May 15, 2024

Merge and Rebase

#git #programming

~ views | ~ words

I have two general rules for working across branches in a git repo these days:

  • squash merge feature branch into main (or rebase, sometimes)
  • merge main into feature branches

https://twitter.com/mehulkar/status/1790822175638765856

squash merge feature branches into main

For most people, I imagine this is done via Pull/Merge Requests on your git hosting provider (e.g. Github or Gitlab), but if you work in one of those projects that merges locally and pushes to main to close out our PR, same goes for you too.

I like Squash & Merge because:

  • it keeps main (or whatever your default branch is) free of merge commits. Merge commits are fine, and it's not the end of the world to have them, but they are harder to work with when you're viewing history, rebasing, or bisecting. And for someone just tracing file histories, they are really just extra noise.
  • It lets you forgo writing good commit messages until the point of merging. The hard truth is that 99% of people don't care about most of your commit messages in your branches. (There is a very important caveat to this that I'll get to in the next section). But in main, your commit messages do matter. So it's important that when you merge into main, you write one good one. Squashing all your commits from your branch into one lets you do that.

I like Rebase and Merge on select occasions when I have taken the time to craft individual commits. That means that the commit message and body is well written and useful to have in main, and the contents are atomic (i.e. cleanly reversible) and a logical portion of work that someone could/should digest independently. I've worked with probably <5 people in my 15 year career that actually organized their work into good commits, and I'm one of them. So it's rare.

These days it mostly comes in handy when I don't want to pay the CI cost of submitting those individual commits in separate PRs. PRs are expensive because CI is often slow and flakey and that is just a fact of life these days for most codebases. So instead of making 5 PRs, I like to submit 5 commits in a PR, smile and wink at the reviewer, and then Rebase & Merge to keep the history and separation intact in main.

I like Rebase and Squash Merges both because they keep a single line of commits in main.

merge main into feature branches

I've written about this before, but not very succinctly. The main purpose of using Rebase/Squash when merging into main is to keep a linear history. Oddly enough the purpose of git merge main into the feature branch is the same: a linear history! The difference is that feature branches have more "events" in their history than just commit messages. Feature branches are often submitted as Pull/Merge Requests. And Pull Requests have comments and reviews. Each of these are part of a timestamped history of the artifact. Rebasing destroys this history. You cannot tell after a rebase if a comment was submitted before or after a commit. Reviewers have to review the entire diff again to see whether their comments were addressed when new commits are added. For this reason, I prefer merge commits in my feature branches. Admittedly, that gets messy and I have to start over with a git reset --soft main, but that's pretty rare. These merge commits show up in your PR, but then if you Squash and merge, they don't travel into main, so there is no real harm there. They are also easier to do on a feature branch, because conflicts only have to be resolved once.

The only caveat to this is if you work on a team where more than one person works on the same feature branch. This has been exceedingly rare in my career, so I never have to think about it. If I do work like that on occasion, I just treat the feature branch like main and make pull requests into it, and the same rules apply.

If you like this post, please share it on Twitter and/or subscribe to my RSS feed. Or don't, that's also ok.