* [The "BSD license"]

*  Copyright (c) 2012 Terence Parr

*  Copyright (c) 2012 Sam Harwell

*  All rights reserved.

*

*  Redistribution and use in source and binary forms, with or without

*  modification, are permitted provided that the following conditions

*  are met:

*

*  1. Redistributions of source code must retain the above copyright

*     notice, this list of conditions and the following disclaimer.

*  2. Redistributions in binary form must reproduce the above copyright

*     notice, this list of conditions and the following disclaimer in the

*     documentation and/or other materials provided with the distribution.

*  3. The name of the author may not be used to endorse or promote products

*     derived from this software without specific prior written permission.

*

*  THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR

*  IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES

*  OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.

*  IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,

*  INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT

*  NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,

*  DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY

*  THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT

*  (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF

*  THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

*/

* The embodiment of the adaptive LL(*), ALL(*), parsing strategy.

*

* <p>

* The basic complexity of the adaptive strategy makes it harder to understand.

* We begin with ATN simulation to build paths in a DFA. Subsequent prediction

* requests go through the DFA first. If they reach a state without an edge for

* the current symbol, the algorithm fails over to the ATN simulation to

* complete the DFA path for the current input (until it finds a conflict state

* or uniquely predicting state).</p>

*

* <p>

* All of that is done without using the outer context because we want to create

* a DFA that is not dependent upon the rule invocation stack when we do a

* prediction. One DFA works in all contexts. We avoid using context not

* necessarily because it's slower, although it can be, but because of the DFA

* caching problem. The closure routine only considers the rule invocation stack

* created during prediction beginning in the decision rule. For example, if

* prediction occurs without invoking another rule's ATN, there are no context

* stacks in the configurations. When lack of context leads to a conflict, we

* don't know if it's an ambiguity or a weakness in the strong LL(*) parsing

* strategy (versus full LL(*)).</p>

*

* <p>

* When SLL yields a configuration set with conflict, we rewind the input and

* retry the ATN simulation, this time using full outer context without adding

* to the DFA. Configuration context stacks will be the full invocation stacks

* from the start rule. If we get a conflict using full context, then we can

* definitively say we have a true ambiguity for that input sequence. If we

* don't get a conflict, it implies that the decision is sensitive to the outer

* context. (It is not context-sensitive in the sense of context-sensitive

* grammars.)</p>

*

* <p>

* The next time we reach this DFA state with an SLL conflict, through DFA

* simulation, we will again retry the ATN simulation using full context mode.

* This is slow because we can't save the results and have to "interpret" the

* ATN each time we get that input.</p>

*

* <p>

* <strong>CACHING FULL CONTEXT PREDICTIONS</strong></p>

*

* <p>

* We could cache results from full context to predicted alternative easily and

* that saves a lot of time but doesn't work in presence of predicates. The set

* of visible predicates from the ATN start state changes depending on the

* context, because closure can fall off the end of a rule. I tried to cache

* tuples (stack context, semantic context, predicted alt) but it was slower

* than interpreting and much more complicated. Also required a huge amount of

* memory. The goal is not to create the world's fastest parser anyway. I'd like

* to keep this algorithm simple. By launching multiple threads, we can improve

* the speed of parsing across a large number of files.</p>

*

* <p>

* There is no strict ordering between the amount of input used by SLL vs LL,

* which makes it really hard to build a cache for full context. Let's say that

* we have input A B C that leads to an SLL conflict with full context X. That

* implies that using X we might only use A B but we could also use A B C D to

* resolve conflict. Input A B C D could predict alternative 1 in one position

* in the input and A B C E could predict alternative 2 in another position in

* input. The conflicting SLL configurations could still be non-unique in the

* full context prediction, which would lead us to requiring more input than the

* original A B C.  To make a  prediction cache work, we have to track  the exact

* input  used during the previous prediction. That amounts to a cache that maps

* X to a specific DFA for that context.</p>

*

* <p>

* Something should be done for left-recursive expression predictions. They are

* likely LL(1) + pred eval. Easier to do the whole SLL unless error and retry

* with full LL thing Sam does.</p>

*

* <p>

* <strong>AVOIDING FULL CONTEXT PREDICTION</strong></p>

*

* <p>

* We avoid doing full context retry when the outer context is empty, we did not

* dip into the outer context by falling off the end of the decision state rule,

* or when we force SLL mode.</p>

*

* <p>

* As an example of the not dip into outer context case, consider as super

* constructor calls versus function calls. One grammar might look like

* this:</p>

*

* <pre>

* ctorBody

*   : '{' superCall? stat* '}'

*   ;

* </pre>

*

* <p>

* Or, you might see something like</p>

*

* <pre>

* stat

*   : superCall ';'

*   | expression ';'

*   | ...

*   ;

* </pre>

*

* <p>

* In both cases I believe that no closure operations will dip into the outer

* context. In the first case ctorBody in the worst case will stop at the '}'.

