>main = do let arr = foo (NB (1.0,1))
>  putStr ("arr==arrS "++show (arr==arrS)++"\n")
>  putStr ("arrS==arr "++show (arrS==arr)++"\n")
>  putStr ("bnds arr arrS "++show ((bounds arr)==(bounds arrS))++"\n")
>  putStr ("bnds +id arr arrS "++show (((bounds.id) arr)==((bounds) arrS))++"\n")
>  putStr ("id +bnds arr arrS "++show (((id.bounds) arr)==((bounds) arrS))++"\n")

>foo :: RD -> Array (Int,Int) (CM Double)
>foo rd = case rd of
>    NB (f,p) -> h where h=hCMt [(p,listArray ((1,1),(1,1)) [f])]
>   where
>   --h0CMt :: Array (Int, Int) (CM Double)
>   h0CMt = accumArray (\a _-> a) ZM ((1,1),(1,2)) []
>   hCMt po_arL = h0CMt // (map (\(po,ar) -> ((1,po),M ar)) po_arL)

>arrS :: Array (Int,Int) (CM Double)
>arrS = listArray ((1,1),(1,2)) [M (listArray ((1,1),(1,1)) [1.0]),ZM]
>-- This is the array that foo (NB (1.0,1)) shows in Hugs.
>-- But (foo (NB (1.0,1)))==arrS is False.
>-- All in main is only to show the strange behaviour.
>-- If I write NB (f,p) -> hCMt [(p,listArray ((1,1),(1,1)) [f])] instead of line 16
>-- the bug disappears. That is also the reason why I have to keep the data declaration RD.
>-- If I put the type signature of line 18 in scope the bug also disappears.
>-- If I write
>-- hCMt po_arL = (accumArray (\a _-> a) ZM ((1,1),(1,2)) []) // (map (\(po,ar) -> ((1,po),M ar)) po_arL)
>-- instead of line 19 and 20 it also vanishes.
>-- So I thing now it is the shortest example that I can find. (Apart from deleting main.)