Add into extended statistics clause WITH and option 'method' which command

postgresql generate ndistinct combinations corresponding to a columns order
in the index.

It is needed to survive ANALYZE if we want to estimate multiple columns and
believe they will be used in a query according to definition of an existing
index.

Author: danolivo

src/backend/catalog/system_views.sql

@@ -283,6 +283,7 @@ CREATE VIEW pg_stats_ext WITH (security_barrier) AS
            ) AS attnames,
            pg_get_statisticsobjdef_expressions(s.oid) as exprs,
            s.stxkind AS kinds,
+		   s.options AS options,
            sd.stxdinherit AS inherited,
            sd.stxdndistinct AS n_distinct,
            sd.stxddependencies AS dependencies,

src/backend/commands/statscmds.c

@@ -55,6 +55,45 @@ compare_int16(const void *a, const void *b)
 	return (av - bv);
 }
 
+/*
+ * Check correctness of the options list and prepare the list to be stored
+ * in the statistics relation.
+ */
+static Datum
+transform_extstat_options(List *defList)
+{
+	ArrayBuildState	   *astate = NULL;
+	Datum				result;
+
+	foreach_ptr(DefElem, def, defList)
+	{
+		const char *value;
+		Size		len;
+		text	   *t;
+
+		if (strcmp(def->defname, "method") != 0 || def->arg == NULL)
+			elog(ERROR, "unrecognized option or value: %s", def->defname);
+
+		value = defGetString(def);
+
+		len = VARHDRSZ + strlen(def->defname) + 1 + strlen(value);
+		/* +1 leaves room for sprintf's trailing null */
+		t = (text *) palloc(len + 1);
+		SET_VARSIZE(t, len);
+		sprintf(VARDATA(t), "%s=%s", def->defname, value);
+
+		astate = accumArrayResult(astate, PointerGetDatum(t),
+								  false, TEXTOID,
+								  CurrentMemoryContext);
+	}
+
+	result = (astate != NULL) ?
+								makeArrayResult(astate, CurrentMemoryContext) :
+								(Datum) 0;
+
+	return result;
+}
+
 /*
  *		CREATE STATISTICS
  */
@@ -504,6 +543,10 @@ CreateStatistics(CreateStatsStmt *stmt)
 	if (exprsDatum == (Datum) 0)
 		nulls[Anum_pg_statistic_ext_stxexprs - 1] = true;
 
+	values[Anum_pg_statistic_ext_options-1] = transform_extstat_options(stmt->options);
+	if (stmt->options == NIL)
+		nulls[Anum_pg_statistic_ext_options - 1] = true;
+
 	/* insert it into pg_statistic_ext */
 	htup = heap_form_tuple(statrel->rd_att, values, nulls);
 	CatalogTupleInsert(statrel, htup);

src/backend/parser/gram.y

@@ -4645,7 +4645,7 @@ ExistingIndex:   USING INDEX name					{ $$ = $3; }
 
 CreateStatsStmt:
 			CREATE STATISTICS opt_qualified_name
-			opt_name_list ON stats_params FROM from_list
+			opt_name_list ON stats_params FROM from_list opt_reloptions
 				{
 					CreateStatsStmt *n = makeNode(CreateStatsStmt);
 
@@ -4655,10 +4655,11 @@ CreateStatsStmt:
 					n->relations = $8;
 					n->stxcomment = NULL;
 					n->if_not_exists = false;
+					n->options = $9;
 					$$ = (Node *) n;
 				}
 			| CREATE STATISTICS IF_P NOT EXISTS any_name
-			opt_name_list ON stats_params FROM from_list
+			opt_name_list ON stats_params FROM from_list opt_reloptions
 				{
 					CreateStatsStmt *n = makeNode(CreateStatsStmt);
 
@@ -4668,6 +4669,7 @@ CreateStatsStmt:
 					n->relations = $11;
 					n->stxcomment = NULL;
 					n->if_not_exists = true;
+					n->options = $12;
 					$$ = (Node *) n;
 				}
 			;

src/backend/parser/parse_utilcmd.c

@@ -2117,6 +2117,12 @@ generateClonedExtStatsStmt(RangeVar *heapRel, Oid heapRelid,
 		pfree(exprsString);
 	}
 