* In the 2nd case it should stop at the ';'. Both cases should stay within the

* entry rule and not dip into the outer context.</p>

*

* <p>

* <strong>PREDICATES</strong></p>

*

* <p>

* Predicates are always evaluated if present in either SLL or LL both. SLL and

* LL simulation deals with predicates differently. SLL collects predicates as

* it performs closure operations like ANTLR v3 did. It delays predicate

* evaluation until it reaches and accept state. This allows us to cache the SLL

* ATN simulation whereas, if we had evaluated predicates on-the-fly during

* closure, the DFA state configuration sets would be different and we couldn't

* build up a suitable DFA.</p>

*

* <p>

* When building a DFA accept state during ATN simulation, we evaluate any

* predicates and return the sole semantically valid alternative. If there is

* more than 1 alternative, we report an ambiguity. If there are 0 alternatives,

* we throw an exception. Alternatives without predicates act like they have

* true predicates. The simple way to think about it is to strip away all

* alternatives with false predicates and choose the minimum alternative that

* remains.</p>

*

* <p>

* When we start in the DFA and reach an accept state that's predicated, we test

* those and return the minimum semantically viable alternative. If no

* alternatives are viable, we throw an exception.</p>

*

* <p>

* During full LL ATN simulation, closure always evaluates predicates and

* on-the-fly. This is crucial to reducing the configuration set size during

* closure. It hits a landmine when parsing with the Java grammar, for example,

* without this on-the-fly evaluation.</p>

*

* <p>

* <strong>SHARING DFA</strong></p>

*

* <p>

* All instances of the same parser share the same decision DFAs through a

* static field. Each instance gets its own ATN simulator but they share the

* same {@link #decisionToDFA} field. They also share a

* {@link PredictionContextCache} object that makes sure that all

* {@link PredictionContext} objects are shared among the DFA states. This makes

* a big size difference.</p>

*

* <p>

* <strong>THREAD SAFETY</strong></p>

*

* <p>

* The {@link ParserATNSimulator} locks on the {@link #decisionToDFA} field when

* it adds a new DFA object to that array. {@link #addDFAEdge}

* locks on the DFA for the current decision when setting the

* {@link DFAState#edges} field. {@link #addDFAState} locks on

* the DFA for the current decision when looking up a DFA state to see if it

* already exists. We must make sure that all requests to add DFA states that

* are equivalent result in the same shared DFA object. This is because lots of

* threads will be trying to update the DFA at once. The

* {@link #addDFAState} method also locks inside the DFA lock

* but this time on the shared context cache when it rebuilds the

* configurations' {@link PredictionContext} objects using cached

* subgraphs/nodes. No other locking occurs, even during DFA simulation. This is

* safe as long as we can guarantee that all threads referencing

* {@code s.edge[t]} get the same physical target {@link DFAState}, or

* {@code null}. Once into the DFA, the DFA simulation does not reference the

* {@link DFA#states} map. It follows the {@link DFAState#edges} field to new

* targets. The DFA simulator will either find {@link DFAState#edges} to be

* {@code null}, to be non-{@code null} and {@code dfa.edges[t]} null, or

* {@code dfa.edges[t]} to be non-null. The

* {@link #addDFAEdge} method could be racing to set the field

* but in either case the DFA simulator works; if {@code null}, and requests ATN

* simulation. It could also race trying to get {@code dfa.edges[t]}, but either

* way it will work because it's not doing a test and set operation.</p>

*

* <p>

* <strong>Starting with SLL then failing to combined SLL/LL (Two-Stage

* Parsing)</strong></p>

*

* <p>

* Sam pointed out that if SLL does not give a syntax error, then there is no

* point in doing full LL, which is slower. We only have to try LL if we get a

* syntax error. For maximum speed, Sam starts the parser set to pure SLL

* mode with the {@link BailErrorStrategy}:</p>

*

* <pre>

* parser.{@link Parser#getInterpreter() getInterpreter()}.{@link #setPredictionMode setPredictionMode}{@code (}{@link PredictionMode#SLL}{@code )};

* parser.{@link Parser#setErrorHandler setErrorHandler}(new {@link BailErrorStrategy}());

* </pre>

*

* <p>

* If it does not get a syntax error, then we're done. If it does get a syntax

* error, we need to retry with the combined SLL/LL strategy.</p>

*

* <p>

* The reason this works is as follows. If there are no SLL conflicts, then the

* grammar is SLL (at least for that input set). If there is an SLL conflict,

* the full LL analysis must yield a set of viable alternatives which is a

* subset of the alternatives reported by SLL. If the LL set is a singleton,

* then the grammar is LL but not SLL. If the LL set is the same size as the SLL

* set, the decision is SLL. If the LL set has size &gt; 1, then that decision

* is truly ambiguous on the current input. If the LL set is smaller, then the

* SLL conflict resolution might choose an alternative that the full LL would

* rule out as a possibility based upon better context information. If that's

* the case, then the SLL parse will definitely get an error because the full LL

* analysis says it's not viable. If SLL conflict resolution chooses an

* alternative within the LL set, them both SLL and LL would choose the same

* alternative because they both choose the minimum of multiple conflicting

* alternatives.</p>

*

* <p>

* Let's say we have a set of SLL conflicting alternatives {@code {1, 2, 3}} and

* a smaller LL set called <em>s</em>. If <em>s</em> is {@code {2, 3}}, then SLL

* parsing will get an error because SLL will pursue alternative 1. If

* <em>s</em> is {@code {1, 2}} or {@code {1, 3}} then both SLL and LL will

* choose the same alternative because alternative one is the minimum of either

* set. If <em>s</em> is {@code {2}} or {@code {3}} then SLL will get a syntax

* error. If <em>s</em> is {@code {1}} then SLL will succeed.</p>

*

* <p>

* Of course, if the input is invalid, then we will get an error for sure in

* both SLL and LL parsing. Erroneous input will therefore require 2 passes over

* the input.</p>

*/

  public static final boolean debug = false;

  public static final boolean debug_list_atn_decisions = false;

  public static final boolean dfa_debug = false;

  public static final boolean retry_debug = false;

  @Nullable

  protected final Parser parser;

  @NotNull

  public final DFA[] decisionToDFA;

  /** SLL, LL, or LL + exact ambig detection? */

  @NotNull

  private PredictionMode mode = PredictionMode.LL;

  /** Each prediction operation uses a cache for merge of prediction contexts.

   *  Don't keep around as it wastes huge amounts of memory. DoubleKeyMap

   *  isn't synchronized but we're ok since two threads shouldn't reuse same

   *  parser/atnsim object because it can only handle one input at a time.

   *  This maps graphs a and b to merged result c. (a,b)&rarr;c. We can avoid

   *  the merge if we ever see a and b again.  Note that (b,a)&rarr;c should

   *  also be examined during cache lookup.

   */

  protected DoubleKeyMap<PredictionContext,PredictionContext,PredictionContext> mergeCache;

  // LAME globals to avoid parameters!!!!! I need these down deep in predTransition

  protected TokenStream _input;

  protected int _startIndex;

  protected ParserRuleContext _outerContext;

  protected DFA _dfa;

  /** Testing only! */

  public ParserATNSimulator(@NotNull ATN atn, @NotNull DFA[] decisionToDFA,

                @NotNull PredictionContextCache sharedContextCache)

  {

    this(null, atn, decisionToDFA, sharedContextCache);

  }

  public ParserATNSimulator(@Nullable Parser parser, @NotNull ATN atn,

                @NotNull DFA[] decisionToDFA,

                @NotNull PredictionContextCache sharedContextCache)

  {

    super(atn, sharedContextCache);

    this.parser = parser;

    this.decisionToDFA = decisionToDFA;

    //    DOTGenerator dot = new DOTGenerator(null);

    //    System.out.println(dot.getDOT(atn.rules.get(0), parser.getRuleNames()));

    //    System.out.println(dot.getDOT(atn.rules.get(1), parser.getRuleNames()));

  }

  @Override

  public void reset() {

  }

  @Override

  public void clearDFA() {

    for (int d = 0; d < decisionToDFA.length; d++) {

      decisionToDFA[d] = new DFA(atn.getDecisionState(d), d);

    }

  }

  public int adaptivePredict(@NotNull TokenStream input, int decision,

                 @Nullable ParserRuleContext outerContext)

