X-Git-Url: http://git.megacz.com/?p=sbp.git;a=blobdiff_plain;f=src%2Fedu%2Fberkeley%2Fsbp%2FParser.java;h=95e3ba52a024c8e31223bf4271aee5b55d1450dd;hp=64ad67ac9abc46b3790a890dfd9398c4694f9121;hb=c805010980fc22bcd66c1684a772f66563cd6b72;hpb=dc9bb3a45ed306e2e35549076842b3e74efecb48 diff --git a/src/edu/berkeley/sbp/Parser.java b/src/edu/berkeley/sbp/Parser.java index 64ad67a..95e3ba5 100644 --- a/src/edu/berkeley/sbp/Parser.java +++ b/src/edu/berkeley/sbp/Parser.java @@ -1,17 +1,14 @@ -// Copyright 2006 all rights reserved; see LICENSE file for BSD-style license +// Copyright 2006-2007 all rights reserved; see LICENSE file for BSD-style license package edu.berkeley.sbp; import edu.berkeley.sbp.util.*; -import edu.berkeley.sbp.Sequence.Position; +import edu.berkeley.sbp.Sequence.Pos; +import edu.berkeley.sbp.Sequence.Pos; import java.io.*; import java.util.*; -// FEATURE: try harder to "fuse" states together along two dimensions: -// - identical (equivalent) states, or states that subsume each other -// - unnecessary intermediate states ("short cut" GLR) - /** a parser which translates an Input<Token> into a Forest<NodeType> */ -public abstract class Parser { +public abstract class Parser implements Serializable { final Table pt; @@ -24,39 +21,83 @@ public abstract class Parser { public abstract Topology emptyTopology(); public String toString() { return pt.toString(); } - Cache cache() { return pt; } /** parse input, and return the shared packed parse forest (or throw an exception) */ public Forest parse(Input input) throws IOException, ParseFailed { + long start = System.currentTimeMillis(); verbose = System.getProperty("sbp.verbose", null) != null; spinpos = 0; + GSS gss = new GSS(input, this); + int idmax = 0; + int[][] count = new int[1024*1024][]; + HashMap ids = new HashMap(); try { - GSS gss = new GSS(input, this); for(GSS.Phase current = gss.new Phase(pt.start); ;) { - - if (verbose) { - // FIXME: clean this up - String s; - s = " " + spin[spinpos++ % (spin.length)]+" parsing "; - s += input.getName(); - s += " "+input.getLocation(); - while(s.indexOf(':') != -1 && s.indexOf(':') < 8) s = " " + s; - String y = "@"+gss.viewPos+" "; - while(y.length() < 9) y = " " + y; - s += y; - s += " nodes="+gss.numOldNodes; - while(s.length() < 50) s = s + " "; - s += " shifted="+gss.numNewNodes; - while(s.length() < 60) s = s + " "; - s += " reductions="+gss.numReductions; - System.err.print("\r"+s+ANSI.clreol()+"\r"); - } - + if (verbose) debug(current.token, gss, input); if (current.isDone()) return (Forest)current.finalResult; - Forest forest = shiftToken((Token)current.token, current.getRegion()); + Input.Region region = current.getLocation().createRegion(current.getNextLocation()); + Forest forest = shiftToken((Token)current.token, region); + /* + int maxid = 0; + for(Reduction r : gss.finishedReductions) + if (ids.get(r.reduction())==null) + ids.put(r.reduction(), idmax++); + count[current.pos] = new int[idmax]; + for(Reduction r : gss.finishedReductions) + count[current.pos][ids.get(r.reduction())]++; + */ current = gss.new Phase(current, forest); } - } finally { if (verbose) System.err.print("\r"+ANSI.clreol()); } + } finally { + if (verbose) { + long time = System.currentTimeMillis() - start; + System.err.println("\r parse time: " + time +"ms "+ ANSI.clreol()); + debug(null, gss, input); + } + /* + PrintWriter pw = new PrintWriter(new OutputStreamWriter(new FileOutputStream("out.plot"))); + boolean[] use = new boolean[idmax]; + for(int i=0; i20) + use[j] = true; + for(int i=0; i=count[i].length ? 