+// 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 java.io.*;
import java.util.*;
-/** a parser which translates streams of Tokens of type T into a Forest<R> */
-public abstract class Parser<T extends Token, R> {
-
- private final Table pt;
+/** a parser which translates an Input<Token> into a Forest<NodeType> */
+public abstract class Parser<Token, NodeType> {
+ protected final Table<Token> pt;
/** create a parser to parse the grammar with start symbol <tt>u</tt> */
- protected Parser(Union u) { this.pt = new Table(u, top()); }
- protected Parser(Table pt) { this.pt = pt; }
+ public Parser(Union u, Topology<Token> top) { this.pt = new Table<Token>(u, top); }
+ Parser(Table<Token> pt) { this.pt = pt; }
/** implement this method to create the output forest corresponding to a lone shifted input token */
- public abstract Forest<R> shiftedToken(T t, Token.Location loc);
+ public abstract Forest<NodeType> shiftToken(Token t, Input.Location newloc);
+
+ boolean helpgc = true;
- /** this method must return an empty topology of the input token type */
- public abstract Topology<T> top();
+ public String toString() { return pt.toString(); }
- /** parse <tt>input</tt>, using the table <tt>pt</tt> to drive the parser */
- public Forest<R> parse(Token.Stream<T> input) throws IOException, ParseFailed {
- GSS gss = new GSS();
- Token.Location loc = input.getLocation();
- GSS.Phase current = gss.new Phase(null, this, null, input.next(1, 0, 0), loc, null);
- current.newNode(null, Forest.leaf(null, null), pt.start, true);
+ /** parse <tt>input</tt>, and return the shared packed parse forest (or throw an exception) */
+ public Forest<NodeType> parse(Input<Token> input) throws IOException, ParseFailed {
+ GSS gss = new GSS(input);
+ Input.Location loc = input.getLocation();
+ Token tok = input.next();
+ GSS.Phase current = gss.new Phase<Token>(null, null, tok, loc, input.getLocation(), null);
+ current.newNode(new Result(Forest.create(loc.createRegion(loc), null, null, false), null, null), pt.start, true);
int count = 1;
- for(;;) {
- loc = input.getLocation();
+ for(int idx=0;;idx++) {
+ Input.Location oldloc = loc;
current.reduce();
- Forest forest = current.token==null ? null : shiftedToken((T)current.token, loc);
- GSS.Phase next = gss.new Phase(current, this, current, input.next(count, gss.resets, gss.waits), loc, forest);
+ Forest forest = current.token==null ? null : shiftToken((Token)current.token, loc);
+ loc = input.getLocation();
+ Token nextToken = input.next();
+ GSS.Phase next = gss.new Phase<Token>(current, current, nextToken, loc, input.getLocation(), forest);
+
+ /*
+ FileOutputStream fos = new FileOutputStream("out-"+idx+".dot");
+ PrintWriter p = new PrintWriter(new OutputStreamWriter(fos));
+ GraphViz gv = new GraphViz();
+ for(Object n : current)
+ ((Node)n).toGraphViz(gv);
+ gv.dump(p);
+ p.flush();
+ p.close();
+ */
+
count = next.size();
- if (current.isDone()) return (Forest<R>)gss.finalResult;
+ if (current.isDone()) return (Forest<NodeType>)gss.finalResult;
current = next;
}
}
// Table //////////////////////////////////////////////////////////////////////////////
/** an SLR(1) parse table which may contain conflicts */
- static class Table extends Walk.Cache {
+ static class Table<Token> extends Walk.Cache {
+
+ public String toString() {
+ StringBuffer sb = new StringBuffer();
+ sb.append("parse table");
+ for(State<Token> state : all_states) {
+ sb.append(" " + state + "\n");
+ for(Topology<Token> t : state.shifts) {
+ sb.append(" shift \""+
+ new edu.berkeley.sbp.chr.CharTopology((IntegerTopology<Character>)t)+"\" => ");
