X-Git-Url: http://git.megacz.com/?p=sbp.git;a=blobdiff_plain;f=src%2Fedu%2Fberkeley%2Fsbp%2FGSS.java;h=0358ef64603ea89b392b28dfd1a6140290f1bd76;hp=77464c25b4f96d28087c836b8d1f99a6bdb27afb;hb=a8478f5ddfbfbc8d910d09f27163cbd55752d3b6;hpb=c263c1e02119917a9d63ab14db9711c7d1cb4d9d diff --git a/src/edu/berkeley/sbp/GSS.java b/src/edu/berkeley/sbp/GSS.java index 77464c2..0358ef6 100644 --- a/src/edu/berkeley/sbp/GSS.java +++ b/src/edu/berkeley/sbp/GSS.java @@ -1,72 +1,125 @@ +// 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() { } + public static Queue removals = new LinkedList(); - private Phase.Node[] reducing_list = null; + static String note = ""; + static int single_newnode = 0; + static int toplevel_reductions = 0; + static int multi_newnode = 0; + static int waiting_newnode = 0; + static int shifts = 0; - /** corresponds to a positions between tokens the input stream; same as Tomita's U_i's */ - public class Phase { + static int count = 0; + static int reductions = 0; + int resets = 0; + int waits = 0; + + Input input; - /** the token immediately after this phase */ - public final Token token; + public GSS(Input input) { this.input = input; } - boolean reducing = false; + private Node[] reducing_list = null; - /** currently this is necessary only for the code() hack -- it doesn't actually correspond to the input */ - private final int pos; + // FIXME: right now, these are the performance bottleneck + HashMapBag waiting = new HashMapBag(); + HashMapBag performed = new HashMapBag(); + HashMapBag lastperformed = new HashMapBag(); + HashMapBag expected = new HashMapBag(); + + /** FIXME */ + Forest.Many finalResult; + + /** corresponds to a positions between tokens the input stream; same as Tomita's U_i's */ + class Phase implements Invokable, IntegerMappable, GraphViz.ToGraphViz, Iterable { + + public int pos() { return pos; } + public boolean closed() { return closed; } + public Tok token() { return token; } + + public Iterator iterator() { return hash.iterator(); } + public void invoke(State st, Forest result, Node n) { + shifts++; + good |= next.newNode(n, result, st, false); + } - /** FIXME */ - public Forest.Ref finalResult = null; + /** the token immediately after this phase */ + final Tok token; - /** all nodes, keyed by the value returned by code() */ - private HashMap hash = new HashMap(); /* ALLOC */ + private final int pos; - /** the number of nodes in this phase */ - private int numNodes = 0; + boolean reducing; + public IntPairMap hash; /* ALLOC */ + private boolean closed; + private boolean good; + private Phase next = null; + private Phase prev; + private Input.Location location; + private Input.Location nextLocation; + private Input.Location prevLocation; + + public final Parser parser; - boolean closed = false; + private Forest forest; - private Token.Location location; - public Phase(Phase previous, Token token, Token.Location location) { + public Phase(Phase prev, Parser parser, Phase previous, Tok token, Input.Location location, + Input.Location nextLocation, Forest forest) throws ParseFailed { + this.prevLocation = prev==null ? location : prev.getLocation(); + this.prev = prev; + this.forest = forest; + this.parser = parser; this.pos = previous==null ? 0 : previous.pos+1; this.token = token; this.location = location; + this.nextLocation = nextLocation; + performed.clear(); + reset(); } - public boolean isDone() { return token == null; } + public void reset() throws ParseFailed { + waiting.clear(); + expected.clear(); + lastperformed.clear(); + lastperformed.addAll(performed); + performed.clear(); + hash = new IntPairMap(); + reset = false; + good = false; + closed = false; + reducing = false; + finalResult = null; + if (prev != null) prev.shift(this, forest); + } - private String error = "generic syntax error"; - public void checkFailure() throws Parser.Failed { - if (numNodes <= 0) - throw new Parser.Failed(error, getLocation()); + + public boolean isDone() throws ParseFailed { + if (token != null) return false; + if (token==null && finalResult==null) + ParseFailed.error("unexpected end of file", + getLocation(), + token, + hash.values(), + getLocation().createRegion(getLocation()), + input, + GSS.this); + return true; } - public Token.Location getLocation() { return location; } + public Input.Location getPrevLocation() { return prevLocation; } + public Input.Location getLocation() { return location; } + public Input.Region getRegion() { return getPrevLocation().createRegion(getLocation()); } + public Input.Location getNextLocation() { return nextLocation; } /** add a new node (merging with existing nodes if possible) * @param parent the parent of the new node @@ -75,207 +128,220 @@ 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); + public boolean newNode(Node parent, Forest pending, State state, boolean fromEmptyReduction) { + Node p = hash.get(state, parent==null?null:parent.phase()); + if (p != null) return newNode2(p, parent, pending, state, fromEmptyReduction); + else return newNode3(parent, pending, state, fromEmptyReduction); } - 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); + public void newNode(Node parent, Forest pending, State state, boolean fromEmptyReduction, Position reduction) { + int pos = parent==null?0:parent.phase()==null?0:parent.phase().pos; + Sequence owner = reduction==null ? null : reduction.owner(); + if (reduction!