Class ParserATNSimulator

  • Direct Known Subclasses:

    public class ParserATNSimulator
    extends ATNSimulator
    The embodiment of the adaptive LL(*), ALL(*), parsing strategy.

    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).

    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(*)).

    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.)

    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.


    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.

    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.

    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.


    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.

    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:

       : '{' superCall? stat* '}'

    Or, you might see something like

       : superCall ';'
       | expression ';'
       | ...

    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.


    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.

    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.

    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.

    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.


    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 decisionToDFA field. They also share a PredictionContextCache object that makes sure that all PredictionContext objects are shared among the DFA states. This makes a big size difference.


    The ParserATNSimulator locks on the decisionToDFA field when it adds a new DFA object to that array. addDFAEdge(org.antlr.v4.runtime.dfa.DFA, org.antlr.v4.runtime.dfa.DFAState, int, org.antlr.v4.runtime.dfa.DFAState) locks on the DFA for the current decision when setting the DFAState.edges field. addDFAState(org.antlr.v4.runtime.dfa.DFA, org.antlr.v4.runtime.dfa.DFAState) 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 addDFAState(org.antlr.v4.runtime.dfa.DFA, org.antlr.v4.runtime.dfa.DFAState) method also locks inside the DFA lock but this time on the shared context cache when it rebuilds the configurations' 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 s.edge[t] get the same physical target DFAState, or null. Once into the DFA, the DFA simulation does not reference the DFA.states map. It follows the DFAState.edges field to new targets. The DFA simulator will either find DFAState.edges to be null, to be non-null and dfa.edges[t] null, or dfa.edges[t] to be non-null. The addDFAEdge(org.antlr.v4.runtime.dfa.DFA, org.antlr.v4.runtime.dfa.DFAState, int, org.antlr.v4.runtime.dfa.DFAState) method could be racing to set the field but in either case the DFA simulator works; if null, and requests ATN simulation. It could also race trying to get dfa.edges[t], but either way it will work because it's not doing a test and set operation.

    Starting with SLL then failing to combined SLL/LL (Two-Stage Parsing)

    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 BailErrorStrategy:

     parser.setErrorHandler(new BailErrorStrategy());

    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.

    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 > 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.

    Let's say we have a set of SLL conflicting alternatives {1, 2, 3} and a smaller LL set called s. If s is {2, 3}, then SLL parsing will get an error because SLL will pursue alternative 1. If s is {1, 2} or {1, 3} then both SLL and LL will choose the same alternative because alternative one is the minimum of either set. If s is {2} or {3} then SLL will get a syntax error. If s is {1} then SLL will succeed.

    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.

    • Field Detail

      • debug

        public static boolean debug
      • trace_atn_sim

        public static boolean trace_atn_sim
      • dfa_debug

        public static boolean dfa_debug
      • retry_debug

        public static boolean retry_debug

        public static final boolean TURN_OFF_LR_LOOP_ENTRY_BRANCH_OPT
        Just in case this optimization is bad, add an ENV variable to turn it off
      • parser

        protected final Parser parser
      • decisionToDFA

        public final DFA[] decisionToDFA
      • mergeCache

        protected DoubleKeyMap<PredictionContext,​PredictionContext,​PredictionContext> mergeCache
        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)→c. We can avoid the merge if we ever see a and b again. Note that (b,a)→c should also be examined during cache lookup.
      • _startIndex

        protected int _startIndex
      • _dfa

        protected DFA _dfa
    • Method Detail

      • clearDFA

        public void clearDFA()
        Description copied from class: ATNSimulator
        Clear the DFA cache used by the current instance. Since the DFA cache may be shared by multiple ATN simulators, this method may affect the performance (but not accuracy) of other parsers which are being used concurrently.
        clearDFA in class ATNSimulator
      • execATN

        protected int execATN​(DFA dfa,
                              DFAState s0,
                              TokenStream input,
                              int startIndex,
                              ParserRuleContext outerContext)
        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
      • getExistingTargetState

        protected DFAState getExistingTargetState​(DFAState previousD,
                                                  int t)
        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 null.
        previousD - The current DFA state
        t - The next input symbol
        The existing target DFA state for the given input symbol t, or null if the target state for this edge is not already cached
      • computeTargetState

