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 class Parser<T extends Token, R> {
14 private final Table pt;
16 //public Parser( Topology top) { this(new Table( top)); }
17 //public Parser(String s, Topology top) { this(new Table(s, top)); }
20 * create a parser to parse the grammar with start symbol <tt>u</tt>
21 * @param top a "sample" Topology<T> that can be cloned (FIXME, demanding this is lame)
23 public Parser(Union u, Topology<T> top) { this(new Table(u, top)); }
25 Parser(Table pt) { this.pt = pt; }
27 /** parse <tt>input</tt> for a exactly one unique result, throwing <tt>Ambiguous</tt> if not unique or <tt>Failed</tt> if none */
28 public Tree<R> parse1(Token.Stream<T> input) throws IOException, Failed, Ambiguous { return parse(input).expand1(); }
30 /** parse <tt>input</tt>, using the table <tt>pt</tt> to drive the parser */
31 public Forest<R> parse(Token.Stream<T> input) throws IOException, Failed {
33 GSS.Phase current = gss.new Phase(null, input.next());
34 current.newNode(null, null, pt.start, true, null);
36 GSS.Phase next = gss.new Phase(current, input.next());
39 if (current.isDone()) return (Forest<R>)current.finalResult;
40 current.checkFailure();
46 // Exceptions //////////////////////////////////////////////////////////////////////////////
48 public static class Failed extends Exception {
49 private final Token.Location location;
50 private final String message;
51 public Failed() { this("", null); }
52 public Failed(String message, Token.Location loc) { this.location = loc; this.message = message; }
53 public Token.Location getLocation() { return location; }
54 public String toString() { return message + (location==null ? "" : (" at " + location + "\n" + location.getContext())); }
57 public static class Ambiguous extends RuntimeException {
58 public final Forest ambiguity;
59 public Ambiguous(Forest ambiguity) { this.ambiguity = ambiguity; }
60 public String toString() {
61 StringBuffer sb = new StringBuffer();
62 sb.append("unresolved ambiguity "/*"at " + ambiguity.getLocation() + ":"*/);
63 for(Object result : ambiguity.expand(false))
64 sb.append("\n " + result);
70 // Table //////////////////////////////////////////////////////////////////////////////
72 /** an SLR(1) parse table which may contain conflicts */
75 private final Union start0 = new Top();
76 private final Sequence start0seq;
77 static class Top extends Union { public Top() { super("0"); } }
79 public final Walk.Cache cache = new Walk.Cache();
81 public HashSet<Position> closure() {
82 HashSet<Position> hp = new HashSet<Position>();
86 public Position firstPosition() { return start0seq.firstp(); }
87 public Position lastPosition() { Position ret = start0seq.firstp(); while(!ret.isLast()) ret = ret.next(); return ret; }
89 private void walk(Element e, HashSet<Element> hs) {
91 if (hs.contains(e)) return;
93 if (e instanceof Atom) return;
94 for(Sequence s : (Union)e) {
96 for(Position p = s.firstp(); p != null; p = p.next())
97 walk(p.element(), hs);
100 public HashSet<Element> walk() {
101 HashSet<Element> ret = new HashSet<Element>();
107 public String toString() {
108 StringBuffer sb = new StringBuffer();
109 for(Element e : walk())
110 if (e instanceof Union)
111 ((Union)e).toString(sb);
112 return sb.toString();
116 /** the start state */
117 public final State start;
119 /** used to generate unique values for State.idx */
120 private int master_state_idx = 0;
122 /** construct a parse table for the given grammar */
123 public Table(Topology top) { this("s", top); }
124 public Table(String startSymbol, Topology top) { this(new Union(startSymbol), top); }
125 public Table(Union u, Topology top) {
126 start0seq = new Sequence.Singleton(u, null, null);
127 start0.add(start0seq);
129 // construct the set of states
130 HashMap<HashSet<Position>,State> all_states = new HashMap<HashSet<Position>,State>();
131 HashSet<Element> all_elements = walk();
132 for(Element e : all_elements)
133 cache.ys.put(e, new Walk.YieldSet(e, cache).walk());
134 this.start = new State(closure(), all_states, all_elements);
136 // for each state, fill in the corresponding "row" of the parse table
137 for(State state : all_states.values())
138 for(Position p : state.hs) {
140 // the Grammar's designated "last position" is the only accepting state
141 if (p==lastPosition())
144 // FIXME: how does right-nullability interact with follow restrictions?
