On 5 May 2009, at 11:27, z_axis wrote:

The following code snippets is from xmonad:
-- Given a window, find the screen it is located on, and compute
-- the geometry of that window wrt. that screen.
floatLocation :: Window -> X (ScreenId, W.RationalRect)
--.......
rr <- snd `fmap` floatLocation w
 
Prelude> :i fmap
class Functor f where fmap :: (a -> b) -> f a -> f b
 
It seems it is different from the definition of fmap ?
sincerely!

As the type signature of fmap explains, it transforms a function.  Specifically, it starts with a function (a -> b), and it transforms it to accept an 'a' inside a functor instead of just an a, and return a 'b' inside the same functor instead of just a b.  In other words, fmap applies functions inside containers.

We can see from floatLocation that it returns a pair inside a container - specifically, an X container.  Fmap takes snd, and transforms it to work on values inside the X.

So, snd has type (a,b) -> b, thus fmap snd has type f (a,b) -> f b.  In this case, the type it's being applied to is X (ScreenId, W.RationalRect), so f unifies with X, a with ScreenID and b with W.RationalRect.  Making snd `fmap` floatLocation w hav the type X W.RationalRect.

Finally, the bind into rr there takes it out of the X monad all together, getting you a W.RationalRect.

You may want to read this article which explains some of Haskell's abstraciton mechanisms:

http://noordering.wordpress.com/2009/03/31/how-you-shouldnt-use-monad/

Bob