// -----------------------------------------------------------------------
// Given a column list providing identifiers for columns of this table,
// this method returns a list of VEG expressions and/or base columns that
// show the equivalence of base columns with index columns.
// -----------------------------------------------------------------------
void TableDesc::getEquivVEGCols (const ValueIdList& columnList,
ValueIdList &VEGColumnList) const
{
for (CollIndex i=0; i < columnList.entries(); i++)
{
ItemExpr *ie = columnList[i].getItemExpr();
BaseColumn *bc = NULL;
switch (ie->getOperatorType())
{
case ITM_BASECOLUMN:
bc = (BaseColumn *) ie;
break;
case ITM_INDEXCOLUMN:
bc = (BaseColumn *) ((IndexColumn *) ie)->getDefinition().
getItemExpr();
CMPASSERT(bc->getOperatorType() == ITM_BASECOLUMN);
break;
default:
ABORT("Invalid argument to TableDesc::getEquivVEGCols()\n");
}
CMPASSERT(bc->getTableDesc() == this);
VEGColumnList.insert(getColumnVEGList()[bc->getColNumber()]);
}
}
ValueIdList AbstractTable::copyValueIds(const size_t row, const field_list_t *fields) const {
ValueIdList valueIdList;
if (fields) {
for (const field_t & field: *fields) {
valueIdList.push_back(getValueId(field, row));
}
} else {
for (size_t i = 0; i < columnCount(); ++i) {
valueIdList.push_back(getValueId(i, row));
}
}
return valueIdList;
}
// -----------------------------------------------------------------------
// TableDesc::getSystemColumnList()
// -----------------------------------------------------------------------
void TableDesc::getSystemColumnList(ValueIdList &columnList) const
{
for (CollIndex i = 0; i < colList_.entries(); i++) {
ValueId valId = colList_[i];
NAColumn *column = valId.getNAColumn();
if (column->isSystemColumn())
columnList.insert(valId);
}
}
NABoolean IndexDesc::isUniqueIndex() const
{
return getNAFileSet()->uniqueIndex();
#pragma nowarn(269) // warning elimination
ValueIdList nonKeyColumnList;
#pragma warn(269) // warning elimination
getNonKeyColumnList(nonKeyColumnList);
// if there are some non-index-key columns(the key of base table),
// then this is a unique index. The primary key of base table is
// not needed to define the key of the index. It is, of course,
// needed to be present in the index as a non-key column.
if (nonKeyColumnList.entries() > 0)
return TRUE;
else
return FALSE;
}
void
IndexDesc::getNonKeyColumnList(ValueIdList& nonKeyColumnList) const
{
const ValueIdList
&indexColumns = getIndexColumns(),
&keyColumns = getIndexKey();
// clean up input:
nonKeyColumnList.clear();
// Add all index columns
CollIndex i = 0;
for (i=0;
i < indexColumns.entries();
i++)
{
nonKeyColumnList.insert(indexColumns[i]);
}
// And remove all key columns:
for (i=0;
i < keyColumns.entries();
i++)
{
nonKeyColumnList.remove(keyColumns[i]);
// if this is a secondary index, the base column
// which is part of the index,
// may also be present, remove it:
const ItemExpr *colPtr = keyColumns[i].getItemExpr();
if (colPtr->getOperatorType()
==
ITM_INDEXCOLUMN)
{
const ValueId & colDef = ((IndexColumn *)(colPtr))->getDefinition();
nonKeyColumnList.remove(colDef);
}
}
} // IndexDesc::getNonKeyColumnList(ValueIdSet& nonKeyColumnSet) const
// -----------------------------------------------------------------------
// TableDesc::getIdentityColumn()
// -----------------------------------------------------------------------
void TableDesc::getIdentityColumn(ValueIdList &columnList) const
{
for (CollIndex i = 0; i < colList_.entries(); i++)
{
ValueId valId = colList_[i];
NAColumn *column = valId.getNAColumn();
if (column->isIdentityColumn())
{
columnList.insert(valId);
break; // Break when you find the first,
// as there can only be one Identity column per table.
}
}
}
static short ft_codegen(Generator *generator,
RelExpr &relExpr,
ComTdbFastExtract *&newTdb,
Cardinality estimatedRowCount,
char * targetName,
char * hdfsHost,
Int32 hdfsPort,
char * hiveTableName,
char * delimiter,
char * header,
char * nullString,
char * recordSeparator,
ULng32 downQueueMaxSize,
ULng32 upQueueMaxSize,
ULng32 outputBufferSize,
ULng32 requestBufferSize,
ULng32 replyBufferSize,
ULng32 numOutputBuffers,
ComTdb * childTdb,
NABoolean isSequenceFile)
{
CmpContext *cmpContext = generator->currentCmpContext();
Space *space = generator->getSpace();
ExpGenerator *exp_gen = generator->getExpGenerator();
MapTable *map_table = generator->getMapTable();
MapTable *last_map_table = generator->getLastMapTable();
ex_expr *input_expr = NULL;
ex_expr *output_expr = NULL;
ex_expr * childData_expr = NULL ;
ex_expr * cnvChildData_expr = NULL ;
ULng32 i;
ULng32 requestRowLen = 0;
ULng32 outputRowLen = 0;
ULng32 childDataRowLen = 0;
ULng32 cnvChildDataRowLen = 0;
ExpTupleDesc *requestTupleDesc = NULL;
ExpTupleDesc *outputTupleDesc = NULL;
ExpTupleDesc *childDataTupleDesc = NULL;
ExpTupleDesc *cnvChildDataTupleDesc = NULL;
newTdb = NULL;
OperatorTypeEnum relExprType = relExpr.getOperatorType();
GenAssert(relExprType == REL_FAST_EXTRACT, "Unexpected RelExpr at FastExtract codegen")
FastExtract * fastExtract = (FastExtract *) &relExpr;
const Int32 workAtpNumber = 1;
ex_cri_desc *given_desc = generator->getCriDesc(Generator::DOWN);
ex_cri_desc *returned_desc = NULL;
ex_cri_desc *work_cri_desc = NULL;
returned_desc = given_desc;
// Setup local variables related to the work ATP
unsigned short numWorkTupps = 0;
unsigned short childDataTuppIndex = 0;
unsigned short cnvChildDataTuppIndex = 0;
numWorkTupps = 3;
childDataTuppIndex = numWorkTupps - 1 ;
numWorkTupps ++;
cnvChildDataTuppIndex = numWorkTupps - 1;
work_cri_desc = new (space) ex_cri_desc(numWorkTupps, space);
ExpTupleDesc::TupleDataFormat childReqFormat = ExpTupleDesc::SQLMX_ALIGNED_FORMAT;
ValueIdList childDataVids;
ValueIdList cnvChildDataVids;
const ValueIdList& childVals = fastExtract->getSelectList();
for (i = 0; i < childVals.entries(); i++)
{
ItemExpr &inputExpr = *(childVals[i].getItemExpr());
const NAType &formalType = childVals[i].getType();
ItemExpr *lmExpr = NULL;
ItemExpr *lmExpr2 = NULL;
int res;
lmExpr = &inputExpr; //CreateCastExpr(inputExpr, *inputExpr.getValueId().getType().newCopy(), cmpContext);
res = CreateAllCharsExpr(formalType, // [IN] Child output type
*lmExpr, // [IN] Actual input value
cmpContext, // [IN] Compilation context
lmExpr2 // [OUT] Returned expression
);
GenAssert(res == 0 && lmExpr != NULL,
"Error building expression tree for LM child Input value");
lmExpr->bindNode(generator->getBindWA());
childDataVids.insert(lmExpr->getValueId());
if (lmExpr2)
{
lmExpr2->bindNode(generator->getBindWA());
cnvChildDataVids.insert(lmExpr2->getValueId());
}
} // for (i = 0; i < childVals.entries(); i++)
if (childDataVids.entries() > 0 &&
cnvChildDataVids.entries()>0) //-- convertedChildDataVids
{
exp_gen->generateContiguousMoveExpr (
childDataVids, //childDataVids// [IN] source ValueIds
TRUE, // [IN] add convert nodes?
workAtpNumber, // [IN] target atp number (0 or 1)
childDataTuppIndex, // [IN] target tupp index
childReqFormat, // [IN] target tuple data format
childDataRowLen, // [OUT] target tuple length
&childData_expr, // [OUT] move expression
&childDataTupleDesc, // [optional OUT] target tuple desc
ExpTupleDesc::LONG_FORMAT // [optional IN] target desc format
);
exp_gen->processValIdList (
cnvChildDataVids, // [IN] ValueIdList
ExpTupleDesc::SQLARK_EXPLODED_FORMAT, // [IN] tuple data format
cnvChildDataRowLen, // [OUT] tuple length
workAtpNumber, // [IN] atp number
cnvChildDataTuppIndex, // [IN] index into atp
&cnvChildDataTupleDesc, // [optional OUT] tuple desc
ExpTupleDesc::LONG_FORMAT // [optional IN] tuple desc format
);
}
//
// Add the tuple descriptor for request values to the work ATP
//
work_cri_desc->setTupleDescriptor(childDataTuppIndex, childDataTupleDesc);
work_cri_desc->setTupleDescriptor(cnvChildDataTuppIndex, cnvChildDataTupleDesc);
// We can now remove all appended map tables
generator->removeAll(last_map_table);
ComSInt32 maxrs = 0;
UInt32 flags = 0;
UInt16 numIoBuffers = (UInt16)(ActiveSchemaDB()->getDefaults()).getAsLong(FAST_EXTRACT_IO_BUFFERS);
UInt16 ioTimeout = (UInt16)(ActiveSchemaDB()->getDefaults()).getAsLong(FAST_EXTRACT_IO_TIMEOUT_SEC);
Int64 hdfsBufSize = (Int64)CmpCommon::getDefaultNumeric(HDFS_IO_BUFFERSIZE);
hdfsBufSize = hdfsBufSize * 1024; // convert to bytes
Int16 replication = (Int16)CmpCommon::getDefaultNumeric(HDFS_REPLICATION);
// Create a TDB
ComTdbFastExtract *tdb = new (space) ComTdbFastExtract (
flags,
estimatedRowCount,
targetName,
hdfsHost,
hdfsPort,
hiveTableName,
delimiter,
header,
nullString,
recordSeparator,
given_desc,
returned_desc,
work_cri_desc,
downQueueMaxSize,
upQueueMaxSize,
(Lng32) numOutputBuffers,
outputBufferSize,
numIoBuffers,
ioTimeout,
input_expr,
output_expr,
requestRowLen,
outputRowLen,
childData_expr,
childTdb,
space,
childDataTuppIndex,
cnvChildDataTuppIndex,
childDataRowLen,
hdfsBufSize,
replication
);
tdb->setSequenceFile(isSequenceFile);
tdb->setHdfsCompressed(CmpCommon::getDefaultNumeric(TRAF_UNLOAD_HDFS_COMPRESS)!=0);
tdb->setSkipWritingToFiles(CmpCommon::getDefault(TRAF_UNLOAD_SKIP_WRITING_TO_FILES) == DF_ON);
tdb->setBypassLibhdfs(CmpCommon::getDefault(TRAF_UNLOAD_BYPASS_LIBHDFS) == DF_ON);
generator->initTdbFields(tdb);
// Generate EXPLAIN info.
if (!generator->explainDisabled())
{
generator->setExplainTuple(relExpr.addExplainInfo(tdb, 0, 0, generator));
}
// Tell the generator about our in/out rows and the new TDB
generator->setCriDesc(given_desc, Generator::DOWN);
generator->setCriDesc(returned_desc, Generator::UP);
generator->setGenObj(&relExpr, tdb);
// Return a TDB pointer to the caller
newTdb = tdb;
return 0;
} // ft_codegen()
short
PhysUnPackRows::codeGen(Generator *generator)
{
// Get handles on expression generator, map table, and heap allocator
//
ExpGenerator *expGen = generator->getExpGenerator();
Space *space = generator->getSpace();
// Allocate a new map table for this operation
//
MapTable *localMapTable = generator->appendAtEnd();
// Generate the child and capture the task definition block and a description
// of the reply composite row layout and the explain information.