0 : count[i][j])); + } + pw.println(); + } + pw.close(); + pw = new PrintWriter(new OutputStreamWriter(new FileOutputStream("test.plot"))); + pw.println("set terminal postscript enhanced color"); + pw.println("set output \"out.ps\""); + pw.println("set pm3d map"); + pw.println("set autoscale"); + pw.println("set view 0,0"); + pw.println("set ytics (\\"); + int q = -1; + for(int j=0; j { System.err.print("\r " + spin[spinpos++ % (spin.length)]+"\r"); } + private int _last = -1; + private String buf = ""; + private void debug(Object t, GSS gss, Input input) { + //FIXME + int c = t==null ? -1 : ((t+"").charAt(0)); + int last = _last; + _last = c; + switch(c) { + case edu.berkeley.sbp.chr.CharAtom.left: + buf += "\033[31m>\033[0m"; + break; + case edu.berkeley.sbp.chr.CharAtom.right: + buf += "\033[31m<\033[0m"; + break; + case -1: // FIXME + case '\n': + if (verbose) { + if (last==' ') buf += ANSI.blue("\\n"); + System.err.println("\r"+ANSI.clreol()+"\r"+buf); + buf = ""; + } + break; + default: + buf += ANSI.cyan(""+((char)c)); + break; + } + if (t==null) return; + + // FIXME: clean this up + String s; + s = " " + spin[spinpos++ % (spin.length)]+" parsing "; + s += input.getName(); + s += " "+input.getLocation(); + while(s.indexOf(':') != -1 && s.indexOf(':') < 8) s = " " + s; + String y = "@"+gss.viewPos+" "; + while(y.length() < 9) y = " " + y; + s += y; + s += " nodes="+gss.numOldNodes; + while(s.length() < 50) s = s + " "; + s += " shifted="+gss.numNewNodes; + while(s.length() < 60) s = s + " "; + s += " reductions="+gss.numReductions; + while(s.length() < 78) s = s + " "; + System.err.print("\r"+ANSI.invert(s+ANSI.clreol())+"\r"); + } + // Table ////////////////////////////////////////////////////////////////////////////// /** an SLR(1) parse table which may contain conflicts */ - class Table extends Cache { + class Table implements Serializable { /** the start state */ final State start; @@ -88,35 +175,36 @@ public abstract class Parser { private int master_state_idx = 0; /** all the states for this table */ - HashSet> all_states = new HashSet>(); + private transient HashSet> all_states = new HashSet>(); /** all the doomed states in this table */ - HashMap,State> doomed_states = new HashMap,State>(); + private transient HashMap,State> doomed_states = new HashMap,State>(); /** all the non-doomed states in this table */ - HashMap,State> normal_states = new HashMap,State>(); + private transient HashMap,State> normal_states = new HashMap,State>(); - Topology emptyTopology() { return Parser.this.emptyTopology(); } - /** construct a parse table for the given grammar */ Table(Union ux) { - super(new Union("0", Sequence.create(ux), true)); + Union rootUnion = new Union("0", Sequence.create(ux), true); + Grammar grammar = new Grammar(rootUnion) { + public Topology emptyTopology() { return Parser.this.emptyTopology(); } + }; // create the "dead state" - this.dead_state = new State(new HashSet(), true); + this.dead_state = new State(new HashSet(), true, grammar); // construct the start state; this will recursively create *all* the states - this.start = new State(reachable(rootUnion), false); + this.start = new State(reachable(rootUnion), false, grammar); - buildReductions(); - sortReductions(); + buildReductions(grammar); + sortReductions(grammar); } /** fill in the reductions table */ - private void buildReductions() { + private void buildReductions(Grammar grammar) { // for each state, fill in the corresponding "row" of the parse table for(State state : all_states) - for(Position p : state.hs) { + for(Pos p : state.hs) { // if the element following this position is an atom, copy the corresponding // set of rows out of the "master" goto table and into this state's shift table @@ -126,9 +214,9 @@ public abstract class Parser { // RNGLR: we can potentially reduce from any "right-nullable" position -- that is, // any position for which all Elements after it in the Sequence are capable of // matching the empty string. - if (!isRightNullable(p)) continue; - Topology follow = follow(p.owner()); - for(Position p2 = p; p2 != null && p2.element() != null; p2 = p2.next()) { + if (!grammar.isRightNullable(p)) continue; + Topology follow = grammar.follow(p.owner()); + for(Pos p2 = p; p2 != null && p2.element() != null; p2 = p2.next()) { if (!