* nodeAgg.c
* Routines to handle aggregate nodes.
*
- * Copyright (c) 1994, Regents of the University of California
+ * ExecAgg evaluates each aggregate in the following steps: (initcond1,
+ * initcond2 are the initial values and sfunc1, sfunc2, and finalfunc are
+ * the transition functions.)
+ *
+ * value1 = initcond1
+ * value2 = initcond2
+ * foreach input_value do
+ * value1 = sfunc1(value1, input_value)
+ * value2 = sfunc2(value2)
+ * value1 = finalfunc(value1, value2)
+ *
+ * If initcond1 is NULL then the first non-NULL input_value is
+ * assigned directly to value1. sfunc1 isn't applied until value1
+ * is non-NULL.
+ *
+ * sfunc1 is never applied when the current tuple's input_value is NULL.
+ * sfunc2 is applied for each tuple if the aggref is marked 'usenulls',
+ * otherwise it is only applied when input_value is not NULL.
+ * (usenulls was formerly used for COUNT(*), but is no longer needed for
+ * that purpose; as of 10/1999 the support for usenulls is dead code.
+ * I have not removed it because it seems like a potentially useful
+ * feature for user-defined aggregates. We'd just need to add a
+ * flag column to pg_aggregate and a parameter to CREATE AGGREGATE...)
*
*
- * NOTE
- * The implementation of Agg node has been reworked to handle legal
- * SQL aggregates. (Do not expect POSTQUEL semantics.) -- ay 2/95
+ * Copyright (c) 1994, Regents of the University of California
*
* IDENTIFICATION
- * $Header: /cvsroot/pgsql/src/backend/executor/nodeAgg.c,v 1.59 1999/10/30 02:35:14 tgl Exp $
+ * $Header: /cvsroot/pgsql/src/backend/executor/nodeAgg.c,v 1.60 1999/12/13 01:26:52 tgl Exp $
*
*-------------------------------------------------------------------------
*/
#include "access/heapam.h"
#include "catalog/pg_aggregate.h"
+#include "catalog/pg_operator.h"
#include "executor/executor.h"
#include "executor/nodeAgg.h"
#include "optimizer/clauses.h"
+#include "parser/parse_expr.h"
+#include "parser/parse_oper.h"
#include "parser/parse_type.h"
#include "utils/syscache.h"
+#include "utils/tuplesort.h"
/*
* AggStatePerAggData - per-aggregate working state for the Agg scan
* thereafter:
*/
+ /* Link to Aggref node this working state is for */
+ Aggref *aggref;
+
/* Oids of transfer functions */
Oid xfn1_oid;
Oid xfn2_oid;
FmgrInfo xfn2;
FmgrInfo finalfn;
/*
+ * Type of input data and Oid of sort operator to use for it;
+ * only set/used when aggregate has DISTINCT flag. (These are not
+ * used directly by nodeAgg, but must be passed to the Tuplesort object.)
+ */
+ Oid inputType;
+ Oid sortOperator;
+ /*
+ * fmgr lookup data for input type's equality operator --- only set/used
+ * when aggregate has DISTINCT flag.
+ */
+ FmgrInfo equalfn;
+ /*
* initial values from pg_aggregate entry
*/
Datum initValue1; /* for transtype1 */
bool initValue1IsNull,
initValue2IsNull;
/*
- * We need the len and byval info for the agg's transition status types
- * in order to know how to copy/delete values.
+ * We need the len and byval info for the agg's input and transition
+ * data types in order to know how to copy/delete values.
*/
- int transtype1Len,
+ int inputtypeLen,
+ transtype1Len,
transtype2Len;
- bool transtype1ByVal,
+ bool inputtypeByVal,
+ transtype1ByVal,
transtype2ByVal;
/*
* These values are working state that is initialized at the start
- * of an input tuple group and updated for each input tuple:
+ * of an input tuple group and updated for each input tuple.
+ *
+ * For a simple (non DISTINCT) aggregate, we just feed the input values
+ * straight to the transition functions. If it's DISTINCT, we pass the
+ * input values into a Tuplesort object; then at completion of the input
+ * tuple group, we scan the sorted values, eliminate duplicates, and run
+ * the transition functions on the rest.
*/
+ Tuplesortstate *sortstate; /* sort object, if a DISTINCT agg */
+
Datum value1, /* current transfer values 1 and 2 */
value2;
bool value1IsNull,
} AggStatePerAggData;
+static void initialize_aggregate (AggStatePerAgg peraggstate);
+static void advance_transition_functions (AggStatePerAgg peraggstate,
+ Datum newVal, bool isNull);
+static void finalize_aggregate (AggStatePerAgg peraggstate,
+ Datum *resultVal, bool *resultIsNull);
+static Datum copyDatum(Datum val, int typLen, bool typByVal);
+
+
/*
- * Helper routine to make a copy of a Datum.
