Git guide for Linux wireless users and developers

This is a quick git-guide for Linux users and developers with emphasis on Linux wireless. The latest Linux wireless development takes place on John Linville's wireless-testing git tree.

Cloning latest wireless-testing

First, clone the wireless-testing.git tree

git clone git://git.kernel.org/pub/scm/linux/kernel/git/linville/wireless-testing.git
cd wireless-testing

Get the latest updates

You will want to update your local git repository to match what John has last committed. You can do this as follows.

git pull

Review the changes last registered

git log

To review changes made to wireless drivers

git log -p drivers/net/wireless/

To review changes made to mac80211

git log -p net/mac80211/

You get the idea.

Hacking on Linux wireless

If you'd like to hack on Linux wireless you can create own branch based on the one you are using. This is so you don't screw your current branch up.

git checkout -b my-fix-for-foo
# hack hack hack
# To get a diff of your work:
git diff > my_changes.diff
# Or if you just want to read them:
git diff
# To revert to the original state of the branch:
git checkout -f
# If instead you want to commit
git commit -a

Check available branches

Suppose you have created a few branches, and just are not sure what you have anymore.

# To view local branches
git branch -l
# To view all remote branches
git branch -r

Reviewing changes between commmits

Suppose you want to get the log and diff between two commits.

# get the SHA of two commits
git log
# Then get the diff of them, by showing the logs in between
git log -p d8a285c8f83f728be2d056e6d4b0909972789d51..9202ec15da36ca060722c363575e0e390d85fb71
# Since SHAs are pretty unique you can just give it a short version
# and it will try to match what is right:
git log -p d8a28..9202e

Merging git branches

Say you have two local branches, and I want to merge them. If you're on local branch my-latest and I want to merge with local branch my-fix-for-foo, you would do:

git pull . my-fix-for-foo

Checkout code as it was from specific commit

Suppose you want to checkout what the codebase looked like at a specific commit SHA. You can do this with branches.

# Long form:
git checkout -b view-commit-foo d8a285c8f83f728be2d056e6d4b0909972789d51
# Or short form:
git checkout -b view-commit-foo d8a28

Delete branches

If you are fed up with a branch delete it. You must not be on that branch so go into another one.

git checkout master
git branch -D old-branch

No need to download more kernel tarballs

You can simply make your current directory look like a specific tag blessed by Linus (or Linville).

git checkout -b v2.6.27-rc7 v2.6.27-rc7

Generate patches

Say you have 3 commits and you want to send the patches now.

git format-patch --cover-letter -o some-dir d8a285c8f83f728be2d056e6d4b0909972789d51..9202ec15da36ca060722c363575e0e390d85fb71
# this is equivalent to, this is the short form
git format-patch --cover-letter -n -o some-dir d8a28..9202e

Where d8a28 was the last commit before you started hacking and 9202e is the current head, meaning the commit ID of your latest commit.

Generating patches for renames

If you are going to rename files you can add "-M" to the arguments to git-format-patch, this way the patches don't generate useless endless removals and adds for a simple rename.

Fixing patches after review

This section tells you how to deal with fixing patches with git after you have sent them out for review or in case you realize you need to go back in history and edit/fix something.

Fixing a patch or commit message

To fix a patch or commit message you have committed you can simply do:

# Edit the file you forgot to add a fix for, and then
# tell git (-a option) all the files you have edited
# should go into the commit, but that you want it to apply
# to the last commit and you also want to review/edit the
# commit message
git commit -a --amend

If you want to ignore all changes you have pending don't use the "-a" option.

Fixing a series of patches

When you a large set of patches and you are not the maintainer chances are pretty high you'll get feedback and you'll need to respin them. A nice trick to avoid having to use quilt/stgit/etc is to use git to edit the patch back in history and continue then. You can do this with git's rebase.

git rebase -i commit-id-foo

This will let you select which patches you want to edit, once done with editing you will have to add the file you fixed

git add drivers/net/wireless/foo/bar.c

And then amend the commit:

git commit --amend

You can skip the 'git add' part by just using 'git commit -a –amend' but keep in mind this will add into the commit *all* changes in your current diff (git diff).

If you didn't have to remove a commit, let the rebase continue.

git rebase --continue

Keep in mind you will have to edit the patches to deal with conflicts if any were found. To deal with them simply edit the files its complaining about, git add them, and do 'git rebase –continue' once done. The conflicts are marked with a set of "<<<<" in the sections. It'll have part from the original file and the part from the new file. You get to mangle with these to figure out what is the right code.

Annotating new revision

If developers raise issues with your patch you are expected to follow up with another iteration of your patch or series of patches. In your new iteration of patches you should specify that these patches are part of a new iteration. You can do this by specifying the iteration number on the subject. For example, for a second iteration you would use:

[PATCH v2]

You can specify this with git by using an argument to git format-patch:

--subject-prefix="PATCH v2"

Removing a commit from a series

If you want to *remove* a commit you can do this trick:

git rebase -i commit-id-foo
git checkout commit-id-before-change
git rebase --continue

Adding a new commit to the series

If you want to add a new commit to the series simply add the commit using the usual commit procedures. Once you are done continue with the rebase.

Sending patches

Read git-send-email man page. But here is a quick summary for those who just want to get it to work. Keep in mind git send-email is a perl script and is usually shipped separately from git core.

You can install your favorite mailer, one option is to use ssmtp.

Setting up ssmtp

Below is an example config that works with an exchange server, in etc/ssmtp/ssmtp.conf:

root=hacker@company.com
mailhub=smtp.company.com
hostname=smtp.company.com
FromLineOverride=YES

UseSTARTTLS=YES
AuthUser=hacker
AuthPass=my-uber-secret-password

Here is an example /etc/ssmtp/revaliases

user:hacker@company.com:smtp.company.com
hacker:hacker@company.com:smtp.company.com

user can be the username (whoami) on the system.

Sending e-mails

Once you have your mailer setup and patches in a directory, review them so they are correct. Once all done send them out using:

# Note new versions of git use: git send-email
git send-email --no-chain-reply-to --from "Random Developer <hacker@company.com>" --to linville@tuxdriver.com --cc linux-wireless@vger.kernel.org --cc maintainer-of-driver@some-cool.org some-dir/

Where some-dir is where you stashed your patches. Keep in mind that if you are submitting a series it helps to send an introductory PATCH [0/n] as well, where n is the number of patches you want to send. You can add this to the git-send-email queue easily using –cover-letter when generating patches using git-format-patch. Be sure to edit the patch 0000-foo then. git-send-email will pick it up when you specify the directory :)