Table of Contents
- Git Users
- Browse the GIT Repository Online
- Using the GIT Repository
- Making Changes
- Working with Branches
- Viewing Changes
- Reverting Changes
- git reset
- git revert
- Merging Changes
- Accessing a developer's repository
- Creating a Patch
- Submitting a Patch
- Configuring git-send-email to use GMail
- Sending Email
- Manage Your Code
- Private Servers
- Learn more about Git
TODO: Managing a (private/public) git mirror, using github, submitting pull requests...
Browse the GIT Repository Online
You can browse all available repositories online by accessing http://git.rtems.org/.
Using the GIT Repository
The following examples demonstrate how to use the RTEMS' git repos. These examples are provided for the main rtems module, but they are also valid for the other modules.
First we need to obtain our own local copy of the RTEMS git repository:
git clone git://git.rtems.org/rtems.git rtems
This command will create a folder named rtems in the current directory. This folder will contain a full featured rtems' git repository and the current HEAD revision checked out. Since all the history is available we can checkout any release of RTEMS. Major RTEMS releases are available as separate branches in the repo.
To see all available remote branches issue the following command:
git branch -r
We can checkout one of those remote branches (e.g. rtems-4.10 branch) using the command:
git checkout -b rtems410 origin/4.10
This will create a local branch named "rtems410", containing the rtems-4.10 release, that will track the remote branch "rtems-4-10-branch" in origin (git://git.rtems.org/rtems.git). The 'git branch' command prints a list of the current local branches, indicating the one currently checked out.
If you want to switch between local branches:
git checkout <branch-name>
With time your local repository will diverge from the main RTEMS repository. To keep your local copy up to date you need to issue:
git pull origin
This command will update all your local branches with any new code revisions available on the central repository.
Git allows you to make changes in the RTEMS source tree and track those changes locally. First you need to clone the repository:
git clone git://git.rtems.org/rtems.git rtems
Or if you already cloned it before, then you might want to update to the latest version before making your changes:
cd rtems git pull
Next make your changes to files. If you add, delete or move/rename files you need to inform Git
git add /some/new/file git rm /some/old/file git mv /some/old/file /some/new/file
When you're satisfied with the changes you made, commit them (locally)
git commit -a
The -a flag commits all the changes that were made, but you can also control which changes to commit by individually adding files as you modify them by using. You can also specify other options to commit, such as a message with the -m flag.
git add /some/changed/files git commit
But you shouldn't do any of this on the branch that you cloned, instead you should create a new branch to work with, so that the cloned branch (master) is a pristine copy of the upstream RTEMS repository.
Working with Branches
Branches facilitate trying out new code and creating patches.
Previous release of RTEMS are available through remote branches. To check out a remote branch, first query the Git repository for the list of branches:
git branch -r
Then checkout the desired remote branch, for example:
git checkout -b rtems410 origin/4.10
Or if you have previously checked out the remote branch then you should see it in your local branches:
You can change to an existing local branch easily:
git checkout rtems410
You can also create a new branch and switch to it:
git branch temporary git checkout temporary
Or more concisely:
git checkout -b temporary
If you forget which branch you are on
shows you by placing a * next to the current one.
When a branch is no longer useful you can delete it.
git checkout master git branch -d temporary
If you have unmerged changes in the old branch git complains and you need to use -D instead of -d.
To view all changes since the last commit:
git diff HEAD
To view all changes between the current branch and another branch, say master:
git diff master..HEAD
To view descriptions of committed changes:
Or view the changeset for some file (or directory):
git log /some/file
To view the changesets made between two branches:
git log master..HEAD
Or for a more brief description use shortlog:
git shortlog master..HEAD
To remove all (uncommitted) changes on a branch
git checkout -f
Or to selectively revert (uncommited) files, for example if you accidentally deleted ./some/file
git checkout -- ./some/file
git checkout HEAD ./some/file
To remove commits there are two useful options, reset and revert. git reset should only be used on local branches that no one else is accessing remotely. git revert is cleaner, and is the right way to revert changes that have already been pushed/pulled remotely.
git reset is a powerful and tricky command that should only be used on local (un-pushed) branches): A good description of what it enables to do can be found here. The following are a few useful examples. Note that adding a ~ after HEAD refers to the most recent commit, and you can add a number after the ~ to refer to commits even further back; HEAD by itself refers to the current working directory (changes since last commit).
git reset HEAD~
Will undo the last commit and unstage those changes. Your working directory will remain the same, therefore a "git status" will yield any changes you made plus the changes made in your last commit. This can be used to fix the last commit. You will need to add the files again.
git reset --soft HEAD~
Will just undo the last commit. The changes from last commit will still be staged (just as if you finished git adding them). This can be used to amend the last commit (e.g. I forgot to add a file to the last commit).
git reset --hard HEAD~
Will revert everything, including the working directory, to the previous commit. This is dangerous and can lead to you losing all your changes; the --hard flag ignores errors.
git reset HEAD
Will unstage any change. This is used to revert a wrong git add. (e.g. I added a file that shouldn't be there, but I haven't 'committed')
git reset --hard HEAD
Will revert your working directory to a HEAD state. You will lose any change you made to files after the last commit. This is used when you just want to destroy all changes you made since the last commit.
git revert does the same as reset but creates a new commit with the reverted changes instead of modifying the local repository directly.
git revert HEAD
This will create a new commit which undoes the change in HEAD. You will be given a chance to edit the commit message for the new commit.
