mirror of
				https://github.com/KevinMidboe/linguist.git
				synced 2025-10-29 17:50:22 +00:00 
			
		
		
		
	What is Frege? ------------- Frege is a non-strict, pure functional programming language in the spirit of Haskell for the JVM. It enjoys a strong static type system with type inference. Higher rank types are supported, though type annotations are required for that. Frege programs are compiled to Java and run in a JVM. Existing Java Classes and Methods can be used seamlessly from Frege. The Frege programming language is named after and in honor of Gottlob Frege. Project State: ------------- The compiler, an Eclipse plugin and a provisional version of the documentation can be downloaded from here https://github.com/Frege/frege/releases. The REPL can be downloaded from here https://github.com/Frege/frege-repl/releases. An online REPL is running here http://try.frege-lang.org/. Examples: -------- 1) Command Line Clock: https://github.com/Frege/frege/blob/master/examples/CommandLineClock.fr 2) Brainfuck: https://github.com/Frege/frege/blob/master/examples/Brainfuck.fr 3) Concurrency: https://github.com/Frege/frege/blob/master/examples/Concurrent.fr 4) Sudoku: https://github.com/Frege/frege/blob/master/examples/Sudoku.fr 5) Java Swing examples: https://github.com/Frege/frege/blob/master/examples/SwingExamples.fr
		
			
				
	
	
		
			562 lines
		
	
	
		
			23 KiB
		
	
	
	
		
			Plaintext
		
	
	
	
	
	
			
		
		
	
	
			562 lines
		
	
	
		
			23 KiB
		
	
	
	
		
			Plaintext
		
	
	
	
	
