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 class GSS {
+ public static int count = 0;
+ public static int shifts = 0;
+ public static int reductions = 0;
+
public GSS() { }
private Phase.Node[] reducing_list = null;
+ public int resets = 0;
+ public int waits = 0;
+
+ // FIXME: right now, these are the performance bottleneck
+ HashMapBag<Sequence,Phase.Waiting> waiting = new HashMapBag<Sequence,Phase.Waiting>();
+ HashMapBag<Integer,Sequence> performed = new HashMapBag<Integer,Sequence>();
+ HashMapBag<Integer,Sequence> lastperformed = new HashMapBag<Integer,Sequence>();
+ HashMapBag<Integer,Sequence> expected = new HashMapBag<Integer,Sequence>();
+
+ /** FIXME */
+ public Forest.Ref finalResult;
/** 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, Forest, Phase<Tok>.Node>, IntegerMappable, GraphViz.ToGraphViz, Iterable<Phase.Node> {
+
+ public Iterator<Phase.Node> iterator() { return hash.iterator(); }
+ public void invoke(State st, Forest result, Node n) {
+ shifts++;
+ good |= next.newNode(n, result, st, false);
+ }
/** the token immediately after this phase */
- public final Token token;
+ final Tok token;
- /** currently this is necessary only for the code() hack -- it doesn't actually correspond to the input */
private final int pos;
- /** 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 */
-
- /** 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) {
+ boolean reducing;
+ private IntPairMap<Phase.Node> hash; /* 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) throws ParseFailed {
+ this.prev = prev;
+ this.forest = forest;
+ this.parser = parser;
this.pos = previous==null ? 0 : previous.pos+1;
this.token = token;
this.location = location;
+ 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<Phase.Node>();
+ 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)
+ 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
* @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;
+ 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;
+ }
+ }
+ }
+ if (!owner.lame)
+ 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.parents().add(parent, true);
- if (p!=parent && !fromEmptyReduction) p.queueReductions(parent);
+ 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;
}
-
- boolean reducing = false;
/** 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];
- reducing_list[i] = null;
- n.queueEmptyReductions();
- n.queueReductions();
+ public void reduce() throws ParseFailed {
+ 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<num; i++) {
+ Node n = reducing_list[i];
+ n.performEmptyReductions();
+ // 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.performReductions();
+ }
+ if (reset) {
+ reset = false;
+ resets++;
+ throw new Reset();
+ }
+ for(int i : expected)
+ for(Sequence s : expected.getAll(i))
+ if (!performed.contains(i, s)) {
+ //System.out.println("resetting due to pos="+i+": " + s + " " + System.identityHashCode(s));
+ resets++;
+ throw new Reset();
+ }
+ } catch (Reset r) {
+ reset();
+ reduce();
}
+ count = 0;
}
+ private boolean reset = false;
+ class Reset extends RuntimeException { }
+
/** perform all shift operations, adding promoted nodes to <tt>next</tt> */
- 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;
+ }
+ 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);
+ 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;
- }
+ 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());
}
+
+ 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 extends FastSet<Node> {
+ final class Node implements Invokable<Position, Node, Node>, IntegerMappable, GraphViz.ToGraphViz {
+ public FastSet<Node> set = new FastSet<Node>();
- private Forest.Ref holder = null;
+
private boolean allqueued = false;
- 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;
+ public final Parser.Table<Tok>.State<Tok> 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 this; }
+ public Phase phase() { return Phase.this; }
+
+ private HashSet<Forest.Ref> resultMap = new HashSet<Forest.Ref>();
+ public Iterable<Forest.Ref> results() { return resultMap; }
+ public FastSet<Node> parents() { return set; }
+ public boolean merge(Node parent, Forest result) {
+ // FIXME: inefficient!
+ for(Forest.Ref f : results()) {
+ if (f.parents.contains(parent) /* UGLY: */ && f.parents.size()==1) {
+ f.merge(result);
+ return true;
+ }
+ }
+ Forest.Ref f = new Forest.Ref();
+ f.parents.add(parent);
+ f.merge(result);
+ resultMap.add(f);
+ set.add(parent, true);
+ return false;
+ }
- /** FIXME */
- public void queueReductions() {
+ public void performReductions() {
if (allqueued) return;
allqueued = true;
- int where = parents().size();
- for(int i=0; i<where; i++) queueReductions(get(i));
+ state.invokeReductions(token, this, this, null);
+ }
+
+ public void performReductions(Node n2) {
+ if (!allqueued) performReductions();
+ else state.invokeReductions(token, this, this, n2);
}
- /** FIXME */
- public void queueReductions(Node n2) {
- if (!allqueued) { queueReductions(); return; }
- 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));
- } else {
- r.reduce(n, n2, Phase.this, null);
+ public void performEmptyReductions() { state.invokeReductions(token, this, null, null); }
+ public final void invoke(Position r, Node n, Node n2) {
+ reductions++;
+ 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());
+ else n.reduce(r, r.pos-1, n.phase(), null);
+ } else {
+ if (r.pos<=0) throw new Error("called wrong form of reduce()");
+ int pos = r.pos-1;
+ n.reduce(r, pos, n.phase(), n2);
}
}
+ public void reduce(Position r, int pos, Phase target, Node only) {
+ Forest[] holder = r.holder;
+ Forest old = holder[pos];
+
+ for(Forest result : results())
+ for(Node child : ((Forest.Ref<?>)result).parents) {
+ if (only != null && child!=only) continue;
+ holder[pos] = result;
+ if (pos==0) child.finish(r, r.rewrite(phase().getLocation()), target);
+ else child.reduce(r, pos-1, target, null);
+ }
+
+ holder[pos] = old;
+ }
- /** FIXME */
- 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, this.phase, r.zero());
+ public void finish(Position r, Forest result, Phase<Tok> target) {
+ Parser.Table<Tok>.State<Tok> 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);
+ this.merge(parent, pending);
+ Phase start = parent==null ? null : parent.phase();
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
+ 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++;
+
+ // GraphViz //////////////////////////////////////////////////////////////////////////////
+
+ public GraphViz.Node toGraphViz(GraphViz gv) {
+ if (gv.hasNode(this)) return gv.createNode(this);
+ GraphViz.Node n = gv.createNode(this);
+ n.label = ""+state.toStringx();
+ n.shape = "rectangle";
+ n.fill = "green";
+ for(Forest result : results()) n.edge(result, "");
+ for(Node parent : parents()) n.edge(parent, "");
+ ((GraphViz.Group)phase().toGraphViz(gv)).add(n);
+ return n;
+ }
+ public boolean isTransparent() { return false; }
+ public boolean isHidden() { return false; }
+
}
+ 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);
- }
+ // 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;
+ }
+ public boolean isTransparent() { return false; }
+ public boolean isHidden() { return false; }
- /** 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;
}
projects / sbp.git / blobdiff