Date: Thu, 14 Feb 2008 17:57:28 -0500
From: "John W. Linville" <linville@tuxdriver.com>
To: linux-wireless@vger.kernel.org
Subject: Upstream wireless process changes
Message-ID: <20080214225728.GC2981@tuxdriver.com> (sfid-20080214_231226_719097_BFD639A8)
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Sender: linux-wireless-owner@vger.kernel.org

Greetings,

Given current discussions, it seems timely to revamp some of the
upstream wireless tree management policies.  I'm sure I still won't
be able to please everyone, but maybe I can change who gets to be
unhappy about the process... :-)

Let's review the current process.  The wireless-2.6 tree has a number
of branches, each with a specific use.  Most of these are for my own
administrative purposes relating to coping with the processes upstream
from me.  These include any "fixes", "upstream", or "pending" branches,
as well as any "master-*" branches.  There are also "merged-*" branches
which I maintain to assist me when rebases are necessary.  In addition,
there have been topic branches such as "at76" and (formerly) "ath5k".
And there is the "mm-master" branch, which primarily just collects
the topic branches together for testing in -mm.

The "everything" branch is an integration branch which collects patches
from all the "fixes", "upstream", and "pending" branches as well
as any topic branches.  This is the branch I ask developers to use.
The purpose of this branch is to avoid having API-level patches miss
any drivers as well as to avoid any similar conflicts.

Finally, in the old process the "master" branch always just pointed
at the most recent full or -rc release from Linus.  This always seemed
to confuse users looking for the "latest and greatest" wireless bits.
Moreover, all the branches created confusion among both users and
developers.

The new process shifts from reliance on branches to the use of
several trees.  Each tree may have some placeholder branches used for
administrative purposes, but the interesting bits will be committed
on the "master" branches of those trees to avoid confusion about
which tree normal people should use.

The wireless-2.6 tree will primarily become a vehicle for pushing
patches to the current -rc release.  This replaces the former "fixes"
branches.  It is my intent that this will not be rebased except in
the most extreme (and unforseen) circumstances.

A new wireless-2.6.26 tree (or 2.6.x as appropriate in the future) will
be the vehicle for queueing patches to net-2.6.26 (or its successors)
in anticipation of the next merge window.  This tree will regularly
be updated to correspond to the current state of net-2.6.26.  I will
avoid rebasing this tree as much as possible, but given its dependence
on net-2.6.26 I will be somewhat at Dave's mercy... :-)

Finally, a new wireless-testing tree will be created to replace the
usage of the former 'everything' branch.  This tree will be based
on a current -rc release in hopes avoiding the churn in between
-rc releases.  The tree may contain topic branches (e.g. "at76")
as appropriate, as well as picked commits from wireless-2.6 and
wireless-2.6.26.  I will attempt to limit rebasing this tree as much as
practical, at the expense of having some ugly history.  However, this
tree almost certainly will be rebased from time to time, and you should
expect any patches in this tree to be re-committed in wireless-2.6
or wireless-2.6.26 before going upstream -- you have been warned!

I hope that this "covers all the bases" for our various process
needs (merging fixes, queueing for the merge window, integration and
on-going development).  Any coments or suggestions you might have
are welcome now. :-)

So, comments?

Thanks,

John
-- 
John W. Linville
linville@tuxdriver.com