X-Git-Url: http://git.megacz.com/?p=sbp.git;a=blobdiff_plain;f=src%2Fedu%2Fberkeley%2Fsbp%2FGSS.java;h=ca8e1d505368093ab6d2f3598a6215291d9bda13;hp=3843f57020efe11b1df60b4612e8c818e89ddb91;hb=dc9bb3a45ed306e2e35549076842b3e74efecb48;hpb=ebb5fe5647046306f415e31e4967b23169c9004e diff --git a/src/edu/berkeley/sbp/GSS.java b/src/edu/berkeley/sbp/GSS.java index 3843f57..ca8e1d5 100644 --- a/src/edu/berkeley/sbp/GSS.java +++ b/src/edu/berkeley/sbp/GSS.java @@ -1,6 +1,9 @@ +// Copyright 2006 all rights reserved; see LICENSE file for BSD-style license + package 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.*; @@ -9,122 +12,153 @@ import java.lang.reflect.*; /** 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 between tokens the input stream; same as Tomita's U_i's */ - public class Phase implements Invokable { - - /** the token immediately after this phase */ - public final Token token; - - boolean reducing = false; - - /** currently this is necessary only for the code() hack -- it doesn't actually correspond to the input */ - private final int pos; + class Phase implements Invokable, IntegerMappable, GraphViz.ToGraphViz, Iterable { - /** FIXME */ - public Forest.Ref finalResult = null; + // FIXME: right now, these are the performance bottleneck + private HashMapBag performed = new HashMapBag(); - /** all nodes, keyed by the value returned by code() */ - /*private*/ HashMap hash = new HashMap(); /* ALLOC */ + public Forest.Many finalResult; + private PriorityQueue reductionQueue = new PriorityQueue(); - /** the number of nodes in this phase */ - private int numNodes = 0; - - boolean closed = false; - - 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; + Parser parser() { return parser; } + public void addReduction(Reduction r) { + //System.out.println("+ " + r); + parser.spin(); + reductionQueue.add(r); } - - public void complain(Node n, HashMap> errors, boolean force) { - if (n.touched) return; - n.touched = true; - for(Position p : n.state) { - //if (!p.isLast()) { - if (((p.isFirst() || p.isLast()) && !force) || p.owner().name==null) { - for(Node n2 : n.parents()) - complain(n2, errors, force | p.isFirst()); - } else { - String seqname = p.owner().name; - HashSet hs = errors.get(seqname); - if (hs==null) errors.put(seqname, hs = new HashSet()); - hs.add(p.element()+""); - //String s = " while parsing " + seqname + ": expected a " + p.element(); - //"\n"; - /* - s += " parsed: "; - for(Position p2 = p.owner().firstp(); p2 != null && p2 != p && !p2.isLast(); p2 = p2.next()) s += (p2.element() + " "); - s += "\n"; - s += " expected: "; - for(Position p2 = p; p2 != null && !p2.isLast(); p2 = p2.next()) s += (p2.element() + " "); - */ - //s += "\n"; - //errors.add(s); - } - } + public void invoke(State st, Result result) { + parser.spin(); + good |= next.newNode(result, st, false); } - public String black(Object o) { return "\033[30m"+o+"\033[0m"; } - public String red(Object o) { return "\033[31m"+o+"\033[0m"; } - public String green(Object o) { return "\033[32m"+o+"\033[0m"; } - public String yellow(Object o) { return "\033[33m"+o+"\033[0m"; } - public String blue(Object o) { return "\033[34m"+o+"\033[0m"; } - public String purple(Object o) { return "\033[35m"+o+"\033[0m"; } - public String cyan(Object o) { return "\033[36m"+o+"\033[0m"; } - public String el(Object e) { - String s = e.toString(); - if (s.length()==0 || s.charAt(0)!='\"' || s.charAt(s.length()-1)!='\"') return yellow(s); - s = s.substring(1); - s = s.substring(0, s.length()-1); - StringBuffer ret = new StringBuffer(); - for(int i=0; i hash = new IntPairMap(); /* 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; + + public Phase(State startState) throws ParseFailed, IOException { + this(null, null); + Result primordealResult = new Result(null, null, null); + newNode(primordealResult, startState, true); } - public String error(String message) { - String lookAhead = token==null ? "" : token.toString(); - StringBuffer ret = new StringBuffer(); - ret.append("\n "); - ret.append(message); - HashMap> errors = new HashMap>(); - for(Node n : hash.values()) complain(n, errors, false); - for(String s : errors.keySet()) { - ret.append(" while parsing " + yellow(s)); - HashSet hs = errors.get(s); - if (hs.size()==1) ret.append(" expected " + yellow(el(hs.iterator().next())) + "\n"); - else { - ret.append(" expected "); - boolean first = true; - for(String s2 : hs) { - if (!first) ret.append(" or "); - first = false; - ret.append(yellow(el(s2))); + 