checkpoint harmony
[sbp.git] / src / edu / berkeley / sbp / Parser.java
index a24c523..18bec5a 100644 (file)
@@ -1,12 +1,9 @@
 package edu.berkeley.sbp;
 import edu.berkeley.sbp.*;
 import edu.berkeley.sbp.util.*;
-import edu.berkeley.sbp.*;
 import edu.berkeley.sbp.Sequence.Position;
-import edu.berkeley.sbp.*;
 import java.io.*;
 import java.util.*;
-import java.lang.reflect.*;
 
 /** a parser which translates streams of Tokens of type T into a Forest<R> */
 public abstract class Parser<T extends Token, R> {
@@ -17,63 +14,29 @@ public abstract class Parser<T extends Token, R> {
     protected Parser(Union u)  { this.pt = new Table(u, top()); }
     protected Parser(Table pt) { this.pt = pt; }
 
+    /** implement this method to create the output forest corresponding to a lone shifted input token */
     public abstract Forest<R> shiftedToken(T t, Token.Location loc);
-    public abstract Topology<T> top();
-
 
-    /** parse <tt>input</tt> for a exactly one unique result, throwing <tt>Ambiguous</tt> if not unique or <tt>Failed</tt> if none */
-    public Tree<R> parse1(Token.Stream<T> input) throws IOException, Failed, Ambiguous {
-        Forest<R> ret = parse(input);
-        try { return ret.expand1(); }
-        catch (Ambiguous a) {
-            System.out.println("while expanding:");
-            System.out.println(ret);
-            throw a;
-        }
-    }
+    /** this method must return an empty topology of the input token type */
+    public abstract Topology<T> top();
 
     /** parse <tt>input</tt>, using the table <tt>pt</tt> to drive the parser */
-    public Forest<R> parse(Token.Stream<T> input) throws IOException, Failed {
+    public Forest<R> parse(Token.Stream<T> input) throws IOException, ParseFailed {
         GSS gss = new GSS();
         Token.Location loc = input.getLocation();
-        GSS.Phase current = gss.new Phase(null, input.next(), loc);
-        current.newNode(null, null, pt.start, true, null);
+        GSS.Phase current = gss.new Phase(null, this, null, input.next(1, 0, 0), loc, null);
+        current.newNode(null, Forest.leaf(null, null), pt.start, true);
+        int count = 1;
         for(;;) {
             loc = input.getLocation();
-            GSS.Phase next = gss.new Phase(current, input.next(), loc);
             current.reduce();
             Forest forest = current.token==null ? null : shiftedToken((T)current.token, loc);
-            current.shift(next, forest);
+            GSS.Phase next = gss.new Phase(current, this, current, input.next(count, gss.resets, gss.waits), loc, forest);
+            count = next.hash.size();
             if (current.isDone()) return (Forest<R>)current.finalResult;
-            current.checkFailure();
             current = next;
         }
     }
-    
-
-    // Exceptions //////////////////////////////////////////////////////////////////////////////
-
-    public static class Failed extends Exception {
-        private final Token.Location location;
-        private final String         message;
-        public Failed() { this("", null); }
-        public Failed(String message, Token.Location loc) { this.location = loc; this.message = message; }
-        public Token.Location getLocation() { return location; }
-        public String toString() { return message + (location==null ? "" : (" at " + location)); }
-    }
-
-    public static class Ambiguous extends RuntimeException {
-        public final Forest ambiguity;
-        public Ambiguous(Forest ambiguity) { this.ambiguity = ambiguity; }
-        public String toString() {
-            StringBuffer sb = new StringBuffer();
-            sb.append("unresolved ambiguity "/*"at " + ambiguity.getLocation() + ":"*/);
-            for(Object result : ambiguity.expand(false))
-                sb.append("\n    " + result);
-            return sb.toString();
-        }
-    }
-
 
     // Table //////////////////////////////////////////////////////////////////////////////
 
@@ -81,7 +44,7 @@ public abstract class Parser<T extends Token, R> {
     static class Table extends Walk.Cache {
 
         public final Walk.Cache cache = this;
-        
+
         private void walk(Element e, HashSet<Element> hs) {
             if (e==null) return;
             if (hs.contains(e)) return;
@@ -115,7 +78,7 @@ public abstract class Parser<T extends Token, R> {
             HashSet<Element>                   all_elements  = new HashSet<Element>();
             walk(start0, all_elements);
             for(Element e : all_elements)
-                cache.ys.put(e, new Walk.YieldSet(e, cache).walk());
+                cache.ys.addAll(e, new Walk.YieldSet(e, cache).walk());
             HashSet<Position> hp = new HashSet<Position>();
             reachable(start0, hp);
             this.start = new State(hp, all_states, all_elements);
@@ -128,12 +91,22 @@ public abstract class Parser<T extends Token, R> {
                     if (start0.contains(p.owner()) && p.next()==null)
                         state.accept = true;
 
