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 { return parse(input).expand1(); }
27 /** parse <tt>input</tt>, using the table <tt>pt</tt> to drive the parser */
28 public Forest<R> parse(Token.Stream<T> input) throws IOException, Failed {
30 Token.Location loc = input.getLocation();
31 GSS.Phase current = gss.new Phase(null, input.next(), loc);
32 current.newNode(null, null, pt.start, true, null);
34 loc = input.getLocation();
35 GSS.Phase next = gss.new Phase(current, input.next(), loc);
37 Forest forest = current.token==null ? null : shiftedToken((T)current.token, loc);
38 current.shift(next, forest);
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)); }
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 */
73 static class Table extends Walk.Cache {
75 public final Walk.Cache cache = this;
77 private void walk(Element e, HashSet<Element> hs) {
79 if (hs.contains(e)) return;
81 if (e instanceof Atom) return;
82 for(Sequence s : (Union)e) {
84 for(Position p = s.firstp(); p != null; p = p.next())
85 walk(p.element(), hs);
89 /** the start state */
90 public final State start;
92 /** used to generate unique values for State.idx */
93 private int master_state_idx = 0;
95 /** construct a parse table for the given grammar */
96 public Table(Topology top) { this("s", top); }
97 public Table(String startSymbol, Topology top) { this(new Union(startSymbol), top); }
98 public Table(Union ux, Topology top) {
99 Union start0 = new Union("0");
100 start0.add(new Sequence.Singleton(ux, null, null));
102 for(Sequence s : start0) cache.eof.put(s, true);
103 cache.eof.put(start0, true);
105 // construct the set of states
106 HashMap<HashSet<Position>,State> all_states = new HashMap<HashSet<Position>,State>();
107 HashSet<Element> all_elements = new HashSet<Element>();
108 walk(start0, all_elements);
109 for(Element e : all_elements)
110 cache.ys.put(e, new Walk.YieldSet(e, cache).walk());
111 HashSet<Position> hp = new HashSet<Position>();
112 reachable(start0, hp);
113 this.start = new State(hp, all_states, all_elements);
115 // for each state, fill in the corresponding "row" of the parse table
116 for(State state : all_states.values())
117 for(Position p : state.hs) {
119 // the Grammar's designated "last position" is the only accepting state
120 if (start0.contains(p.owner()) && p.next()==null)
123 // FIXME: how does right-nullability interact with follow restrictions?
124 // all right-nullable rules get a reduction [Johnstone 2000]
125 if (p.isRightNullable(cache)) {
126 Walk.Follow wf = new Walk.Follow(top.empty(), p.owner(), all_elements, cache);
127 Reduction red = new Reduction(p);
128 state.reductions.put(wf.walk(p.owner()), red);
129 if (wf.includesEof()) state.eofReductions.add(red, true);
132 // if the element following this position is an atom, copy the corresponding
133 // set of rows out of the "master" goto table and into this state's shift table
134 if (p.element() != null && p.element() instanceof Atom)
135 state.shifts.addAll(state.gotoSetTerminals.subset(((Atom)p.element())));
139 /** a single state in the LR table and the transitions possible from it */
140 public class State implements Comparable<Table.State>, Iterable<Position> {
142 public final int idx = master_state_idx++;
143 private final HashSet<Position> hs;
145 private transient HashMap<Element,State> gotoSetNonTerminals = new HashMap<Element,State>();
146 private transient TopologicalBag<Token,State> gotoSetTerminals = new TopologicalBag<Token,State>();
148 private TopologicalBag<Token,Reduction> reductions = new TopologicalBag<Token,Reduction>();
149 private FastSet<Reduction> eofReductions = new FastSet<Reduction>();
150 private TopologicalBag<Token,State> shifts = new TopologicalBag<Token,State>();
151 private boolean accept = false;
153 // Interface Methods //////////////////////////////////////////////////////////////////////////////
155 public boolean canShift(Token t) { return shifts.contains(t); }
156 public Iterable<State> getShifts(Token t) { return shifts.get(t); }
157 public boolean isAccepting() { return accept; }
158 public Iterable<Reduction> getReductions(Token t) { return reductions.get(t); }
159 public Iterable<Reduction> getEofReductions() { return eofReductions; }
160 public Iterator<Position> iterator() { return hs.iterator(); }
162 // Constructor //////////////////////////////////////////////////////////////////////////////
165 * create a new state consisting of all the <tt>Position</tt>s in <tt>hs</tt>
166 * @param hs the set of <tt>Position</tt>s comprising this <tt>State</tt>
167 * @param all_states the set of states already constructed (to avoid recreating states)
168 * @param all_elements the set of all elements (Atom instances need not be included)
170 * In principle these two steps could be merged, but they
171 * are written separately to highlight these two facts:
173 * <li> Non-atom elements either match all-or-nothing, and do not overlap
174 * with each other (at least not in the sense of which element corresponds
175 * to the last reduction performed). Therefore, in order to make sure we
176 * wind up with the smallest number of states and shifts, we wait until
177 * we've figured out all the token-to-position multimappings before creating
180 * <li> In order to be able to run the state-construction algorithm in a single
181 * shot (rather than repeating until no new items appear in any state set),
182 * we need to use the "yields" semantics rather than the "produces" semantics
183 * for non-Atom Elements.
