Applicative and Functor

[Ashley Yakeley]

Ross Paterson wrote:
>>The Prelude is to obvious place, since that's where both Functor and 
>>(->) are introduced.
> 
> Yes, but that would break compatibility with Haskell 98.

In that case, we should just create a new module.

> OK, one client but no instances.

Fair enough. I like the separation of functionality on principle, but 
it's difficult to justify without instances.

I propose:

1. A new module Data.Functor, exporting Functor and all its instances 
from the Prelude, as well as "instance Functor ((->) a)".

2. Making Functor a superclass of Applicative, and getting rid of <$>.

3. Making Functor and Foldable superclasses of Traversable, and getting 
rid of fmapDefault and foldMapDefault.

You might also consider:

4. Adding a method to "class Traversable t" (with a default implementation):

   toList :: t a -> [a]

5. Renaming pure as returnA and <*> as apA. Only it looks like your 
Arrows stole returnA.

Also:

If Functor is a superclass of Traversable, is it better to have traverse 
and mapM as the methods, or sequence and sequenceA?

What's your "instance Applicative []"? Does it use repeat and zapp, or 
is it the "list of successes" (which would be compatible with "instance 
Monad []")?

-- 
Ashley Yakeley, Seattle WA
WWED? http://www.cs.utexas.edu/users/EWD/