Answers

1. There are two reasons, one more practical and the other more principled:
   • Firstly, lazy evaluation makes modeling I/O with ordinary functions practically unworkable. We lose the ability to reason compositionally about programs. Moreover, it is hard to figure out in what order I/O actions will happen, and this is crucial for program correctness.
   • Secondly, pure functions (i.e., without I/O) have a number of essential properties (e.g., predictability and ease of reasoning) that Haskell wants to preserve. Adding I/O in the standard way would destroy that.
2. Conceptually, we separate describing I/O actions from performing them. Haskell provides a type, IO a, that describes I/O actions that, when performed, yield a result of type a. It’s the job of the Haskell program’s main :: IO () function to construct such an I/O description, and the job of the runtime system to perform the I/O actions described.

   To create the IO () description that main should produce, the Haskell...