Class Predicate

    • Field Detail

      • pushable

        boolean pushable
      • referencedSet

        JBitSet referencedSet
      • equivalenceClass

        int equivalenceClass
      • indexPosition

        int indexPosition
      • startKey

        protected boolean startKey
      • stopKey

        protected boolean stopKey
      • isQualifier

        protected boolean isQualifier
      • searchClauses

        private java.util.Set<java.lang.Integer> searchClauses
      • scoped

        private boolean scoped
    • Constructor Detail

      • Predicate

        Predicate​(AndNode andNode,
                  JBitSet referencedSet,
                  ContextManager cm)
        Constructor.
        Parameters:
        andNode - The top of the predicate
        referencedSet - Bit map of referenced tables
        cm - The context manager
    • Method Detail

      • hasEqualOnColumnList

        public int hasEqualOnColumnList​(int[] baseColumnPositions,
                                        Optimizable optTable)
                                 throws StandardException
        Description copied from interface: OptimizablePredicate
        Returns if the predicate involves an equal operator on one of the columns specified in the baseColumnPositions.
        Specified by:
        hasEqualOnColumnList in interface OptimizablePredicate
        Parameters:
        baseColumnPositions - the column numbers on which the user wants to check if the equality condition exists.
        optTable - the table for which baseColumnPositions are given.
        Returns:
        returns the index into baseColumnPositions of the column that has the equality operator.
        Throws:
        StandardException
      • getIndexPosition

        public int getIndexPosition()
        Description copied from interface: OptimizablePredicate
        Get the position of the index column that this predicate restricts. NOTE: This assumes that this predicate is part of an OptimizablePredicateList, and that classify() has been called on the OptimizablePredicateList.
        Specified by:
        getIndexPosition in interface OptimizablePredicate
        Returns:
        The index position that this predicate restricts (zero-based)
        See Also:
        OptimizablePredicate.getIndexPosition()
      • compareTo

        public int compareTo​(Predicate other)
        Specified by:
        compareTo in interface java.lang.Comparable<Predicate>
      • getAndNode

        AndNode getAndNode()
        Return the andNode.
        Returns:
        AndNode The andNode.
      • setAndNode

        void setAndNode​(AndNode andNode)
        Set the andNode.
        Parameters:
        andNode - The new andNode.
      • getPushable

        boolean getPushable()
        Return the pushable.
        Returns:
        boolean Whether or not the predicate is pushable.
      • setPushable

        void setPushable​(boolean pushable)
        Set whether or not this predicate is pushable. This method is intended for use when creating a copy of the predicate, ex for predicate pushdown. We choose not to add this assignment to copyFields() because the comments for that method say that it should copy all fields _except_ the two specified at init time; "pushable" is one of the two specified at init time.
        Parameters:
        pushable - Whether or not the predicate is pushable.
      • getReferencedSet

        JBitSet getReferencedSet()
        Return the referencedSet.
        Returns:
        JBitSet The referencedSet.
      • setEquivalenceClass

        void setEquivalenceClass​(int equivalenceClass)
        Set the equivalence class, if any, for this predicate.
        Parameters:
        equivalenceClass - The equivalence class for this predicate.
      • getEquivalenceClass

        int getEquivalenceClass()
        Get the equivalenceClass for this predicate.
        Returns:
        The equivalenceClass for this predicate.
      • categorize

        void categorize()
                 throws StandardException
        Categorize this predicate. Initially, this means building a bit map of the referenced tables for each predicate.
        Throws:
        StandardException - Thrown on error
      • getRelop

        RelationalOperator getRelop()
        Get the RelationalOperator on the left side of the AND node, if there is one. If the left side is not a RelationalOperator, return null.
        Returns:
        The RelationalOperator on the left side of the AND node, if any.
      • isOrList

        final boolean isOrList()
      • isStoreQualifier

        final boolean isStoreQualifier()
        Is this predicate a possible Qualifier for store?

