+// Copyright 2006-2007 all rights reserved; see LICENSE file for BSD-style license
+
package 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> /*extends PrintableTree<Forest.MyBody<T>>*/
- implements Visitable<Forest.Body<T>>,
- IntegerMappable,
- GraphViz.ToGraphViz {
+/**
+ * <font color=blue>
+ * An efficient representation of a collection of trees (Tomita's
+ * shared packed parse forest).
+ * </font>
+ */
+public abstract class Forest<NodeType> implements GraphViz.ToGraphViz {
+
+ /** assume that this forest contains exactly one tree and return it; otherwise throw an exception */
+ public abstract Tree<NodeType> expand1() throws Ambiguous;
- private static int master_idx = 0;
- private final int idx = master_idx++;
- public int toInt() { return idx; }
+ /** expand this forest into a set of trees */
+ public Iterable<Tree<NodeType>> expand() {
+ HashSet<Tree<NodeType>> ht = new HashSet<Tree<NodeType>>();
+ expand(ht, new HashSet<Forest<NodeType>>(), null);
+ return ht;
+ }
- public abstract void expand(TaskList tl, HashSet<Tree<T>> ht);
- public abstract void gather(TaskList tl, HashSet<Tree<T>>[] ht, HashSet<Tree<T>> target);
+ /** returns the input Region which this Forest was parsed from */
+ public abstract Input.Region getRegion();
- 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); }
+ // Package-Private //////////////////////////////////////////////////////////////////////////////
+
+ public static <NodeType> Forest<NodeType> create(Input.Region region, NodeType head, Forest<NodeType>[] children) {
+ return create(region, head, children, new boolean[children==null ? 0 : children.length]); }
+ public static <NodeType> Forest<NodeType> create(Input.Region region, NodeType head, Forest<NodeType>[] children, boolean[] lifts) {
+ if (region == null) throw new RuntimeException("invoked Forest.create(region=null) -- this should never happen");
+ return new One<NodeType>(region, head, children, lifts);
+ }
+
+ abstract void expand(HashSet<Tree<NodeType>> ht, HashSet<Forest<NodeType>> ignore, Tree<NodeType> bogus);
+ abstract void gather(HashSet<Forest<NodeType>> ignore);
+ abstract void edges(GraphViz.StateNode n);
+ boolean ambiguous() { return false; }
+
+ // One //////////////////////////////////////////////////////////////////////////////
+
+ /** A "single" forest with a head and child subforests */
+ private static class One<NodeType> extends Forest<NodeType> {
+
+ private final Input.Region location;
+ private final NodeType head;
+ private final Forest<NodeType>[] children;
+
+ /** if true, the last child's children are considered children of this node */
+ private final boolean[] lifts;
+
+ public Input.Region getRegion() { return location; }
+
+ private One(Input.Region loc, NodeType head, Forest<NodeType>[] children, boolean[] lifts) {
+ this.location = loc;
+ this.head = head;
+ if (head==null) throw new RuntimeException("invoked Forest.create(,null,,,) -- this should never happen");
+ this.children = children==null ? emptyForestArray : new Forest[children.length];
+ if (children != null) System.arraycopy(children, 0, this.children, 0, children.length);
+ if (children != null) for(int i=0; i<children.length; i++) if (children[i]==null) throw new Error(i+"");
+ this.lifts = lifts;
}
- 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 Tree<NodeType> expand1() throws Ambiguous {
+ if (children.length == 0)
+ return new Tree<NodeType>(location, head);
+ else
+ return expand1ForOneAndMany((Forest<NodeType>)this);
}
- public void expand(Forest f, HashSet hs) {
- add(new ExpandTask(f, hs));
+
+ void gather(HashSet<Forest<NodeType>> hf) {
+ hf.add(this);
+ for(Forest<NodeType> f : children) f.gather(hf);
}
- public void gather(Forest f, HashSet[] ht, HashSet hs) {
