1 // Copyright 2006 all rights reserved; see LICENSE file for BSD-style license
3 package edu.berkeley.sbp;
4 import edu.berkeley.sbp.*;
5 import edu.berkeley.sbp.util.*;
6 import edu.berkeley.sbp.Sequence.Position;
10 /** a parser which translates an Input<Token> into a Forest<NodeType> */
11 public abstract class Parser<Token, NodeType> {
13 protected final Table<Token> pt;
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; }
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);
22 boolean helpgc = true;
24 public String toString() { return pt.toString(); }
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, this, null, tok, loc, input.getLocation(), null);
32 current.newNode(null, Forest.create(loc.createRegion(loc), null, null, false), pt.start, true);
34 for(int idx=0;;idx++) {
35 Input.Location oldloc = loc;
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, this, current, nextToken, loc, input.getLocation(), forest);
42 FileOutputStream fos = new FileOutputStream("out-"+idx+".dot");
43 PrintWriter p = new PrintWriter(new OutputStreamWriter(fos));
44 GraphViz gv = new GraphViz();
46 ((GSS.Phase.Node)n).toGraphViz(gv);
52 if (current.isDone()) return (Forest<NodeType>)gss.finalResult;
57 // Table //////////////////////////////////////////////////////////////////////////////
59 /** an SLR(1) parse table which may contain conflicts */
60 static class Table<Token> extends Walk.Cache {
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+" ");
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");
82 public final Walk.Cache cache = this;
84 private void walk(Element e, HashSet<SequenceOrElement> hs) {
86 if (hs.contains(e)) return;
88 if (e instanceof Atom) return;
89 for(Sequence s : (Union)e)
92 private void walk(Sequence s, HashSet<SequenceOrElement> hs) {
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);
100 /** the start state */
101 public final State<Token> start;
103 /** the state from which no reductions can be done */
104 private final State<Token> dead_state;
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>>();
110 /** construct a parse table for the given grammar */
111 public Table(Topology top) { this("s", top); }
112 public Table(String startSymbol, Topology top) { this(new Union(startSymbol), top); }
113 public Table(Union ux, Topology top) {
114 Union start0 = new Union("0");
115 start0.add(new Sequence.Singleton(ux));
117 for(Sequence s : start0) cache.eof.put(s, true);
118 cache.eof.put(start0, true);
120 // construct the set of states
121 HashSet<SequenceOrElement> all_elements = new HashSet<SequenceOrElement>();
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);
128 this.dead_state = new State<Token>(new HashSet<Position>(), all_states, all_elements);
129 this.start = new State<Token>(hp, all_states, all_elements);
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) {
135 // the Grammar's designated "last position" is the only accepting state
136 if (start0.contains(p.owner()) && p.next()==null)
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 follow = follow.intersect(new Walk.Follow(top.empty(), p2.element(), all_elements, cache).walk(p2.element()));
144 state.reductions.put(follow, p);
145 if (wf.includesEof()) state.eofReductions.add(p);
148 // if the element following this position is an atom, copy the corresponding
149 // set of rows out of the "master" goto table and into this state's shift table
150 if (p.element() != null && p.element() instanceof Atom)
151 state.shifts.addAll(state.gotoSetTerminals.subset(((Atom)p.element()).getTokenTopology()));
153 if (top instanceof IntegerTopology)
154 for(State<Token> state : all_states.values()) {
155 state.oreductions = state.reductions.optimize(((IntegerTopology)top).functor());
156 state.oshifts = state.shifts.optimize(((IntegerTopology)top).functor());
160 private boolean isRightNullable(Position p) {
161 if (p.isLast()) return true;
162 if (!possiblyEpsilon(p.element())) return false;
163 return isRightNullable(p.next());
166 /** a single state in the LR table and the transitions possible from it */
168 class State<Token> implements Comparable<State<Token>>, IntegerMappable, Iterable<Position> {
170 public final int idx = master_state_idx++;
171 private final HashSet<Position> hs;
173 public transient HashMap<Sequence,State<Token>> gotoSetNonTerminals = new HashMap<Sequence,State<Token>>();
174 private transient TopologicalBag<Token,State<Token>> gotoSetTerminals = new TopologicalBag<Token,State<Token>>();
176 private TopologicalBag<Token,Position> reductions = new TopologicalBag<Token,Position>();
177 private HashSet<Position> eofReductions = new HashSet<Position>();
178 private TopologicalBag<Token,State<Token>> shifts = new TopologicalBag<Token,State<Token>>();
179 private boolean accept = false;
181 private VisitableMap<Token,State<Token>> oshifts = null;
182 private VisitableMap<Token,Position> oreductions = null;
184 // Interface Methods //////////////////////////////////////////////////////////////////////////////
186 boolean isAccepting() { return accept; }
187 public Iterator<Position> iterator() { return hs.iterator(); }
189 boolean canShift(Token t) { return oshifts!=null && oshifts.contains(t); }
190 <B,C> void invokeShifts(Token t, Invokable<State<Token>,B,C> irbc, B b, C c) {
191 oshifts.invoke(t, irbc, b, c);
194 boolean canReduce(Token t) { return oreductions != null && (t==null ? eofReductions.size()>0 : oreductions.contains(t)); }
195 <B,C> void invokeReductions(Token t, Invokable<Position,B,C> irbc, B b, C c) {
196 if (t==null) for(Position r : eofReductions) irbc.invoke(r, b, c);
197 else oreductions.invoke(t, irbc, b, c);
200 // Constructor //////////////////////////////////////////////////////////////////////////////
203 * create a new state consisting of all the <tt>Position</tt>s in <tt>hs</tt>
204 * @param hs the set of <tt>Position</tt>s comprising this <tt>State</tt>
205 * @param all_states the set of states already constructed (to avoid recreating states)
206 * @param all_elements the set of all elements (Atom instances need not be included)
208 * In principle these two steps could be merged, but they
209 * are written separately to highlight these two facts:
211 * <li> Non-atom elements either match all-or-nothing, and do not overlap
212 * with each other (at least not in the sense of which element corresponds
213 * to the last reduction performed). Therefore, in order to make sure we
214 * wind up with the smallest number of states and shifts, we wait until
215 * we've figured out all the token-to-position multimappings before creating
218 * <li> In order to be able to run the state-construction algorithm in a single
219 * shot (rather than repeating until no new items appear in any state set),
220 * we need to use the "yields" semantics rather than the "produces" semantics
221 * for non-Atom Elements.
