[Gstreamer-openmax] pre-release-gst-android-0-10-22-01

[Felipe Contreras]

On Sun, Jun 21, 2009 at 4:23 PM, Prajnashi S<prajnashi at gmail.com> wrote:
>
> Hi, Felipe,
> I just went through your changes in
> http://github.com/felipec/gst-openmax/commits/android with following
> questions when rebase to 0.10.0.5
>
> 1) Why you want to remove "BUILD_WITH_ANDROID". Don't you want gstomx
> in android share same base with other platform?

Well, if we follow that approach then we'll have BUILD_WITH_ANDROID,
BUILD_WITH_TI, BUILD_WITH_BELLAGIO, BUILD_WITH_NOKIA, etc. The code
would quickly become a mess.

The promise of OpenMAX IL is that exactly the same code will work on
all omx implementations, but of course that's not the case.
Workarounds are needed, but they need to be minimized.

Have you noticed the 'omap' branch? Instead of having #ifdefs all over
the place all the workarounds are in one branch, so if you don't use
TI's omx implementation you are not bothered by all that noise.
Ideally TI should fix their implementation so that no workarounds are
needed, and I've been pushing them to do just that. For now if you
want to use TI's omx you need to use the 'omap' branch.

I was thinking the same should be done for Android, however I'm
considering other alternatives. If we follow a different path it won't
be BUILD_WITH_ANDROID, the change in behavior should be at run-time,
and the code for the workaround should be separate from the main code.

For now I think it's better to have an 'android' branch just like the
'omap' branch.

> 2) PV omx uses ms as timestamp, which does not compliant with OMX
> spec. I asked it in android group, but no one give me reply. The worse
> thing is, TI OMX in omapzoom follow it. So, in order to avoid further
> changes, I plan to change gstomx to use ms as well.

That's ok. Do as many changes as you need, however, please make each
workaround a separate commit and clearly specify why it's needed in
the commit message.

-- 
Felipe Contreras