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 public abstract Forest<Result> shiftToken(Tok t, Input.Location loc);
20 public boolean helpgc = true;
22 public String toString() { return pt.toString(); }
24 /** parse <tt>input</tt>, using the table <tt>pt</tt> to drive the parser */
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.leaf(null, null, null), pt.start, true);
31 for(int idx=0;;idx++) {
32 loc = input.getLocation();
34 Forest forest = current.token==null ? null : shiftToken((Tok)current.token, loc);
35 GSS.Phase next = gss.new Phase<Tok>(current, this, current, input.next(), loc, forest);
37 FileOutputStream fos = new FileOutputStream("out-"+idx+".dot");
38 PrintWriter p = new PrintWriter(new OutputStreamWriter(fos));
39 GraphViz gv = new GraphViz();
41 ((GSS.Phase.Node)n).toGraphViz(gv);
47 if (current.isDone()) return (Forest<Result>)gss.finalResult;
52 // Table //////////////////////////////////////////////////////////////////////////////
54 /** an SLR(1) parse table which may contain conflicts */
55 static class Table<Tok> extends Walk.Cache {
57 public String toString() {
58 StringBuffer sb = new StringBuffer();
59 sb.append("parse table");
60 for(State<Tok> state : all_states.values()) {
61 sb.append(" " + state + "\n");
62 for(Topology<Tok> t : state.shifts) {
63 sb.append(" shift \""+
64 new edu.berkeley.sbp.chr.CharTopology((IntegerTopology<Character>)t)+"\" => ");
65 for(State st : state.shifts.getAll(t))
66 sb.append(st.idx+" ");
69 for(Topology<Tok> t : state.reductions)
70 sb.append(" reduce \""+
71 new edu.berkeley.sbp.chr.CharTopology((IntegerTopology<Character>)t)+"\" => " +
72 state.reductions.getAll(t) + "\n");
77 public final Walk.Cache cache = this;
79 private void walk(Element e, HashSet<Element> hs) {
81 if (hs.contains(e)) return;
83 if (e instanceof Atom) return;
84 for(Sequence s : (Union)e) {
86 for(Position p = s.firstp(); p != null; p = p.next())
87 walk(p.element(), hs);
91 /** the start state */
92 public final State<Tok> start;
94 /** used to generate unique values for State.idx */
95 private int master_state_idx = 0;
96 HashMap<HashSet<Position>,State<Tok>> all_states = new HashMap<HashSet<Position>,State<Tok>>();
98 /** construct a parse table for the given grammar */
99 public Table(Topology top) { this("s", top); }
100 public Table(String startSymbol, Topology top) { this(new Union(startSymbol), top); }
101 public Table(Union ux, Topology top) {
102 Union start0 = new Union("0");
103 start0.add(new Sequence.Singleton(ux));
105 for(Sequence s : start0) cache.eof.put(s, true);
106 cache.eof.put(start0, true);
108 // construct the set of states
109 HashSet<Element> all_elements = new HashSet<Element>();
110 walk(start0, all_elements);
111 for(Element e : all_elements)
112 cache.ys.addAll(e, new Walk.YieldSet(e, cache).walk());
113 HashSet<Position> hp = new HashSet<Position>();
114 reachable(start0, hp);
115 this.start = new State<Tok>(hp, all_states, all_elements);
117 // for each state, fill in the corresponding "row" of the parse table
118 for(State<Tok> state : all_states.values())
119 for(Position p : state.hs) {
121 // the Grammar's designated "last position" is the only accepting state
122 if (start0.contains(p.owner()) && p.next()==null)
125 if (isRightNullable(p)) {
126 Walk.Follow wf = new Walk.Follow(top.empty(), p.owner(), all_elements, cache);
127 Topology follow = wf.walk(p.owner());
128 for(Position p2 = p; p2 != null && p2.element() != null; p2 = p2.next())
129 follow = follow.intersect(new Walk.Follow(top.empty(), p2.element(), all_elements, cache).walk(p2.element()));
130 state.reductions.put(follow, p);
131 if (wf.includesEof()) state.eofReductions.add(p);
134 // if the element following this position is an atom, copy the corresponding
135 // set of rows out of the "master" goto table and into this state's shift table
136 if (p.element() != null && p.element() instanceof Atom)
137 state.shifts.addAll(state.gotoSetTerminals.subset(((Atom)p.element())));
139 if (top instanceof IntegerTopology)
140 for(State<Tok> state : all_states.values()) {
141 state.oreductions = state.reductions.optimize(((IntegerTopology)top).functor());
142 state.oshifts = state.shifts.optimize(((IntegerTopology)top).functor());
146 private boolean isRightNullable(Position p) {
147 if (p.isLast()) return true;
148 if (!possiblyEpsilon(p.element())) return false;
149 return isRightNullable(p.next());
152 /** a single state in the LR table and the transitions possible from it */
154 class State<Tok> implements Comparable<State<Tok>>, IntegerMappable, Iterable<Position> {
156 public final int idx = master_state_idx++;
157 private final HashSet<Position> hs;
159 public transient HashMap<Element,State<Tok>> gotoSetNonTerminals = new HashMap<Element,State<Tok>>();
160 private transient TopologicalBag<Tok,State<Tok>> gotoSetTerminals = new TopologicalBag<Tok,State<Tok>>();
162 private TopologicalBag<Tok,Position> reductions = new TopologicalBag<Tok,Position>();
163 private HashSet<Position> eofReductions = new HashSet<Position>();
164 private TopologicalBag<Tok,State<Tok>> shifts = new TopologicalBag<Tok,State<Tok>>();
165 private boolean accept = false;
167 private VisitableMap<Tok,State<Tok>> oshifts = null;