- add(new GatherTask(f, ht, hs));
+ void expand(HashSet<Tree<NodeType>> ht, HashSet<Forest<NodeType>> ignore, Tree<NodeType> bogus) {
+ if (ignore.contains(this)) { ht.add(bogus); return; }
+ expand(0, new Tree[children.length], ht, ignore, bogus);
}
- public void run() {
- while(true) {
- if (isEmpty()) return;
- Task task = get(size()-1);
- remove(size()-1);
- task.perform();
+ private void expand(final int i, Tree<NodeType>[] ta, HashSet<Tree<NodeType>> ht, HashSet<Forest<NodeType>> ignore,
+ Tree<NodeType> bogus) {
+ if (i==children.length) {
+ ht.add(new Tree<NodeType>(location, head, ta, lifts));
+ } else {
+ HashSet<Tree<NodeType>> ht2 = new HashSet<Tree<NodeType>>();
+ children[i].expand(ht2, ignore, bogus);
+ for(Tree<NodeType> tc : ht2) {
+ ta[i] = tc;
+ expand(i+1, ta, ht, ignore, bogus);
+ ta[i] = null;
+ }
}
}
- }
-
- /** assume that this forest contains exactly one tree and return it; otherwise throw an exception */
- 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 HashSet<Tree<T>> expand(boolean toss) {
- TaskList tl = new TaskList();
- HashSet<Tree<T>> ret = new HashSet<Tree<T>>();
- tl.expand(this, ret);
- tl.run();
-
- 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(Input.Region loc, Position p) {
- return create(loc, null, new Forest[] { }, false, true, p); }
- static <T> Forest<T> singleton(Input.Region loc, Forest<T> body, Position p) { return body; }
- static <T> Forest<T> leaf(Input.Region loc, T tag, Position p) { return create(loc, tag, null, false, false, p); }
- public static <T> Forest<T> create(Input.Region loc, T tag, Forest<T>[] tokens, boolean unwrap, boolean singleton, Position p) {
- return new MyBody<T>(loc, tag, tokens, unwrap, singleton, p);
- }
- // Body //////////////////////////////////////////////////////////////////////////////
-
- public /*protected*/ static interface Body<T> extends GraphViz.ToGraphViz {
- void expand(int i, TreeBuilder<T> h);
- }
- public abstract void edges(GraphViz.Node n);
- public boolean ambiguous() { return false; }
- public /*protected*/ static class MyBody<T> extends Forest<T> implements Body<T> /* extends PrintableTree<Forest<T>> implements */ {
+ // GraphViz, ToInt //////////////////////////////////////////////////////////////////////////////
public boolean isTransparent() { return false; }
public boolean isHidden() { return false; }
- public GraphViz.Node toGraphViz(GraphViz gv) {
+ public GraphViz.StateNode toGraphViz(GraphViz gv) {
if (gv.hasNode(this)) return gv.createNode(this);
- GraphViz.Node n = gv.createNode(this);
+ GraphViz.StateNode 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) {
+ boolean edges = false; // FIXME ??
+ public void edges(GraphViz.StateNode 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);
+ for(int i=0; i<children.length; i++) {
+ if (lifts[i] && !children[i].ambiguous()) {
+ children[i].edges(n);
} else {
- n.edge(tokens[i], null);
+ n.edge(children[i], null);
}
}
}
- public <B,C> void visit(Invokable<Forest.Body<T>,B,C> ivbc, B b, C c) {
- ivbc.invoke(this, b, c);
- }
+ protected String headToString() { return head==null?null:head.toString(); }
+ protected String headToJava() { return "null"; }
+ protected String left() { return "{"; }
+ protected String right() { return "}"; }
+ protected boolean ignoreSingleton() { return false; }
+ }
- private final Input.Region location;
- private final T tag;
- private final Forest<T>[] tokens;
- private final boolean unwrap;
- private final boolean singleton;
- private final Sequence.Position reduction;
- private MyBody(Input.Region loc, T tag, Forest<T>[] tokens, 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);
- 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;
- }
