On Sat, 21 Jun 2008, Iavor Diatchki wrote:
Hi Jamie,
thanks for doing this---I think that the functions in the API look good. Of course, one can always think of more :-) One that I often find useful when working with Data.Map is a function that removes a key from a map and also returns the associated value---a kind of destructive lookup.
I think that the library would be better if you exposed a non-overloaded API for the map(s?) that you plan to implement. This is useful because (i) most of the type I anticipiate using a single type of map, and concrete types lead to more accurate types in type inference, (ii) it makes the API Haskell'98 and hence likely to work on most Haskell implementations out there. If you really think that the overloading is useful, then you could provide separate modules that provide the class+instances in terms of the non-overloaded API and add a cabal flag to turn these on/off.
That's certainly not a good idea, since other modules must be sure, that if package generic-tries is installed that then the advanced API is available. Thus the advanced API must be moved to a separate package. (I prefer that separation.)