Mutation testing: Too good to be true?
summary
Explains practical results from running mutation testing with Pitest on JVM code: default engine Gregor vs Discarders behavior, Gregor generates many irrelevant mutants for Kotlin and data classes, Discarders finds a missed filter bug; mutation runs are typically 10–30x slower (some projects 2–6 hours). Built-in presets can multiply runtime by 10. Google’s internal system reduces useless mutants from ~80% to ~15% via historical filtering and review integration. Many logic errors (fluent/reactive chains, wrong fields, single-item bugs) remain undetected, so human code review stays necessary.
AI-written summary · about
chapters (9)
- 0:00Introduction to mutation testing
- 3:48What is mutation testing
- 7:23Mutation testing tools overview
- 12:23Projects suitable for mutation testing
- 15:09Performance challenges with mutation testing
- 20:08Case study bad financial code
- 27:32Google's advanced mutation testing
- 35:43Practical use in daily work
- 39:35Final recommendations and summary
