swynx

10 ms·

1. 1.▲
   by swynx 7mo ago

   I built a tool that detects dead code across 26 languages. The honest answer is most people aren't catching it. The code works, the tests pass, and nobody reviews what the LLM left behind. You end up shipping unreachable functions, dup
2. 2.▲
   by swynx 7mo ago

   Agree that test coverage is the ultimate safety net for false positives, but I'd push back on relying on it as the primary strategy. "Delete it and see if tests break" works at small scale, but the feedback cycle gets painful