- *
- * NB: input had better not be a NULL; might cause null-pointer dereference.
+ * Initialize one aggregate for a new set of input values.
*/
-static Datum
-copyDatum(Datum val, int typLen, bool typByVal)
+static void
+initialize_aggregate (AggStatePerAgg peraggstate)
{
- if (typByVal)
- return val;
+ Aggref *aggref = peraggstate->aggref;
+
+ /*
+ * Start a fresh sort operation for each DISTINCT aggregate.
+ */
+ if (aggref->aggdistinct)
+ {
+ /* In case of rescan, maybe there could be an uncompleted
+ * sort operation? Clean it up if so.
+ */
+ if (peraggstate->sortstate)
+ tuplesort_end(peraggstate->sortstate);
+
+ peraggstate->sortstate =
+ tuplesort_begin_datum(peraggstate->inputType,
+ peraggstate->sortOperator,
+ false);
+ }
+
+ /*
+ * (Re)set value1 and value2 to their initial values.
+ */
+ if (OidIsValid(peraggstate->xfn1_oid) &&
+ ! peraggstate->initValue1IsNull)
+ peraggstate->value1 = copyDatum(peraggstate->initValue1,
+ peraggstate->transtype1Len,
+ peraggstate->transtype1ByVal);
+ else
+ peraggstate->value1 = (Datum) NULL;
+ peraggstate->value1IsNull = peraggstate->initValue1IsNull;
+
+ if (OidIsValid(peraggstate->xfn2_oid) &&
+ ! peraggstate->initValue2IsNull)
+ peraggstate->value2 = copyDatum(peraggstate->initValue2,
+ peraggstate->transtype2Len,
+ peraggstate->transtype2ByVal);
else
+ peraggstate->value2 = (Datum) NULL;
+ peraggstate->value2IsNull = peraggstate->initValue2IsNull;
+
+ /* ------------------------------------------
+ * If the initial value for the first transition function
+ * doesn't exist in the pg_aggregate table then we will let
+ * the first value returned from the outer procNode become
+ * the initial value. (This is useful for aggregates like
+ * max{} and min{}.) The noInitValue flag signals that we
+ * still need to do this.
+ * ------------------------------------------
+ */
+ peraggstate->noInitValue = peraggstate->initValue1IsNull;
+}
+
+/*
+ * Given a new input value, advance the transition functions of an aggregate.
+ *
+ * Note: if the agg does not have usenulls set, null inputs will be filtered
+ * out before reaching here.
+ */
+static void
+advance_transition_functions (AggStatePerAgg peraggstate,
+ Datum newVal, bool isNull)
+{
+ Datum args[2];
+
+ if (OidIsValid(peraggstate->xfn1_oid) && !isNull)
{
- char *newVal;
+ if (peraggstate->noInitValue)
+ {
+ /*
+ * value1 has not been initialized. This is the first non-NULL
+ * input value. We use it as the initial value for value1.
+ *
+ * XXX We assume, without having checked, that the agg's input
+ * type is binary-compatible with its transtype1!
+ *
+ * We have to copy the datum since the tuple from which it came
+ * will be freed on the next iteration of the scan.
+ */
+ peraggstate->value1 = copyDatum(newVal,
+ peraggstate->transtype1Len,
+ peraggstate->transtype1ByVal);
+ peraggstate->value1IsNull = false;
+ peraggstate->noInitValue = false;
+ }
+ else
+ {
+ /* apply transition function 1 */
+ args[0] = peraggstate->value1;
+ args[1] = newVal;
+ newVal = (Datum) fmgr_c(&peraggstate->xfn1,
+ (FmgrValues *) args,
+ &isNull);
+ if (! peraggstate->transtype1ByVal)
+ pfree(peraggstate->value1);
+ peraggstate->value1 = newVal;
+ }
+ }
- if (typLen == -1) /* variable length type? */
- typLen = VARSIZE((struct varlena *) DatumGetPointer(val));
- newVal = (char *) palloc(typLen);
- memcpy(newVal, DatumGetPointer(val), typLen);
- return PointerGetDatum(newVal);
+ if (OidIsValid(peraggstate->xfn2_oid))
+ {
+ /* apply transition function 2 */
+ args[0] = peraggstate->value2;
+ isNull = false; /* value2 cannot be null, currently */
+ newVal = (Datum) fmgr_c(&peraggstate->xfn2,
+ (FmgrValues *) args,
+ &isNull);
+ if (! peraggstate->transtype2ByVal)
+ pfree(peraggstate->value2);
+ peraggstate->value2 = newVal;
}
}
+/*
+ * Compute the final value of one aggregate.