+ for(State st : state.shifts.getAll(t))
+ sb.append(st.idx+" ");
+ sb.append("\n");
+ }
+ for(Topology<Token> t : state.reductions)
+ sb.append(" reduce \""+
+ new edu.berkeley.sbp.chr.CharTopology((IntegerTopology<Character>)t)+"\" => " +
+ state.reductions.getAll(t) + "\n");
+ for(Sequence s : state.gotoSetNonTerminals.keySet())
+ sb.append(" goto "+state.gotoSetNonTerminals.get(s)+" from " + s + "\n");
+ }
+ return sb.toString();
+ }
public final Walk.Cache cache = this;
- private void walk(Element e, HashSet<Element> hs) {
+ private void walk(Element e, HashSet<SequenceOrElement> hs) {
if (e==null) return;
if (hs.contains(e)) return;
hs.add(e);
if (e instanceof Atom) return;
- for(Sequence s : (Union)e) {
- hs.add(s);
- for(Position p = s.firstp(); p != null; p = p.next())
- walk(p.element(), hs);
- }
+ for(Sequence s : (Union)e)
+ walk(s, hs);
+ }
+ private void walk(Sequence s, HashSet<SequenceOrElement> hs) {
+ hs.add(s);
+ for(Position p = s.firstp(); p != null; p = p.next())
+ walk(p.element(), hs);
+ for(Sequence ss : s.needs()) walk(ss, hs);
+ for(Sequence ss : s.hates()) walk(ss, hs);
}
/** the start state */
- public final State start;
+ public final State<Token> start;
+
+ /** the state from which no reductions can be done */
+ private final State<Token> dead_state;
/** used to generate unique values for State.idx */
private int master_state_idx = 0;
+ HashSet<State<Token>> all_states = new HashSet<State<Token>>();
+ HashMap<HashSet<Position>,State<Token>> doomed_states = new HashMap<HashSet<Position>,State<Token>>();
+ HashMap<HashSet<Position>,State<Token>> normal_states = new HashMap<HashSet<Position>,State<Token>>();
+ HashSet<SequenceOrElement> all_elements = new HashSet<SequenceOrElement>();
/** construct a parse table for the given grammar */
public Table(Topology top) { this("s", top); }
public Table(String startSymbol, Topology top) { this(new Union(startSymbol), top); }
public Table(Union ux, Topology top) {
Union start0 = new Union("0");
- start0.add(new Sequence.Singleton(ux, null, null));
+ start0.add(new Sequence.Singleton(ux));
for(Sequence s : start0) cache.eof.put(s, true);
cache.eof.put(start0, true);
// construct the set of states
- HashMap<HashSet<Position>,State> all_states = new HashMap<HashSet<Position>,State>();
- HashSet<Element> all_elements = new HashSet<Element>();
walk(start0, all_elements);
- for(Element e : all_elements)
+ for(SequenceOrElement e : all_elements)
cache.ys.addAll(e, new Walk.YieldSet(e, cache).walk());
+ for(SequenceOrElement e : all_elements)
+ cache.ys2.addAll(e, new Walk.YieldSet2(e, cache).walk());
HashSet<Position> hp = new HashSet<Position>();
reachable(start0, hp);
- this.start = new State(hp, all_states, all_elements);
+
+ this.dead_state = new State<Token>(new HashSet<Position>(), true);
+ this.start = new State<Token>(hp);
// for each state, fill in the corresponding "row" of the parse table
- for(State state : all_states.values())
+ for(State<Token> state : all_states)
for(Position p : state.hs) {
// the Grammar's designated "last position" is the only accepting state
- if (start0.contains(p.owner()) && p.next()==null)
+ if (start0.contains(p.owner()) && p.next()==null && !state.doomed)
state.accept = true;
if (isRightNullable(p)) {
Walk.Follow wf = new Walk.Follow(top.empty(), p.owner(), all_elements, cache);
- Reduction red = new Reduction(p);
-
Topology follow = wf.walk(p.owner());
- for(Position p2 = p; p2 != null && p2.element() != null; p2 = p2.next())
+ for(Position p2 = p; p2 != null && p2.element() != null; p2 = p2.next()) {