//
child(0)->codeGen(generator);
ComTdb *childTdb = (ComTdb*)(generator->getGenObj());
ex_cri_desc *childCriDesc = generator->getCriDesc(Generator::UP);
ExplainTuple *childExplainTuple = generator->getExplainTuple();
// Make all of my child's outputs map to ATP 1. Since they are
// not needed above, they will not be in the work ATP (0).
// (Later, they will be removed from the map table)
//
localMapTable->setAllAtp(1);
// Generate the given and returned composite row descriptors.
// unPackRows adds a tupp (for the generated outputs) to the
// row given by the parent. The workAtp will have the 2 more
// tupps (1 for the generated outputs and another for the
// indexValue) than the given.
//
ex_cri_desc *givenCriDesc = generator->getCriDesc(Generator::DOWN);
ex_cri_desc *returnedCriDesc =
#pragma nowarn(1506) // warning elimination
new(space) ex_cri_desc(givenCriDesc->noTuples() + 1, space);
#pragma warn(1506) // warning elimination
ex_cri_desc *workCriDesc =
#pragma nowarn(1506) // warning elimination
new(space) ex_cri_desc(givenCriDesc->noTuples() + 2, space);
#pragma warn(1506) // warning elimination
// unPackCols is the next to the last Tp in Atp 0, the work ATP.
// and the last Tp in the returned ATP.
//
const Int32 unPackColsAtpIndex = workCriDesc->noTuples() - 2;
const Int32 unPackColsAtp = 0;
// The length of the new tuple which will contain the columns
// generated by unPackRows
//
ULng32 unPackColsTupleLen;
// The Tuple Desc describing the tuple containing the new unPacked columns
// It is generated when the expression is generated.
//
ExpTupleDesc *unPackColsTupleDesc = 0;
// indexValue is the last Tp in Atp 0, the work ATP.
//
const Int32 indexValueAtpIndex = workCriDesc->noTuples() - 1;
const Int32 indexValueAtp = 0;
// The length of the work tuple which will contain the value
// of the index. This should always be sizeof(int).
//
ULng32 indexValueTupleLen = 0;
// The Tuple Desc describing the tuple containing the new unPacked columns
// It is generated when the expression is generated.
//
ExpTupleDesc *indexValueTupleDesc = 0;
ValueIdList indexValueList;
if (indexValue() != NULL_VALUE_ID)
{
indexValueList.insert(indexValue());
expGen->processValIdList(indexValueList,
ExpTupleDesc::SQLARK_EXPLODED_FORMAT,
indexValueTupleLen,
indexValueAtp,
indexValueAtpIndex,
&indexValueTupleDesc,
ExpTupleDesc::SHORT_FORMAT);
GenAssert(indexValueTupleLen == sizeof(Int32),
"UnPackRows::codeGen: Internal Error");
}
// If a packingFactor exists, generate a move expression for this.
// It is assumed that the packingFactor expression evaluates to a
// 4 byte integer.
//
ex_expr *packingFactorExpr = 0;
ULng32 packingFactorTupleLen;
if(packingFactor().entries() > 0) {
expGen->generateContiguousMoveExpr(packingFactor(),
-1,
unPackColsAtp,
unPackColsAtpIndex,
ExpTupleDesc::SQLARK_EXPLODED_FORMAT,
packingFactorTupleLen,
&packingFactorExpr);
GenAssert(packingFactorTupleLen == sizeof(Int32),
"UnPackRows::codeGen: Internal Error");
}
// Generate the UnPack expressions.
//
// characteristicOutputs() - refers to the list of expressions
// to be move to another tuple.
//
// 0 - Do not add conv. nodes.
//
// unPackColsAtp - this expression will move data to the
// unPackColsAtp (0) ATP.
//
// unPackColsAtpIndex - within the unPackColsAtp (0) ATP, the destination
// for this move expression will be the unPackColsAtpIndex TP. This should
// be the next to the last TP of the work ATP. (The indexValue will be in
// the last position)
//
// SQLARK_EXPLODED_FORMAT - generate the move expression to construct
// the destination tuple in EXPLODED FORMAT.
//
// unPackColsTupleLen - This is an output which will contain the length
// of the destination Tuple.
//
// &unPackColsExpr - The address of the pointer to the expression
// which will be generated.
//
// &unPackColsTupleDesc - The address of the tuple descriptor which is
// generated. This describes the destination tuple of the move expression.
//
// SHORT_FORMAT - generate the unPackColsTupleDesc in the SHORT FORMAT.
//
ex_expr *unPackColsExpr = 0;
expGen->
genGuardedContigMoveExpr(selectionPred(),
getGroupAttr()->getCharacteristicOutputs(),
0, // No Convert Nodes added
unPackColsAtp,
unPackColsAtpIndex,
ExpTupleDesc::SQLARK_EXPLODED_FORMAT,
unPackColsTupleLen,
&unPackColsExpr,
&unPackColsTupleDesc,
ExpTupleDesc::SHORT_FORMAT);
#pragma nowarn(1506) // warning elimination
workCriDesc->setTupleDescriptor(unPackColsAtpIndex,
#pragma warn(1506) // warning elimination
unPackColsTupleDesc);
#pragma nowarn(1506) // warning elimination
returnedCriDesc->setTupleDescriptor(unPackColsAtpIndex,
#pragma warn(1506) // warning elimination
unPackColsTupleDesc);
// expressions for rowwise rowset implementation.
ex_expr * rwrsInputSizeExpr = 0;
ex_expr * rwrsMaxInputRowlenExpr = 0;
ex_expr * rwrsBufferAddrExpr = 0;
ULng32 rwrsInputSizeExprLen = 0;
ULng32 rwrsMaxInputRowlenExprLen = 0;
ULng32 rwrsBufferAddrExprLen = 0;
const Int32 rwrsAtp = 1;
const Int32 rwrsAtpIndex = workCriDesc->noTuples() - 2;
ExpTupleDesc * rwrsTupleDesc = 0;
ValueIdList rwrsVidList;
if (rowwiseRowset())
{
rwrsVidList.insert(this->rwrsInputSizeExpr()->getValueId());
expGen->generateContiguousMoveExpr(rwrsVidList,
0 /*don't add conv nodes*/,
rwrsAtp, rwrsAtpIndex,
ExpTupleDesc::SQLARK_EXPLODED_FORMAT,
rwrsInputSizeExprLen,
&rwrsInputSizeExpr,
&rwrsTupleDesc,ExpTupleDesc::SHORT_FORMAT);
rwrsVidList.clear();
rwrsVidList.insert(this->rwrsMaxInputRowlenExpr()->getValueId());
expGen->generateContiguousMoveExpr(rwrsVidList,
0 /*don't add conv nodes*/,
rwrsAtp, rwrsAtpIndex,
ExpTupleDesc::SQLARK_EXPLODED_FORMAT,
rwrsMaxInputRowlenExprLen,
&rwrsMaxInputRowlenExpr,
&rwrsTupleDesc,ExpTupleDesc::SHORT_FORMAT);
rwrsVidList.clear();
rwrsVidList.insert(this->rwrsBufferAddrExpr()->getValueId());
expGen->generateContiguousMoveExpr(rwrsVidList,
0 /*don't add conv nodes*/,
rwrsAtp, rwrsAtpIndex,
ExpTupleDesc::SQLARK_EXPLODED_FORMAT,
rwrsBufferAddrExprLen,
&rwrsBufferAddrExpr,
&rwrsTupleDesc,ExpTupleDesc::SHORT_FORMAT);
expGen->assignAtpAndAtpIndex(rwrsOutputVids(),
unPackColsAtp, unPackColsAtpIndex);
}
// Move the generated maptable entries, to the localMapTable,
// so that all other entries can later be removed.
//
for(ValueId outputValId = getGroupAttr()->getCharacteristicOutputs().init();
getGroupAttr()->getCharacteristicOutputs().next(outputValId);
getGroupAttr()->getCharacteristicOutputs().advance(outputValId)) {
generator->addMapInfoToThis(localMapTable, outputValId,
generator->getMapInfo(outputValId)->
getAttr());
// Indicate that code was generated for this map table entry.
//
generator->getMapInfoFromThis(localMapTable, outputValId)->codeGenerated();
}
NABoolean tolerateNonFatalError = FALSE;
if (isRowsetIterator() && (generator->getTolerateNonFatalError())) {
tolerateNonFatalError = TRUE;
setTolerateNonFatalError(RelExpr::NOT_ATOMIC_);
}
// Allocate the UnPack TDB
//
ComTdbUnPackRows *unPackTdb = NULL;
if (rowwiseRowset())
{
unPackTdb =
new (space) ComTdbUnPackRows(NULL, //childTdb,
rwrsInputSizeExpr,
rwrsMaxInputRowlenExpr,
rwrsBufferAddrExpr,
rwrsAtpIndex,
givenCriDesc,
returnedCriDesc,
workCriDesc,
16,
1024,
(Cardinality) getGroupAttr()->
getOutputLogPropList()[0]->
getResultCardinality().value(),
2, 20000);
}
else
{
// Base the initial queue size on the est. cardinality.
// UnPackRows does not do dyn queue resize, so passed in
// queue size values represent initial (and final) queue
// sizes (not max queue sizes).
//
queue_index upQueueSize =
(queue_index)getGroupAttr()->getOutputLogPropList()[0]->getResultCardinality().value();
// Make sure it is at least 1024.
upQueueSize = (upQueueSize < 1024 ? 1024 : upQueueSize);
// Make sure that it is not more the 64K.
upQueueSize = (upQueueSize > 65536 ? 65536 : upQueueSize);
unPackTdb =
new (space) ComTdbUnPackRows(childTdb,
packingFactorExpr,
unPackColsExpr,
#pragma nowarn(1506) // warning elimination
unPackColsTupleLen,
unPackColsAtpIndex,
indexValueAtpIndex,
givenCriDesc,
returnedCriDesc,
workCriDesc,
16,
upQueueSize,
(Cardinality) getGroupAttr()->
getOutputLogPropList()[0]->
getResultCardinality().value(),
isRowsetIterator(),
tolerateNonFatalError);
}
#pragma warn(1506) // warning elimination
generator->initTdbFields(unPackTdb);
// Remove child's outputs from mapTable, They are not needed
// above.
//
generator->removeAll(localMapTable);
// Add the explain Information for this node to the EXPLAIN
// Fragment. Set the explainTuple pointer in the generator so
// the parent of this node can get a handle on this explainTuple.
//
if(!generator->explainDisabled()) {
generator->setExplainTuple(addExplainInfo(unPackTdb,
childExplainTuple,
0,
generator));
}
// Restore the Cri Desc's and set the return object.