+ */
+static void
+finalize_aggregate (AggStatePerAgg peraggstate,
+ Datum *resultVal, bool *resultIsNull)
+{
+ Aggref *aggref = peraggstate->aggref;
+ char *args[2];
+
+ /*
+ * If it's a DISTINCT aggregate, all we've done so far is to stuff the
+ * input values into the sort object. Complete the sort, then run
+ * the transition functions on the non-duplicate values. Note that
+ * DISTINCT always suppresses nulls, per SQL spec, regardless of usenulls.
+ */
+ if (aggref->aggdistinct)
+ {
+ Datum oldVal = (Datum) 0;
+ bool haveOldVal = false;
+ Datum newVal;
+ bool isNull;
+
+ tuplesort_performsort(peraggstate->sortstate);
+ while (tuplesort_getdatum(peraggstate->sortstate, true,
+ &newVal, &isNull))
+ {
+ if (isNull)
+ continue;
+ if (haveOldVal)
+ {
+ Datum equal;
+
+ equal = (Datum) (*fmgr_faddr(&peraggstate->equalfn)) (oldVal,
+ newVal);
+ if (DatumGetInt32(equal) != 0)
+ {
+ if (! peraggstate->inputtypeByVal)
+ pfree(DatumGetPointer(newVal));
+ continue;
+ }
+ }
+ advance_transition_functions(peraggstate, newVal, false);
+ if (haveOldVal && ! peraggstate->inputtypeByVal)
+ pfree(DatumGetPointer(oldVal));
+ oldVal = newVal;
+ haveOldVal = true;
+ }
+ if (haveOldVal && ! peraggstate->inputtypeByVal)
+ pfree(DatumGetPointer(oldVal));
+ tuplesort_end(peraggstate->sortstate);
+ peraggstate->sortstate = NULL;
+ }
+
+ /*
+ * Now apply the agg's finalfn, or substitute the appropriate transition
+ * value if there is no finalfn.
+ *
+ * XXX For now, only apply finalfn if we got at least one
+ * non-null input value. This prevents zero divide in AVG().
+ * If we had cleaner handling of null inputs/results in functions,
+ * we could probably take out this hack and define the result
+ * for no inputs as whatever finalfn returns for null input.
+ */
+ if (OidIsValid(peraggstate->finalfn_oid) &&
+ ! peraggstate->noInitValue)
+ {
+ if (peraggstate->finalfn.fn_nargs > 1)
+ {
+ args[0] = (char *) peraggstate->value1;
+ args[1] = (char *) peraggstate->value2;
+ }
+ else if (OidIsValid(peraggstate->xfn1_oid))
+ args[0] = (char *) peraggstate->value1;
+ else if (OidIsValid(peraggstate->xfn2_oid))
+ args[0] = (char *) peraggstate->value2;
+ else
+ elog(ERROR, "ExecAgg: no valid transition functions??");
+ *resultIsNull = false;
+ *resultVal = (Datum) fmgr_c(&peraggstate->finalfn,
+ (FmgrValues *) args,
+ resultIsNull);
+ }
+ else if (OidIsValid(peraggstate->xfn1_oid))
+ {
+ /* Return value1 */
+ *resultVal = peraggstate->value1;
+ *resultIsNull = peraggstate->value1IsNull;
+ /* prevent pfree below */
+ peraggstate->value1IsNull = true;
+ }
+ else if (OidIsValid(peraggstate->xfn2_oid))
+ {
+ /* Return value2 */
+ *resultVal = peraggstate->value2;
+ *resultIsNull = peraggstate->value2IsNull;
+ /* prevent pfree below */
+ peraggstate->value2IsNull = true;
+ }
+ else
+ elog(ERROR, "ExecAgg: no valid transition functions??");
+
+ /*
+ * Release any per-group working storage, unless we're passing
+ * it back as the result of the aggregate.
+ */
+ if (OidIsValid(peraggstate->xfn1_oid) &&
+ ! peraggstate->value1IsNull &&
+ ! peraggstate->transtype1ByVal)
+ pfree(peraggstate->value1);
+
+ if (OidIsValid(peraggstate->xfn2_oid) &&
+ ! peraggstate->value2IsNull &&
+ ! peraggstate->transtype2ByVal)
+ pfree(peraggstate->value2);
+}
/* ---------------------------------------
*
* the expression context to be used when ExecProject evaluates the
* result tuple.