        protected DFAState computeTargetState​(DFA dfa,
                                              DFAState previousD,
                                              int t)
        Compute a target state for an edge in the DFA, and attempt to add the computed state and corresponding edge to the DFA.
        dfa - The DFA
        previousD - The current DFA state
        t - The next input symbol
        The computed target DFA state for the given input symbol t. If t does not lead to a valid DFA state, this method returns ATNSimulator.ERROR.
      • predicateDFAState

        protected void predicateDFAState​(DFAState dfaState,
                                         DecisionState decisionState)
      • removeAllConfigsNotInRuleStopState

        protected ATNConfigSet removeAllConfigsNotInRuleStopState​(ATNConfigSet configs,
                                                                  boolean lookToEndOfRule)
        Return a configuration set containing only the configurations from configs which are in a RuleStopState. If all configurations in configs are already in a rule stop state, this method simply returns configs.

        When lookToEndOfRule is true, this method uses ATN.nextTokens(org.antlr.v4.runtime.atn.ATNState, org.antlr.v4.runtime.RuleContext) for each configuration in 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.

        configs - the configuration set to update
        lookToEndOfRule - when true, this method checks for rule stop states reachable by epsilon-only transitions from each configuration in configs.
        configs if all configurations in configs are in a rule stop state, otherwise return a new configuration set containing only the configurations from configs which are in a rule stop state
      • applyPrecedenceFilter

        protected ATNConfigSet applyPrecedenceFilter​(ATNConfigSet configs)
        This method transforms the start state computed by computeStartState(org.antlr.v4.runtime.atn.ATNState, org.antlr.v4.runtime.RuleContext, boolean) 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.
        1. Evaluate the precedence predicates for each configuration using SemanticContext.evalPrecedence(org.antlr.v4.runtime.Recognizer<?, ?>, org.antlr.v4.runtime.RuleContext).
        2. When ATNConfig.isPrecedenceFilterSuppressed() is 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:
          • 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.
          • 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 ATNConfig.isPrecedenceFilterSuppressed() property marks ATN configurations which do not meet this condition, and therefore are not eligible for elimination during the filtering process.

        The prediction context must be considered by this filter to address situations like the following.

         grammar TA;
         prog: statement* EOF;
         statement: letterA | statement letterA 'b' ;
         letterA: 'a';

        If the above grammar, the ATN state immediately before the token reference 'a' in letterA is reachable from the left edge of both the primary and closure blocks of the left-recursive rule statement. The prediction context associated with each of these configurations distinguishes between them, and prevents the alternative which stepped out to prog (and then back in to statement from being eliminated by the filter.

        configs - The configuration set computed by computeStartState(org.antlr.v4.runtime.atn.ATNState, org.antlr.v4.runtime.RuleContext, boolean) as the start state for the DFA.
        The transformed configuration set representing the start state for a precedence DFA at a particular precedence level (determined by calling Parser.getPrecedence()).
      • getReachableTarget

        protected ATNState getReachableTarget​(Transition trans,
                                              int ttype)
      • getAltThatFinishedDecisionEntryRule

        protected int getAltThatFinishedDecisionEntryRule​(ATNConfigSet configs)
      • splitAccordingToSemanticValidity

        protected Pair<ATNConfigSet,​ATNConfigSet> splitAccordingToSemanticValidity​(ATNConfigSet configs,
                                                                                         ParserRuleContext outerContext)
        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.
      • evalSemanticContext

        protected BitSet evalSemanticContext​(DFAState.PredPrediction[] predPredictions,
                                             ParserRuleContext outerContext,
                                             boolean complete)
        Look through a list of predicate/alt pairs, returning alts for the pairs that win. A NONE predicate indicates an alt containing an unpredicated config which behaves as "always true." If !complete then we stop at the first predicate that evaluates to true. This includes pairs with null predicates.
      • evalSemanticContext

        protected boolean evalSemanticContext​(SemanticContext pred,
                                              ParserRuleContext parserCallStack,
                                              int alt,
                                              boolean fullCtx)
        Evaluate a semantic context within a specific parser context.

