better reporting of indentation related parse errors
[sbp.git] / src / edu / berkeley / sbp / GSS.java
index a4f3d6a..5c1b92f 100644 (file)
+// Copyright 2006 all rights reserved; see LICENSE file for BSD-style license
+
 package edu.berkeley.sbp;
 import edu.berkeley.sbp.*;
-import edu.berkeley.sbp.*;
-import edu.berkeley.sbp.*;
 import edu.berkeley.sbp.util.*;
+import edu.berkeley.sbp.Parser.Table.*;
+import edu.berkeley.sbp.Sequence.Position;
 import java.io.*;
 import java.util.*;
 import java.lang.reflect.*;
 
-//////////////////////////////////////////////////////////////////////////////
-// TODO:
-//
-//  - fix public/package/private status
-//
-
-//////////////////////////////////////////////////////////////////////////////
-// Optimizations to add
-//
-// ** NOTE: not all of these are appropriate for this class -- it is
-//          simply a list of optimizations not implemented.  This
-//          class is meant to remain simple and easy to understand;
-//          optimizations which obscure that do not belong here (they
-//          should go into the compiled version instead)
-
 /** implements Tomita's Graph Structured Stack */
 class GSS {
 
-    public GSS() { }
+    Input input;
+    private Parser parser;
+    public GSS(Input input, Parser parser) { this.input = input; this.parser = parser;}
+    public Input getInput() { return input; }
 
-    private Phase.Node[] reducing_list = null;
+    int numNewNodes = 0;
+    int numOldNodes = 0;
+    int viewPos = 0;
+    int numReductions = 0;
 
     /** corresponds to a positions <i>between tokens</i> the input stream; same as Tomita's U_i's */
-    public class Phase {
+    class Phase<Tok> implements Invokable<State, Result>, IntegerMappable, GraphViz.ToGraphViz, Iterable<Node> {
 
-        /** the token immediately after this phase */
-        public  final Token token;
-
-        boolean reducing = false;
+        // FIXME: right now, these are the performance bottleneck
+        private HashMapBag<Integer,Sequence>       performed       = new HashMapBag<Integer,Sequence>();
 
-        /** currently this is necessary only for the code() hack -- it doesn't actually correspond to the input */
-        private final int pos;
+        public Forest.Many finalResult;
+        private PriorityQueue<Reduction> reductionQueue = new PriorityQueue<Reduction>();
 
-        /** FIXME */
-        public  Forest.Ref finalResult = null;
-
-        /** all nodes, keyed by the value returned by code() */
-        private HashMap<Long,Phase.Node> hash    = new HashMap<Long,Phase.Node>();  /* ALLOC */
+        Parser parser() { return parser; }
+        public void addReduction(Reduction r) {
+            //System.out.println("+ " + r);
+            parser.spin();
+            reductionQueue.add(r);
+        }
+        public void invoke(State st, Result result) {
+            parser.spin();
+            good |= next.newNode(result, st, false);
+        }
 
-        /** the number of nodes in this phase */
-        private int numNodes = 0;
+        /** the token immediately after this phase */
+        final Tok token;
+        final int pos;
+
+        public IntPairMap<Node> hash = new IntPairMap<Node>();  /* ALLOC */
+        private boolean good = false;
+        private Phase next = null;
+        private Phase prev;
+        private Input.Location location;
+        private Input.Location nextLocation;
+        private Input.Location prevLocation;
+        
+        private Forest forest;
 
