Make pg_statistic and related code account more honestly for collations.

When we first put in collations support, we basically punted on teaching
pg_statistic, ANALYZE, and the planner selectivity functions about that.
They've just used DEFAULT_COLLATION_OID independently of the actual
collation of the data.  It's time to improve that, so:

* Add columns to pg_statistic that record the specific collation associated
with each statistics slot.

* Teach ANALYZE to use the column's actual collation when comparing values
for statistical purposes, and record this in the appropriate slot.  (Note
that type-specific typanalyze functions are now expected to fill
stats->stacoll with the appropriate collation, too.)

* Teach assorted selectivity functions to use the actual collation of
the stats they are looking at, instead of just assuming it's
DEFAULT_COLLATION_OID.

This should give noticeably better results in selectivity estimates for
columns with nondefault collations, at least for query clauses that use
that same collation (which would be the default behavior in most cases).
It's still true that comparisons with explicit COLLATE clauses different
from the stored data's collation won't be well-estimated, but that's no
worse than before.  Also, this patch does make the first step towards
doing better with that, which is that it's now theoretically possible to
collect stats for a collation other than the column's own collation.

Patch by me; thanks to Peter Eisentraut for review.

Discussion: https://postgr.es/m/14706.1544630227@sss.pgh.pa.us

doc/src/sgml/catalogs.sgml

@@ -6394,6 +6394,18 @@ SCRAM-SHA-256$<replaceable>&lt;iteration count&gt;</replaceable>:<replaceable>&l
       </entry>
      </row>
 
+     <row>
+      <entry><structfield>stacoll<replaceable>N</replaceable></structfield></entry>
+      <entry><type>oid</type></entry>
+      <entry><literal><link linkend="catalog-pg-collation"><structname>pg_collation</structname></link>.oid</literal></entry>
+      <entry>
+       The collation used to derive the statistics stored in the
+       <replaceable>N</replaceable>th <quote>slot</quote>.  For example, a
+       histogram slot for a collatable column would show the collation that
+       defines the sort order of the data.  Zero for noncollatable data.
+      </entry>
+     </row>
+
      <row>
       <entry><structfield>stanumbers<replaceable>N</replaceable></structfield></entry>
       <entry><type>float4[]</type></entry>

src/backend/commands/analyze.c

@@ -904,11 +904,22 @@ examine_attribute(Relation onerel, int attnum, Node *index_expr)
 	{
 		stats->attrtypid = exprType(index_expr);
 		stats->attrtypmod = exprTypmod(index_expr);
+
+		/*
+		 * If a collation has been specified for the index column, use that in
+		 * preference to anything else; but if not, fall back to whatever we
+		 * can get from the expression.
+		 */
+		if (OidIsValid(onerel->rd_indcollation[attnum - 1]))
+			stats->attrcollid = onerel->rd_indcollation[attnum - 1];
+		else
+			stats->attrcollid = exprCollation(index_expr);
 	}
 	else
 	{
 		stats->attrtypid = attr->atttypid;
 		stats->attrtypmod = attr->atttypmod;
+		stats->attrcollid = attr->attcollation;
 	}
 
 	typtuple = SearchSysCacheCopy1(TYPEOID,
@@ -1553,6 +1564,11 @@ update_attstats(Oid relid, bool inh, int natts, VacAttrStats **vacattrstats)
 		{
 			values[i++] = ObjectIdGetDatum(stats->staop[k]);	/* staopN */
 		}
+		i = Anum_pg_statistic_stacoll1 - 1;
+		for (k = 0; k < STATISTIC_NUM_SLOTS; k++)
+		{
+			values[i++] = ObjectIdGetDatum(stats->stacoll[k]);	/* stacollN */
+		}
 		i = Anum_pg_statistic_stanumbers1 - 1;
 		for (k = 0; k < STATISTIC_NUM_SLOTS; k++)
 		{
@@ -1993,9 +2009,8 @@ compute_distinct_stats(VacAttrStatsP stats,
 		firstcount1 = track_cnt;
 		for (j = 0; j < track_cnt; j++)
 		{
-			/* We always use the default collation for statistics */
 			if (DatumGetBool(FunctionCall2Coll(&f_cmpeq,
-											   DEFAULT_COLLATION_OID,
+											   stats->attrcollid,
 											   value, track[j].value)))
 			{
 				match = true;
@@ -2202,6 +2217,7 @@ compute_distinct_stats(VacAttrStatsP stats,
 
 			stats->stakind[0] = STATISTIC_KIND_MCV;
 			stats->staop[0] = mystats->eqopr;
+			stats->stacoll[0] = stats->attrcollid;
 			stats->stanumbers[0] = mcv_freqs;
 			stats->numnumbers[0] = num_mcv;
 			stats->stavalues[0] = mcv_values;
@@ -2273,8 +2289,7 @@ compute_scalar_stats(VacAttrStatsP stats,
 
 	memset(&ssup, 0, sizeof(ssup));
 	ssup.ssup_cxt = CurrentMemoryContext;
-	/* We always use the default collation for statistics */
-	ssup.ssup_collation = DEFAULT_COLLATION_OID;
+	ssup.ssup_collation = stats->attrcollid;
 	ssup.ssup_nulls_first = false;
 
 	/*
@@ -2567,6 +2582,7 @@ compute_scalar_stats(VacAttrStatsP stats,
 
 			stats->stakind[slot_idx] = STATISTIC_KIND_MCV;
 			stats->staop[slot_idx] = mystats->eqopr;
+			stats->stacoll[slot_idx] = stats->attrcollid;
 			stats->stanumbers[slot_idx] = mcv_freqs;
 			stats->numnumbers[slot_idx] = num_mcv;
 			stats->stavalues[slot_idx] = mcv_values;
@@ -2682,6 +2698,7 @@ compute_scalar_stats(VacAttrStatsP stats,
 
 			stats->stakind[slot_idx] = STATISTIC_KIND_HISTOGRAM;
 			stats->staop[slot_idx] = mystats->ltopr;
+			stats->stacoll[slot_idx] = stats->attrcollid;
 			stats->stavalues[slot_idx] = hist_values;
 			stats->numvalues[slot_idx] = num_hist;
 
