Micro-committing with Git

I have been using micro-committing for some time now, during which I have adapted the usage of this technique to my needs, bringing it to a level that makes me more productive than ever in software development.

Combining micro-committing with Git, while doing TDD is now my favorite development experience: I like how this workflow helps to deliver changes with speed and confidence.

This is not a one-size-fits-all approach, I’m sharing what works great for me; I hope some parts of what follows will help you and your team as well.

What is micro-committing?

Despite micro-commits are not new, there is not much literature or a formal definition of it. The concept is very simple: when developing or refactoring, use the SCM at every notable step creating a small commit.

Sizing properly micro-commits is what takes more practice mastering: I think it’s a very personal thing, one might be comfortable with lots of very tiny commits, some other with slightly larger bundles of changes, the important factor is usability.

We want to store in a commit every relevant code that produces an increment towards a goal.

Why should you care?

I found that using micro-commits makes working iteratively safer: the confidence I gain in adding or changing code once I am backed by a “diffable history”, is priceless.

Keeping track of changes at a smaller rate enables to move back and forth into history to do local optimizations. Having these changes recorded with Git might not be useful all the time, but when they’re needed they are going to be very valuable.

How to make the best out of it

In theory, the smaller the change bundle recorded in every commit, the better! While keeping a long list of tiny changes locally can have several benefits, it can also turn your review process and CI pipelines into chaos.

That’s why we need to be careful in managing micro-commits not only during coding but also when preparing changes to be submitted for peer review.

During development

During the development phase we want to store locally as many micro-commits as we need to ensure that we can easily move our code back to a working revision. When the development is completed we don’t want to send the micro-commits to the remote repository, because reviewing that history would be difficult.

With these concepts in mind, below the things to remember when mirco-committing during the development of new features.

One thing that I learnt recently thanks to my amazing colleagues is that Git history is precious. We don’t want to pollute the tree with too many commits targeting the same feature. Micro-commits are an aid for local development: they should be squashed, rewritten and edited before sending them into a remote repository (more on this below).

We want that every change with a significant impact on the codebase can be consistently fetched. Any IDE Undo/Redo feature does kind of the same, but a micro-commit plays better with refactoring, for example when changing a signature or renaming a package.

We don’t want to store in a micro-commit changes that do not have a “business” meaning, like renaming a variable. At the same time we don’t want to store in a micro-commit several changes that map to multiple features: doing so would prevent us to go back in time with fine-grained control, and thus lose the benefits or incremental history.

Optionally, include in the commit message the name of the component or sub-system under change. When re-reading the history of commits, it will be helpful to know which commits are part of which bundle of changes.

This is not strictly required: I like to create a simple, non-annotated tag, with the name of the task when I declare it completed. It helps me to point a commit to a task completion in the Git log. Adding a tag can be replaced by details in the commit message: we need to be able to understand which commit resolved a given task.

This is where this technique really shines: firing up a single command

git reset --hard

will allow to re-start from what you chose minutes ago as a standing point, a safe place from which a new iteration can start. If you create tags as in the previous bullet, you can even jump multiple micro-commits back, correcting a wider design mistake or a faulty implementation with a newer idea, just specify the tag name you want to reset to.

Preparing for peer review

It is important to review the Git history before submitting it to the remote repository: we should group multiple micro-commits belonging to the same task into a single commit.

The reasons for doing so:

  1. peer-review done commit by commit helps reviewers understanding more than the code logic, enriching code with the rationale and the decisions made by the author
  2. reverting a single feature should be as easy as reverting a single commit (or a few of them): reducing the cognitive load embedded in the history helps when things go wrong, even if all tests had passed 😁

A good way to achieve a clean history, ready for being reviewed, is with an interactive rebase to the target branch. This is the most straightforward way to create a meaningful list of commits that can tell a story about the development being shipped.

When rebasing interactively, you are presented the list of commits you are going to add on top of the target branch in an editor. From there it’s possible to merge (squash, fixup), remove (drop) and even change the order of commits!

I use these 3 steps to craft the intended Git history:

  1. remove tags used as references to micro-commits

In most CI systems tied to the Git tree, tags can trigger special workflows like creating artifacts for releases: we don’t want this.

If tags have been created to keep track of multiple features, I’ll remove them to avoid accidentally pushing them and triggering such workflows.

