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, Token.Location loc);
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 Token.Location loc = input.getLocation();
44 GSS.Phase current = gss.new Phase(null, input.next(), loc);
45 current.newNode(null, null, pt.start, true, null);
47 loc = input.getLocation();
48 GSS.Phase next = gss.new Phase(current, input.next(), loc);
50 Forest forest = current.token==null ? null : shiftedToken((T)current.token, loc);
51 current.shift(next, forest);
52 if (current.isDone()) return (Forest<R>)current.finalResult;
53 current.checkFailure();
59 // Exceptions //////////////////////////////////////////////////////////////////////////////
61 public static class Failed extends Exception {
62 private final Token.Location location;
63 private final String message;
64 public Failed() { this("", null); }
65 public Failed(String message, Token.Location loc) { this.location = loc; this.message = message; }
66 public Token.Location getLocation() { return location; }
67 public String toString() { return message + (location==null ? "" : (" at " + location)); }
70 public static class Ambiguous extends RuntimeException {
71 public final Forest ambiguity;
72 public Ambiguous(Forest ambiguity) { this.ambiguity = ambiguity; }
73 public String toString() {
74 StringBuffer sb = new StringBuffer();
75 sb.append("unresolved ambiguity "/*"at " + ambiguity.getLocation() + ":"*/);
76 for(Object result : ambiguity.expand(false))
77 sb.append("\n " + result);
83 // Table //////////////////////////////////////////////////////////////////////////////
85 static class Top extends Union { public Top() { super("0"); } }
87 /** an SLR(1) parse table which may contain conflicts */
90 private final Union start0 = new Top();
91 private final Sequence start0seq;
93 public final Walk.Cache cache = new Walk.Cache();
95 public HashSet<Position> closure() {
96 HashSet<Position> hp = new HashSet<Position>();
97 reachable(start0, hp);
100 public Position firstPosition() { return start0seq.firstp(); }
101 public Position lastPosition() { Position ret = start0seq.firstp(); while(!ret.isLast()) ret = ret.next(); return ret; }
103 private void walk(Element e, HashSet<Element> hs) {
105 if (hs.contains(e)) return;
107 if (e instanceof Atom) return;
108 for(Sequence s : (Union)e) {
110 for(Position p = s.firstp(); p != null; p = p.next())
111 walk(p.element(), hs);
114 public HashSet<Element> walk() {
115 HashSet<Element> ret = new HashSet<Element>();
121 public String toString() {
122 StringBuffer sb = new StringBuffer();
123 for(Element e : walk())
124 if (e instanceof Union)
125 ((Union)e).toString(sb);
126 return sb.toString();
130 /** the start state */
131 public final State start;
133 /** used to generate unique values for State.idx */
134 private int master_state_idx = 0;
136 /** construct a parse table for the given grammar */
137 public Table(Topology top) { this("s", top); }
138 public Table(String startSymbol, Topology top) { this(new Union(startSymbol), top); }
139 public Table(Union u, Topology top) {
140 start0seq = new Sequence.Singleton(u, null, null);
141 start0.add(start0seq);
143 // construct the set of states
144 HashMap<HashSet<Position>,State> all_states = new HashMap<HashSet<Position>,State>();
145 HashSet<Element> all_elements = walk();
146 for(Element e : all_elements)
147 cache.ys.put(e, new Walk.YieldSet(e, cache).walk());
148 this.start = new State(closure(), all_states, all_elements);
150 // for each state, fill in the corresponding "row" of the parse table
151 for(State state : all_states.values())
152 for(Position p : state.hs) {
154 // the Grammar's designated "last position" is the only accepting state
155 if (p==lastPosition())
158 // FIXME: how does right-nullability interact with follow restrictions?