@@ -2725,6 +2742,7 @@ compute_scalar_stats(VacAttrStatsP stats,
 
 			stats->stakind[slot_idx] = STATISTIC_KIND_CORRELATION;
 			stats->staop[slot_idx] = mystats->ltopr;
+			stats->stacoll[slot_idx] = stats->attrcollid;
 			stats->stanumbers[slot_idx] = corrs;
 			stats->numnumbers[slot_idx] = 1;
 			slot_idx++;

src/backend/statistics/dependencies.c

@@ -252,6 +252,9 @@ dependency_degree(int numrows, HeapTuple *rows, int k, AttrNumber *dependency,
 	 * (b) split the data into groups by first (k-1) columns
 	 *
 	 * (c) for each group count different values in the last column
+	 *
+	 * We use the column data types' default sort operators and collations;
+	 * perhaps at some point it'd be worth using column-specific collations?
 	 */
 
 	/* prepare the sort function for the first dimension, and SortItem array */
@@ -266,7 +269,7 @@ dependency_degree(int numrows, HeapTuple *rows, int k, AttrNumber *dependency,
 				 colstat->attrtypid);
 
 		/* prepare the sort function for this dimension */
-		multi_sort_add_dimension(mss, i, type->lt_opr);
+		multi_sort_add_dimension(mss, i, type->lt_opr, type->typcollation);
 
 		/* accumulate all the data for both columns into an array and sort it */
 		for (j = 0; j < numrows; j++)

src/backend/statistics/extended_stats.c

@@ -363,18 +363,18 @@ multi_sort_init(int ndims)
 }
 
 /*
- * Prepare sort support info using the given sort operator
+ * Prepare sort support info using the given sort operator and collation
  * at the position 'sortdim'
  */
 void
-multi_sort_add_dimension(MultiSortSupport mss, int sortdim, Oid oper)
+multi_sort_add_dimension(MultiSortSupport mss, int sortdim,
+						 Oid oper, Oid collation)
 {
 	SortSupport ssup = &mss->ssup[sortdim];
 
 	ssup->ssup_cxt = CurrentMemoryContext;
-	ssup->ssup_collation = DEFAULT_COLLATION_OID;
+	ssup->ssup_collation = collation;
 	ssup->ssup_nulls_first = false;
-	ssup->ssup_cxt = CurrentMemoryContext;
 
 	PrepareSortSupportFromOrderingOp(oper, ssup);
 }

src/backend/statistics/mvdistinct.c

@@ -454,6 +454,9 @@ ndistinct_for_combination(double totalrows, int numrows, HeapTuple *rows,
 	/*
 	 * For each dimension, set up sort-support and fill in the values from the
 	 * sample data.
+	 *
+	 * We use the column data types' default sort operators and collations;
+	 * perhaps at some point it'd be worth using column-specific collations?
 	 */
 	for (i = 0; i < k; i++)
 	{
@@ -466,7 +469,7 @@ ndistinct_for_combination(double totalrows, int numrows, HeapTuple *rows,
 				 colstat->attrtypid);
 
 		/* prepare the sort function for this dimension */
-		multi_sort_add_dimension(mss, i, type->lt_opr);
+		multi_sort_add_dimension(mss, i, type->lt_opr, type->typcollation);
 
 		/* accumulate all the data for this dimension into the arrays */
 		for (j = 0; j < numrows; j++)

src/backend/tsearch/ts_typanalyze.c

@@ -14,6 +14,7 @@
 #include "postgres.h"
 
 #include "access/hash.h"
+#include "catalog/pg_collation.h"
 #include "catalog/pg_operator.h"
 #include "commands/vacuum.h"
 #include "tsearch/ts_type.h"
@@ -415,6 +416,7 @@ compute_tsvector_stats(VacAttrStats *stats,
 
 			stats->stakind[0] = STATISTIC_KIND_MCELEM;
 			stats->staop[0] = TextEqualOperator;
+			stats->stacoll[0] = DEFAULT_COLLATION_OID;
 			stats->stanumbers[0] = mcelem_freqs;
 			/* See above comment about two extra frequency fields */
 			stats->numnumbers[0] = num_mcelem + 2;

src/backend/utils/adt/array_selfuncs.c

@@ -46,21 +46,21 @@ static Selectivity mcelem_array_selec(ArrayType *array,
 				   Datum *mcelem, int nmcelem,
 				   float4 *numbers, int nnumbers,
 				   float4 *hist, int nhist,
-				   Oid operator, FmgrInfo *cmpfunc);
+				   Oid operator);
 static Selectivity mcelem_array_contain_overlap_selec(Datum *mcelem, int nmcelem,
 								   float4 *numbers, int nnumbers,
 								   Datum *array_data, int nitems,
-								   Oid operator, FmgrInfo *cmpfunc);
+								   Oid operator, TypeCacheEntry *typentry);
 static Selectivity mcelem_array_contained_selec(Datum *mcelem, int nmcelem,
 							 float4 *numbers, int nnumbers,
 							 Datum *array_data, int nitems,
 							 float4 *hist, int nhist,
-							 Oid operator, FmgrInfo *cmpfunc);
+							 Oid operator, TypeCacheEntry *typentry);
 static float *calc_hist(const float4 *hist, int nhist, int n);
 static float *calc_distr(const float *p, int n, int m, float rest);
 static int	floor_log2(uint32 n);
 static bool find_next_mcelem(Datum *mcelem, int nmcelem, Datum value,
-				 int *index, FmgrInfo *cmpfunc);
+				 int *index, TypeCacheEntry *typentry);
 static int	element_compare(const void *key1, const void *key2, void *arg);
 static int	float_compare_desc(const void *key1, const void *key2);
 