	
package examples.Sudoku where
 | 
						|
 | 
						|
import Data.TreeMap (Tree, keys)
 | 
						|
import Data.List as DL hiding (find, union)
 | 
						|
 | 
						|
 | 
						|
type Element    = Int           -- 1,2,3,4,5,6,7,8,9
 | 
						|
type Zelle      = [Element]     -- set of candidates
 | 
						|
type Position   = Int           -- 0..80
 | 
						|
type Feld       = (Position, Zelle)
 | 
						|
type Brett      = [Feld]
 | 
						|
 | 
						|
--- data type for assumptions and conclusions
 | 
						|
data Assumption =
 | 
						|
              !ISNOT Position Element
 | 
						|
            | !IS    Position Element
 | 
						|
 | 
						|
 | 
						|
derive Eq Assumption
 | 
						|
derive Ord Assumption
 | 
						|
instance Show Assumption where
 | 
						|
    show (IS p e)    = pname p ++ "=" ++ e.show
 | 
						|
    show (ISNOT p e) = pname p ++ "/" ++ e.show
 | 
						|
 | 
						|
showcs cs = joined " " (map Assumption.show cs)
 | 
						|
 | 
						|
elements :: [Element]           -- all possible elements
 | 
						|
elements = [1 .. 9]
 | 
						|
 | 
						|
{-
 | 
						|
    a  b  c   d  e  f   g  h  i
 | 
						|
     0  1  2 | 3  4  5 | 6  7  8    1
 | 
						|
     9 10 11 |12 13 14 |15 16 17    2
 | 
						|
    18 19 20 |21 22 23 |24 25 26    3
 | 
						|
    ---------|---------|--------
 | 
						|
    27 28 29 |30 31 32 |33 34 35    4
 | 
						|
    36 37 38 |39 40 41 |42 43 44    5
 | 
						|
    45 46 47 |48 49 50 |51 52 53    6
 | 
						|
    ---------|---------|--------
 | 
						|
    54 55 56 |57 58 59 |60 61 62    7
 | 
						|
    63 64 65 |66 67 68 |69 70 71    8
 | 
						|
    72 73 74 |75 76 77 |78 79 80    9
 | 
						|
-}
 | 
						|
 | 
						|
positions :: [Position]         -- all possible positions
 | 
						|
positions = [0..80]
 | 
						|
rowstarts :: [Position]         -- all positions where a row is starting
 | 
						|
rowstarts =  [0,9,18,27,36,45,54,63,72]
 | 
						|
colstarts :: [Position]         -- all positions where a column is starting
 | 
						|
colstarts =  [0,1,2,3,4,5,6,7,8]
 | 
						|
boxstarts :: [Position]         -- all positions where a box is starting
 | 
						|
boxstarts =  [0,3,6,27,30,33,54,57,60]
 | 
						|
boxmuster :: [Position]         -- pattern for a box, by adding upper left position results in real box
 | 
						|
boxmuster =  [0,1,2,9,10,11,18,19,20]
 | 
						|
 | 
						|
 | 
						|
--- extract field for position
 | 
						|
getf :: Brett -> Position  -> Feld
 | 
						|
getf (f:fs) p
 | 
						|
    | fst f == p = f
 | 
						|
    | otherwise  = getf fs p
 | 
						|
getf [] p = (p,[])
 | 
						|
 | 
						|
 | 
						|
--- extract cell for position
 | 
						|
getc :: Brett -> Position -> Zelle
 | 
						|
getc b p = snd (getf b p)
 | 
						|
 | 
						|
--- compute the list of all positions that belong to the same row as a given position
 | 
						|
row :: Position -> [Position]
 | 
						|
row p = [z..(z+8)] where z = (p `quot` 9) * 9
 | 
						|
 | 
						|
--- compute the list of all positions that belong to the same col as a given position
 | 
						|
col :: Position -> [Position]
 | 
						|
col p = map (c+) rowstarts where c = p `mod` 9
 | 
						|
 | 
						|
--- compute the list of all positions that belong to the same box as a given position
 | 
						|
box :: Position -> [Position]
 | 
						|
box p  = map (z+) boxmuster where
 | 
						|
    ri = p `div` 27 * 27    -- 0, 27 or 54, depending on row
 | 
						|
    ci = p `mod` 9          -- column index 0..8, 0,1,2 is left, 3,4,5 is middle, 6,7,8 is right
 | 
						|
    cs = ci `div` 3 * 3     -- 0, 3 or 6
 | 
						|
    z  = ri + cs
 | 
						|
 | 
						|
--- check if candidate set has exactly one member, i.e. field has been solved
 | 
						|
single :: Zelle -> Bool
 | 
						|
single [_] = true
 | 
						|
single _   = false
 | 
						|
 | 
						|
unsolved :: Zelle -> Bool
 | 
						|
unsolved [_] = false
 | 
						|
unsolved _   = true
 | 
						|
 | 
						|
-- list of rows, cols, boxes
 | 
						|
allrows = map row rowstarts
 | 
						|
allcols = map col colstarts
 | 
						|
allboxs = map box boxstarts
 | 
						|
allrcb  = zip (repeat "row") allrows
 | 
						|
          ++ zip (repeat "col") allcols
 | 
						|
          ++ zip (repeat "box") allboxs
 | 
						|
 | 
						|
 | 
						|
