ConstantFilter::ConstantFilter(const SOP& op, ReturnedColumn* lhs, ReturnedColumn* rhs)
{
SSFP ssfp(new SimpleFilter(op, lhs, rhs));
fFilterList.push_back(ssfp);
SimpleColumn* sc = dynamic_cast<SimpleColumn*>(lhs);
fCol.reset(sc->clone());
}
void SimpleFilter::replaceRealCol(CalpontSelectExecutionPlan::ReturnedColumnList& derivedColList)
{
SimpleColumn *sc = NULL;
if (fLhs)
{
sc = dynamic_cast<SimpleColumn*>(fLhs);
if (sc)
{
ReturnedColumn* tmp = derivedColList[sc->colPosition()]->clone();
delete fLhs;
fLhs = tmp;
}
else
{
fLhs->replaceRealCol(derivedColList);
}
}
if (fRhs)
{
sc = dynamic_cast<SimpleColumn*>(fRhs);
if (sc)
{
ReturnedColumn* tmp = derivedColList[sc->colPosition()]->clone();
delete fRhs;
fRhs = tmp;
}
else
{
fRhs->replaceRealCol(derivedColList);
}
}
}
void FunctionColumn::setDerivedTable()
{
if (hasAggregate())
{
fDerivedTable = "";
return;
}
setSimpleColumnList();
string derivedTableAlias = "";
for (uint32_t i = 0; i < fSimpleColumnList.size(); i++)
{
SimpleColumn* sc = fSimpleColumnList[i];
sc->setDerivedTable();
if (sc->derivedTable() != derivedTableAlias)
{
if (derivedTableAlias == "")
{
derivedTableAlias = sc->tableName();
}
else
{
derivedTableAlias = "";
break;
}
}
}
fDerivedTable = derivedTableAlias;
}
SimpleColumn::SimpleColumn (const SimpleColumn& rhs,const uint32_t sessionID):
ReturnedColumn(rhs, sessionID),
fSchemaName (rhs.schemaName()),
fTableName (rhs.tableName()),
fColumnName (rhs.columnName()),
fOid (rhs.oid()),
fTableAlias (rhs.tableAlias()),
fData (rhs.data()),
fIndexName (rhs.indexName()),
fViewName (rhs.viewName()),
fIsInfiniDB (rhs.isInfiniDB())
{
}
void SimpleColumn::setOID()
{
boost::shared_ptr<CalpontSystemCatalog> csc = CalpontSystemCatalog::makeCalpontSystemCatalog(fSessionID);
CalpontSystemCatalog::TableColName tcn;
tcn = make_tcn(fSchemaName, fTableName, fColumnName);
fOid = csc->lookupOID (tcn);
if (fOid == -1)
{
// get colMap from CalpontSelectExecutionPlan
// and try to map the schema and table name
CalpontSelectExecutionPlan::ColumnMap::iterator iter;
CalpontSelectExecutionPlan::ColumnMap colMap = CalpontSelectExecutionPlan::colMap();
// if this is the only column name exist in the map, return it ??