  {

    if ( debug || debug_list_atn_decisions )  {

      System.out.println("adaptivePredict decision "+decision+

                   " exec LA(1)=="+ getLookaheadName(input)+

                   " line "+input.LT(1).getLine()+":"+input.LT(1).getCharPositionInLine());

    }

    _input = input;

    _startIndex = input.index();

    _outerContext = outerContext;

    DFA dfa = decisionToDFA[decision];

    _dfa = dfa;

    int m = input.mark();

    int index = _startIndex;

    // Now we are certain to have a specific decision's DFA

    // But, do we still need an initial state?

    try {

      DFAState s0;

      if (dfa.isPrecedenceDfa()) {

        // the start state for a precedence DFA depends on the current

        // parser precedence, and is provided by a DFA method.

        s0 = dfa.getPrecedenceStartState(parser.getPrecedence());

      }

      else {

        // the start state for a "regular" DFA is just s0

        s0 = dfa.s0;

      }

      if (s0 == null) {

        if ( outerContext ==null ) outerContext = ParserRuleContext.EMPTY;

        if ( debug || debug_list_atn_decisions )  {

          System.out.println("predictATN decision "+ dfa.decision+

                     " exec LA(1)=="+ getLookaheadName(input) +

                     ", outerContext="+ outerContext.toString(parser));

        }

        /* If this is not a precedence DFA, we check the ATN start state

         * to determine if this ATN start state is the decision for the

         * closure block that determines whether a precedence rule

         * should continue or complete.

         */

        if (!dfa.isPrecedenceDfa() && dfa.atnStartState instanceof StarLoopEntryState) {

          if (((StarLoopEntryState)dfa.atnStartState).precedenceRuleDecision) {

            dfa.setPrecedenceDfa(true);

          }

        }

        boolean fullCtx = false;

        ATNConfigSet s0_closure =

          computeStartState(dfa.atnStartState,

                    ParserRuleContext.EMPTY,

                    fullCtx);

        if (dfa.isPrecedenceDfa()) {

          /* If this is a precedence DFA, we use applyPrecedenceFilter

           * to convert the computed start state to a precedence start

           * state. We then use DFA.setPrecedenceStartState to set the

           * appropriate start state for the precedence level rather

           * than simply setting DFA.s0.

           */

          s0_closure = applyPrecedenceFilter(s0_closure);

          s0 = addDFAState(dfa, new DFAState(s0_closure));

          dfa.setPrecedenceStartState(parser.getPrecedence(), s0);

        }

        else {

          s0 = addDFAState(dfa, new DFAState(s0_closure));

          dfa.s0 = s0;

        }

      }

      int alt = execATN(dfa, s0, input, index, outerContext);

      if ( debug ) System.out.println("DFA after predictATN: "+ dfa.toString(parser.getVocabulary()));

      return alt;

    }

    finally {

      mergeCache = null; // wack cache after each prediction

      _dfa = null;

      input.seek(index);

      input.release(m);

    }

  }

  /** Performs ATN simulation to compute a predicted alternative based

   *  upon the remaining input, but also updates the DFA cache to avoid

   *  having to traverse the ATN again for the same input sequence.

   There are some key conditions we're looking for after computing a new

   set of ATN configs (proposed DFA state):

         * if the set is empty, there is no viable alternative for current symbol

         * does the state uniquely predict an alternative?

         * does the state have a conflict that would prevent us from

           putting it on the work list?

   We also have some key operations to do:

         * add an edge from previous DFA state to potentially new DFA state, D,

           upon current symbol but only if adding to work list, which means in all

           cases except no viable alternative (and possibly non-greedy decisions?)

         * collecting predicates and adding semantic context to DFA accept states

         * adding rule context to context-sensitive DFA accept states

         * consuming an input symbol

         * reporting a conflict

         * reporting an ambiguity

         * reporting a context sensitivity

         * reporting insufficient predicates

   cover these cases:

      dead end

      single alt

      single alt + preds

      conflict

      conflict + preds

   */

  protected int execATN(@NotNull DFA dfa, @NotNull DFAState s0,

             @NotNull TokenStream input, int startIndex,

             ParserRuleContext outerContext)

  {

    if ( debug || debug_list_atn_decisions) {

      System.out.println("execATN decision "+dfa.decision+

                 " exec LA(1)=="+ getLookaheadName(input)+

                 " line "+input.LT(1).getLine()+":"+input.LT(1).getCharPositionInLine());

    }

    DFAState previousD = s0;

    if ( debug ) System.out.println("s0 = "+s0);

    int t = input.LA(1);

    while (true) { // while more work

      DFAState D = getExistingTargetState(previousD, t);

      if (D == null) {

        D = computeTargetState(dfa, previousD, t);

      }

      if (D == ERROR) {

        // if any configs in previous dipped into outer context, that

        // means that input up to t actually finished entry rule

        // at least for SLL decision. Full LL doesn't dip into outer

        // so don't need special case.

        // We will get an error no matter what so delay until after

        // decision; better error message. Also, no reachable target

        // ATN states in SLL implies LL will also get nowhere.

        // If conflict in states that dip out, choose min since we

        // will get error no matter what.

        NoViableAltException e = noViableAlt(input, outerContext, previousD.configs, startIndex);

        input.seek(startIndex);

        int alt = getSynValidOrSemInvalidAltThatFinishedDecisionEntryRule(previousD.configs, outerContext);

        if ( alt!=ATN.INVALID_ALT_NUMBER ) {

          return alt;

        }

        throw e;

      }

      if ( D.requiresFullContext && mode != PredictionMode.SLL ) {

        // IF PREDS, MIGHT RESOLVE TO SINGLE ALT => SLL (or syntax error)

        BitSet conflictingAlts = D.configs.conflictingAlts;

        if ( D.predicates!=null ) {

          if ( debug ) System.out.println("DFA state has preds in DFA sim LL failover");

          int conflictIndex = input.index();

          if (conflictIndex != startIndex) {

            input.seek(startIndex);

          }

          conflictingAlts = evalSemanticContext(D.predicates, outerContext, true);

          if ( conflictingAlts.cardinality()==1 ) {

            if ( debug ) System.out.println("Full LL avoided");

            return conflictingAlts.nextSetBit(0);

          }

          if (conflictIndex != startIndex) {

            // restore the index so reporting the fallback to full

            // context occurs with the index at the correct spot

            input.seek(conflictIndex);

          }

        }

        if ( dfa_debug ) System.out.println("ctx sensitive state "+outerContext+" in "+D);

        boolean fullCtx = true;