containers :: [(Position -> [Position], String)]
 | 
						|
containers = [(row, "row"), (col, "col"), (box, "box")]
 | 
						|
 | 
						|
-- ----------------- PRINTING ------------------------------------
 | 
						|
-- printable coordinate of field, upper left is a1, lower right is i9
 | 
						|
pname p = packed [chr (ord 'a' + p `mod` 9), chr (ord '1' + p `div` 9)]
 | 
						|
 | 
						|
-- print board
 | 
						|
printb b = mapM_ p1line allrows >> println ""
 | 
						|
    where
 | 
						|
        p1line row = do
 | 
						|
                print (joined "" (map pfld line))
 | 
						|
            where line = map (getc b) row
 | 
						|
 | 
						|
-- print field (brief)
 | 
						|
--   ? = no candidate
 | 
						|
--   5 = field is 5
 | 
						|
--   . = some candidates
 | 
						|
pfld [] = "?"
 | 
						|
pfld [x] = show x
 | 
						|
pfld zs = "0"
 | 
						|
 | 
						|
-- print initial/final board
 | 
						|
result msg b = do
 | 
						|
        println ("Result: " ++ msg)
 | 
						|
        print   ("Board: ")
 | 
						|
        printb b
 | 
						|
        return b
 | 
						|
 | 
						|
res012 b = case concatMap (getc b) [0,1,2] of
 | 
						|
    [a,b,c] -> a*100+b*10+c
 | 
						|
    _ -> 9999999
 | 
						|
 | 
						|
-- -------------------------- BOARD ALTERATION ACTIONS ---------------------------------
 | 
						|
-- print a message about what is done to the board and return the new board
 | 
						|
turnoff1 :: Position -> Zelle -> Brett -> IO Brett
 | 
						|
turnoff1 i off b
 | 
						|
    | single nc = do
 | 
						|
            -- print (pname i)
 | 
						|
            -- print ": set to "
 | 
						|
            -- print (head nc)
 | 
						|
            -- println " (naked single)"
 | 
						|
            return newb
 | 
						|
    | otherwise = return newb
 | 
						|
    where
 | 
						|
        cell   = getc b i
 | 
						|
        nc     = filter (`notElem` off) cell
 | 
						|
        newb   = (i, nc) : [ f | f <- b, fst f != i ]
 | 
						|
 | 
						|
turnoff :: Int -> Zelle -> String -> Brett -> IO Brett
 | 
						|
turnoff i off msg b = do
 | 
						|
        -- print (pname i)
 | 
						|
        -- print ": set to "
 | 
						|
        -- print nc
 | 
						|
        -- print " by clearing "
 | 
						|
        -- print off
 | 
						|
        -- print " "
 | 
						|
        -- println  msg
 | 
						|
        return newb
 | 
						|
    where
 | 
						|
        cell   = getc b i
 | 
						|
        nc     = filter (`notElem` off) cell
 | 
						|
        newb   = (i, nc) : [ f | f <- b, fst f != i ]
 | 
						|
 | 
						|
turnoffh ps off msg b = foldM toh b ps
 | 
						|
    where
 | 
						|
        toh b p = turnoff p off msg b
 | 
						|
 | 
						|
setto :: Position -> Element -> String -> Brett -> IO Brett
 | 
						|
setto i n cname b = do
 | 
						|
        -- print (pname i)
 | 
						|
        -- print ": set to "
 | 
						|
        -- print n
 | 
						|
        -- print " (hidden single in "
 | 
						|
        -- print cname
 | 
						|
        -- println ")"
 | 
						|
        return newb
 | 
						|
    where
 | 
						|
        nf     = [n]
 | 
						|
        newb   = (i, nf) : [ f | f <- b, fst f != i ]
 | 
						|
 | 
						|
 | 
						|
-- ----------------------------- SOLVING STRATEGIES ---------------------------------------------
 | 
						|
-- reduce candidate sets that contains numbers already in same row, col or box
 | 
						|
-- This finds (and logs) NAKED SINGLEs in passing.
 | 
						|
reduce b = [  turnoff1 p sss | (p,cell) <- b,               -- for each field
 | 
						|
                unsolved cell,                              --  with more than 1 candidate
 | 
						|
                --       single fields in containers that are candidates of that field
 | 
						|
                sss = [ s | (rcb, _) <- containers, [s] <- map (getc b) (rcb p), s `elem` cell],
 | 
						|
                sss != [] ]                                     -- collect field index, elements to remove from candidate set
 | 
						|
 | 
						|
-- look for a number that appears in exactly 1 candidate set of a container
 | 
						|
-- this number can go in no other place (HIDDEN SINGLE)
 | 
						|
hiddenSingle b = [ setto i n cname |                     -- select index, number, containername
 | 
						|
