W3C home > Mailing lists > Public > www-dom@w3.org > January to March 1999

Re: comments on DOM2 Event Model

From: Tom Pixley <joki@netscape.com>
Date: Wed, 24 Feb 1999 16:32:41 -0800
Message-ID: <36D49A28.EC537843@netscape.com>
To: www-dom@w3.org
CC: Jean-Michel Leon <jmleon@Eng.Sun.COM>
Jean-Michel Leon wrote:
> 
> [ the following suggestions are based on WD-DOM-Level-2-19990222. ]
> 
> Event Flow
> ==========
> 
> this paragraph suggest that there is a top-down dispatch of events, but
> this dispatch mechanism is not specified. Plus, although it seems quite
> natural for UI events, because you can defined how theses events are
> dispatched down the DOM tree, using x,y coordinates, or knowing which Node
> has the focus, I'm not sure what this means for other kind of events
> (logical & mutation events). Maybe 'dispatching' should be defined only for
> x,y related events
> 
Downward propagation during capturing follows the tree structure down to
the event target just as bubbling follows the tree structure upward. 
Coordinates are not brought into play as the DOM is unware of such
things.

> I think Event Flow should describe both 'Dispatching' and 'Bubbling', and
> both of these should be detailed like bubbling is.
> 
> Capturing:
> =========
> 
> Once you've described how dispatching works, you realize that the capturing
> stuff is not well defined anymore. In fact (especially since this is
> currently using the same interfaces) 'capturing' simply means 'dispatching
> cancellation', just like you have 'bubbling cancellation'. I think this
> should be symmetrical: you should either have a reserved term for 'bubbling
> cancellation', or you should not talk about 'capturing'.
> 
> I also have one special grief against Event.cancelCapture, which is a poor
> name for specifying 'please do not propagate this event to my
> descendents'. I'd rather see something like 'cancelDispatch', which would
> be more consistent to 'cancelBubble'.
> 
> The bottom line is that I think the event flow should talk more about
> dispatching, and dispatching and bubbling should be more symmetrical.
> 
So let me sum up my comments to all of the above here.  I know I need to
clarify the text of capturing and bubbling.  This is on my to do list. 
But let me try to explain a bit here, anyway.

Capturing is intended to be the symmetrical opposite of bubbling.  The
use of the word dispatching may be being used in an improper context
within the spec but it has no special meaning.  Dispatching an event to
a given Node simply implies that the event listeners on that Node will
be called.  The term is equally applicable within the context of either
bubbling or capturing.  I believe, however, , most of your above
statements agree with out intent once you substitute 'capture' for
'dispatch'.

> Event Cancellation
> ==================
> 
> what's the rationale behind using a property for cancellation instead of a
> method ?
> 
I believe the initial rationale (based on carryover from IE4) was that,
during bubbling, higher Nodes in tree could query and/or override the
value set by lower Nodes.  

> this seems more a risk than a feature to me, because it forbids to specify
> cancellation policy when more than one listener are set. Typically, using
> the existing mechanism, you cannot predict what happens if one listener
> sets this attribute to true and the other sets it to false.
> 
That is true.  And if we were maintaining the old policy of a single
listener per Node I'd say that was fine.  But you have a point.  If we
allow multiple unordered listeners per Node (which is the current plan)
then the behavior is totally unpredictable.  If we were to allow
ordering (still on the table) then we might want to keep this, again.

> I can see two different options that would be much easier to specify:
> 
>     - use a method (cancelBubble() or cancel()), and specify that
>     bubbling is cancelled if at least one of the listeners calls this
>     method.
> 
>     - use handleEvent return value to specify that bubbling is cancelled
>     if at least one of the listeners return 'true'
> 
I'll raise these with the group.