//
generator->setCriDesc(givenCriDesc, Generator::DOWN);
generator->setCriDesc(returnedCriDesc, Generator::UP);
generator->setGenObj(this, unPackTdb);
return 0;
}
short TriRelational::mdamPredGen(Generator * generator,
MdamPred ** head,
MdamPred ** tail,
MdamCodeGenHelper & mdamHelper,
ItemExpr * parent)
{
// temp -- haven't been able to unit test this code yet because I haven't been
// able to figure out how to get the Optimizer to pick a TriRelational guy as
// a key predicate -- seems better to have the code abort rather than run unverified
// and possibly fail in strange ways
GenAssert(0, "Reached TriRelational::mdamPredGen");
return -1;
// end temp code
#pragma nowarn(269) // warning elimination
short rc = 0;
#pragma warn(269) // warning elimination
enum MdamPred::MdamPredType predType =
MdamPred::MDAM_EQ; // just to initialize
// Find out what kind of predicate this is. Note that for DESCending
// columns, we reverse the direction of any comparison.
switch (getOperatorType())
{
case ITM_LESS_OR_LE:
{
predType = MdamPred::MDAM_LT;
if (mdamHelper.isDescending())
predType = MdamPred::MDAM_GT;
break;
}
case ITM_GREATER_OR_GE:
{
predType = MdamPred::MDAM_GT;
if (mdamHelper.isDescending())
predType = MdamPred::MDAM_LT;
break;
}
default:
{
GenAssert(0, "mdamPredGen: unsupported TriRelational comparison.");
break;
}
}
ItemExpr * child0 = child(0);
ItemExpr * child1 = child(1);
ValueId keyColumn = mdamHelper.getKeyColumn();
// assume predicate is <key> <compare> <value>
ItemExpr * keyValue = child1;
if (child1->getValueId() == keyColumn)
{
// we guessed wrong -- predicate is <value> <compare> <key>
keyValue = child0;
GenAssert(child0->getValueId() != keyColumn,
"mdamPredGen: unexpected form for key predicate.");
// $$$ Add code here to reverse the comparison?
}
else
{
GenAssert(child0->getValueId() == keyColumn,
"mdamPredGen: unexpected form for key predicate.");
}
// generate an expression to convert the key value to the
// type of the key column (in its key buffer) and encode it
ItemExpr * vnode = 0;
// errorsCanOccur() determines if errors can occur converting the class
// datatype to the target datatype. The object on whose behalf the
// member function is called is expected to be a NAType.
NABoolean generateNarrow =
keyValue->getValueId().getType().errorsCanOccur(*mdamHelper.getTargetType());
if ((generateNarrow) &&
(getenv("NO_NARROWS"))) // for testing -- allows turning off Narrows
generateNarrow = FALSE;
if (generateNarrow)
{
vnode = new(generator->wHeap())
Narrow(keyValue,
mdamHelper.getDataConversionErrorFlag(),
mdamHelper.getTargetType()->newCopy());
}
else
{
vnode = new(generator->wHeap())
Cast(keyValue,mdamHelper.getTargetType()->newCopy());
}
#pragma nowarn(1506) // warning elimination
vnode = new CompEncode(vnode,mdamHelper.isDescending());
#pragma warn(1506) // warning elimination
vnode->bindNode(generator->getBindWA());
// add CASE
// WHEN child(2)
// CAST(round up/round down)
// ELSE
// no-op
ItemExpr * hnode = 0;
if (predType == MdamPred::MDAM_LT)
hnode = new ConstValue(2 /* ex_conv_clause::CONV_RESULT_ROUNDED_UP_TO_MIN */);
else
hnode = new ConstValue(-2 /* ex_conv_clause::CONV_RESULT_ROUNDED_DOWN_TO_MAX */);
hnode = generator->getExpGenerator()->
createExprTree("CASE WHEN @B1 THEN @A2 ELSE @A3 END",
0,
3, // number of subtree parameters
child(2), // @B1
hnode, // @A2
0); // @A3 -- results in no operation
hnode->bindNode(generator->getBindWA());
// Assign attributes for result value
ValueId vnodeId = vnode->getValueId();
ValueId hnodeId = hnode->getValueId();
ULng32 tupleLength = 0;
ValueIdList vnodeList;
vnodeList.insert(vnode->getValueId());
generator->getExpGenerator()->processValIdList(
vnodeList,
mdamHelper.getTupleDataFormat(),
tupleLength, // out
mdamHelper.getAtp(),
mdamHelper.getAtpIndex());
// Assign attributes for modifying data conversion error flag
// Note that all we do is copy the already-assigned attributes
ItemExpr * dataCEF = mdamHelper.getDataConversionErrorFlag();
ValueId dataCEFId = dataCEF->getValueId();
Attributes * dataCEFAttr =
(generator->getMapInfo(dataCEFId))->getAttr();
generator->addMapInfoToThis(generator->getLastMapTable(), hnodeId,dataCEFAttr);
// finally generate the expression and hang it off an MdamPred
ex_expr *vexpr = 0;
vnodeList.insert(hnode->getValueId()); // put hnode in there too
rc = generator->getExpGenerator()->generateListExpr(
vnodeList,
ex_expr::exp_ARITH_EXPR,
&vexpr);
#pragma nowarn(1506) // warning elimination
*head = *tail = new(generator->getSpace())
MdamPred(mdamHelper.getDisjunctNumber(),
predType,
vexpr);
#pragma warn(1506) // warning elimination
return rc;
}
// BiRelat for which the following is called will be a predicate for one of the
// endpoints of an MDAM_BETWEEN.
void BiRelat::getMdamPredDetails(Generator* generator,
MdamCodeGenHelper& mdamHelper,
MdamPred::MdamPredType& predType,
ex_expr** vexpr)
{
// Find out what kind of predicate this is. Inequality preds are not inverted
// for descending keys here; instead, the endpoints of the MDAM_BETWEEN
// interval are switched during creation of the mdam network in the executor.
switch (getOperatorType())
{
case ITM_LESS:
predType = MdamPred::MDAM_LT;
break;
case ITM_LESS_EQ:
predType = MdamPred::MDAM_LE;
break;
case ITM_GREATER:
predType = MdamPred::MDAM_GT;
break;
case ITM_GREATER_EQ:
predType = MdamPred::MDAM_GE;
break;
default:
GenAssert(0, "mdamPredGen: invalid comparison for subrange.");
break;
}
ItemExpr* child0 = child(0);
ItemExpr* child1 = child(1);
ValueId keyColumn = mdamHelper.getKeyColumn();
// Canonical form used by rangespec is <key> <compare> <value>.
ItemExpr* keyValue = child1;
GenAssert(child0->getValueId() == keyColumn,
"mdamPredGen: unexpected form for key predicate.");
// generate an expression to convert the key value to the
// type of the key column (in its key buffer) and encode it
ItemExpr* vnode = NULL;
// errorsCanOccur() determines if errors can occur converting the class
// datatype to the target datatype. The object on whose behalf the
// member function is called is expected to be a NAType.
NABoolean generateNarrow =
keyValue->getValueId().getType().errorsCanOccur(*mdamHelper.getTargetType());
#ifdef _DEBUG
if ((generateNarrow) &&
(getenv("NO_NARROWS"))) // for testing -- allows turning off Narrows
generateNarrow = FALSE;
#endif
if (generateNarrow)
vnode = new(generator->wHeap())
Narrow(keyValue,
mdamHelper.getDataConversionErrorFlag(),
mdamHelper.getTargetType()->newCopy(generator->wHeap()));
else
vnode = new(generator->wHeap())
Cast(keyValue,
mdamHelper.getTargetType()->newCopy(generator->wHeap()));
vnode->bindNode(generator->getBindWA());
vnode->preCodeGen(generator);
#pragma nowarn(1506) // warning elimination
vnode = new(generator->wHeap()) CompEncode(vnode,mdamHelper.isDescending());
#pragma warn(1506) // warning elimination
vnode->bindNode(generator->getBindWA());
ValueIdList vnodeList;
vnodeList.insert(vnode->getValueId());
ULng32 dummyLen = 0;
short rc =
generator->getExpGenerator()
->generateContiguousMoveExpr(vnodeList,
0, // don't add convert nodes
mdamHelper.getAtp(),
mdamHelper.getAtpIndex(),
mdamHelper.getTupleDataFormat(),
dummyLen, // out
vexpr);
GenAssert(rc == 0, "generateContiguousMoveExpr() returned error when called "
"from BiRelat::getMdamPredDetails().");
}
short BiRelat::mdamPredGen(Generator * generator,
MdamPred ** head,
MdamPred ** tail,
MdamCodeGenHelper & mdamHelper,
ItemExpr * parent)
{
short rc = 0; // assume success
enum MdamPred::MdamPredType predType =
MdamPred::MDAM_EQ; // just to initialize
// Find out what kind of predicate this is. Note that for DESCending
// columns, we reverse the direction of any comparison.
switch (getOperatorType())
{
case ITM_EQUAL:
{
predType = MdamPred::MDAM_EQ;
break;
}
case ITM_LESS:
{
predType = MdamPred::MDAM_LT;
if (mdamHelper.isDescending()) predType = MdamPred::MDAM_GT;
break;
}
case ITM_LESS_EQ:
{
predType = MdamPred::MDAM_LE;
if (mdamHelper.isDescending()) predType = MdamPred::MDAM_GE;
break;
}
case ITM_GREATER:
{
predType = MdamPred::MDAM_GT;
if (mdamHelper.isDescending()) predType = MdamPred::MDAM_LT;
break;
}
case ITM_GREATER_EQ:
{
predType = MdamPred::MDAM_GE;
if (mdamHelper.isDescending()) predType = MdamPred::MDAM_LE;
break;
}
case ITM_ITEM_LIST:
{
GenAssert(0, "mdamPredGen: encountered multivalued predicate.");
break;
}
default:
{
GenAssert(0, "mdamPredGen: unsupported comparison.");
break;
}
}
ItemExpr * child0 = child(0);
ItemExpr * child1 = child(1);
ValueId keyColumn = mdamHelper.getKeyColumn();
// assume predicate is <key> <compare> <value>
ItemExpr * keyValue = child1;
if (child1->getValueId() == keyColumn)
{
// we guessed wrong -- predicate is <value> <compare> <key>
keyValue = child0;
GenAssert(child0->getValueId() != keyColumn,
"mdamPredGen: unexpected form for key predicate.");
// Reverse the comparison operator if it is <, <=, > or >=.
switch (predType)
{
case MdamPred::MDAM_LT:
{
predType = MdamPred::MDAM_GT;
break;
}
case MdamPred::MDAM_LE:
{
predType = MdamPred::MDAM_GE;
break;
}
case MdamPred::MDAM_GT:
{
predType = MdamPred::MDAM_LT;
break;
}
case MdamPred::MDAM_GE:
{
predType = MdamPred::MDAM_LE;
break;
}
} // switch (predType)
}
else
{
GenAssert(child0->getValueId() == keyColumn,
"mdamPredGen: unexpected form for key predicate.");
}
// generate an expression to convert the key value to the
// type of the key column (in its key buffer) and encode it
ItemExpr * vnode = 0;
// errorsCanOccur() determines if errors can occur converting the class
// datatype to the target datatype. The object on whose behalf the
// member function is called is expected to be a NAType.
NABoolean generateNarrow =
keyValue->getValueId().getType().errorsCanOccur(*mdamHelper.getTargetType());
#ifdef _DEBUG
if ((generateNarrow) &&
(getenv("NO_NARROWS"))) // for testing -- allows turning off Narrows
generateNarrow = FALSE;
#endif
if (generateNarrow)
{
vnode = new(generator->wHeap()) Narrow(keyValue,
mdamHelper.getDataConversionErrorFlag(),
mdamHelper.getTargetType()->newCopy(generator->wHeap()));
}
else
{
vnode = new(generator->wHeap()) Cast(keyValue,mdamHelper.getTargetType()->newCopy(generator->wHeap()));
}
vnode->bindNode(generator->getBindWA());
vnode->preCodeGen(generator);
#pragma nowarn(1506) // warning elimination
vnode = new(generator->wHeap()) CompEncode(vnode,mdamHelper.isDescending());
#pragma warn(1506) // warning elimination
vnode->bindNode(generator->getBindWA());
ValueIdList vnodeList;
vnodeList.insert(vnode->getValueId());
ex_expr *vexpr = 0;
ULng32 dummyLen = 0;
rc = generator->getExpGenerator()->generateContiguousMoveExpr(
vnodeList,
0, // don't add convert nodes
mdamHelper.getAtp(),
mdamHelper.getAtpIndex(),
mdamHelper.getTupleDataFormat(),
dummyLen, // out
&vexpr);
#pragma nowarn(1506) // warning elimination
*head = *tail = new(generator->getSpace()) MdamPred(
mdamHelper.getDisjunctNumber(),
predType,
vexpr);
#pragma warn(1506) // warning elimination
return rc;
}
void ItemExprList::insertTree(ItemExpr *tree,
OperatorTypeEnum backBoneType,
NABoolean flattenSBQ, NABoolean flattenUDF)
{
if (tree->getOperatorType() == backBoneType)
{
for (Int32 i = 0; i < tree->getArity(); i++)
{
// Check for NULL list for right linear trees. That is, arity may be
// two, but second child is NULL.
