At 2002-05-17 03:54, Johannes Waldmann wrote:

From a type-theoretic viewpoint, instance declarations are relations between (sets of) trees (= elements of the respective data types). So one needs representations of such relations with effective decidability of overlapping-ness, most-sepcific-ness and such.

Yeah but unfortunately GHC doesn't look at the contexts of instance declarations. It has a straightforward but rather conservative concept of "overlapping" and "most specific". I'd like to see that fixed too, but that's a separate issue. -- Ashley Yakeley, Seattle WA