unrolling forests without recursion
[sbp.git] / src / edu / berkeley / sbp / Forest.java
index 6ac1e4b..937a229 100644 (file)
 package edu.berkeley.sbp;
 import edu.berkeley.sbp.*;
-import edu.berkeley.sbp.*;
+import edu.berkeley.sbp.Sequence.Position;
 import edu.berkeley.sbp.util.*;
 import java.io.*;
 import java.util.*;
 import java.lang.reflect.*;
 
 /** an efficient representation of a collection of trees (Tomita's shared packed parse forest) */
-public abstract class Forest<T> {
+public abstract class Forest<T> /*extends PrintableTree<Forest.MyBody<T>>*/
+    implements Visitable<Forest.Body<T>>,
+               IntegerMappable,
+               GraphViz.ToGraphViz {
+
+    private static int master_idx = 0;
+    private final int idx = master_idx++;
+    public int toInt() { return idx; }
+
+    public abstract void expand(TaskList tl, HashSet<Tree<T>> ht);
+    public abstract void gather(TaskList tl, HashSet<Tree<T>>[] ht, HashSet<Tree<T>> target);
+
+    public static class TaskList extends ArrayList<TaskList.Task> {
+        public interface Task {
+            public void perform();
+        }
+        public class ExpandTask implements Task {
+            private Forest f;
+            private HashSet hs;
+            public ExpandTask(Forest f, HashSet hs) { this.f = f; this.hs = hs; }
+            public void perform() { f.expand(TaskList.this, hs); }
+        }
+        public class GatherTask implements Task {
+            private Forest f;
+            private HashSet[] ht;
+            private HashSet hs;
+            public GatherTask(Forest f, HashSet<Tree<?>>[] ht, HashSet<Tree<?>> hs) { this.f = f; this.hs = hs; this.ht = ht;}
+            public void perform() { f.gather(TaskList.this, ht, hs); }
+        }
+        public void expand(Forest f, HashSet hs) {
+            add(new ExpandTask(f, hs));
+        }
+        public void gather(Forest f, HashSet[] ht, HashSet hs) {
+            add(new GatherTask(f, ht, hs));
+        }
+        public void run() {
+            while(true) {
+                if (isEmpty()) return;
+                Task task = get(size()-1);
+                remove(size()-1);
+                task.perform();
+            }
+        }
+    }
 
     /** assume that this forest contains exactly one tree and return it; otherwise throw an exception */
-    public final Tree<T> expand1() throws Parser.Ambiguous, Parser.Failed {
-        Iterator<Tree<T>> it = expand(true).iterator();
-        if (!it.hasNext()) throw new Parser.Failed();
-        return it.next();
+    public final Tree<T> expand1() throws Ambiguous, ParseFailed {
+        try {
+            Iterator<Tree<T>> it = expand(true).iterator();
+            if (!it.hasNext()) throw new ParseFailed();
+            return it.next();
+        } catch (InnerAmbiguous ia) { throw new Ambiguous(ia.f); }
     }
 
     /** expand this forest into a set of trees */
-    public abstract HashSet<Tree<T>>  expand(boolean toss);
+    public HashSet<Tree<T>> expand(boolean toss) {
+        /*
+        final HashSetTreeConsumer<T> ret = new HashSetTreeConsumer<T>();
+        visit(new TreeMaker2<T>(toss, ret), null, null);
+        */
+        TaskList tl = new TaskList();
+        HashSet<Tree<T>> ret = new HashSet<Tree<T>>();
+        tl.expand(this, ret);
+        tl.run();
 
-    abstract boolean valid();
+        if (toss && ret.size() > 1) throw new InnerAmbiguous(this);
+        return ret;
+    }
+
+    private static class InnerAmbiguous extends RuntimeException {
+        public final Forest<?> f;
+        public InnerAmbiguous(Forest<?> f) { this.f = f; }
+    }
 
-    static <T> Forest<T>   singleton(Token.Location loc, Sequence creator)                       { return create(loc, null, new Forest[] { }, creator, false, true); }
-    static <T> Forest<T>   singleton(Token.Location loc, Forest<T> body, Sequence creator)       { return create(loc, null, new Forest[] { body }, creator, false, true); }
-    static <T> Forest<T>   leaf(Token.Location loc, T tag,                     Sequence creator) { return create(loc, tag, null, creator, false, false); }
-    public static <T> Forest<T>   create(Token.Location loc, T tag, Forest<T>[] tokens, Sequence creator, boolean unwrap, boolean singleton) {
-        return new MultiForest<T>(loc, tag, tokens, creator, unwrap, singleton);
+    static interface TreeConsumer<T> {
+        public void addTree(Tree<T> t);
+    }
+    static class HashSetTreeConsumer<T> extends HashSet<Tree<T>> implements TreeConsumer<T> {
+        public void addTree(Tree<T> t) {
+            super.add(t);
+        }
     }
 
