Package org.antlr.analysis

Class DecisionProbe

  • public class DecisionProbe
extends java.lang.Object
    Collection of information about what is wrong with a decision as discovered while building the DFA predictor. The information is collected during NFA→DFA conversion and, while some of this is available elsewhere, it is nice to have it all tracked in one spot so a great error message can be easily had. I also like the fact that this object tracks it all for later perusing to make an excellent error message instead of lots of imprecise on-the-fly warnings (during conversion). A decision normally only has one problem; e.g., some input sequence can be matched by multiple alternatives. Unfortunately, some decisions such as a : ( A | B ) | ( A | B ) | A ; have multiple problems. So in general, you should approach a decision as having multiple flaws each one uniquely identified by a DFAState. For example, statesWithSyntacticallyAmbiguousAltsSet tracks the set of all DFAStates where ANTLR has discovered a problem. Recall that a decision is represented internall with a DFA comprised of multiple states, each of which could potentially have problems. Because of this, you need to iterate over this list of DFA states. You'll note that most of the informational methods like getSampleNonDeterministicInputSequence() require a DFAState. This state will be one of the iterated states from stateToSyntacticallyAmbiguousAltsSet. This class is not thread safe due to shared use of visited maps etc... Only one thread should really need to access one DecisionProbe anyway.

    • Field Detail

      • dfa

        public DFA dfa

      • statesWithSyntacticallyAmbiguousAltsSet

        protected java.util.Set<DFAState> statesWithSyntacticallyAmbiguousAltsSet
        Track all DFA states with nondeterministic alternatives. By reaching the same DFA state, a path through the NFA for some input is able to reach the same NFA state by starting at more than one alternative's left edge. Though, later, we may find that predicates resolve the issue, but track info anyway. Note that from the DFA state, you can ask for which alts are nondeterministic.

      • stateToSyntacticallyAmbiguousTokensRuleAltsMap

        protected java.util.Map<DFAState,​java.util.Set<java.lang.Integer>> stateToSyntacticallyAmbiguousTokensRuleAltsMap
        Track just like stateToSyntacticallyAmbiguousAltsMap, but only for nondeterminisms that arise in the Tokens rule such as keyword vs ID rule. The state maps to the list of Tokens rule alts that are in conflict.

      • statesResolvedWithSemanticPredicatesSet

        protected java.util.Set<DFAState> statesResolvedWithSemanticPredicatesSet
        Was a syntactic ambiguity resolved with predicates? Any DFA state that predicts more than one alternative, must be resolved with predicates or it should be reported to the user.

      • stateToAltSetWithSemanticPredicatesMap

        protected java.util.Map<DFAState,​java.util.Map<java.lang.Integer,​SemanticContext>> stateToAltSetWithSemanticPredicatesMap
        Track the predicates for each alt per DFA state; more than one DFA state might have syntactically ambig alt prediction. Maps DFA state to another map, mapping alt number to a SemanticContext (pred(s) to execute to resolve syntactic ambiguity).

      • stateToIncompletelyCoveredAltsMap

        protected java.util.Map<DFAState,​java.util.Map<java.lang.Integer,​java.util.Set<org.antlr.runtime.Token>>> stateToIncompletelyCoveredAltsMap
        Tracks alts insufficiently covered. For example, p1||true gets reduced to true and so leaves whole alt uncovered. This maps DFA state to the set of alts

      • danglingStates

        protected java.util.Set<DFAState> danglingStates
        The set of states w/o emanating edges and w/o resolving sem preds.

      • altsWithProblem

        protected java.util.Set<java.lang.Integer> altsWithProblem
        The overall list of alts within the decision that have at least one conflicting input sequence.

      • nonLLStarDecision

        public boolean nonLLStarDecision
        If decision with > 1 alt has recursion in > 1 alt, it's (likely) nonregular lookahead. The decision cannot be made with a DFA. the alts are stored in altsWithProblem.

      • stateToRecursionOverflowConfigurationsMap

        protected MultiMap<java.lang.Integer,​NFAConfiguration> stateToRecursionOverflowConfigurationsMap
        Recursion is limited to a particular depth. If that limit is exceeded the proposed new NFAConfiguration is recorded for the associated DFA state.

      • timedOut

        protected boolean timedOut
        Did ANTLR have to terminate early on the analysis of this decision?

      • stateReachable

        protected java.util.Map<java.lang.Integer,​java.lang.Integer> stateReachable
        Used to find paths through syntactically ambiguous DFA. If we've seen statement number before, what did we learn?

