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     protected final Table<Token> pt;
  13
  14     /** create a parser to parse the grammar with start symbol <tt>u</tt> */
  15     public Parser(Union u, Topology<Token> top)  { this.pt = new Table<Token>(u, top); }
  16     Parser(Table<Token> pt)               { this.pt = pt; }
  17
  18     /** implement this method to create the output forest corresponding to a lone shifted input token */
  19     public abstract Forest<NodeType> shiftToken(Token t, Input.Location newloc);
  20
  21     boolean helpgc = true;
  22
  23     public String toString() { return pt.toString(); }
  24
  25     /** parse <tt>input</tt>, and return the shared packed parse forest (or throw an exception) */
  26     public Forest<NodeType> parse(Input<Token> input) throws IOException, ParseFailed {
  27         GSS gss = new GSS(input);
  28         Input.Location loc = input.getLocation();
  29         Token tok = input.next();
  30         GSS.Phase current = gss.new Phase<Token>(null, null, tok, loc, input.getLocation(), null);
  31         current.newNode(new Result(Forest.create(loc.createRegion(loc), null, null, false), null, null), pt.start, true);
  32         int count = 1;
  33         for(int idx=0;;idx++) {
  34             Input.Location oldloc = loc;
  35             current.reduce();
  36             Forest forest = current.token==null ? null : shiftToken((Token)current.token, loc);
  37             loc = input.getLocation();
  38             Token nextToken = input.next();
  39             GSS.Phase next = gss.new Phase<Token>(current, current, nextToken, loc, input.getLocation(), forest);
  40
  41             /*
  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 : current)
  46                 ((Node)n).toGraphViz(gv);
  47             gv.dump(p);
  48             p.flush();
  49             p.close();
  50             */
  51
  52             count = next.size();
  53             if (current.isDone()) return (Forest<NodeType>)gss.finalResult;
  54             current = next;
  55         }
  56     }
  57
  58     // Table //////////////////////////////////////////////////////////////////////////////
  59
  60     /** an SLR(1) parse table which may contain conflicts */
  61     static class Table<Token> extends Walk.Cache {
  62
  63         public String toString() {
  64             StringBuffer sb = new StringBuffer();
  65             sb.append("parse table");
  66             for(State<Token> state : all_states.values()) {
  67                 sb.append("  " + state + "\n");
  68                 for(Topology<Token> t : state.shifts) {
  69                     sb.append("      shift  \""+
  70                               new edu.berkeley.sbp.chr.CharTopology((IntegerTopology<Character>)t)+"\" => ");
  71                     for(State st : state.shifts.getAll(t))
  72                         sb.append(st.idx+"  ");
  73                     sb.append("\n");
  74                 }
  75                 for(Topology<Token> t : state.reductions)
  76                     sb.append("      reduce \""+
  77                               new edu.berkeley.sbp.chr.CharTopology((IntegerTopology<Character>)t)+"\" => " +
  78                               state.reductions.getAll(t) + "\n");
  79                 for(Sequence s : state.gotoSetNonTerminals.keySet())
  80                     sb.append("      goto   "+state.gotoSetNonTerminals.get(s)+" from " + s + "\n");
  81             }
  82             return sb.toString();
  83         }
  84
  85         public final Walk.Cache cache = this;
  86
  87         private void walk(Element e, HashSet<SequenceOrElement> hs) {
  88             if (e==null) return;
  89             if (hs.contains(e)) return;
  90             hs.add(e);
  91             if (e instanceof Atom) return;
  92             for(Sequence s : (Union)e)
  93                 walk(s, hs);
  94         }
  95         private void walk(Sequence s, HashSet<SequenceOrElement> hs) {
  96             hs.add(s);
  97             for(Position p = s.firstp(); p != null; p = p.next())
  98                 walk(p.element(), hs);
  99             for(Sequence ss : s.needs()) walk(ss, hs);
 100             for(Sequence ss : s.hates()) walk(ss, hs);
 101         }
 102
 103         /** the start state */
 104         public  final State<Token>   start;
 105
 106         /** the state from which no reductions can be done */
 107         private final State<Token>   dead_state;
 108
 109         /** used to generate unique values for State.idx */
 110         private int master_state_idx = 0;
 111         HashMap<HashSet<Position>,State<Token>>   all_states    = new HashMap<HashSet<Position>,State<Token>>();
 112         HashSet<SequenceOrElement>                all_elements  = new HashSet<SequenceOrElement>();
 113
 114         /** construct a parse table for the given grammar */
 115         public Table(Topology top) { this("s", top); }
 116         public Table(String startSymbol, Topology top) { this(new Union(startSymbol), top); }
