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>, 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.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 /** used to generate unique values for State.idx */
100 private int master_state_idx = 0;
101 HashMap<HashSet<Position>,State<Tok>> all_states = new HashMap<HashSet<Position>,State<Tok>>();
103 /** construct a parse table for the given grammar */
104 public Table(Topology top) { this("s", top); }
105 public Table(String startSymbol, Topology top) { this(new Union(startSymbol), top); }
106 public Table(Union ux, Topology top) {
107 Union start0 = new Union("0");
108 start0.add(new Sequence.Singleton(ux));
110 for(Sequence s : start0) cache.eof.put(s, true);
111 cache.eof.put(start0, true);
113 // construct the set of states
114 HashSet<Element> all_elements = new HashSet<Element>();
115 walk(start0, all_elements);
116 for(Element e : all_elements)
117 cache.ys.addAll(e, new Walk.YieldSet(e, cache).walk());
118 HashSet<Position> hp = new HashSet<Position>();
119 reachable(start0, hp);
120 this.start = new State<Tok>(hp, all_states, all_elements);
122 // for each state, fill in the corresponding "row" of the parse table
123 for(State<Tok> state : all_states.values())
124 for(Position p : state.hs) {
126 // the Grammar's designated "last position" is the only accepting state
127 if (start0.contains(p.owner()) && p.next()==null)
130 if (isRightNullable(p)) {
131 Walk.Follow wf = new Walk.Follow(top.empty(), p.owner(), all_elements, cache);
132 Topology follow = wf.walk(p.owner());
133 for(Position p2 = p; p2 != null && p2.element() != null; p2 = p2.next())
134 follow = follow.intersect(new Walk.Follow(top.empty(), p2.element(), all_elements, cache).walk(p2.element()));
135 state.reductions.put(follow, p);
136 if (wf.includesEof()) state.eofReductions.add(p);
139 // if the element following this position is an atom, copy the corresponding
140 // set of rows out of the "master" goto table and into this state's shift table
141 if (p.element() != null && p.element() instanceof Atom)
142 state.shifts.addAll(state.gotoSetTerminals.subset(((Atom)p.element())));
144 if (top instanceof IntegerTopology)
145 for(State<Tok> state : all_states.values()) {
146 state.oreductions = state.reductions.optimize(((IntegerTopology)top).functor());
147 state.oshifts = state.shifts.optimize(((IntegerTopology)top).functor());
151 private boolean isRightNullable(Position p) {
152 if (p.isLast()) return true;
153 if (!possiblyEpsilon(p.element())) return false;
154 return isRightNullable(p.next());
157 /** a single state in the LR table and the transitions possible from it */
159 class State<Tok> implements Comparable<State<Tok>>, IntegerMappable, Iterable<Position> {
161 public final int idx = master_state_idx++;
162 private final HashSet<Position> hs;
164 public transient HashMap<Element,State<Tok>> gotoSetNonTerminals = new HashMap<Element,State<Tok>>();
165 private transient TopologicalBag<Tok,State<Tok>> gotoSetTerminals = new TopologicalBag<Tok,State<Tok>>();
167 private TopologicalBag<Tok,Position> reductions = new TopologicalBag<Tok,Position>();
168 private HashSet<Position> eofReductions = new HashSet<Position>();
169 private TopologicalBag<Tok,State<Tok>> shifts = new TopologicalBag<Tok,State<Tok>>();
170 private boolean accept = false;
172 private VisitableMap<Tok,State<Tok>> oshifts = null;
173 private VisitableMap<Tok,Position> oreductions = null;
175 // Interface Methods //////////////////////////////////////////////////////////////////////////////
177 boolean isAccepting() { return accept; }
178 public Iterator<Position> iterator() { return hs.iterator(); }
180 boolean canShift(Tok t) { return oshifts.contains(t); }
181 <B,C> void invokeShifts(Tok t, Invokable<State<Tok>,B,C> irbc, B b, C c) {
182 oshifts.invoke(t, irbc, b, c);
185 boolean canReduce(Tok t) { return t==null ? eofReductions.size()>0 : oreductions.contains(t); }
186 <B,C> void invokeReductions(Tok t, Invokable<Position,B,C> irbc, B b, C c) {
187 if (t==null) for(Position r : eofReductions) irbc.invoke(r, b, c);
188 else oreductions.invoke(t, irbc, b, c);
191 // Constructor //////////////////////////////////////////////////////////////////////////////
194 * create a new state consisting of all the <tt>Position</tt>s in <tt>hs</tt>
195 * @param hs the set of <tt>Position</tt>s comprising this <tt>State</tt>
196 * @param all_states the set of states already constructed (to avoid recreating states)
197 * @param all_elements the set of all elements (Atom instances need not be included)
199 * In principle these two steps could be merged, but they
200 * are written separately to highlight these two facts:
202 * <li> Non-atom elements either match all-or-nothing, and do not overlap
203 * with each other (at least not in the sense of which element corresponds
204 * to the last reduction performed). Therefore, in order to make sure we
205 * wind up with the smallest number of states and shifts, we wait until
206 * we've figured out all the token-to-position multimappings before creating
209 * <li> In order to be able to run the state-construction algorithm in a single
210 * shot (rather than repeating until no new items appear in any state set),
211 * we need to use the "yields" semantics rather than the "produces" semantics
212 * for non-Atom Elements.