*
- * ExecAgg evaluates each aggregate in the following steps: (initcond1,
- * initcond2 are the initial values and sfunc1, sfunc2, and finalfunc are
- * the transition functions.)
- *
- * value1 = initcond1
- * value2 = initcond2
- * foreach tuple do
- * value1 = sfunc1(value1, aggregated_value)
- * value2 = sfunc2(value2)
- * value1 = finalfunc(value1, value2)
- *
- * If initcond1 is NULL then the first non-NULL aggregated_value is
- * assigned directly to value1. sfunc1 isn't applied until value1
- * is non-NULL.
- *
- * sfunc1 is never applied when the current tuple's aggregated_value
- * is NULL. sfunc2 is applied for each tuple if the aggref is marked
- * 'usenulls', otherwise it is only applied when aggregated_value is
- * not NULL. (usenulls was formerly used for COUNT(*), but is no longer
- * needed for that purpose; as of 10/1999 the support for usenulls is
- * dead code. I have not removed it because it seems like a potentially
- * useful feature for user-defined aggregates. We'd just need to add a
- * flag column to pg_aggregate and a parameter to CREATE AGGREGATE...)
- *
* If the outer subplan is a Group node, ExecAgg returns as many tuples
* as there are groups.
*
TupleTableSlot *resultSlot;
HeapTuple inputTuple;
int aggno;
- List *alist;
bool isDone;
bool isNull;
/*
* Initialize working state for a new input tuple group
*/
- aggno = -1;
- foreach(alist, aggstate->aggs)
+ for (aggno = 0; aggno < aggstate->numaggs; aggno++)
{
- AggStatePerAgg peraggstate = &peragg[++aggno];
+ AggStatePerAgg peraggstate = &peragg[aggno];
- /*
- * (Re)set value1 and value2 to their initial values.
- */
- if (OidIsValid(peraggstate->xfn1_oid) &&
- ! peraggstate->initValue1IsNull)
- peraggstate->value1 = copyDatum(peraggstate->initValue1,
- peraggstate->transtype1Len,
- peraggstate->transtype1ByVal);
- else
- peraggstate->value1 = (Datum) NULL;
- peraggstate->value1IsNull = peraggstate->initValue1IsNull;
-
- if (OidIsValid(peraggstate->xfn2_oid) &&
- ! peraggstate->initValue2IsNull)
- peraggstate->value2 = copyDatum(peraggstate->initValue2,
- peraggstate->transtype2Len,
- peraggstate->transtype2ByVal);
- else
- peraggstate->value2 = (Datum) NULL;
- peraggstate->value2IsNull = peraggstate->initValue2IsNull;
-
- /* ------------------------------------------
- * If the initial value for the first transition function
- * doesn't exist in the pg_aggregate table then we will let
- * the first value returned from the outer procNode become
- * the initial value. (This is useful for aggregates like
- * max{} and min{}.) The noInitValue flag signals that we
- * still need to do this.
- * ------------------------------------------
- */
- peraggstate->noInitValue = peraggstate->initValue1IsNull;
+ initialize_aggregate(peraggstate);
}
inputTuple = NULL; /* no saved input tuple yet */
break;
econtext->ecxt_scantuple = outerslot;
- aggno = -1;
- foreach(alist, aggstate->aggs)
+ for (aggno = 0; aggno < aggstate->numaggs; aggno++)
{
- Aggref *aggref = (Aggref *) lfirst(alist);
- AggStatePerAgg peraggstate = &peragg[++aggno];
+ AggStatePerAgg peraggstate = &peragg[aggno];
+ Aggref *aggref = peraggstate->aggref;
Datum newVal;
- Datum args[2];
newVal = ExecEvalExpr(aggref->target, econtext,
&isNull, &isDone);
if (isNull && !aggref->usenulls)
continue; /* ignore this tuple for this agg */
- if (OidIsValid(peraggstate->xfn1_oid) && !isNull)
- {
- if (peraggstate->noInitValue)
- {
- /*
- * value1 has not been initialized. This is the
- * first non-NULL input value. We use it as the
- * initial value for value1. XXX We assume,
- * without having checked, that the agg's input type
- * is binary-compatible with its transtype1!
- *
- * We have to copy the datum since the tuple from
- * which it came will be freed on the next iteration
- * of the scan.
