Date: Mon, 15 Aug 2005 12:17:41 -0700
From: tony.luck@intel.com
Subject: Some tutorial text (was git/cogito workshop/bof at linuxconf au?)
Abstract: In this article, Tony Luck discusses how he uses GIT
 as a Linux subsystem maintainer.

Here's something that I've been putting together on how I'm using
GIT as a Linux subsystem maintainer.

-Tony

Last updated w.r.t. GIT 1.1

Linux subsystem maintenance using GIT
-------------------------------------

My requirements here are to be able to create two public trees:

1) A "test" tree into which patches are initially placed so that they
can get some exposure when integrated with other ongoing development.
This tree is available to Andrew for pulling into -mm whenever he wants.

2) A "release" tree into which tested patches are moved for final
sanity checking, and as a vehicle to send them upstream to Linus
(by sending him a "please pull" request.)

Note that the period of time that each patch spends in the "test" tree
is dependent on the complexity of the change.  Since GIT does not support
cherry picking, it is not practical to simply apply all patches to the
test tree and then pull to the release tree as that would leave trivial
patches blocked in the test tree waiting for complex changes to accumulate
enough test time to graduate.

Back in the BitKeeper days I achieved this by creating small forests of
temporary trees, one tree for each logical grouping of patches, and then
pulling changes from these trees first to the test tree, and then to the
release tree.  At first I replicated this in GIT, but then I realised
that I could so this far more efficiently using branches inside a single
GIT repository.

So here is the step-by-step guide how this all works for me.

First create your work tree by cloning Linus's public tree:

 $ git clone git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux-2.6.git work

Change directory into the cloned tree you just created

 $ cd work

Set up a remotes file so that you can fetch the latest from Linus' master
branch into a local branch named "linus":

 $ cat > .git/remotes/linus
 URL: git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux-2.6.git
 Pull: master:linus
 ^D

and create the linus branch:

 $ git branch linus

The "linus" branch will be used to track the upstream kernel.  To update it,
you simply run:

 $ git fetch linus

you can do this frequently (and it should be safe to do so with pending
work in your tree, but perhaps not if you are in mid-merge).

If you need to keep track of other public trees, you can add remote branches
for them too:

 $ git branch another
 $ cat > .git/remotes/another
 URL: ... insert URL here ...
 Pull: name-of-branch-in-this-remote-tree:another
 ^D

and run:

 $ git fetch another

Now create the branches in which you are going to work, these start
out at the current tip of the linus branch.

 $ git branch test linus
 $ git branch release linus

These can be easily kept up to date by merging from the "linus" branch:

 $ git checkout test && git merge "Auto-update from upstream" test linus
 $ git checkout release && git merge "Auto-update from upstream" release linus

Important note!  If you have any local changes in these branches, then
this merge will create a commit object in the history (with no local
changes git will simply do a "Fast forward" merge).  Many people dislike
the "noise" that this creates in the Linux history, so you should avoid
doing this capriciously in the "release" branch, as these noisy commits
will become part of the permanent history when you ask Linus to pull
from the release branch.

Set up so that you can push upstream to your public tree (you need to
log-in to the remote system and create an empty tree there before the
first push).

 $ cat > .git/remotes/mytree
 URL: master.kernel.org:/pub/scm/linux/kernel/git/aegl/linux-2.6.git
 Push: release
 Push: test
 ^D

and the push both the test and release trees using:

 $ git push mytree

or push just one of the test and release branches using:

 $ git push mytree test
or
 $ git push mytree release

Now to apply some patches from the community.  Think of a short
snappy name for a branch to hold this patch (or related group of
patches), and create a new branch from the current tip of the
linus branch:

 $ git checkout -b speed-up-spinlocks linus

Now you apply the patch(es), run some tests, and commit the change(s).  If
the patch is a multi-part series, then you should apply each as a separate
commit to this branch.

 $ ... patch ... test  ... commit [ ... patch ... test ... commit ]*

When you are happy with the state of this change, you can pull it into the
"test" branch in preparation to make it public:

 $ git checkout test && git merge "Pull speed-up-spinlock changes" test speed-up-spinlocks

It is unlikely that you would have any conflicts here ... but you might if you
spent a while on this step and had also pulled new versions from upstream.