Suppose you commit changes in two different branches, branch1 and branch2, and want to create a new branch containing both sets of changes:
git checkout -b merged git merge branch1 git merge branch2
Or you might want to bring the changes in one branch into the other:
git checkout branch1 git merge branch2
And now that branch2 is merged you might get rid of it:
git branch -d branch2
If you have done work on a branch, say branch1, and have gone out-of-sync with the remote repository, you can pull the changes from the remote repo and then merge them into your branch:
git checkout master git pull git checkout branch1 git merge master
If all goes well the new commits you pulled into your master branch will be merged into your branch1, which will now be up-to-date. However, if branch1 has not been pushed remotely then rebasing might be a good alternative to merging because the merge generates a commit
An alternative to the merge command is rebase, which replays the changes (commits) on one branch onto another. git rebase finds the common ancestor of the two branches, stores each commit of the branch you’re on to temporary files, and applies each commit in order.
git checkout branch1 git rebase master
or more concisely
git rebase master branch1
will bring the changes of master into branch1, and then you can fast-forward master to include branch1 quite easily
git checkout master git merge branch1
Rebasing makes a cleaner history than merging; the log of a rebased branch looks like a linear history as if the work was done serially rather than in parallel. A primary reason to rebase is to ensure commits apply cleanly on a remote branch, e.g. when submitting patches to RTEMS that you create by working on a branch in a personal repository. Using rebase to merge your work with the remote branch eliminates most integration work for the committer/maintainer.
There is one caveat to using rebase: Do not rebase commits that you have pushed to a public repository. Rebase abandons existing commits and creates new ones that are similar but different. If you push commits that others pull down, and then you rewrite those commits with git rebase and push them up again, the others will have to re-merge their work and trying to integrate their work into yours can become messy.
Accessing a developer's repository
Creating a Patch
The recommended way to create a patch is to branch the git repository master and use one commit for each logical change. Then you can use git-format-patch to turn your commits into patches and easily submit them.
git format-patch master
Creates a separate patch for each commit that has been made between the master branch and the current branch and writes them in the current directory. Use the -o flag to redirect the files to a different directory.
If you are re-submitting a patch that has previously been reviewed, you should specify a version number for your patch, for example use
git format-patch -v2 ...
to indicate the second version of a patch, v3 for a third, and so forth.
Patches created using git-format-patch are formatted so they can be emailed and rely on having git configured with your name and email address, for example
git config --global user.name "Your Name" git config --global user.email firstname.lastname@example.org
Submitting a Patch
Using git-send-email you can easily contribute your patches. You will need to install git-send-email first:
yum install git-email
apt-get install git-send-email
Then you will need to configure an SMTP server. You could install one on your localhost, or you can connect to a mail server such as gmail.
Configuring git-send-email to use GMail
Configure git to use gmail:
git config --global sendemail.smtpserver smtp.gmail.com git config --global sendemail.smtpserverport 587 git config --global sendemail.smtpencryption tls git config --global sendemail.smtpuser email@example.com
It will ask for your password each time you use git-send-email. Optionally you can also put it in your git config:
git config --global sendemail.smtppass your_password
To send your patches just
git send-email /path/to/patch --to firstname.lastname@example.org
To send multiple related patches (if you have more than one commit in your branch) specify a path to a directory containing all of the patches created by git-format-patch. git-send-email has some useful options such as:
- --annotate to show/edit your patch
- --cover-letter to prepend a summary
- --cc=<address> to cc someone
You can configure the to address:
git config --global sendemail.to email@example.com
So all you need is:
git send-email /path/to/patch
Some restrictive corporate firewalls block access through the git protocol (git://) If you are unable to reach the server git://git.rtems.org/ you can try accessing through http. To clone the rtems repository using the http protocol use the following command:
git clone http://git.rtems.org/rtems/ rtems
This access through http is slower (way slower!) than through the git protocol therefore the git protocol is preferred.
Manage Your Code
You may prefer to keep your application and development work in a git repository for all the good reasons that come with version control. For public repositories, you may like to try GitHub or BitBucket. RTEMS maintains mirrors on GitHub?, which can make synchronizing with upstream changes relatively simple. If you need to keep your work private, you can use one of those services with private repositories or manage your own server. The details of setting up a server are outside the scope of this document, but if you have a server with ssh access you should be able to find instructions on how to set up git access. Once you have git configured on the server, adding repositories is a snap.
In the following, replace @USER@ with your username on your server, @REPO@ with the name of your repository, and @SERVER@ with your server's name or address.
To push a mirror to your private server, first create a bare repository on your server.
cd /home/@USER@ mkdir git mkdir git/@REPO@.git cd git/@REPO@.git git --bare init
Now from your client machine (e.g. your work laptop/desktop), push a git, perhaps one you cloned from elsewhere, or one that you made locally with 'git init', by adding a remote and pushing:
git remote add @SERVER@ ssh://@SERVER@/home/@USER@/git/@REPO@.git git push @SERVER@ master
You can replace the @SERVER@ with another name for your remote if you like. And now you can push other branches that you might have created. Now you can push and pull between your client and your server. Use SSH keys to authenticate with your server if you want to save on password typing; remember to put a passphrase on your SSH key if there is a risk the private key file might get compromised.
The following is an example scenario that might be useful for RTEMS users that uses a slightly different approach than the one just outlined:
ssh @SERVER@ mkdir git git clone --mirror git://git.rtems.org/rtems.git ## Add your ssh key to ~/.ssh/authorized_keys exit git clone ssh://@SERVER@/home/@USER@/git/rtems.git cd rtems git remote add upstream git://git.rtems.org/rtems.git git fetch upstream git pull upstream master git push ## If you want to track RTEMS on your personal master branch, ## you should only push changes to origin/master that you pull ## from upstream. The basic workflow should look something like: git checkout master git pull upstream master git push git checkout -b anewbranch ## Repeat: do work, git commit -a git push origin anewbranch ## delete a remote branch git push origin :anewbranch ## delete a local branch git branch -d anewbranch
Learn more about Git
Links to site with good Git information.