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);
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 if (p.isRightNullable(cache)) {
132 Walk.Follow wf = new Walk.Follow(top.empty(), p.owner(), all_elements, cache);
133 Reduction red = new Reduction(p);
135 Topology follow = wf.walk(p.owner());
136 if (p.owner() instanceof Sequence.RewritingSequence &&
137 (((Sequence.RewritingSequence)p.owner()).tag+"").equals("emailaddr")) {
138 System.out.println("follow before: " + new edu.berkeley.sbp.misc.CharToken.CharRange(follow));
140 for(Position p2 = p; p2 != null && p2.element() != null; p2 = p2.next())
141 follow = follow.intersect(new Walk.Follow(top.empty(), p2.element(), all_elements, cache).walk(p2.element()));
142 if (p.owner() instanceof Sequence.RewritingSequence &&
143 (((Sequence.RewritingSequence)p.owner()).tag+"").equals("emailaddr")) {
144 System.out.println("follow after: " + new edu.berkeley.sbp.misc.CharToken.CharRange(follow));
146 state.reductions.put(follow, red);
147 if (wf.includesEof()) state.eofReductions.add(red);
150 // if the element following this position is an atom, copy the corresponding
151 // set of rows out of the "master" goto table and into this state's shift table
152 if (p.element() != null && p.element() instanceof Atom)
153 state.shifts.addAll(state.gotoSetTerminals.subset(((Atom)p.element())));
155 for(State state : all_states.values()) {
156 state.oreductions = state.reductions.optimize();
157 state.oshifts = state.shifts.optimize();
161 /** a single state in the LR table and the transitions possible from it */
162 public class State implements Comparable<Table.State>, Iterable<Position> {
165 public boolean isResolvable(Token t) {
166 boolean found = false;
167 for(Reduction r : getReductions(t)) {
168 Position p = r.position;
169 if (!p.isRightNullable(cache)) continue;
170 if (p.owner().firstp()==p) continue;
172 // found two items meeting criteria #1
178 if (p.element()==null) continue;
179 Topology first = new Walk.First(top(), cache).walk(p.element());
180 if (first.contains(t))
185 public final int idx = master_state_idx++;
186 private final HashSet<Position> hs;
188 private transient HashMap<Element,State> gotoSetNonTerminals = new HashMap<Element,State>();
189 private transient TopologicalBag<Token,State> gotoSetTerminals = new TopologicalBag<Token,State>();
191 private TopologicalBag<Token,Reduction> reductions = new TopologicalBag<Token,Reduction>();
192 private HashSet<Reduction> eofReductions = new HashSet<Reduction>();
193 private TopologicalBag<Token,State> shifts = new TopologicalBag<Token,State>();
194 private boolean accept = false;
196 private VisitableMap<Token,State> oshifts = null;
197 private VisitableMap<Token,Reduction> oreductions = null;
199 // Interface Methods //////////////////////////////////////////////////////////////////////////////
201 public boolean isAccepting() { return accept; }
203 public boolean canShift(Token t) { return oshifts.contains(t); }
204 public boolean canReduce(Token t) { return t==null ? eofReductions.size()>0 : oreductions.contains(t); }
206 public Iterator<Position> iterator() { return hs.iterator(); }
208 public <B,C> void invokeShifts(Token t, Invokable<State,B,C> irbc, B b, C c) {
209 oshifts.invoke(t, irbc, b, c);
211 public <B,C> void invokeReductions(Token t, Invokable<Reduction,B,C> irbc, B b, C c) {
212 if (t==null) for(Reduction r : eofReductions) irbc.invoke(r, b, c);
213 else oreductions.invoke(t, irbc, b, c);
216 // Constructor //////////////////////////////////////////////////////////////////////////////
219 * create a new state consisting of all the <tt>Position</tt>s in <tt>hs</tt>
220 * @param hs the set of <tt>Position</tt>s comprising this <tt>State</tt>
221 * @param all_states the set of states already constructed (to avoid recreating states)
222 * @param all_elements the set of all elements (Atom instances need not be included)
224 * In principle these two steps could be merged, but they
225 * are written separately to highlight these two facts:
227 * <li> Non-atom elements either match all-or-nothing, and do not overlap
228 * with each other (at least not in the sense of which element corresponds
229 * to the last reduction performed). Therefore, in order to make sure we
230 * wind up with the smallest number of states and shifts, we wait until
231 * we've figured out all the token-to-position multimappings before creating
234 * <li> In order to be able to run the state-construction algorithm in a single
235 * shot (rather than repeating until no new items appear in any state set),
236 * we need to use the "yields" semantics rather than the "produces" semantics
237 * for non-Atom Elements.
240 public State(HashSet<Position> hs,
241 HashMap<HashSet<Position>,State> all_states,
242 HashSet<Element> all_elements) {
245 // register ourselves in the all_states hash so that no
246 // two states are ever created with an identical position set
247 all_states.put(hs, this);
249 // Step 1a: examine all Position's in this state and compute the mappings from
250 // sets of follow tokens (tokens which could follow this position) to sets
251 // of _new_ positions (positions after shifting). These mappings are
252 // collectively known as the _closure_
254 TopologicalBag<Token,Position> bag0 = new TopologicalBag<Token,Position>();
255 for(Position position : hs) {
256 if (position.isLast() || !(position.element() instanceof Atom)) continue;
257 Atom a = (Atom)position.element();
258 HashSet<Position> hp = new HashSet<Position>();
259 reachable(position.next(), hp);
263 // Step 1b: for each _minimal, contiguous_ set of characters having an identical next-position
264 // set, add that character set to the goto table (with the State corresponding to the
265 // computed next-position set).
