Git Wine Tutorial
Welcome to the Git Wine tutorial! This page describes how to manage Wine code and patches with Git. Git is a fast version control system, originally written for use with large repositories, such as the Linux Kernel source. The Git Wine tree gives you fast access to the entire Wine tree and its history, and allows you to maintain a local tree or patch series and merge it easily with WineHQ.
- 1 Tutorials and guides
- 2 Set up your Git repository
- 3 Glossary
- 4 Managing your changes - the simple way
- 4.1 Committing a change into your local tree
- 4.2 Seeing where you are and what you've done
- 4.3 Reverting changes in your working copy
- 4.4 Undoing commits
- 4.5 Editing commits
- 4.6 Editing commits the hard way
- 4.7 Removing trailing whitespace
- 4.8 Keeping up to date with WineHQ
- 4.9 Resolving merge conflicts
- 4.10 Sending patches: generating a patchset
- 5 Managing branches
- 6 Other useful operations
- 7 Working with GitHub
- 8 Patch stack
- 9 Annoyances
- 10 Other Git repositories
Tutorials and guides
- This Wiki page.
- The Wiki page on Regression Testing (which is aimed at non-programmers).
- The official Git tutorial, and the official Git manual.
- Pro Git, a free e-book by Scott Chacon and Ben Straub.
- Jeff Garzik's Git tutorial (Aug 2008).
- A very comprehensive guide to advanced git usage ("branch wizardry and git grandmastery") called Git Magic available; despite the name it also addresses Basic Trickery for the beginner's needs.
- On git.or.cz, there is a Git crash course for SVN users (Sep 2008).
- LWN's article on Branching and merging with git (Nov 2006 - which is a bit old, but detailed).
- Git for Computer Scientists, a "quick introduction to git internals for people who are not scared by words like Directed Acyclic Graph" (Jul 2009).
- The Git Parable "will take you on a journey through the creation of a Git-like system from the ground up" (May 2009).
Note: To get documentation on a Git command, use
git help or
man. For example, the following two commands are equivalent:
git help format-patch man git-format-patch
Set up your Git repository
The first step to using Git with Wine is to set up a local Git repository.
Downloading and installing Git
It's recommended to install Git via your distribution's package manager. If you want to install from source, you can download the latest version of Git from https://www.kernel.org/pub/software/scm/git/ . It installs into ~/bin by default. See the Pro Git book for build instructions.
Cloning the Wine Git repository
You can browse the WineHQ Git repositories on the web at https://source.winehq.org/git. To download the main Wine repository:
git clone git://source.winehq.org/git/wine.git cd wine
If you are stuck behind an uncooperative firewall, try cloning from https://github.com/wine-mirror/wine.git instead.
The clone operation transfers around 150 MiB of data. If all goes well, the output of git config remote.origin.url should be:
$ git config remote.origin.url git://source.winehq.org/git/wine.git
and you will have a single branch named "master", which can be revealed by running git branch.
If you plan on sharing or committing any patches at all, you should set your name and email address using:
git config --global user.name "Your Name" git config --global user.email "firstname.lastname@example.org"
A useful feature of Git is its ability to color the output of various commands (diff, status, branch, etc). To enable colors in all commands that support it, when running them from a terminal:
git config --global color.ui auto
Particularly useful is the colored output of the git diff command, as it also highlights trailing whitespace in patches.
See also the official Git glossary.
- A commit (or changeset, or revision) is a snapshot of a codebase. Each commit has a name -- its commit id -- which is a SHA1 hash. Commits can have other names (tags, etc). See the "Specifying revisions" section of the git rev-parse page for details on how refer to commits. The verb to commit means to create a commit.
- A diff is a file describing the differences between two sets of files. They are created with the command git diff. If you have code in the "before" state, you can apply the patch (with git apply) and you end up with code in the "after" state.
- A patch is a file that includes authorship information, a description, and a diff. They are created with git format-patch and committed with git am. This similar to applying a patch, but it also creates a commit that preserves the original description and authorship information.
- The HEAD (or tip) of a branch is the most recent commit. On its own, HEAD means the tip of the current branch.
- master is the main branch. SVN/CVS calls this "trunk".