- */
- peraggstate->value1 = copyDatum(newVal,
- peraggstate->transtype1Len,
- peraggstate->transtype1ByVal);
- peraggstate->value1IsNull = false;
- peraggstate->noInitValue = false;
- }
- else
- {
- /* apply transition function 1 */
- args[0] = peraggstate->value1;
- args[1] = newVal;
- newVal = (Datum) fmgr_c(&peraggstate->xfn1,
- (FmgrValues *) args,
- &isNull);
- if (! peraggstate->transtype1ByVal)
- pfree(peraggstate->value1);
- peraggstate->value1 = newVal;
- }
- }
-
- if (OidIsValid(peraggstate->xfn2_oid))
- {
- /* apply transition function 2 */
- args[0] = peraggstate->value2;
- isNull = false; /* value2 cannot be null, currently */
- newVal = (Datum) fmgr_c(&peraggstate->xfn2,
- (FmgrValues *) args,
- &isNull);
- if (! peraggstate->transtype2ByVal)
- pfree(peraggstate->value2);
- peraggstate->value2 = newVal;
- }
+ if (aggref->aggdistinct)
+ tuplesort_putdatum(peraggstate->sortstate,
+ newVal, isNull);
+ else
+ advance_transition_functions(peraggstate,
+ newVal, isNull);
}
/*
* Done scanning input tuple group.
* Finalize each aggregate calculation.
*/
- aggno = -1;
- foreach(alist, aggstate->aggs)
+ for (aggno = 0; aggno < aggstate->numaggs; aggno++)
{
- AggStatePerAgg peraggstate = &peragg[++aggno];
- char *args[2];
-
- /*
- * XXX For now, only apply finalfn if we got at least one
- * non-null input value. This prevents zero divide in AVG().
- * If we had cleaner handling of null inputs/results in functions,
- * we could probably take out this hack and define the result
- * for no inputs as whatever finalfn returns for null input.
- */
- if (OidIsValid(peraggstate->finalfn_oid) &&
- ! peraggstate->noInitValue)
- {
- if (peraggstate->finalfn.fn_nargs > 1)
- {
- args[0] = (char *) peraggstate->value1;
- args[1] = (char *) peraggstate->value2;
- }
- else if (OidIsValid(peraggstate->xfn1_oid))
- args[0] = (char *) peraggstate->value1;
- else if (OidIsValid(peraggstate->xfn2_oid))
- args[0] = (char *) peraggstate->value2;
- else
- elog(ERROR, "ExecAgg: no valid transition functions??");
- aggnulls[aggno] = false;
- aggvalues[aggno] = (Datum) fmgr_c(&peraggstate->finalfn,
- (FmgrValues *) args,
- &(aggnulls[aggno]));
- }
- else if (OidIsValid(peraggstate->xfn1_oid))
- {
- /* Return value1 */
- aggvalues[aggno] = peraggstate->value1;
- aggnulls[aggno] = peraggstate->value1IsNull;
- /* prevent pfree below */
- peraggstate->value1IsNull = true;
- }
- else if (OidIsValid(peraggstate->xfn2_oid))
- {
- /* Return value2 */
- aggvalues[aggno] = peraggstate->value2;
- aggnulls[aggno] = peraggstate->value2IsNull;
- /* prevent pfree below */
- peraggstate->value2IsNull = true;
- }
- else
- elog(ERROR, "ExecAgg: no valid transition functions??");
-
- /*
- * Release any per-group working storage, unless we're passing
- * it back as the result of the aggregate.
- */
- if (OidIsValid(peraggstate->xfn1_oid) &&
- ! peraggstate->value1IsNull &&
- ! peraggstate->transtype1ByVal)
- pfree(peraggstate->value1);
+ AggStatePerAgg peraggstate = &peragg[aggno];
- if (OidIsValid(peraggstate->xfn2_oid) &&
- ! peraggstate->value2IsNull &&
- ! peraggstate->transtype2ByVal)
- pfree(peraggstate->value2);
+ finalize_aggregate(peraggstate,
+ & aggvalues[aggno], & aggnulls[aggno]);
}
/*
/*
* Form a projection tuple using the aggregate results and the
- * representative input tuple. Store it in the result tuple slot,
- * and return it if it meets my qual condition.
+ * representative input tuple. Store it in the result tuple slot.
*/
resultSlot = ExecProject(projInfo, &isDone);
/*
* If the completed tuple does not match the qualifications,
* it is ignored and we loop back to try to process another group.
+ * Otherwise, return the tuple.
*/
}
while (! ExecQual(node->plan.qual, econtext));
/*
* find aggregates in targetlist and quals
+ *
+ * Note: pull_agg_clauses also checks that no aggs contain other agg
+ * calls in their arguments. This would make no sense under SQL semantics
+ * anyway (and it's forbidden by the spec). Because that is true, we
+ * don't need to worry about evaluating the aggs in any particular order.
