Committing code

This section is addressed to the committers and to anyone interested in knowing how code gets committed into Django. If you’re a community member who wants to contribute code to Django, look at Working with Git and GitHub instead.

Handling pull requests

Since Django is now hosted at GitHub, most patches are provided in the form of pull requests.

When committing a pull request, make sure each individual commit matches the commit guidelines described below. Contributors are expected to provide the best pull requests possible. In practice however, committers - who will likely be more familiar with the commit guidelines - may decide to bring a commit up to standard themselves.

You may want to have Jenkins test the pull request with one of the pull request builders that doesn’t run automatically, such as Oracle or Selenium. See the Jenkins wiki page for instructions.

An easy way to checkout a pull request locally is to add an alias to your ~/.gitconfig (upstream is assumed to be django/django):

[alias]
    pr = !sh -c \"git fetch upstream pull/${1}/head:pr/${1} && git checkout pr/${1}\"

Now you can simply run git pr #### to checkout the corresponding pull request.

At this point, you can work on the code. Use git rebase -i and git commit --amend to make sure the commits have the expected level of quality. Once you’re ready:

$ # Pull in the latest changes from master.
$ git checkout master
$ git pull upstream master
$ # Rebase the pull request on master.
$ git checkout pr/####
$ git rebase master
$ git checkout master
$ # Merge the work as "fast-forward" to master to avoid a merge commit.
$ # (in practice, you can omit "--ff-only" since you just rebased)
$ git merge --ff-only pr/XXXX
$ # If you're not sure if you did things correctly, check that only the
$ # changes you expect will be pushed to upstream.
$ git push --dry-run upstream master
$ # Push!
$ git push upstream master
$ # Delete the pull request branch.
$ git branch -d pr/xxxx

...\> REM Pull in the latest changes from master.
...\> git checkout master
...\> git pull upstream master
...\> REM Rebase the pull request on master.
...\> git checkout pr/####
...\> git rebase master
...\> git checkout master
...\> REM Merge the work as "fast-forward" to master to avoid a merge commit.
...\> REM (in practice, you can omit "--ff-only" since you just rebased)
...\> git merge --ff-only pr/XXXX
...\> REM If you're not sure if you did things correctly, check that only the
...\> REM changes you expect will be pushed to upstream.
...\> git push --dry-run upstream master
...\> REM Push!
...\> git push upstream master
...\> REM Delete the pull request branch.
...\> git branch -d pr/xxxx

Force push to the branch after rebasing on master but before merging and pushing to upstream. This allows the commit hashes on master and the branch to match which automatically closes the pull request.

If a pull request doesn’t need to be merged as multiple commits, you can use GitHub’s “Squash and merge” button on the website. Edit the commit message as needed to conform to the guidelines and remove the pull request number that’s automatically appended to the message’s first line.

When rewriting the commit history of a pull request, the goal is to make Django’s commit history as usable as possible:

  • If a patch contains back-and-forth commits, then rewrite those into one. For example, if a commit adds some code and a second commit fixes stylistic issues introduced in the first commit, those commits should be squashed before merging.
  • Separate changes to different commits by logical grouping: if you do a stylistic cleanup at the same time as you do other changes to a file, separating the changes into two different commits will make reviewing history easier.
  • Beware of merges of upstream branches in the pull requests.
  • Tests should pass and docs should build after each commit. Neither the tests nor the docs should emit warnings.
  • Trivial and small patches usually are best done in one commit. Medium to large work may be split into multiple commits if it makes sense.

Practicality beats purity, so it is up to each committer to decide how much history mangling to do for a pull request. The main points are engaging the community, getting work done, and having a usable commit history.

Committing guidelines

In addition, please follow the following guidelines when committing code to Django’s Git repository:

  • Never change the published history of django/django branches by force pushing. If you absolutely must (for security reasons for example), first discuss the situation with the team.

  • For any medium-to-big changes, where “medium-to-big” is according to your judgment, please bring things up on the django-developers mailing list before making the change.

    If you bring something up on django-developers and nobody responds, please don’t take that to mean your idea is great and should be implemented immediately because nobody contested it. Everyone doesn’t always have a lot of time to read mailing list discussions immediately, so you may have to wait a couple of days before getting a response.