145 // all right-nullable rules get a reduction [Johnstone 2000]
146 if (p.isRightNullable(cache)) {
147 Walk.Follow wf = new Walk.Follow(top.fresh(), p.owner(), all_elements, cache);
148 Reduction red = new Reduction(p);
149 state.reductions.put(wf.walk(p.owner()), red);
150 if (wf.includesEof()) state.eofReductions.add(red, true);
153 // if the element following this position is an atom, copy the corresponding
154 // set of rows out of the "master" goto table and into this state's shift table
155 if (p.element() != null && p.element() instanceof Atom)
156 state.shifts.addAll(state.gotoSetTerminals.subset(((Atom)p.element()).top()));
160 /** a single state in the LR table and the transitions possible from it */
161 public class State implements Comparable<Table.State>, Iterable<Position> {
163 public final int idx = master_state_idx++;
164 private final HashSet<Position> hs;
166 private transient HashMap<Element,State> gotoSetNonTerminals = new HashMap<Element,State>();
167 private transient TopologicalBag<Token,State> gotoSetTerminals = new TopologicalBag<Token,State>();
169 private TopologicalBag<Token,Reduction> reductions = new TopologicalBag<Token,Reduction>();
170 private FastSet<Reduction> eofReductions = new FastSet<Reduction>();
171 private TopologicalBag<Token,State> shifts = new TopologicalBag<Token,State>();
172 private boolean accept = false;
174 // Interface Methods //////////////////////////////////////////////////////////////////////////////
176 public boolean canShift(Token t) { return shifts.contains(t); }
177 public Iterable<State> getShifts(Token t) { return shifts.get(t); }
178 public boolean isAccepting() { return accept; }
179 public Iterable<Reduction> getReductions(Token t) { return reductions.get(t); }
180 public Iterable<Reduction> getEofReductions() { return eofReductions; }
181 public Iterator<Position> iterator() { return hs.iterator(); }
183 // Constructor //////////////////////////////////////////////////////////////////////////////
186 * create a new state consisting of all the <tt>Position</tt>s in <tt>hs</tt>
187 * @param hs the set of <tt>Position</tt>s comprising this <tt>State</tt>
188 * @param all_states the set of states already constructed (to avoid recreating states)
189 * @param all_elements the set of all elements (Atom instances need not be included)
191 * In principle these two steps could be merged, but they
192 * are written separately to highlight these two facts:
194 * <li> Non-atom elements either match all-or-nothing, and do not overlap
195 * with each other (at least not in the sense of which element corresponds
196 * to the last reduction performed). Therefore, in order to make sure we
197 * wind up with the smallest number of states and shifts, we wait until
198 * we've figured out all the token-to-position multimappings before creating
201 * <li> In order to be able to run the state-construction algorithm in a single
202 * shot (rather than repeating until no new items appear in any state set),
203 * we need to use the "yields" semantics rather than the "produces" semantics
204 * for non-Atom Elements.
207 public State(HashSet<Position> hs,
208 HashMap<HashSet<Position>,State> all_states,
209 HashSet<Element> all_elements) {
212 // register ourselves in the all_states hash so that no
213 // two states are ever created with an identical position set
214 all_states.put(hs, this);
216 // Step 1a: examine all Position's in this state and compute the mappings from
217 // sets of follow tokens (tokens which could follow this position) to sets
218 // of _new_ positions (positions after shifting). These mappings are
219 // collectively known as the _closure_
221 TopologicalBag<Token,Position> bag0 = new TopologicalBag<Token,Position>();
222 for(Position position : hs) {
223 if (position.isLast() || !(position.element() instanceof Atom)) continue;
224 Atom a = (Atom)position.element();
225 HashSet<Position> hp = new HashSet<Position>();
226 position.next().reachable(hp);
227 bag0.addAll(a.top(), /*clo.walk()*/hp);
230 // Step 1b: for each _minimal, contiguous_ set of characters having an identical next-position
231 // set, add that character set to the goto table (with the State corresponding to the
232 // computed next-position set).