        ATNConfigSet s0_closure =

          computeStartState(dfa.atnStartState, outerContext,

                    fullCtx);

        reportAttemptingFullContext(dfa, conflictingAlts, D.configs, startIndex, input.index());

        int alt = execATNWithFullContext(dfa, D, s0_closure,

                         input, startIndex,

                         outerContext);

        return alt;

      }

      if ( D.isAcceptState ) {

        if (D.predicates == null) {

          return D.prediction;

        }

        int stopIndex = input.index();

        input.seek(startIndex);

        BitSet alts = evalSemanticContext(D.predicates, outerContext, true);

        switch (alts.cardinality()) {

        case 0:

          throw noViableAlt(input, outerContext, D.configs, startIndex);

        case 1:

          return alts.nextSetBit(0);

        default:

          // report ambiguity after predicate evaluation to make sure the correct

          // set of ambig alts is reported.

          reportAmbiguity(dfa, D, startIndex, stopIndex, false, alts, D.configs);

          return alts.nextSetBit(0);

        }

      }

      previousD = D;

      if (t != IntStream.EOF) {

        input.consume();

        t = input.LA(1);

      }

    }

  }

  /**

   * Get an existing target state for an edge in the DFA. If the target state

   * for the edge has not yet been computed or is otherwise not available,

   * this method returns {@code null}.

   *

   * @param previousD The current DFA state

   * @param t The next input symbol

   * @return The existing target DFA state for the given input symbol

   * {@code t}, or {@code null} if the target state for this edge is not

   * already cached

   */

  @Nullable

  protected DFAState getExistingTargetState(@NotNull DFAState previousD, int t) {

    DFAState[] edges = previousD.edges;

    if (edges == null || t + 1 < 0 || t + 1 >= edges.length) {

      return null;

    }

    return edges[t + 1];

  }

  /**

   * Compute a target state for an edge in the DFA, and attempt to add the

   * computed state and corresponding edge to the DFA.

   *

   * @param dfa The DFA

   * @param previousD The current DFA state

   * @param t The next input symbol

   *

   * @return The computed target DFA state for the given input symbol

   * {@code t}. If {@code t} does not lead to a valid DFA state, this method

   * returns {@link #ERROR}.

   */

  @NotNull

  protected DFAState computeTargetState(@NotNull DFA dfa, @NotNull DFAState previousD, int t) {

    ATNConfigSet reach = computeReachSet(previousD.configs, t, false);

    if ( reach==null ) {

      addDFAEdge(dfa, previousD, t, ERROR);

      return ERROR;

    }

    // create new target state; we'll add to DFA after it's complete

    DFAState D = new DFAState(reach);

    int predictedAlt = getUniqueAlt(reach);

    if ( debug ) {

      Collection<BitSet> altSubSets = PredictionMode.getConflictingAltSubsets(reach);

      System.out.println("SLL altSubSets="+altSubSets+

                 ", configs="+reach+

                 ", predict="+predictedAlt+", allSubsetsConflict="+

                   PredictionMode.allSubsetsConflict(altSubSets)+", conflictingAlts="+

                 getConflictingAlts(reach));

    }

    if ( predictedAlt!=ATN.INVALID_ALT_NUMBER ) {

      // NO CONFLICT, UNIQUELY PREDICTED ALT

      D.isAcceptState = true;

      D.configs.uniqueAlt = predictedAlt;

      D.prediction = predictedAlt;

    }

    else if ( PredictionMode.hasSLLConflictTerminatingPrediction(mode, reach) ) {

      // MORE THAN ONE VIABLE ALTERNATIVE

      D.configs.conflictingAlts = getConflictingAlts(reach);

      D.requiresFullContext = true;

      // in SLL-only mode, we will stop at this state and return the minimum alt

      D.isAcceptState = true;

      D.prediction = D.configs.conflictingAlts.nextSetBit(0);

    }

    if ( D.isAcceptState && D.configs.hasSemanticContext ) {

      predicateDFAState(D, atn.getDecisionState(dfa.decision));

      if (D.predicates != null) {

        D.prediction = ATN.INVALID_ALT_NUMBER;

      }

    }

    // all adds to dfa are done after we've created full D state

    D = addDFAEdge(dfa, previousD, t, D);

    return D;

  }

  protected void predicateDFAState(DFAState dfaState, DecisionState decisionState) {

    // We need to test all predicates, even in DFA states that

    // uniquely predict alternative.

    int nalts = decisionState.getNumberOfTransitions();

    // Update DFA so reach becomes accept state with (predicate,alt)

    // pairs if preds found for conflicting alts

    BitSet altsToCollectPredsFrom = getConflictingAltsOrUniqueAlt(dfaState.configs);

    SemanticContext[] altToPred = getPredsForAmbigAlts(altsToCollectPredsFrom, dfaState.configs, nalts);

    if ( altToPred!=null ) {

      dfaState.predicates = getPredicatePredictions(altsToCollectPredsFrom, altToPred);

      dfaState.prediction = ATN.INVALID_ALT_NUMBER; // make sure we use preds

    }

    else {

      // There are preds in configs but they might go away

      // when OR'd together like {p}? || NONE == NONE. If neither

      // alt has preds, resolve to min alt

      dfaState.prediction = altsToCollectPredsFrom.nextSetBit(0);

    }

  }

  // comes back with reach.uniqueAlt set to a valid alt

  protected int execATNWithFullContext(DFA dfa,

                     DFAState D, // how far we got before failing over

                     @NotNull ATNConfigSet s0,

                     @NotNull TokenStream input, int startIndex,

                     ParserRuleContext outerContext)

  {

    if ( debug || debug_list_atn_decisions ) {

      System.out.println("execATNWithFullContext "+s0);

    }

    boolean fullCtx = true;

    boolean foundExactAmbig = false;

    ATNConfigSet reach = null;

    ATNConfigSet previous = s0;

    input.seek(startIndex);

    int t = input.LA(1);

    int predictedAlt;

    while (true) { // while more work

//      System.out.println("LL REACH "+getLookaheadName(input)+

//                 " from configs.size="+previous.size()+

//                 " line "+input.LT(1).getLine()+":"+input.LT(1).getCharPositionInLine());

      reach = computeReachSet(previous, t, fullCtx);

      if ( reach==null ) {

        // if any configs in previous dipped into outer context, that

        // means that input up to t actually finished entry rule

        // at least for LL decision. Full LL doesn't dip into outer

        // so don't need special case.

        // We will get an error no matter what so delay until after

        // decision; better error message. Also, no reachable target

        // ATN states in SLL implies LL will also get nowhere.

