ubuntu-phone team mailing list archive

[Thomas Voß <thomas.voss@xxxxxxxxxxxxx>]

On Tue, Oct 22, 2013 at 8:45 PM, Rasmus Eneman <Rasmus@xxxxxxxxx> wrote:
>>I do disagree with requiring the user to interact with the system to
>>ensure longer battery life. We should try as hard as possible to make
>>this automagically work.
>
>>For a tech-savvyy audience: Yes. For everyday users I do disagree. In
>>the end, users do blame the platform for bad battery life (for a good
>>reason) and we certainly don't want to have battery life preserving
>>apps in our app store (Google Play has got quite a few of them).
> While automagically is great when it works, it's a real hassle when it
> don't.
> I agree that users will be put of if there is a lot of battery drain,
> however
> users will also be put of if they can't use their phone as they would expect
> (Android have become huge and more and more users learn that they can
> do things in the background on their phone too).
>
> A golden path have to be found between features and battery. Personally
> I think Android does pretty good here but I get a feeling that you don't?
>

I think that there is room for improvement on top of the android model
(which I quite like).
The essence here is: If applications are split up in components
(whether you call them activities and services, or front- and
background services does not matter), it is way easier to come up with
a policy that both ensures good battery life _and_ allows for apps (or
components of them) to run in the background.

>>my point of view is still that forcing every little small app to bring its
>> own
>>daemon will:
>>
>>a) scare off people for writing apps for our platform as the communication
>>overhead between a service and the UI is a huge effort and easy to mess up.
> Android have AIDL (http://developer.android.com/guide/components/aidl.html)
> to help developers with this, I have used it and have to say that it's very
> simple
> to use. Hopefully something like this could come to Ubuntu too (it would
> make
> sense on the desktop as well).
>

AIDL is *another* middleware that helps in implementing an
out-of-process component model. That being said, we will need
something comparable. And before people start asking: I think we need
to help developers with a layer on top of "raw" dbus to make this as
convenient and easy as possible. We can probably hijack existing
object hierarchies, but might as well come up with something that is
less coupled to a specific object model.

One thing that has been spinning in my mind for some time is something like:

class MyBackgroundService : public ubuntu::Component
{
     public:
          // From ubuntu::Component
          void on_new_message(const Message& message) override;
          const std::string& name() const override { return
"MyCoolBackgroundService"; }
          bool store_state_for_migration(const Archive& archive) override;
};

ANNOUNCE_COMPONENT_WITH_NAME(MyBackgroundService, "MyCoolBackgroundService");

auto my_ex_process_service =
ubuntu::component_cast<MyBackgroundService>(ubuntu::Component::instantiate_out_of_process(MyBackgroundService));

if (auto)
{
     // We have the service running in a separate process now.
     // Invoke a function of the component
     auto result =
ubuntu::Component::invoke<MyBackgroundService::Sum>(my_ex_process_service).with(42).and(42);

     if (!result.is_error())
     {
           // Prints 84.
           std::cout << *result() << std::endl;
     }
}

Obviously, draft code so take it with a grain of salt. But we could
come up with something like this pretty rapidly.

>>b) be suicide in terms of battery usage. This will cause exactly the
>> opposite
>>than what it should. If everyone brings its own daemon we'll have tons of
>>services running all the time and the user can't even stop them because
>> there
>>is no UI to stop them.
> Again: Android.
> Android kills a service if the resources are needed, if the developer wants
> to
> create an app which have a responsible chance to stay alive even if the user
> does something else he have to create an ongoing notification. This makes
> to user aware that the app is running.
>
>

We can fine-tune resource constraints via cgroups and ensure that a
background service does not trash the system.
Unity8's application manager would be a great place to add this
functionality as it already takes care of suspend/resume and adjusting
oom_score's.

