src/backend/gpopt/translate/CContextDXLToPlStmt.cpp

//---------------------------------------------------------------------------
//	Greenplum Database
//	Copyright (C) 2011 Greenplum, Inc.
//
//	@filename:
//		CContextDXLToPlStmt.cpp
//
//	@doc:
//		Implementation of the functions that provide
//		access to CIdGenerators (needed to number initplans, motion
//		nodes as well as params), list of RangeTableEntires and Subplans
//		generated so far during DXL-->PlStmt translation.
//
//	@test:
//
//
//---------------------------------------------------------------------------

extern "C" {
#include "postgres.h"

#include "nodes/parsenodes.h"
#include "nodes/plannodes.h"
#include "utils/rel.h"
}

#include "gpos/base.h"

#include "gpopt/gpdbwrappers.h"
#include "gpopt/translate/CContextDXLToPlStmt.h"
#include "naucrates/exception.h"

using namespace gpdxl;

//---------------------------------------------------------------------------
//	@function:
//		CContextDXLToPlStmt::CContextDXLToPlStmt
//
//	@doc:
//		Ctor
//
//---------------------------------------------------------------------------
CContextDXLToPlStmt::CContextDXLToPlStmt(
	CMemoryPool *mp, CIdGenerator *plan_id_counter,
	CIdGenerator *motion_id_counter, CIdGenerator *param_id_counter,
	DistributionHashOpsKind distribution_hashops)
	: m_mp(mp),
	  m_plan_id_counter(plan_id_counter),
	  m_motion_id_counter(motion_id_counter),
	  m_param_id_counter(param_id_counter),
	  m_param_types_list(NIL),
	  m_distribution_hashops(distribution_hashops),
	  m_rtable_entries_list(nullptr),
	  m_subplan_entries_list(nullptr),
	  m_subplan_sliceids_list(nullptr),
	  m_slices_list(nullptr),
	  m_result_relation_index(0),
	  m_distribution_policy(nullptr),
	  m_part_selector_to_param_map(nullptr),
	  m_agg_infos(nullptr),
	  m_agg_trans_infos(nullptr)
{
	m_cte_producer_info = GPOS_NEW(m_mp) HMUlCTEProducerInfo(m_mp);
	m_part_selector_to_param_map = GPOS_NEW(m_mp) UlongToUlongMap(m_mp);
	m_used_rte_indexes = GPOS_NEW(m_mp) HMUlIndex(m_mp);
}

//---------------------------------------------------------------------------
//	@function:
//		CContextDXLToPlStmt::~CContextDXLToPlStmt
//
//	@doc:
//		Dtor
//
//---------------------------------------------------------------------------
CContextDXLToPlStmt::~CContextDXLToPlStmt()
{
	m_cte_producer_info->Release();
	m_part_selector_to_param_map->Release();
	m_used_rte_indexes->Release();
}

//---------------------------------------------------------------------------
//	@function:
//		CContextDXLToPlStmt::GetNextPlanId
//
//	@doc:
//		Get the next plan id
//
//---------------------------------------------------------------------------
ULONG
CContextDXLToPlStmt::GetNextPlanId()
{
	return m_plan_id_counter->next_id();
}

//---------------------------------------------------------------------------
//	@function:
//		CContextDXLToPlStmt::GetCurrentMotionId
//
//	@doc:
//		Get the current motion id
//
//---------------------------------------------------------------------------
ULONG
CContextDXLToPlStmt::GetCurrentMotionId()
{
	return m_motion_id_counter->current_id();
}

//---------------------------------------------------------------------------
//	@function:
//		CContextDXLToPlStmt::GetNextMotionId
//
//	@doc:
//		Get the next motion id
//
//---------------------------------------------------------------------------
ULONG
CContextDXLToPlStmt::GetNextMotionId()
{
	return m_motion_id_counter->next_id();
}