ItemExpr *child = tree->child(i);
if (child)
insertTree(tree->child(i), backBoneType, flattenSBQ, flattenUDF);
}
}
else if (tree->getOperatorType() == ITM_ONE_ROW)
{
Aggregate *agr = (Aggregate *)tree;
if (agr->isOneRowTransformed_)
{
for (Int32 i = 0; i < tree->getArity(); i++)
insertTree(tree->child(i), backBoneType, flattenSBQ, flattenUDF);
}
else
{
// do nothing, postpone this processing until OneRow transformation
// is done
}
}
else if ((flattenSBQ AND tree->isASubquery()) OR
(flattenUDF AND
(tree->getOperatorType() == ITM_USER_DEF_FUNCTION)) AND
(NOT tree->nodeIsTransformed()))
// Added the extra check for transformation above to avoid any issues
// where we might flatten a subquery/MVF a second time around while
// we deal with ValueIdProxies.
// The ValueIdProxy->needToTransformChild()
// flag should be sufficient, but it never hurts to be safe.
{
ValueIdList cols;
NABoolean haveRDesc(FALSE);
if (tree->isASubquery())
{
// flatten the subquery select list
RETDesc *retDesc = ((Subquery*)tree)->getSubquery()->getRETDesc();
if (retDesc)
{
retDesc->getColumnList()->getValueIdList(cols);
if (cols.entries() > 1)
{
haveRDesc = TRUE;
}
}
}
else if (tree->getOperatorType() == ITM_USER_DEF_FUNCTION)
{
// flatten the UDF by adding the additional outputs to the tree
const RoutineDesc *rDesc = ((UDFunction *)tree)->getRoutineDesc();
if (rDesc && rDesc->getOutputColumnList().entries() > 1)
{
cols = rDesc->getOutputColumnList();
haveRDesc = TRUE;
}
}
if (haveRDesc == TRUE)
{
for (CollIndex i = 0; i < cols.entries(); i++)
{
ValueId proxyId;
proxyId = cols[i];
// We create a ValueIdProxy for each element in the subquery's
// select list or for each output parameter of a MVF. The first
// one of these will be marked to be transformed. This allows
// us to get the correct degree of statements containing MVFs or
// subquery with degree > 1 at bind time.
ValueIdProxy *proxyOutput =
new (CmpCommon::statementHeap())
ValueIdProxy( tree->getValueId(),
proxyId,
i);
proxyOutput->synthTypeAndValueId();
// Make sure we transform the subquery or MVF
if (i == 0 ) proxyOutput->setTransformChild(TRUE);
insert(proxyOutput);
}
}
else
insert(tree); // we are processing a valueId of a UDFunction
// or subquery before we have bound it. Just insert
// its valueId and we'll have to deal with it later..
}
else
insert(tree);
}
void RangePartitioningFunction::generatePivLayout(
Generator *generator,
Lng32 &partitionInputDataLength,
Lng32 atp,
Lng32 atpIndex,
Attributes ***pivAttrs)
{
// Make a layout of the partition input data record such that
// begin and end key are aligned in the same way.
// (layout = ((beg. key) (filler1) (end key) (filler2) (exclusion flag)))
ExpGenerator *expGen = generator->getExpGenerator();
CollIndex numPartInputs = getPartitionInputValuesLayout().entries();
CollIndex numPartKeyCols = (numPartInputs - 1) / 2;
// the number of partition input variables must be odd
GenAssert(2*numPartKeyCols+1 == numPartInputs,
"NOT 2*numPartKeyCols+1 == numPartInputs");
// ---------------------------------------------------------------------
// Start by processing the begin key PIVs
// ---------------------------------------------------------------------
ValueIdList partialPivs;
Attributes **returnedAttrs = NULL;
Attributes **localPartialAttrs;
Lng32 maxAlignment = 1;
Lng32 alignedPartKeyLen;
if (pivAttrs)
{
returnedAttrs = new(generator->wHeap()) Attributes *[numPartInputs];
}
CollIndex i = 0;
for (i = 0; i < numPartKeyCols; i++)
partialPivs.insert(getPartitionInputValuesLayout()[i]);
expGen->processValIdList(
partialPivs,
ExpTupleDesc::SQLARK_EXPLODED_FORMAT,
(ULng32 &) partitionInputDataLength,
atp,
atpIndex,
NULL,
ExpTupleDesc::SHORT_FORMAT,
0,
&localPartialAttrs);
if (returnedAttrs)
for (i = 0; i < numPartKeyCols; i++)
returnedAttrs[i] = localPartialAttrs[i];
// ---------------------------------------------------------------------
// Now find out the max. alignment that is needed in the begin key,
// make sure that the end key starts on an offset that is a
// multiple of the max. alignment in the partition input values
// ---------------------------------------------------------------------
for (i = 0; i < numPartKeyCols; i++)
{
if (localPartialAttrs[i]->getDataAlignmentSize() > maxAlignment)
maxAlignment = localPartialAttrs[i]->getDataAlignmentSize();
if (localPartialAttrs[i]->getVCIndicatorLength() > maxAlignment)
maxAlignment = localPartialAttrs[i]->getVCIndicatorLength();
if (localPartialAttrs[i]->getNullIndicatorLength() > maxAlignment)
maxAlignment = localPartialAttrs[i]->getNullIndicatorLength();
}
alignedPartKeyLen = partitionInputDataLength;
while (alignedPartKeyLen % maxAlignment != 0)
alignedPartKeyLen++;
// ---------------------------------------------------------------------
// Now that we are starting on a good offset, process the end key
// ---------------------------------------------------------------------
partialPivs.clear();
for (i = numPartKeyCols; i < numPartInputs-1; i++)
partialPivs.insert(getPartitionInputValuesLayout()[i]);
expGen->processValIdList(
partialPivs,
ExpTupleDesc::SQLARK_EXPLODED_FORMAT,
(ULng32 &) partitionInputDataLength,
atp,
atpIndex,
NULL,
ExpTupleDesc::SHORT_FORMAT,
alignedPartKeyLen,
&localPartialAttrs);
if (returnedAttrs)
for (i = numPartKeyCols; i < numPartInputs-1; i++)
returnedAttrs[i] = localPartialAttrs[i-numPartKeyCols];
// ---------------------------------------------------------------------
// Process the exclusion flag at offset 2*alignedPartKeyLen
// ---------------------------------------------------------------------
partialPivs.clear();
partialPivs.insert(getPartitionInputValuesLayout()[numPartInputs-1]);
expGen->processValIdList(
partialPivs,
ExpTupleDesc::SQLARK_EXPLODED_FORMAT,
(ULng32 &) partitionInputDataLength,
atp,
atpIndex,
NULL,
ExpTupleDesc::SHORT_FORMAT,
2*alignedPartKeyLen,
&localPartialAttrs);
// set up return values
if (returnedAttrs)
returnedAttrs[numPartInputs-1] = localPartialAttrs[0];
partitionInputDataLength += 2*alignedPartKeyLen;
if (pivAttrs)
{
*pivAttrs = returnedAttrs;
}
else
{
NADELETEBASIC(returnedAttrs, generator->wHeap());
}
}
short RangePartitioningFunction::codeGen(Generator *generator,
Lng32 partInputDataLength)
{
ExpGenerator * exp_gen = generator->getExpGenerator();
Lng32 myOwnPartInputDataLength;
const Int32 pivMoveAtp = 0; // only one atp is used for this expr
const Int32 pivMoveAtpIndex = 2; // 0: consts, 1: temps, 2: result
const ExpTupleDesc::TupleDataFormat pivFormat = // format of PIVs
ExpTupleDesc::SQLARK_EXPLODED_FORMAT;
ex_cri_desc *partInputCriDesc = new(generator->getSpace())
ex_cri_desc(pivMoveAtpIndex+1,
generator->getSpace());
ExpTupleDesc *partInputTupleDesc;
ExRangePartInputData *generatedObject = NULL;
// get the list of partition input variables
ValueIdList piv(getPartitionInputValuesLayout());
CollIndex numPartInputs = piv.entries();
CollIndex numPartKeyCols = (numPartInputs - 1) / 2;
// the number of partition input variables must be odd
GenAssert(2*numPartKeyCols+1 == numPartInputs,
"NOT 2*numPartKeyCols+1 == numPartInputs");
Attributes **begEndAttrs;
Int32 alignedPartKeyLen;
// make a layout of the partition input data record
generatePivLayout(
generator,
myOwnPartInputDataLength,
pivMoveAtp,
pivMoveAtpIndex,
&begEndAttrs);
// the aligned part key length is where the end key values start
alignedPartKeyLen = (Int32) begEndAttrs[numPartKeyCols]->getOffset();
if (begEndAttrs[numPartKeyCols]->getNullIndicatorLength() > 0)
alignedPartKeyLen = MINOF(
alignedPartKeyLen,
(Int32)begEndAttrs[numPartKeyCols]->getNullIndOffset());
if (begEndAttrs[numPartKeyCols]->getVCIndicatorLength() > 0)
alignedPartKeyLen = MINOF(
alignedPartKeyLen,
begEndAttrs[numPartKeyCols]->getVCLenIndOffset());
// generate a tuple desc for the whole PIV record and a cri desc
partInputTupleDesc = new(generator->getSpace()) ExpTupleDesc(
numPartInputs,
begEndAttrs,
myOwnPartInputDataLength,
pivFormat,
ExpTupleDesc::LONG_FORMAT,
generator->getSpace());
partInputCriDesc->setTupleDescriptor(pivMoveAtpIndex,partInputTupleDesc);
// make sure we fulfill the assertions we made
// optimizer and generator should agree on the part input data length
GenAssert(partInputDataLength == (Lng32) myOwnPartInputDataLength,
"NOT partInputDataLength == myOwnPartInputDataLength");
// the length of the begin key and the end key must be the same
// (compare offsets of their last fields)
// Commented out because this check does not work. The check needs
// to compute the LENGTH of each key field, by subtracting the current
// offset from the next offset, taking into account varchar length
// and null indicator fields (which are not part of the length but
// increase the offset).
