Notice: This Wiki is now read only and edits are no longer possible. Please see: https://gitlab.eclipse.org/eclipsefdn/helpdesk/-/wikis/Wiki-shutdown-plan for the plan.
Git for Committers
Contents
Background
Motivation
DVCS make cheap branches a first class concept. When everyone has their own branch / fork it's up to the VCS to make local changeset management easy and the merge process as painless as possible.
This guide illustrates how you might use git to increase productivity when developing Eclipse. Even committers with access to the centralized repository will benefit. After all a committer can't (shouldn't!) commit work-in-progress to a repo relied on by millions of users. git makes the process of tracking and merging upstream changes painless, and allows quick and easy patch regeneration for bugzilla review.
The guide does not aim to give an exhaustive overview of the commands and how they work. There is large body of existing documentation under #Reference.
Related
Bugs
- bug 257706 Host a git repository on Eclipse Foundation servers, support git as the repository of Eclipse projects
- bug 257706#c63 Mike M. on potential problems with git at the Foundation
- bug 249745 Eclipse Repository Best Practices
Setting up git
Easy-setup
Has someone already imported the repository for you? If so clone or fork theirs! Creating the repository from scratch is straightforward, but beware the initial checkout is very time consuming. (It's also not very friendly to the Eclipse CVS servers; git has to reconstruct the individual commits and checkout each changeset individually...) (See #Known_git_repositories_for_Eclipse_projects).
Cloning a repository is as easy as:
git clone <repo_url> my_checkout_dir_name
e.g. git clone git://github.com/jamesblackburn/eclipse-core-resources-tests.git org.eclipse.core.tests.resources
Importing a repo
The process of creating a repository is the same whether you're an existing committer or not.
Tracking only your own changes
If you're only interested in your own changes and/or don't care about importing existing commit history from a legacy repository, you can just tell git to start tracking changes in a specific directory tree. Turn any directory into a versioned repository with the sequence:
Importing from CVS
git cvsimport -v -C eclipse-core-resources -d :pserver:anonymous@dev.eclipse.org:/cvsroot/eclipse -r cvs org.eclipse.core.resources
This command creates a new git repository importing all the history, in this case, from the org.eclipse.core.resources repository.
This commands says: import the org.eclipse.core.resources module into a git repository in the directory (-C) eclipse-core-resources. The -r switch says to store the loaded repository data under the cvs remote. A remote is like a git branch except you never checkout a remote directly. You'll create your own branches based on remote versions. The remote branches must be kept clean for when you next fetch updates from the remote.
With that one command you now have a fully-fledged repository waiting for you. Take a look:
cd eclipse-core-resources
git branch -a
git log
git branch -a will show all the branches -- you'll see this contains all the upstream branches, potentially a lot! You'll see that you're currently sitting in master which cvsimport has created for you tracking the cvs/master remote branch. git log shows you the full imported commit history on this branch.
Staying fresh
git cvsimport -i -v -C eclipse-core-resources -d :pserver:anonymous@dev.eclipse.org:/cvsroot/eclipse -r cvs org.eclipse.core.resources
To update your repository bringing in recent upstream changes , run the git cvsimport command again. You might want to pass the -i switch to cvsimport. This prevents git from automatically merging (pulling) changes from the remote's master into the branch you've currently checked out. If your working tree isn't clean, has diverged, or you've checked out a different branch such automatic merge may fail.
You can see changes made in a remote with git log <remote_name>/<branch_name>, and compare diffs with git diff. You can merge changes in, or if you haven't shared your branch with anyone else, rebase your branch.
Examples: Some scripts which track Eclipse/e4 core.resources and CDT are in a github repository take a look at the eclipse-sync directory and the sync.sh script. This cvsimports the projects from the Eclipse repository and pushes selected branches (HEAD and 3_4) to github.
Working with git
Creating a topic branch
git checkout -b branch-name original
This creates a new branch branch-name based on original and switches to it. NB branch-name can contain Unix style slashes. So you might have a number of feature branches bug/nnnn.
Committing
git add file_paths
git commit -a
Adds some files to the index (to be tracked).
The git index is a staging post. Files must be added initially if you want git to track them. You then need to git add the files you want to be part of your next commit. If you want to commit all outstanding changes the -a is provided.
Tree Status
Status of the tree including:
git status
Changes to be committed: changes which have been added to the index and scheduled for the next commit.
dirty files: those which have changes since the last git add.
Untracked files which aren'tignored
Committed history:
git log
Diff of dirty, tracked files in the tree to the content added to the index:
git diff
Diff between index and HEAD of the current branch -- i.e. the changes that will be committed on git commit:
git diff --cached
Switching Branch
git checkout branch-name
Generating patches for Bugzilla
git diff --no-prefix origin/cvs/HEAD > my_bugNNN.patch
This creates a patch my_bugNNN.patch against the cvs/HEAD branch in the origin remote to the HEAD of your current branch. This --no-prefix switch is needed to allow the patch to be applied by the Eclipse Apply-Patch wizard.
