src/edu/berkeley/sbp/Parser.java

   1 // Copyright 2006 all rights reserved; see LICENSE file for BSD-style license
   2
   3 package edu.berkeley.sbp;
   4 import edu.berkeley.sbp.*;
   5 import edu.berkeley.sbp.util.*;
   6 import edu.berkeley.sbp.Sequence.Position;
   7 import java.io.*;
   8 import java.util.*;
   9
  10 /** a parser which translates an Input&lt;Token&gt; into a Forest&lt;NodeType&gt; */
  11 public abstract class Parser<Token, NodeType> {
  12
  13     protected final Table<Token> pt;
  14
  15     /** create a parser to parse the grammar with start symbol <tt>u</tt> */
  16     public Parser(Union u, Topology<Token> top)  { this.pt = new Table<Token>(u, top); }
  17     Parser(Table<Token> pt)               { this.pt = pt; }
  18
  19     /** implement this method to create the output forest corresponding to a lone shifted input token */
  20     public abstract Forest<NodeType> shiftToken(Token t, Input.Location newloc);
  21
  22     boolean helpgc = true;
  23
  24     public String toString() { return pt.toString(); }
  25
  26     /** parse <tt>input</tt>, and return the shared packed parse forest (or throw an exception) */
  27     public Forest<NodeType> parse(Input<Token> input) throws IOException, ParseFailed {
  28         GSS gss = new GSS(input);
  29         Input.Location loc = input.getLocation();
  30         Token tok = input.next();
  31         GSS.Phase current = gss.new Phase<Token>(null, null, tok, loc, input.getLocation(), null);
  32         current.newNode(new Result(Forest.create(loc.createRegion(loc), null, null, false), null, null), pt.start, true);
  33         int count = 1;
  34         for(int idx=0;;idx++) {
  35             Input.Location oldloc = loc;
  36             current.reduce();
  37             Forest forest = current.token==null ? null : shiftToken((Token)current.token, loc);
  38             loc = input.getLocation();
  39             Token nextToken = input.next();
  40             GSS.Phase next = gss.new Phase<Token>(current, current, nextToken, loc, input.getLocation(), forest);
  41             if (!helpgc) {
  42                 FileOutputStream fos = new FileOutputStream("out-"+idx+".dot");
  43                 PrintWriter p = new PrintWriter(new OutputStreamWriter(fos));
  44                 GraphViz gv = new GraphViz();
  45                 for(Object n : next)
  46                     ((Node)n).toGraphViz(gv);
  47                 gv.dump(p);
  48                 p.flush();
  49                 p.close();
  50             }
  51             count = next.size();
  52             if (current.isDone()) return (Forest<NodeType>)gss.finalResult;
  53             current = next;
  54         }
  55     }
  56
  57     // Table //////////////////////////////////////////////////////////////////////////////
  58
  59     /** an SLR(1) parse table which may contain conflicts */
  60     static class Table<Token> extends Walk.Cache {
  61
  62         public String toString() {
  63             StringBuffer sb = new StringBuffer();
  64             sb.append("parse table");
  65             for(State<Token> state : all_states.values()) {
  66                 sb.append("  " + state + "\n");
  67                 for(Topology<Token> t : state.shifts) {
  68                     sb.append("      shift  \""+
  69                               new edu.berkeley.sbp.chr.CharTopology((IntegerTopology<Character>)t)+"\" => ");
  70                     for(State st : state.shifts.getAll(t))
  71                         sb.append(st.idx+"  ");
  72                     sb.append("\n");
  73                 }
  74                 for(Topology<Token> t : state.reductions)
  75                     sb.append("      reduce \""+
  76                               new edu.berkeley.sbp.chr.CharTopology((IntegerTopology<Character>)t)+"\" => " +
  77                               state.reductions.getAll(t) + "\n");
  78             }
  79             return sb.toString();
  80         }
  81
  82         public final Walk.Cache cache = this;
  83
  84         private void walk(Element e, HashSet<SequenceOrElement> hs) {
  85             if (e==null) return;
  86             if (hs.contains(e)) return;
  87             hs.add(e);
  88             if (e instanceof Atom) return;
  89             for(Sequence s : (Union)e)
  90                 walk(s, hs);
  91         }
  92         private void walk(Sequence s, HashSet<SequenceOrElement> hs) {
  93             hs.add(s);
  94             for(Position p = s.firstp(); p != null; p = p.next())
  95                 walk(p.element(), hs);
  96             for(Sequence ss : s.needs()) walk(ss, hs);
  97             for(Sequence ss : s.hates()) walk(ss, hs);
  98         }
  99
 100         /** the start state */
 101         public  final State<Token>   start;
 102
 103         /** the state from which no reductions can be done */
 104         private final State<Token>   dead_state;
 105
 106         /** used to generate unique values for State.idx */
 107         private int master_state_idx = 0;
 108         HashMap<HashSet<Position>,State<Token>>   all_states    = new HashMap<HashSet<Position>,State<Token>>();