for (iter = colMap.find(fColumnName); iter != colMap.end(); ++iter)
{
SRCP srcp = iter->second;
SimpleColumn* scp = dynamic_cast<SimpleColumn*>(srcp.get());
if (colMap.count(fColumnName) == 1 ||
scp->tableName().compare(fTableName) == 0)
{
fOid = scp->oid();
//@bug 221 fix
fTableName = scp->tableName();
fSchemaName = scp->schemaName();
//@info assign tableAlias also. watch for other conflict
fTableAlias = scp->tableAlias();
fResultType = scp->resultType();
return;
}
}
}
fResultType = csc->colType(fOid);
}
void* ing_opentbl(void* qctx, const char* sn, const char* tn)
{
try {
if (!qctx) return 0;
if (!sn) return 0;
if (!tn) return 0;
if (strlen(sn) == 0) return 0;
if (strlen(tn) == 0) return 0;
IngQCtx* ingQctx = reinterpret_cast<IngQCtx*>(qctx);
cpsm_conhdl_t* cal_conn_hndl = reinterpret_cast<cpsm_conhdl_t*>(ingQctx->dhcs_ses_ctx);
IngTCtx* tctx = new IngTCtx();
dhcs_tableid_t tableid = 0;
//tableid = tbname2id(sn, tn, cal_conn_hndl);
tableid = tbname2id("te", "f_trans", cal_conn_hndl);
RequiredColOidSet requiredColOidSet;
const CalpontSelectExecutionPlan::ColumnMap colMap = ingQctx->csep.columnMap();
CalpontSelectExecutionPlan::ColumnMap::const_iterator colIter = colMap.begin();
CalpontSelectExecutionPlan::ColumnMap::const_iterator colend = colMap.end();
while (colIter != colend)
{
SRCP srcp = colIter->second;
SimpleColumn* scp = dynamic_cast<SimpleColumn*>(srcp.get());
if (scp)
requiredColOidSet.insert(scp->oid());
++colIter;
}
tctx->dhcs_tpl_ctx = new cpsm_tplh_t();
dhcs_tpl_open(tableid, &tctx->dhcs_tpl_ctx, ingQctx->dhcs_ses_ctx, requiredColOidSet);
tctx->dhcs_tpl_scan_ctx = new cpsm_tplsch_t();
dhcs_tpl_scan_open(tableid, DHCS_TPL_FH_READ, &tctx->dhcs_tpl_scan_ctx, ingQctx->dhcs_ses_ctx);
return tctx;
} catch (...) {
}
return NULL;
}
ParseTree* replaceRefCol(ParseTree*& n, CalpontSelectExecutionPlan::ReturnedColumnList& derivedColList)
{
ParseTree *lhs = n->left();
ParseTree *rhs = n->right();
if (lhs)
n->left(replaceRefCol(lhs, derivedColList));
if (rhs)
n->right(replaceRefCol(rhs, derivedColList));
SimpleFilter *sf = dynamic_cast<SimpleFilter*>(n->data());
ConstantFilter *cf = dynamic_cast<ConstantFilter*>(n->data());
ReturnedColumn *rc = dynamic_cast<ReturnedColumn*>(n->data());
if (sf)
{
sf->replaceRealCol(derivedColList);
}
else if (cf)
{
cf->replaceRealCol(derivedColList);
}
else if (rc)
{
SimpleColumn* sc = dynamic_cast<SimpleColumn*>(rc);
if (sc)
{
ReturnedColumn* tmp = derivedColList[sc->colPosition()]->clone();
delete sc;
n->data(tmp);
}
else
{
rc->replaceRealCol(derivedColList);
}
}
return n;
}
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;
}
void verifyParseTree(const ParseTree* t)
{
const ParseTree *ct, *ct2;
SimpleColumn *s;
