Fixing bug in disambiguation in regexps

diff --git a/basis/regexp/classes/classes-tests.factor b/basis/regexp/classes/classes-tests.factor
index 4cbb2e7a574f54d660a34fcef73664387eaa0934..5eac0ea352710f0515a0de2439695d67170d79f8 100644 (file)
--- a/basis/regexp/classes/classes-tests.factor
+++ b/basis/regexp/classes/classes-tests.factor
@@ -23,3 +23,4 @@ IN: regexp.classes.tests
 [ 1 ] [ { 1 1 } <or-class> ] unit-test
 [ T{ primitive-class { class letter-class } } ] [ letter-class <primitive-class> dup 2array <and-class> ] unit-test
 [ T{ primitive-class { class letter-class } } ] [ letter-class <primitive-class> dup 2array <or-class> ] unit-test
+[ T{ or-class { seq { 2 3 1 } } } ] [ { 1 2 } <or-class> { 2 3 } <or-class> 2array <or-class> ] unit-test

diff --git a/basis/regexp/classes/classes.factor b/basis/regexp/classes/classes.factor
index 978be2c369bbab199e2c008fdb5de3de250da666..33652f760671fdc5251fef5e3fdf84daecf3f44e 100644 (file)
--- a/basis/regexp/classes/classes.factor
+++ b/basis/regexp/classes/classes.factor
@@ -1,7 +1,8 @@
 ! Copyright (C) 2008, 2009 Doug Coleman, Daniel Ehrenberg.
 ! See http://factorcode.org/license.txt for BSD license.
 USING: accessors kernel math math.order words combinators locals
-ascii unicode.categories combinators.short-circuit sequences ;
+ascii unicode.categories combinators.short-circuit sequences
+fry macros arrays ;
 IN: regexp.classes
 
 SINGLETONS: any-char any-char-no-nl
@@ -150,6 +151,12 @@ M: not-class combine-or
 M: integer combine-or
     2dup swap class-member? [ drop t ] [ 2drop f f ] if ;
 
+MACRO: instance? ( class -- ? )
+    "predicate" word-prop ;
+
+: flatten ( seq class -- newseq )
+    '[ dup _ instance? [ seq>> ] [ 1array ] if ] map concat ; inline
+
 : try-combine ( elt1 elt2 quot -- combined/f ? )
     3dup call [ [ 3drop ] dip t ] [ drop swapd call ] if ; inline
 
@@ -160,7 +167,8 @@ M: integer combine-or
     [ seq elt prefix ] if* ; inline
 
 :: combine ( seq quot: ( elt1 elt2 -- combined/f ? ) empty class -- newseq )
-    seq { } [ quot prefix-combining ] reduce
+    seq class flatten
+    { } [ quot prefix-combining ] reduce
     dup length {
         { 0 [ drop empty ] }
         { 1 [ first ] }
@@ -179,12 +187,19 @@ M: and-class class-member?
 M: or-class class-member?
     seq>> [ class-member? ] with any? ;
 
-: <not-class> ( class -- inverse )
-    {
-        { t [ f ] }
-        { f [ t ] }
-        [ dup not-class? [ class>> ] [ not-class boa ] if ]
-    } case ;
+GENERIC: <not-class> ( class -- inverse )
+
+M: object <not-class>
+    not-class boa ;
+
+M: not-class <not-class>
+    class>> ;
+
+M: and-class <not-class>
+    seq>> [ <not-class> ] map <or-class> ;
+
+M: or-class <not-class>
+    seq>> [ <not-class> ] map <and-class> ;
 
 M: not-class class-member?
     class>> class-member? not ;
@@ -192,4 +207,4 @@ M: not-class class-member?
 M: primitive-class class-member?
     class>> class-member? ;
 
-UNION: class primitive-class not-class or-class range ;
+UNION: class primitive-class not-class or-class and-class range ;

diff --git a/basis/regexp/disambiguate/disambiguate.factor b/basis/regexp/disambiguate/disambiguate.factor
index b8c03d7a3b8e396cc32fe72d7c3b710ad45c9032..abfe76d832d217bb884e17c5bae2ed391c7c5e0b 100644 (file)
--- a/basis/regexp/disambiguate/disambiguate.factor
+++ b/basis/regexp/disambiguate/disambiguate.factor
@@ -1,7 +1,7 @@
 ! Copyright (C) 2009 Daniel Ehrenberg.
 ! See http://factorcode.org/license.txt for BSD license.
 USING: kernel accessors regexp.classes math.bits assocs sequences
-arrays sets regexp.dfa math fry regexp.minimize ;
+arrays sets regexp.dfa math fry regexp.minimize regexp.ast ;
 IN: regexp.disambiguate
 
 TUPLE: parts in out ;
@@ -20,22 +20,28 @@ TUPLE: parts in out ;
     prefix <and-class> ;
 
 : get-transitions ( partition state-transitions -- next-states )
-    [ in>> ] dip '[ _ at ] map prune ;
+    [ in>> ] dip '[ _ at ] gather sift ;
 
-: disambiguate ( dfa -- nfa )  
+: new-transitions ( transitions -- assoc ) ! assoc is class, partition
+    values [ keys ] gather
+    [ tagged-epsilon? not ] filter
+    powerset-partition
+    [ [ partition>class ] keep ] { } map>assoc
+    [ drop ] assoc-filter ;
+
+: preserving-epsilon ( state-transitions quot -- new-state-transitions )
+    [ [ drop tagged-epsilon? ] assoc-filter ] bi
+    assoc-union H{ } assoc-like ; inline
+
+: disambiguate ( nfa -- nfa )  
     [
-        [
-            [ keys powerset-partition ] keep '[
-                [ partition>class ]
-                [ _ get-transitions ] bi
-            ] H{ } map>assoc
-            [ drop ] assoc-filter 
+        dup new-transitions '[
+            [
+                _ swap '[ _ get-transitions ] assoc-map
+                [ nip empty? not ] assoc-filter 
+            ] preserving-epsilon
         ] assoc-map
     ] change-transitions ;
 
-USE: sorting
-
 : nfa>dfa ( nfa -- dfa )
-    construct-dfa minimize
-    disambiguate
-    construct-dfa minimize ;
+    disambiguate construct-dfa minimize ;

diff --git a/basis/regexp/nfa/nfa.factor b/basis/regexp/nfa/nfa.factor
index 55147a1d269fca1cc72d2e694de28273cecab5bf..68f7761394c93ca1df45316e4928eff23831b36b 100644 (file)
--- a/basis/regexp/nfa/nfa.factor
+++ b/basis/regexp/nfa/nfa.factor
@@ -11,19 +11,6 @@ IN: regexp.nfa
 ! but case-insensitive matching should be done by case-folding everything
 ! before processing starts
 
-GENERIC: remove-lookahead ( syntax-tree -- syntax-tree' )
-! This is unfinished and does nothing right now!
-
-M: object remove-lookahead ;
-
-M: with-options remove-lookahead
-    [ tree>> remove-lookahead ] [ options>> ] bi <with-options> ;
-
-M: alternation remove-lookahead
-    [ first>> ] [ second>> ] bi [ remove-lookahead ] bi@ alternation boa ;
-
-M: concatenation remove-lookahead ;
-
 SYMBOL: option-stack
 
 SYMBOL: state
@@ -148,7 +135,7 @@ M: with-options nfa-node ( node -- start end )
     [
         0 state set
         <transition-table> nfa-table set
-        remove-lookahead nfa-node
+        nfa-node
         nfa-table get
             swap dup associate >>final-states
             swap >>start-state