Package org.apache.derby.impl.sql.compile

Class BinaryRelationalOperatorNode

    • Constructor Detail

      • BinaryRelationalOperatorNode

        BinaryRelationalOperatorNode​(int kind,
                             ValueNode leftOperand,
                             ValueNode rightOperand,
                             boolean forQueryRewrite,
                             ContextManager cm)
                      throws StandardException
        Constructor. DERBY-6185 Query against view with "where name LIKE 'Col1' ESCAPE '\' " failed. Argument forQueryRewrite can be true only if this node has been added by an internal rewrite of the query. This allows binding to be more liberal when checking it against allowed syntax. This parameter will be passed FALSE when a new instance of the node is being created(which is the majority of the cases). But when an existing node is getting cloned, the value of this parameter should be passed as the originalNode.getForQueryRewrite(). Examples of this can be found in Predicate.Java and PredicateList.java
        Parameters:
        kind - The kind of operator
        leftOperand - The left operand
        rightOperand - The right operand
        forQueryRewrite - See method description
        cm - The context manager
        Throws:
        StandardException

      • BinaryRelationalOperatorNode

        BinaryRelationalOperatorNode​(int kind,
                             ValueNode leftOperand,
                             ValueNode rightOperand,
                             InListOperatorNode inListOp,
                             boolean forQueryRewrite,
                             ContextManager cm)
                      throws StandardException
        Same as constructor above except takes a third argument that is an InListOperatorNode. This version is used during IN-list preprocessing to create a "probe predicate" for the IN-list. See InListOperatorNode.preprocess() for more. DERBY-6185 (Query against view with "where name LIKE 'Col1' ESCAPE '\' " failed) 4th argument forQueryRewrite can be true only if this node has been added by an internal rewrite of the query. This allows binding to be more liberal when checking it against allowed syntax. This parameter will be passed FALSE when a new instance of the node is being created(which is the majority of the cases). But when an existing node is getting cloned, the value of this parameter should be passed as the originalNode.getForQueryRewrite(). Examples of this can be found in Predicate.Java and PredicateList.java
        Throws:
        StandardException

    • Method Detail

      • constructorMinion

        private void constructorMinion()

      • getMethodName

        private static java.lang.String getMethodName​(int kind)

      • getOperatorName

        private static java.lang.String getOperatorName​(int kind)

      • getRelOpType

        private int getRelOpType​(int op)

      • getInListOp

        protected InListOperatorNode getInListOp()
        If this rel op was created for an IN-list probe predicate then return the underlying InListOperatorNode. Will return null if this rel op is a "legitimate" relational operator (as opposed to a disguised IN-list). With the exception of nullability checking via the isInListProbeNode() method, all access to this.inListProbeSource MUST come through this method, as this method ensures that the left operand of the inListProbeSource is set correctly before returning it.

      • getColumnOperand

        public ColumnReference getColumnOperand​(Optimizable optTable,
                                        int columnPosition)
        Description copied from interface: RelationalOperator
        Check whether this RelationalOperator is a comparison of the given column with an expression. If so, return the ColumnReference that corresponds to the given column, and that is on one side of this RelationalOperator or the other (this method copes with the column being on either side of the operator). If the given column does not appear by itself on one side of the comparison, the method returns null.
        Specified by:
        getColumnOperand in interface RelationalOperator
        Parameters:
        optTable - An Optimizable for the base table the column is in
        columnPosition - The ordinal position of the column (one-based)
        Returns:
        The ColumnReference on one side of this RelationalOperator that represents the given columnPosition. Returns null if no such ColumnReference exists by itself on one side of this RelationalOperator.
        See Also:
        RelationalOperator.getColumnOperand(org.apache.derby.iapi.sql.compile.Optimizable, int)

