User Tools

Site Tools


devel:git-commit-guidelines

GIT Commit Guidelines

GIT Repository Location

Starting with December 15, 2014, the main GIT repository of Kamailio project is hosted on Github at:

A read-only mirror is maintained at:

Cloning GIT Repository

You need to install the git command line tool, part of git-core (on Debian/Ubuntu) or git (Fedora/CentOS) package:

apt-get install git-core

Create the folder where you want to clone the repository, for example:

mkdir ~/kamailio
cd ~/kamailio

If you have developer write access to GIT repository, clone the repository with:

git clone https://github.com/kamailio/kamailio.git kamailio-dev

or if you prefer via ssh:

git clone ssh+git://git@github.com/kamailio/kamailio.git kamailio-dev

You may be asked for your Github username and password.

The clone is tracking automatically the remote master branch (the development version).

Working with Github forks of a GIT Repository

If you don't have developer write access, you can fork the Kamailio repository. Then you checkout the forked repository to your local disk and do the changes. Commit the changes to your repository and create a pull request to integrate your changes into the Kamailio repository back. Have a look e.g. to this documentation:

Github help - forking

Github help - pull request

Read-Only Cloning of GIT Repository

Working with a fork of the Kamailio repository and then submitting a pull request is nowadays the preferred way to integrate your contributions, if you don't have write permissions to the Kamailio repository. Please refer to the paragraph above.

If you don't have developer write access, you can clone read-only the repository with:

git clone https://github.com/kamailio/kamailio.git kamailio-dev

You can also clone (read-only, not for developers that want to push back commits) from the real-time mirror server:

git clone --depth 1 git://git.kamailio.org/kamailio kamailio-dev

When cloning in read only mode, you can commit your changes to the local clone, but you cannot push the patches to the remote repository. Export the patches and attach them to an email to mailing list <sr-dev [at] lists.sip-router.org> or to an issue on bug tracker. A developer can pick them and commit in your behalf.

You can export the local commit with git format-patch, for example, exporting last commit:

git format-patch -1

The name of the patch file is printed in the terminal and the file is saved in current directory.

GIT Commit Guidelines to Kamailio Repository

Not an absolute MUST, but it SHOULD be used almost always. It helps to track changes and follow the development easier (better commit mail messages and ChangeLogs are easier to generate).

Where to commit

Note: there is no list of owners and the code parts, the guidelines here are not to prevent contributions to any parts of code, but to keep the coherence of the code and sanity of development. As generic rule, if you are not the initial author of a component, then someone else is the maintainer. If you don't know who is that developer, just ask on mailing list sr-dev [at] lists.sip-router.org.

Changes to your part of code

Whenever possible, commit first into master branch and then backport the patch - a recommended way is to cherry-pick the patch by commit id from master branch in your target branch.

By committing first into master branch, we ensure that the development version is always the most up-to-date. Also, people can review and test the commit from master branch before it is backported to a stable branch.

Changes to other parts of code

If it is a major change, then it is recommended to commit first into a personal or tmp branch. Then send an announcement to sr-dev mailing list, describing shortly the changes and where they are located. In this way, first, we avoid development conflicts between main (other) author(s) of the code and the other developers. Second, other developers can review or test, giving feedback or suggesting improvements or alternatives.

In some cases, the main author of the code may provide a different implementation, for the same purpose, that will fit better in existing architecture or future plans for that component. Very often, the contributions will be just merged in the main devel branch.

TMP and Personal branches

Each developer has access to a personal branch on GIT remote repository hosted on sip-router.org. The name of the branch starts with the GIT username of developer, followed by a forward slash and then the effective branch name. For example:

daniel/tm-updates

Only the respective developer can do commit in such branch.

If many developers need to work on a branch, it can be created in the tmp/ space:

tmp/tm-changes

Commit message

Please create the commit messages following the GIT convention:

  • start with one short line, preferably less then 50 chars summarizing the changes (this is referred later as “first line of the commit message”)
  • then one empty line
  • then a more detailed description

Think of the first line as of an email “Subject” line. In fact it will be used as “Subject” in the generated commit emails and it will also be used when generating the Changelog (e.g. git log –pretty=oneline).

Please start always with the prefix of the subsystem that is modified by the commit, for example:

  • core: typo fixes
  • tcp: stun fixes
  • mem: faster malloc
  • module_name: support for foo
  • lib_name: critical bug fix

Examples of commit messages:

  • change to usrloc module from modules
usrloc: fixed name conflict

- destroy_avps() renamed to reg_destroy_avps() to avoid conflicts
  with the usr_avp.h version.
  • change to core
core: loadpath can now use a list of directories

- loadpath can use a list of directories separated by ':',
  e.g.: loadpath "modules:modules_s:modules_k".
  First match wins (e.g. for loadmodule "textops" if
  modules/textops.so or modules/textops/textops.so exists, it will
  be loaded and the search will stop).

