[maemo-developers] Repositories mess: conclusions and actions
From: Igor Stoppa igor.stoppa at nokia.comDate: Tue Oct 30 09:29:16 EET 2007
- Previous message: Repositories mess: conclusions and actions
- Next message: Repositories mess: conclusions and actions
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Hi, please find my (very biased) comment below. On Mon, 2007-10-29 at 07:42 +0200, ext Quim Gil wrote: > Hi, > > > On Sat, 2007-10-27 at 22:41 +0200, ext Tim Teulings wrote: > > Hmm, you never would get my applications ;-) I'm pretty sure that I > > won't fulfill all the requirements. > > Maybe then rephrasing the sentence? This is not an official > certification process, it is just a tool to get developers aware about > the level of quality expected in extras. > > But. > > > For example being good at battery time is a tricky thing to test > > If this is true then we need to provide better documentation and tools > to test/debug. At least the big issues should be not difficult to detect > (you app is draining the battery in 24h even if running just in the > background). I guess any developer can get some help by getting the > packages in extras-testing and expressing the will to push them to > extras. > > If a developer is not concerned at all about power management and hasn't > got a clue about how his app is treating the battery, then his software > shouldn't get so fast in extras. Indeed. > > We have got already real cases of a popular app getting all kinds of > positive comments in one side, and then 'this battery sucks'-like > threads here and there. Until someone is able to connect one app A with > battery problems B. Developer A is then aware, gets feedback and > implement fixes. Doing this before hitting extras instead of after makes > a whole difference. I'd like to distinguish between 3 major areas: -kernel this _is_ painful as everybody here can testify, but we ship a SoC, therefore there is only this much of drivers to be developed and there is already a very active OMAP community, so I wouldn't consider maemo as the best place where to discuss OMAP related issues. There is also a linux-pm mailing list for generic power management related discussions. Also the cpufreq ml is a good place where to follow power-related discussions, although we are not very active there. -userspace - infrastructure with this i refer to all the stuff that provides the ecosystem to applications (and probably it is the core of what maemo is at the moment): we are talking about sw that is supposed to run (almost) all the time and can significantly affect the performance of the device. At the moment this sw is only about the ARM core in omap, so the usual 2 rules strictly apply: * make your code have low average CPU usage * try to hit 0% CPU usage as often as you can To elaborate it a bit, it's much better to have high activity bursts than spread the activity over longer time (yes, this seems to clash against our DVFS implementation, but it's better to let the system spread the load, than trying to be smart in every single application). What i have described above is indirect power management through CPU load profiling and is simpler to accomplish (no need to instrument your device with a current meter) and Eero has already provided on these lists eccelent examples of how to monitor the CPU activity. -userspace - applications this case is probably both the more common and at the same time the simpler, if we assume that the application will run for a limited amount of time, possibly as main application on the screen and should be easily associated to the battery running flat. The same techniques described at the previous point are effective in this case as well. I think that having developers clearly stating to which class their code belongs to would significantly improve the perception that the user has of the sw he is installing. I don't want to single out Canola, but it is just the first example that comes to my mind of something that is perceived as plain application, while it also has infrastructure components. To conclude, it shouldn't be too difficult to get power management working, as long as we are talking of developing applications. Unfortunately it is not perceived as a _required_ aspect of a quality application. As much as having a usable UI. OMAP3 will make the requirement even stronger; this is the right time to start changing mindset about power management. Doing CPU profiling is a good step down that road. -- Cheers, Igor Igor Stoppa <igor.stoppa at nokia.com> (Nokia Multimedia - CP - OSSO / Helsinki, Finland)
- Previous message: Repositories mess: conclusions and actions
- Next message: Repositories mess: conclusions and actions
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]