  1. squash or fixup corrections and clean-up commits to the uppermost relevant commit

It is good to remove commits that were useful during local development, but do not have a tangible impact for the reviewers: i.e. linting corrections, typos in documentation, etc… These commits should always be conflated with a previous one.

It’s also possible to move commits up and down during an interactive rebase, but note that it can lead to difficulties in setting up the desired history: you might end up in conflicts that will counter-effect the benefits of micro-committing, with time lost on editing during the rebase.

  1. set the commit that is the union of multiple fixup/squash commits with a reword

In this way, Git will open the editor to revrite the commit message. A commit that will be in the history of a repository needs to be explicit about what’s being committed. It has to contain a good high-level description of what it holds in all its parts.

Remember the famous jewelery slogan “a diamond is forever”? A valid statement is also “a commit is forever” (at least it should be!) 😉.

A practical example

I’ll use the history of the repository in which I’m writing this page 😎. If I run git log now I have

commit 91dc03e493fc9ff9426c7250de9d6dd168d58b2b (HEAD -> master, tag: ready-for-conclusion)
Author: inge4pres <xxx@gmail.com>
Date:   Sat May 1 18:18:10 2021 +0200

    refinement before example

commit aaafc4030dae80baa1a6c15905111a69484b9e6d
Author: inge4pres <xxx@gmail.com>
Date:   Tue Apr 6 22:40:58 2021 +0200

    refined central section

commit 1ef9ec1e6b70a61dd1772df3db69cb219dc4609e
Author: inge4pres <xxx@gmail.com>
Date:   Sat Apr 3 19:28:17 2021 +0200

    Resume: update page

commit 869e15fd7c85dcaef1e3e1865f2dcb62a5c06966
Author: inge4pres <xxx@gmail.com>
Date:   Sat Apr 3 17:23:37 2021 +0200

    fixed typos and intro section

commit aa25b348a6a0aa61cc6482beacbe4b96ec96e79f
Author: inge4pres <xxx@gmail.com>
Date:   Sat Apr 3 17:08:21 2021 +0200

    second section: peer review

commit f368028b3dcb38aed86893139ee9230d32fc7f2a
Author: inge4pres <xxx@gmail.com>
Date:   Fri Apr 2 21:25:16 2021 +0200

    first draft: completed intro

where we can note the following:

Removing the tag: git tag -d ready-for-conclusion.

It’s time to run a git rebase -i origin/master to rebase interactively my local branch on top of the remote branch. The output in the editor is the following:

pick f368028 first draft: completed intro
pick aa25b34 second section: peer review
pick 869e15f fixed typos and intro section
pick 1ef9ec1 Resume: update page
pick aaafc40 refined central section
pick 91dc03e refinement before example

# Rebase 01ff245..91dc03e onto 01ff245 (7 commands)

With the considerations done earlier, I will change the history like so:

pick 1ef9ec1 Resume: update page
pick f368028 first draft: completed intro
squash aa25b34 second section: peer review
fixup 869e15f fixed typos and intro section
reword aaafc40 refined central section
fixup 91dc03e refinement before example

The first editor session will be to opened to combine the commits f368028, aa25b34 and 869e15f: I will merge the commit messages, and I will add the slug of the post.

The second editor session will be to rewrite the last 2 commits combined, adding details about the post content.

Now the history looks much better, and it can be pushed:

commit be50a37f5dc63b587a1d105869449a6f67741a6a
Author: inge4pres <xxx@gmail.com>
Date:   Sat May 1 18:35:35 2021 +0200

    microcommitting: best practices and tips

commit 4f351b07ecbf45d1f11268cc98f13506c7fdb81d
Author: inge4pres <xxx@gmail.com>
Date:   Fri Apr 2 21:25:16 2021 +0200

    microcommitting: draft intro and peer review

commit 5d96163f8bf4196a57f01b59de434395fc52a221
Author: inge4pres <xxx@gmail.com>
Date:   Sat Apr 3 19:28:17 2021 +0200

    Resume: update page

Wrap up

Micro-committing is very helpful to collaborate effectively on a project, it can be tedious in first instance and apparently the benefits are not tangible, though adopting it will be very valuable in some occasions.

Try to practice it stick with mataining a clean Git history, then ask for feedback to the people collaborating with you on the repository.


Thanks to all the mentors that helped me be more productive! 🙏