            (cname, rcb) <- allrcb,                 -- FOR rcb IN allrcb
 | 
						|
            n <- elements,                          --  FOR n IN elements
 | 
						|
            fs     = filter (unsolved • snd) (map (getf b) rcb),
 | 
						|
            occurs  = filter ((n `elem`) • snd) fs,
 | 
						|
            length occurs == 1,
 | 
						|
            (i, _) <- occurs ]
 | 
						|
 | 
						|
-- look for NAKED PAIRS, TRIPLES, QUADS
 | 
						|
nakedPair n b = [ turnoff p t ("(naked tuple in " ++ nm ++ ")") |           -- SELECT pos, tuple, name
 | 
						|
            -- n <- [2,3,4],                    //  FOR n IN [2,3,4]
 | 
						|
            (nm, rcb) <- allrcb,             --    FOR rcb IN containers
 | 
						|
            fs = map (getf b) rcb,              --      let fs = fields for rcb positions
 | 
						|
            u  = (fold union [] . filter unsolved . map snd) fs,   -- let u = union of non single candidates
 | 
						|
            t <- n `outof` u,                   --      FOR t IN n-tuples
 | 
						|
            hit = (filter ((`subset` t) . snd) . filter (unsolved . snd)) fs,
 | 
						|
            length hit == n,
 | 
						|
            (p, cell) <- fs,
 | 
						|
            p `notElem` map fst hit,
 | 
						|
            any (`elem` cell) t
 | 
						|
            ]
 | 
						|
 | 
						|
-- look for HIDDEN PAIRS, TRIPLES or QUADS
 | 
						|
hiddenPair n b = [ turnoff p off ("(hidden " ++ show t ++ " in " ++ nm ++ ")") |           -- SELECT pos, tuple, name
 | 
						|
            -- n <- [2,3,4],                    //  FOR n IN [2,3,4]
 | 
						|
            (nm, rcb) <- allrcb,             --    FOR rcb IN containers
 | 
						|
            fs = map (getf b) rcb,              --      let fs = fields for rcb positions
 | 
						|
            u  = (fold union [] . filter ((>1) . length) . map snd) fs,   -- let u = union of non single candidates
 | 
						|
            t <- n `outof` u,                   --      FOR t IN n-tuples
 | 
						|
            hit = (filter (any ( `elem` t) . snd) . filter (unsolved . snd)) fs,
 | 
						|
            length hit == n,
 | 
						|
            off = (fold union [] . map snd) hit `minus` t,
 | 
						|
            off != [],
 | 
						|
            (p, cell) <- hit,
 | 
						|
            ! (cell `subset` t)
 | 
						|
            ]
 | 
						|
 | 
						|
a `subset` b = all (`elem` b) a
 | 
						|
a `union`  b = uniq (sort (a ++ b))
 | 
						|
a `minus`  b = filter (`notElem` b) a
 | 
						|
a `common` b = filter (`elem` b) a
 | 
						|
n `outof` as
 | 
						|
    | length as < n = []
 | 
						|
    | [] <- as      = []
 | 
						|
    | 1 >= n        = map (:[]) as
 | 
						|
    | (a:bs) <- as  = map (a:) ((n-1) `outof` bs) ++ (n `outof` bs)
 | 
						|
    | otherwise     = undefined  -- cannot happen because either as is empty or not
 | 
						|
 | 
						|
same f a b = b `elem` f a
 | 
						|
 | 
						|
intersectionlist = [(allboxs, row, "box/row intersection"), (allboxs, col, "box/col intersection"),
 | 
						|
                    (allrows ++ allcols, box, "line/box intersection")]
 | 
						|
intersections b = [
 | 
						|
    turnoff pos [c] reason |    -- SELECT position, candidate, reson
 | 
						|
        (from, container, reason) <- intersectionlist,
 | 
						|
        rcb <- from,
 | 
						|
        fs = (filter (unsolved . snd) . map (getf b)) rcb,        -- fs = fields in from with more than 1 candidate
 | 
						|
        c <- (fold union [] • map snd) fs,                          -- FOR c IN union of candidates
 | 
						|
        cpos = (map fst • filter ((c `elem`) • snd)) fs,            -- cpos = positions where c occurs
 | 
						|
        cpos != [],                                                 -- WHERE cpos is not empty
 | 
						|
        all (same container (head cpos)) (tail cpos),               -- WHERE all positions are in the intersection
 | 
						|
        -- we can remove all occurences of c that are in container, but not in from
 | 
						|
        (pos, cell) <- map (getf b) (container (head cpos)),
 | 
						|
        c `elem` cell,
 | 
						|
        pos `notElem` rcb ]
 | 
						|
 | 
						|
 | 
						|
-- look for an XY Wing
 | 
						|
--  - there exists a cell A with candidates X and Y
 | 
						|