@@ -166,7 +166,7 @@ scalararraysel_containment(PlannerInfo *root,
 														   sslot.nnumbers,
 														   &constval, 1,
 														   OID_ARRAY_CONTAINS_OP,
-														   cmpfunc);
+														   typentry);
 			else
 				selec = mcelem_array_contained_selec(sslot.values,
 													 sslot.nvalues,
@@ -176,7 +176,7 @@ scalararraysel_containment(PlannerInfo *root,
 													 hslot.numbers,
 													 hslot.nnumbers,
 													 OID_ARRAY_CONTAINED_OP,
-													 cmpfunc);
+													 typentry);
 
 			free_attstatsslot(&hslot);
 			free_attstatsslot(&sslot);
@@ -189,14 +189,14 @@ scalararraysel_containment(PlannerInfo *root,
 														   NULL, 0,
 														   &constval, 1,
 														   OID_ARRAY_CONTAINS_OP,
-														   cmpfunc);
+														   typentry);
 			else
 				selec = mcelem_array_contained_selec(NULL, 0,
 													 NULL, 0,
 													 &constval, 1,
 													 NULL, 0,
 													 OID_ARRAY_CONTAINED_OP,
-													 cmpfunc);
+													 typentry);
 		}
 
 		/*
@@ -212,14 +212,14 @@ scalararraysel_containment(PlannerInfo *root,
 													   NULL, 0,
 													   &constval, 1,
 													   OID_ARRAY_CONTAINS_OP,
-													   cmpfunc);
+													   typentry);
 		else
 			selec = mcelem_array_contained_selec(NULL, 0,
 												 NULL, 0,
 												 &constval, 1,
 												 NULL, 0,
 												 OID_ARRAY_CONTAINED_OP,
-												 cmpfunc);
+												 typentry);
 		/* we assume no nulls here, so no stanullfrac correction */
 	}
 
@@ -385,7 +385,7 @@ calc_arraycontsel(VariableStatData *vardata, Datum constval,
 									   sslot.values, sslot.nvalues,
 									   sslot.numbers, sslot.nnumbers,
 									   hslot.numbers, hslot.nnumbers,
-									   operator, cmpfunc);
+									   operator);
 
 			free_attstatsslot(&hslot);
 			free_attstatsslot(&sslot);
@@ -395,7 +395,7 @@ calc_arraycontsel(VariableStatData *vardata, Datum constval,
 			/* No most-common-elements info, so do without */
 			selec = mcelem_array_selec(array, typentry,
 									   NULL, 0, NULL, 0, NULL, 0,
-									   operator, cmpfunc);
+									   operator);
 		}
 
 		/*
@@ -408,7 +408,7 @@ calc_arraycontsel(VariableStatData *vardata, Datum constval,
 		/* No stats at all, so do without */
 		selec = mcelem_array_selec(array, typentry,
 								   NULL, 0, NULL, 0, NULL, 0,
-								   operator, cmpfunc);
+								   operator);
 		/* we assume no nulls here, so no stanullfrac correction */
 	}
 
@@ -431,7 +431,7 @@ mcelem_array_selec(ArrayType *array, TypeCacheEntry *typentry,
 				   Datum *mcelem, int nmcelem,
 				   float4 *numbers, int nnumbers,
 				   float4 *hist, int nhist,
-				   Oid operator, FmgrInfo *cmpfunc)
+				   Oid operator)
 {
 	Selectivity selec;
 	int			num_elems;
@@ -476,20 +476,20 @@ mcelem_array_selec(ArrayType *array, TypeCacheEntry *typentry,
 
 	/* Sort extracted elements using their default comparison function. */
 	qsort_arg(elem_values, nonnull_nitems, sizeof(Datum),
-			  element_compare, cmpfunc);
+			  element_compare, typentry);
 
 	/* Separate cases according to operator */
 	if (operator == OID_ARRAY_CONTAINS_OP || operator == OID_ARRAY_OVERLAP_OP)
 		selec = mcelem_array_contain_overlap_selec(mcelem, nmcelem,
 												   numbers, nnumbers,
 												   elem_values, nonnull_nitems,
-												   operator, cmpfunc);
+												   operator, typentry);
 	else if (operator == OID_ARRAY_CONTAINED_OP)
 		selec = mcelem_array_contained_selec(mcelem, nmcelem,
 											 numbers, nnumbers,
 											 elem_values, nonnull_nitems,
 											 hist, nhist,
-											 operator, cmpfunc);
+											 operator, typentry);
 	else
 	{
 		elog(ERROR, "arraycontsel called for unrecognized operator %u",
@@ -523,7 +523,7 @@ static Selectivity
 mcelem_array_contain_overlap_selec(Datum *mcelem, int nmcelem,
 								   float4 *numbers, int nnumbers,
 								   Datum *array_data, int nitems,
-								   Oid operator, FmgrInfo *cmpfunc)
+								   Oid operator, TypeCacheEntry *typentry)
 {
 	Selectivity selec,
 				elem_selec;
@@ -586,14 +586,14 @@ mcelem_array_contain_overlap_selec(Datum *mcelem, int nmcelem,
 
 		/* Ignore any duplicates in the array data. */
 		if (i > 0 &&
-			element_compare(&array_data[i - 1], &array_data[i], cmpfunc) == 0)
+			element_compare(&array_data[i - 1], &array_data[i], typentry) == 0)
 			continue;
 
