+// 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.*;
-import edu.berkeley.sbp.*;
import edu.berkeley.sbp.util.*;
+import edu.berkeley.sbp.Parser.Table.*;
+import edu.berkeley.sbp.Sequence.Pos;
+import edu.berkeley.sbp.Sequence.Pos;
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)
-//
-// - most of our time is now spent creating and storing Reduct instances
-// - we should be able to perform Reduct's immediately after creating them...
-//
-
/** 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; }
- /** corresponds to a positions <i>between tokens</i> the input stream; same as Tomita's U_i's */
- public class Phase {
+ /*
+ HashSet<Reduction> finishedReductions = new HashSet<Reduction>();
+ */
+ int numNewNodes = 0;
+ int numOldNodes = 0;
+ int viewPos = 0;
+ int numReductions = 0;
- /** the token immediately after this phase */
- public final Token token;
+ /** corresponds to a positions <i>between tokens</i> the input stream; same as Tomita's U_i's */
+ class Phase<Tok> implements Invokable<State, StateNode, Forest>, IntegerMappable, GraphViz.ToGraphViz, Iterable<StateNode> {
- /** 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
+ private HashMapBag<Integer,Integer> performed = new HashMapBag<Integer,Integer>();
- /** FIXME */
- public Forest.Ref finalResult = null;
+ public Forest.Many finalResult;
+ private PriorityQueue<Reduction> reductionQueue = new PriorityQueue<Reduction>();
- /** all reductions (pending and completed) */
- private HashSet<Phase.Reduct> reductions = new HashSet<Phase.Reduct>(); /* ALLOC */
+ Parser parser() { return parser; }
+ public void addReduction(Reduction r) {
+ //System.out.println("+ " + r);
+ parser.spin();
+ reductionQueue.add(r);
+ }
- /** all nodes, keyed by the value returned by code() */
- private HashMap<Long,Phase.Node> hash = new HashMap<Long,Phase.Node>(); /* ALLOC */
+ public void invoke(State st, StateNode pred, Forest f) {
+ parser.spin();
+ good |= next.newNode(f, null, pred, st, false);
+ }
- /** the number of pending reductions */
- private int pendingReductions = 0;
- private int totalReductions = 0;
- private HashSet<Reduct> pendingReduct = new HashSet<Reduct>();
+ /** the token immediately after this phase */
+ final Tok token;
+ final int pos;
+ public IntPairMap<StateNode> hash = new IntPairMap<StateNode>(); /* ALLOC */
+ private boolean good = false;
+ private Phase next = null;
+ private Phase prev;
+ private Input.Location location;
+ private Input.Location nextLocation;
+
+ private Forest forest;
+
+ public Phase(State startState) throws ParseFailed, IOException {
+ this(null, null);
+ newNode(null, null, null, startState, true);
+ }
+ public Phase(Phase prev, Forest forest) throws ParseFailed, IOException {
+ this.location = input.getLocation();
+ this.token = (Tok)input.next();
+ this.nextLocation = input.getLocation();
+ this.prev = prev;
+ this.forest = forest;
+ this.pos = prev==null ? 0 : prev.pos+1;
+ if (prev != null) prev.shift(this, forest);
+ numReductions = 0;
+ /*
+ finishedReductions.clear();
+ */
+
+ int minPhasePos = Integer.MAX_VALUE;
+ Reduction best = null;
+ //System.out.println("==============================================================================");
+ while(!reductionQueue.isEmpty()) {
+ Reduction r = reductionQueue.poll();
+ //System.out.println("- " + r);
+ if (r.predPhase() != null)
+ if (r.predPhase().pos > minPhasePos)
+ throw new Error();
+ r.perform();
+ if (r.predPhase() != null) {
+ if (r.predPhase().pos < minPhasePos) {
+ minPhasePos = r.predPhase().pos;
+ best = r;
+ } else if (r.predPhase().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));
+ */
+ best = r;
+ }
+ }
+ /*
+ finishedReductions.add(r);
+ */
+ numReductions++;
+ }
+ if (token==null) shift(null, null);
+ }
- /** the number of nodes in this phase */
- private int numNodes = 0;
+ 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;
+ }
- boolean closed = false;