 109         HashSet<SequenceOrElement>                all_elements  = new HashSet<SequenceOrElement>();
 110
 111         /** construct a parse table for the given grammar */
 112         public Table(Topology top) { this("s", top); }
 113         public Table(String startSymbol, Topology top) { this(new Union(startSymbol), top); }
 114         public Table(Union ux, Topology top) {
 115             Union start0 = new Union("0");
 116             start0.add(new Sequence.Singleton(ux));
 117
 118             for(Sequence s : start0) cache.eof.put(s, true);
 119             cache.eof.put(start0, true);
 120
 121             // construct the set of states
 122             walk(start0, all_elements);
 123             for(SequenceOrElement e : all_elements)
 124                 cache.ys.addAll(e, new Walk.YieldSet(e, cache).walk());
 125             HashSet<Position> hp = new HashSet<Position>();
 126             reachable(start0, hp);
 127
 128             this.dead_state = new State<Token>(new HashSet<Position>());
 129             this.start = new State<Token>(hp);
 130
 131             // for each state, fill in the corresponding "row" of the parse table
 132             for(State<Token> state : all_states.values())
 133                 for(Position p : state.hs) {
 134
 135                     // the Grammar's designated "last position" is the only accepting state
 136                     if (start0.contains(p.owner()) && p.next()==null)
 137                         state.accept = true;
 138
 139                     if (isRightNullable(p)) {
 140                         Walk.Follow wf = new Walk.Follow(top.empty(), p.owner(), all_elements, cache);
 141                         Topology follow = wf.walk(p.owner());
 142                         for(Position p2 = p; p2 != null && p2.element() != null; p2 = p2.next()) {
 143                             Atom set = new Walk.EpsilonFollowSet(new edu.berkeley.sbp.chr.CharAtom(top.empty().complement()),
 144                                                                  new edu.berkeley.sbp.chr.CharAtom(top.empty()),
 145                                                                  cache).walk(p2.element());
 146                             follow = follow.intersect(new Walk.Follow(top.empty(), p2.element(), all_elements, cache).walk(p2.element()));
 147                             if (set != null) follow = follow.intersect(set.getTokenTopology());
 148                         }
 149                         state.reductions.put(follow, p);
 150                         if (wf.includesEof()) state.eofReductions.add(p);
 151                     }
 152
 153                     // if the element following this position is an atom, copy the corresponding
 154                     // set of rows out of the "master" goto table and into this state's shift table
 155                     if (p.element() != null && p.element() instanceof Atom)
 156                         state.shifts.addAll(state.gotoSetTerminals.subset(((Atom)p.element()).getTokenTopology()));
 157                 }
 158             if (top instanceof IntegerTopology)
 159                 for(State<Token> state : all_states.values()) {
 160                     state.oreductions = state.reductions.optimize(((IntegerTopology)top).functor());
 161                     state.oshifts = state.shifts.optimize(((IntegerTopology)top).functor());
 162                 }
 163         }
 164
 165         private boolean isRightNullable(Position p) {
 166             if (p.isLast()) return true;
 167             if (!possiblyEpsilon(p.element())) return false;
 168             return isRightNullable(p.next());
 169         }
 170
 171         /** a single state in the LR table and the transitions possible from it */
 172
 173         class State<Token> implements IntegerMappable, Iterable<Position> {
 174
 175             public  final     int               idx    = master_state_idx++;
 176             private final     HashSet<Position> hs;
 177             public HashSet<State<Token>> also = new HashSet<State<Token>>();
 178
 179             public transient HashMap<Sequence,State<Token>>         gotoSetNonTerminals = new HashMap<Sequence,State<Token>>();
 180             private transient TopologicalBag<Token,State<Token>>     gotoSetTerminals    = new TopologicalBag<Token,State<Token>>();
 181
 182             private           TopologicalBag<Token,Position> reductions          = new TopologicalBag<Token,Position>();
 183             private           HashSet<Position>              eofReductions       = new HashSet<Position>();
 184             private           TopologicalBag<Token,State<Token>>     shifts              = new TopologicalBag<Token,State<Token>>();
 185             private           boolean                         accept              = false;
 186
 187             private VisitableMap<Token,State<Token>> oshifts = null;
 188             private VisitableMap<Token,Position> oreductions = null;
 189
 190             // Interface Methods //////////////////////////////////////////////////////////////////////////////
 191