215 public State(HashSet<Position> hs,
216 HashMap<HashSet<Position>,State<Tok>> all_states,
217 HashSet<Element> all_elements) {
220 // register ourselves in the all_states hash so that no
221 // two states are ever created with an identical position set
222 all_states.put(hs, this);
224 // Step 1a: examine all Position's in this state and compute the mappings from
225 // sets of follow tokens (tokens which could follow this position) to sets
226 // of _new_ positions (positions after shifting). These mappings are
227 // collectively known as the _closure_
229 TopologicalBag<Tok,Position> bag0 = new TopologicalBag<Tok,Position>();
230 for(Position position : hs) {
231 if (position.isLast() || !(position.element() instanceof Atom)) continue;
232 Atom a = (Atom)position.element();
233 HashSet<Position> hp = new HashSet<Position>();
234 reachable(position.next(), hp);
238 // Step 1b: for each _minimal, contiguous_ set of characters having an identical next-position
239 // set, add that character set to the goto table (with the State corresponding to the
240 // computed next-position set).
242 for(Topology<Tok> r : bag0) {
243 HashSet<Position> h = new HashSet<Position>();
244 for(Position p : bag0.getAll(r)) h.add(p);
245 gotoSetTerminals.put(r, all_states.get(h) == null ? new State<Tok>(h, all_states, all_elements) : all_states.get(h));
248 // Step 2: for every non-Atom element (ie every Element which has a corresponding reduction),
249 // compute the closure over every position in this set which is followed by a symbol
250 // which could yield the Element in question.
252 // "yields" [in one or more step] is used instead of "produces" [in exactly one step]
253 // to avoid having to iteratively construct our set of States as shown in most
254 // expositions of the algorithm (ie "keep doing XYZ until things stop changing").
255 HashMapBag<Element,Position> move = new HashMapBag<Element,Position>();
256 for(Position p : hs) {
257 Element e = p.element();
258 if (e==null) continue;
259 for(Element y : cache.ys.getAll(e)) {
260 HashSet<Position> hp = new HashSet<Position>();
261 reachable(p.next(), hp);
265 for(Element y : move) {
266 HashSet<Position> h = move.getAll(y);
267 State<Tok> s = all_states.get(h) == null ? new State<Tok>(h, all_states, all_elements) : all_states.get(h);
268 gotoSetNonTerminals.put(y, s);
272 public String toStringx() {
273 StringBuffer st = new StringBuffer();
274 for(Position p : this) {
275 if (st.length() > 0) st.append("\n");
278 return st.toString();
280 public String toString() {
281 StringBuffer ret = new StringBuffer();
282 ret.append("state["+idx+"]: ");
283 for(Position p : this) ret.append("{"+p+"} ");
284 return ret.toString();
287 public int compareTo(State<Tok> s) { return idx==s.idx ? 0 : idx < s.idx ? -1 : 1; }
288 public int toInt() { return idx; }
292 // Helpers //////////////////////////////////////////////////////////////////////////////
294 private static void reachable(Sequence s, HashSet<Position> h) {
295 reachable(s.firstp(), h);
296 for(Sequence ss : s.needs()) reachable(ss, h);
297 for(Sequence ss : s.hates()) reachable(ss, h);
299 private static void reachable(Element e, HashSet<Position> h) {
300 if (e instanceof Atom) return;
301 for(Sequence s : ((Union)e))
304 private static void reachable(Position p, HashSet<Position> h) {
305 if (h.contains(p)) return;
307 if (p.element() != null) reachable(p.element(), h);