Class instance specificity order (was Re: Fundeps and type equality)

[Tyson Whitehead]

On January 10, 2013 13:56:02 Richard Eisenberg wrote:
> Class instances that overlap are chosen among by order of specificity;

Sorry to jump in the middle here, but this caught my attention as this sort of 
specificity determination is exactly what I had in mind when I was working on 
my "The shape of things: a reason for type preference" paper.

I would love to know what approach GHC is currently taking to determine this.  
The document gives the example of matching C Int [Int] against

C Int a
C a Bool
C Int [a]
C Int [Int]

This is easy though.  Obviously the last match is best because it is exact.  
What if we have something like C Int [[Int]] against

C Int b
C a [[b]]

though.  How is the best match determined in this case?

Thanks!  -Tyson