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; } - ret.append("\n"); } + numReductions++; } - return ret.toString(); + if (token==null) shift(null, null); } - - public boolean isDone() throws Parser.Failed { + + public boolean isDone() throws ParseFailed { if (token != null) return false; if (token==null && finalResult==null) - throw new Parser.Failed(error(red("unexpected end of file\n")), - getLocation()); + ParseFailed.error("unexpected end of file", this); return true; } - private String error = "generic syntax error"; + 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; } - public Token.Location getLocation() { return location; } + /** perform all shift operations, adding promoted nodes to next */ + private void shift(Phase next, Forest result) throws ParseFailed { + this.next = next; + // this massively improves GC performance + if (prev != null) { + IntPairMap 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) ParseFailed.error("unexpected character", this); + if (token==null && finalResult==null) ParseFailed.error("unexpected end of file", this); + for(Node n : hash) n.check(); + } + + 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 @@ -133,166 +167,52 @@ 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 boolean newNode(Node parent, Forest pending, Parser.Table.State state, boolean fromEmptyReduction) { - Node p = hash.get(code(state, parent==null?null:parent.phase())); - if (p != null) return newNode2(p, parent, pending, state, fromEmptyReduction); - else return newNode3(parent, pending, state, fromEmptyReduction); - } - private boolean newNode2(Node p, Node parent, Forest pending, Parser.Table.State state, boolean fromEmptyReduction) { - p.holder.merge(pending); - if (p.parents().contains(parent)) return true; - //if (p.fe && p.phase() != parent.phase()) throw new Error("yep yep"); - //if (!p.fe && p.phase() == parent.phase()) throw new Error("yep yep2"); - p.parents().add(parent, true); - if (p!=parent && !fromEmptyReduction) p.queueReductions(parent); - return true; - } - private boolean newNode3(Node parent, Forest pending, Parser.Table.State state, boolean fromEmptyReduction) { + 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; - //if (!state.canReduce(token)) return false; - //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, fromEmptyReduction); // 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 hv = hash.values(); - hv.toArray(reducing_list); - int num = hv.size(); - for(int i=0; inext */ - public void shift(Phase next, Forest result) throws Parser.Failed { - this.next = next; - 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()) { - if (finalResult==null) finalResult = new Forest.Ref(); - finalResult.merge(n.holder); - } - if (!n.holder.valid()) continue; - if (token == null) continue; - n.state.invokeShifts(token, this, result, n); - } - - if (!good && token!=null) - throw new Parser.Failed(error(red("unexpected character")+" "+purple(token)+" encountered at "+green(getLocation())+"\n"), - getLocation()); - if (token==null && finalResult==null) - throw new Parser.Failed(error(red("unexpected end of file\n")), - getLocation()); - - // this massively improves GC performance - hash = null; + 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 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; } - - - // GSS Nodes ////////////////////////////////////////////////////////////////////////////// - - /** a node in the GSS */ - public final class Node extends FastSet implements Invokable { - - public boolean touched = false; - 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 parents() { return this; } - - public void queueReductions() { - if (!reducing) return; - if (allqueued) return; - allqueued = true; - int where = parents().size(); - state.invokeReductions(token, this, this, null); - } - - public void queueReductions(Node n2) { - if (!allqueued) { queueReductions(); return; } - state.invokeReductions(token, this, this, n2); - } - - public final void invoke(Parser.Table.Reduction r, Node n, Node n2) { - if (n==null) { - if (r.numPop==0) r.reduce(this); - return; - } - if (r.numPop==0) return; - if (n2==null) r.reduce(n); - else r.reduce(n, n2); - } - public void queueEmptyReductions() { - if (!reducing) return; - state.invokeReductions(token, this, null, null); - } - - private boolean fe; - private Node(Node parent, Forest pending, Parser.Table.State state, boolean fe) { - this.fe = fe; - this.state = state; - Phase start = parent==null ? null : parent.phase(); - 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+1)); - } }