RE: Wheel events

Why does a mouse-wheel event need to include information about scrolling? I thought we already have specified a scroll event for this purpose? It seems to me that the act of scrolling is merely related to the default action of wheel event (which, according to the wiki, specifies a default action of mousewheel (which itself could cause the default action of dispatching a scroll event).

After re-reading the proposal on the Wiki, I'm convinced that we're making the whole wheelevent much too complicated. It's a wheel. The user rolls it. If the event starts to make too many assumptions about what the intended purpose of the roll is (e.g., pixel scrolling, line scrolling, page scrolling, page navigations, etc.) then we've arbitrary limited the usefulness of the event as a platform feature. I think it makes sense to fire the event when the mousewheel rolls, report "how much", and leave it at that. If the user cancels the wheelevent, then any corresponding side-effects of rolling the mousewheel should also be cancelled. Regarding "how much": it's a wheel. We should pick either degrees or radians and report the degree of rotation--knowing that the precision of the value reported and in what time-increments may differ from user-agent to user-agent. Note, that such an approach effectively limits the usefulness of the actual value of rotation reported; however, in practical use, web applications merely are interested in knowing the direction of the rotation about the axis coordinates, not "how much the wheel was rotated" because multiple incremental rotations generally fire multiple events, and the actions of the web apps are generally driven by the number of events received, not necessarily by the "amount" of rotation, or differences in the amount of rotation (though I can imagine interesting uses for comparing the differences in values of amount of rotation).

About collapsing deltaX/Y/Z... one of the reasons we like the three vs. 1 is that together they specify a coordinate in 3-dimensional space; a vector that could be used in the future to cleanly move/pan in 3-dimensional space. If we went with .delta/.detail then the same application would suffer from the "jaggies" when responding to events sequentially.

-Travis


-----Original Message-----
From: public-webapps-request@w3.org [mailto:public-webapps-request@w3.org] On Behalf Of Olli Pettay
Sent: Wednesday, October 01, 2008 4:46 AM
To: public-webapps@w3.org
Subject: Re: Wheel events


Hi all,

So version 3 got implemented in Mozilla, but in a bit modified way.
There are separate events for line scrolling and pixel scrolling.
In Mozilla the old DOMMouseScroll and a new, currently named
MozMousePixelScroll, but in this message I call those just lineScroll
and pixelScroll:

- If there is only pixel scrolling, dispatch a pixelScroll event
   and scroll the current scrollable area unless .preventDefault() was
   called.

- If there is only line scrolling, dispatch a lineScroll event, and
   dispatch also a pixel scroll event if it is possible to convert line
   data to pixel data. Scroll the scrollable area if .preventDefault()
   wasn't called on either of those events.

- If there is both line scrolling and pixel scrolling, dispatch line
   scroll and pixelScroll and scroll the scrollable are unless
   preventDefault was called on either of those events.


This way applications which want to handle pixel scrolling, can always
get those events, and this doesn't break existing applications which
only want to handle line scrolling (Google maps etc.).
Dispatching only pixel scroll events wouldn't work because many
applications don't want to handle small scrolling and there isn't always
a good way to convert pixel scroll data to line scroll data.
So to help web apps, better to have separate events.

Note, currently in Mozilla there are separate wheel events for
horizontal and vertical scrolling and events have .axis property.

So
http://dev.w3.org/2006/webapi/DOM-Level-3-Events/html/DOM3-Events.html#events-Events-eventgroupings-wheelevents
could be modified a bit. The event name tells whether scrolling is line
scrolling or pixel scrolling and it also specifies what .delta
means. If we want separate events for different axis, we could remove
.deltaX/Y/Z and have just one .delta and use .detail for axis and have
constants X_AXIS, Y_AXIS,  Z_AXIS. This way we could later, if wanted,
add support for pageScrolling and whatever and also add support for any
new axis or other features.

To make this work with existing implementations which have support
for mousewheel event, the spec could say something like:
"A default action of user agent generated event objects of this type
causes implementations to dispatch a mousewheel event iff it supports
that event type and WheelEvent.type is "lineScroll", WheelEvent.delta is
non-zero and WheelEvent.detail is WheelEvent.Y_AXIS."

I'm not sure what could be the right names for the events.
It is perhaps too easy to confuse 'lineScroll' and 'pixelScroll' with
totally different kind of event 'scroll'.

Not so much related to the new WheelEvent:
Note, if mousewheel event must be defined in DOM 3 Events (as a legacy
event, not obligatory to implement), there is still need to define what
its .wheelDelta means, because that doesn't tell how many lines or how
many pixels to scroll.

-Olli


Olli Pettay wrote:
>
> Hi all,
>
> there has been some discussion about how wheel event should work, and
> what kind of information it should expose.
> The current DOM 3 Events (editor) draft has the legacy mousewheel event
> and also the new wheel event which has deltaX/Y/Z. The problem is
> that what delta means (especially in the new "wheel" event).
> Operating systems create events which may contain several kinds
> of information - how many lines to scroll but also how may pixels.
> Currently Safari seems to support also pixel scrolling, but doesn't
> dispatch events for events which don't have line scrolling >= 1.
> And that causes problems like
> http://mozilla.pettay.fi/moztests/pixelscrolling.mov
>
> Web applications should be able handle that situation and IMO
> should know what kind of wheel event just happened.
> So either deltaX/Y/Z (I'm talking about the new event here, not the
> legacy event) should be always in pixel units (or fragment of such)
> or the event should somehow have information about the units.
> In latter the case there is the problem that same system event may
> contain both pixel and line information. So either two DOM events
> should be dispatch or one event should be able to carry information
> about the different kinds of units.
>
> But if the event contains only pixel information, web app can't know
> how many pixels mean a "line-scroll". And that is valuable when trying
> to make web app to act like a native application.
>
> So there are (at least) 5 options
> 1. only pixel delta
> - Web app can't know what delta value means a "line scroll"
> 2. events with both line and pixel scroll information (maybe also page
> scroll etc?)
> - Web apps would get all the needed information in one event,
> but the event interface would probably look pretty ugly
> 3. separate events for pixel scrolls and line scrolls
> - Web app can't know whether user did one or two gestures
> 4. linked events; if the native event contains both line and pixel
> data, dispatch one event for line delta and one for pixel data
> but add some attribute to the event interface which links these
> events together, something like
> readonly attribute WheelEvent relatedEvent;
> or
> readonly attribute WheelEventList relatedEvents;
> 5. no pixel scrolling at all
> - causes http://mozilla.pettay.fi/moztests/pixelscrolling.mov
>
> I just came up with the idea 4, need to think that a bit more.
>
>
> -Olli
>
....
> Or 6.
> Perhaps the type of deltaX/Y/Z shouldn't be just long.
> It could an array of longs or an array of wheelrollvalues (whatever
> those are).

Received on Saturday, 6 December 2008 00:32:51 UTC