In Haskell your code doesn't "do IO", you write (and compose) IO actions which describe what IO will be done at runtime. Only at runtime do the side effects oc

In Haskell your code doesn't "do IO", you write (and compose) IO actions which describe what IO will be done at runtime. Only at runtime do the side effects occur. The code itself is still pure and referentially transparent.