168 private VisitableMap<Tok,Position> oreductions = null;
170 // Interface Methods //////////////////////////////////////////////////////////////////////////////
172 boolean isAccepting() { return accept; }
173 public Iterator<Position> iterator() { return hs.iterator(); }
175 boolean canShift(Tok t) { return oshifts.contains(t); }
176 <B,C> void invokeShifts(Tok t, Invokable<State<Tok>,B,C> irbc, B b, C c) {
177 oshifts.invoke(t, irbc, b, c);
180 boolean canReduce(Tok t) { return t==null ? eofReductions.size()>0 : oreductions.contains(t); }
181 <B,C> void invokeReductions(Tok t, Invokable<Position,B,C> irbc, B b, C c) {
182 if (t==null) for(Position r : eofReductions) irbc.invoke(r, b, c);
183 else oreductions.invoke(t, irbc, b, c);
186 // Constructor //////////////////////////////////////////////////////////////////////////////
189 * create a new state consisting of all the <tt>Position</tt>s in <tt>hs</tt>
190 * @param hs the set of <tt>Position</tt>s comprising this <tt>State</tt>
191 * @param all_states the set of states already constructed (to avoid recreating states)
192 * @param all_elements the set of all elements (Atom instances need not be included)
194 * In principle these two steps could be merged, but they
195 * are written separately to highlight these two facts:
197 * <li> Non-atom elements either match all-or-nothing, and do not overlap
198 * with each other (at least not in the sense of which element corresponds
199 * to the last reduction performed). Therefore, in order to make sure we
200 * wind up with the smallest number of states and shifts, we wait until
201 * we've figured out all the token-to-position multimappings before creating
204 * <li> In order to be able to run the state-construction algorithm in a single
205 * shot (rather than repeating until no new items appear in any state set),
206 * we need to use the "yields" semantics rather than the "produces" semantics
207 * for non-Atom Elements.
210 public State(HashSet<Position> hs,
211 HashMap<HashSet<Position>,State<Tok>> all_states,
212 HashSet<Element> all_elements) {
215 // register ourselves in the all_states hash so that no
216 // two states are ever created with an identical position set
217 all_states.put(hs, this);
219 // Step 1a: examine all Position's in this state and compute the mappings from
220 // sets of follow tokens (tokens which could follow this position) to sets
221 // of _new_ positions (positions after shifting). These mappings are
222 // collectively known as the _closure_
224 TopologicalBag<Tok,Position> bag0 = new TopologicalBag<Tok,Position>();
225 for(Position position : hs) {
226 if (position.isLast() || !(position.element() instanceof Atom)) continue;
227 Atom a = (Atom)position.element();
228 HashSet<Position> hp = new HashSet<Position>();
229 reachable(position.next(), hp);
233 // Step 1b: for each _minimal, contiguous_ set of characters having an identical next-position
234 // set, add that character set to the goto table (with the State corresponding to the
235 // computed next-position set).
237 for(Topology<Tok> r : bag0) {
238 HashSet<Position> h = new HashSet<Position>();
239 for(Position p : bag0.getAll(r)) h.add(p);
240 gotoSetTerminals.put(r, all_states.get(h) == null ? new State<Tok>(h, all_states, all_elements) : all_states.get(h));
243 // Step 2: for every non-Atom element (ie every Element which has a corresponding reduction),
244 // compute the closure over every position in this set which is followed by a symbol
245 // which could yield the Element in question.
247 // "yields" [in one or more step] is used instead of "produces" [in exactly one step]
248 // to avoid having to iteratively construct our set of States as shown in most
249 // expositions of the algorithm (ie "keep doing XYZ until things stop changing").
250 HashMapBag<Element,Position> move = new HashMapBag<Element,Position>();
251 for(Position p : hs) {
252 Element e = p.element();
253 if (e==null) continue;
254 for(Element y : cache.ys.getAll(e)) {
255 HashSet<Position> hp = new HashSet<Position>();
256 reachable(p.next(), hp);
260 for(Element y : move) {
261 HashSet<Position> h = move.getAll(y);
262 State<Tok> s = all_states.get(h) == null ? new State<Tok>(h, all_states, all_elements) : all_states.get(h);
263 gotoSetNonTerminals.put(y, s);
267 public String toStringx() {
268 StringBuffer st = new StringBuffer();
269 for(Position p : this) {
270 if (st.length() > 0) st.append("\n");
273 return st.toString();
275 public String toString() {
276 StringBuffer ret = new StringBuffer();
277 ret.append("state["+idx+"]: ");
278 for(Position p : this) ret.append("{"+p+"} ");
279 return ret.toString();
282 public int compareTo(State<Tok> s) { return idx==s.idx ? 0 : idx < s.idx ? -1 : 1; }
283 public int toInt() { return idx; }
287 // Helpers //////////////////////////////////////////////////////////////////////////////
289 private static void reachable(Element e, HashSet<Position> h) {
290 if (e instanceof Atom) return;
291 for(Sequence s : ((Union)e))
292 reachable(s.firstp(), h);
294 private static void reachable(Position p, HashSet<Position> h) {
295 if (h.contains(p)) return;
297 if (p.element() != null) reachable(p.element(), h);