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);
36 if (current.isDone()) return (Forest<R>)gss.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 Topology follow = wf.walk(p.owner());
97 for(Position p2 = p; p2 != null && p2.element() != null; p2 = p2.next())
98 follow = follow.intersect(new Walk.Follow(top.empty(), p2.element(), all_elements, cache).walk(p2.element()));
99 state.reductions.put(follow, p);
100 if (wf.includesEof()) state.eofReductions.add(p);
103 // if the element following this position is an atom, copy the corresponding
104 // set of rows out of the "master" goto table and into this state's shift table
105 if (p.element() != null && p.element() instanceof Atom)
106 state.shifts.addAll(state.gotoSetTerminals.subset(((Atom)p.element())));
108 for(State state : all_states.values()) {
109 state.oreductions = state.reductions.optimize();
110 state.oshifts = state.shifts.optimize();
114 private boolean isRightNullable(Position p) {
115 if (p.isLast()) return true;
116 if (!p.element().possiblyEpsilon(this)) return false;
117 return isRightNullable(p.next());
120 /** a single state in the LR table and the transitions possible from it */
122 public class State implements Comparable<Table.State>, IntegerMappable, Iterable<Position> {
124 public final int idx = master_state_idx++;
125 private final HashSet<Position> hs;
127 public transient HashMap<Element,State> gotoSetNonTerminals = new HashMap<Element,State>();
128 private transient TopologicalBag<Token,State> gotoSetTerminals = new TopologicalBag<Token,State>();
130 private TopologicalBag<Token,Position> reductions = new TopologicalBag<Token,Position>();
131 private HashSet<Position> eofReductions = new HashSet<Position>();
132 private TopologicalBag<Token,State> shifts = new TopologicalBag<Token,State>();
133 private boolean accept = false;
135 private VisitableMap<Token,State> oshifts = null;
136 private VisitableMap<Token,Position> oreductions = null;
138 // Interface Methods //////////////////////////////////////////////////////////////////////////////
140 boolean isAccepting() { return accept; }
141 public Iterator<Position> iterator() { return hs.iterator(); }
143 boolean canShift(Token t) { return oshifts.contains(t); }
144 <B,C> void invokeShifts(Token t, Invokable<State,B,C> irbc, B b, C c) {
145 oshifts.invoke(t, irbc, b, c);
148 boolean canReduce(Token t) { return t==null ? eofReductions.size()>0 : oreductions.contains(t); }
149 <B,C> void invokeReductions(Token t, Invokable<Position,B,C> irbc, B b, C c) {
150 if (t==null) for(Position r : eofReductions) irbc.invoke(r, b, c);
151 else oreductions.invoke(t, irbc, b, c);
154 // Constructor //////////////////////////////////////////////////////////////////////////////
157 * create a new state consisting of all the <tt>Position</tt>s in <tt>hs</tt>
158 * @param hs the set of <tt>Position</tt>s comprising this <tt>State</tt>
159 * @param all_states the set of states already constructed (to avoid recreating states)
160 * @param all_elements the set of all elements (Atom instances need not be included)
162 * In principle these two steps could be merged, but they
163 * are written separately to highlight these two facts:
165 * <li> Non-atom elements either match all-or-nothing, and do not overlap
166 * with each other (at least not in the sense of which element corresponds
167 * to the last reduction performed). Therefore, in order to make sure we
168 * wind up with the smallest number of states and shifts, we wait until
169 * we've figured out all the token-to-position multimappings before creating
172 * <li> In order to be able to run the state-construction algorithm in a single
173 * shot (rather than repeating until no new items appear in any state set),
174 * we need to use the "yields" semantics rather than the "produces" semantics
175 * for non-Atom Elements.
178 public State(HashSet<Position> hs,
179 HashMap<HashSet<Position>,State> all_states,
180 HashSet<Element> all_elements) {
183 // register ourselves in the all_states hash so that no
184 // two states are ever created with an identical position set
185 all_states.put(hs, this);
187 // Step 1a: examine all Position's in this state and compute the mappings from
188 // sets of follow tokens (tokens which could follow this position) to sets
189 // of _new_ positions (positions after shifting). These mappings are
190 // collectively known as the _closure_
192 TopologicalBag<Token,Position> bag0 = new TopologicalBag<Token,Position>();
193 for(Position position : hs) {
194 if (position.isLast() || !(position.element() instanceof Atom)) continue;
195 Atom a = (Atom)position.element();
196 HashSet<Position> hp = new HashSet<Position>();
197 reachable(position.next(), hp);
201 // Step 1b: for each _minimal, contiguous_ set of characters having an identical next-position
202 // set, add that character set to the goto table (with the State corresponding to the
203 // computed next-position set).
205 for(Topology<Token> r : bag0) {
206 HashSet<Position> h = new HashSet<Position>();
207 for(Position p : bag0.getAll(r)) h.add(p);
208 gotoSetTerminals.put(r, all_states.get(h) == null ? new State(h, all_states, all_elements) : all_states.get(h));
211 // Step 2: for every non-Atom element (ie every Element which has a corresponding reduction),
212 // compute the closure over every position in this set which is followed by a symbol
213 // which could yield the Element in question.
215 // "yields" [in one or more step] is used instead of "produces" [in exactly one step]
216 // to avoid having to iteratively construct our set of States as shown in most
217 // expositions of the algorithm (ie "keep doing XYZ until things stop changing").
218 HashMapBag<Element,Position> move = new HashMapBag<Element,Position>();
219 for(Position p : hs) {
220 Element e = p.element();
221 if (e==null) continue;
222 for(Element y : cache.ys.getAll(e)) {
223 HashSet<Position> hp = new HashSet<Position>();
224 reachable(p.next(), hp);
228 for(Element y : move) {
229 HashSet<Position> h = move.getAll(y);
230 State s = all_states.get(h) == null ? new State(h, all_states, all_elements) : all_states.get(h);
231 gotoSetNonTerminals.put(y, s);
235 public String toString() {
236 StringBuffer ret = new StringBuffer();
237 ret.append("state["+idx+"]: ");
238 for(Position p : this) ret.append("{"+p+"} ");
239 return ret.toString();
242 public int compareTo(Table.State s) { return idx==s.idx ? 0 : idx < s.idx ? -1 : 1; }
243 public int toInt() { return idx; }
247 private static final Forest[] emptyForestArray = new Forest[0];
250 // Helpers //////////////////////////////////////////////////////////////////////////////
252 private static void reachable(Element e, HashSet<Position> h) {
253 if (e instanceof Atom) return;
254 for(Sequence s : ((Union)e))
255 reachable(s.firstp(), h);
257 private static void reachable(Position p, HashSet<Position> h) {
258 if (h.contains(p)) return;
260 if (p.element() != null) reachable(p.element(), h);