1 package edu.berkeley.sbp;
2 import edu.berkeley.sbp.*;
3 import edu.berkeley.sbp.util.*;
4 import edu.berkeley.sbp.Sequence.Position;
8 /** a parser which translates streams of Tokens of type T into a Forest<R> */
9 public abstract class Parser<T extends Token, R> {
11 private final Table pt;
13 /** create a parser to parse the grammar with start symbol <tt>u</tt> */
14 protected Parser(Union u) { this.pt = new Table(u, top()); }
15 protected Parser(Table pt) { this.pt = pt; }
17 /** implement this method to create the output forest corresponding to a lone shifted input token */
18 public abstract Forest<R> shiftedToken(T t, Token.Location loc);
20 /** this method must return an empty topology of the input token type */
21 public abstract Topology<T> top();
23 /** parse <tt>input</tt>, using the table <tt>pt</tt> to drive the parser */
24 public Forest<R> parse(Token.Stream<T> input) throws IOException, ParseFailed {
26 Token.Location loc = input.getLocation();
27 GSS.Phase current = gss.new Phase(null, this, null, input.next(1, 0, 0), loc, null);
28 current.newNode(null, Forest.leaf(null, null), pt.start, true);
31 loc = input.getLocation();
33 Forest forest = current.token==null ? null : shiftedToken((T)current.token, loc);
34 GSS.Phase next = gss.new Phase(current, this, current, input.next(count, gss.resets, gss.waits), loc, forest);
35 count = next.hash.size();
36 if (current.isDone()) return (Forest<R>)current.finalResult;
41 // Table //////////////////////////////////////////////////////////////////////////////
43 /** an SLR(1) parse table which may contain conflicts */
44 static class Table extends Walk.Cache {
46 public final Walk.Cache cache = this;
48 private void walk(Element e, HashSet<Element> hs) {
50 if (hs.contains(e)) return;
52 if (e instanceof Atom) return;
53 for(Sequence s : (Union)e) {
55 for(Position p = s.firstp(); p != null; p = p.next())
56 walk(p.element(), hs);
60 /** the start state */
61 public final State start;
63 /** used to generate unique values for State.idx */
64 private int master_state_idx = 0;
66 /** construct a parse table for the given grammar */
67 public Table(Topology top) { this("s", top); }
68 public Table(String startSymbol, Topology top) { this(new Union(startSymbol), top); }
69 public Table(Union ux, Topology top) {
70 Union start0 = new Union("0");
71 start0.add(new Sequence.Singleton(ux, null, null));
73 for(Sequence s : start0) cache.eof.put(s, true);
74 cache.eof.put(start0, true);
76 // construct the set of states
77 HashMap<HashSet<Position>,State> all_states = new HashMap<HashSet<Position>,State>();
78 HashSet<Element> all_elements = new HashSet<Element>();
79 walk(start0, all_elements);
80 for(Element e : all_elements)
81 cache.ys.addAll(e, new Walk.YieldSet(e, cache).walk());
82 HashSet<Position> hp = new HashSet<Position>();
83 reachable(start0, hp);
84 this.start = new State(hp, all_states, all_elements);
86 // for each state, fill in the corresponding "row" of the parse table
87 for(State state : all_states.values())
88 for(Position p : state.hs) {
90 // the Grammar's designated "last position" is the only accepting state
91 if (start0.contains(p.owner()) && p.next()==null)
94 if (isRightNullable(p)) {
95 Walk.Follow wf = new Walk.Follow(top.empty(), p.owner(), all_elements, cache);
96 Reduction red = new Reduction(p);
98 Topology follow = wf.walk(p.owner());
99 if (p.owner() instanceof Sequence.RewritingSequence &&
100 (((Sequence.RewritingSequence)p.owner()).tag+"").equals("emailaddr")) {
101 System.out.println("follow before: " + new edu.berkeley.sbp.misc.CharToken.CharRange(follow));
103 for(Position p2 = p; p2 != null && p2.element() != null; p2 = p2.next())
104 follow = follow.intersect(new Walk.Follow(top.empty(), p2.element(), all_elements, cache).walk(p2.element()));
105 if (p.owner() instanceof Sequence.RewritingSequence &&
106 (((Sequence.RewritingSequence)p.owner()).tag+"").equals("emailaddr")) {
107 System.out.println("follow after: " + new edu.berkeley.sbp.misc.CharToken.CharRange(follow));
109 state.reductions.put(follow, red);
110 if (wf.includesEof()) state.eofReductions.add(red);
113 // if the element following this position is an atom, copy the corresponding
114 // set of rows out of the "master" goto table and into this state's shift table
115 if (p.element() != null && p.element() instanceof Atom)
116 state.shifts.addAll(state.gotoSetTerminals.subset(((Atom)p.element())));
118 for(State state : all_states.values()) {
119 state.oreductions = state.reductions.optimize();
120 state.oshifts = state.shifts.optimize();
124 private boolean isRightNullable(Position p) {
125 if (p.isLast()) return true;
126 if (!p.element().possiblyEpsilon(this)) return false;
127 return isRightNullable(p.next());
130 /** a single state in the LR table and the transitions possible from it */
132 public class State implements Comparable<Table.State>, IntegerMappable, Iterable<Position> {
134 public int toInt() { return idx; }
136 public boolean lame() {
137 for(Position p : this)
138 for(Position p2 = p; p2!=null; p2=p2.next())
139 if (p2.isLast() && !p2.owner().lame)
144 public boolean isResolvable(Token t) {
