Re: Linux 3.4-rc1

[Ingo Molnar <mingo@kernel.org>]

* Linus Torvalds <torvalds@linux-foundation.org> wrote:

> [...]
> 
> One thing worth pointing out is that the header file cleanups 
> were nice, but let's never do them again. Or at least not for 
> a release or two. They caused a lot of merge conflicts and 
> small annoyances, and while I'm ok with resolving merges, it 
> was annoying enough that I don't want to go through that 
> immediately again. I know they also annoyed some 
> submaintainers that were complaining to me about the pain.

I wasn't amongst those complaining and I agree with the system.h 
elimination cleanup, but I think it's better to do these right 
at -rc1 time instead of during -rc0 ...

There's very little complex testing needed: only build coverage 
on architectures and key configs - one iteration of linux-next 
exposure will do that.

So acks can be gathered, it can be rebased to -rc1 or almost-rc1 
and can be pulled in (or conflict-merged), before folks grow a 
large development tree again.

> That said, I do think they helped. The <asm/system.h> 
> disintegration (and to a smaller degree the bug.h cleanups) 
> may have been painful, but it definitely cleaned things up. 
> [...]

Agreed. We probably need a similar sched.h, fs.h and mm.h 
splitting/elimination/shrinking pass as well :-)

> [...] So I guess we *will* do things like this in the future 
> again, I just want to forget about the pain before we embark 
> on this next time. Ok?

I think this kind of pain is largely avoidable via proper timing 
- this one simply wasn't timed properly - pulling it in in the 
middle of the merge window was rather crazy and I think you 
regretted it on the next morning! ;-)

Thanks,

	Ingo