hotplug plans - issue 1 (ideas welcome)

[Dave Airlie]

On Tue, Mar 6, 2012 at 10:55 PM, Keith Packard <keithp at keithp.com> wrote:
> <#part sign=pgpmime>
> On Tue, 6 Mar 2012 12:27:36 +0000, Dave Airlie <airlied at gmail.com> wrote:
>
>> This is of course only one of the issues I face, but I would need to
>> get some ideas on merging and keeping the tree in some ways stable.
>
> I'd prefer a); either of the other choices means having piles of code
> that most people won't ever use, and which will quickly rot anyways,
> leaving us with a partially broken a) in the end.

Well my major problem is I'm torn between doing this like kernel
development where I could make lots of incremental API changes and fix
the drivers up, and with the needs that people want drivers to build
against old X servers and be maintainable. Also doing it incrementally
means having to work out what to do with XAA and DRI1, whereas if I
just run with dual APIs for a couple of server releases I can ignore
that problem until someone nukes them.

The thing is its very messy to merge any of this and keep the server
stable, if we take the non-duplicated code path, and I don't want to
end up in a crunch in 3-4 months time where the server has bugs we
haven't ironed out and we can't release it until the one person
blocking it (i.e. me) uncrunches it.

I've done (c) here with a bunch of ugly sed and a few ifdefs and I've
reduced the hotplug server diff to a more manageable 20k loc diff.

I just can't see a nice way to do this incrementally and meet the
requirement of the X release process. If we'd merged the drivers it
would be much easier to just evolve things, but since some people
dislike that idea, it means I have to go with the multiple APIs/flag
day approach which really doesn't sit well with me.

Dave.