void makeAntiJoin(const ParseTree* n)
{
TreeNode* tn = n->data();
SimpleFilter* sf = dynamic_cast<SimpleFilter*>(tn);
if (!sf)
return;
uint64_t lJoinInfo = sf->lhs()->joinInfo();
if (lJoinInfo & JOIN_SEMI)
{
lJoinInfo &= ~JOIN_SEMI;
lJoinInfo |= JOIN_ANTI;
if (lJoinInfo & JOIN_NULLMATCH_CANDIDATE)
lJoinInfo |= JOIN_NULL_MATCH;
sf->lhs()->joinInfo(lJoinInfo);
}
uint64_t rJoinInfo = sf->rhs()->joinInfo();
if (rJoinInfo & JOIN_SEMI)
{
rJoinInfo &= ~JOIN_SEMI;
rJoinInfo |= JOIN_ANTI;
if (rJoinInfo & JOIN_NULLMATCH_CANDIDATE)
rJoinInfo |= JOIN_NULL_MATCH;
sf->rhs()->joinInfo(rJoinInfo);
}
}
SimpleFilter* createSimpleFilter
(
CalpontSystemCatalog*& csc,
const CalpontSystemCatalog::TableColName& tcn,
const string& opstr,
ConstantColumn* cc
)
{
SimpleFilter* lsf = new SimpleFilter();
Operator* op = new Operator();
op->data(opstr);
CalpontSystemCatalog::ColType ccct;
ccct = op->resultType();
ccct.colDataType = cc->resultType().colDataType;
op->operationType(ccct);
SOP sop(op);
lsf->op(sop);
CalpontSystemCatalog::OID oid = csc->lookupOID(tcn);
CalpontSystemCatalog::ColType ct = csc->colType(oid);
SimpleColumn* sc = new SimpleColumn();
sc->schemaName(tcn.schema);
sc->tableName(tcn.table);
sc->tableAlias(tcn.table);
sc->columnName(tcn.column);
sc->oid(oid);
sc->resultType(ct);
sc->alias(tcn.toString());
lsf->lhs(sc);
lsf->rhs(cc);
return lsf;
}
/**
* Handle MySQL's plugin functions
* This is mostly for handling the null related functions that MySQL adds to the execution plan
*/
void InSub::handleFunc(gp_walk_info* gwip, Item_func* func)
{
if (func->functype() == Item_func::TRIG_COND_FUNC || func->functype() == Item_func::COND_OR_FUNC)
{
// purpose: remove the isnull() function from the parsetree in ptWorkStack.
// IDB handles the null semantics in the join operation
// trigcond(or_cond) is the only form we recognize for now
if (func->argument_count() > 2)
{
fGwip.fatalParseError = true;
fGwip.parseErrorText = "Unsupported item in IN subquery";
return;
}
Item_cond* cond;
if (func->functype() == Item_func::TRIG_COND_FUNC)
{
Item* item;
if (func->arguments()[0]->type() == Item::REF_ITEM)
item = (Item_ref*)(func->arguments()[0])->real_item();
else
item = func->arguments()[0];
cond = (Item_cond*)(item);
}
else
{
cond = (Item_cond*)(func);
}
if (cond->functype() == Item_func::COND_OR_FUNC)
{
// (cache=item) case. do nothing. ignore trigcond()?
if (cond->argument_list()->elements == 1)
return;
// (cache=item or isnull(item)) case. remove "or isnull()"
if (cond->argument_list()->elements == 2)
{
// don't know how to deal with this. don't think it's a fatal error either.
if (gwip->ptWorkStack.empty())
return;
ParseTree* pt = gwip->ptWorkStack.top();
if (!pt->left() || !pt->right())
return;
SimpleFilter* sf = dynamic_cast<SimpleFilter*>(pt->left()->data());
//assert (sf && sf->op()->op() == execplan::OP_ISNULL);
if (!sf || sf->op()->op() != execplan::OP_ISNULL)
return;
delete sf;
sf = dynamic_cast<SimpleFilter*>(pt->right()->data());
//idbassert(sf && sf->op()->op() == execplan::OP_EQ);
if (!sf || sf->op()->op() != execplan::OP_EQ)
return;
// set NULLMATCH for both operand. It's really a setting for the join.
// should only set NULLMATCH when the subtype is NOT_IN. for some IN subquery
// with aggregation column, MySQL inefficiently convert to:
// (cache=item or item is null) and item is not null, which is equivalent to
// cache = item. Do not set NULLMATCH for this case.