//GenAssert(begEndAttrs[numPartKeyCols-1]->getOffset() + alignedPartKeyLen ==
// begEndAttrs[2*numPartKeyCols-1]->getOffset(),
// "begin/end piv keys have different layouts");
#pragma nowarn(1506) // warning elimination
generatedObject = new(generator->getSpace()) ExRangePartInputData(
partInputCriDesc,
partInputDataLength,
alignedPartKeyLen, //len of one part key + filler
begEndAttrs[numPartInputs-1]->getOffset(),//offset of last field
getCountOfPartitions(),
generator->getSpace(),
TRUE); // uses expressions to calculate ranges in the executor
generatedObject->setPartitionExprAtp(pivMoveAtp);
generatedObject->setPartitionExprAtpIndex(pivMoveAtpIndex);
#pragma warn(1506) // warning elimination
// now fill in the individual partition boundaries
// (NOTE: there is one more than there are partitions)
ULng32 boundaryDataLength = 0;
for (Lng32 i = 0; i <= getCountOfPartitions(); i++)
{
const ItemExprList *iel = partitionBoundaries_->getBoundaryValues(i);
ex_expr * generatedExpr = NULL;
ValueIdList boundaryColValues;
ULng32 checkedBoundaryLength;
// convert the ItemExpressionList iel into a ValueIdList
for (CollIndex kc = 0; kc < iel->entries(); kc++)
{
ItemExpr *boundaryVal = (*iel)[kc];
// create a cast node to convert the boundary value to the
// data type of the column
ItemExpr *castBoundaryVal =
new(generator->wHeap()) Cast(boundaryVal,&piv[kc].getType());
castBoundaryVal->bindNode(generator->getBindWA());
boundaryColValues.insert(castBoundaryVal->getValueId());
}
// Now generate a contiguous move expression. Only for the first time
// generate a tuple desc, since all tuples should be the same.
exp_gen->generateContiguousMoveExpr(
boundaryColValues,
0, // cast nodes created above will do the move, no conv nodes
pivMoveAtp,
pivMoveAtpIndex,
pivFormat,
checkedBoundaryLength,
&generatedExpr);
if (i == 0)
{
// first time set the actual part key data length
boundaryDataLength = checkedBoundaryLength;
}
else
{
// all boundary values (piv tuples) must have the same layout
// and therefore the same length
GenAssert(boundaryDataLength == checkedBoundaryLength,
"Partition boundary tuple layout mismatch");
}
generatedObject->setPartitionStartExpr(i,generatedExpr);
}
NADELETEBASIC(begEndAttrs, generator->wHeap());
generator->setGenObj(NULL, (ComTdb*)generatedObject);
return 0;
}
static short ft_codegen(Generator *generator,
RelExpr &relExpr,
ComTdbFastExtract *&newTdb,
Cardinality estimatedRowCount,
char * targetName,
char * hdfsHost,
Int32 hdfsPort,
char * hiveTableName,
char * delimiter,
char * header,
char * nullString,
char * recordSeparator,
ULng32 downQueueMaxSize,
ULng32 upQueueMaxSize,
ULng32 outputBufferSize,
ULng32 requestBufferSize,
ULng32 replyBufferSize,
ULng32 numOutputBuffers,
ComTdb * childTdb,
NABoolean isSequenceFile)
{
CmpContext *cmpContext = generator->currentCmpContext();
Space *space = generator->getSpace();
ExpGenerator *exp_gen = generator->getExpGenerator();
MapTable *map_table = generator->getMapTable();
MapTable *last_map_table = generator->getLastMapTable();
ex_expr *input_expr = NULL;
ex_expr *output_expr = NULL;
ex_expr * childData_expr = NULL ;
ex_expr * cnvChildData_expr = NULL ;
ULng32 i;
ULng32 requestRowLen = 0;
ULng32 outputRowLen = 0;
ULng32 childDataRowLen = 0;
ULng32 cnvChildDataRowLen = 0;
ExpTupleDesc *requestTupleDesc = NULL;
ExpTupleDesc *outputTupleDesc = NULL;
ExpTupleDesc *childDataTupleDesc = NULL;
ExpTupleDesc *cnvChildDataTupleDesc = NULL;
newTdb = NULL;
OperatorTypeEnum relExprType = relExpr.getOperatorType();
GenAssert(relExprType == REL_FAST_EXTRACT, "Unexpected RelExpr at FastExtract codegen")
FastExtract * fastExtract = (FastExtract *) &relExpr;
const Int32 workAtpNumber = 1;
ex_cri_desc *given_desc = generator->getCriDesc(Generator::DOWN);
ex_cri_desc *returned_desc = NULL;
ex_cri_desc *work_cri_desc = NULL;
returned_desc = given_desc;
// Setup local variables related to the work ATP
unsigned short numWorkTupps = 0;
unsigned short childDataTuppIndex = 0;
unsigned short cnvChildDataTuppIndex = 0;
numWorkTupps = 3;
childDataTuppIndex = numWorkTupps - 1 ;
numWorkTupps ++;
cnvChildDataTuppIndex = numWorkTupps - 1;
work_cri_desc = new (space) ex_cri_desc(numWorkTupps, space);
ExpTupleDesc::TupleDataFormat childReqFormat = ExpTupleDesc::SQLMX_ALIGNED_FORMAT;
ValueIdList childDataVids;
ValueIdList cnvChildDataVids;
const ValueIdList& childVals = fastExtract->getSelectList();
const NATable *hiveNATable = NULL;
const NAColumnArray *hiveNAColArray = NULL;
// hiveInsertErrMode:
// if 0, do not do error checks.
// if 1, do error check and return error.
// if 2, do error check and ignore row, if error
// if 3, insert null if an error occurs
Lng32 hiveInsertErrMode = 0;
if ((fastExtract) && (fastExtract->isHiveInsert()) &&
(fastExtract->getHiveTableDesc()) &&
(fastExtract->getHiveTableDesc()->getNATable()) &&
((hiveInsertErrMode = CmpCommon::getDefaultNumeric(HIVE_INSERT_ERROR_MODE)) > 0))
{
hiveNATable = fastExtract->getHiveTableDesc()->getNATable();
hiveNAColArray = &hiveNATable->getNAColumnArray();
}
for (i = 0; i < childVals.entries(); i++)
{
ItemExpr &inputExpr = *(childVals[i].getItemExpr());
const NAType &formalType = childVals[i].getType();
ItemExpr *lmExpr = NULL;
ItemExpr *lmExpr2 = NULL;
int res;
lmExpr = &inputExpr;
lmExpr = lmExpr->bindNode(generator->getBindWA());
if (!lmExpr || generator->getBindWA()->errStatus())
{
GenAssert(0, "lmExpr->bindNode failed");
}
// Hive insert converts child data into string format and inserts
// it into target table.
// If child type can into an error during conversion, then
// add a Cast to convert from child type to target type before
// converting to string format to be inserted.
if (hiveNAColArray)
{
const NAColumn *hiveNACol = (*hiveNAColArray)[i];
const NAType *hiveNAType = hiveNACol->getType();
// if tgt type was a hive 'string', do not return a conversion err
if ((lmExpr->getValueId().getType().errorsCanOccur(*hiveNAType)) &&
(NOT ((DFS2REC::isSQLVarChar(hiveNAType->getFSDatatype())) &&
(((SQLVarChar*)hiveNAType)->wasHiveString()))))
{
ItemExpr *newExpr =
new(generator->wHeap()) Cast(lmExpr, hiveNAType);
newExpr = newExpr->bindNode(generator->getBindWA());
if (!newExpr || generator->getBindWA()->errStatus())
{
GenAssert(0, "newExpr->bindNode failed");
}
if (hiveInsertErrMode == 3)
((Cast*)newExpr)->setConvertNullWhenError(TRUE);
lmExpr = newExpr;
}
}
res = CreateAllCharsExpr(formalType, // [IN] Child output type
*lmExpr, // [IN] Actual input value
cmpContext, // [IN] Compilation context
lmExpr2 // [OUT] Returned expression
);
GenAssert(res == 0 && lmExpr != NULL,
"Error building expression tree for LM child Input value");
childDataVids.insert(lmExpr->getValueId());
if (lmExpr2)
{
lmExpr2->bindNode(generator->getBindWA());
cnvChildDataVids.insert(lmExpr2->getValueId());
}
} // for (i = 0; i < childVals.entries(); i++)
if (childDataVids.entries() > 0 &&
cnvChildDataVids.entries()>0) //-- convertedChildDataVids
{
UInt16 pcm = exp_gen->getPCodeMode();
if ((hiveNAColArray) &&
(hiveInsertErrMode == 3))
{
// if error mode is 3 (mode null when error), disable pcode.
// this feature is currently not being handled by pcode.
// (added as part of JIRA 1920 in FileScan::codeGenForHive).
exp_gen->setPCodeMode(ex_expr::PCODE_NONE);
}
exp_gen->generateContiguousMoveExpr (
childDataVids, //childDataVids// [IN] source ValueIds
TRUE, // [IN] add convert nodes?
workAtpNumber, // [IN] target atp number (0 or 1)
childDataTuppIndex, // [IN] target tupp index
childReqFormat, // [IN] target tuple data format
childDataRowLen, // [OUT] target tuple length
&childData_expr, // [OUT] move expression
&childDataTupleDesc, // [optional OUT] target tuple desc
ExpTupleDesc::LONG_FORMAT // [optional IN] target desc format
);
exp_gen->setPCodeMode(pcm);
exp_gen->processValIdList (
cnvChildDataVids, // [IN] ValueIdList
ExpTupleDesc::SQLARK_EXPLODED_FORMAT, // [IN] tuple data format
cnvChildDataRowLen, // [OUT] tuple length
workAtpNumber, // [IN] atp number
cnvChildDataTuppIndex, // [IN] index into atp
&cnvChildDataTupleDesc, // [optional OUT] tuple desc
ExpTupleDesc::LONG_FORMAT // [optional IN] tuple desc format
);
}
//
// Add the tuple descriptor for request values to the work ATP
//
work_cri_desc->setTupleDescriptor(childDataTuppIndex, childDataTupleDesc);
work_cri_desc->setTupleDescriptor(cnvChildDataTuppIndex, cnvChildDataTupleDesc);
// We can now remove all appended map tables
generator->removeAll(last_map_table);
ComSInt32 maxrs = 0;
UInt32 flags = 0;
UInt16 numIoBuffers = (UInt16)(ActiveSchemaDB()->getDefaults()).getAsLong(FAST_EXTRACT_IO_BUFFERS);
UInt16 ioTimeout = (UInt16)(ActiveSchemaDB()->getDefaults()).getAsLong(FAST_EXTRACT_IO_TIMEOUT_SEC);
Int64 hdfsBufSize = (Int64)CmpCommon::getDefaultNumeric(HDFS_IO_BUFFERSIZE);
hdfsBufSize = hdfsBufSize * 1024; // convert to bytes
Int16 replication = (Int16)CmpCommon::getDefaultNumeric(HDFS_REPLICATION);
// Create a TDB
ComTdbFastExtract *tdb = new (space) ComTdbFastExtract (
flags,
estimatedRowCount,
targetName,
hdfsHost,
hdfsPort,
hiveTableName,
delimiter,
header,
nullString,
recordSeparator,
given_desc,
returned_desc,
work_cri_desc,
downQueueMaxSize,
upQueueMaxSize,
(Lng32) numOutputBuffers,
outputBufferSize,
numIoBuffers,
ioTimeout,
input_expr,
output_expr,
requestRowLen,
outputRowLen,
childData_expr,
childTdb,
space,
childDataTuppIndex,
cnvChildDataTuppIndex,
childDataRowLen,
hdfsBufSize,
replication
);
tdb->setSequenceFile(isSequenceFile);
tdb->setHdfsCompressed(CmpCommon::getDefaultNumeric(TRAF_UNLOAD_HDFS_COMPRESS)!=0);
tdb->setSkipWritingToFiles(CmpCommon::getDefault(TRAF_UNLOAD_SKIP_WRITING_TO_FILES) == DF_ON);
tdb->setBypassLibhdfs(CmpCommon::getDefault(TRAF_UNLOAD_BYPASS_LIBHDFS) == DF_ON);
if ((hiveNAColArray) &&
(hiveInsertErrMode == 2))
{
tdb->setContinueOnError(TRUE);
}
generator->initTdbFields(tdb);
// Generate EXPLAIN info.
if (!generator->explainDisabled())
{
generator->setExplainTuple(relExpr.addExplainInfo(tdb, 0, 0, generator));
}
// Tell the generator about our in/out rows and the new TDB
generator->setCriDesc(given_desc, Generator::DOWN);
generator->setCriDesc(returned_desc, Generator::UP);
generator->setGenObj(&relExpr, tdb);
// Return a TDB pointer to the caller
newTdb = tdb;
return 0;
} // ft_codegen()
short RelInternalSP::codeGen(Generator * generator)
{
Space * space = generator->getSpace();
ExpGenerator * exp_gen = generator->getExpGenerator();
MapTable * last_map_table = generator->getLastMapTable();
ex_expr * input_expr = NULL;
ex_expr * output_expr = NULL;
////////////////////////////////////////////////////////////////////////////
//
// Returned atp layout:
//
// |--------------------------------|
// | input data | stored proc row |
// | ( I tupps ) | ( 1 tupp ) |
// |--------------------------------|
// <-- returned row to parent ---->
//
// input data: the atp input to this node by its parent.