-                    // FIXME: how does right-nullability interact with follow restrictions?
-                    // all right-nullable rules get a reduction [Johnstone 2000]
-                    if (p.isRightNullable(cache)) {
+                    if (isRightNullable(p)) {
                         Walk.Follow wf = new Walk.Follow(top.empty(), p.owner(), all_elements, cache);
                         Reduction red = new Reduction(p);
-                        state.reductions.put(wf.walk(p.owner()), red);
+
+                        Topology follow = wf.walk(p.owner());
+                        if (p.owner() instanceof Sequence.RewritingSequence &&
+                            (((Sequence.RewritingSequence)p.owner()).tag+"").equals("emailaddr")) {
+                            System.out.println("follow before: " + new edu.berkeley.sbp.misc.CharToken.CharRange(follow));
+                        }
+                        for(Position p2 = p; p2 != null && p2.element() != null; p2 = p2.next())
+                            follow = follow.intersect(new Walk.Follow(top.empty(), p2.element(), all_elements, cache).walk(p2.element()));
+                        if (p.owner() instanceof Sequence.RewritingSequence &&
+                            (((Sequence.RewritingSequence)p.owner()).tag+"").equals("emailaddr")) {
+                            System.out.println("follow after: " + new edu.berkeley.sbp.misc.CharToken.CharRange(follow));
+                        }
+                        state.reductions.put(follow, red);
                         if (wf.includesEof()) state.eofReductions.add(red);
                     }
 
@@ -142,11 +115,31 @@ public abstract class Parser<T extends Token, R> {
                     if (p.element() != null && p.element() instanceof Atom)
                         state.shifts.addAll(state.gotoSetTerminals.subset(((Atom)p.element())));
                 }
+            for(State state : all_states.values()) {
+                state.oreductions = state.reductions.optimize();
+                state.oshifts = state.shifts.optimize();
+            }
+        }
+
+        private boolean isRightNullable(Position p) {
+            if (p.isLast()) return true;
+            if (!p.element().possiblyEpsilon(this)) return false;
+            return isRightNullable(p.next());
         }
 
         /** a single state in the LR table and the transitions possible from it */
-        public class State implements Comparable<Table.State>, Iterable<Position> {
+
+        public class State implements Comparable<Table.State>, IntegerMappable, Iterable<Position> {
         
+            public int toInt() { return idx; }
+
+            public boolean lame() {
+                for(Position p : this)
+                    for(Position p2 = p; p2!=null; p2=p2.next())
+                        if (p2.isLast() && !p2.owner().lame)
+                            return false;
+                return true;
+            }
             /*
             public boolean isResolvable(Token t) {
                 boolean found = false;
@@ -179,15 +172,26 @@ public abstract class Parser<T extends Token, R> {
             private           TopologicalBag<Token,State>     shifts              = new TopologicalBag<Token,State>();
             private           boolean                         accept              = false;
 
+            private VisitableMap<Token,State> oshifts = null;
+            private VisitableMap<Token,Reduction> oreductions = null;
+
             // Interface Methods //////////////////////////////////////////////////////////////////////////////
 
-            public boolean             canShift(Token t)           { return shifts.contains(t); }
-            public Iterable<State>     getShifts(Token t)          { return shifts.get(t); }
             public boolean             isAccepting()               { return accept; }
-            public Iterable<Reduction> getReductions(Token t)      { return t==null ? eofReductions : reductions.get(t); }
-            public Iterable<Reduction> getEofReductions()          { return eofReductions; }
+
+            public boolean             canShift(Token t)           { return oshifts.contains(t); }
+            public boolean             canReduce(Token t)          { return t==null ? eofReductions.size()>0 : oreductions.contains(t); }
+
             public Iterator<Position>  iterator()                  { return hs.iterator(); }
 
+            public <B,C> void          invokeShifts(Token t, Invokable<State,B,C> irbc, B b, C c) {
+                oshifts.invoke(t, irbc, b, c);
+            }
+            public <B,C> void          invokeReductions(Token t, Invokable<Reduction,B,C> irbc, B b, C c) {
+                if (t==null) for(Reduction r : eofReductions) irbc.invoke(r, b, c);
+                else         oreductions.invoke(t, irbc, b, c);
+            }
+
             // Constructor //////////////////////////////////////////////////////////////////////////////
 
             /**
@@ -252,27 +256,14 @@ public abstract class Parser<T extends Token, R> {
                 //         "yields" [in one or more step] is used instead of "produces" [in exactly one step]
                 //         to avoid having to iteratively construct our set of States as shown in most
                 //         expositions of the algorithm (ie "keep doing XYZ until things stop changing").
-                /*
-                for(Element e : all_elements) {
-                    if (e instanceof Atom) continue;
-                    HashSet<Position> h = new Walk.Closure(null, g.cache).closure(e, hs);
-                    State s = all_states.get(h) == null ? new State(h, all_states, all_elements) : all_states.get(h);
-                    if (gotoSetNonTerminals.get(e) != null)
-                        throw new Error("this should not happen");
-                    gotoSetNonTerminals.put(e, s);
-                }
-                */
                 HashMapBag<Element,Position> move = new HashMapBag<Element,Position>();
                 for(Position p : hs) {
                     Element e = p.element();
                     if (e==null) continue;
-                    HashSet<Element> ys = cache.ys.get(e);
-                    if (ys != null) {
-                        for(Element y : ys) {
-                            HashSet<Position> hp = new HashSet<Position>();
-                            reachable(p.next(), hp);
-                            move.addAll(y, hp);
-                        }
+                    for(Element y : cache.ys.getAll(e)) {
+                        HashSet<Position> hp = new HashSet<Position>();
+                        reachable(p.next(), hp);
+                        move.addAll(y, hp);
                     }
                 }
                 for(Element y : move) {
@@ -282,7 +273,13 @@ public abstract class Parser<T extends Token, R> {
                 }
             }
 