// Because we don't know IN or NOTIN yet, set candidate bit and switch to NULLMATCH
// later in handleNot function.
if (sf->lhs()->joinInfo() & JOIN_CORRELATED)
sf->lhs()->joinInfo(sf->lhs()->joinInfo() | JOIN_NULLMATCH_CANDIDATE);
if (sf->rhs()->joinInfo() & JOIN_CORRELATED)
sf->rhs()->joinInfo(sf->rhs()->joinInfo() | JOIN_NULLMATCH_CANDIDATE);
pt = pt->right();
gwip->ptWorkStack.pop();
gwip->ptWorkStack.push(pt);
}
}
else if (cond->functype() == Item_func::EQ_FUNC)
{
// not in (select const ...)
if (gwip->ptWorkStack.empty())
return;
ParseTree* pt = gwip->ptWorkStack.top();
SimpleFilter* sf = dynamic_cast<SimpleFilter*>(pt->data());
if (!sf || sf->op()->op() != execplan::OP_EQ)
return;
if (sf->lhs()->joinInfo() & JOIN_CORRELATED)
sf->lhs()->joinInfo(sf->lhs()->joinInfo() | JOIN_NULLMATCH_CANDIDATE);
if (sf->rhs()->joinInfo() & JOIN_CORRELATED)
sf->rhs()->joinInfo(sf->rhs()->joinInfo() | JOIN_NULLMATCH_CANDIDATE);
}
}
}
SimpleFilter::SimpleFilter(const SimpleFilter& rhs) :
fOp(rhs.op()),
fIndexFlag(rhs.indexFlag()),
fJoinFlag(rhs.joinFlag())
{
fLhs = rhs.lhs()->clone();
fRhs = rhs.rhs()->clone();
fSimpleColumnList.clear();
fAggColumnList.clear();
fWindowFunctionColumnList.clear();
SimpleColumn *lsc = dynamic_cast<SimpleColumn*>(fLhs);
FunctionColumn *lfc = dynamic_cast<FunctionColumn*>(fLhs);
ArithmeticColumn *lac = dynamic_cast<ArithmeticColumn*>(fLhs);
WindowFunctionColumn *laf = dynamic_cast<WindowFunctionColumn*>(fLhs);
AggregateColumn *lagc = dynamic_cast<AggregateColumn*>(fLhs);
SimpleColumn *rsc = dynamic_cast<SimpleColumn*>(fRhs);
FunctionColumn *rfc = dynamic_cast<FunctionColumn*>(fRhs);
ArithmeticColumn *rac = dynamic_cast<ArithmeticColumn*>(fRhs);
AggregateColumn *ragc = dynamic_cast<AggregateColumn*>(fRhs);
WindowFunctionColumn *raf = dynamic_cast<WindowFunctionColumn*>(fRhs);
if (lsc)
{
fSimpleColumnList.push_back(lsc);
}
else if (lagc)
{
fAggColumnList.push_back(lagc);
}
else if (lfc)
{
fSimpleColumnList.insert(fSimpleColumnList.end(), lfc->simpleColumnList().begin(), lfc->simpleColumnList().end());
fAggColumnList.insert(fAggColumnList.end(), lfc->aggColumnList().begin(), lfc->aggColumnList().end());
fWindowFunctionColumnList.insert
(fWindowFunctionColumnList.end(), lfc->windowfunctionColumnList().begin(), lfc->windowfunctionColumnList().end());
}
else if (lac)
{
fSimpleColumnList.insert(fSimpleColumnList.end(), lac->simpleColumnList().begin(), lac->simpleColumnList().end());
fAggColumnList.insert(fAggColumnList.end(), lac->aggColumnList().begin(), lac->aggColumnList().end());
fWindowFunctionColumnList.insert
(fWindowFunctionColumnList.end(), lac->windowfunctionColumnList().begin(), lac->windowfunctionColumnList().end());
}
else if (laf)
{
fWindowFunctionColumnList.push_back(laf);
}
if (rsc)
{
fSimpleColumnList.push_back(rsc);
}
else if (ragc)
{
fAggColumnList.push_back(ragc);
}
else if (rfc)
{
fSimpleColumnList.insert
(fSimpleColumnList.end(), rfc->simpleColumnList().begin(), rfc->simpleColumnList().end());
fAggColumnList.insert
(fAggColumnList.end(), rfc->aggColumnList().begin(), rfc->aggColumnList().end());
fWindowFunctionColumnList.insert
(fWindowFunctionColumnList.end(), rfc->windowfunctionColumnList().begin(), rfc->windowfunctionColumnList().end());
}
else if (rac)
{
fSimpleColumnList.insert(fSimpleColumnList.end(), rac->simpleColumnList().begin(), rac->simpleColumnList().end());
fAggColumnList.insert(fAggColumnList.end(), rac->aggColumnList().begin(), rac->aggColumnList().end());
fWindowFunctionColumnList.insert
(fWindowFunctionColumnList.end(), rac->windowfunctionColumnList().begin(), rac->windowfunctionColumnList().end());
}
else if (raf)
{
fWindowFunctionColumnList.push_back(raf);
}
}
void serializeCSEP()
{
/*
* erydbSelectExecutionPlan
* This is a large class; it makes more sense to write == operators
* for everything than to write a giant equivalance test here.
