void MultiJoin::synthLogPropWithMJReuse(NormWA * normWAPtr)
{
// Check to see whether this GA has already been associated
// with a logExpr for synthesis. If so, no need to resynthesize
// for this equivalent log. expression.
if (getGroupAttr()->existsLogExprForSynthesis())
{
Join * joinExprForSynth =
(Join *) getGroupAttr()->getLogExprForSynthesis();
if(joinExprForSynth->isJoinFromMJSynthLogProp())
return;
}
NABoolean reUseMJ = TRUE;
CMPASSERT ( (jbbSubset_.getGB() == NULL_CA_ID));
const CANodeIdSet & jbbcs = jbbSubset_.getJBBCs();
// Instead of always picking the first JBBC as the right child
// pick the one with minimum JBBC connections. This will avoid
// all unnecessary crossproducts
CANodeId jbbcRight;
jbbcRight = jbbcs.getJBBCwithMinConnectionsToThisJBBSubset();
CANodeIdSet right(jbbcRight);
CANodeIdSet left(jbbcs);
left -= jbbcRight;
Join* join = splitSubset(*(left.jbbcsToJBBSubset()),
*(right.jbbcsToJBBSubset()),
reUseMJ);
//if the left is a MultiJoin, synthesize it using reUse
//this has to be done before join->synthLogProp to avoid
//calling MultiJoin::synthLogProp on the left MultiJoin
//because that does not reUse
if(left.entries() > 1)
{
RelExpr * leftRelExpr = join->child(0)->castToRelExpr();
if(leftRelExpr &&
leftRelExpr->getOperator() == REL_MULTI_JOIN)
((MultiJoin *) leftRelExpr)->synthLogPropWithMJReuse(normWAPtr);
}
join->synthLogProp(normWAPtr);
join->setJoinFromMJSynthLogProp();
getGroupAttr()->setLogExprForSynthesis(join);
jbbSubset_.setSubsetMJ(this);
CMPASSERT ( getGroupAttr()->getNumJoinedTables() >= getArity());
}
void MultiJoin::synthLogProp(NormWA * normWAPtr)
{
// Check to see whether this GA has already been associated
// with a logExpr for synthesis. If so, no need to resynthesize
// for this equivalent log. expression.
if (getGroupAttr()->existsLogExprForSynthesis()) return;
CMPASSERT ( (jbbSubset_.getGB() == NULL_CA_ID));
const CANodeIdSet & jbbcs = jbbSubset_.getJBBCs();
// Instead of always picking the first JBBC as the right child
// pick the one with minimum JBBC connections. This will avoid
// all unnecessary crossproducts
CANodeId jbbcRight;
jbbcRight = jbbcs.getJBBCwithMinConnectionsToThisJBBSubset();
CANodeIdSet right(jbbcRight);
CANodeIdSet left(jbbcs);
left -= jbbcRight;
Join* join = splitSubset(*(left.jbbcsToJBBSubset()),
*(right.jbbcsToJBBSubset()));
join->synthLogProp(normWAPtr);
getGroupAttr()->setLogExprForSynthesis(join);
join->setJoinFromMJSynthLogProp();
jbbSubset_.setSubsetMJ(this);
CASortedList * synthLogPropPath =
new (CmpCommon::statementHeap())
CASortedList(CmpCommon::statementHeap(), jbbcs.entries());
synthLogPropPath->insert((*(left.jbbcsToJBBSubset()->getSynthLogPropPath())));
synthLogPropPath->insert(right.getFirst());
jbbSubset_.setSynthLogPropPath(synthLogPropPath);
CMPASSERT ( getGroupAttr()->getNumJoinedTables() >= getArity());
}
Join* MultiJoin::createLeftLinearJoinTree
(const NAList<CANodeIdSet> * const leftDeepJoinSequence,
NAList<MJJoinDirective *> * joinDirectives) const
{
Join* result = NULL;
Join* currentJoin=NULL;
NABoolean reUseMultiJoins = FALSE;
//Set of all JBBCs in this multi-join.
//This set will be broken up to make the join tree
//representing the substitue.
//The loop below will construct the join tree,
//starting from the top join.