-        boolean closed = false;
+        public Phase(State startState) throws ParseFailed, IOException {
+            this(null, null);
+            Result primordealResult = new Result(null, null, null);
+            newNode(primordealResult, startState, true);
+        }
+        public Phase(Phase prev, Forest forest) throws ParseFailed, IOException {
+            this.prevLocation = input.getLocation();
+            this.token = (Tok)input.next();
+            this.location = input.getLocation();
+            this.prev = prev;
+            this.forest = forest;
+            this.pos = prev==null ? 0 : prev.pos+1;
+            this.nextLocation = input.getLocation();
+            if (prev != null) prev.shift(this, forest);
+            numReductions = 0;
+
+            int minPhasePos = Integer.MAX_VALUE;
+            int maxOrd = -1;
+            Reduction best = null;
+            //System.out.println("==============================================================================");
+            while(!reductionQueue.isEmpty()) {
+                Reduction r = reductionQueue.poll();
+                //System.out.println("- " + r);
+                if (r.parentPhase() != null)
+                    if (r.parentPhase().pos > minPhasePos)
+                        throw new Error();
+                r.perform();
+                if (r.parentPhase() != null) {
+                    if (r.parentPhase().pos < minPhasePos) {
+                        minPhasePos = r.parentPhase().pos;
+                        maxOrd = r.reduction().ord;
+                        best = r;
+                    } else if (r.parentPhase().pos == minPhasePos) {
+                        /*
+                        if (best != null && Parser.mastercache.comparePositions(r.reduction(), best.reduction()) < 0)
+                            throw new Error("\n"+r+"\n"+best+"\n"+
+                                            Parser.mastercache.comparePositions(r.reduction(), best.reduction())+"\n"+r.compareTo(best)+
+                                            "\n"+(r.reduction().ord-best.reduction().ord));
+                        */
+                        maxOrd = r.reduction().ord;
+                        best = r;
+                    }
+                }
+                numReductions++;
+            }
+            if (token==null) shift(null, null);
+        }
 
-        private Token.Location location;
-        public Phase(Phase previous, Token token, Token.Location location) {
-            this.pos = previous==null ? 0 : previous.pos+1;
-            this.token = token;
-            this.location = location;
+        public boolean isDone() throws ParseFailed {
+            if (token != null) return false;
+            if (token==null && finalResult==null)
+                ParseFailed.error("unexpected end of file", this, null,
+                                  getLocation().createRegion(getLocation()));
+            return true;
         }
 
-        public boolean isDone() { return token == null; }
+        public Input.Location getPrevLocation() { return prevLocation; }
+        public Input.Location getLocation() { return location; }
+        public Input.Region getRegion() { return prevLocation.createRegion(location); }
+        public Input.Location getNextLocation() { return nextLocation; }
+        public boolean        isFrontier() { return hash!=null; }
 
-        private String error = "generic syntax error";
-        public void checkFailure() throws Parser.Failed {
-            if (numNodes <= 0)
-                throw new Parser.Failed(error, getLocation());
+        /** perform all shift operations, adding promoted nodes to <tt>next</tt> */
+        private void shift(Phase next, Forest result) throws ParseFailed {
+            this.next = next;
+            // this massively improves GC performance
+            if (prev != null) {
+                IntPairMap<Node> h = prev.hash;
+                prev.hash = null;
+                prev.performed = null;
+                for(Node n : h)
+                    n.check();
+            }
+            numOldNodes = hash.size();
+            for(Node n : hash.values()) {
+                if (token == null && n.state().isAccepting()) {
+                    if (finalResult==null) finalResult = new Forest.Many();
+                    for(Result r : n)
+                        finalResult.merge(r.getForest());
+                }
+                if (token == null) continue;
+                n.state().invokeShifts(token, this, new Result(result, n, null));
+            }
+            numNewNodes = next==null ? 0 : next.hash.size();
+            viewPos = this.pos;
+            if (!good && token!=null) {
+                String toks = token+"";
+                if (toks.length()==1 && toks.charAt(0) == edu.berkeley.sbp.chr.CharAtom.left) {
+                    ParseFailed.error("unexpected increase in indentation", this,
+                                      token, getRegionFromThisToNext());
+                } else if (toks.length()==1 && toks.charAt(0) == edu.berkeley.sbp.chr.CharAtom.right) {
+                    ParseFailed.error("unexpected decrease in indentation", this,
+                                      token, getRegionFromThisToNext());
+                } else {
+                    ParseFailed.error("unexpected character '"+ANSI.cyan(StringUtil.escapify(token+"",
+                                                                                             "\\\'\r\n"))+"'",
+                                      this, token, getRegionFromThisToNext());
+                }
+            }
+            if (token==null && finalResult==null)
+                ParseFailed.error("unexpected end of file", this, null,
+                                  getLocation().createRegion(getLocation()));
+            for(Node n : hash) n.check();
         }
 
-        public Token.Location getLocation() { return location; }
+        void newNodeFromReduction(Result result, State state, Position reduction) {
+            int pos = result.phase().pos;
+            Sequence owner = reduction.owner();
+            for(Sequence s : owner.hates())
+                if (performed.contains(pos, s))
+                    return;
+            for(Sequence s : owner.needs())
+                if (!performed.contains(pos, s))
+                    return;
+            if (owner.needed_or_hated && !performed.contains(pos, owner))
+                performed.add(pos, owner);
+            if (state!=null)
+                newNode(result, state, reduction.pos<=0);
+        }
 
