Syed Mohammed, Khasim wrote:
Hi Dirk, Bill,
Many thanks for this discussion,
Yes, having this discussion is quite good!
I just thought of bringout my views:
1) Beagle foundation is not just for kernel and u-boot, its also meant
for application developers and many others who do not know the GIT usage
so well.
Yes. And "...maybe beagle developers/administrators don't want to
maintain a git repository"
For them giving a complete source to start with will be a
decent solution. So we need a complete kernel + u-boot + x-loader package.
Agree. This is already there.
2) It might not be good to direct these app developers to GIT kernel as
major chunk of Display, camera and audio stuff is not on GIT.
Yes. This was what I meant with "give them a 'validated' complete
kernel for download". And this is already there.
3) Getting our display, audio, camera drivers on GIT will take some
time,
but I can understand this
till then we can't hold other users of Beagle.
Agree.
4) So complete beagle source is good, but we have to decide on how we
give this support to app developers and how we make our lifes easier for
maintaining and pushing this content to Tony's tree / upstream.
"... pushing this content to Tony's tree / upstream": Yes, I think
this is the key point of the discussion:
Do we want to push our patches as fast as possible / as easy as
possible *upstream* with goal "Tony's OMAP kernel git == BeagleKernel
git" (*)? Or do we want to pull Tony's tree regularly into our own
Beagle kernel git and *maybe* sometime push beagle patches upstream?
For me, we should focus our (limited!?) resources on upstream pushing,
and not "wasting" them with maintaining a local git repository and
never having resources for upstream patches.
(*) Sorry but from kernel git point of view BeagleBoard is just one
additional board to the > 10 OMAP based boards already maintained in
Tony's tree. Why should we handle this board other than the way the
other boards are gone into Tony's tree?
5) Giving complete source as a tar package makes our (kernel
developer's) life difficult as we cannot take a diff everytime a new
change is done, we cannot pull in community's changes easily.
6) Maintaining source on a GIT is also complex as we need a maintainer
who actively works with Beagle and OMAP community and takes and
maintains chagnes.
Yes. Dirk says "don't waste time for this complex task" Let us use
existing upstream resources for this.
Sorry, but e.g. learning from DaVinci, it's my feeling that for a lot
of people even sending a more or less clean patch to a ML isn't an
easy job. So concentrate on getting patches, improve them, support
patch review cylces and then send them upstream etc.
7) How about just giving quilt patches (for beagle) against latest
Tony's OMAP GIT tree on code.google.com <http://code.google.com>.
Sorry, which git tree on http://code.google.com? Tony's tree is hosted
by MV mirrored to kernel.org.
If
some one has a fix we can just add that patch to the Series file and
patch folder. These patches can be separated based on functionality and
bug fixes, we can just maintain these patch set. This is easier for
kernel developers and to app developers.
Yes, I think this is a good idea. The exact way we technically handle
this depends on the number of patches we expect. E.g. for DaVinci, it
works quite well to have a up to date list which patches are still not
in git
http://wiki.davincidsp.com/index.php?title=Linux_patches#Pending_patches
With <= 10 patches per week this should work. If we need to improve
this, we can provide patch series.
Any suggestions.
Two additional thoughts:
- Timeframe: How many patches do we expect? Hopefully in the next 2-3
month we should have basic BeagleBoard support patches. Once these are
in upstream (== Tony's git), we can send additional patches to Tony as
all other people do for all other boards. I don't think we really need
a git for this.
- Laziness: With not having a beagle kernel/uboot git tree we force
people (TI? ) to care about upstream sending. If we use a local
git repository our "customers" are happy and there is not really the
need (time?) to care about upstream sending. But if upstream git
repository is the only repository we have, people are "forced" to care
that their stuff will be in there. And, before complaining that it is
hard to get something upstream, remember that Tony is not RMK and
pushes patches more or less in a good shape quite fast.
I'd like to avoid what happend with (TI) OMAP kernels e.g. on
http://focus.ti.com/general/docs/wtbu/wtbusplashcontent.tsp?templateId=6123&contentId=4750
From time to time there are new kernels. Nobody notices changes.
There might be quite nice features and bug fixes in them. But nobody
cares getting them into Tony's git. Sometimes a developer from OMAP ML
takes one kernel, picks some fixes from it and sends them to OMAP
list. But this is a pain and really rare.
Many thanks for this discussion!
Dirk