Re: [GIT PULL] KVM updates for the 3.4 merge window

[Ingo Molnar <mingo@kernel.org>]

* Avi Kivity <avi@redhat.com> wrote:

> Say a fix comes in which needs to be mainlined during -rc.  So 
> I put it in some other branch, to be sent off to Linus in a 
> few days after maturing a little.  Meanwhile developers see an 
> incomplete tree, since that patch is not in it.
> 
> Once Linus pulls, I can merge it back (or even before, if I'm 
> reasonably certain it's not going to change), but it leaves a 
> history of unneeded merges.  Or we can do throwaway merges 
> like tip.git.

We don't do throwaway merges in the -tip development branches 
themselves, i.e. in tip:sched/core, tip:perf/core, 
tip:timers/core, etc.

When a fix goes into tip:sched/urgent then until Linus merges it 
it's not in tip:sched/core. 99% of the fixes don't *have to* go 
into sched/core straight away.

In the odd case where there's some dependency, we can manually 
merge it into tip:sched/core ahead of Linus pulling into an -rc. 
Those rare merges are not a problem, and I explain the reason in 
the merge commit itself.

If you look at:

 gll v3.2..v3.3 | grep -E '/urgent.*/core'

you'll see that I only had to do it once in the previous cycle:

 d6c1c49de577 Merge branch 'perf/urgent' into perf/core

and the changelog explains the background:

Merge reason: Add these cherry-picked commits so that future changes
              on perf/core don't conflict.

it was a rare, oddball situation where we cherry-picked 
perf/core changes into perf/urgent. Extra merges are perfectly 
fine in that case.

The 'throwaway' tip:master branch you are probably referring to 
is basically just a testing branch, a convenient merged tree of 
the one dozen maintainer trees that are hosted in -tip. Since we 
don't want to force Linus's hand of him being able to reject 
individual trees we don't merge them properly - hence the 
integrated tree is a throwaway tree in theory.

In practice I tend to throw it away only once per cycle, around 
-rc1, once all pending trees went to Linus. tip:master is not 
used for any Git based contribution work - it's for testing, 
it's for people who want to work with patches - the commits 
themselves always go into persistent non-rebasing, append-only 
Git trees.

If we mess up bisectability we do a delta fix. When choosing 
between somewhat better bisectability and a proper history that 
others can rely on then proper history wins hands down.

Thanks,

	Ingo