+	datum = SysCacheGetAttr(STATEXTOID, ht_stats,
+							Anum_pg_statistic_ext_options, &isnull);
+
+	if (isnull)
+		datum = (Datum) 0;
+
 	/* finally, build the output node */
 	stats = makeNode(CreateStatsStmt);
 	stats->defnames = NULL;
@@ -2126,6 +2132,7 @@ generateClonedExtStatsStmt(RangeVar *heapRel, Oid heapRelid,
 	stats->stxcomment = NULL;
 	stats->transformed = true;	/* don't need transformStatsStmt again */
 	stats->if_not_exists = false;
+	stats->options = untransformRelOptions(datum);
 
 	/* Clean up */
 	ReleaseSysCache(ht_stats);

src/backend/statistics/extended_stats.c

@@ -19,6 +19,7 @@
 #include "access/detoast.h"
 #include "access/genam.h"
 #include "access/htup_details.h"
+#include "access/reloptions.h"
 #include "access/table.h"
 #include "catalog/indexing.h"
 #include "catalog/pg_statistic_ext.h"
@@ -70,6 +71,7 @@ typedef struct StatExtEntry
 	List	   *types;			/* 'char' list of enabled statistics kinds */
 	int			stattarget;		/* statistics target (-1 for default) */
 	List	   *exprs;			/* expressions */
+	int			method;
 } StatExtEntry;
 
 
@@ -510,6 +512,25 @@ fetch_statentries_for_relation(Relation pg_statext, Oid relid)
 
 		entry->exprs = exprs;
 
+		datum = SysCacheGetAttr(STATEXTOID, htup,
+								Anum_pg_statistic_ext_options, &isnull);
+		if (!isnull)
+		{
+			List *defList = untransformRelOptions(datum);
+
+			foreach_ptr(DefElem, def, defList)
+			{
+				if (strcmp(def->defname, "method") != 0 || def->arg == NULL)
+					elog(ERROR, "unrecognized option or value: %s", def->defname);
+
+				if (strcmp(defGetString(def), "linear") == 0)
+					entry->method = EXTSTAT_METHOD_LINEAR;
+			}
+		}
+
+		if (entry->method == 0)
+			entry->method = EXTSTAT_METHOD_COMBS;
+
 		result = lappend(result, entry);
 	}
 
@@ -2560,6 +2581,8 @@ make_build_data(Relation rel, StatExtEntry *stat, int numrows, HeapTuple *rows,
 		k--;
 	}
 
+	result->method = stat->method;
+
 	/* first extract values for all the regular attributes */
 	for (i = 0; i < numrows; i++)
 	{

src/backend/statistics/mvdistinct.c

@@ -65,9 +65,10 @@ typedef struct CombinationGenerator
 	int			current;		/* index of the next combination to return */
 	int			ncombinations;	/* number of combinations (size of array) */
 	int		   *combinations;	/* array of pre-built combinations */
+	int			method;
 } CombinationGenerator;
 
-static CombinationGenerator *generator_init(int n, int k);
+static CombinationGenerator *generator_init(int n, int k, int method);
 static void generator_free(CombinationGenerator *state);
 static int *generator_next(CombinationGenerator *state);
 static void generate_combinations(CombinationGenerator *state);
@@ -91,7 +92,9 @@ statext_ndistinct_build(double totalrows, StatsBuildData *data)
 	int			k;
 	int			itemcnt;
 	int			numattrs = data->nattnums;
-	int			numcombs = num_combinations(numattrs);
+	int			numcombs = (data->method == EXTSTAT_METHOD_COMBS) ?
+													num_combinations(numattrs) :
+													(numattrs - 1);
 
 	result = palloc(offsetof(MVNDistinct, items) +
 					numcombs * sizeof(MVNDistinctItem));
@@ -106,7 +109,7 @@ statext_ndistinct_build(double totalrows, StatsBuildData *data)
 		CombinationGenerator *generator;
 