-            public String toString() { return "state["+idx+"]"; }
+            public String toString() {
+                //return "state["+idx+"]";
+                StringBuffer ret = new StringBuffer();
+                ret.append("state["+idx+"]: ");
+                for(Position p : this) ret.append("{"+p+"}  ");
+                return ret.toString();
+            }
 
             public int compareTo(Table.State s) { return idx==s.idx ? 0 : idx < s.idx ? -1 : 1; }
         }
@@ -311,24 +308,6 @@ public abstract class Parser<T extends Token, R> {
             }
             public String toString() { return "[reduce " + position + "]"; }
 
-            public Forest reduce(GSS.Phase.Node parent) {
-                if (numPop==0) return finish(parent, zero(), parent.phase());
-                return reduce(parent, numPop-1, null, null, parent.phase());
-            }
-
-            public Forest reduce(GSS.Phase.Node parent, GSS.Phase.Node onlychild) {
-                int pos = numPop-1;
-                if (pos>=0) holder[pos] = parent.pending();
-                Forest rex = null;
-                if (pos==0) {
-                    if (rex==null) {
-                        System.arraycopy(holder, 0, position.holder, 0, holder.length);
-                        rex = position.rewrite(parent.phase().getLocation());
-                    }
-                }
-                return reduce(onlychild, pos-1, rex, null, parent.phase());
-            }
-
             private Forest zero = null;
             public Forest zero() {
                 if (zero != null) return zero;
@@ -336,35 +315,44 @@ public abstract class Parser<T extends Token, R> {
                 return zero = position.rewrite(null);
             }
 
+            public void reduce(GSS.Phase.Node parent) {
+                if (numPop==0) finish(parent, zero(), parent.phase());
+                else           reduce(parent, numPop-1, parent.phase());
+            }
+
+            public void reduce(GSS.Phase.Node parent, GSS.Phase.Node onlychild) {
+                if (numPop<=0) throw new Error("called wrong form of reduce()");
+                int pos = numPop-1;
+                Forest old = holder[pos];
+                holder[pos] = parent.pending();
+                if (pos==0) {
+                    System.arraycopy(holder, 0, position.holder, 0, holder.length);
+                    finish(onlychild, position.rewrite(parent.phase().getLocation()), parent.phase());
+                } else {
+                    reduce(onlychild, pos-1, parent.phase());
+                }
+                holder[pos] = old;
+            }
+
             // FIXME: this could be more elegant and/or cleaner and/or somewhere else
-            private Forest reduce(GSS.Phase.Node parent, int pos, Forest rex, GSS.Phase.Node onlychild, GSS.Phase target) {
-                if (pos>=0) holder[pos] = parent.pending();
+            private void reduce(GSS.Phase.Node parent, int pos, GSS.Phase target) {
+                Forest old = holder[pos];
+                holder[pos] = parent.pending();
                 if (pos==0) {
-                    if (rex==null) {
-                        System.arraycopy(holder, 0, position.holder, 0, holder.length);
-                        rex = position.rewrite(target.getLocation());
-                    }
-                    if (onlychild != null)
-                        reduce(onlychild, pos-1, rex, null, target);
-                    else 
-                        for(GSS.Phase.Node child : parent.parents())
-                            reduce(child, pos-1, rex, null, target);
-                } else if (pos>0) {
-                    if (onlychild != null)
-                        reduce(onlychild, pos-1, rex, null, target);
-                    else 
-                        for(GSS.Phase.Node child : parent.parents())
-                            reduce(child, pos-1, rex, null, target);
+                    System.arraycopy(holder, 0, position.holder, 0, holder.length);
+                    for(int i=0; i<position.pos; i++) if (position.holder[i]==null) throw new Error("realbad");
+                    Forest rex = position.rewrite(target.getLocation());
+                    for(GSS.Phase.Node child : parent.parents()) finish(child, rex, target);
                 } else {
-                    return finish(parent, rex, target);
+                    for(GSS.Phase.Node child : parent.parents()) reduce(child, pos-1, target);
                 }
-                return rex;
+                holder[pos] = old;
             }
-            private Forest finish(GSS.Phase.Node parent, Forest result, GSS.Phase target) {
+            private void finish(GSS.Phase.Node parent, Forest result, GSS.Phase target) {
                 State state = parent.state.gotoSetNonTerminals.get(position.owner());
+                if (result==null) throw new Error();
                 if (state!=null)
-                    target.newNode(parent, result, state, numPop<=0, parent.phase());
-                return result;
+                    target.newNode(parent, result, state, numPop<=0, this);
             }
         }
     }