*/
aggstate->aggs = nconc(pull_agg_clause((Node *) node->plan.targetlist),
pull_agg_clause((Node *) node->plan.qual));
/* Mark Aggref node with its associated index in the result array */
aggref->aggno = aggno;
+ /* Fill in the peraggstate data */
+ peraggstate->aggref = aggref;
+
aggTuple = SearchSysCacheTuple(AGGNAME,
PointerGetDatum(aggname),
ObjectIdGetDatum(aggref->basetype),
{
fmgr_info(finalfn_oid, &peraggstate->finalfn);
}
+
+ if (aggref->aggdistinct)
+ {
+ Oid inputType = exprType(aggref->target);
+ Operator eq_operator;
+ Form_pg_operator pgopform;
+
+ peraggstate->inputType = inputType;
+ typeInfo = typeidType(inputType);
+ peraggstate->inputtypeLen = typeLen(typeInfo);
+ peraggstate->inputtypeByVal = typeByVal(typeInfo);
+
+ eq_operator = oper("=", inputType, inputType, true);
+ if (!HeapTupleIsValid(eq_operator))
+ {
+ elog(ERROR, "Unable to identify an equality operator for type '%s'",
+ typeidTypeName(inputType));
+ }
+ pgopform = (Form_pg_operator) GETSTRUCT(eq_operator);
+ fmgr_info(pgopform->oprcode, &(peraggstate->equalfn));
+ peraggstate->sortOperator = any_ordering_op(inputType);
+ peraggstate->sortstate = NULL;
+ }
}
return TRUE;
ExecReScan(((Plan *) node)->lefttree, exprCtxt, (Plan *) node);
}
+
+
+/*
+ * Helper routine to make a copy of a Datum.
+ *
+ * NB: input had better not be a NULL; might cause null-pointer dereference.
+ */
+static Datum
+copyDatum(Datum val, int typLen, bool typByVal)
+{
+ if (typByVal)
+ return val;
+ else
+ {
+ char *newVal;
+
+ if (typLen == -1) /* variable length type? */
+ typLen = VARSIZE((struct varlena *) DatumGetPointer(val));
+ newVal = (char *) palloc(typLen);
+ memcpy(newVal, DatumGetPointer(val), typLen);
+ return PointerGetDatum(newVal);
+ }
+}
* tuplesort.c
* Generalized tuple sorting routines.
*
- * This module handles sorting of either heap tuples or index tuples
- * (and could fairly easily support other kinds of sortable objects,
+ * This module handles sorting of heap tuples, index tuples, or single
+ * Datums (and could easily support other kinds of sortable objects,
* if necessary). It works efficiently for both small and large amounts
* of data. Small amounts are sorted in-memory using qsort(). Large
* amounts are sorted using temporary files and a standard external sort
* Copyright (c) 1994, Regents of the University of California
*
* IDENTIFICATION
- * $Header: /cvsroot/pgsql/src/backend/utils/sort/tuplesort.c,v 1.2 1999/10/30 17:27:15 tgl Exp $
+ * $Header: /cvsroot/pgsql/src/backend/utils/sort/tuplesort.c,v 1.3 1999/12/13 01:27:04 tgl Exp $
*
*-------------------------------------------------------------------------
*/
#include "access/heapam.h"
#include "access/nbtree.h"
#include "miscadmin.h"
+#include "parser/parse_type.h"
#include "utils/logtape.h"
+#include "utils/lsyscache.h"
#include "utils/tuplesort.h"
/*
*/
Relation indexRel;
bool enforceUnique; /* complain if we find duplicate tuples */
+
+ /*
+ * These variables are specific to the Datum case; they are set
+ * by tuplesort_begin_datum and used only by the DatumTuple routines.
+ */
+ Oid datumType;
+ Oid sortOperator;
+ FmgrInfo sortOpFn; /* cached lookup data for sortOperator */
+ /* we need typelen and byval in order to know how to copy the Datums. */
+ int datumTypeLen;
+ bool datumTypeByVal;
};
#define COMPARETUP(state,a,b) ((*(state)->comparetup) (state, a, b))
*--------------------
*/
+/*
+ * For sorting single Datums, we build "pseudo tuples" that just carry
+ * the datum's value and null flag. For pass-by-reference data types,
+ * the actual data value appears after the DatumTupleHeader (MAXALIGNed,
+ * of course), and the value field in the header is just a pointer to it.