//---------------------------------------------------------------------------
//	@function:
//		CContextDXLToPlStmt::GetNextParamId
//
//	@doc:
//		Get the next param id, for a parameter of type 'typeoid'
//
//---------------------------------------------------------------------------
ULONG
CContextDXLToPlStmt::GetNextParamId(OID typeoid)
{
	m_param_types_list = gpdb::LAppendOid(m_param_types_list, typeoid);

	return m_param_id_counter->next_id();
}

//---------------------------------------------------------------------------
//	@function:
//		CContextDXLToPlStmt::GetParamTypes
//
//	@doc:
//		Get the current param types list
//
//---------------------------------------------------------------------------
List *
CContextDXLToPlStmt::GetParamTypes()
{
	return m_param_types_list;
}

//---------------------------------------------------------------------------
//	@function:
//		CContextDXLToPlStmt::RegisterCTEProducerPlan
//
//	@doc:
//		Register information about the CTE producer
//
//---------------------------------------------------------------------------
void
CContextDXLToPlStmt::RegisterCTEProducerInfo(
	ULONG cte_id, ULongPtrArray *producer_output_colidx_map,
	ShareInputScan *siscan)
{
	ULONG *key = GPOS_NEW(m_mp) ULONG(cte_id);
	BOOL result GPOS_ASSERTS_ONLY = m_cte_producer_info->Insert(
		key, GPOS_NEW(m_mp) SCTEEntryInfo(producer_output_colidx_map, siscan));

	GPOS_ASSERT(result);
}

//---------------------------------------------------------------------------
//	@function:
//		CContextDXLToPlStmt::GetCTEProducerInfo
//
//	@doc:
//		Return the producer in CTE
//---------------------------------------------------------------------------
std::pair<ULongPtrArray *, ShareInputScan *>
CContextDXLToPlStmt::GetCTEProducerInfo(ULONG cte_id) const
{
	SCTEEntryInfo *sctepinfo = m_cte_producer_info->Find(&cte_id);
	GPOS_ASSERT(sctepinfo);

	return std::make_pair(sctepinfo->m_pidxmap, sctepinfo->m_cte_producer_plan);
}

//---------------------------------------------------------------------------
//	@function:
//		CContextDXLToPlStmt::AddRTE
//
//	@doc:
//		Add a RangeTableEntries
//
//---------------------------------------------------------------------------
void
CContextDXLToPlStmt::AddRTE(RangeTblEntry *rte, BOOL is_result_relation)
{
	// add rte to rtable entries list
	m_rtable_entries_list = gpdb::LAppend(m_rtable_entries_list, rte);

	rte->inFromCl = true;

	if (is_result_relation)
	{
		GPOS_ASSERT(0 == m_result_relation_index &&
					"Only one result relation supported");
		rte->inFromCl = false;
		m_result_relation_index = gpdb::ListLength(m_rtable_entries_list);
	}
}

//---------------------------------------------------------------------------
//	@function:
//		CContextDXLToPlStmt::InsertUsedRTEIndexes
//
//	@doc:
//		Add assigned query id --> rte index key-value pair
//		to used rte indexes map
//
//---------------------------------------------------------------------------
void
CContextDXLToPlStmt::InsertUsedRTEIndexes(
	ULONG assigned_query_id_for_target_rel, Index index)
{
	// update used rte indexes map
	m_used_rte_indexes->Insert(GPOS_NEW(m_mp)
								   ULONG(assigned_query_id_for_target_rel),
							   GPOS_NEW(m_mp) Index(index));
}

//---------------------------------------------------------------------------
//	@function:
//		CContextDXLToPlStmt::GetSubplanSliceIdArray
//
//	@doc:
//		Get the slice IDs of each subplan as an array.
//
//---------------------------------------------------------------------------
int *
CContextDXLToPlStmt::GetSubplanSliceIdArray()
{
	int numSubplans = list_length(m_subplan_entries_list);
	int *sliceIdArray;
	ListCell *lc;
	int i;

	sliceIdArray = (int *) gpdb::GPDBAlloc(numSubplans * sizeof(int));