// stored proc row: tupp where the row read from SP is moved.
//
////////////////////////////////////////////////////////////////////////////
ex_cri_desc * given_desc
= generator->getCriDesc(Generator::DOWN);
ex_cri_desc * returned_desc
= new(space) ex_cri_desc(given_desc->noTuples() + 1, space);
// cri descriptor for work atp has 3 entries:
// -- the first two entries for consts and temps.
// -- Entry 3(index #2) is where the input and output rows will be created.
ex_cri_desc * work_cri_desc = new(space) ex_cri_desc(3, space);
const Int32 work_atp = 1;
const Int32 work_atp_index = 2;
ExpTupleDesc * input_tuple_desc = NULL;
ExpTupleDesc * output_tuple_desc = NULL;
// Generate expression to create the input row that will be
// given to the stored proc.
// The input value is in sp->getProcAllParams()
// and has to be converted to sp->procType().
// Generate Cast node to convert procParam to ProcType.
// If procType is a varchar, explode it. This is done
// so that values could be extracted correctly.
ValueIdList procVIDList;
for (CollIndex i = 0; i < procTypes().entries(); i++)
{
Cast * cn;
if ((procTypes())[i].getType().getVarLenHdrSize() > 0)
{
// 5/9/98: add support for VARNCHAR
const CharType& char_type =
(CharType&)((procTypes())[i].getType());
// Explode varchars by moving them to a fixed field
// whose length is equal to the max length of varchar.
cn = new(generator->wHeap())
Cast ((getProcAllParamsVids())[i].getItemExpr(),
(new(generator->wHeap())
SQLChar(generator->wHeap(),
CharLenInfo(char_type.getStrCharLimit(), char_type.getDataStorageSize()),
char_type.supportsSQLnull(),
FALSE, FALSE, FALSE,
char_type.getCharSet(),
char_type.getCollation(),
char_type.getCoercibility()
/*
(procTypes())[i].getType().getNominalSize(),
(procTypes())[i].getType().supportsSQLnull()
*/
)
)
);
// Move the exploded field to a varchar field since
// procType is varchar.
// Can optimize by adding an option to convert node to
// blankpad. TBD.
//
cn = new(generator->wHeap())
Cast(cn, &((procTypes())[i].getType()));
}
else
cn = new(generator->wHeap()) Cast((getProcAllParamsVids())[i].getItemExpr(),
&((procTypes())[i].getType()));
cn->bindNode(generator->getBindWA());
procVIDList.insert(cn->getValueId());
}
ULng32 inputRowlen_ = 0;
exp_gen->generateContiguousMoveExpr(procVIDList, -1, /*add conv nodes*/
work_atp, work_atp_index,
ExpTupleDesc::SQLARK_EXPLODED_FORMAT,
inputRowlen_,
&input_expr,
&input_tuple_desc,
ExpTupleDesc::LONG_FORMAT);
// add all columns from this SP to the map table.
ULng32 tupleLength;
exp_gen->processValIdList(getTableDesc()->getColumnList(),
ExpTupleDesc::SQLARK_EXPLODED_FORMAT,
tupleLength,
work_atp,
work_atp_index);
// Generate expression to move the output row returned by the
// stored proc back to parent.
ULng32 outputRowlen_ = 0;
MapTable * returnedMapTable = 0;
exp_gen->generateContiguousMoveExpr(getTableDesc()->getColumnList(),
-1 /*add conv nodes*/,
0, returned_desc->noTuples() - 1,
ExpTupleDesc::SQLARK_EXPLODED_FORMAT,
outputRowlen_,
&output_expr,
&output_tuple_desc,
ExpTupleDesc::LONG_FORMAT,
&returnedMapTable);
// Now generate expressions used to extract or move input or
// output values. See class ExSPInputOutput.
ExSPInputOutput * extractInputExpr = NULL;
ExSPInputOutput * moveOutputExpr = NULL;
generateSPIOExpr(this, generator,
extractInputExpr,
moveOutputExpr);
// done with expressions at this operator. Remove the appended map tables.
generator->removeAll(last_map_table);
// append the map table containing the returned columns
generator->appendAtEnd(returnedMapTable);
NAString procNameAsNAString(procName_);
char * sp_name =
space->allocateAndCopyToAlignedSpace(procNameAsNAString,
procNameAsNAString.length(), 0);
ExpGenerator *expGen = generator->getExpGenerator();
// expression to conditionally return 0 or more rows.
ex_expr *predExpr = NULL;
// generate tuple selection expression, if present
if(NOT selectionPred().isEmpty())
{
ItemExpr* pred = selectionPred().rebuildExprTree(ITM_AND,TRUE,TRUE);
expGen->generateExpr(pred->getValueId(),ex_expr::exp_SCAN_PRED,&predExpr);
}
ComTdbStoredProc * sp_tdb = new(space)
ComTdbStoredProc(sp_name,
input_expr,
inputRowlen_,
output_expr,
outputRowlen_,
work_cri_desc,
work_atp_index,
given_desc,
returned_desc,
extractInputExpr,
moveOutputExpr,
2,
1024,
(Cardinality) getGroupAttr()->
getOutputLogPropList()[0]->
getResultCardinality().value(),
5,
64000, //10240
predExpr,
(UInt16) arkcmpInfo_);
generator->initTdbFields(sp_tdb);
if(!generator->explainDisabled())
{
generator->setExplainTuple(
addExplainInfo(sp_tdb, 0, 0, generator));
}
// Do not infer that any transaction started can
// be in READ ONLY mode if ISPs are present.
generator->setNeedsReadWriteTransaction(TRUE);
generator->setCriDesc(given_desc, Generator::DOWN);
generator->setCriDesc(returned_desc, Generator::UP);
generator->setGenObj(this, sp_tdb);
// Some built-in functions require a TMF transaction
// because they get their information from catman
generator->setTransactionFlag(getRequiresTMFTransaction());
return 0;
}
short ProbeCache::codeGen(Generator *generator)
{
ExpGenerator * exp_gen = generator->getExpGenerator();
Space * space = generator->getSpace();
MapTable * last_map_table = generator->getLastMapTable();
ex_cri_desc * given_desc
= generator->getCriDesc(Generator::DOWN);
ex_cri_desc * returned_desc
= new(space) ex_cri_desc(given_desc->noTuples() + 1, space);
// cri descriptor for work atp has 5 entries:
// entry #0 for const
// entry #1 for temp
// entry #2 for hash value of probe input data in Probe Cache Manager
// entry #3 for encoded probe input data in Probe Cache Manager
// enrry #4 for inner table row data in this operator's cache buffer
Int32 work_atp = 1;
ex_cri_desc * work_cri_desc = new(space) ex_cri_desc(5, space);
unsigned short hashValIdx = 2;
unsigned short encodedProbeDataIdx = 3;
unsigned short innerRowDataIdx = 4;
// generate code for child tree, and get its tdb and explain tuple.
child(0)->codeGen(generator);
ComTdb * child_tdb = (ComTdb *)(generator->getGenObj());
ExplainTuple *childExplainTuple = generator->getExplainTuple();
//////////////////////////////////////////////////////
// Generate up to 4 runtime expressions.
//////////////////////////////////////////////////////
// Will use child's char. inputs (+ execution count) for the next
// two runtime expressions.
ValueIdList inputsToUse = child(0).getGroupAttr()->getCharacteristicInputs();
inputsToUse.insert(generator->getOrAddStatementExecutionCount());
// Expression #1 gets the hash value of the probe input data
ValueIdList hvAsList;
// Executor has hard-coded assumption that the result is long,
// so add a Cast node to convert result to a long.
ItemExpr *probeHashAsIe = new (generator->wHeap())
HashDistPartHash(inputsToUse.rebuildExprTree(ITM_ITEM_LIST));
probeHashAsIe->bindNode(generator->getBindWA());
NumericType &nTyp = (NumericType &)probeHashAsIe->getValueId().getType();
GenAssert(nTyp.isSigned() == FALSE,
"Unexpected signed HashDistPartHash.");
GenAssert(probeHashAsIe->getValueId().getType().supportsSQLnullLogical()
== FALSE, "Unexpected nullable HashDistPartHash.");
ItemExpr *hvAsIe = new (generator->wHeap()) Cast(
probeHashAsIe,
new (generator->wHeap())
SQLInt(FALSE, // false == unsigned.
FALSE // false == not nullable.
));
hvAsIe->bindNode(generator->getBindWA());
hvAsList.insert(hvAsIe->getValueId());
ex_expr *hvExpr = NULL;
ULng32 hvLength;
exp_gen->generateContiguousMoveExpr(
hvAsList,
0, // don't add convert node
work_atp,
hashValIdx,
ExpTupleDesc::SQLARK_EXPLODED_FORMAT,
hvLength,
&hvExpr);
GenAssert(hvLength == sizeof(Lng32),
"Unexpected length of result of hash function.");
// Expression #2 encodes the probe input data for storage in
// the ProbeCacheManager.
ValueIdList encodeInputAsList;
CollIndex inputListIndex;
for (inputListIndex = 0;
inputListIndex < inputsToUse.entries();
inputListIndex++) {
ItemExpr *inputIe =
(inputsToUse[inputListIndex].getValueDesc())->getItemExpr();
if (inputIe->getValueId().getType().getVarLenHdrSize() > 0)
{
// This logic copied from Sort::codeGen().
// Explode varchars by moving them to a fixed field
// whose length is equal to the max length of varchar.
// 5/8/98: add support for VARNCHAR
const CharType& char_type =
(CharType&)(inputIe->getValueId().getType());
if (!CollationInfo::isSystemCollation(char_type.getCollation()))
{
inputIe = new(generator->wHeap())
Cast (inputIe,
(new(generator->wHeap())
SQLChar(
CharLenInfo(char_type.getStrCharLimit(), char_type.getDataStorageSize()),
char_type.supportsSQLnull(),
FALSE, FALSE, FALSE,
char_type.getCharSet(),
char_type.getCollation(),
char_type.getCoercibility()
)
)
);
}
}
CompEncode * enode = new(generator->wHeap())
CompEncode(inputIe, FALSE /* ascend/descend doesn't matter*/);
enode->bindNode(generator->getBindWA());
encodeInputAsList.insert(enode->getValueId());
}
ex_expr *encodeInputExpr = NULL;
ULng32 encodedInputLength;
exp_gen->generateContiguousMoveExpr(encodeInputAsList,
0, //don't add conv nodes
work_atp, encodedProbeDataIdx,
ExpTupleDesc::SQLARK_EXPLODED_FORMAT,
encodedInputLength, &encodeInputExpr);
// Expression #3 moves the inner table data into a buffer pool.
// This is also the tuple returned to ProbeCache's parent.
ex_expr * innerRecExpr = NULL;
ValueIdList innerTableAsList = getGroupAttr()->getCharacteristicOutputs();
//////////////////////////////////////////////////////
// Describe the returned row and add the returned
// values to the map table.
