[Qemu-devel] Merging invasive changes (Was Re: [PULL][v2] Memory API fix)

[Anthony Liguori]

Hi,

This is a summary from a discussion Avi and I had earlier about merging 
invasive changes (like the memory API) into the tree in a more fluid 
fashion.  Here's my suggestions:

- Invasive changes over a long period of time (multi batches) should be 
done via git pull.  patch (and therefore git-am) sucks at resolving 
conflicts which are a given for these types of changes.  git-merge is 
much better at this.

- Pull requests should be prefixed with '[PULL]'.

- Patches should go to the ML as often as they are created.  Small sets 
of patches can depend on each other.  Don't dump 40 patches and then 
submit a pull request the next day.

- Pull requests should be batched in a 1-2 week period after having 
received reviews or sufficient time to be reviewed.  If it's something 
that really needs review, make sure it gets review before doing the pull 
request.   Good judgement is required here.  Batching is important to 
avoid breaking the tree daily.  Exceptions will happen and that's okay, 
but they should be exceptions.

- Patches destined for a subsystem should be prefixed with a subsystem 
marker.  For instance, '[PATCH] block: fix QED blah blah'.  If I see the 
prefix for a known maintained subsystem, I will delete the patch from my 
own queue.

- Subsystems are 'known maintained' by having an entry in MAINTAINERS 
and by having a responsive maintainer.

- Invasive, but mechanical changes shouldn't be overly split.  60 four 
line patches that makes the same pattern of change are harder to review 
than 6 forty line patches that apply the same pattern repeatedly.

- Non-mechanical changes should be split of course based on logical 
functional areas.

Regards,

Anthony Liguori

On 08/22/2011 11:54 AM, Avi Kivity wrote:
> [v2: fix git url...]
>
> Anthony, please pull from:
>
> git://git.kernel.org/pub/scm/virt/kvm/qemu-kvm.git memory/core
>
> to receive a memory API fix. The fix is not required for any of the
> already merged patches, but is needed by other patches in my queue.
>
> I am sending core patches separately than the conversions to increase
> their visibility.
>
>
> (and I'll take this opportunity to announce a new layout for the branch:
>
> - memory/core - core fixes, usually not rebased
> - memory/batch - pending batch
> - memory/queue - the rest of the patches in my queue
> - memory/master - a merge of the first three; please post patches based
> on this branch
>
> this will allow me to fast-path core fixes without rebasing
> memory/batch, except when a dependency is involved.)
>
> Richard Henderson (1):
> memory: Fix old_portio vs non-zero offset
>
> memory.c | 4 ++--
> 1 files changed, 2 insertions(+), 2 deletions(-)