HTH,

  Thomas

> 2013/10/22 Jamie Strandboge <jamie@xxxxxxxxxxxxx>
>>
>> Pulling in Zoltan, Daniel and David for comment
>>
>> On 10/22/2013 05:40 AM, Thomas Voß wrote:
>> > On Tue, Oct 22, 2013 at 12:33 PM, John Lea <john.lea@xxxxxxxxxxxxx>
>> > wrote:
>> >> From a design point of view, the guidance we are currently following
>> >> is:
>> >>
>> >> - Only the app in the foreground when the phone is unlocked is
>> >> guaranteed to
>> >> be running.
>> >>
>> >> - In all cases where an app requires functionality that needs to run in
>> >> the
>> >> background and/or while the phone is locked, this functionality will
>> >> need to
>> >> be split off into a separate daemon that will be packaged and ship
>> >> together
>> >> with the foreground UI app.  The daemon will have no UI, and it's
>> >> functions
>> >> will be started and stopped using the foreground UI app.
>> >>
>> >> Applying this guidance to the use cases below:
>> >>
>> >>
>> >>
>> >> On 21/10/13 23:46, Jamie Strandboge wrote:
>> >>>
>> >>>   * a metronome app for musicians to practice to (2 are in the app
>> >>> store
>> >>> now)
>> >>
>> >>
>> >> The metronome app could be split into a UI app and daemon.  The UI app
>> >> would
>> >> start/stop the metronome and configure the sound/timing, and the daemon
>> >> would play the sound.  The daemon would continue playing the sound
>> >> irrespective of UI app state until the user switches the metronome off
>> >> via
>> >> the UI app.  This is useful in the case where a musician wants to set
>> >> the
>> >> metronome running, and then focus say a synthesiser app to practice on.
>> >>
>> >> Just forcing the Metronome app to run when the phone is locked and the
>> >> metronome app is in the foreground does not support this use case, the
>> >> 'UI
>> >> app' and 'deamon' approach seems a better fit.
>> >>
>> >>
>> >>>   * a white noise app to help people sleep (1 in the store)
>> >>
>> >>
>> >> Again I think splitting the White Noise app into a separate UI app and
>> >> a
>> >> daemon that plays the sound is advantageous.  This is because the user
>> >> might
>> >> for example also want to run a separate Sleep Pattern monitoring app at
>> >> the
>> >> same time.  Restricting the White Noise app to only running when it is
>> >> in
>> >> the foreground (irrespective of the phone lock state) precludes this
>> >> use
>> >> case.  btw, the Sleep Pattern monitoring app would also be broken into
>> >> a UI
>> >> app and a deamon.
>> >>
>> >>
>> >>>   * a navigation app that speaks the directions to you as you drive
>> >>> (none
>> >>> in the
>> >>>     app store AFAIK, but this would be a wonderful addition)
>> >>
>> >>
>> >> I think the UI app / deamon split is a better solution for the problem
>> >> as
>> >> well (again because the user will sometimes want to perform other tasks
>> >> while the navigation app continues to speak directions).
>> >>
>> >>
>> >>>   * internet radio apps (there are at least 2 in the store)
>> >>
>> >>
>> >> I think the UI app / deamon split works here?
>> >>
>> >>
>> >>>   * a 3rd party alarm clock (perhaps the API that the core app clock
>> >>> uses
>> >>> is
>> >>>     sufficient-- I haven't checked)
>> >>
>> >>
>> >> Same as above.
>> >>
>> >> It looks to me like the UI app / deamon split solves all the use cases
>> >> we
>> >> have currently thought of, are there any reasons why we should not
>> >> follow
>> >> this approach in all these cases?
>> >>
>> >
>> > Nope, I do fully agree with your examples here. And it is close to the
>> > aforementioned Activity/Service approach.
>> >
>>
>> It sounds like what we are saying is that we don't have to change anything
>> in
>> the system, we just need to communicate to developers how they should
>> program
>> within the system for these sorts of apps.
>>
>> Sounds like what is needed here is:
>>  * the SDK to treat compiled apps as first class citizens (I know this is
>>    planned, perhaps Zoltan can comment on its status)
>>  * documentation on developer.ubuntu.com that explains in developer terms
>> how
>>    power management and application lifecycle work, specifically for
>> HTML5,
>>    Cordova, pure QML and this new QML ui with background service. I did
>>    something similar for application confinement[1] which has been quite
>> helpful
>>  * provide an example QML ui with background daemon. We have a pure QML
>>    tutorial[2][3] for a currency converter. Perhaps a simple white noise
>> app
>>    that uses qtmultimedia in the backend service would be a good example
>>    (though a simple metronome might be better since the white noise app
>> could
>>    in theory use the media service). The docs[2] could call this 'QML with
>>    background service' (or something)
>>
>> If we did the above, app developers know where they stand and they can get
>> their
>> work done now while we build out various system services and SDK support
>> to
>> better support common use cases. Daniel and David, what do you think about
>> changes like this for http://developer.ubuntu.com? Does it make sense?
>>
>>
>> [1]http://developer.ubuntu.com/publish/apps/security-policy-for-click-packages/
>> [2]http://developer.ubuntu.com/apps/
>> [3]http://developer.ubuntu.com/apps/qml/tutorial/
>>
>> --
>> Jamie Strandboge                 http://www.ubuntu.com/
>>
>>
>> --
>> Mailing list: https://launchpad.net/~ubuntu-phone
>> Post to     : ubuntu-phone@xxxxxxxxxxxxxxxxxxx
>> Unsubscribe : https://launchpad.net/~ubuntu-phone
>> More help   : https://help.launchpad.net/ListHelp
>>
>
>
>
> --
> Rasmus Eneman