 192             boolean             isAccepting()           { return accept; }
 193             public Iterator<Position>  iterator()       { return hs.iterator(); }
 194
 195             boolean             canShift(Token t)         { return oshifts!=null && oshifts.contains(t); }
 196             <B,C> void          invokeShifts(Token t, Invokable<State<Token>,B,C> irbc, B b, C c) {
 197                 oshifts.invoke(t, irbc, b, c);
 198             }
 199
 200             boolean             canReduce(Token t)        { return oreductions != null && (t==null ? eofReductions.size()>0 : oreductions.contains(t)); }
 201             <B,C> void          invokeReductions(Token t, Invokable<Position,B,C> irbc, B b, C c) {
 202                 if (t==null) for(Position r : eofReductions) irbc.invoke(r, b, c);
 203                 else         oreductions.invoke(t, irbc, b, c);
 204             }
 205
 206             // Constructor //////////////////////////////////////////////////////////////////////////////
 207
 208             /**
 209              *  create a new state consisting of all the <tt>Position</tt>s in <tt>hs</tt>
 210              *  @param hs           the set of <tt>Position</tt>s comprising this <tt>State</tt>
 211              *  @param all_elements the set of all elements (Atom instances need not be included)
 212              *
 213              *   In principle these two steps could be merged, but they
 214              *   are written separately to highlight these two facts:
 215              * <ul>
 216              * <li> Non-atom elements either match all-or-nothing, and do not overlap
 217              *      with each other (at least not in the sense of which element corresponds
 218              *      to the last reduction performed).  Therefore, in order to make sure we
 219              *      wind up with the smallest number of states and shifts, we wait until
 220              *      we've figured out all the token-to-position multimappings before creating
 221              *      any new states
 222              *
 223              * <li> In order to be able to run the state-construction algorithm in a single
 224              *      shot (rather than repeating until no new items appear in any state set),
 225              *      we need to use the "yields" semantics rather than the "produces" semantics
 226              *      for non-Atom Elements.
 227              *  </ul>
 228              */
 229             public State(HashSet<Position> hs) { this(hs, false); }
 230             public boolean special;
 231             public State(HashSet<Position> hs, boolean special) {
 232                 this.hs = hs;
 233                 this.special = special;
 234
 235                 // register ourselves in the all_states hash so that no
 236                 // two states are ever created with an identical position set
 237                 ((HashMap)all_states).put(hs, this);
 238
 239                 for(Position p : hs) {
 240                     if (!p.isFirst()) continue;
 241                     for(Sequence s : p.owner().needs()) {
 242                         if (hs.contains(s.firstp())) continue;
 243                         HashSet<Position> h2 = new HashSet<Position>();
 244                         reachable(s.firstp(), h2);
 245                         also.add((State<Token>)(all_states.get(h2) == null ? (State)new State<Token>(h2,true) : (State)all_states.get(h2)));
 246                     }
 247                     for(Sequence s : p.owner().hates()) {
 248                         if (hs.contains(s.firstp())) continue;
 249                         HashSet<Position> h2 = new HashSet<Position>();
 250                         reachable(s, h2);
 251                         also.add((State<Token>)(all_states.get(h2) == null ? (State)new State<Token>(h2,true) : (State)all_states.get(h2)));
 252                     }
 253                 }
 254
 255                 // Step 1a: examine all Position's in this state and compute the mappings from
 256                 //          sets of follow tokens (tokens which could follow this position) to sets
 257                 //          of _new_ positions (positions after shifting).  These mappings are
 258                 //          collectively known as the _closure_
 259
 260                 TopologicalBag<Token,Position> bag0 = new TopologicalBag<Token,Position>();
 261                 for(Position position : hs) {
 262                     if (position.isLast() || !(position.element() instanceof Atom)) continue;
 263                     Atom a = (Atom)position.element();
 264                     HashSet<Position> hp = new HashSet<Position>();
 265                     reachable(position.next(), hp);
 266                     bag0.addAll(a.getTokenTopology(), hp);
 267                 }
 268
 269                 // Step 1b: for each _minimal, contiguous_ set of characters having an identical next-position
 270                 //          set, add that character set to the goto table (with the State corresponding to the
 271                 //          computed next-position set).