- The parent of a commit is the one before it in the history. (Commits can technically have more than one parent, but the Wine repo avoids using this feature, preferring a linear history.) Given a commit
abcde, its parent is referred to as
abcde^, and its great-grandparent is referred to as
- A repository (or repo) is a database storing the source code of every available version of a program, and the author of every change.
- A tree is a git technical term meaning "directory" (sort of), and sometimes means the whole codebase of a project ("the Wine tree").
- The working copy or working tree refers to the files and directories on your file system, the ones you can see and change with the file manager. The git status command will refer to changes to these files as "Changed but not updated".
Managing your changes - the simple way
Committing a change into your local tree
After editing the checked out tree, you can use git status to see which files have changed:
Or you can examine the difference by using git diff:
To then commit all changed files to your local tree, use the git commit command with the -a option:
git commit -a
If you only wish to commit some files, use:
git commit <file1> <file2>
git add <file1> <file2>... git commit
Commit early, commit often: Your local Git tree is yours. You should feel free to commit patches frequently, as it's not until you mail them in that they have a chance of being committed upstream.
Seeing where you are and what you've done
git whatchanged # list of commits, which shows what files were altered git log # list of commits git log --decorate # list of commits, with the last upstream commit marked git log --stat # list of commits, with diffstats git log --stat -p # list of commits, with diffstats and patches
You can get a list of files that you have changed but have not yet committed using git status:
git status # list of uncommitted changes
There's a nice tool to view your Git repository named gitk, written in tcl/tk. AJ suggests using something like
gitk wine-1.0.. to make it go faster (please note that the trailing
.. is important). gitk visualizes both committed and uncommitted changes.
qgit also provides similar functionality, but in a Qt based interface. It appears to be faster than gitk and has additional features such as an annotation facility to identify which change introduced each line of a file.
If you prefer using your terminal, "git lol" and "git lola" are useful aliases you can add to your ~/.gitconfig file:
[alias] lol = log --graph --decorate --pretty=oneline --abbrev-commit lola = log --graph --decorate --pretty=oneline --abbrev-commit --all
If you have a webserver running, git instaweb allows you to view your local tree in your web browser.
Reverting changes in your working copy
If you have edited some files, but decided you don't like the changes you've made and want to undo all the changes that you've made to your working copy, you can use git checkout:
git checkout -f # revert everything git checkout file-name # revert one file
Alternatively, use git reset:
git reset --hard HEAD # revert everything
If you want to undo your most recent commit, you can use the git reset command:
git reset HEAD^ # undo commit, but keep changes to working files git reset --hard HEAD^ # undo commit, and reset working files as well git reset --hard HEAD~5 # undo 5 commits git reset --hard origin # scrap all changes and start all over again
To edit the most recent commit:
vi file.c # edit the file git commit --amend file.c # redo the commit
To edit earlier commits (or reorder or delete them) use the -i (aka --interactive) option to git rebase. So if you are interested in altering the 5 most recent commits use:
git rebase -i HEAD~5
This will open your editor, with a list of commits prefixed with pick. To delete a commit, just remove its line. To reorder them, just rearrange the lines. To edit commits, change pick to edit.
Be sure to follow the instructions carefully when doing
git rebase -i. Specifically, when you are editing a commit (that you explicitly requested to edit) and are satisfied with the changes, you must use:
git add <file1> <file2>... git commit --amend git rebase --continue
git-rebase -i asks you to edit a commit that you have not requested to edit (e.g. there is a conflict), you must use:
git add file1 file2... git rebase --continue
This is quite important - not following this carefully will result in merged patches.