      • REACHABLE_BUSY

        public static final java.lang.Integer REACHABLE_BUSY

      • REACHABLE_NO

        public static final java.lang.Integer REACHABLE_NO

      • REACHABLE_YES

        public static final java.lang.Integer REACHABLE_YES

      • statesVisitedAtInputDepth

        protected java.util.Set<java.lang.String> statesVisitedAtInputDepth
        Used while finding a path through an NFA whose edge labels match an input sequence. Tracks the input position we were at the last time at this node. If same input position, then we'd have reached same state without consuming input...probably an infinite loop. Stop. Set<String>. The strings look like stateNumber_labelIndex.

      • statesVisitedDuringSampleSequence

        protected java.util.Set<java.lang.Integer> statesVisitedDuringSampleSequence

      • verbose

        public static boolean verbose

    • Constructor Detail

      • DecisionProbe

        public DecisionProbe​(DFA dfa)

    • Method Detail

      • getDescription

        public java.lang.String getDescription()
        Return a string like "3:22: ( A {;} | B )" that describes this decision.

      • isReduced

        public boolean isReduced()

      • isCyclic

        public boolean isCyclic()

      • isDeterministic

        public boolean isDeterministic()
        If no states are dead-ends, no alts are unreachable, there are no nondeterminisms unresolved by syn preds, all is ok with decision.

      • analysisOverflowed

        public boolean analysisOverflowed()
        Took too long to analyze a DFA

      • isNonLLStarDecision

        public boolean isNonLLStarDecision()
        Found recursion in > 1 alt

      • getNumberOfStates

        public int getNumberOfStates()
        How many states does the DFA predictor have?

      • getUnreachableAlts

        public java.util.List<java.lang.Integer> getUnreachableAlts()
        Get a list of all unreachable alternatives for this decision. There may be multiple alternatives with ambiguous input sequences, but this is the overall list of unreachable alternatives (either due to conflict resolution or alts w/o accept states).

      • getDanglingStates

        public java.util.Set<DFAState> getDanglingStates()
        return set of states w/o emanating edges and w/o resolving sem preds. These states come about because the analysis algorithm had to terminate early to avoid infinite recursion for example (due to left recursion perhaps).

      • getNonDeterministicAlts

        public java.util.Set<java.lang.Integer> getNonDeterministicAlts()

      • getNonDeterministicAltsForState

        public java.util.List<java.lang.Integer> getNonDeterministicAltsForState​(DFAState targetState)
        Return the sorted list of alts that conflict within a single state. Note that predicates may resolve the conflict.

      • getDFAStatesWithSyntacticallyAmbiguousAlts

        public java.util.Set<DFAState> getDFAStatesWithSyntacticallyAmbiguousAlts()
        Return all DFA states in this DFA that have NFA configurations that conflict. You must report a problem for each state in this set because each state represents a different input sequence.

      • getDisabledAlternatives

        public java.util.Set<java.lang.Integer> getDisabledAlternatives​(DFAState d)
        Which alts were specifically turned off to resolve nondeterminisms? This is different than the unreachable alts. Disabled doesn't mean that the alternative is totally unreachable necessarily, it just means that for this DFA state, that alt is disabled. There may be other accept states for that alt that make an alt reachable.

      • removeRecursiveOverflowState

        public void removeRecursiveOverflowState​(DFAState d)
        If a recursion overflow is resolve with predicates, then we need to shut off the warning that would be generated.

      • getSampleNonDeterministicInputSequence

        public java.util.List<Label> getSampleNonDeterministicInputSequence​(DFAState targetState)
        Return a List<Label> indicating an input sequence that can be matched from the start state of the DFA to the targetState (which is known to have a problem).

      • getInputSequenceDisplay

        public java.lang.String getInputSequenceDisplay​(java.util.List<? extends Label> labels)
        Given List<Label>, return a String with a useful representation of the associated input string. One could show something different for lexers and parsers, for example.

      • getNFAPathStatesForAlt

        public java.util.List<? extends NFAState> getNFAPathStatesForAlt​(int firstAlt,
                                                                 int alt,
                                                                 java.util.List<? extends Label> labels)
        Given an alternative associated with a nondeterministic DFA state, find the path of NFA states associated with the labels sequence. Useful tracing where in the NFA, a single input sequence can be matched. For different alts, you should get different NFA paths. The first NFA state for all NFA paths will be the same: the starting NFA state of the first nondeterministic alt. Imagine (A|B|A|A): 5->9-A->o | 6->10-B->o | 7->11-A->o | 8->12-A->o There are 3 nondeterministic alts. The paths should be: 5 9 ... 5 6 7 11 ... 5 6 7 8 12 ... The NFA path matching the sample input sequence (labels) is computed using states 9, 11, and 12 rather than 5, 7, 8 because state 5, for example can get to all ambig paths. Must isolate for each alt (hence, the extra state beginning each alt in my NFA structures). Here, firstAlt=1.

      • getSemanticContextForAlt

        public SemanticContext getSemanticContextForAlt​(DFAState d,
                                                int alt)
        Each state in the DFA represents a different input sequence for an alt of the decision. Given a DFA state, what is the semantic predicate context for a particular alt.