        // If conflict in states that dip out, choose min since we

        // will get error no matter what.

        NoViableAltException e = noViableAlt(input, outerContext, previous, startIndex);

        input.seek(startIndex);

        int alt = getSynValidOrSemInvalidAltThatFinishedDecisionEntryRule(previous, outerContext);

        if ( alt!=ATN.INVALID_ALT_NUMBER ) {

          return alt;

        }

        throw e;

      }

      Collection<BitSet> altSubSets = PredictionMode.getConflictingAltSubsets(reach);

      if ( debug ) {

        System.out.println("LL altSubSets="+altSubSets+

                   ", predict="+PredictionMode.getUniqueAlt(altSubSets)+

                   ", resolvesToJustOneViableAlt="+

                     PredictionMode.resolvesToJustOneViableAlt(altSubSets));

      }

      reach.uniqueAlt = getUniqueAlt(reach);

      // unique prediction?

      if ( reach.uniqueAlt!=ATN.INVALID_ALT_NUMBER ) {

        predictedAlt = reach.uniqueAlt;

        break;

      }

      if ( mode != PredictionMode.LL_EXACT_AMBIG_DETECTION ) {

        predictedAlt = PredictionMode.resolvesToJustOneViableAlt(altSubSets);

        if ( predictedAlt != ATN.INVALID_ALT_NUMBER ) {

          break;

        }

      }

      else {

        // In exact ambiguity mode, we never try to terminate early.

        // Just keeps scarfing until we know what the conflict is

        if ( PredictionMode.allSubsetsConflict(altSubSets) &&

           PredictionMode.allSubsetsEqual(altSubSets) )

        {

          foundExactAmbig = true;

          predictedAlt = PredictionMode.getSingleViableAlt(altSubSets);

          break;

        }

        // else there are multiple non-conflicting subsets or

        // we're not sure what the ambiguity is yet.

        // So, keep going.

      }

      previous = reach;

      if (t != IntStream.EOF) {

        input.consume();

        t = input.LA(1);

      }

    }

    // If the configuration set uniquely predicts an alternative,

    // without conflict, then we know that it's a full LL decision

    // not SLL.

    if ( reach.uniqueAlt != ATN.INVALID_ALT_NUMBER ) {

      reportContextSensitivity(dfa, predictedAlt, reach, startIndex, input.index());

      return predictedAlt;

    }

    // We do not check predicates here because we have checked them

    // on-the-fly when doing full context prediction.

    /*

    In non-exact ambiguity detection mode, we might  actually be able to

    detect an exact ambiguity, but I'm not going to spend the cycles

    needed to check. We only emit ambiguity warnings in exact ambiguity

    mode.

    For example, we might know that we have conflicting configurations.

    But, that does not mean that there is no way forward without a

    conflict. It's possible to have nonconflicting alt subsets as in:

       LL altSubSets=[{1, 2}, {1, 2}, {1}, {1, 2}]

    from

       [(17,1,[5 $]), (13,1,[5 10 $]), (21,1,[5 10 $]), (11,1,[$]),

      (13,2,[5 10 $]), (21,2,[5 10 $]), (11,2,[$])]

    In this case, (17,1,[5 $]) indicates there is some next sequence that

    would resolve this without conflict to alternative 1. Any other viable

    next sequence, however, is associated with a conflict.  We stop

    looking for input because no amount of further lookahead will alter

    the fact that we should predict alternative 1.  We just can't say for

    sure that there is an ambiguity without looking further.

    */

    reportAmbiguity(dfa, D, startIndex, input.index(), foundExactAmbig, null, reach);

    return predictedAlt;

  }

  protected ATNConfigSet computeReachSet(ATNConfigSet closure, int t,

                       boolean fullCtx)

  {

    if ( debug ) System.out.println("in computeReachSet, starting closure: " + closure);

    if (mergeCache == null) {

      mergeCache = new DoubleKeyMap<PredictionContext, PredictionContext, PredictionContext>();

    }

    ATNConfigSet intermediate = new ATNConfigSet(fullCtx);

    /* Configurations already in a rule stop state indicate reaching the end

     * of the decision rule (local context) or end of the start rule (full

     * context). Once reached, these configurations are never updated by a

     * closure operation, so they are handled separately for the performance

     * advantage of having a smaller intermediate set when calling closure.

     *

     * For full-context reach operations, separate handling is required to

     * ensure that the alternative matching the longest overall sequence is

     * chosen when multiple such configurations can match the input.

     */

    List<ATNConfig> skippedStopStates = null;

    // First figure out where we can reach on input t

    for (ATNConfig c : closure) {

      if ( debug ) System.out.println("testing "+getTokenName(t)+" at "+c.toString());

      if (c.state instanceof RuleStopState) {

        assert c.context.isEmpty();

        if (fullCtx || t == IntStream.EOF) {

          if (skippedStopStates == null) {

            skippedStopStates = new ArrayList<ATNConfig>();

          }

          skippedStopStates.add(c);

        }

        continue;

      }

      int n = c.state.getNumberOfTransitions();

      for (int ti=0; ti<n; ti++) {               // for each transition

        Transition trans = c.state.transition(ti);

        ATNState target = getReachableTarget(trans, t);

        if ( target!=null ) {

          intermediate.add(new ATNConfig(c, target), mergeCache);

        }

      }

    }

    // Now figure out where the reach operation can take us...

    ATNConfigSet reach = null;

    /* This block optimizes the reach operation for intermediate sets which

     * trivially indicate a termination state for the overall

     * adaptivePredict operation.

     *

     * The conditions assume that intermediate

     * contains all configurations relevant to the reach set, but this

     * condition is not true when one or more configurations have been

     * withheld in skippedStopStates, or when the current symbol is EOF.

     */

    if (skippedStopStates == null && t != Token.EOF) {

      if ( intermediate.size()==1 ) {

        // Don't pursue the closure if there is just one state.

        // It can only have one alternative; just add to result

        // Also don't pursue the closure if there is unique alternative

        // among the configurations.

        reach = intermediate;

      }

      else if ( getUniqueAlt(intermediate)!=ATN.INVALID_ALT_NUMBER ) {

        // Also don't pursue the closure if there is unique alternative

        // among the configurations.

        reach = intermediate;

      }

    }

    /* If the reach set could not be trivially determined, perform a closure

     * operation on the intermediate set to compute its initial value.