Some time later when enough time has passed and testing done, you can pull the
same branch into the "release" tree ready to go upstream.  This is where you
see the value of keeping each patch (or patch series) in its own branch.  It
means that the patches can be moved into the "release" tree in any order.

 $ git checkout release && git merge "Pull speed-up-spinlock changes" release speed-up-spinlocks

After a while, you will have a number of branches, and despite the
well chosen names you picked for each of them, you may forget what
they are for, or what status they are in.  To get a reminder of what
changes are in a specific branch, use:

 $ git-whatchanged branchname ^linus | git-shortlog

To see whether it has already been merged into the test or release branches
use:

 $ git-rev-list branchname ^test
or
 $ git-rev-list branchname ^release

[If this branch has not yet been merged you will see a set of SHA1 values
for the commits, if it has been merged, then there will be no output]

Once a patch completes the great cycle (moving from test to release, then
pulled by Linus, and finally coming back into your local "linus" branch)
the branch for this change is no longer needed.  You detect this when the
output from:

 $ git-rev-list branchname ^linus

is empty.  At this point the branch can be deleted:

 $ git branch -d branchname

Some changes are so trivial that it is not necessary to create a separate
branch and then merge into each of the test and release branches.  For
these changes, just apply directly to the "release" branch, and then
merge that into the "test" branch.

To create diffstat and shortlog summaries of changes to include in a "please
pull" request to Linus you can use:

 $ git-whatchanged -p release ^linus | diffstat -p1
and
 $ git-whatchanged release ^linus | git-shortlog


Here are some of the scripts that I use to simplify all this even further.

==== update script ====
# Update a branch in my GIT tree.  If the branch to be updated
# is "linus", then pull from kernel.org.  Otherwise merge local
# linus branch into test|release branch

case "$1" in
test|release)
	git checkout $1 && git merge "Auto-update from upstream" $1 linus
	;;
linus)
	before=$(cat .git/refs/heads/linus)
	git fetch linus
	after=$(cat .git/refs/heads/linus)
	if [ $before != $after ]
	then
		git-whatchanged $after ^$before | git-shortlog
	fi
	;;
*)
	echo "Usage: $0 linus|test|release" 1>&2
	exit 1
	;;
esac

==== merge script ====
# Merge a branch into either the test or release branch

pname=$0

usage()
{
	echo "Usage: $pname branch test|release" 1>&2
	exit 1
}

if [ ! -f .git/refs/heads/"$1" ]
then
	echo "Can't see branch <$1>" 1>&2
	usage
fi

case "$2" in
test|release)
	if [ $(git-rev-list $1 ^$2 | wc -c) -eq 0 ]
	then
		echo $1 already merged into $2 1>&2
		exit 1
	fi
	git checkout $2 && git merge "Pull $1 into $2 branch" $2 $1
	;;
*)
	usage
	;;
esac

==== status script ====
# report on status of my ia64 GIT tree

gb=$(tput setab 2)
rb=$(tput setab 1)
restore=$(tput setab 9)

if [ `git-rev-list release ^test | wc -c` -gt 0 ]
then
	echo $rb Warning: commits in release that are not in test $restore
	git-whatchanged release ^test
fi

for branch in `ls .git/refs/heads`
do
	if [ $branch = linus -o $branch = test -o $branch = release ]
	then
		continue
	fi

	echo -n $gb ======= $branch ====== $restore " "
	status=
	for ref in test release linus
	do
		if [ `git-rev-list $branch ^$ref | wc -c` -gt 0 ]
		then
			status=$status${ref:0:1}
		fi
	done
	case $status in
	trl)
		echo $rb Need to pull into test $restore
		;;
	rl)
		echo "In test"
		;;
	l)
		echo "Waiting for linus"
		;;
	"")
		echo $rb All done $restore
		;;
	*)
		echo $rb "<$status>" $restore
		;;
	esac
	git-whatchanged $branch ^linus | git-shortlog
done