267 for(Topology<Token> r : bag0) {
268 HashSet<Position> h = new HashSet<Position>();
269 for(Position p : bag0.getAll(r)) h.add(p);
270 gotoSetTerminals.put(r, all_states.get(h) == null ? new State(h, all_states, all_elements) : all_states.get(h));
273 // Step 2: for every non-Atom element (ie every Element which has a corresponding reduction),
274 // compute the closure over every position in this set which is followed by a symbol
275 // which could yield the Element in question.
277 // "yields" [in one or more step] is used instead of "produces" [in exactly one step]
278 // to avoid having to iteratively construct our set of States as shown in most
279 // expositions of the algorithm (ie "keep doing XYZ until things stop changing").
281 for(Element e : all_elements) {
282 if (e instanceof Atom) continue;
283 HashSet<Position> h = new Walk.Closure(null, g.cache).closure(e, hs);
284 State s = all_states.get(h) == null ? new State(h, all_states, all_elements) : all_states.get(h);
285 if (gotoSetNonTerminals.get(e) != null)
286 throw new Error("this should not happen");
287 gotoSetNonTerminals.put(e, s);
290 HashMapBag<Element,Position> move = new HashMapBag<Element,Position>();
291 for(Position p : hs) {
292 Element e = p.element();
293 if (e==null) continue;
294 HashSet<Element> ys = cache.ys.get(e);
296 for(Element y : ys) {
297 HashSet<Position> hp = new HashSet<Position>();
298 reachable(p.next(), hp);
303 for(Element y : move) {
304 HashSet<Position> h = move.getAll(y);
305 State s = all_states.get(h) == null ? new State(h, all_states, all_elements) : all_states.get(h);
306 gotoSetNonTerminals.put(y, s);
310 public String toString() { return "state["+idx+"]"; }
312 public int compareTo(Table.State s) { return idx==s.idx ? 0 : idx < s.idx ? -1 : 1; }
316 * the information needed to perform a reduction; copied here to
317 * avoid keeping references to <tt>Element</tt> objects in a Table
319 public class Reduction {
320 // FIXME: cleanup; almost everything in here could go in either Sequence.Position.getRewrite() or else in GSS.Reduct
321 public final int numPop;
322 /*private*/ final Position position;
323 private final Forest[] holder; // to avoid constant reallocation
324 public int hashCode() { return position.hashCode(); }
325 public boolean equals(Object o) {
326 if (o==null) return false;
327 if (o==this) return true;
328 if (!(o instanceof Reduction)) return false;
329 Reduction r = (Reduction)o;
330 return r.position == position;
332 public Reduction(Position p) {
335 this.holder = new Forest[numPop];
337 public String toString() { return "[reduce " + position + "]"; }
339 private Forest zero = null;
340 public Forest zero() {
341 if (zero != null) return zero;
342 if (numPop > 0) throw new Error();
343 return zero = position.rewrite(null);
346 public void reduce(GSS.Phase.Node parent) {
347 if (numPop==0) finish(parent, zero(), parent.phase());
348 else reduce(parent, numPop-1, parent.phase());
351 public void reduce(GSS.Phase.Node parent, GSS.Phase.Node onlychild) {
352 if (numPop<=0) throw new Error("called wrong form of reduce()");
354 Forest old = holder[pos];
355 holder[pos] = parent.pending();
357 System.arraycopy(holder, 0, position.holder, 0, holder.length);
358 finish(onlychild, position.rewrite(parent.phase().getLocation()), parent.phase());
360 reduce(onlychild, pos-1, parent.phase());
365 // FIXME: this could be more elegant and/or cleaner and/or somewhere else
366 private void reduce(GSS.Phase.Node parent, int pos, GSS.Phase target) {
367 Forest old = holder[pos];
368 holder[pos] = parent.pending();
370 System.arraycopy(holder, 0, position.holder, 0, holder.length);
371 for(int i=0; i<position.pos; i++) if (position.holder[i]==null) throw new Error("realbad");
372 Forest rex = position.rewrite(target.getLocation());
373 for(GSS.Phase.Node child : parent.parents()) finish(child, rex, target);
375 for(GSS.Phase.Node child : parent.parents()) reduce(child, pos-1, target);
379 private void finish(GSS.Phase.Node parent, Forest result, GSS.Phase target) {
380 State state = parent.state.gotoSetNonTerminals.get(position.owner());
382 target.newNode(parent, result, state, numPop<=0);
387 private static final Forest[] emptyForestArray = new Forest[0];
390 // Helpers //////////////////////////////////////////////////////////////////////////////
392 private static void reachable(Element e, HashSet<Position> h) {
393 if (e instanceof Atom) return;
394 for(Sequence s : ((Union)e))
395 reachable(s.firstp(), h);
397 private static void reachable(Position p, HashSet<Position> h) {
398 if (h.contains(p)) return;
400 if (p.element() != null) reachable(p.element(), h);