+ Atom set = new Walk.EpsilonFollowSet(new edu.berkeley.sbp.chr.CharAtom(top.empty().complement()),
+ new edu.berkeley.sbp.chr.CharAtom(top.empty()),
+ cache).walk(p2.element());
follow = follow.intersect(new Walk.Follow(top.empty(), p2.element(), all_elements, cache).walk(p2.element()));
- state.reductions.put(follow, red);
- if (wf.includesEof()) state.eofReductions.add(red);
+ if (set != null) follow = follow.intersect(set.getTokenTopology());
+ }
+ state.reductions.put(follow, p);
+ if (wf.includesEof()) state.eofReductions.add(p);
}
// if the element following this position is an atom, copy the corresponding
// set of rows out of the "master" goto table and into this state's shift table
if (p.element() != null && p.element() instanceof Atom)
- state.shifts.addAll(state.gotoSetTerminals.subset(((Atom)p.element())));
+ state.shifts.addAll(state.gotoSetTerminals.subset(((Atom)p.element()).getTokenTopology()));
+ }
+
+ if (top instanceof IntegerTopology)
+ for(State<Token> state : all_states) {
+ state.oreductions = state.reductions.optimize(((IntegerTopology)top).functor());
+ state.oshifts = state.shifts.optimize(((IntegerTopology)top).functor());
+ }
+
+ // crude algorithm to assing an ordinal ordering to every position
+ ArrayList<Sequence.Position> al = new ArrayList<Sequence.Position>();
+ for(State s : all_states) {
+ for(Object po : s) {
+ Sequence.Position p = (Sequence.Position)po;
+ if (al.contains(p)) continue;
+ int i=0;
+ for(; i<al.size(); i++) {
+ if (p.compareTo(al.get(i), cache) > 0)
+ break;
+ }
+ al.add(i, p);
}
- for(State state : all_states.values()) {
- state.oreductions = state.reductions.optimize();
- state.oshifts = state.shifts.optimize();
}
+ for(int i=0; i<al.size(); i++)
+ al.get(i).ord = i;
}
private boolean isRightNullable(Position p) {
if (p.isLast()) return true;
- if (!p.element().possiblyEpsilon(this)) return false;
+ if (!possiblyEpsilon(p.element())) return false;
return isRightNullable(p.next());
}
/** a single state in the LR table and the transitions possible from it */
- public class State implements Comparable<Table.State>, IntegerMappable, Iterable<Position> {
+ class State<Token> implements IntegerMappable, Iterable<Position> {
public final int idx = master_state_idx++;
private final HashSet<Position> hs;
+ public HashSet<State<Token>> also = new HashSet<State<Token>>();
- public transient HashMap<Element,State> gotoSetNonTerminals = new HashMap<Element,State>();
- private transient TopologicalBag<Token,State> gotoSetTerminals = new TopologicalBag<Token,State>();
+ public transient HashMap<Sequence,State<Token>> gotoSetNonTerminals = new HashMap<Sequence,State<Token>>();
+ private transient TopologicalBag<Token,State<Token>> gotoSetTerminals = new TopologicalBag<Token,State<Token>>();
- private TopologicalBag<Token,Reduction> reductions = new TopologicalBag<Token,Reduction>();
- private HashSet<Reduction> eofReductions = new HashSet<Reduction>();
- private TopologicalBag<Token,State> shifts = new TopologicalBag<Token,State>();
+ private TopologicalBag<Token,Position> reductions = new TopologicalBag<Token,Position>();
+ private HashSet<Position> eofReductions = new HashSet<Position>();
+ private TopologicalBag<Token,State<Token>> shifts = new TopologicalBag<Token,State<Token>>();
private boolean accept = false;
- private VisitableMap<Token,State> oshifts = null;
- private VisitableMap<Token,Reduction> oreductions = null;
+ private VisitableMap<Token,State<Token>> oshifts = null;
+ private VisitableMap<Token,Position> oreductions = null;
// Interface Methods //////////////////////////////////////////////////////////////////////////////