+ */
+
+typedef struct
+{
+ Datum val;
+ bool isNull;
+} DatumTuple;
+
+
static Tuplesortstate *tuplesort_begin_common(bool randomAccess);
+static void puttuple_common(Tuplesortstate *state, void *tuple);
static void inittapes(Tuplesortstate *state);
static void selectnewtape(Tuplesortstate *state);
static void mergeruns(Tuplesortstate *state);
static void *readtup_index(Tuplesortstate *state, int tapenum,
unsigned int len);
static unsigned int tuplesize_index(Tuplesortstate *state, void *tup);
+static int comparetup_datum(Tuplesortstate *state,
+ const void *a, const void *b);
+static void *copytup_datum(Tuplesortstate *state, void *tup);
+static void writetup_datum(Tuplesortstate *state, int tapenum, void *tup);
+static void *readtup_datum(Tuplesortstate *state, int tapenum,
+ unsigned int len);
+static unsigned int tuplesize_datum(Tuplesortstate *state, void *tup);
/*
* Since qsort(3) will not pass any context info to qsort_comparetup(),
* have been supplied. After performsort, retrieve the tuples in sorted
* order by calling tuplesort_gettuple until it returns NULL. (If random
* access was requested, rescan, markpos, and restorepos can also be called.)
+ * For Datum sorts, putdatum/getdatum are used instead of puttuple/gettuple.
* Call tuplesort_end to terminate the operation and release memory/disk space.
*/
return state;
}
+Tuplesortstate *
+tuplesort_begin_datum(Oid datumType,
+ Oid sortOperator,
+ bool randomAccess)
+{
+ Tuplesortstate *state = tuplesort_begin_common(randomAccess);
+ Type typeInfo;
+
+ state->comparetup = comparetup_datum;
+ state->copytup = copytup_datum;
+ state->writetup = writetup_datum;
+ state->readtup = readtup_datum;
+ state->tuplesize = tuplesize_datum;
+
+ state->datumType = datumType;
+ state->sortOperator = sortOperator;
+ /* lookup the function that implements the sort operator */
+ fmgr_info(get_opcode(sortOperator), &state->sortOpFn);
+ /* lookup necessary attributes of the datum type */
+ typeInfo = typeidType(datumType);
+ state->datumTypeLen = typeLen(typeInfo);
+ state->datumTypeByVal = typeByVal(typeInfo);
+
+ return state;
+}
+
/*
* tuplesort_end
*
{
/*
* Copy the given tuple into memory we control, and decrease availMem.
+ * Then call the code shared with the Datum case.
*/
tuple = COPYTUP(state, tuple);
+ puttuple_common(state, tuple);
+}
+
+/*
+ * Accept one Datum while collecting input data for sort.
+ *
+ * If the Datum is pass-by-ref type, the value will be copied.
+ */
+void
+tuplesort_putdatum(Tuplesortstate *state, Datum val, bool isNull)
+{
+ DatumTuple *tuple;
+
+ /*
+ * Build pseudo-tuple carrying the datum, and decrease availMem.
+ */
+ if (isNull || state->datumTypeByVal)
+ {
+ USEMEM(state, sizeof(DatumTuple));
+ tuple = (DatumTuple *) palloc(sizeof(DatumTuple));
+ tuple->val = val;
+ tuple->isNull = isNull;
+ }
+ else
+ {
+ int datalen = state->datumTypeLen;
+ int tuplelen;
+ char *newVal;
+
+ if (datalen == -1) /* variable length type? */
+ datalen = VARSIZE((struct varlena *) DatumGetPointer(val));
+ tuplelen = datalen + MAXALIGN(sizeof(DatumTuple));
+ USEMEM(state, tuplelen);
+ newVal = (char *) palloc(tuplelen);
+ tuple = (DatumTuple *) newVal;
+ newVal += MAXALIGN(sizeof(DatumTuple));
+ memcpy(newVal, DatumGetPointer(val), datalen);
+ tuple->val = PointerGetDatum(newVal);
+ tuple->isNull = false;
+ }
+
+ puttuple_common(state, (void *) tuple);
+}
+
+/*
+ * Shared code for tuple and datum cases.
+ */
+static void
+puttuple_common(Tuplesortstate *state, void *tuple)
+{
switch (state->status)
{
case TSS_INITIAL:
}
/*
+ * Fetch the next Datum in either forward or back direction.
+ * Returns FALSE if no more datums.
+ *
+ * If the Datum is pass-by-ref type, the returned value is freshly palloc'd
+ * and is now owned by the caller.
