X-Git-Url: http://git.megacz.com/?p=sbp.git;a=blobdiff_plain;f=src%2Fedu%2Fberkeley%2Fsbp%2FGSS.java;h=89a16ed8c5c3e9b22e07be5fe5d7e1aa7ed1d6f0;hp=8165ccdff0d4dc6f544b443b5f199a64c77a67c5;hb=5d18f5606c9296e6b0c5749f05fc68f358ace2f6;hpb=6c2a17e674b32389d281261796e9e962e9d9380c
diff --git a/src/edu/berkeley/sbp/GSS.java b/src/edu/berkeley/sbp/GSS.java
index 8165ccd..89a16ed 100644
--- a/src/edu/berkeley/sbp/GSS.java
+++ b/src/edu/berkeley/sbp/GSS.java
@@ -1,68 +1,87 @@
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() { }
- /** corresponds to a positions between tokens the input stream; same as Tomita's U_i's */
- public class Phase {
+ private Phase.Node[] reducing_list = null;
+ public int resets = 0;
+ public int waits = 0;
- /** the token immediately after this phase */
- public final Token token;
-
- /** currently this is necessary only for the code() hack -- it doesn't actually correspond to the input */
- private final int pos;
+ HashMapBag inhibited = new HashMapBag();
+ HashMapBag expectedInhibit = new HashMapBag();
+ HashMapBag waiting = new HashMapBag();
+ HashMapBag performed = new HashMapBag();
+
+ /** FIXME */
+ public Forest.Ref finalResult;
- /** FIXME */
- public Forest.Ref finalResult = null;
+ /** corresponds to a positions between tokens the input stream; same as Tomita's U_i's */
+ public class Phase implements Invokable.Node>, IntegerMappable {
- /** all nodes, keyed by the value returned by code() */
- private HashMap hash = new HashMap(); /* ALLOC */
+ public void invoke(State st, Forest result, Node n) {
+ good |= next.newNode(n, result, st, false);
+ }
- /** the number of nodes in this phase */
- private int numNodes = 0;
+ /** the token immediately after this phase */
+ final Tok token;
- boolean closed = false;
+ private final int pos;
- private Token.Location location;
- public Phase(Phase previous, Token token, Token.Location location) {
+ boolean reducing;
+ private IntPairMap hash; /* ALLOC */
+ private IntPairMap singularReductions; /* ALLOC */
+ private boolean closed;
+ private boolean good;
+ private Phase next = null;
+ private Phase prev;
+ private Input.Location location;
+ public final Parser parser;
+
+ private Forest forest;
+
+ public Phase(Phase prev, Parser parser, Phase previous, Tok token, Input.Location location, Forest forest) {
+ this.prev = prev;
+ this.forest = forest;
+ this.parser = parser;
this.pos = previous==null ? 0 : previous.pos+1;
this.token = token;
this.location = location;
+ inhibited.clear();
+ reset();
}
- public boolean isDone() { return token == null; }
+ public void reset() {
+ waiting.clear();
+ performed.clear();
+ hash = new IntPairMap();
+ singularReductions = new IntPairMap();
+ expectedInhibit.clear();
+ expectedInhibit.addAll(inhibited);
+ 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)
+ throw new ParseFailed(ParseFailed.error(ANSI.red("unexpected end of file\n"), token, hash.values()), getLocation());
+ return true;
}
- public Token.Location getLocation() { return location; }
+ public Input.Location getLocation() { return location; }
/** add a new node (merging with existing nodes if possible)
* @param parent the parent of the new node
@@ -71,197 +90,283 @@ 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);
+ }
+ 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 (inhibited.contains(pos, owner)) 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 ((owner.needed != null && owner.needed.size()>0) ||
+ (owner.hated != null && owner.hated.size()>0) ||
+ (owner.hates != null && owner.hates.size()>0))
+ performed.add(pos, owner);
+ }
+ if (!owner.lame)
+ newNode(parent, pending, state, fromEmptyReduction);
+ if (reduction!=null) inhibit(reduction, parent==null?0:parent.phase().pos);
+ 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 void newNode2(Node p, Node parent, Forest pending, Parser.Table.State state, boolean fromEmptyReduction, Phase start) {
+ private boolean newNode2(Node p, Node parent, Forest pending, State state, boolean fromEmptyReduction) {
p.holder.merge(pending);
- if (p.parents.contains(parent)) return;
- p.parents.add(parent, true);
- if (p!=parent && !fromEmptyReduction) p.queueReductions(parent);
+ if (p.parents().contains(parent)) return true;
+ p.parents().add(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(parent, pending, state); // ALLOC
+ if (reducing) {
+ n.performEmptyReductions();
+ if (!fromEmptyReduction) n.performReductions(parent);