See:

Content of the Commit Message

  • the message must be addressed for humans, using common sense language. All commit messages are part of the ChangeLog file for each release and that file is read by sysadmins and other non-developer people
  • the message must be strictly related to the changes in the C code (e.g., don't give instructions what another developer must do with your patch)
  • credits to persons or companies that helped in the context of the commit are allowed and it is fair to be added whenever is the case. The place is at the end of the commit message (or end of the part describing the related change)
  • don't use slang and don't abuse using abbreviations, especially when it is not an official abbreviation
  • don't use system or tools commands to express the changes of the commit. They can be given as an example in the detailed message to show how the changes were done, but text explanation must be first. E.g., if you renamed a function, don't use as commit message 's/oldfunc/newfunc/'.

Generic Guidelines

The suggestions here target to make easier to track the changes and do the backporting to stable branches whenever is the case:

  • commit big changes across many parts of the code in smaller chunks that affect only one part. Examples of major code parts: parser, transport layers, config interpreter, memory manager, libraries, modules, …
  • do the commits per one module when changing in a module and other modules are affected and were updates as well
  • when updating the documentation and it is intended to do a backport, split the commit in one for xml files and one for the README file. Usually the commit to xml files can be backported without conflict (with git cherry-pick …). README files are many times generated with different character sets, a matter of the system where they are generated, resulting in backporting conflicts. Thus is better to backport only the commit for xml files and generate the README for that branch, then push it to remote GIT repository
    • IMPORTANT: the README files for modules must no longer be generated by the developer and pushed to GIT. There is a tool running on server auto-generating the README file shortly after the xml files of the module's docs are updated and pushed to GIT repository
  • if it is a change that has to be repeated in many places, then the commit can be done at once to all changes. For example, renaming a token that is present in all docbook files, removing or adding a new field in a structure such as module exports.

Splitting commits

When you finish doing the changes, you can see affected files with:

git status

If you are in the root of source tree, you can commit all changes with:

git commit .

In case you added new files, first you have to add them with:

git add path/to/file

To split the commits, you can give the list of files as argument to git commit command:

git commit path/to/file1 /path/to/file2

You can commit the changes in a directory with:

git commit path/to/directory/

For example, commit the changes for module auth:

git commit modules/auth/

Core Commits

  • prefix the first line of the commit message with “core:”, or with the subsystem prefix (e.g. “tcp:”, “dns:”, “mem:”, “script:”, a.s.o).

Module Commits

  • prefix the first line of the commit message with “module name:”
  • if there are modules with same name in different directories, use a distinctive marker to give a hint about which one was updated. Recommended markers are:
    • (k) - for modules located in modules_k
    • (s) - for modules located in modules_s

Example commit message to auth module:

auth: improved documentation for www_challenge()

Library Commits

  • prefix the first line of the commit message with “library name:”

Example commit message:

lib/srutils: more detailed log message for uuid generation

Backporting Commits

Commits that include fixes to issues affecting stable branches must be backported using:

git cherry-pick -x _COMMIT_HASH_

Example:

git cherry-pick -x a8bc3d

Parameter -x is important to be able to keep track easier of backported commits.

The git cherry-pick command must be executed inside the branch where the commit has to be backported, after doing a git pull origin.

Note: the commit must be done first time in master, then backported to the latest stable branch, using cherry-pick -x with commit id from master. The backported commit gets a new id in the latest stable branch, that commit id must be used when cherry-picking to the previous stable branch. When willing to backport to older versions, keep using the commit id from the next stable branch.

Useful Commands

This section collects some useful GIT commands.

Reset Last Local Commit

  • hard reset - purge the commit (warning: changes done for the commit are lost):
git reset --hard HEAD~1
  • commit reset - drop the last commit, without losing the changes in the files:
git reset HEAD~1
  • soft reset - leave the last commit and changes in the files – next commit command will do the same commit
git reset --soft HEAD~1

Checkout Remote Branch

git checkout -b local-branch-name origin/remote-branch-name

Abort File Changes

git checkout -- file-name

Modify Previous Commit Message

git commit --amend

Partial Change Checkin

git add --edit

Undo Previous Commit

git revert HEAD^

Delete Remote Branch

git push origin :<branchName>

As of Git v1.7.0, a remote branch can be deleted using:

git push origin --delete <branchName>

Revert already pushed commit

This will produce a commit reverting the previous commit <hash>

git revert <hash>
git push origin

Rename and commit file

Create minimal commit when renaming file

git mv module.c module_mod.c
git commit -m '...: renamed module file to match recommended pattern' module.c module_mod.c
devel/git-commit-guidelines.txt · Last modified: 2018/10/24 22:33 by henningw