      • getExpressionOperand

        public ValueNode getExpressionOperand​(int tableNumber,
                                      int columnPosition,
                                      Optimizable ft)
        Description copied from interface: RelationalOperator
        Check whether this RelationalOperator is a comparison of the given column with an expression. If so, return the expression the column is being compared to.
        Specified by:
        getExpressionOperand in interface RelationalOperator
        Parameters:
        tableNumber - The table number of the base table the column is in
        columnPosition - The ordinal position of the column (one-based)
        ft - We'll look for the column in all tables at and beneath ft. This is useful if ft is, say, a ProjectRestrictNode over a subquery-- then we want to look at all of the FROM tables in the subquery to try to find the right column.
        Returns:
        The ValueNode for the expression the column is being compared to - null if the column is not being compared to anything.
        See Also:
        RelationalOperator.getExpressionOperand(int, int, org.apache.derby.iapi.sql.compile.Optimizable)

      • getOperand

        public ValueNode getOperand​(ColumnReference cRef,
                            int refSetSize,
                            boolean otherSide)
        Description copied from interface: RelationalOperator
        Find the operand (left or right) that points to the same table as the received ColumnReference, and then return either that operand or the "other" operand, depending on the value of otherSide. This presumes it will be found only on one side. If not found, it will return null.
        Specified by:
        getOperand in interface RelationalOperator
        Parameters:
        cRef - The ColumnReference for which we're searching.
        refSetSize - Size of the referenced map for the predicate represented by this RelationalOperator node. This is used for storing base table numbers when searching for cRef.
        otherSide - Assuming we find an operand that points to the same table as cRef, then we will return the *other* operand if otherSide is true; else we'll return the operand that matches cRef.
        See Also:
        RelationalOperator.getOperand(org.apache.derby.impl.sql.compile.ColumnReference, int, boolean)

      • keyColumnOnLeft

        protected boolean keyColumnOnLeft​(Optimizable optTable)
        Return true if a key column for the given table is found on the left side of this operator, false if it is found on the right side of this operator. NOTE: This method assumes that a key column will be found on one side or the other. If you don't know whether a key column exists, use the columnOnOneSide() method (below).
        Parameters:
        optTable - The Optimizable table that we're looking for a key column on.
        Returns:
        true if a key column for the given table is on the left side of this operator, false if one is found on the right side of this operator.

      • columnOnOneSide

        protected int columnOnOneSide​(Optimizable optTable)
        Determine whether there is a column from the given table on one side of this operator, and if so, which side is it on?
        Parameters:
        optTable - The Optimizable table that we're looking for a key column on.
        Returns:
        LEFT if there is a column on the left, RIGHT if there is a column on the right, NEITHER if no column found on either side.

      • getAbsoluteColumnPosition

        private int getAbsoluteColumnPosition​(Optimizable optTable)
        Get the absolute 0-based column position of the ColumnReference from the conglomerate for this Optimizable.
        Parameters:
        optTable - The Optimizable
        Returns:
        The absolute 0-based column position of the ColumnReference

      • isQualifier

        public boolean isQualifier​(Optimizable optTable,
                           boolean forPush)
                    throws StandardException
        Description copied from interface: RelationalOperator
        Return true if this operator can be compiled into a Qualifier for the given Optimizable table. This means that there is a column from that table on one side of this relop, and an expression that does not refer to the table on the other side of the relop. Note that this method has two uses: 1) see if this operator (or more specifically, the predicate to which this operator belongs) can be used as a join predicate (esp. for a hash join), and 2) see if this operator can be pushed to the target optTable. We use the parameter "forPush" to distinguish between the two uses because in some cases (esp. situations where we have subqueries) the answer to "is this a qualifier?" can differ depending on whether or not we're pushing. In particular, for binary ops that are join predicates, if we're just trying to find an equijoin predicate then this op qualifies if it references either the target table OR any of the base tables in the table's subtree. But if we're planning to push the predicate down to the target table, this op only qualifies if it references the target table directly. This difference in behavior is required because in case 1 (searching for join predicates), the operator remains at its current level in the tree even if its operands reference nodes further down; in case 2, though, we'll end up pushing the operator down the tree to child node(s) and that requires additional logic, such as "scoping" consideration. Until that logic is in place, we don't search a subtree if the intent is to push the predicate to which this operator belongs further down that subtree. See BinaryRelationalOperatorNode for an example of where this comes into play.
        Specified by:
        isQualifier in interface RelationalOperator
        Parameters:
        optTable - The Optimizable table in question.
        forPush - Are we asking because we're trying to push?
        Returns:
        true if this operator can be compiled into a Qualifier for the given Optimizable table.
        Throws:
        StandardException - Thrown on error
        See Also:
        RelationalOperator.isQualifier(org.apache.derby.iapi.sql.compile.Optimizable, boolean)