> Listening vs. Capturing
> =======================
> 
> Right now, if I set a listener on a given Node, there is no way to know if
> the listener is triggered during the dispatch phase, or the bubbling phase
> (well, you implicitly know it, because you set the useCapture flag
> accordingly to what you want this listener to do - but nothing prevents you
> to try to use a listener designed for capture as a bubbling listener).
> 
> This also prevents you to set one Listener both as 'capturer' and
> 'listener', since the same method is going to be called in both contexts.
> 
> I think there should be a distinction somewhere for these different
> contexts. I can see two different options:
> 
>    - define different Listener types (you would add Filters for listening
>    to the dispatch phase, and you would add Listeners for listening to the
>    the bubbling phase).
> 
>     so EventTarget would be:
> 
>     interface EventTarget {
>         void addEventFilter(in DOMString type,
>                             in boolean postProcess,
>                             in EventFilter listener);
> 
>         void removeEventFilter(in DOMString type,
>                                in boolean postProcess,
>                                in EventFilter listener);
> 
>         void addEventListener(in DOMString type,
>                               in boolean postProcess,
>                               in EventListener listener);
> 
>         void removeEventListener(in DOMString type,
>                                  in boolean postProcess,
>                                  in EventListener listener);
>     };
> 
>    so we could have:
> 
>    interface EventListener {
>        void handleEvent(in Event event);
>    };
> 
>    interface EventFilter {
>        void filterEvent(in Event event);
>    };
> 
>     and Event could simply have a cancel() method, specified as 'cancelling
>     the event flow (both dispatching and bubbling) as soon as one listener
>     calls it.
> 
>     - have two methods on Listeners, one for capturing, and one for
>     listening.
> 
>    interface EventListener {
>        void handleEvent(in Event event);
>        void filterEvent(in Event event);
>    };
> 
>     But this would imply calling the filtering method many times when it's
>     not required.
> 
Another option would be to provide the user sufficient information to
know if they're being called in the capturing or bubbling phase.

> EventTarget.removeEventListener
> ===============================
> 
> the spec should say what happens if a listener removes itself, while it is
> handling an event.
> 
Okay.  Offhand I'd say it completes its handling.  The existence of the
event handler would only need to be checked when the event is trying to
find the next listener to be called.

> On the same topic, it should also say what happens to the Event Flow if you
> remove the dispatching node (or any ancestor) inside an event handler.
> 
Good question.  This has come up before and we don't have a good answer
yet.

> addEventListener / postProcess
> ==============================
> 
> I'd like to understand what's behind this. My feeling is that when you
> write a script, you are relying on an unknown DOM implementation. So how
> would you know if you should set postProcess to true or not ?
> 
Well postProcess isn't based too much on implementation.  In the case of
mutation events its easy.  If a Node is being added, I can react either
before or after the addition.  In the case of UI events, its less
clear.  And in some cases, invalid.  We've discussed that some event
might need to be handled as preProcessed only and that such would be
noted on the event description but this hasn't made it into the spec
yet.

> Actions
> =======
> 
> these objects are mentioned when describing. I think this is linked to my
> question about postProcessing: whether we have an idea what a DOM
> implementation is doing (what the default actions are), and the spec should
> mention where/when and how these actions take in the event flow, or actions
> are not well defined, and the spec shouldn't mention them. My feeling is
> that something's missing here, maybe it's just an entry in the terminology.
> 
No, unfortunately you are somewhat correct.  We do not plan on defing
the default actions and they will likely vary by implementation.  As far
as where they fit in terms of event flow, yes can be more clear.  I'll
try to make it so.  But the actions themselves are left to the
implementation.

> Mutating Events
> ===============
> 
> the spec should say if events can be mutated (their properties can be
> changed) by a filter or a handler, and what effect this has on the event
> flow. Typically, what happens if I, as an event filter, modify the X,V
> coordinates of an event ? Does this modifies subsequent dispatch or not ?
> 
We've had discussion both ways on this.  That's largely why its
undefined.

> Misc questions
> ==============
> 
> can a script generate an event from a given node ?
> 
That essentially deals with the same subject as changing the event
properties.  There has been discussion of allows creation and dispatch
of events.  This is one of the reasons the Event objects properties are
read/write.  However, its unknown whether this will be in the spec or
not.  If not, we could possibly make the Event object properties
read-only and solve the above question as well.

> jm.
Received on Wednesday, 24 February 1999 19:31:26 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Friday, 22 June 2012 06:13:46 GMT