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<Tok, Result> {
11 protected final Table<Tok> pt;
13 /** create a parser to parse the grammar with start symbol <tt>u</tt> */
14 protected Parser(Union u, Topology<Tok> top) { this.pt = new Table<Tok>(u, top); }
15 protected Parser(Table<Tok> pt) { this.pt = pt; }
17 /** implement this method to create the output forest corresponding to a lone shifted input token */
18 protected abstract Forest<Result> shiftToken(Tok t, Input.Location newloc);
20 boolean helpgc = true;
22 public String toString() { return pt.toString(); }
24 /** parse <tt>input</tt>, and return the shared packed parse forest (or throw an exception) */
25 public Forest<Result> parse(Input<Tok> input) throws IOException, ParseFailed {
27 Input.Location loc = input.getLocation();
28 GSS.Phase current = gss.new Phase<Tok>(null, this, null, input.next(), loc, null);
29 current.newNode(null, Forest.create(null, null, null, false), pt.start, true);
31 for(int idx=0;;idx++) {
32 Input.Location oldloc = loc;
33 loc = input.getLocation();
35 Forest forest = current.token==null ? null : shiftToken((Tok)current.token, loc);
36 GSS.Phase next = gss.new Phase<Tok>(current, this, current, input.next(), loc, forest);
38 FileOutputStream fos = new FileOutputStream("out-"+idx+".dot");
39 PrintWriter p = new PrintWriter(new OutputStreamWriter(fos));
40 GraphViz gv = new GraphViz();
42 ((GSS.Phase.Node)n).toGraphViz(gv);
48 if (current.isDone()) return (Forest<Result>)gss.finalResult;
53 // Table //////////////////////////////////////////////////////////////////////////////
55 /** an SLR(1) parse table which may contain conflicts */
56 static class Table<Tok> extends Walk.Cache {
58 public String toString() {
59 StringBuffer sb = new StringBuffer();
60 sb.append("parse table");
61 for(State<Tok> state : all_states.values()) {
62 sb.append(" " + state + "\n");
63 for(Topology<Tok> t : state.shifts) {
64 sb.append(" shift \""+
65 new edu.berkeley.sbp.chr.CharTopology((IntegerTopology<Character>)t)+"\" => ");
66 for(State st : state.shifts.getAll(t))
67 sb.append(st.idx+" ");
70 for(Topology<Tok> t : state.reductions)
71 sb.append(" reduce \""+
72 new edu.berkeley.sbp.chr.CharTopology((IntegerTopology<Character>)t)+"\" => " +
73 state.reductions.getAll(t) + "\n");
78 public final Walk.Cache cache = this;
80 private void walk(Element e, HashSet<Element> hs) {
82 if (hs.contains(e)) return;
84 if (e instanceof Atom) return;
85 for(Sequence s : (Union)e)
88 private void walk(Sequence s, HashSet<Element> hs) {
90 for(Position p = s.firstp(); p != null; p = p.next())
91 walk(p.element(), hs);
92 for(Sequence ss : s.needs()) walk(ss, hs);
93 for(Sequence ss : s.hates()) walk(ss, hs);
96 /** the start state */
97 public final State<Tok> start;
99 /** the state from which no reductions can be done */
100 private final State<Tok> dead_state;
102 /** used to generate unique values for State.idx */
103 private int master_state_idx = 0;
104 HashMap<HashSet<Position>,State<Tok>> all_states = new HashMap<HashSet<Position>,State<Tok>>();
106 /** construct a parse table for the given grammar */
107 public Table(Topology top) { this("s", top); }
108 public Table(String startSymbol, Topology top) { this(new Union(startSymbol), top); }
109 public Table(Union ux, Topology top) {
110 Union start0 = new Union("0");
111 start0.add(new Sequence.Singleton(ux));
113 for(Sequence s : start0) cache.eof.put(s, true);
114 cache.eof.put(start0, true);
116 // construct the set of states
117 HashSet<Element> all_elements = new HashSet<Element>();
118 walk(start0, all_elements);
119 for(Element e : all_elements)
120 cache.ys.addAll(e, new Walk.YieldSet(e, cache).walk());
121 HashSet<Position> hp = new HashSet<Position>();
122 reachable(start0, hp);
124 this.dead_state = new State<Tok>(new HashSet<Position>(), all_states, all_elements);
125 this.start = new State<Tok>(hp, all_states, all_elements);
127 // for each state, fill in the corresponding "row" of the parse table
128 for(State<Tok> state : all_states.values())
129 for(Position p : state.hs) {
