Completed unoptimized SLF generation.

1 files changed, 23 insertions, 127 deletions

diff --git a/src/GF/Speech/TransformCFG.hs b/src/GF/Speech/TransformCFG.hs
index 5f1e4fb97..57d8ec87b 100644
--- a/src/GF/Speech/TransformCFG.hs
+++ b/src/GF/Speech/TransformCFG.hs
@@ -5,9 +5,9 @@
 -- Stability   : (stable)
 -- Portability : (portable)
 --
--- > CVS $Date: 2005/09/08 15:45:17 $ 
+-- > CVS $Date: 2005/09/12 15:46:44 $ 
 -- > CVS $Author: bringert $
--- > CVS $Revision: 1.19 $
+-- > CVS $Revision: 1.20 $
 --
 --  This module does some useful transformations on CFGs.
 --
@@ -16,12 +16,12 @@
 -- peb thinks: most of this module should be moved to GF.Conversion...
 -----------------------------------------------------------------------------
 
-module GF.Speech.TransformCFG (CFRule_, CFRules, 
+-- FIXME: lots of this stuff is used by CFGToFiniteState, thus
+-- the missing explicit expot list.
+module GF.Speech.TransformCFG {- (CFRule_, CFRules, 
 			       cfgToCFRules, getStartCat,
 			       removeLeftRecursion,
-			       removeEmptyCats,
-			       makeRegular, 
-			       compileAutomaton) where
+			       removeEmptyCats, removeIdenticalRules) -} where
 
 import GF.Infra.Ident
 import GF.Formalism.CFG 
@@ -62,8 +62,6 @@ groupProds = fmToList . addListToFM_C (++) emptyFM . map (\r -> (lhsCat r,[r]))
 ungroupProds :: CFRules -> [CFRule_]
 ungroupProds = concat . map snd
 
-catRules :: CFRules -> Cat_ -> [CFRule_]
-catRules rs c = fromMaybe [] (lookup c rs)
 
 -- | Remove productions which use categories which have no productions
 removeEmptyCats :: CFRules -> CFRules
@@ -77,13 +75,18 @@ removeEmptyCats = fix removeEmptyCats'
 	emptyCats = filter (nothingOrNull . flip lookup rs) allCats
 	k' = map (\ (c,xs) -> (c, filter (not . anyUsedBy emptyCats) xs)) keep
 
+-- | Remove rules which are identical, not caring about the rule names.
+removeIdenticalRules :: CFRules -> CFRules
+removeIdenticalRules g = [(c,nubBy sameCatAndRhs rs) | (c,rs) <- g]
+    where sameCatAndRhs (CFRule c1 ss1 _) (CFRule c2 ss2 _) = c1 == c2 && ss1 == ss2
+
 removeLeftRecursion :: CFRules -> CFRules
 removeLeftRecursion rs = concatMap removeDirectLeftRecursion $ map handleProds rs
     where 
     handleProds (c, r) = (c, concatMap handleProd r)
     handleProd (CFRule ai (Cat aj:alpha) n) | aj < ai  =
               -- FIXME: this will give multiple rules with the same name
-             [CFRule ai (beta ++ alpha) n | CFRule _ beta _ <- fromJust (lookup aj rs)]
+             [CFRule ai (beta ++ alpha) n | CFRule _ beta _ <- lookup' aj rs]
     handleProd r = [r]
 
 removeDirectLeftRecursion :: (Cat_,[CFRule_]) -- ^ All productions for a category
@@ -103,92 +106,22 @@ isDirectLeftRecursive (CFRule c (Cat c':_) _) = c == c'
 isDirectLeftRecursive _ = False
 
 
--- Use the transformation algorithm from \"Regular Approximation of Context-free
--- Grammars through Approximation\", Mohri and Nederhof, 2000
--- to create an over-generating regular frammar for a context-free 
--- grammar
-makeRegular :: CFRules -> CFRules
-makeRegular g = groupProds $ concatMap trSet (mutRecCats g)
-  where trSet cs | allXLinear cs rs = rs
-		 | otherwise = concatMap handleCat cs
-	    where rs = concatMap (catRules g) cs
-		  handleCat c = [CFRule c' [] (mkName (c++"-empty"))] -- introduce A' -> e
-				++ concatMap (makeRightLinearRules c) (catRules g c)
-		      where c' = newCat c
-		  makeRightLinearRules b' (CFRule c ss n) = 
-		      case ys of
-			      [] -> [CFRule b' (xs ++ [Cat (newCat c)]) n] -- no non-terminals left
-			      (Cat b:zs) -> CFRule b' (xs ++ [Cat b]) n 
-					: makeRightLinearRules (newCat b) (CFRule c zs n)
-		      where (xs,ys) = break (`catElem` cs) ss
-	newCat c = c ++ "$"
-
-
--- | Get the sets of mutually recursive non-terminals for a grammar.
-mutRecCats :: CFRules -> [[Cat_]]
-mutRecCats g = equivalenceClasses $ symmetricSubrelation $ transitiveClosure $ reflexiveClosure allCats r
-  where r = nub [(c,c') | (_,rs) <- g, CFRule c ss _ <- rs, Cat c' <- ss]
-	allCats = map fst g
-
-
-
--- Convert a strongly regular grammar to a finite automaton.
-compileAutomaton :: Cat_ -- ^ Start category
-		 -> CFRules
-		 -> FA () (Maybe Token)
-compileAutomaton start g = make_fa s [Cat start] f g fa''
-  where fa = newFA ()
-	s = startState fa
-	(fa',f) = newState () fa
-	fa'' = addFinalState f fa'
-
--- | The make_fa algorithm from \"Regular approximation of CFLs: a grammatical view\",
---   Mark-Jan Nederhof. International Workshop on Parsing Technologies, 1997.
-make_fa :: State -> [Symbol Cat_ Token] -> State 
-	-> CFRules -> FA () (Maybe Token) -> FA () (Maybe Token)
-make_fa q0 a q1 g fa = 
-    case a of
-	   []       -> newTrans q0 q1 Nothing fa
-	   [Tok t]  -> newTrans q0 q1 (Just t) fa
-	   [Cat c]  -> undefined
-	   (x:beta) -> let (fa',q) = newState () fa
-			   fa'' = make_fa q0 [x] q g fa'
-			   fa''' = make_fa q beta q1 g fa''
-			   in fa'''
-
 --
 -- * CFG rule utilities
 --
 
