Xserver driver merging pros & cons

[Peter Hutterer]

On Fri, Sep 16, 2011 at 09:46:29PM -0700, Jeremy Huddleston wrote:
> On Sep 16, 2011, at 13:10, Stéphane Marchesin wrote:
> > 2011/9/16 Jeremy Huddleston <jeremyhu at apple.com>:
> >> 
> >> On Sep 16, 2011, at 12:45 PM, Stéphane Marchesin wrote:
> >> 
> >>> Well, the issue is not separate. Driver repos are straightforward to
> >>> get changes in and therefore can move forward easily.
> >> 
> >> How would that be different from an xorg-server tree with the exact same privileges?  I don't see any difference as far as individual contributors are concerned.
> >> 
> >>> For xserver changes, with the amount of latency/difficulty involved, people don't
> >>> even try.
> >> 
> >> Can you be specific?  What latency are you talking about?
> >> 
> >> 1) Is it latency getting a [PULL] or a ready [PATCH] applied?  Perhaps we can address that.
> >> 
> > 
> > The difficulty of getting reviews, the difficulty to attract attention
> > to a patch and then get it pulled by someone who doesn't understand
> > it.
> 
> I pull patches that I don't understand.  I just make sure it was reviewed by someone who does understand it.
> 
> >> 2) Is the latency involved due to missing reviews?  We need more people to step up and actually review.  Yes, this is a problem which we need to discuss.
> > 
> > Extending a model which is currently not working won't fix the model.
> 
> If the reason the model doesn't work is because people don't review, then
> some how getting people to step up and give reviews would fix the model.

I think the one thing we don't cover right now are patches that cannot
really be reviewed by anyone but the author. Are they left to linger or will 
they be pulled regardless? IIRC I've gotten patches into the tree with
little or no review after pestering keith for a bit but those patches are
still quite slow to go in.

I believe most of your changes go in mostly unreviewed, mostly so because
your subsystem affects hardly any other upstream developers. Simply spelling
that out may alraedy be a step towards a better process.

> > My point is, for drivers, that review system will slow
> > things down (because you need to wait for a pull from someone who has
> > no familiarity with your code)
> 
> No you don't.  The code is already in your tree?  Why do you need to wait
> for it to land on master?  Why do you CARE when it lands on master?  The
> only people who care when it lands on master are release managers and
> distributions.  Driver developers won't even be using master, they'll be
> using their branch off of master.

if you're working on a bug, you can't close the bug until the patch lands on
master. if that is delayed by review delays, pull delays and then
cherry-pick + pull to stable delays, it can be quite annoying (more
bugreporter comments, more duplicates, remembering to close the bug, etc).

Cheers,
  Peter