     */

    if (reach == null) {

      reach = new ATNConfigSet(fullCtx);

      Set<ATNConfig> closureBusy = new HashSet<ATNConfig>();

      boolean treatEofAsEpsilon = t == Token.EOF;

      for (ATNConfig c : intermediate) {

        closure(c, reach, closureBusy, false, fullCtx, treatEofAsEpsilon);

      }

    }

    if (t == IntStream.EOF) {

      /* After consuming EOF no additional input is possible, so we are

       * only interested in configurations which reached the end of the

       * decision rule (local context) or end of the start rule (full

       * context). Update reach to contain only these configurations. This

       * handles both explicit EOF transitions in the grammar and implicit

       * EOF transitions following the end of the decision or start rule.

       *

       * When reach==intermediate, no closure operation was performed. In

       * this case, removeAllConfigsNotInRuleStopState needs to check for

       * reachable rule stop states as well as configurations already in

       * a rule stop state.

       *

       * This is handled before the configurations in skippedStopStates,

       * because any configurations potentially added from that list are

       * already guaranteed to meet this condition whether or not it's

       * required.

       */

      reach = removeAllConfigsNotInRuleStopState(reach, reach == intermediate);

    }

    /* If skippedStopStates is not null, then it contains at least one

     * configuration. For full-context reach operations, these

     * configurations reached the end of the start rule, in which case we

     * only add them back to reach if no configuration during the current

     * closure operation reached such a state. This ensures adaptivePredict

     * chooses an alternative matching the longest overall sequence when

     * multiple alternatives are viable.

     */

    if (skippedStopStates != null && (!fullCtx || !PredictionMode.hasConfigInRuleStopState(reach))) {

      assert !skippedStopStates.isEmpty();

      for (ATNConfig c : skippedStopStates) {

        reach.add(c, mergeCache);

      }

    }

    if ( reach.isEmpty() ) return null;

    return reach;

  }

  /**

   * Return a configuration set containing only the configurations from

   * {@code configs} which are in a {@link RuleStopState}. If all

   * configurations in {@code configs} are already in a rule stop state, this

   * method simply returns {@code configs}.

   *

   * <p>When {@code lookToEndOfRule} is true, this method uses

   * {@link ATN#nextTokens} for each configuration in {@code configs} which is

   * not already in a rule stop state to see if a rule stop state is reachable

   * from the configuration via epsilon-only transitions.</p>

   *

   * @param configs the configuration set to update

   * @param lookToEndOfRule when true, this method checks for rule stop states

   * reachable by epsilon-only transitions from each configuration in

   * {@code configs}.

   *

   * @return {@code configs} if all configurations in {@code configs} are in a

   * rule stop state, otherwise return a new configuration set containing only

   * the configurations from {@code configs} which are in a rule stop state

   */

  @NotNull

  protected ATNConfigSet removeAllConfigsNotInRuleStopState(@NotNull ATNConfigSet configs, boolean lookToEndOfRule) {

    if (PredictionMode.allConfigsInRuleStopStates(configs)) {

      return configs;

    }

    ATNConfigSet result = new ATNConfigSet(configs.fullCtx);

    for (ATNConfig config : configs) {

      if (config.state instanceof RuleStopState) {

        result.add(config, mergeCache);

        continue;

      }

      if (lookToEndOfRule && config.state.onlyHasEpsilonTransitions()) {

        IntervalSet nextTokens = atn.nextTokens(config.state);

        if (nextTokens.contains(Token.EPSILON)) {

          ATNState endOfRuleState = atn.ruleToStopState[config.state.ruleIndex];

          result.add(new ATNConfig(config, endOfRuleState), mergeCache);

        }

      }

    }

    return result;

  }

  @NotNull

  protected ATNConfigSet computeStartState(@NotNull ATNState p,

                      @Nullable RuleContext ctx,

                      boolean fullCtx)

  {

    // always at least the implicit call to start rule

    PredictionContext initialContext = PredictionContext.fromRuleContext(atn, ctx);

    ATNConfigSet configs = new ATNConfigSet(fullCtx);

    for (int i=0; i<p.getNumberOfTransitions(); i++) {

      ATNState target = p.transition(i).target;

      ATNConfig c = new ATNConfig(target, i+1, initialContext);

      Set<ATNConfig> closureBusy = new HashSet<ATNConfig>();

      closure(c, configs, closureBusy, true, fullCtx, false);

    }

    return configs;

  }

  /* parrt internal source braindump that doesn't mess up

   * external API spec.

    applyPrecedenceFilter is an optimization to avoid highly

    nonlinear prediction of expressions and other left recursive

    rules. The precedence predicates such as {3>=prec}? Are highly

    context-sensitive in that they can only be properly evaluated

    in the context of the proper prec argument. Without pruning,

    these predicates are normal predicates evaluated when we reach

    conflict state (or unique prediction). As we cannot evaluate

    these predicates out of context, the resulting conflict leads

    to full LL evaluation and nonlinear prediction which shows up

    very clearly with fairly large expressions.

    Example grammar:

    e : e '*' e

      | e '+' e

      | INT

      ;

    We convert that to the following:

    e[int prec]

      :   INT

        ( {3>=prec}? '*' e[4]

        | {2>=prec}? '+' e[3]

        )*

      ;

    The (..)* loop has a decision for the inner block as well as

    an enter or exit decision, which is what concerns us here. At

    the 1st + of input 1+2+3, the loop entry sees both predicates

    and the loop exit also sees both predicates by falling off the

    edge of e.  This is because we have no stack information with

    SLL and find the follow of e, which will hit the return states

    inside the loop after e[4] and e[3], which brings it back to

    the enter or exit decision. In this case, we know that we

    cannot evaluate those predicates because we have fallen off

    the edge of the stack and will in general not know which prec

    parameter is the right one to use in the predicate.

    Because we have special information, that these are precedence

    predicates, we can resolve them without failing over to full

    LL despite their context sensitive nature. We make an

    assumption that prec[-1] <= prec[0], meaning that the current

    precedence level is greater than or equal to the precedence

    level of recursive invocations above us in the stack. For

    example, if predicate {3>=prec}? is true of the current prec,

    then one option is to enter the loop to match it now. The

    other option is to exit the loop and the left recursive rule

    to match the current operator in rule invocation further up

    the stack. But, we know that all of those prec are lower or

    the same value and so we can decide to enter the loop instead

    of matching it later. That means we can strip out the other

    configuration for the exit branch.

    So imagine we have (14,1,$,{2>=prec}?) and then

    (14,2,$-dipsIntoOuterContext,{2>=prec}?). The optimization

    allows us to collapse these two configurations. We know that

    if {2>=prec}? is true for the current prec parameter, it will

    also be true for any prec from an invoking e call, indicated

    by dipsIntoOuterContext. As the predicates are both true, we

    have the option to evaluate them early in the decision start

    state. We do this by stripping both predicates and choosing to

    enter the loop as it is consistent with the notion of operator

    precedence. It's also how the full LL conflict resolution

    would work.