(p2.element() instanceof Union)) throw new Error("impossible -- only Unions can be nullable"); @@ -136,15 +224,15 @@ public abstract class Parser { // not just the follow-set of the last non-nullable element, but the // follow-sets of the nulled elements as well. for(Sequence s : ((Union)p2.element())) - follow = follow.intersect(follow(s)); - Topology set = epsilonFollowSet((Union)p2.element()); + follow = follow.intersect(grammar.follow(s)); + Topology set = grammar.epsilonFollowSet((Union)p2.element()); if (set != null) follow = follow.intersect(set); } // indicate that when the next token is in the set "follow", nodes in this - // state should reduce according to Position "p" + // state should reduce according to Pos "p" state.reductions.put(follow, p); - if (followEof.contains(p.owner())) state.eofReductions.add(p); + if (grammar.followEof.contains(p.owner())) state.eofReductions.add(p); } // optimize the reductions table @@ -157,17 +245,17 @@ public abstract class Parser { } // FIXME: this method needs to be cleaned up and documented - private void sortReductions() { + private void sortReductions(Grammar grammar) { // crude algorithm to assing an ordinal ordering to every position // al will be sorted in DECREASING order (al[0] >= al[1]) - ArrayList al = new ArrayList(); + ArrayList al = new ArrayList(); for(State s : all_states) { - for(Object po : s) { - Sequence.Position p = (Sequence.Position)po; + for(Object po : s.positions()) { + Sequence.Pos p = (Sequence.Pos)po; if (al.contains(p)) continue; int i=0; for(; i { OUTER: while(true) { for(int i=0; i 0) { - Sequence.Position p = al.remove(j); + if (grammar.comparePositions(al.get(i), al.get(j)) > 0) { + Sequence.Pos p = al.remove(j); al.add(i, p); continue OUTER; } @@ -191,7 +279,7 @@ public abstract class Parser { for(int i=0; i 0) + if (grammar.comparePositions(al.get(k), al.get(i)) > 0) { inc = true; break; } } inc = true; @@ -207,13 +295,13 @@ public abstract class Parser { * A single state in the LR table and the transitions * possible from it * - * A state corresponds to a set of Sequence.Position's. Each + * A state corresponds to a set of Sequence.Pos's. Each * Node in the GSS has a State; the Node represents a set of - * possible parses, one for each Position in the State. + * possible parses, one for each Pos in the State. * - * Every state is either "doomed" or "normal". If a Position + * Every state is either "doomed" or "normal". If a Pos * is part of a Sequence which is a conjunct (that is, it was - * passed to Sequence.{and(),andnot()}), then that Position + * passed to Sequence.{and(),andnot()}), then that Pos * will appear only in doomed States. Furthermore, any set * of Positions reachable from a doomed State also forms a * doomed State. Note that in this latter case, a doomed @@ -233,46 +321,49 @@ public abstract class Parser { * space+time complexity in otherwise simple grammars. There * is an example of this in the regression suite. */ - class State implements IntegerMappable, Iterable { + class State implements IntegerMappable, Serializable { public final int idx = master_state_idx++; - private final HashSet hs; + private final transient HashSet hs; public HashSet> conjunctStates = new HashSet>(); - HashMap> gotoSetNonTerminals = new HashMap>(); + HashMap> gotoSetNonTerminals = new HashMap>(); private transient TopologicalBag> gotoSetTerminals = new