Triaging AI generated code
After years of using git, I discovered git bisect today. A very timely find. Back in the day code was generated by people. People had at least a trace memory of what they created and why. Some even documented that. You get a bug in production, it was not very difficult to isolate what branch, which function might be the culprit. This triage dependent some on asking the contributors to share their knowledge.
Switch to today, most code is AI generated. Nobody has the phantom memory that used to help crawl to that exact chunk of code causing the bug. AI generated code needs AI testing tools and AI analysis tools. While those evolve, git bisect is there to help isolate which PR caused the bug.
- Run
git bisect start - Then
git bisect badtells git the current version is being naughty git bisect good <commit_id>tells git the last working version
The idea is to narrow down to the exact code commit that started the issue.
Test the version. If it is still misbehaving, git bisect bad tells git the problem was introduced before the version under test. git bisect good tells git the problem is in a later version.
Good to know:
git bisect skipskips the version. Useful if the commit was broken.git bisect resetaborts the process.