Disposition of comments for the Device and Sensors Working Group

Hi,

I would kindly suggest you reorder the requirement for queueing with
the requirement for firing. E.g. "The user agent must queue a task to
fire a device proximity event." (and then remove the requirement for
queueing from the firing steps). This matches phraseology in other
specifications better and allows for reuse of the firing definition in
other contexts.

You defined "ondeviceproximity" and "onuserproximity" twice. This
seems to be because of the legacy-DOM-style formatting.

Do not use RFC 2119 terms in notes, such as "may".

If your device is not doing anything, e.g. completely stationary,
these sensors would theoretically not change and you would never be
able to get the actual state. That might be a mostly academic problem,
but this seems like another set of events that violate the spirit of
DOM Events. Kinda ambivalent on whether that's good or bad, but I
think it at least ought to be pointed out more clearly. And perhaps we
ought to define this more explicitly by having some kind of hook in
addEventListener?

Kind regards,

I just noticed your specification has the same issue as the Ambient
thingie. You're not defining default values for the event interface
members. (new UserProximityEvent("haha")).near is undefined. I suggest
adding that to the testsuite. (It's one of the basic things to cover
when testing event constructors.)

On Fri, Dec 7, 2012 at 5:39 AM, Anne van Kesteren <annevk@annevk.nl> wrote:

>
> If your device is not doing anything, e.g. completely stationary,
> these sensors would theoretically not change and you would never be
> able to get the actual state. That might be a mostly academic problem,
> but this seems like another set of events that violate the spirit of
> DOM Events. Kinda ambivalent on whether that's good or bad, but I
> think it at least ought to be pointed out more clearly. And perhaps we
> ought to define this more explicitly by having some kind of hook in
> addEventListener?
>


I've been thinking about this since the last call yesterday as well.

Early in my development of the Johnny-Five framework, I decided that all
sensors must:

1. Generate events

- sensor.on("change", .... ); // Any change to the reading value(s)
- sensor.on("read", .... ); // Every sensor reading

2. Allow value read accessors

- sensor.value; // sensors w/ a single value
- sensor.axis.{ x, y, z } // multiple values (ie. accelerometer, as
shown here)

3. Stream (in progress)

- http://maxogden.com/node-streams.html


But of course, this design is meant to facilitate an arbitrary number of
sensors, eg. you might have an array of 4 proximity or sonar sensors on a
robot to aid in navigation. This level of scalability may not
seem practical for mobile devices, but then I'm reminded of
navigator.getUserMedia that was originally spec'ed with no thought about
multiple cameras—implementations then have to figure it out.


I understand there is platform precedence to consider, ie. DeviceMotion and
DeviceOrientation APIs and this nicely matches, but If I was asked to
start from scratch, I'd start with a Proximity (or
DeviceProximity...something similar) constructor that initialized proximity
objects that had properties: max, min and value and could dispatch events
by inheriting EventTarget. I have no expectation that the current system
would be changed like this, so consider this entirely "just for fun". :)


Rick

I see the distance measurement unit "in centimeters" is mentioned only
once, in the description of "current device proximity", I would suggest
being explicit and including it in "minimum proximity distance" and
"maximum proximity distance" as well.

Rick