Editing commits the hard way
Instructions for those who don't want to use git rebase -i:
If the commit is not the most recent one, but say 5th from the top then you can:
git checkout -b tmp HEAD~5 # rewind to the commit in question vi file.c # edit the file git commit --amend file.c # redo the commit without deleting the commit git rebase --onto tmp master~5 master # replay the later changes git branch -D tmp # clean up the temporary branch
Where there are a number of files to amend you are probably better off using:
git checkout -b tmp HEAD~5 # rewind to the commit in question git reset HEAD^ # delete the commit at the now current point vi file1.c # edit vi file2.c # the files git commit -a -c ORIG_HEAD # redo the commit incorporating all changed files git rebase --onto tmp master~5 master # replay the later changes git branch -D tmp # clean up the temporary branch
Where the commit is not the most recent one, but say 5th from the top and you wish to insert a new commit, then you can:
git checkout -b tmp HEAD~5 # rewind to the commit in question vi new_file.c # create the new file git commit -m "New commit of file new_file.c" new-file.c # create a new commit or a series of commits git rebase --onto tmp master~5 master # replay the later changes git branch -D tmp # clean up the temporary branch
Likewise if you want to delete a commit that is not the most recent one, then you can:
git checkout -b tmp HEAD~5 # rewind to the commit in question git reset HEAD^ # delete the commit at the now current point git checkout path/file1 path/file2 etc # delete the changed files git rebase --onto tmp master~5 master # replay the later changes git branch -D tmp # clean up the temporary branch
and the commit is gone. You need to checkout all the changed files though and the rebase may throw some errors for you to resolve as it applies later commits.
Removing trailing whitespace
Thank you to Mike Kaplinskiy for this very helpful hint:
git rebase --whitespace=fix origin/master
"It fixes whitespace on all the commits that you've made. I think it's pretty good about merge conflicts due to whitespace as well. I don't know of a way of doing this at commit time though."
This is essential for submitting your patches to the Wine project. Please see this post for more information: https://www.winehq.org/pipermail/wine-devel/2010-July/084870.html
Keeping up to date with WineHQ
Now that you have a copy of the Wine Git repository, you will periodically need to receive new commits from the original repository. You do this using:
git pull --rebase
This is equivalent to:
git fetch git rebase origin
git fetch retrieves new files from the WineHQ Git repository; this should always be a safe operation as it does not change your local file system.
git rebase origin reapplies any local commits you have made onto the latest WineHQ branch. (Technically, it creates a new branch on 'origin', reapplies all the patches in your current HEAD to the new branch, then changes HEAD to the new branch.) Patches already applied upstream will not be reapplied.
A common mistake is to use
git fetch by itself. It will only download updates but will not apply them. Another common problem is trying to rebase while having uncommitted changes. One way to fix this is to:
git commit -a # commit changes git rebase origin # rebase git reset HEAD^ # uncommit changes
When you send patches, inevitably, some of your patches will be rejected, while others will be accepted. If you have written a series of patches, but only some of those are rejected, it can be annoying to reorder them, fix one or two problems and resubmit. The main git tools that you can use to help solve this problem are git rebase and git cherry-pick. See here for a discussion on the Git mailing list about rebasing on local branches.
Resolving merge conflicts
When rebasing, sometimes upstream changes prevent your patches from applying. If there is a conflict, you will see something like this:
Applying <patchname> error: patch failed: <file>:<line> error: <file>: patch does not apply Using index info to reconstruct a base tree... Falling back to patching base and 3-way merge... Auto-merged <file> CONFLICT (content): Merge conflict in <file> Failed to merge in the changes. Patch failed at <msgnum>. When you have resolved this problem run "git rebase --continue". If you would prefer to skip this patch, instead run "git rebase --skip". To restore the original branch and stop rebasing run "git rebase --abort".
There are two choices now: resolve the conflict or skip the patch. The file in question will contain conflict markers where the patch failed:
<<<<<<< [code that caused patch not to be applied] ======= [what would have been here if the patch had been applied] >>>>>>>
To resolve the conflict you have to manually merge the code between the conflict markers, leaving the file in a compilable state. After that, run
git add <file> git rebase --continue
to remove the merge-conflict state and continue with the operation.
Patches can be skipped as follows:
(git reset --hard # removes the patch) git rebase --skip
Sending patches: generating a patchset
After checking in your local changes (in multiple small commits), you can generate a list of the patches you need to send upstream (i.e. to wine-devel) with the git format-patch command:
git format-patch origin
"origin" is short for "origin/master", the default name of the upstream WineHQ branch. One file per patch will be created in the current directory. You can change the output directory with
-o <directory>, for example
-o out to save the patches in the out directory.