* For now this is mostly a regression test.
*/
erydbSelectExecutionPlan csep1, csep2;
erydbSelectExecutionPlan::ReturnedColumnList colList;
ParseTree* filterList;
erydbExecutionPlan *cep;
ByteStream b;
cep = &csep2;
CPPUNIT_ASSERT(csep1 == csep2);
CPPUNIT_ASSERT(!(csep1 != csep2));
CPPUNIT_ASSERT(csep1 == cep);
CPPUNIT_ASSERT(!(csep1 != cep));
// returned columns
SimpleColumn *sc = new SimpleColumn("tpch.region.r_regionkey");
colList.push_back(sc);
// filters
erydbSelectExecutionPlan::Parser parser;
std::vector<Token> tokens;
Token t;
SimpleFilter *sf = new SimpleFilter();
SimpleColumn *lhs = new SimpleColumn(*sc);
SimpleColumn *rhs = new SimpleColumn("tpch.nation.n_regionkey");
Operator *op = new Operator("=");
sf->op(op);
sf->lhs(lhs);
sf->rhs(rhs);
t.value = sf;
tokens.push_back(t);
Operator *op1 = new Operator ("and");
t.value = op1;
tokens.push_back(t);
SimpleFilter *sf1 = new SimpleFilter();
SimpleColumn *lhs1 = new SimpleColumn (*rhs);
ConstantColumn *constCol = new ConstantColumn("3", ConstantColumn::NUM);
Operator *op2 = new Operator("!=");
sf1->op(op2);
sf1->lhs(lhs1);
sf1->rhs(constCol);
t.value = sf1;
tokens.push_back(t);
filterList = parser.parse(tokens.begin(), tokens.end());
// draw filterList tree
filterList->drawTree("selectExecutionPlan_1.dot");
// erydb execution plan
csep1.returnedCols (colList);
csep1.filters (filterList);
CPPUNIT_ASSERT(csep1 != csep2);
CPPUNIT_ASSERT(!(csep1 == csep2));
CPPUNIT_ASSERT(csep1 != cep);
CPPUNIT_ASSERT(!(csep1 == cep));
csep1.serialize(b);
csep2.unserialize(b);
CPPUNIT_ASSERT(b.length() == 0);
CPPUNIT_ASSERT(csep1 == csep2);
CPPUNIT_ASSERT(!(csep1 != csep2));
CPPUNIT_ASSERT(csep1 == cep);
CPPUNIT_ASSERT(!(csep1 != cep));
erydbSelectExecutionPlan csep3, csep4;
// subselect
erydbSelectExecutionPlan *subselect = new erydbSelectExecutionPlan;
subselect->location(erydbSelectExecutionPlan::WHERE);
subselect->dependent (false);
CPPUNIT_ASSERT (subselect->location() == erydbSelectExecutionPlan::WHERE);
CPPUNIT_ASSERT (subselect->dependent() == false);
erydbSelectExecutionPlan::SelectList selectList;
selectList.push_back(subselect);
csep3.subSelects(selectList);
// exist filter
erydbSelectExecutionPlan* cep1 = new erydbSelectExecutionPlan();
ExistsFilter *filter = new ExistsFilter();
delete filter;
filter = new ExistsFilter(cep1);
filter->exists(cep1);
//erydbSelectExecutionPlan* cep2 = const_cast<erydbSelectExecutionPlan*>(filter->exists());
erydbSelectExecutionPlan::Parser parser1;
std::vector<Token> tokens1;
Token t1;
t1.value = filter;
tokens1.push_back(t1);
csep3.filters(parser1.parse(tokens1.begin(), tokens1.end()));
csep3.serialize(b);
csep4.unserialize(b);
CPPUNIT_ASSERT(csep3 == csep4);
CPPUNIT_ASSERT(!(csep3 != csep4));
}
void selectExecutionPlan_1() {
cout << "SQL: select region.r_regionkey from region, nation where nation.n_regionkey = region.r_regionkey and nation.n_regionkey != 3;" << endl;
erydbSelectExecutionPlan csep;
CPPUNIT_ASSERT (csep.location() == erydbSelectExecutionPlan::MAIN);
CPPUNIT_ASSERT (csep.dependent() == false);
CPPUNIT_ASSERT (csep.subSelects().size() == 0);