//////////////////////////////////////////////////////
// determine internal format
NABoolean useCif = FALSE;
ExpTupleDesc::TupleDataFormat tupleFormat = generator->getInternalFormat();
//tupleFormat = determineInternalFormat( innerTableAsList, this, useCif,generator);
ULng32 innerRecLength = 0;
ExpTupleDesc * innerRecTupleDesc = 0;
MapTable * returnedMapTable = NULL;
exp_gen->generateContiguousMoveExpr(innerTableAsList,
-1, // do add conv nodes
work_atp, innerRowDataIdx,
tupleFormat,
innerRecLength, &innerRecExpr,
&innerRecTupleDesc, ExpTupleDesc::SHORT_FORMAT,
&returnedMapTable);
returned_desc->setTupleDescriptor(returned_desc->noTuples() - 1,
innerRecTupleDesc);
// remove all appended map tables and return the returnedMapTable
generator->removeAll(last_map_table);
generator->appendAtEnd(returnedMapTable);
// This returnedMapTable will contain the value ids that are being returned
// (the inner table probed).
// Massage the atp and atp_index of the innerTableAsList.
for (CollIndex i = 0; i < innerTableAsList.entries(); i++)
{
ValueId innerValId = innerTableAsList[i];
Attributes *attrib =
generator->getMapInfo(innerValId)->getAttr();
// All reference to the returned values from this point on
// will be at atp = 0, atp_index = last entry in returned desc.
attrib->setAtp(0);
attrib->setAtpIndex(returned_desc->noTuples() - 1);
}
// Expression #4 is a selection predicate, to be applied
// before returning rows to the parent
ex_expr * selectPred = NULL;
if (!selectionPred().isEmpty())
{
ItemExpr * selPredTree =
selectionPred().rebuildExprTree(ITM_AND,TRUE,TRUE);
exp_gen->generateExpr(selPredTree->getValueId(),
ex_expr::exp_SCAN_PRED,
&selectPred);
}
//////////////////////////////////////////////////////
// Prepare params for ComTdbProbeCache.
//////////////////////////////////////////////////////
queue_index pDownSize = (queue_index)getDefault(GEN_PROBE_CACHE_SIZE_DOWN);
queue_index pUpSize = (queue_index)getDefault(GEN_PROBE_CACHE_SIZE_UP);
// Make sure that the ProbeCache queues can support the childs queues.
if(pDownSize < child_tdb->getInitialQueueSizeDown()) {
pDownSize = child_tdb->getInitialQueueSizeDown();
pDownSize = MINOF(pDownSize, 32768);
}
if(pUpSize < child_tdb->getInitialQueueSizeUp()) {
pUpSize = child_tdb->getInitialQueueSizeUp();
pUpSize = MINOF(pUpSize, 32768);
}
ULng32 pcNumEntries = numCachedProbes_;
// Number of entries in the probe cache cannot be less than
// max parent down queue size. Before testing and adjusting the
// max queue size, it is necessary to make sure it is a power of
// two, rounding up if necessary. This is to match the logic in
// ex_queue::resize.
queue_index pdq2 = 1;
queue_index bits = pDownSize;
while (bits && pdq2 < pDownSize) {
bits = bits >> 1;
pdq2 = pdq2 << 1;
}
if (pcNumEntries < pdq2)
pcNumEntries = pdq2;
numInnerTuples_ = getDefault(GEN_PROBE_CACHE_NUM_INNER);
if (innerRecExpr == NULL)
{
// For semi-join and anti-semi-join, executor need not allocate
// a buffer. Set the tdb's buffer size to 0 to be consistent.
numInnerTuples_ = 0;
}
else if (numInnerTuples_ == 0)
{
// Handle special value, 0, which tells code gen to
// decided on buffer size: i.e., large enough to accomodate
// all parent up queue entries and all probe cache entries
// having a different inner table row.
// As we did for the down queue, make sure the up queue size
// specified is a power of two.
queue_index puq2 = 1;
queue_index bits = pUpSize;
while (bits && puq2 < pUpSize) {
bits = bits >> 1;
puq2 = puq2 << 1;
}
numInnerTuples_ = puq2 + pcNumEntries;
}
short ExpGenerator::buildKeyInfo(keyRangeGen ** keyInfo, // out -- generated object
Generator * generator,
const NAColumnArray & keyColumns,
const ValueIdList & listOfKeyColumns,
const ValueIdList & beginKeyPred,
const ValueIdList & endKeyPred,
const SearchKey * searchKey,
const MdamKey * mdamKeyPtr,
const NABoolean reverseScan,
unsigned short keytag,
const ExpTupleDesc::TupleDataFormat tf,
// the next few parameters are here
// as part of a horrible kludge for
// the PartitionAccess::codeGen()
// method, which lacks a SearchKey
// object and therefore exposes
// things like the exclusion
// expressions; with luck, later work
// in the Optimizer will result in a
// much cleaner interface
const NABoolean useTheHorribleKludge,
ItemExpr * beginKeyExclusionExpr,
ItemExpr * endKeyExclusionExpr,
ex_expr_lean ** unique_key_expr,
ULng32 *uniqueKeyLen,
NABoolean doKeyEncodeOpt,
Lng32 * firstKeyColOffset,
Int32 in_key_atp_index
)
{
Space * space = generator->getSpace();
const Int32 work_atp = 1;
const Int32 key_atp_index = (in_key_atp_index <= 0 ? 2 : in_key_atp_index);
const Int32 exclude_flag_atp_index = 3;
const Int32 data_conv_error_atp_index = 4;
const Int32 key_column_atp_index = 5; // used only for Mdam
const Int32 key_column2_atp_index = 6; // used only for Mdam MDAM_BETWEEN pred;
// code in BiLogic::mdamPredGenSubrange
// and MdamColumn::buildDisjunct
// requires this to be 1 more than
// key_column_atp_index
ULng32 keyLen;
// add an entry to the map table for work Atp
MapTable *keyBufferPartMapTable = generator->appendAtEnd();
// generate a temporary variable, which will be used for handling
// data conversion errors during key building
ValueIdList temp_varb_list;
ItemExpr * dataConversionErrorFlag = new(generator->wHeap())
HostVar("_sys_dataConversionErrorFlag",
new(generator->wHeap()) SQLInt(TRUE,FALSE), // int not null
TRUE);
ULng32 temp_varb_tupp_len;
dataConversionErrorFlag->bindNode(generator->getBindWA());
temp_varb_list.insert(dataConversionErrorFlag->getValueId());
processValIdList(temp_varb_list,
ExpTupleDesc::SQLARK_EXPLODED_FORMAT,
temp_varb_tupp_len, // out
work_atp,
data_conv_error_atp_index);
NABoolean doEquiKeyPredOpt = FALSE;
#ifdef _DEBUG
if (getenv("DO_EQUI_KEY_PRED_OPT"))
doEquiKeyPredOpt
= (searchKey ? searchKey->areAllChosenPredsEqualPreds() : FALSE);
#endif
if (mdamKeyPtr == NULL)
{
// check to see if there is a begin key expression; if there
// isn't, don't generate a key object
if (beginKeyPred.entries() == 0)
*keyInfo = 0;
else
{
// For subset and range operators, generate the begin key
// expression, end key expression, begin key exclusion expression
// and end key exclusion expression. For unique operators,
// generate only the begin key exppression.
ex_expr *bk_expr = 0;
ex_expr *ek_expr = 0;
ex_expr *bk_excluded_expr = 0;
ex_expr *ek_excluded_expr = 0;
short bkey_excluded = 0;
short ekey_excluded = 0;
generateKeyExpr(keyColumns,
beginKeyPred,
work_atp,
key_atp_index,
dataConversionErrorFlag,
tf,
keyLen, // out
&bk_expr, // out
doKeyEncodeOpt,
firstKeyColOffset,
doEquiKeyPredOpt);
if (&endKeyPred)
generateKeyExpr(keyColumns,
endKeyPred,
work_atp,
key_atp_index,
dataConversionErrorFlag,
tf,
keyLen, // out -- should be the same as above
&ek_expr, // out
doKeyEncodeOpt,
firstKeyColOffset,
doEquiKeyPredOpt);
if (reverseScan)
{
// reverse scan - swap the begin and end key predicates
// Note: evidently, the Optimizer has already switched
// the key predicates in this case, so what we are
// really doing is switching them back.
ex_expr *temp = bk_expr;
bk_expr = ek_expr;
ek_expr = temp;
}
if (searchKey)
{
generateExclusionExpr(searchKey->getBeginKeyExclusionExpr(),
work_atp,
exclude_flag_atp_index,
&bk_excluded_expr); // out
bkey_excluded = (short) searchKey->isBeginKeyExclusive();
generateExclusionExpr(searchKey->getEndKeyExclusionExpr(),
work_atp,
exclude_flag_atp_index,
&ek_excluded_expr); // out
ekey_excluded = (short) searchKey->isEndKeyExclusive();
if (reverseScan)
{
NABoolean x = bkey_excluded;
bkey_excluded = ekey_excluded;
#pragma nowarn(1506) // warning elimination
ekey_excluded = x;
#pragma warn(1506) // warning elimination
ex_expr* temp = bk_excluded_expr;
bk_excluded_expr = ek_excluded_expr;
bk_excluded_expr = temp;
}
} // if searchKey
else if (useTheHorribleKludge)
{
generateExclusionExpr(beginKeyExclusionExpr,
work_atp,
exclude_flag_atp_index,
&bk_excluded_expr); // out
generateExclusionExpr(endKeyExclusionExpr,
work_atp,
exclude_flag_atp_index,
&ek_excluded_expr); // out
// note that the old PartitionAccess::codeGen() code didn't
// set values for bkey_excluded and ekey_excluded, so the
// safest choice is to choose inclusion, i.e. let the flags
// retain their initial value of 0.
}
// Build key info
if (keytag > 0)
keyLen += sizeof(short);
if ((unique_key_expr == NULL) ||
(NOT generator->genLeanExpr()))
{
// the work cri desc is used to build key values (entry 2) and
// to compute the exclusion flag (entry 3) to monitor for data
// conversion errors (entry 4) and to compute values on a column
// basis (entry 5 - Mdam only)
ex_cri_desc * work_cri_desc
= new(space) ex_cri_desc(6, space);
*keyInfo = new(space) keySingleSubsetGen(
keyLen,
work_cri_desc,
key_atp_index,
exclude_flag_atp_index,
data_conv_error_atp_index,
bk_expr,
ek_expr,
bk_excluded_expr,
ek_excluded_expr,
// static exclude flags (if exprs are NULL)
bkey_excluded,
ekey_excluded);
if (unique_key_expr)
*unique_key_expr = NULL;
}
else
{
if (keyInfo)
*keyInfo = NULL;
*unique_key_expr = (ex_expr_lean*)bk_expr;
*uniqueKeyLen = keyLen;
}
}
} // end of non-mdam case
else // Mdam case
{
// the work cri desc is used to build key values (entry 2) and
// to compute the exclusion flag (entry 3) to monitor for data
// conversion errors (entry 4) and to compute values on a column
// basis (entry 5 - Mdam only, and entry 6 - Mdam only, and only
// for MDAM_BETWEEN predtype)
ex_cri_desc * work_cri_desc
= new(space) ex_cri_desc(7, space);
// compute the format of the key buffer -- We need this
// so that Mdam will know, for each column, where in the buffer
// to move a value and how many bytes that value takes. The
// next few lines of code result in this information being stored
// in the attrs array.
