1 package edu.berkeley.sbp;
2 import edu.berkeley.sbp.*;
3 import edu.berkeley.sbp.util.*;
4 import edu.berkeley.sbp.*;
5 import edu.berkeley.sbp.Sequence.Position;
6 import edu.berkeley.sbp.*;
9 import java.lang.reflect.*;
11 /** a parser which translates streams of Tokens of type T into a Forest<R> */
12 public abstract class Parser<T extends Token, R> {
14 private final Table pt;
16 /** create a parser to parse the grammar with start symbol <tt>u</tt> */
17 protected Parser(Union u) { this.pt = new Table(u, top()); }
18 protected Parser(Table pt) { this.pt = pt; }
20 public abstract Forest<R> shiftedToken(T t, Token.Location loc);
21 public abstract Topology<T> top();
24 /** parse <tt>input</tt> for a exactly one unique result, throwing <tt>Ambiguous</tt> if not unique or <tt>Failed</tt> if none */
25 public Tree<R> parse1(Token.Stream<T> input) throws IOException, Failed, Ambiguous {
26 Forest<R> ret = parse(input);
27 try { return ret.expand1(); }
29 System.out.println("while expanding:");
30 System.out.println(ret);
35 /** parse <tt>input</tt>, using the table <tt>pt</tt> to drive the parser */
36 public Forest<R> parse(Token.Stream<T> input) throws IOException, Failed {
38 Token.Location loc = input.getLocation();
39 GSS.Phase current = gss.new Phase(null, input.next(), loc);
40 current.newNode(null, null, pt.start, true, null);
42 loc = input.getLocation();
43 GSS.Phase next = gss.new Phase(current, input.next(), loc);
45 Forest forest = current.token==null ? null : shiftedToken((T)current.token, loc);
46 current.shift(next, forest);
47 if (current.isDone()) return (Forest<R>)current.finalResult;
48 current.checkFailure();
54 // Exceptions //////////////////////////////////////////////////////////////////////////////
56 public static class Failed extends Exception {
57 private final Token.Location location;
58 private final String message;
59 public Failed() { this("", null); }
60 public Failed(String message, Token.Location loc) { this.location = loc; this.message = message; }
61 public Token.Location getLocation() { return location; }
62 public String toString() { return message + (location==null ? "" : (" at " + location)); }
65 public static class Ambiguous extends RuntimeException {
66 public final Forest ambiguity;
67 public Ambiguous(Forest ambiguity) { this.ambiguity = ambiguity; }
68 public String toString() {
69 StringBuffer sb = new StringBuffer();
70 sb.append("unresolved ambiguity "/*"at " + ambiguity.getLocation() + ":"*/);
71 for(Object result : ambiguity.expand(false))
72 sb.append("\n " + result);
78 // Table //////////////////////////////////////////////////////////////////////////////
80 /** an SLR(1) parse table which may contain conflicts */
81 static class Table extends Walk.Cache {
83 public final Walk.Cache cache = this;
85 private void walk(Element e, HashSet<Element> hs) {
87 if (hs.contains(e)) return;
89 if (e instanceof Atom) return;
90 for(Sequence s : (Union)e) {
92 for(Position p = s.firstp(); p != null; p = p.next())
93 walk(p.element(), hs);
97 /** the start state */
98 public final State start;
100 /** used to generate unique values for State.idx */
101 private int master_state_idx = 0;
103 /** construct a parse table for the given grammar */
104 public Table(Topology top) { this("s", top); }
105 public Table(String startSymbol, Topology top) { this(new Union(startSymbol), top); }
106 public Table(Union ux, Topology top) {
107 Union start0 = new Union("0");
108 start0.add(new Sequence.Singleton(ux, null, null));
110 for(Sequence s : start0) cache.eof.put(s, true);
111 cache.eof.put(start0, true);
113 // construct the set of states
114 HashMap<HashSet<Position>,State> all_states = new HashMap<HashSet<Position>,State>();
115 HashSet<Element> all_elements = new HashSet<Element>();
116 walk(start0, all_elements);
117 for(Element e : all_elements)
118 cache.ys.put(e, new Walk.YieldSet(e, cache).walk());
119 HashSet<Position> hp = new HashSet<Position>();
120 reachable(start0, hp);
121 this.start = new State(hp, all_states, all_elements);
123 // for each state, fill in the corresponding "row" of the parse table
124 for(State state : all_states.values())
125 for(Position p : state.hs) {
127 // the Grammar's designated "last position" is the only accepting state
128 if (start0.contains(p.owner()) && p.next()==null)
131 // FIXME: how does right-nullability interact with follow restrictions?