// returned columns
erydbSelectExecutionPlan::ReturnedColumnList colList;
SimpleColumn *sc = new SimpleColumn("tpch.region.r_regionkey", 0);
colList.push_back(sc);
ArithmeticColumn *ac = new ArithmeticColumn("a+sum(r_regionkey)", 0);
colList.push_back(ac);
csep.returnedCols (colList);
CPPUNIT_ASSERT(csep.returnedCols().size() == 2);
// filters
erydbSelectExecutionPlan::FilterTokenList filterTokenList;
SimpleFilter *sf = new SimpleFilter();
SimpleColumn *lhs = new SimpleColumn();
*lhs = *sc;
SimpleColumn *rhs = new SimpleColumn("tpch.nation.n_regionkey", 0);
CPPUNIT_ASSERT (*lhs == *sc);
CPPUNIT_ASSERT (*rhs != *lhs);
Operator *op = new Operator("=");
sf->op(op);
sf->lhs(lhs);
sf->rhs(rhs);
filterTokenList.push_back (sf);
filterTokenList.push_back( new Operator ("And") );
SimpleFilter *sf1 = new SimpleFilter (new Operator("="), sc->clone(), ac->clone());
filterTokenList.push_back (sf1);
csep.filterTokenList (filterTokenList);
ParseTree *filterList = const_cast<ParseTree*> (csep.filters());
// draw filterList tree
filterList->drawTree("selectExecutionPlan_1.dot");
csep.filters (filterList);
// Group by
erydbSelectExecutionPlan::GroupByColumnList groupByList;
groupByList.push_back(sc->clone());
csep.groupByCols (groupByList);
CPPUNIT_ASSERT(csep.groupByCols().size() == 1);
// Having
erydbSelectExecutionPlan::FilterTokenList havingTokenList;
SimpleFilter *having = new SimpleFilter( new Operator("="),
new ArithmeticColumn("sum(volumn)", 0),
new ConstantColumn(8));
havingTokenList.push_back (having);
csep.havingTokenList (havingTokenList);
CPPUNIT_ASSERT (*sf1 != *having);
CPPUNIT_ASSERT (csep.havingTokenList().size() == 1);
// Order by
erydbSelectExecutionPlan::OrderByColumnList orderByList;
ArithmeticColumn *o1 = new ArithmeticColumn(*ac);
o1->asc(false);
orderByList.push_back(o1);
csep.orderByCols(orderByList);
CPPUNIT_ASSERT(csep.orderByCols().size() == 1);
// another csep
erydbSelectExecutionPlan *newcsep = new erydbSelectExecutionPlan(erydbSelectExecutionPlan::FROM);
erydbSelectExecutionPlan::ReturnedColumnList ncolList;
SimpleColumn *newsc = new SimpleColumn("tpch.region.r_regionkey", 0);
ncolList.push_back(newsc);
newcsep->returnedCols (ncolList);
erydbSelectExecutionPlan::FilterTokenList nfilterTokenList;
SimpleFilter *newsf = new SimpleFilter ( new Operator (">"),
sc->clone(),
newsc->clone());
nfilterTokenList.push_back(newsf);
newcsep->filterTokenList (nfilterTokenList);
erydbSelectExecutionPlan::FilterTokenList nhavingTokenList;
SimpleFilter *newhaving = new SimpleFilter ( new Operator (">"),
sc->clone(),
newsc->clone());
CPPUNIT_ASSERT (*newsf == *newhaving);
nhavingTokenList.push_back(newhaving);
newcsep->havingTokenList (nhavingTokenList);
CPPUNIT_ASSERT (*newcsep != csep);
CPPUNIT_ASSERT (*newcsep->filters() == *newcsep->having());
ByteStream b;
csep.serialize (b);
newcsep->unserialize (b);
CPPUNIT_ASSERT (csep == *newcsep);
erydbSelectExecutionPlan::SelectList selectList;
selectList.push_back(newcsep);
csep.subSelects(selectList);
cout << "\nerydb Execution Plan:" << endl;
cout << csep;
cout << " --- end of test 1 ---" << endl;
}
