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 private static void reachable(Element e, HashSet<Position> h) {
17 if (e instanceof Atom) return;
18 for(Sequence s : ((Union)e))
19 reachable(s.firstp(), h);
21 private static void reachable(Position p, HashSet<Position> h) {
22 if (h.contains(p)) return;
24 if (p.element() != null) reachable(p.element(), h);
28 * create a parser to parse the grammar with start symbol <tt>u</tt>
30 protected Parser(Union u) { this.pt = new Table(u, top()); }
31 protected Parser(Table pt) { this.pt = pt; }
33 public abstract Forest<R> shiftedToken(T t);
34 public abstract Topology<T> top();
37 /** parse <tt>input</tt> for a exactly one unique result, throwing <tt>Ambiguous</tt> if not unique or <tt>Failed</tt> if none */
38 public Tree<R> parse1(Token.Stream<T> input) throws IOException, Failed, Ambiguous { return parse(input).expand1(); }
40 /** parse <tt>input</tt>, using the table <tt>pt</tt> to drive the parser */
41 public Forest<R> parse(Token.Stream<T> input) throws IOException, Failed {
43 GSS.Phase current = gss.new Phase(null, input.next());
44 current.newNode(null, null, pt.start, true, null);
46 GSS.Phase next = gss.new Phase(current, input.next());
48 Forest forest = current.token==null ? null : shiftedToken((T)current.token);
49 current.shift(next, forest);
50 if (current.isDone()) return (Forest<R>)current.finalResult;
51 current.checkFailure();
57 // Exceptions //////////////////////////////////////////////////////////////////////////////
59 public static class Failed extends Exception {
60 private final Token.Location location;
61 private final String message;
62 public Failed() { this("", null); }
63 public Failed(String message, Token.Location loc) { this.location = loc; this.message = message; }
64 public Token.Location getLocation() { return location; }
65 public String toString() { return message + (location==null ? "" : (" at " + location + "\n" + location.getContext())); }
68 public static class Ambiguous extends RuntimeException {
69 public final Forest ambiguity;
70 public Ambiguous(Forest ambiguity) { this.ambiguity = ambiguity; }
71 public String toString() {
72 StringBuffer sb = new StringBuffer();
73 sb.append("unresolved ambiguity "/*"at " + ambiguity.getLocation() + ":"*/);
74 for(Object result : ambiguity.expand(false))
75 sb.append("\n " + result);
81 // Table //////////////////////////////////////////////////////////////////////////////
83 static class Top extends Union { public Top() { super("0"); } }
85 /** an SLR(1) parse table which may contain conflicts */
88 private final Union start0 = new Top();
89 private final Sequence start0seq;
91 public final Walk.Cache cache = new Walk.Cache();
93 public HashSet<Position> closure() {
94 HashSet<Position> hp = new HashSet<Position>();
95 reachable(start0, hp);
98 public Position firstPosition() { return start0seq.firstp(); }
99 public Position lastPosition() { Position ret = start0seq.firstp(); while(!ret.isLast()) ret = ret.next(); return ret; }
101 private void walk(Element e, HashSet<Element> hs) {
103 if (hs.contains(e)) return;
105 if (e instanceof Atom) return;
106 for(Sequence s : (Union)e) {
108 for(Position p = s.firstp(); p != null; p = p.next())
109 walk(p.element(), hs);
112 public HashSet<Element> walk() {
113 HashSet<Element> ret = new HashSet<Element>();
119 public String toString() {
120 StringBuffer sb = new StringBuffer();
121 for(Element e : walk())
122 if (e instanceof Union)
123 ((Union)e).toString(sb);
124 return sb.toString();
128 /** the start state */
129 public final State start;
131 /** used to generate unique values for State.idx */
132 private int master_state_idx = 0;
134 /** construct a parse table for the given grammar */
135 public Table(Topology top) { this("s", top); }
136 public Table(String startSymbol, Topology top) { this(new Union(startSymbol), top); }
137 public Table(Union u, Topology top) {
138 start0seq = new Sequence.Singleton(u, null, null);
139 start0.add(start0seq);
141 // construct the set of states
142 HashMap<HashSet<Position>,State> all_states = new HashMap<HashSet<Position>,State>();
143 HashSet<Element> all_elements = walk();
144 for(Element e : all_elements)
145 cache.ys.put(e, new Walk.YieldSet(e, cache).walk());
146 this.start = new State(closure(), all_states, all_elements);
148 // for each state, fill in the corresponding "row" of the parse table
149 for(State state : all_states.values())
150 for(Position p : state.hs) {
152 // the Grammar's designated "last position" is the only accepting state
153 if (p==lastPosition())
156 // FIXME: how does right-nullability interact with follow restrictions?