    The solution requires a different DFA start state for each

    precedence level.

    The basic filter mechanism is to remove configurations of the

    form (p, 2, pi) if (p, 1, pi) exists for the same p and pi. In

    other words, for the same ATN state and predicate context,

    remove any configuration associated with an exit branch if

    there is a configuration associated with the enter branch.

    It's also the case that the filter evaluates precedence

    predicates and resolves conflicts according to precedence

    levels. For example, for input 1+2+3 at the first +, we see

    prediction filtering

    [(11,1,[$],{3>=prec}?), (14,1,[$],{2>=prec}?), (5,2,[$],up=1),

     (11,2,[$],up=1), (14,2,[$],up=1)],hasSemanticContext=true,dipsIntoOuterContext

    to

    [(11,1,[$]), (14,1,[$]), (5,2,[$],up=1)],dipsIntoOuterContext

    This filters because {3>=prec}? evals to true and collapses

    (11,1,[$],{3>=prec}?) and (11,2,[$],up=1) since early conflict

    resolution based upon rules of operator precedence fits with

    our usual match first alt upon conflict.

    We noticed a problem where a recursive call resets precedence

    to 0. Sam's fix: each config has flag indicating if it has

    returned from an expr[0] call. then just don't filter any

    config with that flag set. flag is carried along in

    closure(). so to avoid adding field, set bit just under sign

    bit of dipsIntoOuterContext (SUPPRESS_PRECEDENCE_FILTER).

    With the change you filter "unless (p, 2, pi) was reached

    after leaving the rule stop state of the LR rule containing

    state p, corresponding to a rule invocation with precedence

    level 0"

   */

 

  /**

   * This method transforms the start state computed by

   * {@link #computeStartState} to the special start state used by a

   * precedence DFA for a particular precedence value. The transformation

   * process applies the following changes to the start state's configuration

   * set.

   *

   * <ol>

   * <li>Evaluate the precedence predicates for each configuration using

   * {@link SemanticContext#evalPrecedence}.</li>

   * <li>When {@link ATNConfig#isPrecedenceFilterSuppressed} is {@code false},

   * remove all configurations which predict an alternative greater than 1,

   * for which another configuration that predicts alternative 1 is in the

   * same ATN state with the same prediction context. This transformation is

   * valid for the following reasons:

   * <ul>

   * <li>The closure block cannot contain any epsilon transitions which bypass

   * the body of the closure, so all states reachable via alternative 1 are

   * part of the precedence alternatives of the transformed left-recursive

   * rule.</li>

   * <li>The "primary" portion of a left recursive rule cannot contain an

   * epsilon transition, so the only way an alternative other than 1 can exist

   * in a state that is also reachable via alternative 1 is by nesting calls

   * to the left-recursive rule, with the outer calls not being at the

   * preferred precedence level. The

   * {@link ATNConfig#isPrecedenceFilterSuppressed} property marks ATN

   * configurations which do not meet this condition, and therefore are not

   * eligible for elimination during the filtering process.</li>

   * </ul>

   * </li>

   * </ol>

   *

   * <p>

   * The prediction context must be considered by this filter to address

   * situations like the following.

   * </p>

   * <code>

   * <pre>

   * grammar TA;

   * prog: statement* EOF;

   * statement: letterA | statement letterA 'b' ;

   * letterA: 'a';

   * </pre>

   * </code>

   * <p>

   * If the above grammar, the ATN state immediately before the token

   * reference {@code 'a'} in {@code letterA} is reachable from the left edge

   * of both the primary and closure blocks of the left-recursive rule

   * {@code statement}. The prediction context associated with each of these

   * configurations distinguishes between them, and prevents the alternative

   * which stepped out to {@code prog} (and then back in to {@code statement}

   * from being eliminated by the filter.

   * </p>

   *

   * @param configs The configuration set computed by

   * {@link #computeStartState} as the start state for the DFA.

   * @return The transformed configuration set representing the start state

   * for a precedence DFA at a particular precedence level (determined by

   * calling {@link Parser#getPrecedence}).

   */

  @NotNull

  protected ATNConfigSet applyPrecedenceFilter(@NotNull ATNConfigSet configs) {

    Map<Integer, PredictionContext> statesFromAlt1 = new HashMap<Integer, PredictionContext>();

    ATNConfigSet configSet = new ATNConfigSet(configs.fullCtx);

    for (ATNConfig config : configs) {

      // handle alt 1 first

      if (config.alt != 1) {

        continue;

      }

      SemanticContext updatedContext = config.semanticContext.evalPrecedence(parser, _outerContext);

      if (updatedContext == null) {

        // the configuration was eliminated

        continue;

      }

      statesFromAlt1.put(config.state.stateNumber, config.context);

      if (updatedContext != config.semanticContext) {

        configSet.add(new ATNConfig(config, updatedContext), mergeCache);

      }

      else {

        configSet.add(config, mergeCache);

      }

    }

    for (ATNConfig config : configs) {

      if (config.alt == 1) {

        // already handled

        continue;

      }

      if (!config.isPrecedenceFilterSuppressed()) {

        /* In the future, this elimination step could be updated to also

         * filter the prediction context for alternatives predicting alt>1

         * (basically a graph subtraction algorithm).

         */

        PredictionContext context = statesFromAlt1.get(config.state.stateNumber);

        if (context != null && context.equals(config.context)) {

          // eliminated

          continue;

        }

      }

      configSet.add(config, mergeCache);

    }

    return configSet;

  }

  @Nullable

  protected ATNState getReachableTarget(@NotNull Transition trans, int ttype) {

    if (trans.matches(ttype, 0, atn.maxTokenType)) {

      return trans.target;

    }

    return null;

  }

  protected SemanticContext[] getPredsForAmbigAlts(@NotNull BitSet ambigAlts,

                          @NotNull ATNConfigSet configs,

                          int nalts)

  {

    // REACH=[1|1|[]|0:0, 1|2|[]|0:1]

    /* altToPred starts as an array of all null contexts. The entry at index i

     * corresponds to alternative i. altToPred[i] may have one of three values:

     *   1. null: no ATNConfig c is found such that c.alt==i

     *   2. SemanticContext.NONE: At least one ATNConfig c exists such that

     *      c.alt==i and c.semanticContext==SemanticContext.NONE. In other words,

     *      alt i has at least one unpredicated config.

     *   3. Non-NONE Semantic Context: There exists at least one, and for all

     *      ATNConfig c such that c.alt==i, c.semanticContext!=SemanticContext.NONE.