 		/* Find the smallest MCELEM >= this array item. */
 		if (use_bsearch)
 		{
 			match = find_next_mcelem(mcelem, nmcelem, array_data[i],
-									 &mcelem_index, cmpfunc);
+									 &mcelem_index, typentry);
 		}
 		else
 		{
@@ -601,7 +601,7 @@ mcelem_array_contain_overlap_selec(Datum *mcelem, int nmcelem,
 			{
 				int			cmp = element_compare(&mcelem[mcelem_index],
 												  &array_data[i],
-												  cmpfunc);
+												  typentry);
 
 				if (cmp < 0)
 					mcelem_index++;
@@ -699,7 +699,7 @@ mcelem_array_contained_selec(Datum *mcelem, int nmcelem,
 							 float4 *numbers, int nnumbers,
 							 Datum *array_data, int nitems,
 							 float4 *hist, int nhist,
-							 Oid operator, FmgrInfo *cmpfunc)
+							 Oid operator, TypeCacheEntry *typentry)
 {
 	int			mcelem_index,
 				i,
@@ -765,7 +765,7 @@ mcelem_array_contained_selec(Datum *mcelem, int nmcelem,
 
 		/* Ignore any duplicates in the array data. */
 		if (i > 0 &&
-			element_compare(&array_data[i - 1], &array_data[i], cmpfunc) == 0)
+			element_compare(&array_data[i - 1], &array_data[i], typentry) == 0)
 			continue;
 
 		/*
@@ -777,7 +777,7 @@ mcelem_array_contained_selec(Datum *mcelem, int nmcelem,
 		{
 			int			cmp = element_compare(&mcelem[mcelem_index],
 											  &array_data[i],
-											  cmpfunc);
+											  typentry);
 
 			if (cmp < 0)
 			{
@@ -1130,7 +1130,7 @@ floor_log2(uint32 n)
  */
 static bool
 find_next_mcelem(Datum *mcelem, int nmcelem, Datum value, int *index,
-				 FmgrInfo *cmpfunc)
+				 TypeCacheEntry *typentry)
 {
 	int			l = *index,
 				r = nmcelem - 1,
@@ -1140,7 +1140,7 @@ find_next_mcelem(Datum *mcelem, int nmcelem, Datum value, int *index,
 	while (l <= r)
 	{
 		i = (l + r) / 2;
-		res = element_compare(&mcelem[i], &value, cmpfunc);
+		res = element_compare(&mcelem[i], &value, typentry);
 		if (res == 0)
 		{
 			*index = i;
@@ -1158,7 +1158,7 @@ find_next_mcelem(Datum *mcelem, int nmcelem, Datum value, int *index,
 /*
  * Comparison function for elements.
  *
- * We use the element type's default btree opclass, and the default collation
+ * We use the element type's default btree opclass, and its default collation
  * if the type is collation-sensitive.
  *
  * XXX consider using SortSupport infrastructure
@@ -1168,10 +1168,11 @@ element_compare(const void *key1, const void *key2, void *arg)
 {
 	Datum		d1 = *((const Datum *) key1);
 	Datum		d2 = *((const Datum *) key2);
-	FmgrInfo   *cmpfunc = (FmgrInfo *) arg;
+	TypeCacheEntry *typentry = (TypeCacheEntry *) arg;
+	FmgrInfo   *cmpfunc = &typentry->cmp_proc_finfo;
 	Datum		c;
 
-	c = FunctionCall2Coll(cmpfunc, DEFAULT_COLLATION_OID, d1, d2);
+	c = FunctionCall2Coll(cmpfunc, typentry->typcollation, d1, d2);
 	return DatumGetInt32(c);
 }
 

src/backend/utils/adt/array_typanalyze.c

@@ -15,7 +15,6 @@
 #include "postgres.h"
 
 #include "access/tuptoaster.h"
-#include "catalog/pg_collation.h"
 #include "commands/vacuum.h"
 #include "utils/array.h"
 #include "utils/builtins.h"
@@ -39,6 +38,7 @@ typedef struct
 	/* Information about array element type */
 	Oid			type_id;		/* element type's OID */
 	Oid			eq_opr;			/* default equality operator's OID */
+	Oid			coll_id;		/* collation to use */
 	bool		typbyval;		/* physical properties of element type */
 	int16		typlen;
 	char		typalign;
@@ -135,6 +135,7 @@ array_typanalyze(PG_FUNCTION_ARGS)
 	extra_data = (ArrayAnalyzeExtraData *) palloc(sizeof(ArrayAnalyzeExtraData));
 	extra_data->type_id = typentry->type_id;
 	extra_data->eq_opr = typentry->eq_opr;
+	extra_data->coll_id = stats->attrcollid;	/* collation we should use */
 	extra_data->typbyval = typentry->typbyval;
 	extra_data->typlen = typentry->typlen;
 	extra_data->typalign = typentry->typalign;
@@ -560,6 +561,7 @@ compute_array_stats(VacAttrStats *stats, AnalyzeAttrFetchFunc fetchfunc,
 
 			stats->stakind[slot_idx] = STATISTIC_KIND_MCELEM;
 			stats->staop[slot_idx] = extra_data->eq_opr;
+			stats->stacoll[slot_idx] = extra_data->coll_id;
 			stats->stanumbers[slot_idx] = mcelem_freqs;
 			/* See above comment about extra stanumber entries */
 			stats->numnumbers[slot_idx] = num_mcelem + 3;
@@ -661,6 +663,7 @@ compute_array_stats(VacAttrStats *stats, AnalyzeAttrFetchFunc fetchfunc,
 
 			stats->stakind[slot_idx] = STATISTIC_KIND_DECHIST;
 			stats->staop[slot_idx] = extra_data->eq_opr;
+			stats->stacoll[slot_idx] = extra_data->coll_id;
 			stats->stanumbers[slot_idx] = hist;
 			stats->numnumbers[slot_idx] = num_hist + 1;
 			slot_idx++;
@@ -703,7 +706,7 @@ prune_element_hashtable(HTAB *elements_tab, int b_current)
 /*
  * Hash function for elements.
  *
- * We use the element type's default hash opclass, and the default collation
+ * We use the element type's default hash opclass, and the column collation
  * if the type is collation-sensitive.
  */
 static uint32
@@ -712,7 +715,9 @@ element_hash(const void *key, Size keysize)
 	Datum		d = *((const Datum *) key);
 	Datum		h;
 