157 // all right-nullable rules get a reduction [Johnstone 2000]
158 if (p.isRightNullable(cache)) {
159 Walk.Follow wf = new Walk.Follow(top.empty(), p.owner(), all_elements, cache);
160 Reduction red = new Reduction(p);
161 state.reductions.put(wf.walk(p.owner()), red);
162 if (wf.includesEof()) state.eofReductions.add(red, true);
165 // if the element following this position is an atom, copy the corresponding
166 // set of rows out of the "master" goto table and into this state's shift table
167 if (p.element() != null && p.element() instanceof Atom)
168 state.shifts.addAll(state.gotoSetTerminals.subset(((Atom)p.element())));
172 /** a single state in the LR table and the transitions possible from it */
173 public class State implements Comparable<Table.State>, Iterable<Position> {
175 public final int idx = master_state_idx++;
176 private final HashSet<Position> hs;
178 private transient HashMap<Element,State> gotoSetNonTerminals = new HashMap<Element,State>();
179 private transient TopologicalBag<Token,State> gotoSetTerminals = new TopologicalBag<Token,State>();
181 private TopologicalBag<Token,Reduction> reductions = new TopologicalBag<Token,Reduction>();
182 private FastSet<Reduction> eofReductions = new FastSet<Reduction>();
183 private TopologicalBag<Token,State> shifts = new TopologicalBag<Token,State>();
184 private boolean accept = false;
186 // Interface Methods //////////////////////////////////////////////////////////////////////////////
188 public boolean canShift(Token t) { return shifts.contains(t); }
189 public Iterable<State> getShifts(Token t) { return shifts.get(t); }
190 public boolean isAccepting() { return accept; }
191 public Iterable<Reduction> getReductions(Token t) { return reductions.get(t); }
192 public Iterable<Reduction> getEofReductions() { return eofReductions; }
193 public Iterator<Position> iterator() { return hs.iterator(); }
195 // Constructor //////////////////////////////////////////////////////////////////////////////
198 * create a new state consisting of all the <tt>Position</tt>s in <tt>hs</tt>
199 * @param hs the set of <tt>Position</tt>s comprising this <tt>State</tt>
200 * @param all_states the set of states already constructed (to avoid recreating states)
201 * @param all_elements the set of all elements (Atom instances need not be included)
203 * In principle these two steps could be merged, but they
204 * are written separately to highlight these two facts:
206 * <li> Non-atom elements either match all-or-nothing, and do not overlap
207 * with each other (at least not in the sense of which element corresponds
208 * to the last reduction performed). Therefore, in order to make sure we
209 * wind up with the smallest number of states and shifts, we wait until
210 * we've figured out all the token-to-position multimappings before creating
213 * <li> In order to be able to run the state-construction algorithm in a single
214 * shot (rather than repeating until no new items appear in any state set),
215 * we need to use the "yields" semantics rather than the "produces" semantics
216 * for non-Atom Elements.
219 public State(HashSet<Position> hs,
220 HashMap<HashSet<Position>,State> all_states,
221 HashSet<Element> all_elements) {
224 // register ourselves in the all_states hash so that no
225 // two states are ever created with an identical position set
226 all_states.put(hs, this);
228 // Step 1a: examine all Position's in this state and compute the mappings from
229 // sets of follow tokens (tokens which could follow this position) to sets
230 // of _new_ positions (positions after shifting). These mappings are
231 // collectively known as the _closure_
233 TopologicalBag<Token,Position> bag0 = new TopologicalBag<Token,Position>();
234 for(Position position : hs) {
235 if (position.isLast() || !(position.element() instanceof Atom)) continue;
236 Atom a = (Atom)position.element();
237 HashSet<Position> hp = new HashSet<Position>();
238 reachable(position.next(), hp);
239 bag0.addAll(a, /*clo.walk()*/hp);
242 // Step 1b: for each _minimal, contiguous_ set of characters having an identical next-position
243 // set, add that character set to the goto table (with the State corresponding to the
244 // computed next-position set).