- boolean isAccepting() { return accept; }
- public Iterator<Position> iterator() { return hs.iterator(); }
+ boolean isAccepting() { return accept; }
+ public Iterator<Position> iterator() { return hs.iterator(); }
- boolean canShift(Token t) { return oshifts.contains(t); }
- <B,C> void invokeShifts(Token t, Invokable<State,B,C> irbc, B b, C c) {
+ boolean canShift(Token t) { return oshifts!=null && oshifts.contains(t); }
+ <B,C> void invokeShifts(Token t, Invokable<State<Token>,B,C> irbc, B b, C c) {
oshifts.invoke(t, irbc, b, c);
}
- boolean canReduce(Token t) { return t==null ? eofReductions.size()>0 : oreductions.contains(t); }
- <B,C> void invokeReductions(Token t, Invokable<Reduction,B,C> irbc, B b, C c) {
- if (t==null) for(Reduction r : eofReductions) irbc.invoke(r, b, c);
+ boolean canReduce(Token t) { return oreductions != null && (t==null ? eofReductions.size()>0 : oreductions.contains(t)); }
+ <B,C> void invokeReductions(Token t, Invokable<Position,B,C> irbc, B b, C c) {
+ if (t==null) for(Position r : eofReductions) irbc.invoke(r, b, c);
else oreductions.invoke(t, irbc, b, c);
}
/**
* create a new state consisting of all the <tt>Position</tt>s in <tt>hs</tt>
* @param hs the set of <tt>Position</tt>s comprising this <tt>State</tt>
- * @param all_states the set of states already constructed (to avoid recreating states)
* @param all_elements the set of all elements (Atom instances need not be included)
*
* In principle these two steps could be merged, but they
* for non-Atom Elements.
* </ul>
*/
- public State(HashSet<Position> hs,
- HashMap<HashSet<Position>,State> all_states,
- HashSet<Element> all_elements) {
+ public State(HashSet<Position> hs) { this(hs, false); }
+ public boolean doomed;
+ public State(HashSet<Position> hs, boolean doomed) {
this.hs = hs;
+ this.doomed = doomed;
// register ourselves in the all_states hash so that no
// two states are ever created with an identical position set
- all_states.put(hs, this);
+ ((HashMap)(doomed ? doomed_states : normal_states)).put(hs, this);
+ ((HashSet)all_states).add(this);
+
+ for(Position p : hs) {
+ if (!p.isFirst()) continue;
+ for(Sequence s : p.owner().needs()) {
+ if (hs.contains(s.firstp())) continue;
+ HashSet<Position> h2 = new HashSet<Position>();
+ reachable(s.firstp(), h2);
+ also.add(mkstate(h2, true));
+ }
+ for(Sequence s : p.owner().hates()) {
+ if (hs.contains(s.firstp())) continue;
+ HashSet<Position> h2 = new HashSet<Position>();
+ reachable(s, h2);
+ also.add(mkstate(h2, true));
+ }
+ }
// Step 1a: examine all Position's in this state and compute the mappings from
// sets of follow tokens (tokens which could follow this position) to sets
Atom a = (Atom)position.element();
HashSet<Position> hp = new HashSet<Position>();
reachable(position.next(), hp);
- bag0.addAll(a, hp);
+ bag0.addAll(a.getTokenTopology(), hp);
}
// Step 1b: for each _minimal, contiguous_ set of characters having an identical next-position
for(Topology<Token> r : bag0) {
HashSet<Position> h = new HashSet<Position>();
for(Position p : bag0.getAll(r)) h.add(p);
- gotoSetTerminals.put(r, all_states.get(h) == null ? new State(h, all_states, all_elements) : all_states.get(h));
+ ((TopologicalBag)gotoSetTerminals).put(r, mkstate(h, doomed));
}
// Step 2: for every non-Atom element (ie every Element which has a corresponding reduction),
// "yields" [in one or more step] is used instead of "produces" [in exactly one step]
// to avoid having to iteratively construct our set of States as shown in most
// expositions of the algorithm (ie "keep doing XYZ until things stop changing").