-	h = FunctionCall1Coll(array_extra_data->hash, DEFAULT_COLLATION_OID, d);
+	h = FunctionCall1Coll(array_extra_data->hash,
+						  array_extra_data->coll_id,
+						  d);
 	return DatumGetUInt32(h);
 }
 
@@ -729,7 +734,7 @@ element_match(const void *key1, const void *key2, Size keysize)
 /*
  * Comparison function for elements.
  *
- * We use the element type's default btree opclass, and the default collation
+ * We use the element type's default btree opclass, and the column collation
  * if the type is collation-sensitive.
  *
  * XXX consider using SortSupport infrastructure
@@ -741,7 +746,9 @@ element_compare(const void *key1, const void *key2)
 	Datum		d2 = *((const Datum *) key2);
 	Datum		c;
 
-	c = FunctionCall2Coll(array_extra_data->cmp, DEFAULT_COLLATION_OID, d1, d2);
+	c = FunctionCall2Coll(array_extra_data->cmp,
+						  array_extra_data->coll_id,
+						  d1, d2);
 	return DatumGetInt32(c);
 }
 

src/backend/utils/adt/rangetypes_typanalyze.c

@@ -320,6 +320,7 @@ compute_range_stats(VacAttrStats *stats, AnalyzeAttrFetchFunc fetchfunc,
 			num_hist = 0;
 		}
 		stats->staop[slot_idx] = Float8LessOperator;
+		stats->stacoll[slot_idx] = InvalidOid;
 		stats->stavalues[slot_idx] = length_hist_values;
 		stats->numvalues[slot_idx] = num_hist;
 		stats->statypid[slot_idx] = FLOAT8OID;

src/backend/utils/adt/selfuncs.c

@@ -87,11 +87,12 @@
  * For both oprrest and oprjoin functions, the operator's input collation OID
  * (if any) is passed using the standard fmgr mechanism, so that the estimator
  * function can fetch it with PG_GET_COLLATION().  Note, however, that all
- * statistics in pg_statistic are currently built using the database's default
+ * statistics in pg_statistic are currently built using the relevant column's
  * collation.  Thus, in most cases where we are looking at statistics, we
- * should ignore the actual operator collation and use DEFAULT_COLLATION_OID.
+ * should ignore the operator collation and use the stats entry's collation.
  * We expect that the error induced by doing this is usually not large enough
- * to justify complicating matters.
+ * to justify complicating matters.  In any case, doing otherwise would yield
+ * entirely garbage results for ordered stats data such as histograms.
  *----------
  */
 
@@ -181,7 +182,8 @@ static double eqjoinsel_semi(Oid opfuncoid,
 			   RelOptInfo *inner_rel);
 static bool estimate_multivariate_ndistinct(PlannerInfo *root,
 								RelOptInfo *rel, List **varinfos, double *ndistinct);
-static bool convert_to_scalar(Datum value, Oid valuetypid, double *scaledvalue,
+static bool convert_to_scalar(Datum value, Oid valuetypid, Oid collid,
+				  double *scaledvalue,
 				  Datum lobound, Datum hibound, Oid boundstypid,
 				  double *scaledlobound, double *scaledhibound);
 static double convert_numeric_to_scalar(Datum value, Oid typid, bool *failure);
@@ -201,7 +203,8 @@ static double convert_one_string_to_scalar(char *value,
 							 int rangelo, int rangehi);
 static double convert_one_bytea_to_scalar(unsigned char *value, int valuelen,
 							int rangelo, int rangehi);
-static char *convert_string_datum(Datum value, Oid typid, bool *failure);
+static char *convert_string_datum(Datum value, Oid typid, Oid collid,
+					 bool *failure);
 static double convert_timevalue_to_scalar(Datum value, Oid typid,
 							bool *failure);
 static void examine_simple_variable(PlannerInfo *root, Var *var,
@@ -370,12 +373,12 @@ var_eq_const(VariableStatData *vardata, Oid operator,
 				/* be careful to apply operator right way 'round */
 				if (varonleft)
 					match = DatumGetBool(FunctionCall2Coll(&eqproc,
-														   DEFAULT_COLLATION_OID,
+														   sslot.stacoll,
 														   sslot.values[i],
 														   constval));
 				else
 					match = DatumGetBool(FunctionCall2Coll(&eqproc,
-														   DEFAULT_COLLATION_OID,
+														   sslot.stacoll,
 														   constval,
 														   sslot.values[i]));
 				if (match)
@@ -666,11 +669,11 @@ mcv_selectivity(VariableStatData *vardata, FmgrInfo *opproc,
 		{
 			if (varonleft ?
 				DatumGetBool(FunctionCall2Coll(opproc,
-											   DEFAULT_COLLATION_OID,
+											   sslot.stacoll,
 											   sslot.values[i],
 											   constval)) :
 				DatumGetBool(FunctionCall2Coll(opproc,
-											   DEFAULT_COLLATION_OID,
+											   sslot.stacoll,
 											   constval,
 											   sslot.values[i])))
 				mcv_selec += sslot.numbers[i];
@@ -744,11 +747,11 @@ histogram_selectivity(VariableStatData *vardata, FmgrInfo *opproc,
 			{
 				if (varonleft ?
 					DatumGetBool(FunctionCall2Coll(opproc,
-												   DEFAULT_COLLATION_OID,
+												   sslot.stacoll,
 												   sslot.values[i],
 												   constval)) :
 					DatumGetBool(FunctionCall2Coll(opproc,
-												   DEFAULT_COLLATION_OID,
+												   sslot.stacoll,
 												   constval,
 												   sslot.values[i])))
 					nmatch++;
@@ -873,7 +876,7 @@ ineq_histogram_selectivity(PlannerInfo *root,
 														 &sslot.values[probe]);
 