+    static        <T> Forest<T> singleton(Input.Location loc, Position p) {
+        return create(loc, null, new Forest[] { }, new Object[0], false, true, p); }
+    static        <T> Forest<T> singleton(Input.Location loc, Forest<T> body, Position p) {
+        //return create(loc, null, new Forest[] { body },  false, true, p);
+        return body;
+    }
+    static        <T> Forest<T> leaf(Input.Location loc, T tag, Position p) { return create(loc, tag, null, null, false, false, p); }
+    public static <T> Forest<T> create(Input.Location loc, T tag, Forest<T>[] tokens, Object[] labels, boolean unwrap, boolean singleton, Position p) {
+        return new MyBody<T>(loc, tag, tokens, labels, unwrap, singleton, p);
+    }
     // Body //////////////////////////////////////////////////////////////////////////////
 
-    protected static class Body<T> {
+    protected static interface Body<T> extends GraphViz.ToGraphViz {
+        void expand(int i, TreeMaker<T> h);
+    }
+    public abstract void edges(GraphViz.Node n);
+    public boolean ambiguous() { return false; }
+    protected static class MyBody<T> extends Forest<T> implements Body<T> /* extends PrintableTree<Forest<T>> implements */ {
+
+        public boolean isTransparent() { return false; }
+        public boolean isHidden() { return false; }
+        public GraphViz.Node toGraphViz(GraphViz gv) {
+            if (gv.hasNode(this)) return gv.createNode(this);
+            GraphViz.Node n = gv.createNode(this);
+            n.label = headToString()==null?"":headToString();
+            n.directed = true;
+            n.comment = reduction==null?null:reduction+"";
+            edges(n);
+            return n;
+        }
+        boolean edges = false;
+        public void edges(GraphViz.Node n) {
+            if (edges) return;
+            edges = true;
+            for(int i=0; i<tokens.length; i++) {
+                if (i==tokens.length-1 && unwrap && !tokens[i].ambiguous()) {
+                    tokens[i].edges(n);
+                } else {
+                    n.edge(tokens[i], labels==null?null:labels[i]);
+                }
+            }
+        }
+
+        public <B,C> void visit(Invokable<Forest.Body<T>,B,C> ivbc, B b, C c) {
+            ivbc.invoke(this, b, c);
+        }
 
-        private final Token.Location          location;
+        private final Input.Location    location;
         private final T                 tag;
         private final Forest<T>[]       tokens;
-        private final Sequence          creator;
+        private final Object[]          labels;
         private final boolean           unwrap;
         private final boolean           singleton;
+        private final Sequence.Position reduction;
 
-        private Body(Token.Location loc, T tag, Forest<T>[] tokens, Sequence creator, boolean unwrap, boolean singleton) {
+        private MyBody(Input.Location loc, T tag, Forest<T>[] tokens, Object[] labels, boolean unwrap, boolean singleton, Position reduction) {
             this.location = loc;
             this.tag = tag;
             this.tokens = tokens==null ? emptyForestArray : new Forest[tokens.length];
             if (tokens != null) System.arraycopy(tokens, 0, this.tokens, 0, tokens.length);
-            this.creator = creator;
+            if (tokens != null) for(int i=0; i<tokens.length; i++) if (tokens[i]==null) throw new Error(i+"");
             this.unwrap = unwrap;
             this.singleton = singleton;
+            this.reduction = reduction;
+            this.labels = labels;
+        }
+        public void gather(TaskList tl, HashSet<Tree<T>>[] ht, HashSet<Tree<T>> target) {
+            gather(tl, ht, target, new Tree[ht.length], 0);
+        }
+        private void gather(TaskList tl, HashSet<Tree<T>>[] ht, HashSet<Tree<T>> target, Tree[] trees, int i) {
+            if (i==ht.length) {
+                target.add(new Tree<T>(null, tag, trees));
+                return;
+            }
+            for(Tree<T> tree : ht[i]) {
+                if (unwrap && i==trees.length-1) {
+                    // I think this is wrong
+                    Tree[] trees2 = new Tree[trees.length - 1 + tree.numChildren()];
+                    System.arraycopy(trees, 0, trees2, 0, trees.length-1);
+                    for(int j=0; j<tree.numChildren(); j++)
+                        trees2[trees.length-1+j] = tree.child(j);
+                    target.add(new Tree<T>(null, tag, trees2));
+                } else {
+                    trees[i] = tree;
+                    gather(tl, ht, target, trees, i+1);
+                    trees[i] = null;
+                }
+            }
+        }
+        public void expand(TaskList tl, HashSet<Tree<T>> ht) {
+            if (singleton) {
+                tokens[0].expand(tl, ht);
+                return;
+            }
+            HashSet<Tree<T>>[] children = new HashSet[tokens.length];
+            tl.gather(this, children, ht);
+            for(int i=0; i<children.length; i++) {
+                children[i] = new HashSet<Tree<T>>();
+                tl.expand(tokens[i], children[i]);
+            }
         }
 
