[CentOS-virt] RFC: Use git am for patchqueue

Tue May 20 11:20:25 UTC 2014
George Dunlap <dunlapg at umich.edu>

On Tue, May 20, 2014 at 9:42 AM, Karanbir Singh <mail-lists at karan.org> wrote:
> On 05/15/2014 07:05 PM, George Dunlap wrote:
>> OK, I've got a basic initial port of the xen package up to 4.4.0.  But
>> before I post that, I wanted to get comments on a new method of
>> managing the Xen part of the patchqueue.  You can find the RFC branch
>> here:
>> repo: http://github.com/gwd/sig-virt-xen
>> branch: out/git-am-patchqueue-RFC-v1
>> The basic idea is to put as much as possible into a 'git am' file.
>> This file can easily be imported into git on top of a Xen git
>> repository with "git am [filename]", rebased / edited like any branch,
>> and the exported again with "git format-patch -N --stdout
>> [release-tag]".
>> I've got a copy of the Xen repo with the patchqueue applied as a branch here:
>> repo: http://github.com/gwd/xen
>> branch: centos/pq/4.2.4-v1
>> Unfortunately this only works for patches to things in the core Xen
>> repo; patches to qemu, or the out-of-tree blktap will need to be
>> managed separately still.  However, at the moment those patches are
>> kind of small.
>> Having the patchqueue in this format made rebasing to 4.4 a *lot* easier.
> assume that we have a git repo per package, then the deps ( like qemu
> ?pxe etc ) could also use the same process ?

I don't see why not -- although we could just "fault them in" lazily as-needed.

> Also, would it make sense to consolidate this queue under the
> github.com/CentOS/<package> repo ?

I don't really understand the question.  I was assuming the workflow would be:

* George something on github.com/gwd/sig-virt-xen
* George sends pull request to github.com/CentOS/sig-virt-xen
(possibly cc'ing centos-virt mailing list)
* Opportunity for discussion, review, objections, suggestions, &c
* Eventually it gets pulled into CentOS/sig-virt-xen
* ...then pulled into git.centos.org for testing.

Did you mean something else?

> I'm also thinking that we might need a process like this for pretty much
> everyone else - there isnt a precidence for the SIG's in this regard as
> yet ( and anecdotally, we use a more manual version of exactly the same
> process for handling branding/blacklists in the centos distro rpms queue
> as well ).

Well each SIG can come up with a method of managing patches that works
for them; but feel free to suggest this to people. :-)