 117         public Table(Union ux, Topology top) {
 118             Union start0 = new Union("0");
 119             start0.add(new Sequence.Singleton(ux));
 120
 121             for(Sequence s : start0) cache.eof.put(s, true);
 122             cache.eof.put(start0, true);
 123
 124             // construct the set of states
 125             walk(start0, all_elements);
 126             for(SequenceOrElement e : all_elements)
 127                 cache.ys.addAll(e, new Walk.YieldSet(e, cache).walk());
 128             for(SequenceOrElement e : all_elements)
 129                 cache.ys2.addAll(e, new Walk.YieldSet2(e, cache).walk());
 130             HashSet<Position> hp = new HashSet<Position>();
 131             reachable(start0, hp);
 132
 133             this.dead_state = new State<Token>(new HashSet<Position>());
 134             this.start = new State<Token>(hp);
 135
 136             // for each state, fill in the corresponding "row" of the parse table
 137             for(State<Token> state : all_states.values())
 138                 for(Position p : state.hs) {
 139
 140                     // the Grammar's designated "last position" is the only accepting state
 141                     if (start0.contains(p.owner()) && p.next()==null)
 142                         state.accept = true;
 143
 144                     if (isRightNullable(p)) {
 145                         Walk.Follow wf = new Walk.Follow(top.empty(), p.owner(), all_elements, cache);
 146                         Topology follow = wf.walk(p.owner());
 147                         for(Position p2 = p; p2 != null && p2.element() != null; p2 = p2.next()) {
 148                             Atom set = new Walk.EpsilonFollowSet(new edu.berkeley.sbp.chr.CharAtom(top.empty().complement()),
 149                                                                  new edu.berkeley.sbp.chr.CharAtom(top.empty()),
 150                                                                  cache).walk(p2.element());
 151                             follow = follow.intersect(new Walk.Follow(top.empty(), p2.element(), all_elements, cache).walk(p2.element()));
 152                             if (set != null) follow = follow.intersect(set.getTokenTopology());
 153                         }
 154                         state.reductions.put(follow, p);
 155                         if (wf.includesEof()) state.eofReductions.add(p);
 156                     }
 157
 158                     // if the element following this position is an atom, copy the corresponding
 159                     // set of rows out of the "master" goto table and into this state's shift table
 160                     if (p.element() != null && p.element() instanceof Atom)
 161                         state.shifts.addAll(state.gotoSetTerminals.subset(((Atom)p.element()).getTokenTopology()));
 162                 }
 163             if (top instanceof IntegerTopology)
 164                 for(State<Token> state : all_states.values()) {
 165                     state.oreductions = state.reductions.optimize(((IntegerTopology)top).functor());
 166                     state.oshifts = state.shifts.optimize(((IntegerTopology)top).functor());
 167                 }
 168         }
 169
 170         private boolean isRightNullable(Position p) {
 171             if (p.isLast()) return true;
 172             if (!possiblyEpsilon(p.element())) return false;
 173             return isRightNullable(p.next());
 174         }
 175
 176         /** a single state in the LR table and the transitions possible from it */
 177
 178         class State<Token> implements IntegerMappable, Iterable<Position> {
 179
 180             public  final     int               idx    = master_state_idx++;
 181             private final     HashSet<Position> hs;
 182             public HashSet<State<Token>> also = new HashSet<State<Token>>();
 183
 184             public transient HashMap<Sequence,State<Token>>         gotoSetNonTerminals = new HashMap<Sequence,State<Token>>();
 185             private transient TopologicalBag<Token,State<Token>>     gotoSetTerminals    = new TopologicalBag<Token,State<Token>>();
 186
 187             private           TopologicalBag<Token,Position> reductions          = new TopologicalBag<Token,Position>();
 188             private           HashSet<Position>              eofReductions       = new HashSet<Position>();
 189             private           TopologicalBag<Token,State<Token>>     shifts              = new TopologicalBag<Token,State<Token>>();
 190             private           boolean                         accept              = false;
 191
 192             private VisitableMap<Token,State<Token>> oshifts = null;
 193             private VisitableMap<Token,Position> oreductions = null;
 194
 195             // Interface Methods //////////////////////////////////////////////////////////////////////////////
 196
 197             boolean             isAccepting()           { return accept; }