-        private HashSet<Tree<T>> expand(boolean toss, ArrayList<Tree<T>> toks, int i, HashSet<Tree<T>> h) {
+        public void expand(final int i, final TreeMaker<T> h) {
             if (singleton) {
-                for(Body<T> b : (IterableForest<T>)tokens[0]) b.expand(toss, toks, i, h);
+                tokens[0].visit(h, null, i);
+                return;
+            }
+            if (i==0) h.start(tag, location);
 
-            } else if (i==tokens.length) {
-                h.add(new Tree<T>(null, tag, toks.toArray(tree_hint)));
+            if (i==tokens.length) {
+                h.finish(tag, location);
 
             } else if (unwrap && i==tokens.length-1) {
                 if (tokens[i] != null)
-                    for(Body b : (IterableForest<T>)tokens[i])
-                        b.expand(toss, toks, 0, h);
+                    tokens[i].visit(h, null, 0);
 
             } else {
-                if (tokens[i]!=null) {
-                    HashSet<Tree<T>> exp = tokens[i].expand(toss);
-                    if (exp != null)
-                        for(Tree<T> r : exp) {
-                            int old = toks.size();
-                            toks.add(r);
-                            expand(toss, toks, i+1, h);
-                            while(toks.size() > old) toks.remove(toks.size()-1);
-                        }
-                }
-            }
-            return h;
-        }
-        void addTo(HashSet<Body> h) {
-            if (!singleton) h.add(this);
-            else for(Body b : (IterableForest<T>)tokens[0]) b.addTo(h);
-        }
-        void addTo(FastSet<Body> h) {
-            if (!singleton) h.add(this, true);
-            else for(Body b : (IterableForest<T>)tokens[0]) b.addTo(h);
-        }
-
-        private boolean kcache = false;
-        private boolean keep = false;
-        public boolean keep() {
-            if (kcache) return keep;
-            kcache = true;
-            for(Forest<T> token : tokens) if (!token.valid()) return keep = false;
-            return keep = creator==null || (creator.needs.size()==0 && creator.hates.size()==0);
-        }
-        public boolean keep(Iterable<Body<T>> h) {
-            if (keep()) return true;
-            for(Forest<T> token : tokens) if (!token.valid()) return false;
-            int needs = 0;
-            for(Body<T> b : h) {
-                if (creator.hates.contains(b.creator) && b.keep(h)) return false;
-                if (creator.needs.contains(b.creator) && b.keep(h)) needs--;
+                tokens[i].visit(new TreeMaker<T>(h.toss) {
+                    public void start(T head, Input.Location loc) { }
+                    public void addTree(Tree<T> t, Object label) { toks.add(t); labs.add(label); }
+                    public void finish(T head, Input.Location loc) {
+                        int old = h.toks.size();
+                        h.addTree(new Tree<T>(loc, head, toks.toArray(tree_hint), labs.toArray(string_hint)), labels==null?null:labels[i]);
+                        expand(i+1, h);
+                        while(h.toks.size() > old) h.toks.remove(h.toks.size()-1);
+                        while(h.labs.size() > old) h.labs.remove(h.labs.size()-1);
+                    }
+                }, null, null);
             }
-            return needs <= -1 * creator.needs.size();
         }
 