131 // the Grammar's designated "last position" is the only accepting state
132 if (start0.contains(p.owner()) && p.next()==null)
135 if (isRightNullable(p)) {
136 Walk.Follow wf = new Walk.Follow(top.empty(), p.owner(), all_elements, cache);
137 Topology follow = wf.walk(p.owner());
138 for(Position p2 = p; p2 != null && p2.element() != null; p2 = p2.next())
139 follow = follow.intersect(new Walk.Follow(top.empty(), p2.element(), all_elements, cache).walk(p2.element()));
140 state.reductions.put(follow, p);
141 if (wf.includesEof()) state.eofReductions.add(p);
144 // if the element following this position is an atom, copy the corresponding
145 // set of rows out of the "master" goto table and into this state's shift table
146 if (p.element() != null && p.element() instanceof Atom)
147 state.shifts.addAll(state.gotoSetTerminals.subset(((Atom)p.element()).getTokenTopology()));
149 if (top instanceof IntegerTopology)
150 for(State<Tok> state : all_states.values()) {
151 state.oreductions = state.reductions.optimize(((IntegerTopology)top).functor());
152 state.oshifts = state.shifts.optimize(((IntegerTopology)top).functor());
156 private boolean isRightNullable(Position p) {
157 if (p.isLast()) return true;
158 if (!possiblyEpsilon(p.element())) return false;
159 return isRightNullable(p.next());
162 /** a single state in the LR table and the transitions possible from it */
164 class State<Tok> implements Comparable<State<Tok>>, IntegerMappable, Iterable<Position> {
166 public final int idx = master_state_idx++;
167 private final HashSet<Position> hs;
169 public transient HashMap<Element,State<Tok>> gotoSetNonTerminals = new HashMap<Element,State<Tok>>();
170 private transient TopologicalBag<Tok,State<Tok>> gotoSetTerminals = new TopologicalBag<Tok,State<Tok>>();
172 private TopologicalBag<Tok,Position> reductions = new TopologicalBag<Tok,Position>();
173 private HashSet<Position> eofReductions = new HashSet<Position>();
174 private TopologicalBag<Tok,State<Tok>> shifts = new TopologicalBag<Tok,State<Tok>>();
175 private boolean accept = false;
177 private VisitableMap<Tok,State<Tok>> oshifts = null;
178 private VisitableMap<Tok,Position> oreductions = null;
180 // Interface Methods //////////////////////////////////////////////////////////////////////////////
182 boolean isAccepting() { return accept; }
183 public Iterator<Position> iterator() { return hs.iterator(); }
185 boolean canShift(Tok t) { return oshifts!=null && oshifts.contains(t); }
186 <B,C> void invokeShifts(Tok t, Invokable<State<Tok>,B,C> irbc, B b, C c) {
187 oshifts.invoke(t, irbc, b, c);
190 boolean canReduce(Tok t) { return oreductions != null && (t==null ? eofReductions.size()>0 : oreductions.contains(t)); }
191 <B,C> void invokeReductions(Tok t, Invokable<Position,B,C> irbc, B b, C c) {
192 if (t==null) for(Position r : eofReductions) irbc.invoke(r, b, c);
193 else oreductions.invoke(t, irbc, b, c);
196 // Constructor //////////////////////////////////////////////////////////////////////////////
199 * create a new state consisting of all the <tt>Position</tt>s in <tt>hs</tt>
200 * @param hs the set of <tt>Position</tt>s comprising this <tt>State</tt>
201 * @param all_states the set of states already constructed (to avoid recreating states)
202 * @param all_elements the set of all elements (Atom instances need not be included)
204 * In principle these two steps could be merged, but they
205 * are written separately to highlight these two facts:
207 * <li> Non-atom elements either match all-or-nothing, and do not overlap
208 * with each other (at least not in the sense of which element corresponds
209 * to the last reduction performed). Therefore, in order to make sure we
210 * wind up with the smallest number of states and shifts, we wait until
211 * we've figured out all the token-to-position multimappings before creating
214 * <li> In order to be able to run the state-construction algorithm in a single
215 * shot (rather than repeating until no new items appear in any state set),
216 * we need to use the "yields" semantics rather than the "produces" semantics
217 * for non-Atom Elements.