 				ltcmp = DatumGetBool(FunctionCall2Coll(opproc,
-													   DEFAULT_COLLATION_OID,
+													   sslot.stacoll,
 													   sslot.values[probe],
 													   constval));
 				if (isgt)
@@ -958,7 +961,8 @@ ineq_histogram_selectivity(PlannerInfo *root,
 				 * values to a uniform comparison scale, and do a linear
 				 * interpolation within this bin.
 				 */
-				if (convert_to_scalar(constval, consttype, &val,
+				if (convert_to_scalar(constval, consttype, sslot.stacoll,
+									  &val,
 									  sslot.values[i - 1], sslot.values[i],
 									  vardata->vartype,
 									  &low, &high))
@@ -2499,7 +2503,7 @@ eqjoinsel_inner(Oid opfuncoid,
 				if (hasmatch2[j])
 					continue;
 				if (DatumGetBool(FunctionCall2Coll(&eqproc,
-												   DEFAULT_COLLATION_OID,
+												   sslot1->stacoll,
 												   sslot1->values[i],
 												   sslot2->values[j])))
 				{
@@ -2711,7 +2715,7 @@ eqjoinsel_semi(Oid opfuncoid,
 				if (hasmatch2[j])
 					continue;
 				if (DatumGetBool(FunctionCall2Coll(&eqproc,
-												   DEFAULT_COLLATION_OID,
+												   sslot1->stacoll,
 												   sslot1->values[i],
 												   sslot2->values[j])))
 				{
@@ -4066,7 +4070,7 @@ estimate_multivariate_ndistinct(PlannerInfo *root, RelOptInfo *rel,
  * converted to measurements expressed in seconds.
  */
 static bool
-convert_to_scalar(Datum value, Oid valuetypid, double *scaledvalue,
+convert_to_scalar(Datum value, Oid valuetypid, Oid collid, double *scaledvalue,
 				  Datum lobound, Datum hibound, Oid boundstypid,
 				  double *scaledlobound, double *scaledhibound)
 {
@@ -4131,11 +4135,11 @@ convert_to_scalar(Datum value, Oid valuetypid, double *scaledvalue,
 		case NAMEOID:
 			{
 				char	   *valstr = convert_string_datum(value, valuetypid,
-														  &failure);
+														  collid, &failure);
 				char	   *lostr = convert_string_datum(lobound, boundstypid,
-														 &failure);
+														 collid, &failure);
 				char	   *histr = convert_string_datum(hibound, boundstypid,
-														 &failure);
+														 collid, &failure);
 
 				/*
 				 * Bail out if any of the values is not of string type.  We
@@ -4404,7 +4408,7 @@ convert_one_string_to_scalar(char *value, int rangelo, int rangehi)
  * before continuing, so as to generate correct locale-specific results.
  */
 static char *
-convert_string_datum(Datum value, Oid typid, bool *failure)
+convert_string_datum(Datum value, Oid typid, Oid collid, bool *failure)
 {
 	char	   *val;
 
@@ -4432,7 +4436,7 @@ convert_string_datum(Datum value, Oid typid, bool *failure)
 			return NULL;
 	}
 
-	if (!lc_collate_is_c(DEFAULT_COLLATION_OID))
+	if (!lc_collate_is_c(collid))
 	{
 		char	   *xfrmstr;
 		size_t		xfrmlen;
@@ -5407,14 +5411,14 @@ get_variable_range(PlannerInfo *root, VariableStatData *vardata, Oid sortop,
 				continue;
 			}
 			if (DatumGetBool(FunctionCall2Coll(&opproc,
-											   DEFAULT_COLLATION_OID,
+											   sslot.stacoll,
 											   sslot.values[i], tmin)))
 			{
 				tmin = sslot.values[i];
 				tmin_is_mcv = true;
 			}
 			if (DatumGetBool(FunctionCall2Coll(&opproc,
-											   DEFAULT_COLLATION_OID,
+											   sslot.stacoll,
 											   tmax, sslot.values[i])))
 			{
 				tmax = sslot.values[i];
@@ -6014,6 +6018,7 @@ prefix_selectivity(PlannerInfo *root, VariableStatData *vardata,
 	Selectivity prefixsel;
 	Oid			cmpopr;
 	FmgrInfo	opproc;
+	AttStatsSlot sslot;
 	Const	   *greaterstrcon;
 	Selectivity eq_sel;
 
@@ -6036,16 +6041,23 @@ prefix_selectivity(PlannerInfo *root, VariableStatData *vardata,
 
 	/*-------
 	 * If we can create a string larger than the prefix, say
-	 *	"x < greaterstr".
+	 * "x < greaterstr".  We try to generate the string referencing the
+	 * collation of the var's statistics, but if that's not available,
+	 * use DEFAULT_COLLATION_OID.
 	 *-------
 	 */
+	if (HeapTupleIsValid(vardata->statsTuple) &&
+		get_attstatsslot(&sslot, vardata->statsTuple,
+						 STATISTIC_KIND_HISTOGRAM, InvalidOid, 0))
+		 /* sslot.stacoll is set up */ ;
+	else
+		sslot.stacoll = DEFAULT_COLLATION_OID;
 	cmpopr = get_opfamily_member(opfamily, vartype, vartype,
 								 BTLessStrategyNumber);
 	if (cmpopr == InvalidOid)
 		elog(ERROR, "no < operator for opfamily %u", opfamily);
 	fmgr_info(get_opcode(cmpopr), &opproc);
-	greaterstrcon = make_greater_string(prefixcon, &opproc,
-										DEFAULT_COLLATION_OID);
+	greaterstrcon = make_greater_string(prefixcon, &opproc, sslot.stacoll);
 	if (greaterstrcon)
 	{
 		Selectivity topsel;