-
-        public String toString() {
-            StringBuffer ret = new StringBuffer();
-            for(int i=0; i<tokens.length; i++) {
-                String q = tokens[i]==null ? "null" : tokens[i].toString();
-                if (q.length() > 0) {
-                    ret.append(q);
-                    ret.append(" ");
-                }
-            }
-            String tail = ret.toString().trim();
-            String head = (tag!=null && !tag.toString().equals("")) ? (tail.length() > 0 ? tag+":" : tag+"") : "";
-            if (tail.length() > 0) tail = "{" + tail + "}";
-            return head + tail;
-        }
+        protected String  headToString()         { return tag==null?null:tag.toString(); }
+        protected String  headToJava()           { return null; }
+        protected String  left()                 { return "{"; }
+        protected String  right()                { return "}"; }
+        protected boolean ignoreSingleton()      { return false; }
     }
 
 
     // Ref //////////////////////////////////////////////////////////////////////////////
 
-    static abstract class IterableForest<T> extends Forest<T> implements Iterable<Forest.Body<T>> {
-        public abstract Iterator<Forest.Body<T>> iterator();
-    }
-
     /**
      *  This class represents a partially complete collection of
      *  forests to be viewed as a forest at some later date; once
      *  viewed, it becomes immutable
      */
-    static class Ref<T> extends IterableForest<T> {
-        private FastSet<Forest> hp = new FastSet<Forest>();
-        private Forest res = null;
-        public boolean valid = false;
+    static class Ref<T> extends Forest<T> {
+        public void expand(TaskList tl, HashSet<Tree<T>> ht) {
+            for (Forest<T> f : hp) f.expand(tl, ht);
+        }
+        public void gather(TaskList tl, HashSet<Tree<T>>[] ht, HashSet<Tree<T>> target) {
+            throw new Error();
+        }
+        public HashSet<GSS.Phase.Node> parents = new HashSet<GSS.Phase.Node>();
+        public boolean contains(Forest f) {
+            return hp.contains(f);
+        }
+        public boolean ambiguous() {
+            if (hp.size()==0) return false;
+            if (hp.size()==1) return hp.iterator().next().ambiguous();
+            return true;
+        }
+        private FastSet<Forest<T>> hp = new FastSet<Forest<T>>();
         public Ref() { }
-        public void merge(Forest p) {
-            if (res != null) throw new Error("already resolved!");
-            if (p==null) throw new Error();
-            if (p!=this) hp.add(p);
-        }
-        public Iterator<Body<T>> iterator() { return ((IterableForest<T>)resolve()).iterator(); }
-        public HashSet<Tree<T>> expand(boolean toss) { return resolve().expand(toss); }
-        public boolean valid() { if (valid) return true; resolve(); return valid; }
-        public String toString() { return resolve().toString(); }
-        public Forest resolve() {
-            if (hp==null) return res;
-            HashSet<Body> results = null;
-            FastSet<Body> nh      = new FastSet<Body>();
-            for(Forest<?> p : hp)
-                for(Body<?> b : (IterableForest<?>)p) {
-                    if (b.keep() && (b.creator==null || !b.creator.lame)) { valid = true; b.addTo(nh); }
-                    else results = new HashSet<Body>();
-                }
-            if (results != null) {
-                for(Forest<?> p : hp)
-                    for(Body<?> b : (IterableForest<?>)p)
-                        results.add(b);
-                for(Body b : results) {
-                    if (b.keep() && (b.creator==null || !b.creator.lame)) continue;
-                    if (b.creator!=null && b.creator.lame) continue;
-                    if (!b.keep(results)) continue;
-                    valid = true;
-                    b.addTo(nh);
-                }
-            }
-            hp = null;
-            return res = new MultiForest(nh, valid);
+        public int toInt() {
+            if (hp.size()==1) return hp.iterator().next().toInt();
+            return super.toInt();
+        }
+        public void merge(Forest p) { if (p!=this) hp.add(p, true); }
+
+        public boolean isTransparent() { return hp.size()==1; }
+        public boolean isHidden() { return hp.size()==0; }
+        public void edges(GraphViz.Node n) {
+            if (hp.size()==1) { hp.iterator().next().edges(n); return; }
+            for(Forest f : hp) f.edges(n);
         }
+        public GraphViz.Node toGraphViz(GraphViz gv) {
+            //if (hp.size()==0) return null;
+            if (hp.size()==1) return hp.iterator().next().toGraphViz(gv);
+            if (gv.hasNode(this)) return gv.createNode(this);
+            GraphViz.Node n = gv.createNode(this);
+            n.label = "?";
+            n.color = "red";
+            for(Forest f : hp) n.edge(f, null);
+            return n;
+        }
+
+        public <B,C> void visit(Invokable<Forest.Body<T>,B,C> ivbc, B b, C c) {
+            if (hp==null) return;
+            for(Forest<T> f : hp)
+                f.visit(ivbc, b, c);
+        }
+        public Forest resolve() { return this; }
     }
 
