Git is a popular version control system. In Git, an atomic change of one or several files is called a commit, and several consecutive commits are combined into a branch. Branches are used to implement new ideas (features).
It happens that the idea is a dead end and the developer has turned the wrong way, so he needs to roll back to the original version. He should forget about the new branch, switch to the main dev or master branch and continue working. In this case, the "scion" will hang forever, as well as the desire to remove it. But how to remove the part of the history? This branch shows the efforts of the hard-working programmer, even if in vain. So it will be easier to report to the boss, because an unsuccessful result is also a result!
I hasten to rejoice that Git developers are going to introduce a new command to close such "homeless" branches in the third version. The current version is 2.21.0.
How to use this command, what benefits does it give and what do IT companies think? The article answers these and other questions.
Description
Now it is possible to close an unsuccessful branch over one of the previous commits. Closure arcs are yellow colored in the pictures below.
The commit 4
is the last one for the unsuccessful feature. It was closed over the commit 1
, and then we return to the master and go the other way (the commit 5
).
You can also close a commit over itself, thus creating loops:
You can close the branch over any commit — Git is smart, it calculates the differences and merges everything correctly:
How to use?
The merge
command does not include the functionality of closures, since for the first case the branch will be fast-forwarded, and for the second case nothing will be done (git already up to date
).
In order not to change the old behavior, the developers decided to introduce a closure command:
git closure -s $source_commit -d $dest_commit -m $message
The first argument -s $source_commit
sets the hash of the commit from which you want to stretch the loop, and the second (optional) -d $dest_commit
sets the commit into which the loop will be closed. If it is absent, the closure occurs in the current check-out branch. The -m $message
argument sets a closure message, like failed feature, revert to origin
. However, the --allow-empty-message
option is also available, which allows commits without messages. By default, Git allows only one closure for a pair of commits. To bypass this limitation, the --allow-multiple-closures
option is available.
After the command is executed, Git calculates the changes, and in the final commit, a double diff will be shown: from the base and closing branches. In the general case, it is an n-dimensional diff, that is, there can be as many closures as you wish. closure-commit is similar to merge-commit with the only difference that it contains several messages, not one.
Unfortunately, existing Git GUI's do not have good support of closures. GitExtensions preview version displays merge curves instead of elegant arcs. Take a look at the new fields such as Closure message
and Closure diff
:
It is worth noting that the closure
command always changes history (as now Git is a full-fledged time machine!), so now it is possible to push branches only with the --force
option, or the safe --force-with-lease
option.
Rebase is also available for looped branches, although the logic for recalculating commits is complicated.
Also the auto
option allows automatic closures of all the old branches. In this case, the closing commit is the one starting the branch. With Git IDE plugins, closures can be run periodically. In GitExtensions there is a similar plugin Delete obsolete branches.
What IT companies think
Large IT companies: Google, Facebook, Apple, DeepMind, Positive Technologies, and especially Microsoft, are eagerly awaiting closures, because now it will be possible to formalize the life cycle of branches, including unmerged ones.
One of Microsoft's top managers, Michael Richter, wrote:
The new feature of Git, of course, will reduce the chaos in the world of open source development (and not only). There are a lot of "hanging" branches in our repositories. For example, in vscode we have more than 200 of them, and in TypeScript more than 300! And this problem is not only ours. Closures not only improve organization, but also make it possible to track programmer's reasoning, sometimes completely incomprehensible even to colleagues :) Closures reminded me of the movie "Back to the Future", where the characters traveled to the past and the future. I like this movie, I watched it several times. And I think I will love Git even more because if this feature :)
Note
If earlier the graph of commits was a directed acyclic graph (DAG), then closures extend it to a general directed graph. Using Git, you can describe regular expressions in which the states are commits, and the alphabet is the set of all messages. But this is the topic for the hub "Abnormal programming", and therefore goes beyond the scope of this article. However, if this sounds interesting to you, check out the article on how to store family trees inside Git.
eyalsh
It's a cool feature, but I foresee that it will only be useful in tools like GitHub, GitLab, BitBucket and the likes.
If I'm using the terminal, than I'll just delete the branch and move on, but I might want to reject PR and keep the commit history for a later time.