The data dependency is circular. The "case e of" Str and Brk are not-circular:
layout examines the input parameters to determine column'. Then column'
is used to compute columnOut and s'. Then the current data is
prepended to s'. The Blo case is the circular one. Pushing the circular
definitions in to this case (and using the mapSnd helper) results in:
> layout :: [Pretty] -> (Int,String) -> (Int,String)
> layout [] columnIn'sIn'pair = columnIn'sIn'pair
> layout (e:es) (!columnIn,sIn) =
> case e of
> Str n str -> mapSnd (showString str) $ layout es (columnIn+n,sIn)
> Brk n | columnIn + n + breakDist es after <= margin ->
> mapSnd (prepend n ' ') $ layout es (columnIn+n,sIn)
> | 0 <= startColumn ->
> mapSnd (('\n':).prepend startColumn ' ') $ layout es
> (startColumn,sIn)
> | otherwise -> mapSnd ('\n':) $ layout es (0,sIn)
> Blo _n indent es' -> let startColumn' = indent + columnIn
> after' = breakDist es after
> (columnOut,s') = layout es (column',sIn)
> (column',sOut) = block startColumn' after'
> es' (columnIn,s')
> in (columnOut,sOut)
>
> mapSnd f (a,b) = (a,f b)
The circular usage of column' and s' can be unwound by "importing
Control.Monad.Fix(fix)"
and writing a definition with "fix" that explicitly feeds back the s':
> Blo _n indent es' -> let startColumn' = indent + columnIn
> after' = breakDist es after
> withS ~(_,s') = let (column',sOut) = block
> startColumn' after' es' (columnIn,s')
> (columnOut,s'') =
> layout es (column',sIn)
> in ((columnOut,sOut),s'')
> in fst (fix withS)
In withS above, the column' is created by the call to block and consumed by the
call to layout.
The s'' is fed back to become s' by the use of fix. The actual answer is the
fst component.
It is also possible to avoid the lazy '~' matching by using "snd":
> Blo _n indent es' -> let startColumn' = indent + columnIn
> after' = breakDist es after
> withS ans's' = let (column',sOut) = block
> startColumn' after' es' (columnIn,snd ans's')
> (columnOut,s'') = layout
> es (column',sIn)
> in ((columnOut,sOut),s'')
> in fst (fix withS)
>
Whether any of these three versions is clearer to the previous message is a
matter of taste.
Cheers,
Chris
_______________________________________________
Haskell-Cafe mailing list
Haskell-Cafe@haskell.org
http://www.haskell.org/mailman/listinfo/haskell-cafe