        Current 2 types of predicates can be pushed to store: 1) RelationalOperator - represented with by left operand as instance of RelationalOperator. 2) A single And'd term of a list of OR terms represented by left operand as instance of OrNode. More checking specific operator's terms to see if they are finally pushable to store. In the final push at execution each term of the AND or OR must be a Relational operator with a column reference on one side and a constant on the other.

        Returns:
        true if term is wither a AND of a RelationalOperator, or an OR of one or more Relational Operators.
        Throws:
        StandardException - Standard exception policy.
      • isPushableOrClause

        final boolean isPushableOrClause​(Optimizable optTable)
                                  throws StandardException
        Is this predicate an pushable OR list?

        Does the predicate represent a AND'd list of OR term's, all of which are pushable. To be pushable each of OR terms must be a legal qualifier, which is a column reference on one side of a Relational operator and a constant on the other.

        Returns:
        true if the predicate is a pushable set of OR clauses.
        Throws:
        StandardException - Standard exception policy.
      • transitiveSearchClauseAdded

        boolean transitiveSearchClauseAdded​(RelationalOperator ro)
        Return whether or not this predicate has been used to add a new search clause of the specified type via transitive closure. NOTE: This can only be true if this is an equijoin between 2 column references.
        Parameters:
        ro - The search clause that we are currently considering as the source for transitive closure
        Returns:
        Whether or not this predicate has been used to add a new search clause of the specified type via transitive closure.
      • setTransitiveSearchClauseAdded

        void setTransitiveSearchClauseAdded​(RelationalOperator ro)
        Mark this predicate as having been used to add a new predicate of the specified type via transitive closure on search clauses.
        Parameters:
        ro - The search clause that we are currently considering as the source for transitive closure
      • getStartOperator

        int getStartOperator​(Optimizable optTable)
        Get the start operator for this predicate for a scan.
        Parameters:
        optTable - The optimizable table, so we can tell which side of the operator the search column is on.
        Returns:
        The start operator for a start key on this column.
      • getStopOperator

        int getStopOperator​(Optimizable optTable)
      • setIndexPosition

        void setIndexPosition​(int indexPosition)
        Set the position of the index column that this predicate restricts
        Parameters:
        indexPosition - The position of the index column that this predicate restricts.
      • clearScanFlags

        void clearScanFlags()
        Clear the start/stop position and qualifier flags
      • toString

        public java.lang.String toString()
        Convert this object to a String. See comments in QueryTreeNode.java for how this should be done for tree printing.
        Overrides:
        toString in class QueryTreeNode
        Returns:
        This object as a String
      • binaryRelOpColRefsToString

        java.lang.String binaryRelOpColRefsToString()
        Get a string version of the column references for this predicate IF it's a binary relational operator. We only print out the names of the operands if they are column references; otherwise we just print a dummy value. This is for debugging purposes only--it's a convenient way to see what columns the predicate is referencing, especially when tracing through code and printing assert failure.
      • printSubNodes

        void printSubNodes​(int depth)
        Prints the sub-nodes of this object. See QueryTreeNode.java for how tree printing is supposed to work.
        Overrides:
        printSubNodes in class QueryTreeNode
        Parameters:
        depth - The depth of this node in the tree
      • copyFields

        void copyFields​(Predicate otherPred)
        Copy all fields of this Predicate (except the two that are set from 'init').
      • pushableToSubqueries

        protected boolean pushableToSubqueries()
                                        throws StandardException
        Determine whether or not this predicate is eligible for push-down into subqueries. Right now the only predicates we consider to be eligible are those which 1) are Binary Relational operator nodes and 2) have a column reference on BOTH sides, each of which has a reference to a base table somewhere beneath it.
        Returns:
        Whether or not this predicate is eligible to be pushed into subqueries.
        Throws:
        StandardException
      • isJoinPredicate