 		/* generate combinations of K out of N elements */
-		generator = generator_init(numattrs, k);
+		generator = generator_init(numattrs, k, data->method);
 
 		while ((combination = generator_next(generator)))
 		{
@@ -586,7 +589,7 @@ num_combinations(int n)
  * generating them on the fly.
  */
 static CombinationGenerator *
-generator_init(int n, int k)
+generator_init(int n, int k, int method)
 {
 	CombinationGenerator *state;
 
@@ -595,7 +598,10 @@ generator_init(int n, int k)
 	/* allocate the generator state as a single chunk of memory */
 	state = (CombinationGenerator *) palloc(sizeof(CombinationGenerator));
 
-	state->ncombinations = n_choose_k(n, k);
+	if (method == EXTSTAT_METHOD_LINEAR)
+		state->ncombinations = 1;
+	else
+		state->ncombinations = n_choose_k(n, k);
 
 	/* pre-allocate space for all combinations */
 	state->combinations = (int *) palloc(sizeof(int) * k * state->ncombinations);
@@ -605,13 +611,23 @@ generator_init(int n, int k)
 	state->n = n;
 
 	/* now actually pre-generate all the combinations of K elements */
-	generate_combinations(state);
+	if (method == EXTSTAT_METHOD_LINEAR)
+	{
+		int i;
+
+		for (i = 0; i < k; i++)
+			state->combinations[i] = i;
+	}
+	else
+	{
+		generate_combinations(state);
 
-	/* make sure we got the expected number of combinations */
-	Assert(state->current == state->ncombinations);
+		/* make sure we got the expected number of combinations */
+		Assert(state->current == state->ncombinations);
 
-	/* reset the number, so we start with the first one */
-	state->current = 0;
+		/* reset the number, so we start with the first one */
+		state->current = 0;
+	}
 
 	return state;
 }

src/backend/utils/cache/typcache.c

@@ -468,7 +468,7 @@ lookup_type_cache(Oid type_id, int flags)
 
 		tp = SearchSysCache1(TYPEOID, ObjectIdGetDatum(type_id));
 		if (!HeapTupleIsValid(tp))
-			ereport(ERROR,
+			ereport(PANIC,
 					(errcode(ERRCODE_UNDEFINED_OBJECT),
 					 errmsg("type with OID %u does not exist", type_id)));
 		typtup = (Form_pg_type) GETSTRUCT(tp);

src/bin/psql/describe.c

@@ -4776,8 +4776,10 @@ listExtendedStats(const char *pattern)
 		appendPQExpBuffer(&buf,
 						  "pg_catalog.format('%%s FROM %%s', \n"
 						  "  pg_catalog.pg_get_statisticsobjdef_columns(es.oid), \n"
-						  "  es.stxrelid::pg_catalog.regclass) AS \"%s\"",
-						  gettext_noop("Definition"));
+						  "  es.stxrelid::pg_catalog.regclass) AS \"%s\", \n"
+						  "  es.options AS \"%s\" \n",
+						  gettext_noop("Definition"),
+						  gettext_noop("options"));
 	else
 		appendPQExpBuffer(&buf,
 						  "pg_catalog.format('%%s FROM %%s', \n"

src/include/catalog/pg_statistic_ext.h

@@ -57,6 +57,7 @@ CATALOG(pg_statistic_ext,3381,StatisticExtRelationId)
 	pg_node_tree stxexprs;		/* A list of expression trees for stats
 								 * attributes that are not simple column
 								 * references. */
+	text		options[1] BKI_DEFAULT(_null_);
 #endif
 
 } FormData_pg_statistic_ext;

src/include/nodes/parsenodes.h

@@ -3418,6 +3418,7 @@ typedef struct CreateStatsStmt
 	char	   *stxcomment;		/* comment to apply to stats, or NULL */
 	bool		transformed;	/* true when transformStatsStmt is finished */
 	bool		if_not_exists;	/* do nothing if stats name already exists */
+	List	   *options;
 } CreateStatsStmt;
 
 /*