+ public Input.Location getLocation() { return location; }
+ public Input.Location getNextLocation() { return nextLocation; }
+ public boolean isFrontier() { return hash!=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;
+ /** perform all shift operations, adding promoted nodes to <tt>next</tt> */
+ private void shift(Phase next, Forest f) throws ParseFailed {
+ this.next = next;
+ // this massively improves GC performance
+ if (prev != null) {
+ IntPairMap<StateNode> h = prev.hash;
+ prev.hash = null;
+ prev.performed = null;
+ for(StateNode n : h) n.check();
+ }
+ numOldNodes = hash.size();
+ for(StateNode n : hash.values()) {
+ if (token == null && n.state().isAccepting()) {
+ if (finalResult==null) finalResult = new Forest.Many();
+ for(ResultNode r : n)
+ finalResult.merge(r.getForest());
+ }
+ if (token == null) continue;
+ n.state().invokeShifts(token, this, n, f);
+ }
+ 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(StateNode n : hash) n.check();
}
- public boolean isDone() { return token == null; }
-
- private String error = "generic syntax error";
- public void checkFailure() throws Parser.Failed {
- if (numNodes <= 0)
- throw new Parser.Failed(error, getLocation());
+ Input.Region getRegionFromThisToNext() {
+ return getLocation().createRegion(getNextLocation());
}
- public Token.Location getLocation() { return location; }
+ void newNodeFromReduction(Forest f, Pos reduction, StateNode pred) {
+ int pos = pred.phase().pos;
+ for(int s : reduction.hates())
+ if (performed.contains(pos, s))
+ return;
+ for(int s : reduction.needs())
+ if (!performed.contains(pos, s))
+ return;
+ if (reduction.owner_needed_or_hated() && !performed.contains(pos, reduction.provides()))
+ performed.add(pos, reduction.provides());
+ Parser.Table.State state = (Parser.Table.State)pred.state().gotoSetNonTerminals.get(reduction);
+ if (state!=null)
+ newNode(f, reduction, pred, state, reduction.numPops()<=0);
+ }
/** add a new node (merging with existing nodes if possible)
* @param parent the parent of the new node
* @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.parents.add(parent, true);
- if (p!=parent && !fromEmptyReduction) p.queueReductions(parent);
- }
- private void newNode3(Node parent, Forest pending, Parser.Table.State state, boolean fromEmptyReduction, Phase start) {
+ private boolean newNode(Forest f, Pos reduction, StateNode pred, State state, boolean fromEmptyReduction) {
+ StateNode p = pred==null ? null : hash.get(state, pred.phase());
+ if (p != null) {
+ p.addPred(f, reduction, pred);
+ return !state.doomed();
+ }
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();
+ StateNode n = new StateNode(Phase.this, f, reduction, pred, state, fromEmptyReduction); // ALLOC
+ /** FIXME: this null-result can be used to notice bogus/dead states */
+ for(Object s : state.conjunctStates)
+ newNode(null, null, n, (State)s, fromEmptyReduction);
+ return !n.state().doomed();
}
- /** perform all reduction operations */
- public void reduce() {
- for(Phase.Node n : hash.values()) {
- n.queueEmptyReductions();
- n.queueReductions();
- }
- while(pendingReduct.size()>0)
- pendingReduct.iterator().next().go();
+ 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<StateNode> iterator() { return hash.iterator(); }
+ public GSS getGSS() { return GSS.this; }
+
+ // GraphViz //////////////////////////////////////////////////////////////////////////////
+
+ public GraphViz.StateNode 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()) {
- 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
- reductions = null;
- hash = null;
+ 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)
+ ((StateNode)n).toGraphViz(gv);
+ gv.dump(p);
+ p.flush();
+ p.close();
}
-
- // GSS Nodes //////////////////////////////////////////////////////////////////////////////
-
- private HashMap<Parser.Table.Reduction,Forest> pcache = new HashMap<Parser.Table.Reduction,Forest>();
- /** a node in the GSS */
- public class Node {