CPPUNIT_ASSERT(t != NULL);
s = dynamic_cast<SimpleColumn*>(t->data());
CPPUNIT_ASSERT(s != NULL);
CPPUNIT_ASSERT(s->schemaName() == "Schema Name 0");
ct = t->left();
CPPUNIT_ASSERT(ct != NULL);
s = dynamic_cast<SimpleColumn*>(ct->data());
CPPUNIT_ASSERT(s != NULL);
CPPUNIT_ASSERT(s->schemaName() == "Schema Name 1");
ct2 = ct->left();
CPPUNIT_ASSERT(ct2 != NULL);
s = dynamic_cast<SimpleColumn*>(ct2->data());
CPPUNIT_ASSERT(s != NULL);
CPPUNIT_ASSERT(s->schemaName() == "Schema Name 3");
CPPUNIT_ASSERT(ct->right() == NULL);
CPPUNIT_ASSERT(ct2->left() == NULL);
CPPUNIT_ASSERT(ct2->right() == NULL);
ct = t->right();
CPPUNIT_ASSERT(ct != NULL);
s = dynamic_cast<SimpleColumn*>(ct->data());
CPPUNIT_ASSERT(s != NULL);
CPPUNIT_ASSERT(s->schemaName() == "Schema Name 2");
ct2 = ct->right();
CPPUNIT_ASSERT(ct2 != NULL);
s = dynamic_cast<SimpleColumn*>(ct2->data());
CPPUNIT_ASSERT(s != NULL);
CPPUNIT_ASSERT(s->schemaName() == "Schema Name 4");
CPPUNIT_ASSERT(ct->left() == NULL);
CPPUNIT_ASSERT(ct2->left() == NULL);
CPPUNIT_ASSERT(ct2->right() == NULL);
}
execplan::ParseTree* ScalarSub::buildParseTree(PredicateOperator* op)
{
idbassert(fColumn.get() && fSub && fFunc);
vector<SRCP> cols;
Filter* filter;
RowColumn* rcol = dynamic_cast<RowColumn*>(fColumn.get());
if (rcol)
{
// IDB only supports (c1,c2..) =/!= (subquery)
if (fFunc->functype() != Item_func::EQ_FUNC && fFunc->functype() != Item_func::NE_FUNC)
{
fGwip.fatalParseError = true;
fGwip.parseErrorText = IDBErrorInfo::instance()->errorMsg(ERR_INVALID_OPERATOR_WITH_LIST);
return NULL;
}
cols = rcol->columnVec();
}
else
cols.push_back(fColumn);
SCSEP csep(new CalpontSelectExecutionPlan());
csep->sessionID(fGwip.sessionid);
csep->location(CalpontSelectExecutionPlan::WHERE);
csep->subType (CalpontSelectExecutionPlan::SINGLEROW_SUBS);
// gwi for the sub query
gp_walk_info gwi;
gwi.thd = fGwip.thd;
gwi.subQuery = this;
// @4827 merge table list to gwi in case there is FROM sub to be referenced
// in the FROM sub
gwi.derivedTbCnt = fGwip.derivedTbList.size();
uint32_t tbCnt = fGwip.tbList.size();
gwi.tbList.insert(gwi.tbList.begin(), fGwip.tbList.begin(), fGwip.tbList.end());
gwi.derivedTbList.insert(gwi.derivedTbList.begin(), fGwip.derivedTbList.begin(), fGwip.derivedTbList.end());
if (getSelectPlan(gwi, *(fSub->get_select_lex()), csep) != 0)
{
//@todo more in error handling
if (!gwi.fatalParseError)
{
fGwip.fatalParseError = true;
fGwip.parseErrorText = "Error occured in ScalarSub::transform()";
}
else
{
fGwip.fatalParseError = gwi.fatalParseError;
fGwip.parseErrorText = gwi.parseErrorText;
}
return NULL;
}
fGwip.subselectList.push_back(csep);
// error out non-support case for now: comparison out of semi join tables.
