Tuesday, June 16, 2009

Draft paper on Derive, comments wanted

It's been a long time since I last blogged (about 3 months). Since then I've had a paper on Firstify accepted in to the Haskell Symposium (I'll post the final version to my website shortly). I've also been writing a paper on Derive to go with my invited talk at Approaches and Applications of Inductive Programming (co-located with ICFP this year). I have to submit a final version by the 22nd of June (6 days time), but any comments on this draft would be gratefully received - either add them as comments to this post or send an email to ndmitchell AT gmail DOT com.

Download link: http://community.haskell.org/~ndm/temp/derive_draft.pdf

Title: Deriving a DSL from One Example

Abstract: Given an appropriate domain specific language (DSL), it is possible to describe the relationship between Haskell data types and many generic functions, typically type class instances. While describing the relationship is possible, it is not always an easy task. There is an alternative -- simply give one example output for a carefully chosen input, and have the relationship derived.

When deriving a relationship from only one example, it is important that the derived relationship is the intended one. We identify general restrictions on the DSL, and on the provided example, to ensure a level of predictability. We then apply these restrictions in practice, to derive the relationship between Haskell data types and generic functions. We have used our scheme in the Derive tool, where over 60% of type classes are derived from a single example.

Home page: http://community.haskell.org/~ndm/derive/

Darcs repo: http://community.haskell.org/~ndm/darcs/derive

The work presented in this paper will become the basis of Derive 2.0. Many thanks for any comments!