=null) { + if (owner.hates!=null) { + for (Sequence s : performed.getAll(pos)) + if (owner.hates.contains(s)) + return; + for (Sequence s : lastperformed.getAll(pos)) + if (owner.hates.contains(s)) { + //System.out.println("now expecting ["+pos+"] => " + s); + expected.add(pos, s); + return; + } + } + if (owner.needs != null) + for(Sequence s : owner.needs) + if (!performed.contains(pos, s)) { + waiting.add(s, new Waiting(parent, pending, state, fromEmptyReduction, reduction)); + return; + } + if (!performed.contains(pos, owner)) { + performed.add(pos, owner); + if (owner.hated != null) + for(Sequence seq : owner.hated) + if (performed.contains(pos, seq)) { + performed.remove(pos, seq); + reset = true; + } + } + } + newNode(parent, pending, state, fromEmptyReduction); + if (reduction != null) { + boolean redo = true; + while(redo) { + redo = false; + for(Waiting w : waiting.getAll(owner)) { + if (w.parent==parent || (parent!=null&&w.parent!=null&&w.parent.phase()==parent.phase())) { + waiting.remove(owner, w); + w.perform(); + redo = true; + break; + } + } + } + } + } + + private boolean newNode2(Node p, Node parent, Forest pending, State state, boolean fromEmptyReduction) { + if (p.merge(parent, pending)) return true; + p.addParent(parent, true); + if (p!=parent && !fromEmptyReduction && reducing) p.performReductions(parent); + return true; } - private void newNode3(Node parent, Forest pending, Parser.Table.State state, boolean fromEmptyReduction, Phase start) { + + private boolean newNode3(Node parent, Forest pending, State state, boolean fromEmptyReduction) { 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 (!state.canReduce(token)) return false; //if (count > 1) break; //if (r.numPop == 0) break; //r.reduce(pending, parent, null, Phase.this, null); //return; } while(false); - Node n = new Node(parent, pending, state, start); // ALLOC - n.queueEmptyReductions(); - if (!fromEmptyReduction) n.queueReductions(parent); + Node n = new Node(Phase.this, parent, pending, state); // ALLOC + if (reducing) { + n.performEmptyReductions(); + if (!fromEmptyReduction) n.performReductions(parent); + } + return true; } - + public LinkedList reductionQueue = new LinkedList(); + /** 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) { + public void shift(Phase next, Forest result) throws ParseFailed { + // this massively improves GC performance + if (prev!=null && parser.helpgc) { + //prev.hash = null; + //System.out.println("\r" + /*shifts + " " + */ single_newnode /*+ "/"+multi_newnode + " " + waiting_newnode*/); + //System.out.println("\r" + shifts + " " + note); + //System.out.println("\r" + shifts); + //System.out.println("\r" + toplevel_reductions); + //System.out.println("\r" + multi_newnode); + single_newnode = 0; + note = ""; + multi_newnode = 0; + toplevel_reductions = 0; + waiting_newnode = 0; + shifts = 0; + } + 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()) { - ok = true; - if (finalResult==null) finalResult = new Forest.Ref(); - finalResult.merge(n.holder); + int count = 0; + for(Node n : hash.values()) { + if (token == null && n.state().isAccepting()) { + if (finalResult==null) finalResult = new Forest.Many(); + for(Object f : n.results()) + finalResult.merge((Forest)f); } - 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(); + n.state().invokeShifts(token, this, result, n); } - - // this massively improves GC performance - hash = null; + //System.out.println(next.hash.size()); + if (!good && token!=null) + ParseFailed.error("unexpected character", + getLocation(), + token, + hash.values(), + getRegion(), + input, + GSS.this); + + if (token==null && finalResult==null) + ParseFailed.error("unexpected end of file", + getLocation(), + token, + hash.values(), + getLocation().createRegion(getLocation()), + input, + GSS.this); } - - // GSS Nodes ////////////////////////////////////////////////////////////////////////////// - - /** a node in the GSS */ - public final class Node extends FastSet { - - 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; - - private HashMap cache = null; - - /** the set of nodes to which there is an edge starting at this node */ - //public final FastSet parents = new FastSet(); /* ALLOC */ - /** 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 HashMap cache() { - return cache==null ? (cache = new HashMap()) : cache; - } - 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; } - - /** FIXME */ - public void queueReductions() { - if (allqueued) return; - allqueued = true; - int where = parents().size(); - for(Parser.Table.Reduction r : token==null ? state.getEofReductions() : state.getReductions(token)) - if (r.numPop > 1) - r.reduce(this, null, null); - for(int i=0; i