README
Background
At a previous job, we frequently ran into issues when hotfixes were deployed because they didn’t include all of the relevant dependencies. The patching team also had no way of rolling back the changes. This led to after-hours calls that I helped field. To reduce those issues, I created a series of scripts to identify dependencies based on Git commit history and generate the patch tarball with pre-install validations and rollback capabilities.
The dependency inference capability is broadly useful, though, as part of an introduction to a Git repo so I rebuilt it. I also added some additional functionality that I thought would be useful (eg. identifying documentation, etc.).
Theory
The core theory behind it is that the more two files appear together in commits, the more likely they are to be coupled. Understanding those relationships is useful for a couple of reasons:
- You know that if you change the public interface of one, then you likely need to update the other
- If there are too many interdependencies, then that indicates the system itself is fairly brittle and may need to be refactored
Final Thoughts
This is a project that I think had a lot of promise, but I didn’t invest enough time in it. Nowadays, though LLMs like Copilot may now do a better job of finding dependencies than a hard-coded script.