Sharing changes with other git'ers
Add a 'remote':
git remote add remote_name url
Fetch changes from it:
git fetch remote_name
Create a new branch from remote branch:
git checkout -b branch-name remote_name/remote_branch
... Do some stuff ...
Push changes
git push remote_name my_branch:remote_branch
... Do some stuff ...
Pull changes (git fetch followed by git merge):
git pull remote_name/branch
Git magic
Un-forking e4
The e4 repository was forked from 3.x by copying the history in the eclipse.org repository. Both repositories therefore share the same initial history, and are ideal for placing in two different branches in git.
To do this, we trick cvs-import into importing both repos into different remotes in the same git repository.
e.g. org.eclipse.core.resources:
1) import org.eclipse.core.resources:
git cvsimport -i -v -C eclipse-core-resources -d :pserver:anonymous@dev.eclipse.org:/cvsroot/eclipse -r cvs org.eclipse.core.resources
2) start to import e4 core resources:
git cvsimport -i -v -C e4-core-resources -d :pserver:anonymous@dev.eclipse.org:/cvsroot/eclipse -r cvs_e4 e4/org.eclipse.e4.resources/bundles/org.eclipse.core.resources
When you see the first commit happen during the e4 cvs-import, cancel the process.
3) cat e4-core-resources/.git/refs/remotes/cvs_e4/master
This will show the hash that the cvs_e4 master currently imported. git is essentially a UI on a hashed object store (DAG). The branches, tags, etc. are simply pointers to tree state hashes. Do read git for computer scientists -- it's neat and simple!
4) cd eclipse-core-resources
5) git log
Search for the hash id from e4. It should be there (after all both repos share the same common ancestry.
6) In git log note the hash of the first commit: first_commit_hash.
mkdir .git/refs/remotes/cvs_e4
echo first_commit_hash > .git/refs/remotes/cvs_e4/master
7) git cvsimport -i -v -C eclipse-core-resources -d :pserver:anonymous@dev.eclipse.org:/cvsroot/eclipse -r cvs_e4 e4/org.eclipse.e4.resources/bundles/org.eclipse.core.resources
When the history (tree hash) diverge, so will the e4 and 3.x branches.
From then on it's as simple as:
git cvsimport -i -v -C eclipse-core-resources -d :pserver:anonymous@dev.eclipse.org:/cvsroot/eclipse -r cvs org.eclipse.core.resources
git cvsimport -i -v -C eclipse-core-resources -d :pserver:anonymous@dev.eclipse.org:/cvsroot/eclipse -r cvs_e4 e4/org.eclipse.e4.resources/bundles/org.eclipse.core.resources
See the import scripts that track various eclipse repos.
Merging HEAD changes to e4
With Eclipse HEAD and e4 in the same repository, git can be used to do most the heavy merg lifting.
Taking org.eclipse.ui.ide as an example:
mkdir 3.5m6-merge/
Open Eclipse using with workspace 3.5m6-merge as workspace...
git clone git://github.com/jamesblackburn/eclipse-ui-ide.git org.eclipse.ui.ide
cd org.eclipse.ui.ide
Eclipse: Import Project org.eclipse.ui.ide into workspace...
git checkout -b merge_e4 origin/cvs_e4/HEAD
git merge origin/cvs/HEAD
... In Eclipse: Refresh and resolve conflicts by...
git diff <filename> <== will show the hunks requiring merge
git log --merge -p <filename> <== will show the commits which touch file (commit id / comment + diff)
git add <merged_file>
git status <== lists no more merge failures
git commit
Produce diff for upstream bugzill: git diff --no-prefix origin/cvs_e4/HEAD > merge_diff.patch
NB having performed a resource changing operation externally, remember to Refresh in Eclipse => Right click on Project > Refresh. Check for build errors.
Reference
Help for other VCS refugees
Using the EGit integration
Contributing to EGit or JGit
If you want to get involved with EGit or JGit, please see the EGit Contributor Guide.
Loading it into Eclipse is as simple as importing the existing projects from the cloned repository.
Known git repositories for Eclipse projects
- Eclipse.org Hosted Git Repositories: http://git.eclipse.org/c/
- Read-only git of all Eclipse CVS: See http://dev.eclipse.org/git/ (Projects that have migrated to Git will not be current here)
- Swordfish Documentation: GIT Repository
- SWTBot: http://github.com/ketan/swtbot/
- BPMN Modeler http://github.com/atoulme/bpmn-modeler/