The design of a pretty-printing library

Abstract

In this chapter we have considered the design of combinator libraries. We saw how studying the algebraic properties of the combinators desired can both help to suggest natural choices of representation, and guide the implementation of the operators. We saw several examples — lists, monads, and a pretty-printing library. For this kind of program development we need a language with higher-order functions and lazy evaluation, for which equational reasoning is valid; in other words, Haskell is ideally suited.

In the case of pretty-printing, studying the algebra led to the correction of a subtle error in the combinators' behaviour, and to the development of much more efficient implementations. The pretty-printing algebra is just too intricate to rely on intuition alone: working informally I could not see how to implement the optimisation considered in section 9, nor could I invent the representation used there. The formal approach has been invaluable.

The pretty-printing library itself has proved useable, despite its simplicity. Indeed, versions of it have seen quite extensive use, in program transformation tools, proof assistants, and compilers. All the pretty-printers in both the Chalmers and the Glasgow Haskell compilers are written using variants of this design.