  • Write detailed commit messages in the past tense, not present tense.

    • Good: “Fixed Unicode bug in RSS API.”
    • Bad: “Fixes Unicode bug in RSS API.”
    • Bad: “Fixing Unicode bug in RSS API.”

    The commit message should be in lines of 72 chars maximum. There should be a subject line, separated by a blank line and then paragraphs of 72 char lines. The limits are soft. For the subject line, shorter is better. In the body of the commit message more detail is better than less:

    Fixed #18307 -- Added git workflow guidelines.

Refactored the Django's documentation to remove mentions of SVN
specific tasks. Added guidelines of how to use Git, GitHub, and
how to use pull request together with Trac instead.

    If the patch wasn’t a pull request, you should credit the contributors in the commit message: “Thanks A for report, B for the patch and C for the review.”

  • For commits to a branch, prefix the commit message with the branch name. For example: “[1.4.x] Fixed #xxxxx – Added support for mind reading.”

  • Limit commits to the most granular change that makes sense. This means, use frequent small commits rather than infrequent large commits. For example, if implementing feature X requires a small change to library Y, first commit the change to library Y, then commit feature X in a separate commit. This goes a long way in helping everyone follow your changes.

  • Separate bug fixes from feature changes. Bugfixes may need to be backported to the stable branch, according to Supported versions.

  • If your commit closes a ticket in the Django ticket tracker, begin your commit message with the text “Fixed #xxxxx”, where “xxxxx” is the number of the ticket your commit fixes. Example: “Fixed #123 – Added whizbang feature.”. We’ve rigged Trac so that any commit message in that format will automatically close the referenced ticket and post a comment to it with the full commit message.

    For the curious, we’re using a Trac plugin for this.

Note

Note that the Trac integration doesn’t know anything about pull requests. So if you try to close a pull request with the phrase “closes #400” in your commit message, GitHub will close the pull request, but the Trac plugin will also close the same numbered ticket in Trac.

  • If your commit references a ticket in the Django ticket tracker but does not close the ticket, include the phrase “Refs #xxxxx”, where “xxxxx” is the number of the ticket your commit references. This will automatically post a comment to the appropriate ticket.

  • Write commit messages for backports using this pattern:

    [<Django version>] Fixed <ticket> -- <description>

Backport of <revision> from <branch>.

    For example:

    [1.3.x] Fixed #17028 -- Changed diveintopython.org -> diveintopython.net.

Backport of 80c0cbf1c97047daed2c5b41b296bbc56fe1d7e3 from master.

    There’s a script on the wiki to automate this.

    If the commit fixes a regression, include this in the commit message:

    Regression in 6ecccad711b52f9273b1acb07a57d3f806e93928.

    (use the commit hash where the regression was introduced).

Reverting commits

Nobody’s perfect; mistakes will be committed.

But try very hard to ensure that mistakes don’t happen. Just because we have a reversion policy doesn’t relax your responsibility to aim for the highest quality possible. Really: double-check your work, or have it checked by another committer, before you commit it in the first place!

When a mistaken commit is discovered, please follow these guidelines:

  • If possible, have the original author revert their own commit.
  • Don’t revert another author’s changes without permission from the original author.
  • Use git revert – this will make a reverse commit, but the original commit will still be part of the commit history.
  • If the original author can’t be reached (within a reasonable amount of time – a day or so) and the problem is severe – crashing bug, major test failures, etc. – then ask for objections on the django-developers mailing list then revert if there are none.
  • If the problem is small (a feature commit after feature freeze, say), wait it out.
  • If there’s a disagreement between the committer and the reverter-to-be then try to work it out on the django-developers mailing list. If an agreement can’t be reached then it should be put to a vote.
  • If the commit introduced a confirmed, disclosed security vulnerability then the commit may be reverted immediately without permission from anyone.
  • The release branch maintainer may back out commits to the release branch without permission if the commit breaks the release branch.
  • If you mistakenly push a topic branch to django/django, just delete it. For instance, if you did: git push upstream feature_antigravity, just do a reverse push: git push upstream :feature_antigravity.
« previous | up | next »