132 // all right-nullable rules get a reduction [Johnstone 2000]
133 if (p.isRightNullable(cache)) {
134 Walk.Follow wf = new Walk.Follow(top.empty(), p.owner(), all_elements, cache);
135 Reduction red = new Reduction(p);
136 state.reductions.put(wf.walk(p.owner()), red);
137 if (wf.includesEof()) state.eofReductions.add(red);
140 // if the element following this position is an atom, copy the corresponding
141 // set of rows out of the "master" goto table and into this state's shift table
142 if (p.element() != null && p.element() instanceof Atom)
143 state.shifts.addAll(state.gotoSetTerminals.subset(((Atom)p.element())));
147 /** a single state in the LR table and the transitions possible from it */
148 public class State implements Comparable<Table.State>, Iterable<Position> {
151 public boolean isResolvable(Token t) {
152 boolean found = false;
153 for(Reduction r : getReductions(t)) {
154 Position p = r.position;
155 if (!p.isRightNullable(cache)) continue;
156 if (p.owner().firstp()==p) continue;
158 // found two items meeting criteria #1
164 if (p.element()==null) continue;
165 Topology first = new Walk.First(top(), cache).walk(p.element());
166 if (first.contains(t))
171 public final int idx = master_state_idx++;
172 private final HashSet<Position> hs;
174 private transient HashMap<Element,State> gotoSetNonTerminals = new HashMap<Element,State>();
175 private transient TopologicalBag<Token,State> gotoSetTerminals = new TopologicalBag<Token,State>();
177 private TopologicalBag<Token,Reduction> reductions = new TopologicalBag<Token,Reduction>();
178 private HashSet<Reduction> eofReductions = new HashSet<Reduction>();
179 private TopologicalBag<Token,State> shifts = new TopologicalBag<Token,State>();
180 private boolean accept = false;
182 // Interface Methods //////////////////////////////////////////////////////////////////////////////
184 public boolean canShift(Token t) { return shifts.contains(t); }
185 public Iterable<State> getShifts(Token t) { return shifts.get(t); }
186 public boolean isAccepting() { return accept; }
187 public Iterable<Reduction> getReductions(Token t) { return t==null ? eofReductions : reductions.get(t); }
188 public Iterable<Reduction> getEofReductions() { return eofReductions; }
189 public Iterator<Position> iterator() { return hs.iterator(); }
191 // Constructor //////////////////////////////////////////////////////////////////////////////
194 * create a new state consisting of all the <tt>Position</tt>s in <tt>hs</tt>
195 * @param hs the set of <tt>Position</tt>s comprising this <tt>State</tt>
196 * @param all_states the set of states already constructed (to avoid recreating states)
197 * @param all_elements the set of all elements (Atom instances need not be included)
199 * In principle these two steps could be merged, but they
200 * are written separately to highlight these two facts:
202 * <li> Non-atom elements either match all-or-nothing, and do not overlap
203 * with each other (at least not in the sense of which element corresponds
204 * to the last reduction performed). Therefore, in order to make sure we
205 * wind up with the smallest number of states and shifts, we wait until
206 * we've figured out all the token-to-position multimappings before creating
209 * <li> In order to be able to run the state-construction algorithm in a single
210 * shot (rather than repeating until no new items appear in any state set),
211 * we need to use the "yields" semantics rather than the "produces" semantics
212 * for non-Atom Elements.
215 public State(HashSet<Position> hs,
216 HashMap<HashSet<Position>,State> all_states,
217 HashSet<Element> all_elements) {
220 // register ourselves in the all_states hash so that no
221 // two states are ever created with an identical position set
222 all_states.put(hs, this);
224 // Step 1a: examine all Position's in this state and compute the mappings from
225 // sets of follow tokens (tokens which could follow this position) to sets
226 // of _new_ positions (positions after shifting). These mappings are
227 // collectively known as the _closure_
229 TopologicalBag<Token,Position> bag0 = new TopologicalBag<Token,Position>();
230 for(Position position : hs) {
231 if (position.isLast() || !(position.element() instanceof Atom)) continue;
232 Atom a = (Atom)position.element();
233 HashSet<Position> hp = new HashSet<Position>();
234 reachable(position.next(), hp);
238 // Step 1b: for each _minimal, contiguous_ set of characters having an identical next-position
239 // set, add that character set to the goto table (with the State corresponding to the
240 // computed next-position set).
242 for(Topology<Token> r : bag0) {
243 HashSet<Position> h = new HashSet<Position>();
244 for(Position p : bag0.getAll(r)) h.add(p);
245 gotoSetTerminals.put(r, all_states.get(h) == null ? new State(h, all_states, all_elements) : all_states.get(h));
248 // Step 2: for every non-Atom element (ie every Element which has a corresponding reduction),
249 // compute the closure over every position in this set which is followed by a symbol
250 // which could yield the Element in question.
