Managing Xen Patches with Git
This document assumes that you are familiar with the following documents
- Xen Project Repositories
- Submitting Xen Project Patches - the document explains conventions related to cover letters, *-by tags, etc. as well as the tooling involved in sending patches and patch series
This document lays out basic examples and best practice of how to use Git to manage Xen patches as part of the patch submission process.
Similar documents exist for
- StGit, a Python application providing similar functionality to Quilt (i.e. pushing/popping patches to/from a stack) on top of Git. You can find instructions at Managing Xen Patches with StGit.
- 1 Generating an initial Patch or Patch Series
- 2 Sending a Patch or Patch Series to xen-devel@
- 3 Addressing Review Comments
- 4 References
Generating an initial Patch or Patch Series
Begin by cloning the git repo from XenProject.org:
$ git clone git://xenbits.xenproject.org/xen.git xen.git
At this point you should have
xenbits set up as the remote repostiory called "origin":
$ git branch -a * master staging ...
This process automatically creates a local branch called
master that will track the XenProject.org branch called
The branch called
staging is the bleeding-edge of commits: this branch is tested regularly with the
xenproject.org build and regression testing system, and when it passes, changes are pushed to the
master branch. However,
master can be significantly behind
From the contributor's point of view, this gives a choice of
stagingas a development baseline and have it apply easily to the tree when all changes are approved. This exposes you to the risk of importing showstopper bugs which prevent you from building or testing. This happens very infrequently, which is why we recommend that people develop against staging.
masterwhich may mean that your change wont apply to
staging. When this occurs your committer ought to try and resolve merge conflicts: however this does not always happen. Thus, it makes sense to rebase against
stagingwhen you are close to completing the code review process.
The remainder of this document assumes you develop against the
Create a branch for your changes
When you want to introduce a change, start by making a new branch based on the most recent change in the
$ git checkout -b staging Switched to branch 'staging'
$ git checkout -b out/trondle-calls staging Switched to a new branch 'out/trondle-calls'
Develop and Test your change
Then edit the source files you want to change. You can see which files have changed by using
Commit your change to your branch
When you're done editing, use
git add to specify which file changes you want included in the changeset, and then use
git commit to make a commit. You will be prompted to make a changset description. The conventions related to what should be in commit messages are described in Submitting Xen Project Patches. The example below is merely intended to explain the necessary git commands: when you submit patches you will likely need more detail than shown in this document.
$ git status # On branch out/trondle-calls # Changes not staged for commit: # (use "git add <file>..." to update what will be committed) # (use "git checkout -- <file>..." to discard changes in working directory) # # modified: foobar/zot.c # modified: foobar/zot.h # no changes added to commit (use "git add" and/or "git commit -a") $ git add foobar/zot.c foobar/zot.h $ git commit
Alternatively, you can commit all changes using "git commit -a":
$ git commit -a
foobar: Add a new trondle calls Add a some new trondle calls to the foobar interface to support the new zot feature. Signed-off-by: Joe Smith <firstname.lastname@example.org>
Patches vs. Patch Series
Every single commit you make on your branch becomes a patch when you submit it. If your change consists of multiple commits, you will be committing a patch series. Information related to Xen Project conventions around patches and patch series can be found here.
Sending a Patch or Patch Series to xen-devel@
You can find instructions on how to send patches in our Patch Submission Guide.
Addressing Review Comments
Once you sent the initial patch or patch series, you are likely to get review comments on each patch or just some of them. Let's assume you have just sent v1 of a series on
my-feature which is made up of the following three patches.
$ git log --oneline --decorate 6711b87 (HEAD -> my-feature) My third patch 8db8b21 My second patch 8efb660 My first patch e18393d (staging) Old stuff on staging
You got feedback that needs to be addressed on the first patch and the second patch which requires re-working these patches. When re-sending the series, you are expected to retain the order and number of patches in the second revision, unless the maintainer asked you to refactor the series.
You could just commit these two changes with a message like “Fix first patch” and "Fix second patch" and worry about squashing it later, but then you have to remember which commit fixes which patch and manually re-order the list of commits in an interactive rebase. Git can do all of this automatically.
Creating outbound version branches
Generally, it is a good idea to create an outbound branch after you sent a patch for review. This allows you to go back to previous versions easily. If you share a branch of a complex series in your cover letter as suggested here, it is important to do this, such that the branch remains static and code reviewers do not get confused by unexpected changes in the shared git branch.
$ git branch my-feature-v1 $ git branch -a * my-feature my-feature-v1 staging master
autosquash to optimize the workflow
During an interactive rebase there are two ways to combine commits:
squash. There are two corresponding options for the
git-commit command called
--squash. These options instruct Git to write a commit message for us, expressing the intention that this new commit will eventually be squashed (or fixed up) with some existing commit.
For your first fix, there is no need to modify the original commit message. Thus, you can use
--fixup and pass the commit that you want the changes to become part of:
$ git add . $ git commit --fixup 8efb660 [my-feature cf02946] fixup! My first patch
The git history now looks like:
cf02946 (HEAD -> my-feature) fixup! My first patch 6711b87 (my-feature-v1) My third patch 8db8b21 My second patch 8efb660 My first patch e18393d (staging) Old stuff on staging
For your second fix, it is necessary to change the commit message. Thus, you can use
--squash to record what needs to be changed when you do the final rebase. For the first fix, you had to tell Git which commit the new changes should be merged with by using the respective SHA from the output of the git log. However, you could have referred to the commit in any of the various ways Git supports. One of the more useful one in code reviews is referring to the commit using some text that appears in its commit message: Git will interpret
:/foo as the most recent commit that contained the string
foo in the first line of its commit message.
To use this in your second code review, you can use:
$ git add . $ git commit --squash :/second [my-feature 5d1ea0a] squash! My second patch
The git history now looks like:
$ git log --oneline --decorate 5d1ea0a (HEAD -> my-feature) squash! My second patch cf02946 fixup! My first patch 6711b87 (my-feature-v1) My third patch 8db8b21 My second patch 8efb660 My first patch e18393d (staging) Old stuff on staging
You can now prepare for sending out v2 of the series by rebasing the series with the following command, which only picks up on commits that begin with
squash!. Git still gives you the chance to move things around in your editor like a regular interactive rebase.
$ git rebase --interactive --autosquash HEAD~5
The git history now looks like:
$ git log --oneline --decorate 7e82927 (HEAD -> my-feature) My third patch 7b4d943 My second patch 32b81e1 My first patch e18393d (staging) Old stuff on staging