Important: If you have not read it already, please read Submitting Patches.
Adding a cover letter
--cover-letter option to
git format-patch allows you to create a "patch 0" that describes the patchset. This first "patch file" contains no actual changes, but can be easily sent as an email before the real patches.
Sending the patches using SMTP
Before you can use the
git send-email command, you have to add information about your SMTP server to ~/.gitconfig. For example, if you are sending from Gmail, put:
[sendemail] smtpencryption = tls smtpserver = smtp.gmail.com smtpuser = email@example.com smtpserverport = 587
You can then send the patches with the command:
git send-email *.patch
If you want to check your patches first (e.g. to alter the subject line), just use the
--annotate option to
Sending the patches using IMAP
You should be able to put patches directly into an IMAP drafts folder using git imap-send.
git format-patch --stdout origin | git imap-send
Set up the IMAP server by editing wine/.git/config and adding entries something like this:
[user] name = "Your Name Here" email = "firstname.lastname@example.org" [imap] folder = "INBOX.Drafts" tunnel = "ssh -C -q email@example.com /usr/bin/imapd ./Maildir 2> /dev/null" [format] headers = "To: wine-devel <firstname.lastname@example.org>\n"
The above works for Courier IMAP; for Dovecot, change the [imap] section to something like this:
[imap] host = dummy folder = "Drafts" tunnel = "ssh email@example.com /usr/libexec/dovecot/imap 2> /dev/null"
If you would like to send patches directly to your Drafts folder on gmail, change the [imap] section to something like the following:
[imap] folder = "[Gmail]/Drafts" host = imaps://imap.gmail.com user = user pass = pass port = 993 sslverify = false
Using Mozilla, sending patches is then just a matter of clicking on "Edit Draft", reviewing the mail and then clicking "Send". If you're using Evolution, you can drag and drop the .patch files into your drafts folder. Patches in the Drafts folder will have the date and time of the timestamp of the commit, hence if you generate multiple times you will have many copies of the same patch with the same date and time. This will be the case until you amend the commit and get a new commit timestamp.
Important: Many email clients will alter your email in ways that will prevent them from applying. For example, Thunderbird will wrap emails and send them with 'format=flowed', which will make them unusable by Git. For Thunderbird, you can follow these instructions to send patches. If all else fails, you can add
--attach to the
git format-patch command to send the emails as attachments.
Setting up ssh simplifies the patch generation by removing the need to enter a password. Use ssh-keygen to create your keys and copy ~/.ssh/id_rsa.pub to ~/.ssh/authorized_keys to allow the tunnel to be created without entering a password.
Using local Thunderbird folders, you can use the following approach to add your patches to the Drafts folder (without using IMAP):
git format-patch --stdout --attach origin | formail -ds >>"/home/username/.thunderbird/12345678.default/Mail/Local Folders/Drafts"
Note: Sometimes you have to rebuild the index of the Drafts folder in Thunderbird to see the mails added this way.
Using local KMail folders, you can use the following approach:
git format-patch --stdout origin | formail -s procmail
Assuming you don't already use procmail to sort your email, you can use the following .procmailrc
Now, all you need to do is to set up a new receiving account in KMail that collects mail from /home/username/.maildir and filter emails coming in on that account to your drafts folder.
Git branches can be useful if you are working on a lot of bugs at the same time. However, they are completely optional and overkill if you are only working on one thing.
Creating a branch
To create a branch, use git checkout with the -b option. For example:
git checkout -b new-branch HEAD~5
This winds back the working directory to HEAD~5, and forks the history there, creating the branch new-branch. The new branch becomes the current branch.
Use git branch to list all branches.
git branch # local branches only git branch -a # both local and remote branches
To change branches, use git checkout:
git checkout master # change to branch master
Merging and rebasing branches
Git allows you to merge branches together; this is not done in the WineHQ repository, so it is easier to just rebase/cherry-pick instead.
git branch -D new-branch
For more information on managing branches in git, see the Git Branches page.
Other useful operations
Picking patches from another branch
You can cherry pick (apply) a patch from another branch into your current branch using:
git cherry-pick <commit-id>
This will create a new commit, but with authorship information from the original patch.
