Re: [PATCH v3 0/9] Media Controller capture driver for DM365

[Sylwester Nawrocki <sylvester.nawrocki@gmail.com>]

On 11/28/2012 10:29 PM, Dan Carpenter wrote:
> On Wed, Nov 28, 2012 at 08:30:04PM +0100, Sylwester Nawrocki wrote:
>> On 11/28/2012 01:22 PM, Dan Carpenter wrote:
>>> In the end this is just a driver, and I don't especially care.  But
>>> it's like not just this one which makes me frustrated.  I really
>>> believe in linux-next and I think everything should spend a couple
>>> weeks there before being merged.
>>
>> Couple of weeks in linux-next plus a couple of weeks of final patch
>> series version awaiting to being reviewed and picked up by a maintainer
>> makes almost entire kernel development cycle. These are huge additional
>> delays, especially in the embedded world. Things like these certainly
>> aren't encouraging for moving over from out-of-tree to the mainline
>> development process. And in this case we are talking only about merging
>> driver to the staging tree...
>
> Yeah.  A couple weeks is probably too long.  But I think a week is
> totally reasonable.

Agreed, exactly that couple weeks requirement seemed a bit long to me.

> You have the process wrong.  The maintainer reviews it first, merges
> it into his -next git tree.  It sits in linux-next for a bit.  The
> merge window opens up.  It is merged.  It gets tested for 3 months.
> It is released.

I believe what you're describing is true for most subsystems.  At
linux-media the process looks roughly that you prepare a patch and post
it to the mailing list for review.  Regular developers review it, you
address the comments and submit again.  Repeat these steps until
everyone's happy.  Then, when the patch looks like it is ready for
merging it is preferred to send the maintainer a pull request.
Now there can be a delay of up to couple weeks.  Afterwards the patch
in most cases gets merged, with a few possible change requests. However
it may happen the maintainer has different views on what's has been
agreed before and you start everything again.

With a few sub-maintainers recently appointed hopefully there can be
seen some improvement.

> It should work as a continuous even flow.  It shouldn't be a rush to
> submit drivers right before the merge window opens.  It's not hard,
> you can submit a driver to linux-next at any time.  It magically
> flows through the process and is released some months later.
>
> It does suck to add a 3 month delay for people who miss the cut off.
> Don't wait until the last minute.  In the embedded world you can
> use git cherry-pick to get the driver from linux-next.

Yeah, it's not unusual to work with specific -rc tree with multiple
subsystem -next branches on top of it.

It's just those cases where you're told to get feature A in the kernel
release X and it is already late in the development cycle... But it
might just be a matter of planning the work adequately with proper
understanding of the whole process.

--

Thanks,
Sylwester