// only check for simplecolumn
if (!gwi.correlatedTbNameVec.empty())
{
for (uint32_t i = 0; i < cols.size(); i++)
{
SimpleColumn* sc = dynamic_cast<SimpleColumn*>(cols[i].get());
if (sc)
{
CalpontSystemCatalog::TableAliasName tan = make_aliastable(sc->schemaName(), sc->tableName(), sc->tableAlias());
uint32_t j = 0;
for (; j < gwi.correlatedTbNameVec.size(); j++)
if (tan == gwi.correlatedTbNameVec[j])
break;
if (j == gwi.correlatedTbNameVec.size())
{
fGwip.fatalParseError = true;
fGwip.parseErrorText = IDBErrorInfo::instance()->errorMsg(ERR_NON_SUPPORT_SCALAR);
return NULL;
}
}
}
}
// remove outer query tables
CalpontSelectExecutionPlan::TableList tblist;
if (csep->tableList().size() >= tbCnt)
tblist.insert(tblist.begin(), csep->tableList().begin() + tbCnt, csep->tableList().end());
CalpontSelectExecutionPlan::SelectList derivedTbList;
if (csep->derivedTableList().size() >= gwi.derivedTbCnt)
derivedTbList.insert(derivedTbList.begin(), csep->derivedTableList().begin() + gwi.derivedTbCnt, csep->derivedTableList().end());
csep->tableList(tblist);
csep->derivedTableList(derivedTbList);
// if (fSub->is_correlated)
if (fSub->unit->first_select()->master_unit()->uncacheable)
{
SelectFilter* subFilter = new SelectFilter();
subFilter->correlated(true);
subFilter->cols(cols);
subFilter->sub(csep);
subFilter->op(SOP(op));
subFilter->returnedColPos(fReturnedColPos);
filter = subFilter;
}
else
{
SimpleScalarFilter* subFilter = new SimpleScalarFilter();
subFilter->cols(cols);
subFilter->sub(csep);
subFilter->op(SOP(op));
filter = subFilter;
}
return new ParseTree(filter);
}
void VirtualTable::addColumn(const SRCP& column)
{
// As of bug3695, make sure varbinary is not used in subquery.
if (column->resultType().colDataType == CalpontSystemCatalog::VARBINARY && !fVarBinOK)
throw runtime_error ("VARBINARY in subquery is not supported.");
AggregateColumn* agc = NULL;
ArithmeticColumn* arc = NULL;
ConstantColumn* cc = NULL;
FunctionColumn* fc = NULL;
SimpleColumn* sc = NULL;
WindowFunctionColumn* wc = NULL;
string columnName;
ostringstream oss;
UniqId colId;
if ((sc = dynamic_cast<SimpleColumn*>(column.get())) != NULL)
{
columnName = sc->columnName();
colId = UniqId(sc);
}
else if ((agc = dynamic_cast<AggregateColumn*>(column.get())) != NULL)
{
// oss << agc->functionName() << "_" << agc->expressionId();
// oss << "Aggregate_" << agc->expressionId();
columnName = agc->data();
colId = UniqId(agc->expressionId(), "", "", "");
}
else if ((wc = dynamic_cast<WindowFunctionColumn*>(column.get())) != NULL)
{
// oss << wc->functionName() << "_" << wc->expressionId();
// oss << "Window_" << wc->expressionId();
columnName = wc->data();
colId = UniqId(wc->expressionId(), "", "", "");
}
else if ((arc = dynamic_cast<ArithmeticColumn*>(column.get())) != NULL)
{
// oss << "Arithmetic_" << arc->expressionId();
columnName = arc->data();
colId = UniqId(arc->expressionId(), "", "", "");
}
else if ((fc = dynamic_cast<FunctionColumn*>(column.get())) != NULL)
{
// oss << fc->functionName() << "_" << fc->expressionId();
columnName = fc->data();
colId = UniqId(fc->expressionId(), "", "", "");
}
else if ((cc = dynamic_cast<ConstantColumn*>(column.get())) != NULL)
{
// oss << "Constant_" << cc->expressionId();
columnName = cc->data();
colId = UniqId(cc->expressionId(), cc->alias(), "", fView);
}
else // new column type has added, but this code is not updated.
