The simpler alternative, wich disables the definition of new modes for attributes in other modules, but is also easier to use, is: ------- data Readable = Readable data Writeable = Writeable data Attr m w a = Attr (w -> IO a) (w -> a -> IO ()) newAttr :: (w -> IO a) -> (w -> a -> IO ()) -> Attr (Readable,Writeable) w a readAttr :: (w -> IO a) -> Attr (Readable,()) w a writeAttr :: (w -> a -> IO ()) -> Attr ((),Writeable) w a get :: w -> Attr (Readable,m) w a -> IO a (=:) (Attr getter setter) x = newProp (\w -> setter w x) getter setter ... ------- Also, again by using typeclasses, it can be enforced that modes are in a fixed set, for example to avoid that one writes "Attr (Writeable,Readable)" in place of "Attr (Readable,Writeable)". As usual, I have compiled htoolkit with the modifications. This time I have not used any non H98 feature, I just have the exact type hierarchy of htoolkit but with well-typed read-only and write-only attributes. The system can be extended to other switches, like "listenable" or "write-once" or similar with a linear number of types, if I am not wrong (so this proposition is always true, uh?). I don't know if it's better to leave around the ReadMode and WriteMode typeclasses or to leave things as in this message, wich looks clearer also in type declarations (maybe extensible is better, anyways). Vincenzo