Listen to this story
Front End Web Developer @hanko.io and owner of a cat named Ubuntu
Rebasing is one of the two Git processes of integrating changes from one branch to another.
While merging always moves a branch forward in history by simply adding a new commit representing each branch integration, rebasing can be much more powerful (and dangerous) as it allows us to rewrite history.
Rebasing is the process of moving or combining (i.e. squashing) a sequence of commits to a new base commit.
Imagine youāre working on a feature branch and after some time you realise the main (aka "master") branch has been updated by someone else's code. This means your branch has now diverged from the main branch. At some point, you will want to include said changes in your feature branch. A common way of doing this would be to simply do a git pull from main which would add a merge commit to your feature branch.
The issues with this method can be:
Merge commits
can be very vague, providing very little information in the git tree which makes debugging a much harder taskpulling/ merging
- multiple times can result in a git history similar to this:Reading this history provides very little (if any) information on what changes have been made to the project.
The goal of rebasing is to keep a clean, clear project history. So instead of pulling the
main
changes and add a merge commit we can tell our branch to move its commits to the top of the latest changesFeature branch JIRA-123 is being rebased to the top of the master branch
By rebasing our feature branch to the top of the
main
branch, we keep all the changes and commits we made unaltered while getting the latest main updates
. In order to achieve this, we can execute the following command while in our feature branch:$ git rebase main
This will start the rebase process. If no conflicts are found you should see a success message and your feature branch is now up-to-date with the main branch and we're one step closer to having a project history similar to this:
Interactive rebasing brings the power of rebasing to a whole new level! It allows us to interactively change multiple commits in different ways.
You can start an interactive rebase by executing a rebase command followed by the -i parameter and the commit we want to modify:
$ git rebase -i <commit hash | HEAD position>
Interactive rebase CLI showing the three last commits and always in order from oldest (top) to newest (bottom)
We can tell just by looking at the list of commands displayed in the CLI how powerful this tool can be! We can edit commits, squash them, remove them, amend them and so much more. We can even permanently change the commit order!
Squashing allows us to specify which commits we want to combine in order to help maintain clean history.
Does something like this look familiar?
Well, if you like committing a lot while you work you probably donāt like thinking about being super descriptive on every single commit message. Squashing can be extremely helpful in these situations because starting an interactive rebase you can then tell git which commits will be āmergedā into one (or more) and then edit its commit message.
Keep in mind that we must always squash a commit āintoā another one.
So say we have 6 commits and we think only two of them are relevant we could squash the other 4 like:
$ git rebase -i HEAD~6
āsā is simply a shorthand name for āsquashā
In this case weāre saying weād like to merge commits 2ā5 and include them in the first one ( 77497f5) and keep the last commit as-is.
Next git will rebase each commit to their new base ( 77497f5) and if there are no conflicts, you'll be able to change their final commit messages, and voila! Only 2 commits total!
Similarly to what happens with a Git merge, updating a branch with another can result in code conflicts.
A big difference in rebase conflicts is that weāre not dealing with one merge commit but with (possibly) multiple commits being moved.
This means rebase conflict solving is an iterative process that goes through each commit being rebased.
In a feature branch with two commits, if a conflict is found while rebasing commit A we fix it, tell git to continue the process to the next commit, fix the new conflicts on commit B and only then weāll be done
Rebasing a local feature branch means weāre rewriting history, so what this actually means is that weāre not cherry-picking the feature branch commits to the new base. Although visually everything looks the same, Git creates new commits (with new hashes) on the targeted branch.
This means whenever we try simply pushing our changes to our remote repo we get an error stating the local and remote branches have diverged.
To fix this problem itās good practice to push using the --force-with-lease parameter.
You can check -force vs -force with lease to learn more about this.
Have you heard of the beautiful command:
$ git reflog
Reflog allows us to check every change we did to our local branch. Running the command will return something like this:
Reflog output showing the last 3 actions
In this case, if Iād like to go back before I ever did the rebase I could jump back to HEAD~2 or using its hash instead:
$ git checkout 682c215
At this point, Iām in the past and can undo the mistakes Iād made before. Time travel!! šš„
At any point in our interactive rebase if things get too messy and we no longer know how to fix them, we can always abort the whole process in the terminal and the interactive rebase will be canceled completely:
$ git rebase ā abort
As you can tell rebase can be used for a lot of different purposes, especially using the interactive rebase functionality. Hopefully, this article provides you with just enough to get started messing around with it.
Donāt worry if you run into issues understanding some of the nuances of Git rebase. It has its learning curve and thatās why I included some escape methods so you can go back and try again šŖ
Also published on Alex's Medium