--  - there exists a cell B with candidates X and Z that shares a container with A
 | 
						|
--  - there exists a cell C with candidates Y and Z that shares a container with A
 | 
						|
-- reasoning
 | 
						|
--  - if A is X, B will be Z
 | 
						|
--  - if A is Y, C will be Z
 | 
						|
--  - since A will indeed be X or Y -> B or C will be Z
 | 
						|
--  - thus, no cell that can see B and C can be Z
 | 
						|
xyWing board = [ turnoff p [z] ("xy wing " ++ pname b ++ " " ++ pname c ++ " because of " ++ pname a) |
 | 
						|
        (a, [x,y]) <- board,                            -- there exists a cell a with candidates x and y
 | 
						|
        rcba = map (getf board) (row a ++ col a ++ box a),  -- rcba = all fields that share a container with a
 | 
						|
        (b, [b1, b2]) <- rcba,
 | 
						|
        b != a,
 | 
						|
        b1 == x && b2 != y || b2 == x && b1 != y,       -- there exists a cell B with candidates x and z
 | 
						|
        z = if b1 == x then b2 else b1,
 | 
						|
        (c, [c1, c2]) <- rcba,
 | 
						|
        c != a, c!= b,
 | 
						|
        c1 == y && c2 == z || c1 == z && c2 == y,       -- there exists a cell C with candidates y and z
 | 
						|
        ps = (uniq . sort) ((row b ++ col b ++ box b) `common` (row c ++ col c ++ box c)),
 | 
						|
        -- remove z in ps
 | 
						|
        (p, cs) <- map (getf board) ps,
 | 
						|
        p != b, p != c,
 | 
						|
        z `elem` cs ]
 | 
						|
 | 
						|
-- look for a N-Fish (2: X-Wing, 3: Swordfish, 4: Jellyfish)
 | 
						|
-- When all candidates for a particular digit in N rows are located
 | 
						|
-- in only N columns, we can eliminate all candidates from those N columns
 | 
						|
--  which are not located on those N rows
 | 
						|
fish n board = fish "row" allrows row col ++ fish "col" allcols col row where
 | 
						|
    fishname 2 = "X-Wing"
 | 
						|
    fishname 3 = "Swordfish"
 | 
						|
    fishname 4 = "Jellyfish"
 | 
						|
    fishname _ = "unknown fish"
 | 
						|
    fish nm allrows row col = [ turnoff p [x] (fishname n ++ " in " ++ nm ++ " " ++ show (map (pname . head) rset)) |
 | 
						|
        rset <- n `outof` allrows,          -- take n rows (or cols)
 | 
						|
        x <- elements,                      -- look for certain number
 | 
						|
        rflds = map (filter ((>1) . length . snd) . map (getf board)) rset,       -- unsolved fields in the rowset
 | 
						|
        colss  = (map (map (head . col . fst) . filter ((x `elem`) . snd)) rflds),   -- where x occurs in candidates
 | 
						|
        all ((>1) . length) colss,         -- x must appear in at least 2 cols
 | 
						|
        cols = fold union [] colss,
 | 
						|
        length cols == n,
 | 
						|
        cstart <- cols,
 | 
						|
        (p, cell) <- map (getf board) (col cstart),
 | 
						|
        x `elem` cell,
 | 
						|
        all (p `notElem`) rset]
 | 
						|
 | 
						|
 | 
						|
-- compute immediate consequences of an assumption of the form (p `IS` e) or (p `ISNOT` e)
 | 
						|
conseq board (IS p e) = uniq (sort ([ p `ISNOT` x | x <- getc board p, x != e ] ++
 | 
						|
    [ a `ISNOT` e |
 | 
						|
        (a,cs) <- map (getf board) (row p ++ col p ++ box p),
 | 
						|
        a != p,
 | 
						|
        e `elem` cs
 | 
						|
    ]))
 | 
						|
conseq board (ISNOT p  e) = uniq (sort ([ p `IS` x | cs = getc board p, length cs == 2, x <- cs, x != e ] ++
 | 
						|
    [ a `IS` e |
 | 
						|
        cp <- [row p, box p, col p],
 | 
						|
        as = (filter ((e `elem`) . getc board) . filter (p!=)) cp,
 | 
						|
        length as == 1,
 | 
						|
        a = head as
 | 
						|
    ]))
 | 
						|
 | 
						|
-- check if two assumptions contradict each other
 | 
						|
contradicts (IS a x)    (IS b y)    = a==b && x!=y
 | 
						|
contradicts (IS a x)    (ISNOT b y) = a==b && x==y
 | 
						|
contradicts (ISNOT a x) (IS b y)    = a==b && x==y
 | 
						|
contradicts (ISNOT _ _) (ISNOT _ _) = false
 | 
						|
 | 
						|
-- get the Position of an Assumption
 | 
						|
aPos (IS p _)    = p
 | 
						|
aPos (ISNOT p _) = p
 | 
						|
 | 
						|
-- get List of elements that must be turned off when assumption is true/false
 | 
						|
toClear board true  (IS p x)    = filter (x!=) (getc board p)
 | 
						|
toClear board false (IS p x)    = [x]
 | 