      • hasPredicate

        public boolean hasPredicate()
        At least one alt refs a sem or syn pred

      • getNondeterministicStatesResolvedWithSemanticPredicate

        public java.util.Set<DFAState> getNondeterministicStatesResolvedWithSemanticPredicate()

      • getIncompletelyCoveredAlts

        public java.util.Map<java.lang.Integer,​java.util.Set<org.antlr.runtime.Token>> getIncompletelyCoveredAlts​(DFAState d)
        Return a list of alts whose predicate context was insufficient to resolve a nondeterminism for state d.

      • issueWarnings

        public void issueWarnings()

      • stripWildCardAlts

        protected void stripWildCardAlts​(java.util.Set<java.lang.Integer> disabledAlts)
        Get the last disabled alt number and check in the grammar to see if that alt is a simple wildcard. If so, treat like an else clause and don't emit the error. Strip out the last alt if it's wildcard.

      • issueRecursionWarnings

        protected void issueRecursionWarnings()

      • reportDanglingState

        public void reportDanglingState​(DFAState d)
        Report the fact that DFA state d is not a state resolved with predicates and yet it has no emanating edges. Usually this is a result of the closure/reach operations being unable to proceed

      • reportNonLLStarDecision

        public void reportNonLLStarDecision​(DFA dfa)
        Report that at least 2 alts have recursive constructs. There is no way to build a DFA so we terminated.

      • reportRecursionOverflow

        public void reportRecursionOverflow​(DFAState d,
                                    NFAConfiguration recursionNFAConfiguration)

      • reportNondeterminism

        public void reportNondeterminism​(DFAState d,
                                 java.util.Set<java.lang.Integer> nondeterministicAlts)

      • reportLexerRuleNondeterminism

        public void reportLexerRuleNondeterminism​(DFAState d,
                                          java.util.Set<java.lang.Integer> nondeterministicAlts)
        Currently the analysis reports issues between token definitions, but we don't print out warnings in favor of just picking the first token definition found in the grammar ala lex/flex.

      • reportNondeterminismResolvedWithSemanticPredicate

        public void reportNondeterminismResolvedWithSemanticPredicate​(DFAState d)

      • reportAltPredicateContext

        public void reportAltPredicateContext​(DFAState d,
                                      java.util.Map<java.lang.Integer,​? extends SemanticContext> altPredicateContext)
        Report the list of predicates found for each alternative; copy the list because this set gets altered later by the method tryToResolveWithSemanticPredicates() while flagging NFA configurations in d as resolved.

      • reportIncompletelyCoveredAlts

        public void reportIncompletelyCoveredAlts​(DFAState d,
                                          java.util.Map<java.lang.Integer,​java.util.Set<org.antlr.runtime.Token>> altToLocationsReachableWithoutPredicate)

      • reachesState

        protected boolean reachesState​(DFAState startState,
                               DFAState targetState,
                               java.util.Set<DFAState> states)
        Given a start state and a target state, return true if start can reach target state. Also, compute the set of DFA states that are on a path from start to target; return in states parameter.

      • getDFAPathStatesToTarget

        protected java.util.Set<DFAState> getDFAPathStatesToTarget​(DFAState targetState)

      • getSampleInputSequenceUsingStateSet

        protected void getSampleInputSequenceUsingStateSet​(State startState,
                                                   State targetState,
                                                   java.util.Set<DFAState> states,
                                                   java.util.List<Label> labels)
        Given a start state and a final state, find a list of edge labels between the two ignoring epsilon. Limit your scan to a set of states passed in. This is used to show a sample input sequence that is nondeterministic with respect to this decision. Return List<Label> as a parameter. The incoming states set must be all states that lead from startState to targetState and no others so this algorithm doesn't take a path that eventually leads to a state other than targetState. Don't follow loops, leading to short (possibly shortest) path.

      • getNFAPath

        protected boolean getNFAPath​(NFAState s,
                             int labelIndex,
                             java.util.List<? extends Label> labels,
                             java.util.List<? super NFAState> path)
        Given a sample input sequence, you usually would like to know the path taken through the NFA. Return the list of NFA states visited while matching a list of labels. This cannot use the usual interpreter, which does a deterministic walk. We need to be able to take paths that are turned off during nondeterminism resolution. So, just do a depth-first walk traversing edges labeled with the current label. Return true if a path was found emanating from state s.

      • getStateLabelIndexKey

        protected java.lang.String getStateLabelIndexKey​(int s,
                                                 int i)

      • getTokenNameForTokensRuleAlt

        public java.lang.String getTokenNameForTokensRuleAlt​(int alt)
        From an alt number associated with artificial Tokens rule, return the name of the token that is associated with that alt.

      • reset

        public void reset()