+ */
+bool
+tuplesort_getdatum(Tuplesortstate *state, bool forward,
+ Datum *val, bool *isNull)
+{
+ DatumTuple *tuple;
+ bool should_free;
+
+ tuple = (DatumTuple *) tuplesort_gettuple(state, forward, &should_free);
+
+ if (tuple == NULL)
+ return false;
+
+ if (tuple->isNull || state->datumTypeByVal)
+ {
+ *val = tuple->val;
+ *isNull = tuple->isNull;
+ }
+ else
+ {
+ int datalen = state->datumTypeLen;
+ char *newVal;
+
+ if (datalen == -1) /* variable length type? */
+ datalen = VARSIZE((struct varlena *) DatumGetPointer(tuple->val));
+ newVal = (char *) palloc(datalen);
+ memcpy(newVal, DatumGetPointer(tuple->val), datalen);
+ *val = PointerGetDatum(newVal);
+ *isNull = false;
+ }
+
+ if (should_free)
+ pfree(tuple);
+
+ return true;
+}
+
+
+/*
* inittapes - initialize for tape sorting.
*
* This is called only if we have found we don't have room to sort in memory.
return tuplen;
}
+
+
+/*
+ * Routines specialized for DatumTuple case
+ */
+
+static int
+comparetup_datum(Tuplesortstate *state, const void *a, const void *b)
+{
+ DatumTuple *ltup = (DatumTuple *) a;
+ DatumTuple *rtup = (DatumTuple *) b;
+
+ if (ltup->isNull)
+ {
+ if (!rtup->isNull)
+ return 1; /* NULL sorts after non-NULL */
+ return 0;
+ }
+ else if (rtup->isNull)
+ return -1;
+ else
+ {
+ int result;
+
+ if (!(result = - (int) (*fmgr_faddr(&state->sortOpFn)) (ltup->val,
+ rtup->val)))
+ result = (int) (*fmgr_faddr(&state->sortOpFn)) (rtup->val,
+ ltup->val);
+ return result;
+ }
+}
+
+static void *
+copytup_datum(Tuplesortstate *state, void *tup)
+{
+ /* Not currently needed */
+ elog(ERROR, "copytup_datum() should not be called");
+ return NULL;
+}
+
+static void
+writetup_datum(Tuplesortstate *state, int tapenum, void *tup)
+{
+ DatumTuple *tuple = (DatumTuple *) tup;
+ unsigned int tuplen = tuplesize_datum(state, tup);
+ unsigned int writtenlen = tuplen + sizeof(unsigned int);
+
+ LogicalTapeWrite(state->tapeset, tapenum,
+ (void*) &writtenlen, sizeof(writtenlen));
+ LogicalTapeWrite(state->tapeset, tapenum,
+ (void*) tuple, tuplen);
+ if (state->randomAccess) /* need trailing length word? */
+ LogicalTapeWrite(state->tapeset, tapenum,
+ (void*) &writtenlen, sizeof(writtenlen));
+
+ FREEMEM(state, tuplen);
+ pfree(tuple);
+}
+
+static void *
+readtup_datum(Tuplesortstate *state, int tapenum, unsigned int len)
+{
+ unsigned int tuplen = len - sizeof(unsigned int);
+ DatumTuple *tuple = (DatumTuple *) palloc(tuplen);
+
+ USEMEM(state, tuplen);
+ if (LogicalTapeRead(state->tapeset, tapenum, (void *) tuple,
+ tuplen) != tuplen)
+ elog(ERROR, "tuplesort: unexpected end of data");
+ if (state->randomAccess) /* need trailing length word? */
+ if (LogicalTapeRead(state->tapeset, tapenum, (void *) &tuplen,
+ sizeof(tuplen)) != sizeof(tuplen))
+ elog(ERROR, "tuplesort: unexpected end of data");
+
+ if (!tuple->isNull && !state->datumTypeByVal)
+ tuple->val = PointerGetDatum(((char *) tuple) +
+ MAXALIGN(sizeof(DatumTuple)));
+ return (void *) tuple;
+}
+
+static unsigned int
+tuplesize_datum(Tuplesortstate *state, void *tup)
+{
+ DatumTuple *tuple = (DatumTuple *) tup;
+
+ if (tuple->isNull || state->datumTypeByVal)
+ {
+ return (unsigned int) sizeof(DatumTuple);
+ }
+ else
+ {
+ int datalen = state->datumTypeLen;
+ int tuplelen;
+
+ if (datalen == -1) /* variable length type? */
+ datalen = VARSIZE((struct varlena *) DatumGetPointer(tuple->val));
+ tuplelen = datalen + MAXALIGN(sizeof(DatumTuple));
+ return (unsigned int) tuplelen;
+ }
+}
OSDN Git Service
