The tragedy of adding side-effects to a once pure function

Slid a persistToDB() inside what used to be a pure function - now the memoization cache has trust issues and CI treats referential transparency like a rumor

Twenty years later, you're still maintaining that "pure" utility function that writes to a global logger, updates metrics, and occasionally mutates a config object - but hey, at least the return value is deterministic

The function started as a beautiful pure expression - same inputs, same outputs, every single time. Then someone added a database call, sprinkled in some logging, threw in a cache update, and suddenly it's mutating global state like it's going out of style. Now it's impossible to test without mocking half the universe, and good luck reasoning about what it does when called twice. The function didn't just lose its purity - it became that legacy method everyone's afraid to touch because 'it works in production... somehow.'

The first “just log it” turned our pure function into an IO-bound microservice with retries, backoff, and a pager - referential transparency doesn’t have SLAs

Side-effects in pure functions: FP's original sin, swapping referential transparency for 'works on my machine' roulette

The fastest migration from FP to debug‑driven architecture is a println inside a “pure” function - congrats, you just invented time‑dependent integration tests