        protected boolean isJoinPredicate()
        Is this predicate a join predicate? In order to be so, it must be a binary relational operator node that has a column reference on both sides.
        Returns:
        Whether or not this is a join predicate.
      • getPredScopedForResultSet

        protected Predicate getPredScopedForResultSet​(JBitSet parentRSNsTables,
                                                      ResultSetNode childRSN,
                                                      int[] whichRC)
                                               throws StandardException
        If this predicate's operator is a BinaryRelationalOperatorNode, then look at the operands and return a new, equivalent predicate that is "scoped" to the received ResultSetNode. By "scoped" we mean that the operands, which shold be column references, have been mapped to the appropriate result columns in the received RSN. This is useful for pushing predicates from outer queries down into inner queries, in which case the column references need to be remapped. For example, let V1 represent select i,j from t1 UNION select i,j from t2 and V2 represent select a,b from t3 UNION select a,b from t4 Then assume we have the following query: select * from V1, V2 where V1.j = V2.b Let's further assume that this Predicate object represents the "V1.j = V2.b" operator and that the childRSN we received as a parameter represents one of the subqueries to which we want to push the predicate; let's say it's: select i,j from t1 Then this method will return a new predicate whose binary operator represents the expression "T1.j = V2.b" (that is, V1.j will be mapped to the corresponding column in T1). For more on how that mapping is made, see the "getScopedOperand()" method in BinaryRelationalOperatorNode.java. ASSUMPTION: We should only get to this method if we know that at least one operand in this predicate can and should be mapped to the received childRSN. For an example of where that check is made, see the pushOptPredicate() method in SetOperatorNode.java.
        Parameters:
        parentRSNsTables - Set of all table numbers referenced by the ResultSetNode that is _parent_ to the received childRSN. We need this to make sure we don't scope the operands to a ResultSetNode to which they don't apply.
        childRSN - The result set node for which we want to create a scoped predicate.
        whichRC - If not -1 then this tells us which ResultColumn in the received childRSN we need to use for the scoped predicate; if -1 then the column position of the scoped column reference will be stored in this array and passed back to the caller.
        Returns:
        A new predicate whose operands have been scoped to the received childRSN.
        Throws:
        StandardException
      • markAsScopedForPush

        protected void markAsScopedForPush()
        Indicate that this predicate is a scoped copy of some other predicate (i.e. it was created as the result of a call to getPredScopedForResultSet() on some other predicate).
      • isScopedForPush

        protected boolean isScopedForPush()
        Return whether or not this predicate is a scoped copy of another predicate.
      • remapScopedPred

        protected boolean remapScopedPred()
        When remapping a "normal" (i.e. non-scoped) predicate both of the predicate's operands are remapped and that's it. But when remapping a scoped predicate, things are slightly different. This method handles remapping of scoped predicates. We know that, for a scoped predicate, exactly one operand has been scoped for a specific target result set; the other operand is pointing to some other instance of FromTable with which the target result set is to be joined (see getScopedOperand() in BinaryRelationalOperatorNode.java). For every level of the query through which the scoped predicate is pushed, we have to perform a remap operation of the scoped operand. We do *not*, however, remap the non-scoped operand. The reason is that the non-scoped operand is already pointing to the result set against which it must be evaluated. As the scoped predicate is pushed down the query tree, the non-scoped operand should not change where it's pointing and thus should not be remapped. For example, assume we have a query whose tree has the following form: SELECT[0] / \ PRN PRN | | SELECT[4] UNION | / \ PRN SELECT[1] SELECT[2] | | | [FBT:T1] PRN PRN | | SELECT[3] [FromBaseTable:T2] | PRN | [FromBaseTable:T3] Assume also that we have some predicate "SELECT[4].i = .j". If the optimizer decides to push the predicate to the UNION node, it (the predicate) will be scoped to the UNION's children, yielding something like "SELECT[4].i = SELECT[1].j" for the left child and "SELECT[4].i = SELECT[2].j" for the right child. These scoped predicates will then be pushed to the PRNs above SELECT[3] and T2, respectively. As part of that pushing process a call to PRN.pushOptPredicate() will occur, which brings us to this method. So let's assume we're here for the scoped predicate "SELECT[4].i = SELECT[1].j". Then we want to remap the scoped operand, "SELECT[1].j", so that it will point to the correct column in "SELECT[3]". We do NOT, however, want to remap the non-scoped operand "SELECT[4].i" because that operand is already pointing to the correct result set--namely, to a column in SELECT[4]. That non-scoped operand should not change regardless of how far down the UNION subtree the scoped predicate is pushed. If we did try to remap the non-scoped operand, it would end up pointing to result sets too low in the tree, which could lead to execution-time errors. So when we remap a scoped predicate, we have to make sure we only remap the scoped operand. That's what this method does.
        Returns:
        True if this predicate is a scoped predicate, in which case we performed a one-sided remap. False if the predicate is not scoped; the caller can then make the calls to perform a "normal" remap on this predicate.
      • isScopedToSourceResultSet