     *

     * From this, it is clear that NONE||anything==NONE.

     */

    SemanticContext[] altToPred = new SemanticContext[nalts + 1];

    for (ATNConfig c : configs) {

      if ( ambigAlts.get(c.alt) ) {

        altToPred[c.alt] = SemanticContext.or(altToPred[c.alt], c.semanticContext);

      }

    }

    int nPredAlts = 0;

    for (int i = 1; i <= nalts; i++) {

      if (altToPred[i] == null) {

        altToPred[i] = SemanticContext.NONE;

      }

      else if (altToPred[i] != SemanticContext.NONE) {

        nPredAlts++;

      }

    }

//    for (int i = 0; i < altToPred.length; i++) {

//      altToPred[i] = altToPred[i].optimize();

//    }

    // nonambig alts are null in altToPred

    if ( nPredAlts==0 ) altToPred = null;

    if ( debug ) System.out.println("getPredsForAmbigAlts result "+Arrays.toString(altToPred));

    return altToPred;

  }

  protected DFAState.PredPrediction[] getPredicatePredictions(BitSet ambigAlts,

                               SemanticContext[] altToPred)

  {

    List<DFAState.PredPrediction> pairs = new ArrayList<DFAState.PredPrediction>();

    boolean containsPredicate = false;

    for (int i = 1; i < altToPred.length; i++) {

      SemanticContext pred = altToPred[i];

      // unpredicated is indicated by SemanticContext.NONE

      assert pred != null;

      if (ambigAlts!=null && ambigAlts.get(i)) {

        pairs.add(new DFAState.PredPrediction(pred, i));

      }

      if ( pred!=SemanticContext.NONE ) containsPredicate = true;

    }

    if ( !containsPredicate ) {

      return null;

    }

    return pairs.toArray(new DFAState.PredPrediction[pairs.size()]);

  }

  /**

   * This method is used to improve the localization of error messages by

   * choosing an alternative rather than throwing a

   * {@link NoViableAltException} in particular prediction scenarios where the

   * {@link #ERROR} state was reached during ATN simulation.

   *

   * <p>

   * The default implementation of this method uses the following

   * algorithm to identify an ATN configuration which successfully parsed the

   * decision entry rule. Choosing such an alternative ensures that the

   * {@link ParserRuleContext} returned by the calling rule will be complete

   * and valid, and the syntax error will be reported later at a more

   * localized location.</p>

   *

   * <ul>

   * <li>If a syntactically valid path or paths reach the end of the decision rule and

   * they are semantically valid if predicated, return the min associated alt.</li>

   * <li>Else, if a semantically invalid but syntactically valid path exist

   * or paths exist, return the minimum associated alt.

   * </li>

   * <li>Otherwise, return {@link ATN#INVALID_ALT_NUMBER}.</li>

   * </ul>

   *

   * <p>

   * In some scenarios, the algorithm described above could predict an

   * alternative which will result in a {@link FailedPredicateException} in

   * the parser. Specifically, this could occur if the <em>only</em> configuration

   * capable of successfully parsing to the end of the decision rule is

   * blocked by a semantic predicate. By choosing this alternative within

   * {@link #adaptivePredict} instead of throwing a

   * {@link NoViableAltException}, the resulting

   * {@link FailedPredicateException} in the parser will identify the specific

   * predicate which is preventing the parser from successfully parsing the

   * decision rule, which helps developers identify and correct logic errors

   * in semantic predicates.

   * </p>

   *

   * @param configs The ATN configurations which were valid immediately before

   * the {@link #ERROR} state was reached

   * @param outerContext The is the \gamma_0 initial parser context from the paper

   * or the parser stack at the instant before prediction commences.

   *

   * @return The value to return from {@link #adaptivePredict}, or

   * {@link ATN#INVALID_ALT_NUMBER} if a suitable alternative was not

   * identified and {@link #adaptivePredict} should report an error instead.

   */

  protected int getSynValidOrSemInvalidAltThatFinishedDecisionEntryRule(ATNConfigSet configs,

                                      ParserRuleContext outerContext)

  {

    Pair<ATNConfigSet,ATNConfigSet> sets =

      splitAccordingToSemanticValidity(configs, outerContext);

    ATNConfigSet semValidConfigs = sets.a;

    ATNConfigSet semInvalidConfigs = sets.b;

    int alt = getAltThatFinishedDecisionEntryRule(semValidConfigs);

    if ( alt!=ATN.INVALID_ALT_NUMBER ) { // semantically/syntactically viable path exists

      return alt;

    }

    // Is there a syntactically valid path with a failed pred?

    if ( semInvalidConfigs.size()>0 ) {

      alt = getAltThatFinishedDecisionEntryRule(semInvalidConfigs);

      if ( alt!=ATN.INVALID_ALT_NUMBER ) { // syntactically viable path exists

        return alt;

      }

    }

    return ATN.INVALID_ALT_NUMBER;

  }

  protected int getAltThatFinishedDecisionEntryRule(ATNConfigSet configs) {

    IntervalSet alts = new IntervalSet();

    for (ATNConfig c : configs) {

      if ( c.getOuterContextDepth()>0 || (c.state instanceof RuleStopState && c.context.hasEmptyPath()) ) {

        alts.add(c.alt);

      }

    }

    if ( alts.size()==0 ) return ATN.INVALID_ALT_NUMBER;

    return alts.getMinElement();

  }

  /** Walk the list of configurations and split them according to

   *  those that have preds evaluating to true/false.  If no pred, assume

   *  true pred and include in succeeded set.  Returns Pair of sets.

   *

   *  Create a new set so as not to alter the incoming parameter.

   *

   *  Assumption: the input stream has been restored to the starting point

   *  prediction, which is where predicates need to evaluate.

    */

  protected Pair<ATNConfigSet,ATNConfigSet> splitAccordingToSemanticValidity(

    ATNConfigSet configs,

    ParserRuleContext outerContext)

  {

    ATNConfigSet succeeded = new ATNConfigSet(configs.fullCtx);

    ATNConfigSet failed = new ATNConfigSet(configs.fullCtx);

    for (ATNConfig c : configs) {

      if ( c.semanticContext!=SemanticContext.NONE ) {

        boolean predicateEvaluationResult = evalSemanticContext(c.semanticContext, outerContext, c.alt, configs.fullCtx);

        if ( predicateEvaluationResult ) {

          succeeded.add(c);

        }

        else {

          failed.add(c);

        }

      }

      else {

        succeeded.add(c);

      }

    }

    return new Pair<ATNConfigSet, ATNConfigSet>(succeeded,failed);

  }