TopologicalBag>(); - private TopologicalBag reductions = new TopologicalBag(); - private HashSet eofReductions = new HashSet(); + TopologicalBag reductions = new TopologicalBag(); + HashSet eofReductions = new HashSet(); private TopologicalBag> shifts = new TopologicalBag>(); private boolean accept = false; private VisitableMap> oshifts = null; - private VisitableMap oreductions = null; + private VisitableMap oreductions = null; public final boolean doomed; // Interface Methods ////////////////////////////////////////////////////////////////////////////// + public boolean doomed() { return doomed; } boolean isAccepting() { return accept; } - public Iterator iterator() { return hs.iterator(); } + + Iterable positions() { return hs; } + boolean canShift(Token t) { return oshifts!=null && oshifts.contains(t); } - void invokeShifts(Token t, GSS.Phase phase, Result r) { oshifts.invoke(t, phase, r); } + void invokeShifts(Token t, GSS.Phase phase, Node pred, Forest f) { oshifts.invoke(t, phase, pred, f); } boolean canReduce(Token t) { return oreductions != null && (t==null ? eofReductions.size()>0 : oreductions.contains(t)); } void invokeEpsilonReductions(Token t, Node node) { - if (t==null) for(Position r : eofReductions) node.invoke(r, null); - else oreductions.invoke(t, node, null); + if (t==null) for(Pos r : eofReductions) node.invoke(r, null, null); + else oreductions.invoke(t, node, null, null); } void invokeReductions(Token t, Node node, Result b) { - if (t==null) for(Position r : eofReductions) node.invoke(r, b); - else oreductions.invoke(t, node, b); + if (t==null) for(Pos r : eofReductions) node.invoke(r, b, null); + else oreductions.invoke(t, node, b, null); } // Constructor ////////////////////////////////////////////////////////////////////////////// /** - * create a new state consisting of all the Positions in hs - * @param hs the set of Positions comprising this State + * create a new state consisting of all the Poss in hs + * @param hs the set of Poss comprising this State * @param all the set of all elements (Atom instances need not be included) * * In principle these two steps could be merged, but they @@ -291,7 +382,7 @@ public abstract class Parser { * for non-Atom Elements. * */ - public State(HashSet hs, boolean doomed) { + public State(HashSet hs, boolean doomed, Grammar grammar) { this.hs = hs; this.doomed = doomed; @@ -300,35 +391,35 @@ public abstract class Parser { ((HashMap)(doomed ? doomed_states : normal_states)).put(hs, this); ((HashSet)all_states).add(this); - for(Position p : hs) { + for(Pos p : hs) { // Step 1a: take note if we are an accepting state // (last position of the root Union's sequence) - if (p.next()==null && !doomed && rootUnion.contains(p.owner())) + if (p.next()==null && !doomed && grammar.rootUnion.contains(p.owner())) accept = true; - // Step 1b: If any Position in the set is the first position of its sequence, then this + // Step 1b: If any Pos in the set is the first position of its sequence, then this // state is responsible for spawning the "doomed" states for each of the // Sequence's conjuncts. This obligation is recorded by adding the to-be-spawned // states to conjunctStates. if (!p.isFirst()) continue; for(Sequence s : p.owner().needs()) if (!hs.contains(s.firstp())) - conjunctStates.add(mkstate(reachable(s.firstp()), true)); + conjunctStates.add(mkstate(reachable(s.firstp()), true, grammar)); for(Sequence s : p.owner().hates()) if (!hs.contains(s.firstp())) - conjunctStates.add(mkstate(reachable(s.firstp()), true)); + conjunctStates.add(mkstate(reachable(s.firstp()), true, grammar)); } - // Step 2a: examine all Position's in this state and compute the mappings from + // Step 2a: examine all Pos's in this state and compute the mappings from // sets of follow tokens (tokens which could follow this position) to sets // of _new_ positions (positions after shifting). These mappings are // collectively known as the _closure_ - TopologicalBag bag0 = new TopologicalBag(); - for(Position position : hs) { + TopologicalBag bag0 = new TopologicalBag(); + for(Pos position : hs) { if (position.isLast() || !