+ }
+ return true;
}
+ public void uninhibit(int p, Sequence s) {
+ if (s.hated!=null)
+ for(Sequence s2 : s.hated)
+ inhibited.remove(p, s2);
+ }
+
+ public void inhibit(Position r, int p) {
+ if (r.owner().hated == null) return;
+ // remember that dead states are still allowed to shift -- just not allowed to reduce
+ boolean reset = false;
+ for(Sequence seq : r.owner().hated) {
+ if (performed.contains(p,seq)) {
+ uninhibit(p, seq);
+ //System.out.println("\nresetting due to " + r.owner() + " killing " + seq);
+ //inhibited.clear();
+ inhibited.add(p, seq);
+ //inhibited = new HashMapBag();
+ reset = true;
+ resets++;
+ throw new Reset();
+ }
+ inhibited.add(p, seq);
+ expectedInhibit.remove(p, seq);
+ }
+ }
- boolean reducing = false;
/** perform all reduction operations */
public void reduce() {
- reducing = true;
- HashSet s = new HashSet();
- s.addAll(hash.values());
- for(Phase.Node n : s) n.queueEmptyReductions();
- for(Phase.Node n : s) n.queueReductions();
+ try {
+ reducing = true;
+ if (reducing_list==null || reducing_list.length < hash.size())
+ reducing_list = new Phase.Node[hash.size() * 4];
+ hash.toArray(reducing_list);
+ int num = hash.size();
+ for(int i=0; i 0) {
+ inhibited.removeAll(expectedInhibit);
+ System.out.println("\n!!!!\n");
+ throw new Reset();
+ }
+ } catch (Reset r) {
+ reset();
+ reduce();
+ }
}
+ class Reset extends RuntimeException { }
+
/** perform all shift operations, adding promoted nodes to next */
- public void shift(Phase next, Forest result) {
+ public void shift(Phase next, Forest result) throws ParseFailed {
+ // this massively improves GC performance
+ if (prev!=null) {
+ prev.hash = null;
+ prev.singularReductions = null;
+ }
+ 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);
}
- 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;
- }
+ n.state.invokeShifts(token, this, result, n);
}
- 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;
+ if (!good && token!=null)
+ throw new ParseFailed(ParseFailed.error(ANSI.red("unexpected character ")+" \'"+
+ ANSI.purple(StringUtil.escapify(token+"", "\\\'\r\n"))+
+ "\' encountered at "+
+ ANSI.green(getLocation())+"\n", token, hash.values()),
+ getLocation());
+ if (token==null && finalResult==null)
+ throw new ParseFailed(ParseFailed.error(ANSI.red("unexpected end of file\n"), token, hash.values()),
+ getLocation());
}
+
+ public class Waiting {
+ Node parent;
+ Forest pending;
+ State state;
+ boolean fromEmptyReduction;
+ Position reduction;
+ public Waiting(Node parent, Forest pending, State state, boolean fromEmptyReduction, Position reduction) {
+ waits++;
+ this.parent = parent;
+ this.pending = pending;
+ this.state = state;
+ this.fromEmptyReduction = fromEmptyReduction;
+ this.reduction = reduction;
+ }
+ public void perform() {
+ //System.out.println("performing: " + reduction.position);
+ newNode(parent, pending, state, fromEmptyReduction, reduction);
+ }
+ }
- // GSS Nodes //////////////////////////////////////////////////////////////////////////////
+ // Node /////////////////////////////////////////////////////////////////////////////////
/** a node in the GSS */
- public final class Node {
+ public final class Node extends FastSet implements Invokable, IntegerMappable {
private Forest.Ref holder = null;
-
- 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 */
+ 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 final Phase phase = Phase.this;
+ public final Parser.Table.State state;
- public HashMap cache() {
- return cache==null ? (cache = new HashMap()) : cache; }
+ /** 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 parents; }
+ public FastSet parents() { return this; }
- /** FIXME */
- public void queueReductions() {
+ public void performReductions() {
if (allqueued) return;
allqueued = true;
- FastSet h = new FastSet();
- for(Node n : parents) h.add(n);
- for(Node n : h) queueReductions(n);
+ state.invokeReductions(token, this, this, null);
}
- private boolean allqueued = false;
- private HashSet queued = new HashSet();
- /** FIXME */
- public void queueReductions(Node n2) {
- if (!allqueued) { queueReductions(); return; }
- if (queued.contains(n2)) return;
- queued.add(n2);
- Node n = this;
- for(Parser.Table.Reduction r : token==null ? n.state.getEofReductions() : n.state.getReductions(token)) {
-
- // UGLY HACK
- // The problem here is that a "reduction of length 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.