234 for(Topology<Token> r : bag0) {
235 HashSet<Position> h = new HashSet<Position>();
236 for(Position p : bag0.getAll(r)) h.add(p);
237 gotoSetTerminals.put(r, all_states.get(h) == null ? new State(h, all_states, all_elements) : all_states.get(h));
240 // Step 2: for every non-Atom element (ie every Element which has a corresponding reduction),
241 // compute the closure over every position in this set which is followed by a symbol
242 // which could yield the Element in question.
244 // "yields" [in one or more step] is used instead of "produces" [in exactly one step]
245 // to avoid having to iteratively construct our set of States as shown in most
246 // expositions of the algorithm (ie "keep doing XYZ until things stop changing").
248 for(Element e : all_elements) {
249 if (e instanceof Atom) continue;
250 HashSet<Position> h = new Walk.Closure(null, g.cache).closure(e, hs);
251 State s = all_states.get(h) == null ? new State(h, all_states, all_elements) : all_states.get(h);
252 if (gotoSetNonTerminals.get(e) != null)
253 throw new Error("this should not happen");
254 gotoSetNonTerminals.put(e, s);
257 HashMapBag<Element,Position> move = new HashMapBag<Element,Position>();
258 for(Position p : hs) {
259 Element e = p.element();
260 if (e==null) continue;
261 HashSet<Element> ys = cache.ys.get(e);
263 for(Element y : ys) {
264 HashSet<Position> hp = new HashSet<Position>();
265 p.next().reachable(hp);
270 for(Element y : move) {
271 HashSet<Position> h = move.getAll(y);
272 State s = all_states.get(h) == null ? new State(h, all_states, all_elements) : all_states.get(h);
273 gotoSetNonTerminals.put(y, s);
277 public String toString() { return "state["+idx+"]"; }
279 public int compareTo(Table.State s) { return idx==s.idx ? 0 : idx < s.idx ? -1 : 1; }
283 * the information needed to perform a reduction; copied here to
284 * avoid keeping references to <tt>Element</tt> objects in a Table
286 public class Reduction {
287 // FIXME: cleanup; almost everything in here could go in either Sequence.Position.getRewrite() or else in GSS.Reduct
288 public final int numPop;
289 private final Position position;
290 private final Forest[] holder; // to avoid constant reallocation
291 public int hashCode() { return position.hashCode(); }
292 public boolean equals(Object o) {
293 if (o==null) return false;
294 if (o==this) return true;
295 if (!(o instanceof Reduction)) return false;
296 Reduction r = (Reduction)o;
297 return r.position == position;
299 public Reduction(Position p) {
302 this.holder = new Forest[numPop];
304 public String toString() { return "[reduce " + position + "]"; }
305 public Forest reduce(Forest f, GSS.Phase.Node parent, GSS.Phase.Node onlychild, GSS.Phase target, Forest rex) {
306 holder[numPop-1] = f;
307 return reduce(parent, numPop-2, rex, onlychild, target);
309 public Forest reduce(GSS.Phase.Node parent, GSS.Phase.Node onlychild, GSS.Phase target, Forest rex) {
310 return reduce(parent, numPop-1, rex, onlychild, target);
313 // FIXME: this could be more elegant and/or cleaner and/or somewhere else
314 private Forest reduce(GSS.Phase.Node parent, int pos, Forest rex, GSS.Phase.Node onlychild, GSS.Phase target) {
315 if (pos>=0) holder[pos] = parent.pending();
316 if (pos<=0 && rex==null) {
317 System.arraycopy(holder, 0, position.holder, 0, holder.length);
318 rex = position.rewrite(target.getLocation());
321 if (onlychild != null)
322 reduce(onlychild, pos-1, rex, null, target);
324 for(GSS.Phase.Node child : parent.parents())
325 reduce(child, pos-1, rex, null, target);
327 State state = parent.state.gotoSetNonTerminals.get(position.owner());
329 target.newNode(parent, rex, state, numPop<=0, parent.phase);
336 private static final Forest[] emptyForestArray = new Forest[0];