src/backend/utils/cache/lsyscache.c

@@ -2881,6 +2881,7 @@ get_attavgwidth(Oid relid, AttrNumber attnum)
  *
  * If a matching slot is found, true is returned, and *sslot is filled thus:
  * staop: receives the actual STAOP value.
+ * stacoll: receives the actual STACOLL value.
  * valuetype: receives actual datatype of the elements of stavalues.
  * values: receives pointer to an array of the slot's stavalues.
  * nvalues: receives number of stavalues.
@@ -2893,6 +2894,10 @@ get_attavgwidth(Oid relid, AttrNumber attnum)
  *
  * If no matching slot is found, false is returned, and *sslot is zeroed.
  *
+ * Note that the current API doesn't allow for searching for a slot with
+ * a particular collation.  If we ever actually support recording more than
+ * one collation, we'll have to extend the API, but for now simple is good.
+ *
  * The data referred to by the fields of sslot is locally palloc'd and
  * is independent of the original pg_statistic tuple.  When the caller
  * is done with it, call free_attstatsslot to release the palloc'd data.
@@ -2927,6 +2932,20 @@ get_attstatsslot(AttStatsSlot *sslot, HeapTuple statstuple,
 		return false;			/* not there */
 
 	sslot->staop = (&stats->staop1)[i];
+	sslot->stacoll = (&stats->stacoll1)[i];
+
+	/*
+	 * XXX Hopefully-temporary hack: if stacoll isn't set, inject the default
+	 * collation.  This won't matter for non-collation-aware datatypes.  For
+	 * those that are, this covers cases where stacoll has not been set.  In
+	 * the short term we need this because some code paths involving type NAME
+	 * do not pass any collation to prefix_selectivity and related functions.
+	 * Even when that's been fixed, it's likely that some add-on typanalyze
+	 * functions won't get the word right away about filling stacoll during
+	 * ANALYZE, so we'll probably need this for awhile.
+	 */
+	if (sslot->stacoll == InvalidOid)
+		sslot->stacoll = DEFAULT_COLLATION_OID;
 
 	if (flags & ATTSTATSSLOT_VALUES)
 	{

src/backend/utils/cache/typcache.c

@@ -388,6 +388,7 @@ lookup_type_cache(Oid type_id, int flags)
 		typentry->typtype = typtup->typtype;
 		typentry->typrelid = typtup->typrelid;
 		typentry->typelem = typtup->typelem;
+		typentry->typcollation = typtup->typcollation;
 
 		/* If it's a domain, immediately thread it into the domain cache list */
 		if (typentry->typtype == TYPTYPE_DOMAIN)

src/include/catalog/catversion.h

@@ -53,6 +53,6 @@
  */
 
 /*							yyyymmddN */
-#define CATALOG_VERSION_NO	201812091
+#define CATALOG_VERSION_NO	201812141
 
 #endif

src/include/catalog/pg_statistic.h

@@ -74,12 +74,13 @@ CATALOG(pg_statistic,2619,StatisticRelationId)
 	 * statistical data can be placed.  Each slot includes:
 	 *		kind			integer code identifying kind of data (see below)
 	 *		op				OID of associated operator, if needed
+	 *		coll			OID of relevant collation, or 0 if none
 	 *		numbers			float4 array (for statistical values)
 	 *		values			anyarray (for representations of data values)
-	 * The ID and operator fields are never NULL; they are zeroes in an
-	 * unused slot.  The numbers and values fields are NULL in an unused
-	 * slot, and might also be NULL in a used slot if the slot kind has
-	 * no need for one or the other.
+	 * The ID, operator, and collation fields are never NULL; they are zeroes
+	 * in an unused slot.  The numbers and values fields are NULL in an
+	 * unused slot, and might also be NULL in a used slot if the slot kind
+	 * has no need for one or the other.
 	 * ----------------
 	 */
 
@@ -95,6 +96,12 @@ CATALOG(pg_statistic,2619,StatisticRelationId)
 	Oid			staop4;
 	Oid			staop5;
 
+	Oid			stacoll1;
+	Oid			stacoll2;
+	Oid			stacoll3;
+	Oid			stacoll4;
+	Oid			stacoll5;
+
 #ifdef CATALOG_VARLEN			/* variable-length fields start here */
 	float4		stanumbers1[1];
 	float4		stanumbers2[1];
@@ -159,7 +166,8 @@ typedef FormData_pg_statistic *Form_pg_statistic;
 
 /*
  * In a "most common values" slot, staop is the OID of the "=" operator
- * used to decide whether values are the same or not.  stavalues contains
+ * used to decide whether values are the same or not, and stacoll is the
+ * collation used (same as column's collation).  stavalues contains
  * the K most common non-null values appearing in the column, and stanumbers
  * contains their frequencies (fractions of total row count).  The values
  * shall be ordered in decreasing frequency.  Note that since the arrays are
@@ -171,9 +179,11 @@ typedef FormData_pg_statistic *Form_pg_statistic;
 
