This section is addressed to the mergers and to anyone interested in knowing\nhow code gets committed into Django. The Committing guidelines apply\nto all contributors, with or without commit rights.
\nIf you’re a community member who wants to contribute code to Django, look at\nWorking with Git and GitHub instead.
\nSince Django is hosted on GitHub, patches are provided in the form of pull\nrequests.
\nWhen committing a pull request, make sure each individual commit matches the\ncommit guidelines described below. Contributors are expected to provide the\nbest pull requests possible. In practice mergers - who will likely be more\nfamiliar with the commit guidelines - may decide to bring a commit up to\nstandard themselves.
\nYou may want to have Jenkins or GitHub actions test the pull request with one\nof the pull request builders that doesn’t run automatically, such as Oracle or\nSelenium. See the CI wiki page for instructions.
\nIf you find yourself checking out pull requests locally more often, this git\nalias will be helpful:
\n[alias]\n pr = !sh -c \\"git fetch upstream pull/${1}/head:pr/${1} && git checkout pr/${1}\\"\nAdd it to your ~/.gitconfig, and set upstream to be django/django.\nThen you can run git pr #### to checkout the corresponding pull request.
At this point, you can work on the code. Use git rebase -i and git\ncommit --amend to make sure the commits have the expected level of quality.\nOnce you’re ready:
Linux / macOS
$ # Pull in the latest changes from main.\n$ git checkout main\n$ git pull upstream main\n$ # Rebase the pull request on main.\n$ git checkout pr/####\n$ git rebase main\n$ git checkout main\n$ # Merge the work as "fast-forward" to main to avoid a merge commit.\n$ # (in practice, you can omit "--ff-only" since you just rebased)\n$ git merge --ff-only pr/XXXX\n$ # If you're not sure if you did things correctly, check that only the\n$ # changes you expect will be pushed to upstream.\n$ git push --dry-run upstream main\n$ # Push!\n$ git push upstream main\n$ # Delete the pull request branch.\n$ git branch -d pr/xxxx\nWindows
...\\> REM Pull in the latest changes from main.\n...\\> git checkout main\n...\\> git pull upstream main\n...\\> REM Rebase the pull request on main.\n...\\> git checkout pr/####\n...\\> git rebase main\n...\\> git checkout main\n...\\> REM Merge the work as "fast-forward" to main to avoid a merge commit.\n...\\> REM (in practice, you can omit "--ff-only" since you just rebased)\n...\\> git merge --ff-only pr/XXXX\n...\\> REM If you're not sure if you did things correctly, check that only the\n...\\> REM changes you expect will be pushed to upstream.\n...\\> git push --dry-run upstream main\n...\\> REM Push!\n...\\> git push upstream main\n...\\> REM Delete the pull request branch.\n...\\> git branch -d pr/xxxx\nForce push to the branch after rebasing on main but before merging and pushing\nto upstream. This allows the commit hashes on main and the branch to match\nwhich automatically closes the pull request.
\nIf a pull request doesn’t need to be merged as multiple commits, you can use\nGitHub’s «Squash and merge» button on the website. Edit the commit message as\nneeded to conform to the guidelines and remove\nthe pull request number that’s automatically appended to the message’s first\nline.
\nWhen rewriting the commit history of a pull request, the goal is to make\nDjango’s commit history as usable as possible:
\nIf a patch contains back-and-forth commits, then rewrite those into one.\nFor example, if a commit adds some code and a second commit fixes stylistic\nissues introduced in the first commit, those commits should be squashed\nbefore merging.
Separate changes to different commits by logical grouping: if you do a\nstylistic cleanup at the same time as you do other changes to a file,\nseparating the changes into two different commits will make reviewing\nhistory easier.
Beware of merges of upstream branches in the pull requests.
Tests should pass and docs should build after each commit. Neither the\ntests nor the docs should emit warnings.
Trivial and small patches usually are best done in one commit. Medium to\nlarge work may be split into multiple commits if it makes sense.
Practicality beats purity, so it is up to each merger to decide how much\nhistory mangling to do for a pull request. The main points are engaging the\ncommunity, getting work done, and having a usable commit history.
\nThese guidelines apply to all commits to Django’s Git repository, whether\nsubmitted by a contributor via a pull request or landed directly by a merger:
\nNever change the published history of django/django branches by force\npushing. If you absolutely must (for security reasons for example), first\ndiscuss the situation with the team.
For any medium-to-big changes, where «medium-to-big» is according to\nyour judgment, please bring things up on the Django Forum before making\nthe change.
\nIf you bring something up and nobody responds, please don’t take that\nto mean your idea is great and should be implemented immediately because\nnobody contested it. Everyone doesn’t always have a lot of time to read\ndiscussions immediately, so you may have to wait a couple of days before\ngetting a response.
\nWrite detailed commit messages in the past tense, not present tense, and\nend the subject line with a period.
\nCorrect: «Fixed Unicode bug in RSS API.»
Incorrect: «Fixes Unicode bug in RSS API.» (present tense)
Incorrect: «Fixing Unicode bug in RSS API.» («-ing» form)
Incorrect: «Fixed Unicode bug in RSS API» (missing trailing period)
The commit message should be in lines of 72 chars maximum. There should be\na subject line, separated by a blank line and then paragraphs of 72 char\nlines. The limits are soft. For the subject line, shorter is better.