        This method might not be called for every semantic context evaluated during the prediction process. In particular, we currently do not evaluate the following but it may change in the future:

        • Precedence predicates (represented by SemanticContext.PrecedencePredicate) are not currently evaluated through this method.
        • Operator predicates (represented by SemanticContext.AND and SemanticContext.OR) are evaluated as a single semantic context, rather than evaluating the operands individually. Implementations which require evaluation results from individual predicates should override this method to explicitly handle evaluation of the operands within operator predicates.
        pred - The semantic context to evaluate
        parserCallStack - The parser context in which to evaluate the semantic context
        alt - The alternative which is guarded by pred
        fullCtx - true if the evaluation is occurring during LL prediction; otherwise, false if the evaluation is occurring during SLL prediction
      • closure

        protected void closure​(ATNConfig config,
                               ATNConfigSet configs,
                               Set<ATNConfig> closureBusy,
                               boolean collectPredicates,
                               boolean fullCtx,
                               boolean treatEofAsEpsilon)
      • closureCheckingStopState

        protected void closureCheckingStopState​(ATNConfig config,
                                                ATNConfigSet configs,
                                                Set<ATNConfig> closureBusy,
                                                boolean collectPredicates,
                                                boolean fullCtx,
                                                int depth,
                                                boolean treatEofAsEpsilon)
      • closure_

        protected void closure_​(ATNConfig config,
                                ATNConfigSet configs,
                                Set<ATNConfig> closureBusy,
                                boolean collectPredicates,
                                boolean fullCtx,
                                int depth,
                                boolean treatEofAsEpsilon)
        Do the actual work of walking epsilon edges
      • canDropLoopEntryEdgeInLeftRecursiveRule

        protected boolean canDropLoopEntryEdgeInLeftRecursiveRule​(ATNConfig config)
        Implements first-edge (loop entry) elimination as an optimization during closure operations. See antlr/antlr4#1398. The optimization is to avoid adding the loop entry config when the exit path can only lead back to the same StarLoopEntryState after popping context at the rule end state (traversing only epsilon edges, so we're still in closure, in this same rule). We need to detect any state that can reach loop entry on epsilon w/o exiting rule. We don't have to look at FOLLOW links, just ensure that all stack tops for config refer to key states in LR rule. To verify we are in the right situation we must first check closure is at a StarLoopEntryState generated during LR removal. Then we check that each stack top of context is a return state from one of these cases: 1. 'not' expr, '(' type ')' expr. The return state points at loop entry state 2. expr op expr. The return state is the block end of internal block of (...)* 3. 'between' expr 'and' expr. The return state of 2nd expr reference. That state points at block end of internal block of (...)*. 4. expr '?' expr ':' expr. The return state points at block end, which points at loop entry state. If any is true for each stack top, then closure does not add a config to the current config set for edge[0], the loop entry branch. Conditions fail if any context for the current config is: a. empty (we'd fall out of expr to do a global FOLLOW which could even be to some weird spot in expr) or, b. lies outside of expr or, c. lies within expr but at a state not the BlockEndState generated during LR removal Do we need to evaluate predicates ever in closure for this case? No. Predicates, including precedence predicates, are only evaluated when computing a DFA start state. I.e., only before the lookahead (but not parser) consumes a token. There are no epsilon edges allowed in LR rule alt blocks or in the "primary" part (ID here). If closure is in StarLoopEntryState any lookahead operation will have consumed a token as there are no epsilon-paths that lead to StarLoopEntryState. We do not have to evaluate predicates therefore if we are in the generated StarLoopEntryState of a LR rule. Note that when making a prediction starting at that decision point, decision d=2, compute-start-state performs closure starting at edges[0], edges[1] emanating from StarLoopEntryState. That means it is not performing closure on StarLoopEntryState during compute-start-state. How do we know this always gives same prediction answer? Without predicates, loop entry and exit paths are ambiguous upon remaining input +b (in, say, a+b). Either paths lead to valid parses. Closure can lead to consuming + immediately or by falling out of this call to expr back into expr and loop back again to StarLoopEntryState to match +b. In this special case, we choose the more efficient path, which is to take the bypass path. The lookahead language has not changed because closure chooses one path over the other. Both paths lead to consuming the same remaining input during a lookahead operation. If the next token is an operator, lookahead will enter the choice block with operators. If it is not, lookahead will exit expr. Same as if closure had chosen to enter the choice block immediately. Closure is examining one config (some loopentrystate, some alt, context) which means it is considering exactly one alt. Closure always copies the same alt to any derived configs. How do we know this optimization doesn't mess up precedence in our parse trees? Looking through expr from left edge of stat only has to confirm that an input, say, a+b+c; begins with any valid interpretation of an expression. The precedence actually doesn't matter when making a decision in stat seeing through expr. It is only when parsing rule expr that we must use the precedence to get the right interpretation and, hence, parse tree.
      • getRuleName