Getting rid of timestamp changes
Git considers a file changed if its date is different from that in the Git index file. "git diff-index HEAD" may show files have changed if you have edited them and reverted the changes (or even just touched the file). You can remove this difference using:
Regression testing is really easy with Git. It's done with the help of git bisect that does all the magic. So all that's left to do is to compile and test. Even non-developers can do it. See Regression Testing for instructions.
To see a list of committers in the last 5 years, sorted by number of commits:
git shortlog -s -n --since="(5years)"
Finding who changed what
git log /path/ # log of changes to files in /path/ git log /path/file # log of changes to /path/file git blame /path/file # show per-line authorship info git blame -w /path/file # ditto, but ignoring whitespace changes
Working with GitHub
GitHub, through its ability to "fork" an existing project, provides a good way to work on the Wine project when, for example, the project itself is in "code freeze" due to proximity to a release milestone.
You will find the WineHQ repository on a !GitHub mirror.
To fork, simply register. Once registered, click on the Fork button in the upper right hand side of the Wine repository link above.
Once you have forked
You will find helpful information on forking a project at !GitHub.
Once forking is complete, you will find an SSH URL listed, that has read/write access, of the form:
First, you must check out your new repository. To do so, type:
git clone firstname.lastname@example.org:username/wine.git
It is advantageous to add the WineHQ Git repository as an upstream repository, rather than the !GitHub mirror. Additionally, we will rebase with this repository to ensure that the HEAD of our repository is up to date with WineHQ. To do this, use the following commands:
git remote add upstream git://source.winehq.org/git/wine.git git pull --rebase upstream
to include the upstream (main) Wine repository as a "remote" in your fork.
Once you have made some local changes, committed them locally, and want to commit them to your !GitHub fork, use the following command:
git push origin master
Finally, to update your local branch to match your !GitHub repository, use:
git pull --rebase
Output all your patches in the directory out:
git format-patch -o out upstream/master
Save all your patches as one big file (for your own easy reference/viewing):
git log upstream/master..origin -p > /path/to/file 2>&1
Count the number of patches in your repository but not in upstream Wine (useful for keeping track of any disappearing patches):
git rev-list upstream/master..origin | wc -l
Perform an interactive rebase of your commits using git rebase:
git rebase -i upstream/master
Finally, if you would like to force commits onto !GitHub (e.g. edited patches), use:
git push -f origin master
Personally, I (MishaKoshelev) use the following each time I have some new patches that I would like to send to my !GitHub repository. This ensures that all patches are forced, and trailing whitespace is removed:
git fetch git rebase --whitespace=fix upstream/master git push -f origin master
Stacked Git is similar to Quilt, just on top of Git. It manages a stack of applied and unapplied patches on top of a Git branch. Patches can be pushed on the applied stack or popped off the applied stack onto the unapplied stack. The topmost applied patch can be edited and the stack can be rebased onto an updated branch. This makes keeping around and refining local changesets (changeset->patch) until they are applied upstream much easier. The history of changes to a patch are also kept in Git.
It can be quite hard to get a real understanding of the underlying concepts of Git, and if you are a git newb you are almost certain to run into in a situation where things break and you have no idea how to get out of the mess. At this point: don't panic! You might be tempted to just trash your repository and manually remerge your patches, but it is very probable you will run into the same situation again so you'd be better off making sure you have understood how Git works. Look through the list of tutorials at the top of this page.
Other Git repositories
|Branch URL||Branch Description|
|https://source.winehq.org/git/wine.git||Standard Wine repository|
|https://github.com/wine-compholio/wine-staging||Wine Staging repository|
|https://source.winehq.org/git/website.git||WineHQ website (https://winehq.org/)|
|https://source.winehq.org/git/tools.git||Tools used on the WineHQ website|
|https://source.winehq.org/git/appdb.git||AppDB website (https://appdb.winehq.org/)|
|https://source.winehq.org/git/bugzilla.git||Bugzilla website (https://bugs.winehq.org/)|
|http://git.etersoft.ru/projects/?p=wine/eterwine.git||Contains versions of Wine used by Etersoft for running programs under Linux|