	i = 0;
	foreach (lc, m_subplan_sliceids_list)
	{
		sliceIdArray[i++] = lfirst_int(lc);
	}

	return sliceIdArray;
}

//---------------------------------------------------------------------------
//	@function:
//		CContextDXLToPlStmt::GetSlices
//
//	@doc:
//		Get the slice table as an array
//
//---------------------------------------------------------------------------
PlanSlice *
CContextDXLToPlStmt::GetSlices(int *numSlices_p)
{
	int numSlices = list_length(m_slices_list);
	PlanSlice *sliceArray;
	ListCell *lc;
	int i;

	sliceArray = (PlanSlice *) gpdb::GPDBAlloc(numSlices * sizeof(PlanSlice));

	i = 0;
	foreach (lc, m_slices_list)
	{
		PlanSlice *src = (PlanSlice *) lfirst(lc);

		memcpy(&sliceArray[i], src, sizeof(PlanSlice));

		i++;
	}

	m_current_slice = nullptr;
	gpdb::ListFreeDeep(m_slices_list);

	*numSlices_p = numSlices;
	return sliceArray;
}

//---------------------------------------------------------------------------
//	@function:
//		CContextDXLToPlStmt::AddSubplan
//
//	@doc:
//		Add a subplan
//
//---------------------------------------------------------------------------
void
CContextDXLToPlStmt::AddSubplan(Plan *plan)
{
	m_subplan_entries_list = gpdb::LAppend(m_subplan_entries_list, plan);
	m_subplan_sliceids_list =
		gpdb::LAppendInt(m_subplan_sliceids_list, m_current_slice->sliceIndex);
}

//---------------------------------------------------------------------------
//	@function:
//		CContextDXLToPlStmt::AddSlice
//
//	@doc:
//		Add a plan slice
//
//---------------------------------------------------------------------------
int
CContextDXLToPlStmt::AddSlice(PlanSlice *slice)
{
	slice->sliceIndex = list_length(m_slices_list);
	m_slices_list = gpdb::LAppend(m_slices_list, slice);

	return slice->sliceIndex;
}

//---------------------------------------------------------------------------
//	@function:
//		CContextDXLToPlStmt::AddCtasInfo
//
//	@doc:
//		Add CTAS info
//
//---------------------------------------------------------------------------
void
CContextDXLToPlStmt::AddCtasInfo(GpPolicy *distribution_policy)
{
	GPOS_ASSERT(nullptr != distribution_policy);

	m_distribution_policy = distribution_policy;
}

//---------------------------------------------------------------------------
//	@function:
//		CContextDXLToPlStmt::GetDistributionHashOpclassForType
//
//	@doc:
//		Return a hash operator class to use for computing
//		distribution key values for the given datatype.
//
//		This returns either the default opclass, or the legacy
//		opclass of the type, depending on what opclasses we have
//		seen being used in tables so far.
//---------------------------------------------------------------------------
Oid
CContextDXLToPlStmt::GetDistributionHashOpclassForType(Oid typid)
{
	Oid opclass = InvalidOid;

	switch (m_distribution_hashops)
	{
		case DistrUseDefaultHashOps:
			opclass = gpdb::GetDefaultDistributionOpclassForType(typid);
			break;

		case DistrUseLegacyHashOps:
			opclass = gpdb::GetLegacyCdbHashOpclassForBaseType(typid);
			break;