        public String getRuleName​(int index)
      • getEpsilonTarget

        protected ATNConfig getEpsilonTarget​(ATNConfig config,
                                             Transition t,
                                             boolean collectPredicates,
                                             boolean inContext,
                                             boolean fullCtx,
                                             boolean treatEofAsEpsilon)
      • getConflictingAlts

        protected BitSet getConflictingAlts​(ATNConfigSet configs)
        Gets a BitSet containing the alternatives in configs which are part of one or more conflicting alternative subsets.
        configs - The ATNConfigSet to analyze.
        The alternatives in configs which are part of one or more conflicting alternative subsets. If configs does not contain any conflicting subsets, this method returns an empty BitSet.
      • getConflictingAltsOrUniqueAlt

        protected BitSet getConflictingAltsOrUniqueAlt​(ATNConfigSet configs)
        Sam pointed out a problem with the previous definition, v3, of ambiguous states. If we have another state associated with conflicting alternatives, we should keep going. For example, the following grammar s : (ID | ID ID?) ';' ; When the ATN simulation reaches the state before ';', it has a DFA state that looks like: [12|1|[], 6|2|[], 12|2|[]]. Naturally 12|1|[] and 12|2|[] conflict, but we cannot stop processing this node because alternative to has another way to continue, via [6|2|[]]. The key is that we have a single state that has config's only associated with a single alternative, 2, and crucially the state transitions among the configurations are all non-epsilon transitions. That means we don't consider any conflicts that include alternative 2. So, we ignore the conflict between alts 1 and 2. We ignore a set of conflicting alts when there is an intersection with an alternative associated with a single alt state in the state→config-list map. It's also the case that we might have two conflicting configurations but also a 3rd nonconflicting configuration for a different alternative: [1|1|[], 1|2|[], 8|3|[]]. This can come about from grammar: a : A | A | A B ; After matching input A, we reach the stop state for rule A, state 1. State 8 is the state right before B. Clearly alternatives 1 and 2 conflict and no amount of further lookahead will separate the two. However, alternative 3 will be able to continue and so we do not stop working on this state. In the previous example, we're concerned with states associated with the conflicting alternatives. Here alt 3 is not associated with the conflicting configs, but since we can continue looking for input reasonably, I don't declare the state done. We ignore a set of conflicting alts when we have an alternative that we still need to pursue.
      • getTokenName

        public String getTokenName​(int t)
      • dumpDeadEndConfigs

        public void dumpDeadEndConfigs​(NoViableAltException nvae)
        Used for debugging in adaptivePredict around execATN but I cut it out for clarity now that alg. works well. We can leave this "dead" code for a bit.
      • getUniqueAlt

        protected static int getUniqueAlt​(ATNConfigSet configs)
      • addDFAState

        protected DFAState addDFAState​(DFA dfa,
                                       DFAState D)
        Add state D to the DFA if it is not already present, and return the actual instance stored in the DFA. If a state equivalent to D is already in the DFA, the existing state is returned. Otherwise this method returns D after adding it to the DFA.

        If D is ATNSimulator.ERROR, this method returns ATNSimulator.ERROR and does not change the DFA.

        dfa - The dfa
        D - The DFA state to add
        The state stored in the DFA. This will be either the existing state if D is already in the DFA, or D itself if the state was not already present.
      • reportAttemptingFullContext

        protected void reportAttemptingFullContext​(DFA dfa,
                                                   BitSet conflictingAlts,
                                                   ATNConfigSet configs,
                                                   int startIndex,
                                                   int stopIndex)
      • reportContextSensitivity

        protected void reportContextSensitivity​(DFA dfa,
                                                int prediction,
                                                ATNConfigSet configs,
                                                int startIndex,
                                                int stopIndex)
      • reportAmbiguity

        protected void reportAmbiguity​(DFA dfa,
                                       DFAState D,
                                       int startIndex,
                                       int stopIndex,
                                       boolean exact,
                                       BitSet ambigAlts,
                                       ATNConfigSet configs)
        If context sensitive parsing, we know it's ambiguity not conflict
      • setPredictionMode

        public final void setPredictionMode​(PredictionMode mode)
      • getParser

        public Parser getParser()
      • getSafeEnv

        public static String getSafeEnv​(String envName)