\nIn the body of the commit message more detail is better than less, and should\nexplain why the change was made, not what was changed or how. The code\nitself shows what changed; the commit message should provide the context and\nreasoning that the code cannot.
\nCredit the contributors in the commit message: «Thanks A for the report and B\nfor review.» Use git’s Co-Authored-By as appropriate, including anyone\nwhose earlier work the change builds on.
\nFor example:
\nFixed #18307 -- Added git workflow guidelines.\n\nRefactored the Django's documentation to remove mentions of SVN\nspecific tasks. Added guidelines of how to use Git, GitHub, and\nhow to use pull request together with Trac instead.\n\nThanks to Full Name for the report, and to Reviewer for reviews.\nLimit commits to the most granular change that makes sense. This means,\nuse frequent small commits rather than infrequent large commits. For\nexample, if implementing feature X requires a small change to library Y,\nfirst commit the change to library Y, then commit feature X in a separate\ncommit. This goes a long way in helping everyone follow your changes.
Separate bug fixes from feature changes. Bugfixes may need to be\nbackported.
If your commit closes a ticket in the Django ticket tracker, begin\nyour commit message with the text «Fixed #xxxxx», where «xxxxx» is the\nnumber of the ticket your commit fixes. Example: «Fixed #123 – Added\nwhizbang feature.». We’ve rigged Trac so that any commit message in that\nformat will automatically close the referenced ticket and post a comment\nto it with the full commit message.
\nFor the curious, we’re using a Trac plugin for this.
\nIf your commit references a ticket in the Django ticket tracker but\ndoes not close the ticket, include the phrase «Refs #xxxxx», where «xxxxx»\nis the number of the ticket your commit references. This will automatically\npost a comment to the appropriate ticket.
Bug fix backports to stable branches are done exclusively by mergers, following\nthe Supported versions. A backport consists of cherry-picking a\ncommit from main onto the target stable/A.B.x branch.
A backport commit must include two things beyond the original commit message:
\nA prefix [A.B.x] on the subject line, where A.B.x is the name of the\nstable/A.B.x branch being targeted.
A suffix Backport of <sha> from main. line in the commit body, pointing\nto the original commit hash.
For example:
\n[1.3.x] Fixed #17028 -- Changed diveintopython.org -> diveintopython.net.\n\nBackport of 80c0cbf1c97047daed2c5b41b296bbc56fe1d7e3 from main.\nIf the backport also fixes a regression, add a line identifying the commit that\nintroduced it:
\nRegression in 6ecccad711b52f9273b1acb07a57d3f806e93928.\nThere are three ways to do a backport, depending on your workflow:
\nOption 1: fully manual
\nCherry-pick the commit onto the stable branch, then amend the commit message to\nadd the required prefix and backport note:
\ngit cherry-pick <sha>\ngit commit --amend\nIf the cherry-pick produces conflicts, resolve them, stage the changes, then\nrun git cherry-pick --continue before amending. This option requires no\nsetup but relies entirely on remembering to format the message correctly.
Option 2: using the helper script in the repo
\nThe scripts/backport.sh Bash script automates the cherry-pick and rewrites\nthe commit message with the correct prefix and backport note. Run it from the\ntarget stable branch:
bash scripts/backport.sh <sha>\nThis is straightforward for clean cherry-picks, but if conflicts are produced,\nyou will need to resolve them manually and add the prefix and backport note to\nthe commit message yourself.
\nOption 3: using the prepare-commit-msg git hook
The scripts/prepare_commit_msg.py Python script can be installed as a\nprepare-commit-msg git hook. It automatically adds the [A.B.x] prefix\nto the subject line, appends Backport of <sha> from main. to the commit\nbody, and ensures the subject line ends with a period, even when the\ncherry-pick produces conflicts. To install it, create an executable file\n.git/hooks/prepare-commit-msg containing:
#!/bin/sh\nexec python scripts/prepare_commit_msg.py "$@"\nOnce installed, a plain git cherry-pick <sha> on a stable branch is\nsufficient; the hook handles the message formatting in all cases.
Nobody’s perfect; mistakes will be committed.
\nBut try very hard to ensure that mistakes don’t happen. Just because we have a\nreversion policy doesn’t relax your responsibility to aim for the highest\nquality possible. Really: double-check your work, or have it checked by\nanother merger before you commit it in the first place!
\nWhen a mistaken commit is discovered, please follow these guidelines:
\nIf possible, have the original author revert their own commit.
Don’t revert another author’s changes without permission from the\noriginal author.
Use git revert – this will make a reverse commit, but the original\ncommit will still be part of the commit history.
If the original author can’t be reached (within a reasonable amount\nof time – a day or so) and the problem is severe – crashing bug,\nmajor test failures, etc. – then ask for objections on the Django Forum\nthen revert if there are none.
If the problem is small (a feature commit after feature freeze,\nsay), wait it out.
If there’s a disagreement between the merger and the reverter-to-be then try\nto work it out on the Django Forum . If an agreement can’t be reached then\nit should be put to a vote.
If the commit introduced a confirmed, disclosed security\nvulnerability then the commit may be reverted immediately without\npermission from anyone.
The release branch maintainer may back out commits to the release\nbranch without permission if the commit breaks the release branch.
If you mistakenly push a topic branch to django/django, delete it.\nFor instance, if you did: git push upstream feature_antigravity,\ndo a reverse push: git push upstream :feature_antigravity.