-    // Implementations //////////////////////////////////////////////////////////////////////////////
-
-    private static class MultiForest<T> extends IterableForest<T> {
-        private final FastSet<Body<T>> results;
-        private boolean valid;
-        public boolean valid() { return valid; }
-        private MultiForest(FastSet<Body<T>> results, boolean valid) { this.results = results; this.valid = valid; }
-        public MultiForest(Token.Location loc, T tag, Forest<T>[] tokens, Sequence creator, boolean unwrap, boolean singleton) {
-            this.results = new FastSet<Body<T>>(new Body(loc, tag, tokens, creator, unwrap, singleton));
-            this.valid = true;
-        }
-        public Iterator<Body<T>> iterator() { return results.iterator(); }
-
-        public HashSet<Tree<T>> expand(boolean toss) {
-            HashSet<Tree<T>> ret = new HashSet<Tree<T>>();
-            for(Body<T> b : results)
-                ret.addAll(b.expand(toss, new ArrayList<Tree<T>>(), 0, new HashSet<Tree<T>>()));
-            if (toss && ret.size() > 1) throw new Parser.Ambiguous(this);
-            return ret;
-        }
-        
-        // Display //////////////////////////////////////////////////////////////////////////////
-        
-        private String toString = null;
-        public String toString() {
-            if (toString != null) return toString;
-            StringBuffer ret = new StringBuffer();
-            if (results.size()==1) {
-                for(Forest.Body<T> r : results)
-                    ret.append(r);
-                return toString = ret.toString();
-            }
-            ret.append("<?");
-            boolean first = true;
-            for(Forest.Body<T> r : results) {
-                if (!first) ret.append(' ');
-                first = false;
-                ret.append(r);
+    public abstract <B,C> void visit(Invokable<Forest.Body<T>,B,C> ivbc, B b, C c);
+    private static class TreeMaker2<T> extends TreeMaker<T> {
+        private TreeConsumer<T> tc;
+        public TreeMaker2(boolean toss, TreeConsumer<T> tc) { super(toss); this.tc = tc; }
+        public void finish(T head, Input.Location loc) { tc.addTree(new Tree<T>(loc, head, toks.toArray(tree_hint), labs.toArray(string_hint)));; }
+        public void start(T head, Input.Location loc) { }
+        public void addTree(Tree<T> t, Object label) { toks.add(t); labs.add(label); }
+    }
+    private static abstract class TreeMaker<T> implements Invokable<Forest.Body<T>,Boolean,Integer>/*, TreeConsumer<T>*/ {
+        public ArrayList<Tree<T>> toks = new ArrayList<Tree<T>>();
+        public ArrayList<Object>  labs = new ArrayList<Object>();
+        private boolean toss;
+        protected T head;
+        public TreeMaker(boolean toss) { this.toss = toss; }
+        public abstract void start(T head, Input.Location loc);
+        public abstract void finish(T head, Input.Location loc);
+        public abstract void addTree(Tree<T> t, Object label);
+        public void invoke(Forest.Body<T> bod, Boolean o, Integer i) {
+            if (i==null) {
+                ArrayList<Tree<T>> toks = this.toks;
+                this.toks = new ArrayList<Tree<T>>();
+                ArrayList<Object> labs = this.labs;
+                this.labs = new ArrayList<Object>();
+                bod.expand(0, this);
+                this.toks = toks;
+                this.labs = labs;
+            } else {
+                bod.expand(i, this);
             }
-            ret.append("?>");
-            return toString = ret.toString();
         }
     }
 
     // Statics //////////////////////////////////////////////////////////////////////////////
 
     private static Tree[] tree_hint = new Tree[0];
-    private static Body[] body_hint = new Body[0];
+    private static String[] string_hint = new String[0];
     private static final Forest[] emptyForestArray = new Forest[0];
+
+    protected String  headToString()    { return null; }
+    protected String  headToJava()      { return null; }
+    protected String  left()            { return "<?"; }
+    protected String  right()           { return "?>"; }
+    protected boolean ignoreSingleton() { return true; }
 }