246 for(Topology<Token> r : bag0) {
247 HashSet<Position> h = new HashSet<Position>();
248 for(Position p : bag0.getAll(r)) h.add(p);
249 gotoSetTerminals.put(r, all_states.get(h) == null ? new State(h, all_states, all_elements) : all_states.get(h));
252 // Step 2: for every non-Atom element (ie every Element which has a corresponding reduction),
253 // compute the closure over every position in this set which is followed by a symbol
254 // which could yield the Element in question.
256 // "yields" [in one or more step] is used instead of "produces" [in exactly one step]
257 // to avoid having to iteratively construct our set of States as shown in most
258 // expositions of the algorithm (ie "keep doing XYZ until things stop changing").
260 for(Element e : all_elements) {
261 if (e instanceof Atom) continue;
262 HashSet<Position> h = new Walk.Closure(null, g.cache).closure(e, hs);
263 State s = all_states.get(h) == null ? new State(h, all_states, all_elements) : all_states.get(h);
264 if (gotoSetNonTerminals.get(e) != null)
265 throw new Error("this should not happen");
266 gotoSetNonTerminals.put(e, s);
269 HashMapBag<Element,Position> move = new HashMapBag<Element,Position>();
270 for(Position p : hs) {
271 Element e = p.element();
272 if (e==null) continue;
273 HashSet<Element> ys = cache.ys.get(e);
275 for(Element y : ys) {
276 HashSet<Position> hp = new HashSet<Position>();
277 reachable(p.next(), hp);
282 for(Element y : move) {
283 HashSet<Position> h = move.getAll(y);
284 State s = all_states.get(h) == null ? new State(h, all_states, all_elements) : all_states.get(h);
285 gotoSetNonTerminals.put(y, s);
289 public String toString() { return "state["+idx+"]"; }
291 public int compareTo(Table.State s) { return idx==s.idx ? 0 : idx < s.idx ? -1 : 1; }
295 * the information needed to perform a reduction; copied here to
296 * avoid keeping references to <tt>Element</tt> objects in a Table
298 public class Reduction {
299 // FIXME: cleanup; almost everything in here could go in either Sequence.Position.getRewrite() or else in GSS.Reduct
300 public final int numPop;
301 private final Position position;
302 private final Forest[] holder; // to avoid constant reallocation
303 public int hashCode() { return position.hashCode(); }
304 public boolean equals(Object o) {
305 if (o==null) return false;
306 if (o==this) return true;
307 if (!(o instanceof Reduction)) return false;
308 Reduction r = (Reduction)o;
309 return r.position == position;
311 public Reduction(Position p) {
314 this.holder = new Forest[numPop];
316 public String toString() { return "[reduce " + position + "]"; }
317 public Forest reduce(Forest f, GSS.Phase.Node parent, GSS.Phase.Node onlychild, GSS.Phase target, Forest rex) {
318 holder[numPop-1] = f;
319 return reduce(parent, numPop-2, rex, onlychild, target);
321 public Forest reduce(GSS.Phase.Node parent, GSS.Phase.Node onlychild, GSS.Phase target, Forest rex) {
322 return reduce(parent, numPop-1, rex, onlychild, target);
325 // FIXME: this could be more elegant and/or cleaner and/or somewhere else
326 private Forest reduce(GSS.Phase.Node parent, int pos, Forest rex, GSS.Phase.Node onlychild, GSS.Phase target) {
327 if (pos>=0) holder[pos] = parent.pending();
328 if (pos<=0 && rex==null) {
329 System.arraycopy(holder, 0, position.holder, 0, holder.length);
330 rex = position.rewrite(target.getLocation());
333 if (onlychild != null)
334 reduce(onlychild, pos-1, rex, null, target);
336 for(GSS.Phase.Node child : parent.parents())
337 reduce(child, pos-1, rex, null, target);
339 State state = parent.state.gotoSetNonTerminals.get(position.owner());
341 target.newNode(parent, rex, state, numPop<=0, parent.phase);
348 private static final Forest[] emptyForestArray = new Forest[0];