         /** add a new node (merging with existing nodes if possible)
          *  @param parent             the parent of the new node
@@ -75,173 +183,52 @@ class GSS {
          *  @param fromEmptyReduction true iff this node is being created as a result of a reduction of length zero (see GRMLR paper)
          *  @param start              the earliest part of the input contributing to this node (used to make merging decisions)
          */
-        public void newNode(Node parent, Forest pending, Parser.Table.State state, boolean fromEmptyReduction, Phase start) {
-            Node p = hash.get(code(state, start));
-            if (p != null)  newNode2(p, parent, pending, state, fromEmptyReduction, start);
-            else            newNode3(parent, pending, state, fromEmptyReduction, start);
-        }
-        private void newNode2(Node p, Node parent, Forest pending, Parser.Table.State state, boolean fromEmptyReduction, Phase start) {
-            p.holder.merge(pending);
-            if (p.parents().contains(parent)) return;
-            p.addParent(parent, fromEmptyReduction);
-        }
-        private void newNode3(Node parent, Forest pending, Parser.Table.State state, boolean fromEmptyReduction, Phase start) {
+        private boolean newNode(Result result, State state, boolean fromEmptyReduction) {
+            Node p = hash.get(state, result.phase());
+            if (p != null) { p.addResult(result); return true; }
             do {
                 if (token != null && state.canShift(token)) break;
                 if (state.isAccepting()) break;
                 if (token==null) break;
-                int count = 0;
-                Parser.Table.Reduction r = null;
-                for(Parser.Table.Reduction red : token==null ? state.getEofReductions() : state.getReductions(token)) { r = red; count++; }
-                if (count==0) return;     // BEWARE! this optimization is suspected to cause really nasty heisenbugs
-                //if (count > 1) break;
-                //if (r.numPop == 0) break;
-                //r.reduce(pending, parent, null, Phase.this, null);
-                //return;
+                if (!state.canReduce(token)) return false;
             } while(false);
-
-            Node n = new Node(parent, pending, state, start);  // ALLOC
-            n.queueEmptyReductions();
-            if (!fromEmptyReduction) n.queueReductions(parent);
+            Node n = new Node(Phase.this, result, state, fromEmptyReduction);  // ALLOC
+            for(Object s : state.conjunctStates)
+                newNode(new Result(null, n, null), (State)s, fromEmptyReduction);
+            return true;
         }
 