						|
toClear board true  (ISNOT p x) = [x]
 | 
						|
toClear board false (ISNOT p x) = filter (x!=) (getc board p)
 | 
						|
 | 
						|
 | 
						|
-- look for assumptions whose implications contradict themself
 | 
						|
chain board paths = [ solution a (head cs) (reverse cs) |
 | 
						|
        (a, css) <-  paths,
 | 
						|
        cs <- take 1 [ cs | cs <- css, contradicts a (head cs) ]
 | 
						|
        ]
 | 
						|
    where
 | 
						|
        solution a c cs = turnoff (aPos a) (toClear board false a) reason where
 | 
						|
            reason = "Assumption " ++ show a ++ " implies " ++ show c ++ "\n\t"
 | 
						|
                ++ showcs cs ++ "\n\t"
 | 
						|
                ++ "Therefore, " ++ show a ++ " must be false."
 | 
						|
 | 
						|
-- look for an assumption that yields to contradictory implications
 | 
						|
-- this assumption must be false
 | 
						|
chainContra board paths = [ solution a (reverse pro) (reverse contra) |
 | 
						|
        (a, css) <- paths,          -- FOR ALL assumptions "a" with list of conlusions "css"
 | 
						|
        (pro, contra) <- take 1 [ (pro, contra) |
 | 
						|
            pro <- (uniqBy (using head) . sortBy (comparing head)) css,                 -- FOR ALL conslusion chains "pro"
 | 
						|
            c = head pro,               -- LET "c" BE the final conclusion
 | 
						|
            contra <- take 1 (filter ((contradicts c) . head) css)   -- THE FIRST conclusion that contradicts c
 | 
						|
        ]
 | 
						|
      ]
 | 
						|
    where
 | 
						|
        solution a pro con = turnoff (aPos a) (toClear board false a) reason where
 | 
						|
            reason = ("assumption " ++ show a ++ " leads to contradictory conclusions\n\t"
 | 
						|
                        ++ showcs pro ++ "\n\t" ++ showcs con)
 | 
						|
 | 
						|
 | 
						|
 | 
						|
-- look for a common implication c of some assumptions ai, where at least 1 ai is true
 | 
						|
-- so that (a0 OR a1 OR a2 OR ...) IMPLIES c
 | 
						|
-- For all cells pi in same container that have x as candidate, we can construct (p0==x OR p1==x OR ... OR pi==x)
 | 
						|
-- For a cell p with candidates ci, we can construct (p==c0 OR p==c1)
 | 
						|
cellRegionChain board paths = [ solution b as (map head os) |
 | 
						|
        as <- cellas ++ regionas,           -- one of as must be true
 | 
						|
        iss = filter ((`elem` as) . fst) paths,    -- the implications for as
 | 
						|
        (a, ass) <- take 1 iss,             -- implications for first assumption
 | 
						|
        fs <- (uniqBy (using head) . sortBy (comparing head)) ass,
 | 
						|
        b = head fs,                        -- final conclusions of first assumption
 | 
						|
        os = [fs] : map (take 1 . filter ((b==) . head) . snd) (tail iss), -- look for implications with same conclusion
 | 
						|
        all ([]!=) os]
 | 
						|
    where
 | 
						|
        cellas   = [ map (p `IS`) candidates | (p, candidates@(_:_:_)) <- board ]
 | 
						|
        regionas = [ map (`IS` e) ps |
 | 
						|
            region <- map (map (getf board)) (allrows ++ allcols ++ allboxs),
 | 
						|
            e <- elements,
 | 
						|
            ps = map fst (filter ((e `elem`) . snd) region),
 | 
						|
            length ps > 1 ]
 | 
						|
        solution b as oss = turnoff (aPos b) (toClear board true b) reason where
 | 
						|
            reason = "all of the assumptions " ++ joined ", " (map show as) ++ " imply " ++ show b ++ "\n\t"
 | 
						|
                ++ joined "\n\t" (map (showcs . reverse) oss) ++ "\n\t"
 | 
						|
                ++ "One of them must be true, so " ++ show b ++ " must be true."
 | 
						|
 | 
						|
 | 
						|
{-
 | 
						|
    Wir brauchen für einige Funktionen eine Datenstruktur wie
 | 
						|
        [ (Assumption, [[Assumption]]) ]
 | 
						|
    d.i. eine Liste von möglichen Annahmen samt aller Schlußketten.
 | 
						|
    Idealerweise sollte die Schlußkette in umgekehrter Reihenfolge vorliegen,
 | 
						|
    dann kann man einfach finden:
 | 
						|
    - Annahmen, die zum Selbstwiderspruch führen.
 | 
						|
    - alles, was aus einer bestimmten Annahme folgt (map (map head) [[a]])
 | 
						|
    -...
 | 
						|
-}
 | 
						|
--- Liste aller Annahmen für ein bestimmtes Brett
 | 
						|