-
- // 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 <= 0) continue;
- if (r.numPop == 1) {
- Forest ret = n.cache().get(r);
- if (ret != null) r.reduce(n, n2, n.phase, ret);
- else n.cache().put(r, r.reduce(n, n2, n.phase, null));
+ public void performReductions(Node n2) {
+ if (!allqueued) performReductions();
+ else state.invokeReductions(token, this, this, n2);
+ }
+
+ public void performEmptyReductions() { state.invokeReductions(token, this, null, null); }
+ public final void invoke(Position r, Node n, Node n2) {
+ if (n==null || n2==null || r.pos==0) {
+ if (r.pos==0) {
+ if (n==null) n = this;
+ else return;
+ }
+ if (n==null) return;
+ Forest[] holder = new Forest[r.pos];
+ if (r.pos==0) n.finish(r, r.zero(), n.phase(), holder);
+ else n.reduce(r, r.pos-1, n.phase(), holder);
+ } else {
+ Forest[] holder = new Forest[r.pos];
+ if (r.pos<=0) throw new Error("called wrong form of reduce()");
+ int pos = r.pos-1;
+ Forest old = holder[pos];
+ holder[pos] = n.pending();
+ if (pos==0) {
+ System.arraycopy(holder, 0, r.holder, 0, holder.length);
+ Forest rex = null;
+ if (r.pos==1) rex = singularReductions.get(this, r);
+ if (rex==null) {
+ rex = r.rewrite(n.phase().getLocation());
+ if (r.pos==1) singularReductions.put(this, r, rex);
+ }
+ n2.finish(r, rex, n.phase(), holder);
} else {
- r.reduce(n, n2, Phase.this, null);
+ n2.reduce(r, pos-1, n.phase(), holder);
}
+ holder[pos] = old;
}
}
+ public void reduce(Position r, int pos, Phase target, Forest[] holder) {
+ Forest old = holder[pos];
+ holder[pos] = this.pending();
+ if (pos==0) {
+ System.arraycopy(holder, 0, r.holder, 0, holder.length);
+ for(int i=0; i target, Forest[] holder) {
+ Parser.Table.State state0 = state.gotoSetNonTerminals.get(r.owner());
+ if (result==null) throw new Error();
+ if (state0!=null)
+ target.newNode(this, result, state0, r.pos<=0, r);
}
- private Node(Node parent, Forest pending, Parser.Table.State state, Phase start) {
+ private Node(Node parent, Forest pending, State state) {
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
+ this.holder().merge(pending);
+ Phase start = parent==null ? null : parent.phase();
+ if (parent != null) parents().add(parent, true);
+ if (Phase.this.hash.get(state, start) != null) throw new Error("severe problem!");
+ Phase.this.hash.put(state, start, this);
}
+ public int toInt() { return idx; }
+ private final int idx = node_idx++;
}
+ private int node_idx = 0;
+ public int toInt() { return pos+1; }
+ public int size() { return hash==null ? 0 : hash.size(); }
}
- /** 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);
- }
- public boolean yak = false;
}