186 public State(HashSet<Position> hs,
187 HashMap<HashSet<Position>,State> all_states,
188 HashSet<Element> all_elements) {
191 // register ourselves in the all_states hash so that no
192 // two states are ever created with an identical position set
193 all_states.put(hs, this);
195 // Step 1a: examine all Position's in this state and compute the mappings from
196 // sets of follow tokens (tokens which could follow this position) to sets
197 // of _new_ positions (positions after shifting). These mappings are
198 // collectively known as the _closure_
200 TopologicalBag<Token,Position> bag0 = new TopologicalBag<Token,Position>();
201 for(Position position : hs) {
202 if (position.isLast() || !(position.element() instanceof Atom)) continue;
203 Atom a = (Atom)position.element();
204 HashSet<Position> hp = new HashSet<Position>();
205 reachable(position.next(), hp);
209 // Step 1b: for each _minimal, contiguous_ set of characters having an identical next-position
210 // set, add that character set to the goto table (with the State corresponding to the
211 // computed next-position set).
213 for(Topology<Token> r : bag0) {
214 HashSet<Position> h = new HashSet<Position>();
215 for(Position p : bag0.getAll(r)) h.add(p);
216 gotoSetTerminals.put(r, all_states.get(h) == null ? new State(h, all_states, all_elements) : all_states.get(h));
219 // Step 2: for every non-Atom element (ie every Element which has a corresponding reduction),
220 // compute the closure over every position in this set which is followed by a symbol
221 // which could yield the Element in question.
223 // "yields" [in one or more step] is used instead of "produces" [in exactly one step]
224 // to avoid having to iteratively construct our set of States as shown in most
225 // expositions of the algorithm (ie "keep doing XYZ until things stop changing").
227 for(Element e : all_elements) {
228 if (e instanceof Atom) continue;
229 HashSet<Position> h = new Walk.Closure(null, g.cache).closure(e, hs);
230 State s = all_states.get(h) == null ? new State(h, all_states, all_elements) : all_states.get(h);
231 if (gotoSetNonTerminals.get(e) != null)
232 throw new Error("this should not happen");
233 gotoSetNonTerminals.put(e, s);
236 HashMapBag<Element,Position> move = new HashMapBag<Element,Position>();
237 for(Position p : hs) {
238 Element e = p.element();
239 if (e==null) continue;
240 HashSet<Element> ys = cache.ys.get(e);
242 for(Element y : ys) {
243 HashSet<Position> hp = new HashSet<Position>();
244 reachable(p.next(), hp);
249 for(Element y : move) {
250 HashSet<Position> h = move.getAll(y);
251 State s = all_states.get(h) == null ? new State(h, all_states, all_elements) : all_states.get(h);
252 gotoSetNonTerminals.put(y, s);
256 public String toString() { return "state["+idx+"]"; }
258 public int compareTo(Table.State s) { return idx==s.idx ? 0 : idx < s.idx ? -1 : 1; }
262 * the information needed to perform a reduction; copied here to
263 * avoid keeping references to <tt>Element</tt> objects in a Table
265 public class Reduction {
266 // FIXME: cleanup; almost everything in here could go in either Sequence.Position.getRewrite() or else in GSS.Reduct
267 public final int numPop;
268 private final Position position;
269 private final Forest[] holder; // to avoid constant reallocation
270 public int hashCode() { return position.hashCode(); }
271 public boolean equals(Object o) {
272 if (o==null) return false;
273 if (o==this) return true;
274 if (!(o instanceof Reduction)) return false;
275 Reduction r = (Reduction)o;
276 return r.position == position;
278 public Reduction(Position p) {
281 this.holder = new Forest[numPop];
283 public String toString() { return "[reduce " + position + "]"; }
284 public Forest reduce(Forest f, GSS.Phase.Node parent, GSS.Phase.Node onlychild, GSS.Phase target, Forest rex) {
285 holder[numPop-1] = f;
286 return reduce(parent, numPop-2, rex, onlychild, target);
288 public Forest reduce(GSS.Phase.Node parent, GSS.Phase.Node onlychild, GSS.Phase target, Forest rex) {
289 return reduce(parent, numPop-1, rex, onlychild, target);
292 // FIXME: this could be more elegant and/or cleaner and/or somewhere else
293 private Forest reduce(GSS.Phase.Node parent, int pos, Forest rex, GSS.Phase.Node onlychild, GSS.Phase target) {
294 if (pos>=0) holder[pos] = parent.pending();
295 if (pos<=0 && rex==null) {
296 System.arraycopy(holder, 0, position.holder, 0, holder.length);
297 rex = position.rewrite(target.getLocation());
300 if (onlychild != null)
301 reduce(onlychild, pos-1, rex, null, target);
303 for(GSS.Phase.Node child : parent.parents())
304 reduce(child, pos-1, rex, null, target);
306 State state = parent.state.gotoSetNonTerminals.get(position.owner());
308 target.newNode(parent, rex, state, numPop<=0, parent.phase);
315 private static final Forest[] emptyForestArray = new Forest[0];
318 // Helpers //////////////////////////////////////////////////////////////////////////////
320 private static void reachable(Element e, HashSet<Position> h) {
321 if (e instanceof Atom) return;
322 for(Sequence s : ((Union)e))
323 reachable(s.firstp(), h);
325 private static void reachable(Position p, HashSet<Position> h) {
326 if (h.contains(p)) return;
328 if (p.element() != null) reachable(p.element(), h);