[Xmonad] xmonad is a zipper

[Donald Bruce Stewart]

ndmitchell:
> Hi
> 
> And once again, firing Catch (google:catch haskell) off on this code:
> 
> * new calls error, not a big surprise that this makes it incomplete
> 
> * view also calls error
> 
> * swap calls tail, which isn't safe unless you know some seriously
> deep stuff about index, and keep a detailed mapping between elements.
> The basic problem is that Catch can't know that successive calls to
> Map.lookup with the same key give the same results. You can probably
> eliminate this by only performing one lookup, which is also faster.

Good suggestion.

> 
> * delete and focusWindow both call focus, which is incomplete. Both
> feature very similar code, and could do with abstracting out. The
> basic pattern is:
> 
>   focus (stack (current s))
> 
> which is unsafe - since stack may return Empty, for which focus is 
> incomplete.

A bug! Well spotted.

> 
> * new relies on the Integral type having sensible semantics, if you
> pass Int or Integer its fine (machine checked), but if you create a
> crazy type, make it an instance of Integral then [0..n-1] may generate
> no elements, and your (h:t) match will fail.

Tricky.

> 
> All the rest of the program is safe, including the partial 'with'
> function, as specifically annotated by dons to check.

Very good Neil, thanks!

I've written up the idea behind implementing xmonad as a zipper-based
data structure which tracks focus for us here,

    http://cgi.cse.unsw.edu.au/~dons/blog/2007/05/17#xmonad_part1b_zipper

For those wondering how the new StackSet data type will work. The plan
is to merge in my zipper branch in the next few days.

-- Don