Johannes Waldmann wrote:

I use Text.PrettyPrint.HughesPJ and Text.ParserCombinators.Parsec heavily so I made some DrIFT rules for deriving the "obvious" instances for

class ToDoc a where toDoc :: a -> Doc class Reader a where readerPrec :: Int -> Parser a

Perhaps someone has solved a similar problem before.

I was looking for the same thing because I didn't like the way derived instances of Read depend on whether I define my datatypes in prefix, infix or record form. I ended up writing my own Read derivation using Template Haskell. Here it is. Basically it can read prefix, infix or record syntax where applicable. It is based on ReadPrec, not Parsec though. Some further notes and questions: 1. Instead of Read I actually use another class, Parse, which is a clone of Read. The reason is that I wanted to have new instances for some standard types that already have their Read instances defined. This introduces some hackery and requires you to use -fallow-overlapping-instances. If you don't like it, it should be quite easy to go back to Read. 2. My instance context derivation is very primitive: I simply repeat all the field types in the context, say

data T a b = T1 Int | T2 (a b) (Maybe b) instance (Parse Int, Parse (a b), Parse (Maybe b)) => Parse (T a b)

The result is that you need -fallow-undecidable-instances to chew through all that :) Here a question: are there any pitfalls about writing such instances that I don't currently see? Any feedback, improvements, fixes, etc. always welcome! Cheers, Misha