From Fedora Project Wiki
m (categorize)
(Pull Requests: add section about the review)
Line 25: Line 25:
  
 
Set up this way, pulling from upstream will attempt to apply your private commits in order on top of the new upstream ones, one after another. If that fails at some point, e.g. because of conflicts, it'll interrupt the rebasing process, so that you can resolve the issue, and continue with <code>git rebase --cont</code>. Alternatively, you could also restore the previous state by running <code>git rebase --abort</code>, e.g. to assess the differences between your (unrebased) branch and upstream before giving it a go again.
 
Set up this way, pulling from upstream will attempt to apply your private commits in order on top of the new upstream ones, one after another. If that fails at some point, e.g. because of conflicts, it'll interrupt the rebasing process, so that you can resolve the issue, and continue with <code>git rebase --cont</code>. Alternatively, you could also restore the previous state by running <code>git rebase --abort</code>, e.g. to assess the differences between your (unrebased) branch and upstream before giving it a go again.
 +
 +
=== The Review ===
 +
 +
When you've submitted your changes as a pull request, hopefully someone will pick it up soon (if not, poke some people on IRC: [irc://irc.freenode.net/#fedora-modularity #fedora-modularity on Freenode]) and give you feedback in form of comments, questions or suggestions. The comment section of a pull request isn't very suitable for longer discussions, so you might switch to email, IRC or another medium to discuss a topic, and then summarize in the PR. Consulting other contributors is encouraged, if additional opinions are needed. The job of a reviewer is not just to act as a gatekeeper for the project, but also to assist you in getting your changes into an acceptable state. This can go as far as making minor fixes on the fly rather than asking you to do it, or bringing the stack of commits "into shape" before merging the pull request.
  
 
== Individual commits ==
 
== Individual commits ==

Revision as of 18:00, 16 September 2016

Apart from writing nice code, there are some things that you should keep in mind regarding the changes you submit. Normally you'd develop your changes in a private branch on your fork of a repository and, when you're done, submit them as pull requests ("PR") against a public branch of the repository. The following guidelines concentrate on changes in this format, their goal is to enable you to groom the commits forming your pull request so that another person can review it without great effort, that the changes can be integrated well with the existing code and can be easily debugged later if necessary.

Pull Requests

Scope

One pull request should really be about implementing one feature or solving one problem. For instance, when developing your changes you might spot a bug in existing code and fix it. Mixing these changes with your new feature make reviewing them more work because the person doing it needs to assess if a chunk of your changes is related to the feature, or the bug fix. Similarly, if the review of your feature drags out, the bug fix might take that much longer before it's available to others. In most cases you should therefore create separate pull requests for both sets of changes.

As an exception to that, merely janitorial changes to the parts of the code your pull request touches anyway—say, fixing trailing whitespace or indentation, superficial changes that make the code you worked on better to read or understand—are acceptable as long as you put these changes in a commit or commits of their own, ideally put before your "real" changes in the commit order. This makes it easier to cope with other PRs that might fix the same things.

Describing your changes

The bigger the changes you submit are, the more important it is to give the reviewer a high level summary of what it is they are reviewing. If a pull request consists only of one commit, then its commit log should be sufficient in most cases and the forges hosting our repositories (Pagure and GitHub) use it as the default description text on submission. If it is longer, you may need to condense the individual changes of your commits, and maybe lose some comments about the problem you wanted to solve and your approach. If you are unsure about parts of your changes, this is also the place give the reviewer a heads-up.

Linear History and Rebasing

The changes you submit for reviewing should be a linear string of commits, please don't have merges in there. Therefore, in order to track upstream changes while you are still developing in a private branch, you should rebase it on top of the upstream branch you track. You can of course do that manually, but it's easier to tell git to automatically attempt to rebase your changes on top of the branch from which you pull (replace $branchname with the actual name of your local branch):

   git config branch.$branchname.rebase true


You can also set this globally for any newly created branch (you'd have to do the above for all existing branches, though):

   git config --global branch.autosetuprebase always

Set up this way, pulling from upstream will attempt to apply your private commits in order on top of the new upstream ones, one after another. If that fails at some point, e.g. because of conflicts, it'll interrupt the rebasing process, so that you can resolve the issue, and continue with git rebase --cont. Alternatively, you could also restore the previous state by running git rebase --abort, e.g. to assess the differences between your (unrebased) branch and upstream before giving it a go again.

The Review

When you've submitted your changes as a pull request, hopefully someone will pick it up soon (if not, poke some people on IRC: #fedora-modularity on Freenode) and give you feedback in form of comments, questions or suggestions. The comment section of a pull request isn't very suitable for longer discussions, so you might switch to email, IRC or another medium to discuss a topic, and then summarize in the PR. Consulting other contributors is encouraged, if additional opinions are needed. The job of a reviewer is not just to act as a gatekeeper for the project, but also to assist you in getting your changes into an acceptable state. This can go as far as making minor fixes on the fly rather than asking you to do it, or bringing the stack of commits "into shape" before merging the pull request.

Individual commits

Commit Scope and Size

Like a pull request itself, a commit should also be about just one thing. So you would e.g. split up the introduction of a new class from its unit tests, where existing code is changed to use the new class instead of some legacy code and the removal of the latter, too (if it isn't used anymore). The reverse—one concern should be dealt with in one commit—also holds true, so if you discover bugs in a newly introduced piece of code while you're still developing it, the buggy commit introducing it and the fix should be rolled into one. This keeps the number of broken commits down which e.g. makes it easier to use git bisect at a later point.

Note.png
"Commit early, commit often."
It's much easier to merge smaller commits into larger ones if they belong together, rather than disassembling a commit that actually addresses more than one concern.

Commit Log Messages

The purpose of a commit log message is to briefly summarize the changes in the commit, but it's also where background information can be put, e.g. why some approach was used and not another. It's where someone would look to find about the "what and why" vs. the "how".

Building a Commit

Often you'll want to pick only parts of your uncommitted changes, in order to follow these guidelines, or to leave out debugging statements which you don't want to submit. You can select the parts in your changes you want to commit by using git add --patch which presents the differences as hunks in unified diff format and lets you choose which ones to add to the staging area and which to skip. After committing these staged changes, you can repeat the process until all changes you want to submit are taken care of. There are ways to separate a large commit into smaller ones, but this approach is often more difficult one of the two.

Tools

  • Adding using patch mode: With git add --patch ... you can pick which changes you want to commit.
  • Interactive rebasing: Use git rebase -i ... @{u} to reorder your commits, reword their commit messages, merge or amend them. It's important to not do this to upstream commits, therefore @{u} specifies the point where your branch split off from upstream.
Important.png
If all else fails:
GIT remembers the history of revisions you had checked out in your repository, refer to the output of git reflog to find a "know good" one.
Stop (medium size).png
Using git reset
You can use git reset [--hard] $some_sha1_commit to bring you back to a known good state. Be careful, though: using the --hard option will lose any changes made to files under the control of GIT.

See also

  • The Pro Git book
    • The "Rewriting History" chapter for more detailed information about amending, interactive rebasing, and other advanced ways of screwing up your repository ;)