		case DistrHashOpsNotDeterminedYet:
			// None of the tables we have seen so far have been
			// hash distributed, so we haven't made up our mind
			// on which opclasses to use yet. But we have to
			// pick something now.
			//
			// FIXME: It's quite unoptimal that this ever happens.
			// To avoid this we should make a pass over the tree to
			// determine the opclasses, before translating
			// anything. But there is no convenient way to "walk"
			// the DXL representation AFAIK.
			//
			// Example query where this happens:
			// select * from dd_singlecol_1 t1,
			//               generate_series(1,10) g
			// where t1.a=g.g and t1.a=1 ;
			//
			// The ORCA plan consists of a join between
			// Result+FunctionScan and TableScan. The
			// Result+FunctionScan side is processed first, and
			// this gets called to generate a "hash filter" for
			// it Result. The TableScan is encountered and
			// added to the range table only later. If it uses
			// legacy ops, we have already decided to use default
			// ops here, and we fall back unnecessarily.
			//
			// On the other hand, when the opclass is not specified in the
			// distributed-by clause one should be decided according to the
			// gp_use_legacy_hashops setting.
			opclass = gpdb::GetColumnDefOpclassForType(NIL, typid);
			// update m_distribution_hashops accordingly
			if (opclass == gpdb::GetDefaultDistributionOpclassForType(typid))
			{
				m_distribution_hashops = DistrUseDefaultHashOps;
			}
			else if (opclass == gpdb::GetLegacyCdbHashOpclassForBaseType(typid))
			{
				m_distribution_hashops = DistrUseLegacyHashOps;
			}
			else
			{
				GPOS_RAISE(
					gpdxl::ExmaMD, gpdxl::ExmiMDObjUnsupported,
					GPOS_WSZ_LIT("Unsupported distribution hashops policy"));
			}
			break;
	}

	return opclass;
}

//---------------------------------------------------------------------------
//	@function:
//		CContextDXLToPlStmt::GetDistributionHashFuncForType
//
//	@doc:
//		Return a hash function to use for computing distribution key
//		values for the given datatype.
//
//		This returns the hash function either from the default
//		opclass, or the legacy opclass of the type, depending on
//		what opclasses we have seen being used in tables so far.
//---------------------------------------------------------------------------
Oid
CContextDXLToPlStmt::GetDistributionHashFuncForType(Oid typid)
{
	Oid opclass;
	Oid opfamily;
	Oid hashproc;

	opclass = GetDistributionHashOpclassForType(typid);

	if (opclass == InvalidOid)
	{
		GPOS_RAISE(gpdxl::ExmaMD, gpdxl::ExmiMDObjUnsupported,
				   GPOS_WSZ_LIT("no default hash opclasses found"));
	}

	opfamily = gpdb::GetOpclassFamily(opclass);
	hashproc = gpdb::GetHashProcInOpfamily(opfamily, typid);

	return hashproc;
}

ULONG
CContextDXLToPlStmt::GetParamIdForSelector(OID oid_type, ULONG selectorId)
{
	ULONG *param_id = m_part_selector_to_param_map->Find(&selectorId);
	if (nullptr == param_id)
	{
		param_id = GPOS_NEW(m_mp) ULONG(GetNextParamId(oid_type));
		ULONG *selector_id = GPOS_NEW(m_mp) ULONG(selectorId);
		m_part_selector_to_param_map->Insert(selector_id, param_id);
	}
	return *param_id;
}

Index
CContextDXLToPlStmt::FindRTE(Oid reloid)
{
	ListCell *lc;
	int idx = 0;

	ForEachWithCount(lc, m_rtable_entries_list, idx)
	{
		RangeTblEntry *rte = (RangeTblEntry *) lfirst(lc);
		if (rte->relid == reloid)
		{
			return idx + 1;
		}
	}
	return -1;
}

RangeTblEntry *
CContextDXLToPlStmt::GetRTEByIndex(Index index)
{
	return (RangeTblEntry *) gpdb::ListNth(m_rtable_entries_list,
										   int(index - 1));
}