224 public State(HashSet<Position> hs,
225 HashMap<HashSet<Position>,State<Token>> all_states,
226 HashSet<SequenceOrElement> all_elements) {
229 // register ourselves in the all_states hash so that no
230 // two states are ever created with an identical position set
231 all_states.put(hs, this);
233 // Step 1a: examine all Position's in this state and compute the mappings from
234 // sets of follow tokens (tokens which could follow this position) to sets
235 // of _new_ positions (positions after shifting). These mappings are
236 // collectively known as the _closure_
238 TopologicalBag<Token,Position> bag0 = new TopologicalBag<Token,Position>();
239 for(Position position : hs) {
240 if (position.isLast() || !(position.element() instanceof Atom)) continue;
241 Atom a = (Atom)position.element();
242 HashSet<Position> hp = new HashSet<Position>();
243 reachable(position.next(), hp);
244 bag0.addAll(a.getTokenTopology(), hp);
247 // Step 1b: for each _minimal, contiguous_ set of characters having an identical next-position
248 // set, add that character set to the goto table (with the State corresponding to the
249 // computed next-position set).
251 for(Topology<Token> r : bag0) {
252 HashSet<Position> h = new HashSet<Position>();
253 for(Position p : bag0.getAll(r)) h.add(p);
254 gotoSetTerminals.put(r, all_states.get(h) == null ? new State<Token>(h, all_states, all_elements) : all_states.get(h));
257 // Step 2: for every non-Atom element (ie every Element which has a corresponding reduction),
258 // compute the closure over every position in this set which is followed by a symbol
259 // which could yield the Element in question.
261 // "yields" [in one or more step] is used instead of "produces" [in exactly one step]
262 // to avoid having to iteratively construct our set of States as shown in most
263 // expositions of the algorithm (ie "keep doing XYZ until things stop changing").
265 HashMapBag<SequenceOrElement,Position> move = new HashMapBag<SequenceOrElement,Position>();
266 for(Position p : hs) {
267 Element e = p.element();
268 if (e==null) continue;
269 for(SequenceOrElement y : cache.ys.getAll(e)) {
270 HashSet<Position> hp = new HashSet<Position>();
271 reachable(p.next(), hp);
275 OUTER: for(SequenceOrElement y : move) {
276 HashSet<Position> h = move.getAll(y);
277 State<Token> s = all_states.get(h) == null ? new State<Token>(h, all_states, all_elements) : all_states.get(h);
278 // if a reduction is "lame", it should wind up in the dead_state after reducing
279 if (y instanceof Sequence) {
280 for(Position p : hs) {
281 if (p.element() != null && (p.element() instanceof Union)) {
282 Union u = (Union)p.element();
283 for(Sequence seq : u)
284 if (seq.needs.contains((Sequence)y) || seq.hates.contains((Sequence)y)) {
285 // FIXME: what if there are two "routes" to get to the sequence?
286 ((HashMap)gotoSetNonTerminals).put((Sequence)y, dead_state);
291 gotoSetNonTerminals.put((Sequence)y, s);
296 public String toStringx() {
297 StringBuffer st = new StringBuffer();
298 for(Position p : this) {
299 if (st.length() > 0) st.append("\n");
302 return st.toString();
304 public String toString() {
305 StringBuffer ret = new StringBuffer();
306 ret.append("state["+idx+"]: ");
307 for(Position p : this) ret.append("{"+p+"} ");
308 return ret.toString();
311 public int compareTo(State<Token> s) { return idx==s.idx ? 0 : idx < s.idx ? -1 : 1; }
312 public int toInt() { return idx; }
316 // Helpers //////////////////////////////////////////////////////////////////////////////
318 private static void reachable(Sequence s, HashSet<Position> h) {
319 reachable(s.firstp(), h);
320 for(Sequence ss : s.needs()) reachable(ss, h);
321 for(Sequence ss : s.hates()) reachable(ss, h);
323 private static void reachable(Element e, HashSet<Position> h) {
324 if (e instanceof Atom) return;
325 for(Sequence s : ((Union)e))
328 private static void reachable(Position p, HashSet<Position> h) {
329 if (h.contains(p)) return;
331 if (p.element() != null) reachable(p.element(), h);