(position.element() instanceof Atom)) continue; Atom a = (Atom)position.element(); - HashSet hp = new HashSet(); + HashSet hp = new HashSet(); reachable(position.next(), hp); bag0.addAll(a.getTokenTopology(), hp); } @@ -338,9 +429,9 @@ public abstract class Parser { // computed next-position set). for(Topology r : bag0) { - HashSet h = new HashSet(); - for(Position p : bag0.getAll(r)) h.add(p); - ((TopologicalBag)gotoSetTerminals).put(r, mkstate(h, doomed)); + HashSet h = new HashSet(); + for(Pos p : bag0.getAll(r)) h.add(p); + ((TopologicalBag)gotoSetTerminals).put(r, mkstate(h, doomed, grammar)); } // Step 3: for every Sequence, compute the closure over every position in this set which @@ -350,61 +441,69 @@ public abstract class Parser { // to avoid having to iteratively construct our set of States as shown in most // expositions of the algorithm (ie "keep doing XYZ until things stop changing"). - HashMapBag move = new HashMapBag(); - for(Position p : hs) + HashMapBag move = new HashMapBag(); + for(Pos p : hs) if (!p.isLast() && p.element() instanceof Union) for(Sequence s : ((Union)p.element())) { - HashSet hp = new HashSet(); + HashSet hp = new HashSet(); reachable(p.next(), hp); move.addAll(s, hp); } OUTER: for(Sequence y : move) { // if a reduction is "lame", it should wind up in the dead_state after reducing - HashSet h = move.getAll(y); - State s = mkstate(h, doomed); - for(Position p : hs) + HashSet h = move.getAll(y); + State s = mkstate(h, doomed, grammar); + for(Pos p : hs) if (p.element() != null && (p.element() instanceof Union)) for(Sequence seq : ((Union)p.element())) if (seq.needs.contains(y) || seq.hates.contains(y)) { // FIXME: assumption that no sequence is ever both usefully (non-lamely) matched // and also directly lamely matched - ((HashMap)gotoSetNonTerminals).put(y, dead_state); + for(Pos pp = y.firstp(); pp != null; pp = pp.next()) + ((HashMap)gotoSetNonTerminals).put(pp, dead_state); continue OUTER; } - gotoSetNonTerminals.put(y, s); + for(Pos pp = y.firstp(); pp != null; pp = pp.next()) + gotoSetNonTerminals.put(pp, s); } } - private State mkstate(HashSet h, boolean b) { + private State mkstate(HashSet h, boolean b, Grammar grammar) { State ret = (b?doomed_states:normal_states).get(h); - if (ret==null) ret = new State(h,b); + if (ret==null) ret = new State(h,b, grammar); return ret; } public int toInt() { return idx; } + public String toString() { + StringBuffer ret = new StringBuffer(); + for(Pos p : hs) + ret.append(p+"\n"); + return ret.toString(); + } } } // Helpers ////////////////////////////////////////////////////////////////////////////// - private static HashSet reachable(Element e) { - HashSet h = new HashSet(); + private static HashSet reachable(Element e) { + HashSet h = new HashSet(); reachable(e, h); return h; } - private static void reachable(Element e, HashSet h) { + private static void reachable(Element e, HashSet h) { if (e instanceof Atom) return; for(Sequence s : ((Union)e)) reachable(s.firstp(), h); } - private static void reachable(Position p, HashSet h) { + private static void reachable(Pos p, HashSet h) { if (h.contains(p)) return; h.add(p); if (p.element() != null) reachable(p.element(), h); } - private static HashSet reachable(Position p) { - HashSet ret = new HashSet(); + private static HashSet reachable(Pos p) { + HashSet ret = new HashSet(); reachable(p, ret); return ret; }