 /*
  * A "histogram" slot describes the distribution of scalar data.  staop is
- * the OID of the "<" operator that describes the sort ordering.  (In theory,
- * more than one histogram could appear, if a datatype has more than one
- * useful sort operator.)  stavalues contains M (>=2) non-null values that
+ * the OID of the "<" operator that describes the sort ordering, and stacoll
+ * is the relevant collation.  (In theory more than one histogram could appear,
+ * if a datatype has more than one useful sort operator or we care about more
+ * than one collation.  Currently the collation will always be that of the
+ * underlying column.)  stavalues contains M (>=2) non-null values that
  * divide the non-null column data values into M-1 bins of approximately equal
  * population.  The first stavalues item is the MIN and the last is the MAX.
  * stanumbers is not used and should be NULL.  IMPORTANT POINT: if an MCV
@@ -190,11 +200,12 @@ typedef FormData_pg_statistic *Form_pg_statistic;
 /*
  * A "correlation" slot describes the correlation between the physical order
  * of table tuples and the ordering of data values of this column, as seen
- * by the "<" operator identified by staop.  (As with the histogram, more
- * than one entry could theoretically appear.)	stavalues is not used and
- * should be NULL.  stanumbers contains a single entry, the correlation
- * coefficient between the sequence of data values and the sequence of
- * their actual tuple positions.  The coefficient ranges from +1 to -1.
+ * by the "<" operator identified by staop with the collation identified by
+ * stacoll.  (As with the histogram, more than one entry could theoretically
+ * appear.)  stavalues is not used and should be NULL.  stanumbers contains
+ * a single entry, the correlation coefficient between the sequence of data
+ * values and the sequence of their actual tuple positions.  The coefficient
+ * ranges from +1 to -1.
  */
 #define STATISTIC_KIND_CORRELATION	3
 
@@ -203,7 +214,8 @@ typedef FormData_pg_statistic *Form_pg_statistic;
  * except that it stores the most common non-null *elements* of the column
  * values.  This is useful when the column datatype is an array or some other
  * type with identifiable elements (for instance, tsvector).  staop contains
- * the equality operator appropriate to the element type.  stavalues contains
+ * the equality operator appropriate to the element type, and stacoll
+ * contains the collation to use with it.  stavalues contains
  * the most common element values, and stanumbers their frequencies.  Unlike
  * MCV slots, frequencies are measured as the fraction of non-null rows the
  * element value appears in, not the frequency of all rows.  Also unlike
@@ -226,7 +238,8 @@ typedef FormData_pg_statistic *Form_pg_statistic;
  * A "distinct elements count histogram" slot describes the distribution of
  * the number of distinct element values present in each row of an array-type
  * column.  Only non-null rows are considered, and only non-null elements.
- * staop contains the equality operator appropriate to the element type.
+ * staop contains the equality operator appropriate to the element type,
+ * and stacoll contains the collation to use with it.
  * stavalues is not used and should be NULL.  The last member of stanumbers is
  * the average count of distinct element values over all non-null rows.  The
  * preceding M (>=2) members form a histogram that divides the population of

src/include/commands/vacuum.h

@@ -52,9 +52,11 @@
  * careful to allocate any pointed-to data in anl_context, which will NOT
  * be CurrentMemoryContext when compute_stats is called.
  *
- * Note: for the moment, all comparisons done for statistical purposes
- * should use the database's default collation (DEFAULT_COLLATION_OID).
- * This might change in some future release.
+ * Note: all comparisons done for statistical purposes should use the
+ * underlying column's collation (attcollation), except in situations
+ * where a noncollatable container type contains a collatable type;
+ * in that case use the type's default collation.  Be sure to record
+ * the appropriate collation in stacoll.
  *----------
  */
 typedef struct VacAttrStats *VacAttrStatsP;
@@ -78,11 +80,13 @@ typedef struct VacAttrStats
 	 * because some index opclasses store a different type than the underlying
 	 * column/expression.  Instead use attrtypid, attrtypmod, and attrtype for
 	 * information about the datatype being fed to the typanalyze function.
+	 * Likewise, use attrcollid not attr->attcollation.
 	 */
 	Form_pg_attribute attr;		/* copy of pg_attribute row for column */
 	Oid			attrtypid;		/* type of data being analyzed */
 	int32		attrtypmod;		/* typmod of data being analyzed */
 	Form_pg_type attrtype;		/* copy of pg_type row for attrtypid */
+	Oid			attrcollid;		/* collation of data being analyzed */
 	MemoryContext anl_context;	/* where to save long-lived data */
 
 	/*
@@ -103,6 +107,7 @@ typedef struct VacAttrStats
 	float4		stadistinct;	/* # distinct values */
 	int16		stakind[STATISTIC_NUM_SLOTS];
 	Oid			staop[STATISTIC_NUM_SLOTS];
+	Oid			stacoll[STATISTIC_NUM_SLOTS];
 	int			numnumbers[STATISTIC_NUM_SLOTS];
 	float4	   *stanumbers[STATISTIC_NUM_SLOTS];
 	int			numvalues[STATISTIC_NUM_SLOTS];

src/include/statistics/extended_stats_internal.h

@@ -59,7 +59,7 @@ extern MVDependencies *statext_dependencies_deserialize(bytea *data);
 
 extern MultiSortSupport multi_sort_init(int ndims);
 extern void multi_sort_add_dimension(MultiSortSupport mss, int sortdim,
-						 Oid oper);
+						 Oid oper, Oid collation);
 extern int	multi_sort_compare(const void *a, const void *b, void *arg);
 extern int multi_sort_compare_dim(int dim, const SortItem *a,
 					   const SortItem *b, MultiSortSupport mss);

src/include/utils/lsyscache.h

@@ -44,6 +44,7 @@ typedef struct AttStatsSlot
 {
 	/* Always filled: */
 	Oid			staop;			/* Actual staop for the found slot */
+	Oid			stacoll;		/* Actual collation for the found slot */
 	/* Filled if ATTSTATSSLOT_VALUES is specified: */
 	Oid			valuetype;		/* Actual datatype of the values */
 	Datum	   *values;			/* slot's "values" array, or NULL if none */

src/include/utils/typcache.h

@@ -41,6 +41,7 @@ typedef struct TypeCacheEntry
 	char		typtype;
 	Oid			typrelid;
 	Oid			typelem;
+	Oid			typcollation;
 
 	/*
 	 * Information obtained from opfamily entries