220 public State(HashSet<Position> hs,
221 HashMap<HashSet<Position>,State<Tok>> all_states,
222 HashSet<Element> all_elements) {
225 // register ourselves in the all_states hash so that no
226 // two states are ever created with an identical position set
227 all_states.put(hs, this);
229 // Step 1a: examine all Position's in this state and compute the mappings from
230 // sets of follow tokens (tokens which could follow this position) to sets
231 // of _new_ positions (positions after shifting). These mappings are
232 // collectively known as the _closure_
234 TopologicalBag<Tok,Position> bag0 = new TopologicalBag<Tok,Position>();
235 for(Position position : hs) {
236 if (position.isLast() || !(position.element() instanceof Atom)) continue;
237 Atom a = (Atom)position.element();
238 HashSet<Position> hp = new HashSet<Position>();
239 reachable(position.next(), hp);
240 bag0.addAll(a.getTokenTopology(), hp);
243 // Step 1b: for each _minimal, contiguous_ set of characters having an identical next-position
244 // set, add that character set to the goto table (with the State corresponding to the
245 // computed next-position set).
247 for(Topology<Tok> r : bag0) {
248 HashSet<Position> h = new HashSet<Position>();
249 for(Position p : bag0.getAll(r)) h.add(p);
250 gotoSetTerminals.put(r, all_states.get(h) == null ? new State<Tok>(h, all_states, all_elements) : all_states.get(h));
253 // Step 2: for every non-Atom element (ie every Element which has a corresponding reduction),
254 // compute the closure over every position in this set which is followed by a symbol
255 // which could yield the Element in question.
257 // "yields" [in one or more step] is used instead of "produces" [in exactly one step]
258 // to avoid having to iteratively construct our set of States as shown in most
259 // expositions of the algorithm (ie "keep doing XYZ until things stop changing").
261 HashMapBag<Element,Position> move = new HashMapBag<Element,Position>();
262 for(Position p : hs) {
263 Element e = p.element();
264 if (e==null) continue;
265 for(Element y : cache.ys.getAll(e)) {
266 HashSet<Position> hp = new HashSet<Position>();
267 reachable(p.next(), hp);
271 OUTER: for(Element y : move) {
272 HashSet<Position> h = move.getAll(y);
273 State<Tok> s = all_states.get(h) == null ? new State<Tok>(h, all_states, all_elements) : all_states.get(h);
274 // if a reduction is "lame", it should wind up in the dead_state after reducing
275 if (y instanceof Sequence) {
276 for(Position p : hs) {
277 if (p.element() != null && (p.element() instanceof Union)) {
278 Union u = (Union)p.element();
279 for(Sequence seq : u)
280 if (seq.needs.contains((Sequence)y) || seq.hates.contains((Sequence)y)) {
281 // FIXME: what if there are two "routes" to get to the sequence?
282 ((HashMap)gotoSetNonTerminals).put(y, dead_state);
288 gotoSetNonTerminals.put(y, s);
292 public String toStringx() {
293 StringBuffer st = new StringBuffer();
294 for(Position p : this) {
295 if (st.length() > 0) st.append("\n");
298 return st.toString();
300 public String toString() {
301 StringBuffer ret = new StringBuffer();
302 ret.append("state["+idx+"]: ");
303 for(Position p : this) ret.append("{"+p+"} ");
304 return ret.toString();
307 public int compareTo(State<Tok> s) { return idx==s.idx ? 0 : idx < s.idx ? -1 : 1; }
308 public int toInt() { return idx; }
312 // Helpers //////////////////////////////////////////////////////////////////////////////
314 private static void reachable(Sequence s, HashSet<Position> h) {
315 reachable(s.firstp(), h);
316 for(Sequence ss : s.needs()) reachable(ss, h);
317 for(Sequence ss : s.hates()) reachable(ss, h);
319 private static void reachable(Element e, HashSet<Position> h) {
320 if (e instanceof Atom) return;
321 for(Sequence s : ((Union)e))
324 private static void reachable(Position p, HashSet<Position> h) {
325 if (h.contains(p)) return;
327 if (p.element() != null) reachable(p.element(), h);