assumptions :: Brett -> [Assumption]
 | 
						|
assumptions board = [ a |
 | 
						|
                (p, cs) <- board,
 | 
						|
                !(single cs),
 | 
						|
                a <- map (ISNOT p) cs ++ map (IS p) cs ]
 | 
						|
 | 
						|
consequences :: Brett -> [Assumption] -> [[Assumption]]
 | 
						|
consequences board as = map (conseq board) as
 | 
						|
 | 
						|
acstree :: Brett -> Tree Assumption [Assumption]
 | 
						|
acstree board = Tree.fromList (zip as cs)
 | 
						|
    where
 | 
						|
        as = assumptions  board
 | 
						|
        cs = consequences board as
 | 
						|
 | 
						|
-- bypass maybe on tree lookup
 | 
						|
find :: Tree Assumption [Assumption] -> Assumption -> [Assumption]
 | 
						|
find t a
 | 
						|
    | Just cs <- t.lookup a = cs
 | 
						|
    | otherwise = error ("no consequences for " ++ show a)
 | 
						|
 | 
						|
-- for performance resons, we confine ourselves to implication chains of length 20 per assumption
 | 
						|
mkPaths :: Tree Assumption [Assumption] -> [ (Assumption, [[Assumption]]) ]
 | 
						|
mkPaths acst = map impl  (keys acst)   -- {[a1], [a2], [a3] ]
 | 
						|
    where
 | 
						|
        -- [Assumption] -> [(a, [chains, ordered by length]
 | 
						|
        impl a = (a, impls [[a]])
 | 
						|
        impls ns = (take 1000 • concat • takeUntil null • iterate expandchain) ns
 | 
						|
        -- expandchain :: [[Assumption]] -> [[Assumption]]
 | 
						|
        expandchain css = [ (n:a:as) |
 | 
						|
            (a : as) <- css,               -- list of assumptions
 | 
						|
            n <- find acst a,              -- consequences of a
 | 
						|
            n `notElem` as                 -- avoid loops
 | 
						|
          ]
 | 
						|
        -- uni (a:as) = a : uni (filter ((head a !=) • head) as)
 | 
						|
        -- uni [] = empty
 | 
						|
        -- empty = []
 | 
						|
 | 
						|
 | 
						|
-- ------------------ SOLVE A SUDOKU --------------------------
 | 
						|
-- Apply all available strategies until nothing changes anymore
 | 
						|
-- Strategy functions are supposed to return a list of
 | 
						|
-- functions, which, when applied to a board, give a changed board.
 | 
						|
-- When a strategy does not find anything to alter,
 | 
						|
-- it returns [], and the next strategy can be tried.
 | 
						|
solve b
 | 
						|
    | all (single . snd) b       = result "Solved" b
 | 
						|
    | any (([]==) . snd) b       = result "not solvable" b
 | 
						|
    | res@(_:_) <- reduce b       = apply b res >>=solve       -- compute smallest candidate sets
 | 
						|
    -- comment "candidate sets are up to date" = ()
 | 
						|
    | res@(_:_) <- hiddenSingle b  = apply b res >>= solve     -- find HIDDEN SINGLES
 | 
						|
    -- comment "no more hidden singles" = ()
 | 
						|
    | res@(_:_) <- intersections b = apply b res >>= solve     -- find locked candidates
 | 
						|
    -- comment "no more intersections" = ()
 | 
						|
    | res@(_:_) <- nakedPair 2 b     = apply b res >>= solve     -- find NAKED PAIRS, TRIPLES or QUADRUPELS
 | 
						|
    -- comment "no more naked pairs" = ()
 | 
						|
    | res@(_:_) <- hiddenPair  2 b   = apply b res >>= solve      -- find HIDDEN PAIRS, TRIPLES or QUADRUPELS
 | 
						|
    -- comment "no more hidden pairs" = ()
 | 
						|
    -- res@(_:_) <- nakedPair 3 b     = apply b res >>= solve       // find NAKED PAIRS, TRIPLES or QUADRUPELS
 | 
						|
    -- | comment "no more naked triples" = ()
 | 
						|
    -- res@(_:_) <- hiddenPair  3 b    = apply b res >>= solve      // find HIDDEN PAIRS, TRIPLES or QUADRUPELS
 | 
						|
    -- | comment "no more hidden triples" = ()
 | 
						|
    -- res@(_:_) <- nakedPair 4 b     = apply b res >>=solve       // find NAKED PAIRS, TRIPLES or QUADRUPELS
 | 
						|
    -- | comment "no more naked quadruples" = ()
 | 
						|
    -- res@(_:_) <- hiddenPair  4 b    = apply b res >>=solve      // find HIDDEN PAIRS, TRIPLES or QUADRUPELS
 | 
						|
    -- | comment "no more hidden quadruples" = ()
 | 
						|
    | res@(_:_) <- xyWing b            = apply b res >>=solve      -- find XY WINGS
 | 
						|
    -- comment "no more xy wings"       = ()
 | 
						|
    | res@(_:_) <- fish 2 b            = apply b res >>=solve      -- find 2-FISH
 | 
