This is an episode transcript. Visit the podcast website for the full episode, show notes, and freebies.
The Pre-Commit hook is a powerful tool for increasing quality in the continuous delivery pipeline. The commit is the first step in the pipeline, so any work done at this stage has down stream ramifications. The hook provides developers with immediate feedback by blocking bad commits. So what constitutes a bad commit?
Consider the question from a lean manufacturing perspective. One fundamental idea lean is assuring quality at the source. Imagine an automobile assembly line.
One step produces lug nuts. The next step uses the lug nut to mount the wheels on the car. If the lug nuts are the wrong size, then the wheels cannot be mounted. The assembly line stops as a result. This is an outage. The assembly line may only resume after the lug nuts are produced correctly and given to the next step.
There's a metric for this scenario. It’s "percent complete; percent accurate", also known as percent-C-A. This metric measures quality in the value stream. The lug nut in the example was complete but not accurate because it was not useable as-is. Checking quality at the source prevents these regressions which could even turn into full blown outages.
So what does percent-complete-percent-accurate mean for commits? It means proposed changes are usable as is by downstream consumers. There are two downstream consumers of commits: the deployment pipeline and other engineers.
Commits that knowingly fail in the deployment pipeline should be rejected. Allowing these commit to proceed through the pipeline only creates waste. We’ve all been there waiting behind the build that takes too long then fails on some dumb mistake like a typo of a malformed build configuration file. That build should have never have started. Running static analysis in the pre-commit hook easily prevents these regressions.
Commits are also code reviewed. This requires that other people read and understand the code. Inconsistent style is jarring to reviewers. It increases the time and effort to complete code reviews. It also creates rework for the original authors. You know, just a few minutes here and there, fixing this and that, then waiting for builds, and eventually getting an approval. Again, this is a waste. These regressions are easily prevented by adding linting and formatting in the pre-commit hook. Then it’s not possible to push commits containing known rework conditions.
This is just static analysis and automated error correction. The key is automation. If the process is not automated then it’s just not happening.
After static analysis then run some tests. I say some because it’s likely impractical to run the entire suite in pre-commit hook. Instead run the fastest tests or the most frequently failed tests, or just the tests associated with files staged in the commit. The objective is quickly fail known bad commits. In other words, fail fast.
Theses practices will increase the quality of individual commits and provide developers fast feedback on correctness. Here’s an outline for a pre-commit hook you can add to your projects:
- Run static analysis on all files staged for a commit. Use tools like
jsonlint, etc. Use the "autocorrect" option to automatically make stylistic changes.
- Run static analysis on all configuration files staged for the commit. I’m referring to files like
travis.yml. These files are read by the pipeline itself, so errors here would definitely break the pipeline. Validating them with the provided CLI prevents obvious regressions.
- Auto-format code using tools like
go fmtand be free from code formatting bike shedding.
- Run those fast tests! Any file staged in the commit should be caught by some linter or static analysis tool. If not, then the pre-commit hook should fail. This requires developers to either update the linting configuration to handle the new file or explicitly ignore it. This ensures that files introduced to the codebase are covered by the pre-commit hook.
When downstream failures occur in the deployment pipeline, consider preventing them by improving the pre-commit hook. Overtime the percent-complete-percent accurate will increase. Feels good right?