Re: [poky] [PATCH 3/3] meta-yocto: add pieces removed from oe-core for beagleboard & atom-pc

[Richard Purdie <richard.purdie@linuxfoundation.org>]

On Thu, 2011-05-05 at 14:31 +0200, Koen Kooi wrote:
> Op 5 mei 2011, om 14:21 heeft Richard Purdie het volgende geschreven:
> 
> > On Thu, 2011-05-05 at 13:46 +0200, Koen Kooi wrote:
> >> Op 5 mei 2011, om 13:38 heeft Richard Purdie het volgende geschreven:
> >> 
> >>> Hi Paul,
> >>> 
> >>> Great work in doing this, thanks. I was just looking at it with a view
> >>> to making machine support cleaner and I think there are still things we
> >>> can likely to do help with this. As one example:
> >>> 
> >>> On Wed, 2011-05-04 at 15:51 +0100, Paul Eggleton wrote:
> >>>> +++ b/meta-yocto/recipes-qt/qt4/qt4-x11-free_4.6.3.bbappend
> >>>> @@ -0,0 +1,2 @@
> >>>> +QT_GLFLAGS_atom-pc = "-opengl"
> >>>> +
> >>>> diff --git a/meta-yocto/recipes-qt/qt4/qt4-x11-free_4.7.2.bbappend b/meta-yocto/recipes-qt/qt4/qt4-x11-free_4.7.2.bbappend
> >>>> new file mode 100644
> >>>> index 0000000..076ade2
> >>>> --- /dev/null
> >>>> +++ b/meta-yocto/recipes-qt/qt4/qt4-x11-free_4.7.2.bbappend
> >>>> @@ -0,0 +1,2 @@
> >>>> +QT_GLFLAGS_atom-pc = "-opengl"
> >>> 
> >>> could we set QT_GLFLAGS in the machine.conf file instead of using a
> >>> bbappend?
> >> 
> >> Putting such USEFLAGS in machines sounds like a bad idea. In this case
> >> enabling it globally and falling back to mesa sw rendering at runtime
> >> is a better idea. The GL flag only enables extra API and libs, so it's
> >> good to have.
> > 
> > Agreed, longer term I think this is going to be the better way to handle
> > this. In this day and age, defaulting to sw rendering is probably the
> > sane thing to do.
> 
> Something like SOC_FAMILY would help here.

In some cases.

> > Equally, moving this from a .bbappend to the machine file is a bit
> > cleaner too though :)
> 
> Is it? Do you really want the machine to know about all the knobs in
> all the different layers? The .bbappends clearly signals a change to
> the recipe, hiding it in the machine.conf will just confuse people.
> And when changing options you need to edit the machine and then use
> PR_INC in a different file. Using a bbappend limits that to a single
> file

I'm not saying this makes sense for every bbappend option. For something
like QT and which is a part of OECore, I'm leaning towards being less
concerned about it though and making *every* QT machine write a bbappend
seems a little extreme the other way.

Cheers,

Richard