159 // all right-nullable rules get a reduction [Johnstone 2000]
160 if (p.isRightNullable(cache)) {
161 Walk.Follow wf = new Walk.Follow(top.empty(), p.owner(), all_elements, cache);
162 Reduction red = new Reduction(p);
163 state.reductions.put(wf.walk(p.owner()), red);
164 if (wf.includesEof()) state.eofReductions.add(red, true);
167 // if the element following this position is an atom, copy the corresponding
168 // set of rows out of the "master" goto table and into this state's shift table
169 if (p.element() != null && p.element() instanceof Atom)
170 state.shifts.addAll(state.gotoSetTerminals.subset(((Atom)p.element())));
174 /** a single state in the LR table and the transitions possible from it */
175 public class State implements Comparable<Table.State>, Iterable<Position> {
177 public final int idx = master_state_idx++;
178 private final HashSet<Position> hs;
180 private transient HashMap<Element,State> gotoSetNonTerminals = new HashMap<Element,State>();
181 private transient TopologicalBag<Token,State> gotoSetTerminals = new TopologicalBag<Token,State>();
183 private TopologicalBag<Token,Reduction> reductions = new TopologicalBag<Token,Reduction>();
184 private FastSet<Reduction> eofReductions = new FastSet<Reduction>();
185 private TopologicalBag<Token,State> shifts = new TopologicalBag<Token,State>();
186 private boolean accept = false;
188 // Interface Methods //////////////////////////////////////////////////////////////////////////////
190 public boolean canShift(Token t) { return shifts.contains(t); }
191 public Iterable<State> getShifts(Token t) { return shifts.get(t); }
192 public boolean isAccepting() { return accept; }
193 public Iterable<Reduction> getReductions(Token t) { return reductions.get(t); }
194 public Iterable<Reduction> getEofReductions() { return eofReductions; }
195 public Iterator<Position> iterator() { return hs.iterator(); }
197 // Constructor //////////////////////////////////////////////////////////////////////////////
200 * create a new state consisting of all the <tt>Position</tt>s in <tt>hs</tt>
201 * @param hs the set of <tt>Position</tt>s comprising this <tt>State</tt>
202 * @param all_states the set of states already constructed (to avoid recreating states)
203 * @param all_elements the set of all elements (Atom instances need not be included)
205 * In principle these two steps could be merged, but they
206 * are written separately to highlight these two facts:
208 * <li> Non-atom elements either match all-or-nothing, and do not overlap
209 * with each other (at least not in the sense of which element corresponds
210 * to the last reduction performed). Therefore, in order to make sure we
211 * wind up with the smallest number of states and shifts, we wait until
212 * we've figured out all the token-to-position multimappings before creating
215 * <li> In order to be able to run the state-construction algorithm in a single
216 * shot (rather than repeating until no new items appear in any state set),
217 * we need to use the "yields" semantics rather than the "produces" semantics
218 * for non-Atom Elements.
221 public State(HashSet<Position> hs,
222 HashMap<HashSet<Position>,State> all_states,
223 HashSet<Element> all_elements) {
226 // register ourselves in the all_states hash so that no
227 // two states are ever created with an identical position set
228 all_states.put(hs, this);
230 // Step 1a: examine all Position's in this state and compute the mappings from
231 // sets of follow tokens (tokens which could follow this position) to sets
232 // of _new_ positions (positions after shifting). These mappings are
233 // collectively known as the _closure_
235 TopologicalBag<Token,Position> bag0 = new TopologicalBag<Token,Position>();
236 for(Position position : hs) {
237 if (position.isLast() || !(position.element() instanceof Atom)) continue;
238 Atom a = (Atom)position.element();
239 HashSet<Position> hp = new HashSet<Position>();
240 reachable(position.next(), hp);
241 bag0.addAll(a, /*clo.walk()*/hp);
244 // Step 1b: for each _minimal, contiguous_ set of characters having an identical next-position
245 // set, add that character set to the goto table (with the State corresponding to the
246 // computed next-position set).