145 boolean found = false;
146 for(Reduction r : getReductions(t)) {
147 Position p = r.position;
148 if (!p.isRightNullable(cache)) continue;
149 if (p.owner().firstp()==p) continue;
151 // found two items meeting criteria #1
157 if (p.element()==null) continue;
158 Topology first = new Walk.First(top(), cache).walk(p.element());
159 if (first.contains(t))
164 public final int idx = master_state_idx++;
165 private final HashSet<Position> hs;
167 private transient HashMap<Element,State> gotoSetNonTerminals = new HashMap<Element,State>();
168 private transient TopologicalBag<Token,State> gotoSetTerminals = new TopologicalBag<Token,State>();
170 private TopologicalBag<Token,Reduction> reductions = new TopologicalBag<Token,Reduction>();
171 private HashSet<Reduction> eofReductions = new HashSet<Reduction>();
172 private TopologicalBag<Token,State> shifts = new TopologicalBag<Token,State>();
173 private boolean accept = false;
175 private VisitableMap<Token,State> oshifts = null;
176 private VisitableMap<Token,Reduction> oreductions = null;
178 // Interface Methods //////////////////////////////////////////////////////////////////////////////
180 public boolean isAccepting() { return accept; }
182 public boolean canShift(Token t) { return oshifts.contains(t); }
183 public boolean canReduce(Token t) { return t==null ? eofReductions.size()>0 : oreductions.contains(t); }
185 public Iterator<Position> iterator() { return hs.iterator(); }
187 public <B,C> void invokeShifts(Token t, Invokable<State,B,C> irbc, B b, C c) {
188 oshifts.invoke(t, irbc, b, c);
190 public <B,C> void invokeReductions(Token t, Invokable<Reduction,B,C> irbc, B b, C c) {
191 if (t==null) for(Reduction r : eofReductions) irbc.invoke(r, b, c);
192 else oreductions.invoke(t, irbc, b, c);
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);
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").
259 HashMapBag<Element,Position> move = new HashMapBag<Element,Position>();
260 for(Position p : hs) {
261 Element e = p.element();
262 if (e==null) continue;
263 for(Element y : cache.ys.getAll(e)) {
264 HashSet<Position> hp = new HashSet<Position>();
265 reachable(p.next(), hp);
269 for(Element y : move) {
270 HashSet<Position> h = move.getAll(y);
271 State s = all_states.get(h) == null ? new State(h, all_states, all_elements) : all_states.get(h);
272 gotoSetNonTerminals.put(y, s);
276 public String toString() {
277 //return "state["+idx+"]";
278 StringBuffer ret = new StringBuffer();
279 ret.append("state["+idx+"]: ");
280 for(Position p : this) ret.append("{"+p+"} ");
281 return ret.toString();
284 public int compareTo(Table.State s) { return idx==s.idx ? 0 : idx < s.idx ? -1 : 1; }
288 * the information needed to perform a reduction; copied here to
289 * avoid keeping references to <tt>Element</tt> objects in a Table
291 public class Reduction {
292 // FIXME: cleanup; almost everything in here could go in either Sequence.Position.getRewrite() or else in GSS.Reduct
293 public final int numPop;
294 /*private*/ final Position position;
295 private final Forest[] holder; // to avoid constant reallocation
296 public int hashCode() { return position.hashCode(); }
297 public boolean equals(Object o) {
298 if (o==null) return false;
299 if (o==this) return true;
300 if (!(o instanceof Reduction)) return false;
301 Reduction r = (Reduction)o;
302 return r.position == position;
304 public Reduction(Position p) {
307 this.holder = new Forest[numPop];
309 public String toString() { return "[reduce " + position + "]"; }
311 private Forest zero = null;
312 public Forest zero() {
313 if (zero != null) return zero;
314 if (numPop > 0) throw new Error();
315 return zero = position.rewrite(null);
318 public void reduce(GSS.Phase.Node parent) {
319 if (numPop==0) finish(parent, zero(), parent.phase());
320 else reduce(parent, numPop-1, parent.phase());
323 public void reduce(GSS.Phase.Node parent, GSS.Phase.Node onlychild) {
324 if (numPop<=0) throw new Error("called wrong form of reduce()");
326 Forest old = holder[pos];
327 holder[pos] = parent.pending();
329 System.arraycopy(holder, 0, position.holder, 0, holder.length);
330 finish(onlychild, position.rewrite(parent.phase().getLocation()), parent.phase());
332 reduce(onlychild, pos-1, parent.phase());
337 // FIXME: this could be more elegant and/or cleaner and/or somewhere else
338 private void reduce(GSS.Phase.Node parent, int pos, GSS.Phase target) {
339 Forest old = holder[pos];
340 holder[pos] = parent.pending();
342 System.arraycopy(holder, 0, position.holder, 0, holder.length);
343 for(int i=0; i<position.pos; i++) if (position.holder[i]==null) throw new Error("realbad");
344 Forest rex = position.rewrite(target.getLocation());
345 for(GSS.Phase.Node child : parent.parents()) finish(child, rex, target);
347 for(GSS.Phase.Node child : parent.parents()) reduce(child, pos-1, target);
351 private void finish(GSS.Phase.Node parent, Forest result, GSS.Phase target) {
352 State state = parent.state.gotoSetNonTerminals.get(position.owner());
353 if (result==null) throw new Error();
355 target.newNode(parent, result, state, numPop<=0, this);
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);