-
- private Forest.Ref holder = null;
-
- private HashMap<Parser.Table.Reduction,Forest> cache = null;
-
- /** the set of nodes to which there is an edge starting at this node */
- public final FastSet<Node> parents = new FastSet<Node>(); /* 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 final Phase phase = Phase.this;
-
- public HashMap<Parser.Table.Reduction,Forest> cache() { return cache==null ? (cache = new HashMap<Parser.Table.Reduction,Forest>()) : 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<Node> parents() { return parents; }
-
- /** FIXME */
- public void queueReductions() {
- for(Node n2 : parents)
- queueReductions(n2);
- }
-
- /** FIXME */
- public void queueReductions(Node n2) {
- new Reduct(this, n2, null);
- for(Parser.Table.Reduction r : token==null ? state.getEofReductions() : state.getReductions(token)) {
-
- // currently we have this weird problem where we
- // have to do an individual reduct for each child
- // when the reduction length is one (ie the
- // children wind up being children of the newly
- // created node rather than part of the popped
- // sequence
-
- if (r.numPop == 1) new Reduct(this, n2, r);
- }
- }
-
-
- /** FIXME */
- public void queueEmptyReductions() {
- for(Parser.Table.Reduction r : token==null ? state.getEofReductions() : state.getReductions(token)) {
- if (r.numPop==0)
- new Reduct(this, null, r); /* ALLOC */
- }
- }
-
- 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
- }
- }
-
-
- // Forest / Completed Reductions //////////////////////////////////////////////////////////////////////////////
-
- /** a pending or completed reduction */
- class Reduct {
-
- /** the node from which the reduction should begin */
- public Node n = null;
-
- /** the node on the other end of the edge to be reduced along (either: null, the second node of the reduction,
- * or the parent of the result of a length-one reduction)
- */
- public Node n2 = null;
-
- /** true iff the reduction has already been performed */
- private boolean done = false;
-
- /** the reduction to be applied */
- public Parser.Table.Reduction r;
-
- public Tree<String> result = null;
-
- public Reduct(Node n, Node n2, Parser.Table.Reduction r) {
- this.n = n;
- this.n2 = n2;
- this.r = r;
- if (reductions.contains(this)) { done = true; return; }
- reductions.add(this);
- pendingReduct.add(this);
- pendingReductions++;
- }
-
- /** perform the reduction */
- public void go() {
- if (done) return;
- done = true;
- pendingReduct.remove(this);
- pendingReductions--;
-
- // FIXME: explain this
- if (r==null) {
- for(Parser.Table.Reduction r : token==null ? n.state.getEofReductions() : n.state.getReductions(token)) {
- if (r.numPop <= 1) continue;
- r.reduce(n, n2, Phase.this, null);
- }
- } else if (r.numPop<=1) {
- // UGLY HACK
- // The problem here is that a "reduction of length 0/1"
- // performed twice with different values of n2 needs
- // to only create a *single* new result, but must add
- // multiple parents to the node holding that result.
- // The current reducer doesn't differentiate between
- // the next node of an n-pop reduction and the
- // ultimate parent of the last pop, so we need to
- // cache instances here as a way of avoiding
- // recreating them.
-
- Forest ret = (r.numPop==0 ? pcache : n.cache()).get(r);
- if (ret != null) r.reduce(n, n2, n.phase, ret);
- else (r.numPop==0 ? pcache : n.cache()).put(r, r.reduce(n, n2, n.phase, null));
-
- } else {
- r.reduce(n, n2, Phase.this, null);
- }
- }
-
- // FIXME: this is a PITA
- public int hashCode() { return n.hashCode() ^ (r==null ? 0 : r.hashCode()) ^ (n2==null ? 0 : n2.hashCode()); }
- public boolean equals(Object o) {
- if (o==null) return false;
- if (o==this) return true;
- if (!(o instanceof Reduct)) return false;
- Reduct other = (Reduct)o;
- return equal(r, other.r) && equal(n, other.n) && equal(n2, other.n2);
- }
- }
-
- }
-
- /** 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);
}
}