      • booleanSelectivity

        protected double booleanSelectivity​(Optimizable optTable)
                             throws StandardException
        Return 50% if this is a comparison with a boolean column, a negative selectivity otherwise.
        Throws:
        StandardException

      • getReceiverInterfaceName

        java.lang.String getReceiverInterfaceName()
        The methods generated for this node all are on Orderable. Overrides this method in BooleanOperatorNode for code generation purposes.
        Overrides:
        getReceiverInterfaceName in class BinaryOperatorNode

      • newBool

        private ValueNode newBool​(boolean b)
                   throws StandardException
        Create a Boolean constant node with a specified value.
        Parameters:
        b - the value of the constant
        Returns:
        a node representing a Boolean constant
        Throws:
        StandardException

      • getNegationNode

        private int getNegationNode()

      • getKindForSwap

        private int getKindForSwap()
        Return the node type that must be used in order to construct an equivalent expression if the operands are swapped. For symmetric operators (= and <>), the same node type is returned. Otherwise, the direction of the operator is switched in order to preserve the meaning (for instance, a node representing less-than will return the node type for greater-than).
        Returns:
        a node type that preserves the meaning of the expression if the operands are swapped

      • usefulStartKey

        protected boolean usefulStartKey​(boolean columnOnLeft)
        is this is useful start key? for example a predicate of the from column Lessthan 5 is not a useful start key but is a useful stop key. However 5 Lessthan column is a useful start key.
        Parameters:
        columnOnLeft - is true if the column is the left hand side of the binary operator.

      • isRelationalOperator

        boolean isRelationalOperator()
        Description copied from class: ValueNode
        Returns true if this ValueNode is a relational operator. Relational Operators are <, <=, =, >, >=, <> as well as IS NULL and IS NOT NULL. This is the preferred way of figuring out if a ValueNode is relational or not.
        Overrides:
        isRelationalOperator in class ValueNode
        See Also:
        ValueNode.isRelationalOperator()

      • optimizableEqualityNode

        boolean optimizableEqualityNode​(Optimizable optTable,
                                int columnNumber,
                                boolean isNullOkay)
                         throws StandardException
        Description copied from class: ValueNode
        Return true if the predicate represents an optimizable equality node. an expression is considered to be an optimizable equality node if all the following conditions are met:
        1. the operator is an = or IS NULL operator
        2. one of the operands is a column specified by optTable/columnNumber
        3. Both operands are not the same column; i.e tab.col = tab.col
        4. There are no implicit varchar comparisons of the operands; i.e either both operands are string like (varchar, char, longvarchar) or neither operand is string like
        Overrides:
        optimizableEqualityNode in class ValueNode
        Parameters:
        optTable - the table being optimized. Column reference must be from this table.
        columnNumber - the column number. One of the operands of this predicate must be the column number specified by optTable/columnNumber
        isNullOkay - if set to true we also consider IS NULL predicates; otherwise consider only = predicates.
        Throws:
        StandardException
        See Also:
        ValueNode.optimizableEqualityNode(org.apache.derby.iapi.sql.compile.Optimizable, int, boolean)

      • implicitVarcharComparison

        private boolean implicitVarcharComparison()
                                   throws StandardException
        Return whether or not this binary relational predicate requires an implicit (var)char conversion. This is important when considering hash join since this type of equality predicate is not currently supported for a hash join.
        Returns:
        Whether or not an implicit (var)char conversion is required for this binary relational operator.
        Throws:
        StandardException - Thrown on error