-        
-        /** perform all reduction operations */
-        public void reduce() {
-            reducing = true;
-            if (reducing_list==null || reducing_list.length < hash.size())
-                reducing_list = new Phase.Node[hash.size() * 4];
-            Collection<Node> hv = hash.values();
-            hv.toArray(reducing_list);
-            int num = hv.size();
-            for(int i=0; i<num; i++) {
-                Node n = reducing_list[i];
-                n.queueEmptyReductions();
-                // INVARIANT: we never "see" a node until its parent-set is complete, modulo merges
-            }
-            for(int i=0; i<num; i++) {
-                Node n = reducing_list[i];
-                reducing_list[i] = null;
-                n.queueReductions();
-            }
+        public int toInt() { return pos+1; }
+        public int size() { return hash==null ? 0 : hash.size(); }
+        public int pos() { return pos; }
+        public Tok getToken() { return token; }
+        public Iterator<Node> iterator() { return hash.iterator(); }
+        public GSS getGSS() { return GSS.this; }
+
+        // GraphViz //////////////////////////////////////////////////////////////////////////////
+
+        public GraphViz.Node toGraphViz(GraphViz gv) {
+            if (gv.hasNode(this)) return gv.createNode(this);
+            GraphViz.Group g = gv.createGroup(this);
+            g.label = "Phase " + pos;
+            g.color = "gray";
+            g.cluster = true;
+            return g;
         }
-
-        /** perform all shift operations, adding promoted nodes to <tt>next</tt> */
-        public void shift(Phase next, Forest result) {
-            closed = true;
-            Forest res = null;
-            boolean ok = false;
-            for(Phase.Node n : hash.values()) {
-                if (n.holder==null) continue;
-                n.holder.resolve();
-                if (token == null && n.state.isAccepting()) {
-                    ok = true;
-                    if (finalResult==null) finalResult = new Forest.Ref();
-                    finalResult.merge(n.holder);
-                }
-                if (!n.holder.valid()) continue;
-                if (token == null) continue;
-                for(Parser.Table.State st : n.state.getShifts(token)) {
-                    if (res == null) res = result;
-                    next.newNode(n, res, st, true, this);
-                    ok = true;
-                }
-            }
-
-            if (!ok && token != null) {
-                StringBuffer error = new StringBuffer();
-                error.append("error: unable to shift token \"" + token + "\"\n");
-                //error.append("  before: " +pendingReductions+ "\n");
-                //error.append("  before: " +totalReductions+ "\n");
-                //for(Phase.Node n : hash.values()) {
-                //n.queueReductions();
-                //n.queueEmptyReductions();
-                //}
-                //error.append("  after: " +pendingReductions+ "\n");
-                //error.append("  candidate states:\n");
-                //for(Phase.Node n : hash.values()) {
-                    //for(Sequence.Position p : n.state) error.append("        " + p + "\n");
-                    //error.append("        --\n");
-                //for(Parser.Table.Reduction r : n.state.getReductions(token)) error.append("        " + r + "\n");
-                    //error.append("        ==\n");
-                //}
-                next.error = error.toString();
-            }
-
-            // this massively improves GC performance
-            hash = null;
-        }
-
-       
-        // GSS Nodes //////////////////////////////////////////////////////////////////////////////
-
-        /** a node in the GSS */
-        public final class Node extends FastSet<Node> {
-
-            public void addParent(Node parent, boolean fromEmptyReduction) {
-                parents().add(parent, true);
-                if (this!=parent && !fromEmptyReduction) queueReductions(parent);
-            }
-
-            private Forest.Ref holder = null;
-            private boolean allqueued = false;
-
-            /** what state this node is in */
-            public final Parser.Table.State state;
-
-            /** which Phase this Node belongs to (node that Node is also a non-static inner class of Phase) */
-            public Phase phase() { return Phase.this; }
-
-            public  Forest.Ref holder() { return holder==null ? (holder = new Forest.Ref()) : holder; }
-            public  Forest pending() { return Phase.this.closed ? holder().resolve() : holder; }
-            public  FastSet<Node> parents() { return this; }
-
-            public void queueReductions() {
-                if (allqueued) return;
-                allqueued = true;
-                int where = parents().size();
-                for(Parser.Table.Reduction r : state.getReductions(token))
-                    if (r.numPop >= 1)
-                        r.reduce(this, null, null);
-            }
-
-            public void queueReductions(Node n2) {
-                if (!allqueued) { queueReductions(); return; }
-                for(Parser.Table.Reduction r : state.getReductions(token))
-                    if (r.numPop > 0)
-                        r.reduce(this, n2);
-            }
-
-
-            public void queueEmptyReductions() {
-                if (reducing)
-                    for(Parser.Table.Reduction r : token==null ? state.getEofReductions() : state.getReductions(token))
-                        if (r.numPop==0)
-                            r.reduce(this, null, r.zero());
-            }
-
-            private Node(Node parent, Forest pending, Parser.Table.State state, Phase start) {
-                this.state = state;
-                if (pending != null) this.holder().merge(pending);
-                if (parent != null) parents().add(parent, true);
-                if (Phase.this.hash.get(code(state, start)) != null) throw new Error("severe problem!");
-                Phase.this.hash.put(code(state, start), this);
-                Phase.this.numNodes++;
-                if (parent==null) holder().valid = true; // hack to make sure that the "base" node is always considered valid
-            }
+        public boolean isTransparent() { return false; }
+        public boolean isHidden() { return false; }
+
+        public void dumpGraphViz(String filename) throws IOException {
+            FileOutputStream fos = new FileOutputStream(filename);
+            PrintWriter p = new PrintWriter(new OutputStreamWriter(fos));
+            GraphViz gv = new GraphViz();
+            for(Object n : this)
+                ((Node)n).toGraphViz(gv);
+            gv.dump(p);
+            p.flush();
+            p.close();
         }
 
     }
 
-    /** helper method */
-    private static boolean equal(Object a, Object b) {
-        if (a==null && b==null) return true;
-        if (a==null || b==null) return false;
-        return a.equals(b);
-    }
-
-    /** this is something of a hack right now */
-    private static long code(Parser.Table.State state, Phase start) {
-        return (((long)state.idx) << 32) | (start==null ? 0 : start.pos);
-    }
 }