-{-
--- | Get all the rules for a given category.
-catRules :: Eq c => [CFRule c n t] -> c -> [CFRule c n t]
-catRules rs c = [r | r@(CFRule c' _ _) <- rs, c' == c]
--}
+catRules :: CFRules -> Cat_ -> [CFRule_]
+catRules rs c = fromMaybe [] (lookup c rs)
 
--- | Gets the set of LHS categories of a set of rules.
-lhsCats :: Eq c => [CFRule c n t] -> [c]
-lhsCats = nub . map lhsCat
+catSetRules :: CFRules -> [Cat_] -> [CFRule_]
+catSetRules g s = concatMap (catRules g) s
 
 lhsCat :: CFRule c n t -> c
 lhsCat (CFRule c _ _) = c
 
--- | Check if all the rules are right-linear, or all the rules are
---   left-linear, with respect to given categories.
-allXLinear :: Eq c => [c] -> [CFRule c n t] -> Bool
-allXLinear cs rs = all (isRightLinear cs) rs || all (isLeftLinear cs) rs
+ruleRhs :: CFRule c n t ->  [Symbol c t]
+ruleRhs (CFRule _ ss _) = ss
 
--- | Checks if a context-free rule is right-linear.
-isRightLinear :: Eq c => [c]  -- ^ The categories to consider
-	      -> CFRule c n t -- ^ The rule to check for right-linearity
-	      -> Bool
-isRightLinear cs (CFRule _ ss _) = all (not . (`catElem` cs)) (safeInit ss)
-
--- | Checks if a context-free rule is left-linear.
-isLeftLinear ::  Eq c => [c]  -- ^ The categories to consider
-	      -> CFRule c n t -- ^ The rule to check for right-linearity
-	      -> Bool
-isLeftLinear cs (CFRule _ ss _) = all (not . (`catElem` cs)) (drop 1 ss)
 
 -- | Checks if a symbol is a non-terminal of one of the given categories.
 catElem :: Eq c => Symbol c t -> [c] -> Bool
@@ -202,37 +135,14 @@ anyUsedBy cs (CFRule _ ss _) = any (`elem` cs) (filterCats ss)
 mkName :: String -> Name
 mkName n = Name (IC n) []
 
---
--- * Relations
---
-
--- FIXME: these could use a more efficent data structures and algorithms.
-
-isRelatedTo :: Eq a => [(a,a)] -> a  -> a -> Bool
-isRelatedTo r x y = (x,y) `elem` r
-
-transitiveClosure :: Eq a => [(a,a)] -> [(a,a)]
-transitiveClosure r = fix (\r -> r `union` [ (x,w) | (x,y) <- r, (z,w) <- r, y == z ]) r
-
-reflexiveClosure :: Eq a => [a] -- ^ The set over which the relation is defined.
-		 -> [(a,a)] -> [(a,a)]
-reflexiveClosure u r = [(x,x) | x <- u] `union` r
-
-symmetricSubrelation :: Eq a => [(a,a)] -> [(a,a)]
-symmetricSubrelation r = [p | p@(x,y) <- r, (y,x) `elem` r]
-
--- | Get the equivalence classes from an equivalence relation. Since
---   the relation is relexive, the set can be recoved from the relation.
-equivalenceClasses :: Eq a => [(a,a)] -> [[a]]
-equivalenceClasses r = equivalenceClasses_ (nub (map fst r)) r
- where equivalenceClasses_ [] _ = []
-       equivalenceClasses_ (x:xs) r = (x:ys):equivalenceClasses_ zs r
-	   where (ys,zs) = partition (isRelatedTo r x) xs
 
 --
 -- * Utilities
 --
 
+findSet :: Eq c => c -> [[c]] -> Maybe [c]
+findSet x = find (x `elem`)
+
 fix :: Eq a => (a -> a) -> a -> a
 fix f x = let x' = f x in if x' == x then x else fix f x'
 
@@ -240,26 +150,12 @@ nothingOrNull :: Maybe [a] -> Bool
 nothingOrNull Nothing = True
 nothingOrNull (Just xs) = null xs
 
-safeInit :: [a] -> [a]
-safeInit [] = []
-safeInit xs = init xs
-
 unionAll :: Eq a => [[a]] -> [a]
 unionAll = nub . concat
 
 whenMP :: MonadPlus m => Bool -> a -> m a
 whenMP b x = if b then return x else mzero
 
---
--- * Testing stuff, can be removed
--- 
-
-c --> ss = CFRule c ss (mkName "")
-
-prGr g = putStrLn $ showGr g
-
-showGr g = unlines $ map showRule g
-
-showRule (CFRule c ss _) = c ++ " --> " ++ unwords (map showSym ss) 
+lookup' :: Eq a => a -> [(a,b)] -> b
+lookup' x = fromJust . lookup x
 
-showSym s = symbol id show s
\ No newline at end of file