Re: [RFC] Health Thermometer Profile API

On Jun 22, 2011, at 7:20 AM, Santiago Carot wrote:

Hi,

2011/6/21 Elvis Pfutzenreuter [off-list ref]:

quoted

On Jun 21, 2011, at 12:56 PM, Santiago Carot wrote:

(cut)

quoted

quoted

Another option would be to guarantee that MeasureReceived() is called
right after agent registering, in order to give the client an
initial value to show (having in mind that notification interval might
be very long).

I would prefer to use the Final and Intermediate properties too. The
problem that we should address occurs when an application wich hasn't
got a watcher registered sets Intermediate or Final measurement
property and the last watcher is removed. I think that the best option
may be turn off measurments for power safe due that no watcher will
get the measures.

I'm not sure I follow. Enabling intermediate and/or final notifications
is done per-agent basis; if any of active agents needs intermediate
measurements, device is configured to send them. Disabling it on one
agent should not hurt another.

I mean the case when a process (it is not neccesary it has to be an
agent) is enabling intermediate and/or final notifications through
setProperty method in D-Bus API. Please, note that any process can do
this in current API because no extra control is done for this
operation.

May be I'm misundertanding what you want to said with per-agent basis
but as far as I see the only way to manage enabling/disabling
properties in this way is by adding a new parameter in setProperty
method with the agent path and checkicng if it is already registered
before enabling or disabling intermediate and final notifications. IMO
If we do this we would break the BlueZ APIs where no extra parameters
are required in such methods. I'm not sure if this would be a problem
but I don't feel comfortable with this solution.

On the other hands, if we discard adding the agent parameter in that
method (which I think that it fits better in BlueZ API), we can ignore
the case when there aren't any agents registered for the thermometer
service and a process enables some of this properties. In this case it
wouldn't be necessary to activate these properties since no one will
receive notifications for the measures.

True, I did not remember that those were properties. They would have
to go, otherwise it is too easy for an app to "break" another, by
disabling notifications.

A pair of Enable/DisableIntermediate methods taking an agent as
parameter would be more adequate.

IMHO there would not be a need of such
methods for final measure, because the final measurement (which is
the only one guaranteed to be precise) is the final objective, there
is no point in registering an agent and not wanting a final measurement.

So, inside the driver, final measurement indications are enabled if there
is at least one agent; and intermediate notifications are enabled if at
least one agent has requested it.

quoted

quoted

We can change the API to do that. What do you think?

quoted

quoted

+
+
+Health Thermometer Watcher hierarchy
+====================================
+Service              unique name
+Interface    org.bluez.ThermometerWatcher
+Object path  freely definable
+
+Methods              void MeasureReceived(dict measure)
+
+                     This callback gets called when a measure has been
+                     scanned in the thermometer. The Value entry corresponds
+                     to IEEE-11073 32-bit FLOAT.The Time entry in the dict
+                     will be only present if the device supports storing of
+                     data. The time value is expressed in seconds since epoch.

Using such a special type (FLOAT) looks funny. Not sure whether it is the best option
for this API.

At the very least, a description of FLOAT type should be added to the API, since
IEEE 11073 specs are not public/free, and API users would have to resort to unofficial
source to learn it.

I understand why you used this type (keeps the thermometer transcodable through this
API and there are the special values), so BTW the lack of device information service
access through this API should at least be mentioned, so we ammend that later (either
here or adding another API).

...or we can split the value in their exponent and mantissa
components. Any suggestion?

I'd go for that, but we need to take a look on special values to see if it makes sense.
It would be really ugly to say something like "if mantissa=5 and exponent=-3 then
temperature is actually zero" :)

We could provide special values in the API so that end users can know
when they are treating with those special cases. As far as I remember
the only special cases are NaN, NRes, +INIFINITY, - INFINITY and RFU.
Do you see anything else?

I think that these are all. Can't see a way to represent those special
values in a clean fashion. For the next RFC just include the mantissa/exponent
pairs related to them, and see if someone else complains.