www-dom/2003JanMar/0019.html

Comment 7

[Alex Vincent [:WeirdAl]]

http://lists.w3.org/Archives/Public/www-dom/2003JanMar/0022.html

The DOM WG's response.

I think Mr. Hegaret is telling us that for the XUL testcase's second part, the
Event object should still go to the target node.

Comment 8

[Olli Pettay [:smaug][[e-mail]]]

This is how KHTML handles event propagation:
http://lxr.kde.org/source/kdelibs/khtml/xml/dom_nodeimpl.cpp#L416
But is that a bit slow way to do it, creating a node chain for each event...

Comment 9

[Olli Pettay [:smaug][[e-mail]]]

Attachment: Create event target chain in stack

Any comments on this one?
The idea is to create an event target chain in stack before starting to handle
the event.
This would be quite big change to event handling, but I think something like
this is needed. And hopefully it wouldn't slow down the event handling.

Comment 10

[Boris Zbarsky [:bzbarsky]]

Yes, something like this is absolutely needed; there is no other way to
implement the spec.  Yes, this would be a major change to event handling (and
need changes to all the HandleDOMEvent mathods, if nothing else).  See bug
234455.  Any such code should go there, probably.

Comment 11

[Olli Pettay [:smaug][[e-mail]]]

fixed in bug 234455