 198             public Iterator<Position>  iterator()       { return hs.iterator(); }
 199
 200             boolean             canShift(Token t)         { return oshifts!=null && oshifts.contains(t); }
 201             <B,C> void          invokeShifts(Token t, Invokable<State<Token>,B,C> irbc, B b, C c) {
 202                 oshifts.invoke(t, irbc, b, c);
 203             }
 204
 205             boolean             canReduce(Token t)        { return oreductions != null && (t==null ? eofReductions.size()>0 : oreductions.contains(t)); }
 206             <B,C> void          invokeReductions(Token t, Invokable<Position,B,C> irbc, B b, C c) {
 207                 if (t==null) for(Position r : eofReductions) irbc.invoke(r, b, c);
 208                 else         oreductions.invoke(t, irbc, b, c);
 209             }
 210
 211             // Constructor //////////////////////////////////////////////////////////////////////////////
 212
 213             /**
 214              *  create a new state consisting of all the <tt>Position</tt>s in <tt>hs</tt>
 215              *  @param hs           the set of <tt>Position</tt>s comprising this <tt>State</tt>
 216              *  @param all_elements the set of all elements (Atom instances need not be included)
 217              *
 218              *   In principle these two steps could be merged, but they
 219              *   are written separately to highlight these two facts:
 220              * <ul>
 221              * <li> Non-atom elements either match all-or-nothing, and do not overlap
 222              *      with each other (at least not in the sense of which element corresponds
 223              *      to the last reduction performed).  Therefore, in order to make sure we
 224              *      wind up with the smallest number of states and shifts, we wait until
 225              *      we've figured out all the token-to-position multimappings before creating
 226              *      any new states
 227              *
 228              * <li> In order to be able to run the state-construction algorithm in a single
 229              *      shot (rather than repeating until no new items appear in any state set),
 230              *      we need to use the "yields" semantics rather than the "produces" semantics
 231              *      for non-Atom Elements.
 232              *  </ul>
 233              */
 234             public State(HashSet<Position> hs) { this(hs, false); }
 235             public boolean special;
 236             public State(HashSet<Position> hs, boolean special) {
 237                 this.hs = hs;
 238                 this.special = special;
 239
 240                 // register ourselves in the all_states hash so that no
 241                 // two states are ever created with an identical position set
 242                 ((HashMap)all_states).put(hs, this);
 243
 244                 for(Position p : hs) {
 245                     if (!p.isFirst()) continue;
 246                     for(Sequence s : p.owner().needs()) {
 247                         if (hs.contains(s.firstp())) continue;
 248                         HashSet<Position> h2 = new HashSet<Position>();
 249                         reachable(s.firstp(), h2);
 250                         also.add((State<Token>)(all_states.get(h2) == null ? (State)new State<Token>(h2,true) : (State)all_states.get(h2)));
 251                     }
 252                     for(Sequence s : p.owner().hates()) {
 253                         if (hs.contains(s.firstp())) continue;
 254                         HashSet<Position> h2 = new HashSet<Position>();
 255                         reachable(s, h2);
 256                         also.add((State<Token>)(all_states.get(h2) == null ? (State)new State<Token>(h2,true) : (State)all_states.get(h2)));
 257                     }
 258                 }
 259
 260                 // Step 1a: examine all Position's in this state and compute the mappings from
 261                 //          sets of follow tokens (tokens which could follow this position) to sets
 262                 //          of _new_ positions (positions after shifting).  These mappings are
 263                 //          collectively known as the _closure_
 264
 265                 TopologicalBag<Token,Position> bag0 = new TopologicalBag<Token,Position>();
 266                 for(Position position : hs) {
 267                     if (position.isLast() || !(position.element() instanceof Atom)) continue;
 268                     Atom a = (Atom)position.element();
 269                     HashSet<Position> hp = new HashSet<Position>();
 270                     reachable(position.next(), hp);
 271                     bag0.addAll(a.getTokenTopology(), hp);
 272                 }
 273
 274                 // Step 1b: for each _minimal, contiguous_ set of characters having an identical next-position
 275                 //          set, add that character set to the goto table (with the State corresponding to the
 276                 //          computed next-position set).