// Some words on the technique: We create expressions whose
// result datatype matches the key buffer datatypes for each key
// column. Then we use the datatypes of these expressions to
// compute buffer format. The expressions themselves are not
// used any further; they do not result in compiled expressions
// in the plan. At run time we use string moves to move key
// values instead.
const CollIndex keyCount = listOfKeyColumns.entries();
CollIndex i;
// assert at least one column
GenAssert(keyCount > 0,"MDAM: at least one key column required.");
Attributes ** attrs = new(generator->wHeap()) Attributes * [keyCount];
for (i = 0; i < keyCount; i++)
{
ItemExpr * col_node =
listOfKeyColumns[i].getItemExpr();
ItemExpr *enode = col_node;
if ((tf == ExpTupleDesc::SQLMX_KEY_FORMAT) &&
(enode->getValueId().getType().getVarLenHdrSize() > 0))
{
// varchar keys in SQL/MP tables are converted to
// fixed length chars in key buffers
const CharType& char_type =
(CharType&)(enode->getValueId().getType());
if (!CollationInfo::isSystemCollation(char_type.getCollation()))
{
enode = new(generator->wHeap())
Cast(enode,
(new (generator->wHeap())
SQLChar( CharLenInfo(char_type.getStrCharLimit(),
char_type.getDataStorageSize()),
char_type.supportsSQLnull(),
FALSE, FALSE, FALSE,
char_type.getCharSet(),
char_type.getCollation(),
char_type.getCoercibility())));
}
}
NABoolean desc_flag;
if (keyColumns.isAscending(i))
desc_flag = reverseScan;
else
desc_flag = !reverseScan;
#pragma nowarn(1506) // warning elimination
enode = new(generator->wHeap()) CompEncode(enode,desc_flag);
#pragma warn(1506) // warning elimination
enode->bindNode(generator->getBindWA());
attrs[i] =
(generator->
addMapInfoToThis(keyBufferPartMapTable, enode->getValueId(), 0))->getAttr();
} // for, over keyCount
// Compute offsets, lengths, etc. and assign them to the right
// atp and atp index
processAttributes((ULng32)keyCount,
attrs, tf,
keyLen,
work_atp, key_atp_index);
// Now we have key column offsets and lengths stored in attrs.
// Next, for each column, generate expressions to compute hi,
// lo, non-null hi and non-null lo values, and create
// MdamColumnGen structures.
// Notes: In the Mdam network itself, all key values are
// encoded. Hence, we generate CompEncode nodes in all of the
// expressions, regardless of tuple format. In the Simulator
// case, we must at run-time decode the encoded values when
// moving them to the key buffer. $$$ We need an expression to
// do this. This decoding work has not yet been done, so the
// simulator only works correctly for columns that happen to be
// correctly aligned and whose encoding function does not change
// the value. $$$
MdamColumnGen * first = 0;
MdamColumnGen * last = 0;
LIST(NAType *) keyTypeList(generator->wHeap());//to keep the type of the keys for later
for (i = 0; i < keyCount; i++)
{
// generate expressions to compute hi, lo, non-null hi, non-null lo
NAType * targetType = (keyColumns[i]->getType())->newCopy(generator->wHeap());
// Genesis case 10-971031-9814 fix: desc_flag must take into account
// both the ASC/DESC attribute of the key column and the reverseScan
// attribute. Before this fix, it only took into account the first of
// these.
NABoolean desc_flag;
if (keyColumns.isAscending(i))
desc_flag = reverseScan;
else
desc_flag = !reverseScan;
// End Genesis case 10-971031-9814 fix.
if ((tf == ExpTupleDesc::SQLMX_KEY_FORMAT) &&
(targetType->getVarLenHdrSize() > 0))
{
// 5/9/98: add support for VARNCHAR
const CharType* char_type = (CharType*)(targetType);
if (!CollationInfo::isSystemCollation(char_type->getCollation()))
{
targetType = new(generator->wHeap())
SQLChar( CharLenInfo(char_type->getStrCharLimit(),
char_type->getDataStorageSize()),
char_type -> supportsSQLnull(),
FALSE, FALSE, FALSE,
char_type -> getCharSet(),
char_type -> getCollation(),
char_type -> getCoercibility());
/*
targetType->getNominalSize(),
targetType->supportsSQLnull()
*/
}
}
keyTypeList.insert(targetType); // save in ith position for later
// don't need to make copy of targetType in next call
ItemExpr * lo = new(generator->wHeap()) ConstValue(targetType,
!desc_flag,
TRUE /* allow NULL */);
#pragma nowarn(1506) // warning elimination
lo = new(generator->wHeap()) CompEncode(lo,desc_flag);
#pragma warn(1506) // warning elimination
lo->bindNode(generator->getBindWA());
ValueIdList loList;
loList.insert(lo->getValueId());
ex_expr *loExpr = 0;
ULng32 dataLen = 0;
generateContiguousMoveExpr(loList,
0, // don't add convert nodes
work_atp,
key_column_atp_index,
tf,
dataLen,
&loExpr);
ItemExpr * hi = new(generator->wHeap()) ConstValue(targetType->newCopy(generator->wHeap()),
desc_flag,
TRUE /* allow NULL */);
#pragma nowarn(1506) // warning elimination
hi = new(generator->wHeap()) CompEncode(hi,desc_flag);
#pragma warn(1506) // warning elimination
hi->bindNode(generator->getBindWA());
ValueIdList hiList;
hiList.insert(hi->getValueId());
ex_expr *hiExpr = 0;
generateContiguousMoveExpr(hiList,
0, // don't add convert nodes
work_atp,
key_column_atp_index,
tf,
dataLen,
&hiExpr);
ex_expr *nonNullLoExpr = loExpr;
ex_expr *nonNullHiExpr = hiExpr;
if (targetType->supportsSQLnull())
{
if (desc_flag)
{
ItemExpr * nonNullLo = new(generator->wHeap())
ConstValue(targetType->newCopy(generator->wHeap()),
!desc_flag,
FALSE /* don't allow NULL */);
#pragma nowarn(1506) // warning elimination
nonNullLo = new(generator->wHeap()) CompEncode(nonNullLo,desc_flag);
#pragma warn(1506) // warning elimination
nonNullLo->bindNode(generator->getBindWA());
ValueIdList nonNullLoList;
nonNullLoList.insert(nonNullLo->getValueId());
nonNullLoExpr = 0; // so we will get an expression back
generateContiguousMoveExpr(nonNullLoList,
0, // don't add convert nodes
work_atp,
key_column_atp_index,
tf,
dataLen,
&nonNullLoExpr);
}
else
{
ItemExpr * nonNullHi = new(generator->wHeap())
ConstValue(targetType->newCopy(generator->wHeap()),
desc_flag,
FALSE /* don't allow NULL */);
#pragma nowarn(1506) // warning elimination
nonNullHi = new(generator->wHeap()) CompEncode(nonNullHi,desc_flag);
#pragma warn(1506) // warning elimination
nonNullHi->bindNode(generator->getBindWA());
ValueIdList nonNullHiList;
nonNullHiList.insert(nonNullHi->getValueId());
nonNullHiExpr = 0; // so we will get an expression back
generateContiguousMoveExpr(nonNullHiList,
0, // don't add convert nodes
work_atp,
key_column_atp_index,
tf,
dataLen,
&nonNullHiExpr);
}
}
NABoolean useSparseProbes = mdamKeyPtr->isColumnSparse(i);
// calculate offset to the beginning of the column value
// (including the null indicator and the varchar length
// indicator if present)
ULng32 column_offset = attrs[i]->getOffset();
if (attrs[i]->getNullFlag())
column_offset = attrs[i]->getNullIndOffset();
else if (attrs[i]->getVCIndicatorLength() > 0)
column_offset = attrs[i]->getVCLenIndOffset();
last = new(space) MdamColumnGen(last,
dataLen,
column_offset,
useSparseProbes,
loExpr,
hiExpr,
nonNullLoExpr,
nonNullHiExpr);
if (first == 0)
first = last;
} // for over keyCount
// generate MdamPred's and attach to MdamColumnGen's
const ColumnOrderListPtrArray &columnOrderListPtrArray =
mdamKeyPtr->getColumnOrderListPtrArray();
#ifdef _DEBUG
// Debug print stataments below depend on this
// variable:
char *ev = getenv("MDAM_PRINT");
const NABoolean mdamPrintOn = (ev != NULL AND strcmp(ev,"ON")==0);
#endif
#ifdef _DEBUG
if (mdamPrintOn)
{
fprintf(stdout, "\n\n***Generating the MDAM key for table with index"
" columns: ");
listOfKeyColumns.display();
}
#endif
for (CollIndex n = 0; n < columnOrderListPtrArray.entries(); n++)
{
// get the list of key predicates associated with the n disjunct:
const ColumnOrderList &columnOrderList = *columnOrderListPtrArray[n];
#ifdef _DEBUG
if (mdamPrintOn)
{
fprintf(stdout,"\nDisjunct[%d]:----------------\n",n);
columnOrderList.print();
}
#endif
MdamColumnGen * cc = first;
CMPASSERT(keyCount == columnOrderList.entries());
const ValueIdSet *predsPtr = NULL;
for (i = 0; i < keyCount; i++)
{
#ifdef _DEBUG
if (mdamPrintOn)
{
fprintf(stdout, "Column(%d) using: ", i);
if ( mdamKeyPtr->isColumnSparse(i) )
fprintf(stdout,"SPARSE probes\n");
else
fprintf(stdout, "DENSE probes\n");
}
#endif
// get predicates for column order i:
predsPtr = columnOrderList[i];
NAType * keyType = keyTypeList[i];
NABoolean descending;
if (keyColumns.isAscending(i))
descending = reverseScan;
else
descending = !reverseScan;
ValueId keyColumn = listOfKeyColumns[i];
MdamCodeGenHelper mdamHelper(
n,
keyType,
descending,
work_atp,
key_column_atp_index,
tf,
dataConversionErrorFlag,
keyColumn);
MdamPred * lastPred = cc->getLastPred();
if (predsPtr != NULL)
{
for (ValueId predId = predsPtr->init();
predsPtr->next(predId); predsPtr->advance(predId))
{
MdamPred * head = 0; // head of generated MdamPred's
MdamPred * tail = 0;
ItemExpr * orGroup = predId.getItemExpr();
orGroup->mdamPredGen(generator,&head,&tail,mdamHelper,NULL);
if (lastPred)
{
if ( CmpCommon::getDefault(RANGESPEC_TRANSFORMATION) == DF_ON )
{
MdamPred* curr = lastPred;
while(curr->getNext() != NULL)
curr=curr->getNext();
curr->setNext(head);
}
else
lastPred->setNext(head);
}
cc->setLastPred(tail);
lastPred = tail; //@ZXmdam if 1st pred has head != tail, head is lost
} // for over preds
} // if (predsPtr != NULL)
cc = cc->getNext();
} // for every order...
} // for every column order list in the array (of disjuncts)
// build the Mdam key info
if (keytag > 0)
keyLen += sizeof(short);
*keyInfo = new(space) keyMdamGen(keyLen,
work_cri_desc,
key_atp_index,
exclude_flag_atp_index,
data_conv_error_atp_index,
key_column_atp_index,
first,
last,
reverseScan,
generator->wHeap());
} // end of mdam case
if (*keyInfo)
(*keyInfo)->setKeytag(keytag);
// reset map table to forget about the key object's work Atp
// aside: this logic is more bloody than it should be because the
// map table implementation doesn't accurately reflect the map table
// abstraction
generator->removeAll(keyBufferPartMapTable); // deletes anything that might have been
// added after keyBufferPartMapTable (at
// this writing we don't expect there to
// be anything, but we want to be safe)
// at this point keyBufferPartMapTable should be the last map table in the
// global map table chain
generator->removeLast(); // unlinks keyBufferPartMapTable and deletes it
return 0;
};