- 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>(location, 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>(location, 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]);
- }
- }
+ // Many //////////////////////////////////////////////////////////////////////////////
- public void expand(final int i, final TreeBuilder<T> h) {
- if (singleton) {
- tokens[0].visit(h, null, i);
- return;
- }
- if (i==0) h.start(tag, location);
+ /** An "ambiguity node"; this is immutable once it has been "looked at" */
+ static class Many<NodeType> extends Forest<NodeType> {
- if (i==tokens.length) {
- h.finish(tag, location);
+ private FastSet<Forest<NodeType>> hp = new FastSet<Forest<NodeType>>();
+ private boolean touched = false;
- } else if (unwrap && i==tokens.length-1) {
- if (tokens[i] != null)
- tokens[i].visit(h, null, 0);
+ public Many() { }
- } else {
- tokens[i].visit(new TreeBuilder<T>(h.toss) {
- public void start(T head, Input.Region loc) { }
- public void addTree(Tree<T> t) { toks.add(t); }
- public void finish(T head, Input.Region loc) {
- int old = h.toks.size();
- h.addTree(new Tree<T>(loc, head, toks.toArray(tree_hint)));
- expand(i+1, h);
- while(h.toks.size() > old) h.toks.remove(h.toks.size()-1);
- }
- }, null, null);
+ public Input.Region getRegion() { return hp.iterator().next().getRegion(); } // all should be identical
+
+ Forest<NodeType> simplify() throws Ambiguous {
+ touched();
+ if (hp.size() > 1) {
+ HashSet<Forest<NodeType>> hf0 = new HashSet<Forest<NodeType>>();
+ Iterator<Forest<NodeType>> ih = hp.iterator();
+ ih.next().gather(hf0);
+ for(Forest<NodeType> f : hp) {
+ HashSet<Forest<NodeType>> hf1 = new HashSet<Forest<NodeType>>();
+ f.gather(hf1);
+ hf0.retainAll(hf1);
+ }
+ HashSet<Tree<NodeType>> ht = new HashSet<Tree<NodeType>>();
+ expand(ht, hf0, new Tree(null, "*"));
+ throw new Ambiguous((Forest<?>)this,
+ (HashSet<Tree<?>>)(Object)ht);
}
- }
- 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; }
- }
+ return hp.iterator().next();
+ }
+ public Tree<NodeType> expand1() throws Ambiguous {
+ return simplify().expand1();
+ }
+
+ void gather(HashSet<Forest<NodeType>> ht) {
+ touched();
- // Ref //////////////////////////////////////////////////////////////////////////////
+ // FIXME: do something more sensible here
+ if (ht.contains(this)) {
+ System.err.println("WARNING: grammar produced a circular forest\n" + this);
+ //throw new Error("grammar produced a circular forest:\n" + this);
+ return;
+ }
- /**
- * 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 Forest<T> {
- public void expand(TaskList tl, HashSet<Tree<T>> ht) {
- for (Forest<T> f : hp) f.expand(tl, ht);
+ ht.add(this);
+ for(Forest<NodeType> f : hp) f.gather(ht);
}
- public void gather(TaskList tl, HashSet<Tree<T>>[] ht, HashSet<Tree<T>> target) {
- throw new Error();
+
+ private void touched() {
+ if (touched) return;
+ touched = true;
+ /*
+ FastSet<Forest<NodeType>> f2 = new FastSet<Forest<NodeType>>();
+ for(Forest f : hp)
+ if (f instanceof Forest.One) f2.add(f);
+ else for(Forest ff : ((Forest.Many<NodeType>)f))
+ f2.add(ff);
+ hp = f2;
+ */
}
- public HashSet<GSS.Phase.Node> parents = new HashSet<GSS.Phase.Node>();
public boolean contains(Forest f) {
+ touched();
return hp.contains(f);
}
- public boolean ambiguous() {
+ public void merge(Forest p) {
+ if (touched) throw new RuntimeException("attempt to merge() on a Forest.Many that has already been examined");
+ if (p==this) throw new RuntimeException("attempt to merge() a Forest.Many to itself!");
+ hp.add(p, true);
+ }
+ boolean ambiguous() {
+ touched();
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 int toInt() {
- if (hp.size()==1) return hp.iterator().next().toInt();
- return super.toInt();