//---------------------------------------------------------------------------
//	@function: of associated
//		CContextDXLToPlStmt::GetRTEIndexByAssignedQueryId
//
//	@doc:
//
//		For given assigned query id, this function returns the index of rte in
//		m_rtable_entries_list for further processing and sets is_rte_exists
//		flag that rte was processed.
//
//		assigned query id is a "tag" of a table descriptor. It marks the query
//		structure that contains the result relation. If two table descriptors
//		have the same assigned query id, these two table descriptors point to
//		the same result relation in `ModifyTable` operation. If the assigned
//		query id is positive, it indicates the query is an INSERT/UPDATE/DELETE
//		operation (`ModifyTable` operation). If the assigned query id zero
// 		(UNASSIGNED_QUERYID), usually it indicates the operation is not
// 		INSERT/UPDATE/DELETE, and doesn't have a result relation. Or sometimes,
// 		the operation is INSERT/UPDATE/DELETE, but the table descriptor tagged
//		with the assigned query id doesn't point to a result relation.
//
//		m_used_rte_indexes is a hash map. It maps the assigned query id (key)
//		to the rte index (value) as in m_rtable_entries_list. The hash map only
//		stores positive id's, because id 0 is reserved for operations that
//		don't have a result relation.
//
//		To look up the rte index, we may encounter three scenarios:
//
//		1. If assigned query id == 0, since the hash map only stores positive
//		id's, the rte index isn't stored in the hash map. In this case, we
//		return rtable entries list length+1. This means we will add a new rte
//		to the rte list.
//
//		2. If assigned query id > 0, it indicates the operation is
//		INSERT/UPDATE/DELETE. We look for its index in the hash map. If the
//		index is found, we return the index. This means we can reuse the rte
//		that's already in the list.
//
// 		3. If assigned query id > 0, but the index isn't found in the hash map,
// 		it means the id hasn't been processed. We return rtable entries list
// 		length+1. This means we will add a new rte to the rte list. At the same
// 		time, we will add the assigned query id --> list_length+1 key-value pair
// 		to the hash map for future look ups.
//
//		Here's an example
//		```sql
//		create table b (i int, j int);
//		create table c (i int);
//		insert into b(i,j) values (1,2), (2,3), (3,4);
//		insert into c(i) values (1), (2);
//		delete from b where i in (select i from c);
//		```
//
//		`b` is a result relation. Table descriptors for `b` will have the same
//		positive assigned query id. `c` is not a result relation. Table
//		descriptors for `c` will have zero assigned query id.
//---------------------------------------------------------------------------

Index
CContextDXLToPlStmt::GetRTEIndexByAssignedQueryId(
	ULONG assigned_query_id_for_target_rel, BOOL *is_rte_exists)
{
	*is_rte_exists = false;

	if (assigned_query_id_for_target_rel == UNASSIGNED_QUERYID)
	{
		return gpdb::ListLength(m_rtable_entries_list) + 1;
	}

	Index *usedIndex =
		m_used_rte_indexes->Find(&assigned_query_id_for_target_rel);

	//	`usedIndex` is a non NULL value in next case: table descriptor with
	//	the same `assigned_query_id_for_target_rel` was processed previously
	//	(so no need to create a new index for result relation like the relation
	//	itself)
	if (usedIndex)
	{
		*is_rte_exists = true;
		return *usedIndex;
	}

	//	`assigned_query_id_for_target_rel` of table descriptor which points to
	//	result relation wasn't previously processed - create a new index.
	return gpdb::ListLength(m_rtable_entries_list) + 1;
}

void
CContextDXLToPlStmt::AppendAggInfos(AggInfo *agginfo)
{
	m_agg_infos = gpdb::LAppend(m_agg_infos, agginfo);
}

void
CContextDXLToPlStmt::AppendAggTransInfos(AggTransInfo *transinfo)
{
	m_agg_trans_infos = gpdb::LAppend(m_agg_trans_infos, transinfo);
}

void
CContextDXLToPlStmt::ResetAggInfosAndTransInfos()
{
	ListCell *lc;
	foreach (lc, m_agg_infos)
	{
		AggInfo *agginfo = (AggInfo *) lfirst(lc);
		gpdb::GPDBFree(agginfo);
	}

	gpdb::ListFree(m_agg_infos);

	foreach (lc, m_agg_trans_infos)
	{
		AggTransInfo *aggtransinfo = (AggTransInfo *) lfirst(lc);
		gpdb::GPDBFree(aggtransinfo);
	}

	gpdb::ListFree(m_agg_trans_infos);

	m_agg_infos = nullptr;
	m_agg_trans_infos = nullptr;
}

// EOF