CANodeIdSet childSet = getJBBSubset().getJBBCs();
// variables used in loop below
MultiJoin * currentMJoin = (MultiJoin *) this;
// in an iteration this is the parent join
// e.g. when we are create JOIN3, this will
// be JOIN2
Join * parentJoin = NULL;
#ifdef _DEBUG
if ( CmpCommon::getDefault( NSK_DBG ) == DF_ON &&
CmpCommon::getDefault( NSK_DBG_MJRULES_TRACKING ) == DF_ON )
{
// LCOV_EXCL_START - dpm
CURRCONTEXT_OPTDEBUG->stream() << "Following is left deep join sequence: " << endl;
CURRCONTEXT_OPTDEBUG->stream() << endl;
// LCOV_EXCL_STOP
}
#endif
UInt32 numJoinChildren = leftDeepJoinSequence->entries();
CANodeId currentTable = NULL_CA_ID;
for (UInt32 i = 0; i < (numJoinChildren-1); i++)
{
//create JBBSubset representing a comprising component of the
//leftDeepJoinSequence.
JBBSubset * joinRightChild = ((*leftDeepJoinSequence)[i]).computeJBBSubset();
MJJoinDirective * joinDirective = (*joinDirectives)[i];
//remove all tables that will become right side of join
childSet.remove((*leftDeepJoinSequence)[i]);
#ifdef _DEBUG
//print the right child of the current join
if ( CmpCommon::getDefault( NSK_DBG ) == DF_ON &&
CmpCommon::getDefault( NSK_DBG_MJRULES_TRACKING ) == DF_ON )
{
CURRCONTEXT_OPTDEBUG->stream() << ((*leftDeepJoinSequence)[i]).getText() << endl; // LCOV_EXCL_LINE - dpm
}
#endif
//Get JBBSubset for left side of join
JBBSubset * joinLeftChild = childSet.computeJBBSubset();
//create the join by doing a split of the multi-join
currentJoin = currentMJoin->splitSubset(*joinLeftChild, *joinRightChild, reUseMultiJoins);
joinDirective->setupJoin(currentJoin);
if ( CmpCommon::getDefault(COMP_BOOL_120) == DF_OFF)
currentJoin->updatePotential(-3);
// if this is the first iteration
// set the result to be the top join
if (i == 0)
result = currentJoin;
//set the current multi-join to the left child of the
//join just created
//change getChild to child().getPointer
currentMJoin = (MultiJoin*) currentJoin->getChild(0);
//if there was a parent join, set the left child to
//point to the new join we just created i.e. currentJoin.
if (parentJoin)
parentJoin->setChild(0,currentJoin);
//set currentJoin to be the parent for the next iteration
parentJoin = currentJoin;
}
#ifdef _DEBUG
//print the left most child
if ( CmpCommon::getDefault( NSK_DBG ) == DF_ON &&
CmpCommon::getDefault( NSK_DBG_MJRULES_TRACKING ) == DF_ON )
{
// LCOV_EXCL_START - dpm
CURRCONTEXT_OPTDEBUG->stream() << ((*leftDeepJoinSequence)[(numJoinChildren-1)]).getText() << endl;
CURRCONTEXT_OPTDEBUG->stream() << endl;
// LCOV_EXCL_STOP
}
#endif
// end - construct the join tree
// synth the join
result->synthLogProp();
//if the right child of the top-most join is a multi-Join,
//synthesize_it
if(result->child(1))
if(result->child(1)->getOperatorType()==REL_MULTI_JOIN)
result->child(1)->synthLogProp();
// synth the left child too
result->child(0)->synthLogProp();
return result;
} // MultiJoin::createLeftLinearJoinTree()
// This method forms the join expression for join on JBBC specified by jbbcId
// inputEstLogProp should not be cacheable
Join * AppliedStatMan::formJoinExprForJoinOnJBBC(
CANodeIdSet jbbSubset,
CANodeId jbbcId,
const ValueIdSet * jbbcLocalPreds,
const ValueIdSet * joinPreds,
const EstLogPropSharedPtr& inputEstLogProp,
const NABoolean cacheable)
{
NABoolean origInputIsCacheable = inputEstLogProp->isCacheable();