						|
    -- comment "no more x-wings"        = ()
 | 
						|
    -- res@(_:_) <- fish 3 b            = apply b res >>=solve      // find 3-FISH
 | 
						|
    -- | comment "no more swordfish"      = ()
 | 
						|
    -- res@(_:_) <- fish 4 b            = apply b res >>=solve      // find 4-FISH
 | 
						|
    -- | comment "no more jellyfish"      = ()
 | 
						|
    -- | comment pcomment                 = ()
 | 
						|
    | res@(_:_) <- chain b paths             = apply b (take 9 res) >>= solve  -- find forcing chains
 | 
						|
    | res@(_:_) <- cellRegionChain b paths   = apply b (take 9 res) >>= solve  -- find common conclusion for true assumption
 | 
						|
    | res@(_:_) <- chainContra b paths       = apply b (take 9 res) >>= solve  -- find assumptions that allow to infer both a and !a
 | 
						|
    -- comment "consistent conclusions only"       = ()
 | 
						|
 | 
						|
    | otherwise = result "ambiguous" b
 | 
						|
    where
 | 
						|
        apply brd fs = foldM (\b\f -> f b) brd fs
 | 
						|
        paths = mkPaths (acstree b)
 | 
						|
        -- pcomment = show (length paths) ++ " assumptions with " ++ show (fold (+) 0 (map (length <~ snd) paths))
 | 
						|
        --    ++ " implication chains"
 | 
						|
 | 
						|
-- comment com = do stderr << com << "\n" for false
 | 
						|
-- log com     = do stderr << com << "\n" for true
 | 
						|
 | 
						|
--- turn a string into a row
 | 
						|
mkrow :: String -> [Zelle]
 | 
						|
mkrow s = mkrow1 xs
 | 
						|
    where
 | 
						|
        xs = s ++ "---------" -- make sure at least 9 elements
 | 
						|
        mkrow1 xs = (take 9 • filter ([]!=) • map f • unpacked) xs
 | 
						|
        f x | x >= '1' && x <= '9'  =  [ord x - ord '0']
 | 
						|
            | x == ' '  = []    -- ignored
 | 
						|
            | otherwise = elements
 | 
						|
 | 
						|
main ["-h"]    = main []
 | 
						|
main ["-help"] = main []
 | 
						|
main [] = do
 | 
						|
        mapM_ stderr.println [
 | 
						|
            "usage: java Sudoku file ...",
 | 
						|
            "       java Sudoku position",
 | 
						|
            "where position is a 81 char string consisting of digits",
 | 
						|
            "One can get such a string by going to",
 | 
						|
            "http://www.sudokuoftheday.com/pages/s-o-t-d.php",
 | 
						|
            "Right click on the puzzle and open it in new tab",
 | 
						|
            "Copy the 81 digits from the URL in the address field of your browser.",
 | 
						|
            "",
 | 
						|
            "There is also a file with hard sudokus in examples/top95.txt\n"]
 | 
						|
        return ()
 | 
						|
 | 
						|
 | 
						|
main [s@#^[0-9\W]{81}$#] = solve board >> return ()
 | 
						|
    where
 | 
						|
        board = zip positions felder
 | 
						|
        felder = decode s
 | 
						|
 | 
						|
main files = forM_ files sudoku
 | 
						|
    where
 | 
						|
        sudoku file = do
 | 
						|
            br <- openReader file
 | 
						|
            lines <- BufferedReader.getLines br
 | 
						|
            bs <- process lines
 | 
						|
            ss <- mapM (\b -> print "Puzzle: " >> printb b >> solve b) bs
 | 
						|
            println ("Euler: " ++ show (sum (map res012 ss)))
 | 
						|
            return ()
 | 
						|
 | 
						|
-- "--3-" => [1..9, 1..9, [3], 1..9]
 | 
						|
decode s = map candi (unpacked s) where
 | 
						|
        candi c | c >= '1' && c <= '9'  = [(ord c - ord '0')]
 | 
						|
                | otherwise = elements
 | 
						|
process [] = return []
 | 
						|
process (s:ss)
 | 
						|
    | length s == 81 = consider b1
 | 
						|
    | length s == 9,
 | 
						|
      length acht == 8,
 | 
						|
      all ((9==) • length) acht = consider b2
 | 
						|
    | otherwise = do
 | 
						|
            stderr.println ("skipped line: " ++ s)
 | 
						|
            process ss
 | 
						|
    where
 | 
						|
        acht = take 8 ss
 | 
						|
        neun = fold (++) "" (s:acht)
 | 
						|
        b1 = zip positions (decode s)
 | 
						|
        b2 = zip positions (decode neun)
 | 
						|
        consider b = do
 | 
						|
            -- print "Puzzle: "
 | 
						|
            -- printb b
 | 
						|
            bs <- process ss
 | 
						|
            return (b:bs)
 | 
						|
 |