 272
 273                 for(Topology<Token> r : bag0) {
 274                     HashSet<Position> h = new HashSet<Position>();
 275                     for(Position p : bag0.getAll(r)) h.add(p);
 276                     ((TopologicalBag)gotoSetTerminals).put(r, all_states.get(h) == null
 277                                                            ? new State<Token>(h) : all_states.get(h));
 278                 }
 279
 280                 // Step 2: for every non-Atom element (ie every Element which has a corresponding reduction),
 281                 //         compute the closure over every position in this set which is followed by a symbol
 282                 //         which could yield the Element in question.
 283                 //
 284                 //         "yields" [in one or more step] is used instead of "produces" [in exactly one step]
 285                 //         to avoid having to iteratively construct our set of States as shown in most
 286                 //         expositions of the algorithm (ie "keep doing XYZ until things stop changing").
 287
 288                 HashMapBag<SequenceOrElement,Position> move = new HashMapBag<SequenceOrElement,Position>();
 289                 for(Position p : hs) {
 290                     Element e = p.element();
 291                     if (e==null) continue;
 292                     for(SequenceOrElement y : cache.ys.getAll(e)) {
 293                         HashSet<Position> hp = new HashSet<Position>();
 294                         reachable(p.next(), hp);
 295                         move.addAll(y, hp);
 296                     }
 297                 }
 298                 OUTER: for(SequenceOrElement y : move) {
 299                     HashSet<Position> h = move.getAll(y);
 300                     State<Token> s = all_states.get(h) == null ? (State)new State<Token>(h) : (State)all_states.get(h);
 301                     // if a reduction is "lame", it should wind up in the dead_state after reducing
 302                     if (y instanceof Sequence) {
 303                         for(Position p : hs) {
 304                             if (p.element() != null && (p.element() instanceof Union)) {
 305                                 Union u = (Union)p.element();
 306                                 for(Sequence seq : u)
 307                                     if (seq.needs.contains((Sequence)y) || seq.hates.contains((Sequence)y)) {
 308                                         // FIXME: what if there are two "routes" to get to the sequence?
 309                                         ((HashMap)gotoSetNonTerminals).put((Sequence)y, dead_state);
 310                                         continue OUTER;
 311                                     }
 312                             }
 313                         }
 314                         gotoSetNonTerminals.put((Sequence)y, s);
 315                     }
 316                 }
 317             }
 318
 319             public String toStringx() {
 320                 StringBuffer st = new StringBuffer();
 321                 for(Position p : this) {
 322                     if (st.length() > 0) st.append("\n");
 323                     st.append(p);
 324                 }
 325                 return st.toString();
 326             }
 327             public String toString() {
 328                 StringBuffer ret = new StringBuffer();
 329                 ret.append("state["+idx+"]: ");
 330                 for(Position p : this) ret.append("{"+p+"}  ");
 331                 return ret.toString();
 332             }
 333
 334             public Walk.Cache cache() { return cache; }
 335             public int toInt() { return idx; }
 336         }
 337     }
 338
 339     // Helpers //////////////////////////////////////////////////////////////////////////////
 340
 341     private static void reachable(Sequence s, HashSet<Position> h) {
 342         reachable(s.firstp(), h);
 343         //for(Sequence ss : s.needs()) reachable(ss, h);
 344         //for(Sequence ss : s.hates()) reachable(ss, h);
 345     }
 346     private static void reachable(Element e, HashSet<Position> h) {
 347         if (e instanceof Atom) return;
 348         for(Sequence s : ((Union)e))
 349             reachable(s, h);
 350     }
 351     private static void reachable(Position p, HashSet<Position> h) {
 352         if (h.contains(p)) return;
 353         h.add(p);
 354         if (p.element() != null) reachable(p.element(), h);
 355     }
 356
 357 }