 277
 278                 for(Topology<Token> r : bag0) {
 279                     HashSet<Position> h = new HashSet<Position>();
 280                     for(Position p : bag0.getAll(r)) h.add(p);
 281                     ((TopologicalBag)gotoSetTerminals).put(r, all_states.get(h) == null
 282                                                            ? new State<Token>(h) : all_states.get(h));
 283                 }
 284
 285                 // Step 2: for every non-Atom element (ie every Element which has a corresponding reduction),
 286                 //         compute the closure over every position in this set which is followed by a symbol
 287                 //         which could yield the Element in question.
 288                 //
 289                 //         "yields" [in one or more step] is used instead of "produces" [in exactly one step]
 290                 //         to avoid having to iteratively construct our set of States as shown in most
 291                 //         expositions of the algorithm (ie "keep doing XYZ until things stop changing").
 292
 293                 HashMapBag<SequenceOrElement,Position> move = new HashMapBag<SequenceOrElement,Position>();
 294                 for(Position p : hs) {
 295                     Element e = p.element();
 296                     if (e==null) continue;
 297                     for(SequenceOrElement y : cache.ys2.getAll(e)) {
 298                         //System.out.println(e + " yields " + y);
 299                         HashSet<Position> hp = new HashSet<Position>();
 300                         reachable(p.next(), hp);
 301                         move.addAll(y, hp);
 302                     }
 303                 }
 304                 OUTER: for(SequenceOrElement y : move) {
 305                     HashSet<Position> h = move.getAll(y);
 306                     State<Token> s = all_states.get(h) == null ? (State)new State<Token>(h) : (State)all_states.get(h);
 307                     // if a reduction is "lame", it should wind up in the dead_state after reducing
 308                     if (y instanceof Sequence) {
 309                         for(Position p : hs) {
 310                             if (p.element() != null && (p.element() instanceof Union)) {
 311                                 Union u = (Union)p.element();
 312                                 for(Sequence seq : u)
 313                                     if (seq.needs.contains((Sequence)y) || seq.hates.contains((Sequence)y)) {
 314                                         // FIXME: what if there are two "routes" to get to the sequence?
 315                                         ((HashMap)gotoSetNonTerminals).put((Sequence)y, dead_state);
 316                                         continue OUTER;
 317                                     }
 318                             }
 319                         }
 320                         gotoSetNonTerminals.put((Sequence)y, s);
 321                     }
 322                 }
 323             }
 324
 325             public String toStringx() {
 326                 StringBuffer st = new StringBuffer();
 327                 for(Position p : this) {
 328                     if (st.length() > 0) st.append("\n");
 329                     st.append(p);
 330                 }
 331                 return st.toString();
 332             }
 333             public String toString() {
 334                 StringBuffer ret = new StringBuffer();
 335                 ret.append("state["+idx+"]: ");
 336                 for(Position p : this) ret.append("{"+p+"}  ");
 337                 return ret.toString();
 338             }
 339
 340             public Walk.Cache cache() { return cache; }
 341             public int toInt() { return idx; }
 342         }
 343     }
 344
 345     // Helpers //////////////////////////////////////////////////////////////////////////////
 346
 347     private static void reachable(Sequence s, HashSet<Position> h) {
 348         reachable(s.firstp(), h);
 349         //for(Sequence ss : s.needs()) reachable(ss, h);
 350         //for(Sequence ss : s.hates()) reachable(ss, h);
 351     }
 352     private static void reachable(Element e, HashSet<Position> h) {
 353         if (e instanceof Atom) return;
 354         for(Sequence s : ((Union)e))
 355             reachable(s, h);
 356     }
 357     private static void reachable(Position p, HashSet<Position> h) {
 358         if (h.contains(p)) return;
 359         h.add(p);
 360         if (p.element() != null) reachable(p.element(), h);
 361     }
 362
 363 }