- HashMapBag<Element,Position> move = new HashMapBag<Element,Position>();
+
+ HashMapBag<SequenceOrElement,Position> move = new HashMapBag<SequenceOrElement,Position>();
for(Position p : hs) {
Element e = p.element();
if (e==null) continue;
- for(Element y : cache.ys.getAll(e)) {
+ for(SequenceOrElement y : cache.ys2.getAll(e)) {
+ //System.out.println(e + " yields " + y);
HashSet<Position> hp = new HashSet<Position>();
reachable(p.next(), hp);
move.addAll(y, hp);
}
}
- for(Element y : move) {
+ OUTER: for(SequenceOrElement y : move) {
HashSet<Position> h = move.getAll(y);
- State s = all_states.get(h) == null ? new State(h, all_states, all_elements) : all_states.get(h);
- gotoSetNonTerminals.put(y, s);
+ State<Token> s = mkstate(h, doomed);
+ // if a reduction is "lame", it should wind up in the dead_state after reducing
+ if (y instanceof Sequence) {
+ for(Position p : hs) {
+ if (p.element() != null && (p.element() instanceof Union)) {
+ Union u = (Union)p.element();
+ for(Sequence seq : u)
+ if (seq.needs.contains((Sequence)y) || seq.hates.contains((Sequence)y)) {
+ // FIXME: what if there are two "routes" to get to the sequence?
+ ((HashMap)gotoSetNonTerminals).put((Sequence)y, dead_state);
+ continue OUTER;
+ }
+ }
+ }
+ gotoSetNonTerminals.put((Sequence)y, s);
+ }
}
}
+ private State<Token> mkstate(HashSet<Position> h, boolean b) {
+ if (b) return doomed_states.get(h) == null ? (State)new State<Token>(h,b) : (State)doomed_states.get(h);
+ else return normal_states.get(h) == null ? (State)new State<Token>(h,b) : (State)normal_states.get(h);
+ }
+
+ public String toStringx() {
+ StringBuffer st = new StringBuffer();
+ for(Position p : this) {
+ if (st.length() > 0) st.append("\n");
+ st.append(p);
+ }
+ return st.toString();
+ }
public String toString() {
StringBuffer ret = new StringBuffer();
ret.append("state["+idx+"]: ");
return ret.toString();
}
- public int compareTo(Table.State s) { return idx==s.idx ? 0 : idx < s.idx ? -1 : 1; }
+ public Walk.Cache cache() { return cache; }
public int toInt() { return idx; }
}
-
- /**
- * the information needed to perform a reduction; copied here to
- * avoid keeping references to <tt>Element</tt> objects in a Table
- */
- public class Reduction {
- // FIXME: cleanup; almost everything in here could go in either Sequence.Position.getRewrite() or else in GSS.Reduct
- /*private*/ final Position position;
- public int hashCode() { return position.hashCode(); }
- public boolean equals(Object o) {
- if (o==null) return false;
- if (o==this) return true;
- if (!(o instanceof Reduction)) return false;
- Reduction r = (Reduction)o;
- return r.position == position;
- }
- public Reduction(Position p) {
- this.position = p;
- }
- public String toString() { return "[reduce " + position + "]"; }
-
- private Forest zero = null;
- public Forest zero() {
- if (zero != null) return zero;
- if (position.pos > 0) throw new Error();
- return zero = position.rewrite(null);
- }
-
- }
}
- private static final Forest[] emptyForestArray = new Forest[0];
-
-
// Helpers //////////////////////////////////////////////////////////////////////////////
-
+
+ private static void reachable(Sequence s, HashSet<Position> h) {
+ reachable(s.firstp(), h);
+ //for(Sequence ss : s.needs()) reachable(ss, h);
+ //for(Sequence ss : s.hates()) reachable(ss, h);
+ }
private static void reachable(Element e, HashSet<Position> h) {
if (e instanceof Atom) return;
for(Sequence s : ((Union)e))
- reachable(s.firstp(), h);
+ reachable(s, h);
}
private static void reachable(Position p, HashSet<Position> h) {
if (h.contains(p)) return;