{
oss << "not supported column type: " << typeid(*(column.get())).name();
throw runtime_error(oss.str());
}
if (columnName.empty())
columnName = column->alias();
SimpleColumn* vc = new SimpleColumn();
vc->tableName(fName);
vc->tableAlias(fAlias);
vc->columnName(columnName);
vc->alias(column->alias());
vc->viewName(fView);
uint32_t index = fColumns.size();
vc->colPosition(index);
vc->oid(fTableOid+index+1);
vc->resultType(column->resultType());
SSC ssc(vc);
fColumns.push_back(ssc);
fColumnTypes.push_back(column->resultType());
fColumnMap.insert(make_pair(colId, index));
}
execplan::ReturnedColumn* buildPseudoColumn(Item* item,
gp_walk_info& gwi,
bool& nonSupport,
uint32_t pseudoType)
{
Item_func* ifp = (Item_func*)item;
// idblocalpm is replaced by constant
if (pseudoType == PSEUDO_LOCALPM)
{
int64_t localPm = idblocalpm();
ConstantColumn* cc;
if (localPm)
cc = new ConstantColumn(localPm);
else
cc = new ConstantColumn("", ConstantColumn::NULLDATA);
cc->alias(ifp->name? ifp->name : "");
return cc;
}
// convert udf item to pseudocolumn item.
// adjust result type
// put arg col to column map
string funcName = ifp->func_name();
if (ifp->arg_count != 1 ||
!(ifp->arguments()) ||
!(ifp->arguments()[0]) ||
ifp->arguments()[0]->type() != Item::FIELD_ITEM)
return nullOnError(gwi, funcName);
Item_field* field = (Item_field*)(ifp->arguments()[0]);
// @todo rule out derive table
if (!field->field || !field->db_name || strlen(field->db_name) == 0)
return nullOnError(gwi, funcName);
SimpleColumn *sc = buildSimpleColumn(field, gwi);
if (!sc)
return nullOnError(gwi, funcName);
if ((pseudoType == PSEUDO_EXTENTMIN || pseudoType == PSEUDO_EXTENTMAX) &&
(sc->colType().colDataType == CalpontSystemCatalog::VARBINARY ||
(sc->colType().colDataType == CalpontSystemCatalog::VARCHAR && sc->colType().colWidth > 7) ||
(sc->colType().colDataType == CalpontSystemCatalog::CHAR && sc->colType().colWidth > 8)))
return nullOnError(gwi, funcName);
// put arg col to column map
if (gwi.clauseType == SELECT || gwi.clauseType == GROUP_BY) // select clause
{
SRCP srcp(sc);
gwi.columnMap.insert(CalpontSelectExecutionPlan::ColumnMap::value_type(sc->columnName(), srcp));
gwi.tableMap[make_aliastable(sc->schemaName(), sc->tableName(), sc->tableAlias(), sc->isInfiniDB())] =
make_pair(1, field->cached_table);
}
else if (!gwi.rcWorkStack.empty())
{
gwi.rcWorkStack.pop();
}
if (pseudoType == PSEUDO_PARTITION)
{
// parms: psueducolumn dbroot, segmentdir, segment
SPTP sptp;
FunctionColumn *fc = new FunctionColumn(funcName);
funcexp::FunctionParm parms;
PseudoColumn *dbroot = new PseudoColumn(*sc, PSEUDO_DBROOT);
sptp.reset(new ParseTree(dbroot));
parms.push_back(sptp);
PseudoColumn *pp = new PseudoColumn(*sc, PSEUDO_SEGMENTDIR);
sptp.reset(new ParseTree(pp));
parms.push_back(sptp);
PseudoColumn* seg = new PseudoColumn(*sc, PSEUDO_SEGMENT);
sptp.reset(new ParseTree(seg));
parms.push_back(sptp);
fc->functionParms(parms);
fc->expressionId(gwi.expressionId++);
// string result type
CalpontSystemCatalog::ColType ct;
ct.colDataType = CalpontSystemCatalog::VARCHAR;
ct.colWidth = 256;
fc->resultType(ct);
// operation type integer
funcexp::Func_idbpartition* idbpartition = new funcexp::Func_idbpartition();
fc->operationType(idbpartition->operationType(parms, fc->resultType()));
fc->alias(ifp->name? ifp->name : "");
return fc;
}
PseudoColumn *pc = new PseudoColumn(*sc, pseudoType);
// @bug5892. set alias for derived table column matching.
pc->alias(ifp->name? ifp->name : "");
return pc;
}
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;
}