248 for(Topology<Token> r : bag0) {
249 HashSet<Position> h = new HashSet<Position>();
250 for(Position p : bag0.getAll(r)) h.add(p);
251 gotoSetTerminals.put(r, all_states.get(h) == null ? new State(h, all_states, all_elements) : all_states.get(h));
254 // Step 2: for every non-Atom element (ie every Element which has a corresponding reduction),
255 // compute the closure over every position in this set which is followed by a symbol
256 // which could yield the Element in question.
258 // "yields" [in one or more step] is used instead of "produces" [in exactly one step]
259 // to avoid having to iteratively construct our set of States as shown in most
260 // expositions of the algorithm (ie "keep doing XYZ until things stop changing").
262 for(Element e : all_elements) {
263 if (e instanceof Atom) continue;
264 HashSet<Position> h = new Walk.Closure(null, g.cache).closure(e, hs);
265 State s = all_states.get(h) == null ? new State(h, all_states, all_elements) : all_states.get(h);
266 if (gotoSetNonTerminals.get(e) != null)
267 throw new Error("this should not happen");
268 gotoSetNonTerminals.put(e, s);
271 HashMapBag<Element,Position> move = new HashMapBag<Element,Position>();
272 for(Position p : hs) {
273 Element e = p.element();
274 if (e==null) continue;
275 HashSet<Element> ys = cache.ys.get(e);
277 for(Element y : ys) {
278 HashSet<Position> hp = new HashSet<Position>();
279 reachable(p.next(), hp);
284 for(Element y : move) {
285 HashSet<Position> h = move.getAll(y);
286 State s = all_states.get(h) == null ? new State(h, all_states, all_elements) : all_states.get(h);
287 gotoSetNonTerminals.put(y, s);
291 public String toString() { return "state["+idx+"]"; }
293 public int compareTo(Table.State s) { return idx==s.idx ? 0 : idx < s.idx ? -1 : 1; }
297 * the information needed to perform a reduction; copied here to
298 * avoid keeping references to <tt>Element</tt> objects in a Table
300 public class Reduction {
301 // FIXME: cleanup; almost everything in here could go in either Sequence.Position.getRewrite() or else in GSS.Reduct
302 public final int numPop;
303 private final Position position;
304 private final Forest[] holder; // to avoid constant reallocation
305 public int hashCode() { return position.hashCode(); }
306 public boolean equals(Object o) {
307 if (o==null) return false;
308 if (o==this) return true;
309 if (!(o instanceof Reduction)) return false;
310 Reduction r = (Reduction)o;
311 return r.position == position;
313 public Reduction(Position p) {
316 this.holder = new Forest[numPop];
318 public String toString() { return "[reduce " + position + "]"; }
319 public Forest reduce(Forest f, GSS.Phase.Node parent, GSS.Phase.Node onlychild, GSS.Phase target, Forest rex) {
320 holder[numPop-1] = f;
321 return reduce(parent, numPop-2, rex, onlychild, target);
323 public Forest reduce(GSS.Phase.Node parent, GSS.Phase.Node onlychild, GSS.Phase target, Forest rex) {
324 return reduce(parent, numPop-1, rex, onlychild, target);
327 // FIXME: this could be more elegant and/or cleaner and/or somewhere else
328 private Forest reduce(GSS.Phase.Node parent, int pos, Forest rex, GSS.Phase.Node onlychild, GSS.Phase target) {
329 if (pos>=0) holder[pos] = parent.pending();
330 if (pos<=0 && rex==null) {
331 System.arraycopy(holder, 0, position.holder, 0, holder.length);
332 rex = position.rewrite(target.getLocation());
335 if (onlychild != null)
336 reduce(onlychild, pos-1, rex, null, target);
338 for(GSS.Phase.Node child : parent.parents())
339 reduce(child, pos-1, rex, null, target);
341 State state = parent.state.gotoSetNonTerminals.get(position.owner());
343 target.newNode(parent, rex, state, numPop<=0, parent.phase);
350 private static final Forest[] emptyForestArray = new Forest[0];