252 // "yields" [in one or more step] is used instead of "produces" [in exactly one step]
253 // to avoid having to iteratively construct our set of States as shown in most
254 // expositions of the algorithm (ie "keep doing XYZ until things stop changing").
256 for(Element e : all_elements) {
257 if (e instanceof Atom) continue;
258 HashSet<Position> h = new Walk.Closure(null, g.cache).closure(e, hs);
259 State s = all_states.get(h) == null ? new State(h, all_states, all_elements) : all_states.get(h);
260 if (gotoSetNonTerminals.get(e) != null)
261 throw new Error("this should not happen");
262 gotoSetNonTerminals.put(e, s);
265 HashMapBag<Element,Position> move = new HashMapBag<Element,Position>();
266 for(Position p : hs) {
267 Element e = p.element();
268 if (e==null) continue;
269 HashSet<Element> ys = cache.ys.get(e);
271 for(Element y : ys) {
272 HashSet<Position> hp = new HashSet<Position>();
273 reachable(p.next(), hp);
278 for(Element y : move) {
279 HashSet<Position> h = move.getAll(y);
280 State s = all_states.get(h) == null ? new State(h, all_states, all_elements) : all_states.get(h);
281 gotoSetNonTerminals.put(y, s);
285 public String toString() { return "state["+idx+"]"; }
287 public int compareTo(Table.State s) { return idx==s.idx ? 0 : idx < s.idx ? -1 : 1; }
291 * the information needed to perform a reduction; copied here to
292 * avoid keeping references to <tt>Element</tt> objects in a Table
294 public class Reduction {
295 // FIXME: cleanup; almost everything in here could go in either Sequence.Position.getRewrite() or else in GSS.Reduct
296 public final int numPop;
297 /*private*/ final Position position;
298 private final Forest[] holder; // to avoid constant reallocation
299 public int hashCode() { return position.hashCode(); }
300 public boolean equals(Object o) {
301 if (o==null) return false;
302 if (o==this) return true;
303 if (!(o instanceof Reduction)) return false;
304 Reduction r = (Reduction)o;
305 return r.position == position;
307 public Reduction(Position p) {
310 this.holder = new Forest[numPop];
312 public String toString() { return "[reduce " + position + "]"; }
314 private Forest zero = null;
315 public Forest zero() {
316 if (zero != null) return zero;
317 if (numPop > 0) throw new Error();
318 return zero = position.rewrite(null);
321 public Forest reduce(GSS.Phase.Node parent) {
322 if (numPop==0) return finish(parent, zero(), parent.phase());
323 return reduce(parent, numPop-1, null, parent.phase());
326 public Forest reduce(GSS.Phase.Node parent, GSS.Phase.Node onlychild) {
327 if (numPop<=0) throw new Error("called wrong form of reduce()");
329 holder[pos] = parent.pending();
331 System.arraycopy(holder, 0, position.holder, 0, holder.length);
332 return finish(onlychild, position.rewrite(parent.phase().getLocation()), parent.phase());
334 return reduce(onlychild, pos-1, null, parent.phase());
337 // FIXME: this could be more elegant and/or cleaner and/or somewhere else
338 private Forest reduce(GSS.Phase.Node parent, int pos, Forest rex, GSS.Phase target) {
339 if (pos<0) return finish(parent, rex, target);
340 holder[pos] = parent.pending();
341 if (pos==0 && rex==null) {
342 System.arraycopy(holder, 0, position.holder, 0, holder.length);
343 rex = position.rewrite(target.getLocation());
345 for(GSS.Phase.Node child : parent.parents())
346 //if (pos==0) finish(parent, rex, target);
348 reduce(child, pos-1, rex, target);
351 private Forest finish(GSS.Phase.Node parent, Forest result, GSS.Phase target) {
352 State state = parent.state.gotoSetNonTerminals.get(position.owner());
354 target.newNode(parent, result, state, numPop<=0, parent.phase());
360 private static final Forest[] emptyForestArray = new Forest[0];
363 // Helpers //////////////////////////////////////////////////////////////////////////////
365 private static void reachable(Element e, HashSet<Position> h) {
366 if (e instanceof Atom) return;
367 for(Sequence s : ((Union)e))
368 reachable(s.firstp(), h);
370 private static void reachable(Position p, HashSet<Position> h) {
371 if (h.contains(p)) return;
373 if (p.element() != null) reachable(p.element(), h);