(does replying to my own message constitute a discussion?-) I would have hoped for someone to say any of the following: - no need for that, Haskell' is going to have proper records - Haskell' is going to avoid the records discussion, but I still don't like your idea, because.. - perhaps that idea isn't entirely wrong, let's see.. but, of course, I realise now that I made a fundamental mistake. The established procedure to generate discussion about language design issues is to propose either syntactic sugar or complex type system extensions, and my proposal was careful to do neither:-( To make up for this, I've now extended the example code to include more operations, cryptic infix operators, and more type class tricks (we have examples of record field selection, symmetric record concatenation, record field removal, record field update, record field renaming). Will that be sufficient to encourage one or two responses, even though the proposal is still the same? If nothing else, you might want to look through the code for usage patterns and usage issues wrt type classes/instances and various "extensions". How much of that will Haskell' be able to handle? Will it remove the need for workarounds? Having first-class labels would also simplify and make more useable things like the HList effort, which plays similar tricks, and the particular record system variation we get with the attached code is not unlike the one presented at last year's TFP by Daan. So here are two more related links: Extensible records with scoped labels. (draft) http://www.cs.uu.nl/~daan/pubs.html Strongly typed heterogeneous collections http://homepages.cwi.nl/~ralf/HList/ Cheers, Claus