if(origInputIsCacheable)
{
inputEstLogProp->setCacheableFlag(FALSE);
CCMPASSERT("Expecting Non Cacheable Input");
}
RelExpr * jbbcExpr = getExprForCANodeId(jbbcId, inputEstLogProp, jbbcLocalPreds);
jbbcExpr->getGroupAttr()->outputLogProp(inputEstLogProp);
RelExpr * jbbSubsetExpr = jbbSubset.jbbcsToJBBSubset()->getPreferredJoin();
if(!jbbSubsetExpr)
if(jbbSubset.entries()==1)
if(!inputEstLogProp->isCacheable())
{
inputEstLogProp->setCacheableFlag(TRUE);
jbbSubsetExpr = getExprForCANodeId(jbbSubset.getFirst(), inputEstLogProp);
inputEstLogProp->setCacheableFlag(FALSE);
}
else
jbbSubsetExpr = getExprForCANodeId(jbbSubset.getFirst(), inputEstLogProp);
else
{
CCMPASSERT("No Subset expression, need at least one entry in set");
}
RelExpr * leftChildExpr = jbbSubsetExpr;
RelExpr * rightChildExpr = jbbcExpr;
GroupAttributes * galeft = jbbSubsetExpr->getGroupAttr();
GroupAttributes * garight = jbbcExpr->getGroupAttr();
// xxx
JBBC * jbbc = jbbcId.getNodeAnalysis()->getJBBC();
Join * jbbcParentJoin = jbbc->getOriginalParentJoin();
ValueIdSet leftOuterJoinFilterPreds;
Join * joinExpr = NULL;
if(jbbcParentJoin)
{
if(jbbcParentJoin->isSemiJoin())
joinExpr = new STMTHEAP Join(leftChildExpr, rightChildExpr, REL_SEMIJOIN, NULL);
if(jbbcParentJoin->isAntiSemiJoin())
joinExpr = new STMTHEAP Join(leftChildExpr, rightChildExpr, REL_ANTI_SEMIJOIN, NULL);
if(jbbcParentJoin->isLeftJoin())
{
joinExpr = new STMTHEAP Join(leftChildExpr, rightChildExpr, REL_LEFT_JOIN, NULL);
leftOuterJoinFilterPreds += jbbc->getLeftJoinFilterPreds();
}
if(joinExpr)
{
joinExpr->setJoinPred(jbbc->getPredsWithPredecessors());
joinExpr->nullInstantiatedOutput().insert(jbbc->nullInstantiatedOutput());
}
}
if(!joinExpr)
{
// now form a JoinExpr with these left and right children.
joinExpr = new STMTHEAP Join(leftChildExpr, rightChildExpr, REL_JOIN, NULL);
}
ValueIdSet selPredsAndLOJFilter = leftOuterJoinFilterPreds;
selPredsAndLOJFilter += (*joinPreds);
joinExpr->setSelectionPredicates(selPredsAndLOJFilter);
// set groupAttr of this Join expression
GroupAttributes * gaJoin = new STMTHEAP GroupAttributes();
// set required outputs of Join as sum of characteristic
// outputs of the left and the right children
ValueIdSet requiredOutputs;
requiredOutputs.addSet(getPotentialOutputs(jbbSubset));
requiredOutputs.addSet(getPotentialOutputs(jbbcId));
gaJoin->setCharacteristicOutputs(requiredOutputs);
// set JBBSubset for this group, if all estLogProps are cacheable.
// Else JBBSubset is NULL
CANodeIdSet combinedSet = jbbSubset;
combinedSet += jbbcId;
if (cacheable)
gaJoin->getGroupAnalysis()->setLocalJBBView(combinedSet.jbbcsToJBBSubset());
gaJoin->setMinChildEstRowCount(MINOF(garight->getMinChildEstRowCount(), galeft->getMinChildEstRowCount() ) );
// if there are some probes coming into the join
// then join type = tsj.
if ((inputEstLogProp->getResultCardinality() > 1) ||
(inputEstLogProp->getColStats().entries() > 1))
{
if (cacheable)
{
CANodeIdSet inputNodeSet = *(inputEstLogProp->getNodeSet());
gaJoin->setCharacteristicInputs(getPotentialOutputs(inputNodeSet));
}
}
joinExpr->setGroupAttr(gaJoin);
gaJoin->setLogExprForSynthesis(joinExpr);
joinExpr->synthLogProp();
inputEstLogProp->setCacheableFlag(origInputIsCacheable);
return joinExpr;
} // AppliedStatMan::formJoinExprForJoinOnJBBC