+
+ void expand(HashSet<Tree<NodeType>> ht, HashSet<Forest<NodeType>> ignore, Tree<NodeType> bogus) {
+ touched();
+ if (ignore.contains(this)) { ht.add(bogus); return; }
+ for (Forest<NodeType> f : hp) f.expand(ht, ignore, bogus);
}
- public void merge(Forest p) { if (p!=this) hp.add(p, true); }
+
+
+ // GraphViz, ToInt //////////////////////////////////////////////////////////////////////////////
public boolean isTransparent() { return hp.size()==1; }
public boolean isHidden() { return hp.size()==0; }
- public void edges(GraphViz.Node n) {
+ public void edges(GraphViz.StateNode 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;
+ public GraphViz.StateNode toGraphViz(GraphViz gv) {
if (hp.size()==1) return hp.iterator().next().toGraphViz(gv);
if (gv.hasNode(this)) return gv.createNode(this);
- GraphViz.Node n = gv.createNode(this);
+ GraphViz.StateNode n = gv.createNode(this);
n.label = "?";
n.color = "red";
for(Forest f : hp) n.edge(f, null);
return n;
}
+ }
+
+
+ // Statics //////////////////////////////////////////////////////////////////////////////
+
+ /** Depth-first expansion, handling .One and .Many without recursion. */
+ private static <NodeType> Tree<NodeType> expand1ForOneAndMany(
+ Forest<NodeType> forest) throws Ambiguous {
+ List<Tree<NodeType>> nodes = new ArrayList<Tree<NodeType>>();
+ List<Tree<NodeType>> build = new ArrayList<Tree<NodeType>>();
+
+ // path stack
+ List<Forest.One<NodeType>> path = new ArrayList<Forest.One<NodeType>>();
+ List<Integer> sizes = new ArrayList<Integer>(),
+ poss = new ArrayList<Integer>();
+
+ // handle the left-most leaf
+ expand1Descend(path, nodes, sizes, poss, forest);
+
+ for (Forest.One<NodeType> f; path.size() > 0;) {
+ // up one
+ f = pop(path);
+ int size = pop(sizes) + 1;
+ int pos = pop(poss) + 1;
+ Forest<NodeType>[] children = f.children;
+ if (pos < children.length) {
+ // down the next branch
+ path.add(f);
+ sizes.add(new Integer(size));
+ poss.add(new Integer(pos));
- 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);
+ // handle the left-most leaf
+ expand1Descend(path, nodes, sizes, poss, children[pos]);
+ } else {
+ if (path.size() > 0 && peek(path).lifts[peek(poss)]) {
+ // skip assembling this node, lift children
+ sizes.add(pop(sizes) + size - 1);
+ continue;
+ }
+ // assemble this node
+ for (int i=size; i > 0; i--)
+ build.add(nodes.remove(nodes.size() - i));
+ nodes.add(new Tree<NodeType>(f.location, f.head, build));
+ build.clear();
+ }
}
- public Forest resolve() { return this; }
+
+ return pop(nodes);
}
- public abstract <B,C> void visit(Invokable<Forest.Body<T>,B,C> ivbc, B b, C c);
+ /** Descend to the left-most leaf, building the path. */
+ private static <NodeType> void expand1Descend(
+ List<Forest.One<NodeType>> path, List<Tree<NodeType>> nodes,
+ List<Integer> sizes, List<Integer> poss, Forest<NodeType> f)
+ throws Ambiguous {
+ while (true) {
+ if (f instanceof Forest.Many)
+ f = ((Forest.Many<NodeType>)f).simplify();
+ else if (f instanceof Forest.One) {
+ Forest.One<NodeType> one = (Forest.One<NodeType>)f;
+ if (one.children.length == 0)
+ break;
+ path.add(one);
+ sizes.add(0);
+ poss.add(0);
+ f = one.children[0];
+ } else {
+ nodes.add(f.expand1());
+ return;
+ }
+ }
- // Statics //////////////////////////////////////////////////////////////////////////////
+ if (path.size() > 0 && peek(path).lifts[peek(poss)])
+ sizes.add(pop(sizes) - 1); // ignore lifted leafs
+ else
+ nodes.add(f.expand1());
+ }
+
+ private static <T> T pop(List<T> list) {
+ return list.remove(list.size() - 1); }
+ private static <T> T peek(List<T> list) {
+ return list.get(list.size() - 1); }
private static Tree[] tree_hint = new Tree[0];
private static String[] string_hint = new String[0];