How I Git
After Wednesday's email, someone said, "It sounds like you and I use git very differently." So, I wanted to explain what my typical Git workflow is.
I used to use Git Flow, but now, I almost never create a new branch when starting a new task.
I keep my workflow as simple as possible by using trunk-based development and working on a single branch as much as I can.
Before I start, I make sure any uncommitted changes are committed or reset and that the automated tests, static analysis, coding standards checks, etc., are passing so I know I'm starting from a good place.
Then, I start working on the task.
I like to work in small steps and make small, regular commits, but I don't always push each individual commit to the remote repository.
Sometimes, I'll make a number of "work in progress" commits and squash them into one before pushing them.
I want the time between making and pushing the commit to be as short as possible, and I want each commit to be deployable.
If I'm doing test-driven development, I'll typically commit each time a test is passing - whether it's adding a new test or extending one.
I run any tests often whilst writing code to ensure they pass, either using a watch command or a keybinding in Neovim.
I won't push a commit that would cause the code to not work, a test to fail, or block any other (potentially more urgent) changes from being pushed to production.
If I push a commit that breaks the CI pipeline, I'll fix it quickly, which is usually possible as the changes are small.
If not, I'll revert the commit to get back to a deployable state as quickly as possible.
If I'm going to add a feature flag, I'll usually know that in advance and avoid rushing to add one later if a more urgent task comes in.
By keeping each commit in a working and deployable state, a change can be feature flagged and deployed but not activated until the feature flag is enabled.
- Oliver