        protected boolean isScopedToSourceResultSet()
                                             throws StandardException
        Return true if this predicate is scoped AND the scoped operand is a ColumnReference that points to a source result set. If the scoped operand is not a ColumnReference that points to a source result set then it must be pointing to some kind of expression, such as a literal (ex. 'strlit'), an aggregate value (ex. "count(*)"), or the result of a function (ex. "sin(i)") or operator (ex. "i+1"). This method is used when pushing predicates to determine how far down the query tree a scoped predicate needs to be pushed to allow for successful evaluation of the scoped operand. If the scoped operand is not pointing to a source result set then it should not be pushed any further down tree. The reason is that evaluation of the expression to which the operand is pointing may depend on other values from the current level in the tree (ex. "sin(i)" depends on the value of "i", which could be a column at the predicate's current level). If we pushed the predicate further down, those values could become inaccessible, leading to execution-time errors. If, on the other hand, the scoped operand *is* pointing to a source result set, then we want to push it further down the tree until it reaches that result set, which allows evaluation of this predicate to occur as close to store as possible. This method doesn't actually do the push, it just returns "true" and then the caller can push as appropriate.
        Throws:
        StandardException
      • isRelationalOpPredicate

        protected boolean isRelationalOpPredicate()
        Return whether or not this predicate corresponds to a legitimate relational operator.
        Returns:
        False if there is no relational operator for this predicate OR if this predicate is an internal "probe predicate" (in which case it "looks" like we have a relational operator but in truth it's a disguised IN-list operator). True otherwise.
      • isInListProbePredicate

        protected boolean isInListProbePredicate()
        Return whether or not this predicate is an IN-list probe predicate.
      • getSourceInList

        protected InListOperatorNode getSourceInList()
        If this predicate corresponds to an IN-list, return the underlying InListOperatorNode from which it was built. There are two forms to check for: 1. This predicate is an IN-list "probe predicate", in which case the underlying InListOpNode is stored within the binary relational operator that is the left operand of this predicate's AND node. 2. This predicate corresponds to an IN-list that could _not_ be transformed into a "probe predicate" (i.e. the IN-list contains one or more non-parameter, non-constant values). In that case the underlying InListOpNode is simply the left operand of this predicate's AND node. If this predicate does not correspond to an IN-list in any way, this method will return null.
      • getSourceInList

        protected InListOperatorNode getSourceInList​(boolean probePredOnly)
        Does the work of getSourceInList() above, but can also be called directly with an argument to indicate whether or not we should limit ourselves to probe predicates.
        Parameters:
        probePredOnly - If true, only get the source IN list for this predicate *if* it is an IN-list probe predicate. If false, return the underlying InListOperatorNode (if it exists) regardless of whether this is a probe predicate or an un-transformed IN-list pred.
        Returns:
        Underlying InListOp for this predicate (depending on the value of probePredOnly), or null if this predicate does not correspond to an IN-list in any way.