      • getScopedOperand

        ValueNode getScopedOperand​(int whichSide,
                           JBitSet parentRSNsTables,
                           ResultSetNode childRSN,
                           int[] whichRC)
                    throws StandardException
        Take a ResultSetNode and return a column reference that is scoped for for the received ResultSetNode, where "scoped" means that the column reference points to a specific column in the RSN. This is used for remapping predicates from an outer query down to a subquery. For example, assume we have the following query: select * from (select i,j from t1 union select i,j from t2) X1, (select a,b from t3 union select a,b from t4) X2 where X1.j = X2.b; Then assume that this BinaryRelationalOperatorNode represents the "X1.j = X2.b" predicate 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 what we want to do in this method is map one of the operands X1.j or X2.b (depending on the 'whichSide' parameter) to the childRSN, if possible. Note that in our example, "X2.b" should _NOT_ be mapped because it doesn't apply to the childRSN for the subquery "select i,j from t1"; thus we should leave it as it is. "X1.j", however, _does_ need to be scoped, and so this method will return a ColumnReference pointing to "T1.j" (or whatever the corresponding column in T1 is). ASSUMPTION: We should only get to this method if we know that exactly one operand in the predicate to which this operator belongs can and should be mapped to the received childRSN.
        Parameters:
        whichSide - The operand are we trying to scope (LEFT or RIGHT)
        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 operand to a ResultSetNode to which it doesn't apply.
        childRSN - The result set node to 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 column reference scoped to the received childRSN, if possible. If the operand is a ColumnReference that is not supposed to be scoped, we return a _clone_ of the reference--this is necessary because the reference is going to be pushed to two places (left and right children of the parentRSN) and if both children are referencing the same instance of the column reference, they'll interfere with each other during optimization.
        Throws:
        StandardException

      • valNodeReferencesOptTable

        private boolean valNodeReferencesOptTable​(ValueNode valNode,
                                          Optimizable optTable,
                                          boolean forPush,
                                          boolean walkOptTableSubtree)
        Determine whether or not the received ValueNode (which will usually be a ColumnReference) references either the received optTable or else a base table in the subtree beneath that optTable.
        Parameters:
        valNode - The ValueNode that has the reference(s).
        optTable - The table/subtree node to which we're trying to find a reference.
        forPush - Whether or not we are searching with the intent to push this operator to the target table.
        walkOptTableSubtree - Should we walk the subtree beneath optTable to find base tables, or not? Will be false if we've already done it for the left operand and now we're here for the right operand.
        Returns:
        True if valNode contains a reference to optTable or to a base table in the subtree beneath optTable; false otherwise.

      • initBaseTableVisitor

        private void initBaseTableVisitor​(int numTablesInQuery,
                                  boolean initOptBaseTables)
        Initialize the fields used for retrieving base tables in subtrees, which allows us to do a more extensive search for table references. If the fields have already been created, then just reset their values.
        Parameters:
        numTablesInQuery - Used for creating JBitSets that can hold table numbers for the query.
        initOptBaseTables - Whether or not we should clear out or initialize the optBaseTables bit set.

      • buildTableNumList

        private void buildTableNumList​(Optimizable ft,
                               boolean forPush)
                        throws StandardException
        Create a set of table numbers to search when trying to find which (if either) of this operator's operands reference the received target table. At the minimum this set should contain the target table's own table number. After that, if we're _not_ attempting to push this operator (or more specifically, the predicate to which this operator belongs) to the target table, we go on to search the subtree beneath the target table and add any base table numbers to the searchable list.
        Parameters:
        ft - Target table for which we're building the search list.
        forPush - Whether or not we are searching with the intent to push this operator to the target table.
        Throws:
        StandardException

      • isSameNodeKind

        boolean isSameNodeKind​(ValueNode o)
        Description copied from class: ValueNode
        Some node classes represent several logical node types (to reduce footprint), which we call kinds. This means that implementations of ValueNode.isEquivalent(org.apache.derby.impl.sql.compile.ValueNode) cannot always just use instanceof to check if the other node represents the same kind. Hence this method needs to be overridden by all node classes that represent several kinds. This default implementation does not look at kinds. It is only called from implementations of isEquivalent.
        Overrides:
        isSameNodeKind in class BinaryOperatorNode
        Parameters:
        o - The other value node whose kind we want to compare with.
        Returns:
        true if this and o represent the same logical node type, i.e. kind.