Add support for sorted gist index builds to btree_gist

This enables sortsupport in the btree_gist extension for faster
builds of gist indexes.

Sorted gist index build strategy is the new default now. Regression
tests are unchanged (except for one small change in the 'enum' test to
add coverage for enum values added later) and are using the sorted
build strategy instead.

One version of this was committed a long time ago already, in commit
9f984ba, but it was reverted because of buildfarm failures. The
failures were presumably caused by some small bugs, but we never got
around to debug and commit it again. This patch was written from
scratch, implementing the same idea, with some fragments copied from
the original patch.

Author: Bernd Helmle <mailings@oopsware.de>
Author: Andrey Borodin <x4mmm@yandex-team.ru>
Discussion: https://www.postgresql.org/message-id/64d324ce2a6d535d3f0f3baeeea7b25beff82ce4.camel@oopsware.de

Author: hlinnaka

contrib/btree_gist/Makefile

@@ -34,7 +34,7 @@ DATA = btree_gist--1.0--1.1.sql \
        btree_gist--1.1--1.2.sql btree_gist--1.2.sql btree_gist--1.2--1.3.sql \
        btree_gist--1.3--1.4.sql btree_gist--1.4--1.5.sql \
        btree_gist--1.5--1.6.sql btree_gist--1.6--1.7.sql \
-       btree_gist--1.7--1.8.sql
+       btree_gist--1.7--1.8.sql btree_gist--1.8--1.9.sql
 PGFILEDESC = "btree_gist - B-tree equivalent GiST operator classes"
 
 REGRESS = init int2 int4 int8 float4 float8 cash oid timestamp timestamptz \

contrib/btree_gist/btree_bit.c

@@ -6,6 +6,7 @@
 #include "btree_gist.h"
 #include "btree_utils_var.h"
 #include "utils/fmgrprotos.h"
+#include "utils/sortsupport.h"
 #include "utils/varbit.h"
 
 
@@ -18,10 +19,33 @@ PG_FUNCTION_INFO_V1(gbt_bit_picksplit);
 PG_FUNCTION_INFO_V1(gbt_bit_consistent);
 PG_FUNCTION_INFO_V1(gbt_bit_penalty);
 PG_FUNCTION_INFO_V1(gbt_bit_same);
+PG_FUNCTION_INFO_V1(gbt_bit_sortsupport);
+PG_FUNCTION_INFO_V1(gbt_varbit_sortsupport);
 
 
 /* define for comparison */
 
+static int
+gbt_bit_ssup_cmp(Datum x, Datum y, SortSupport ssup)
+{
+	GBT_VARKEY *key1 = PG_DETOAST_DATUM(x);
+	GBT_VARKEY *key2 = PG_DETOAST_DATUM(y);
+
+	GBT_VARKEY_R arg1 = gbt_var_key_readable(key1);
+	GBT_VARKEY_R arg2 = gbt_var_key_readable(key2);
+	Datum		result;
+
+	/* for leaf items we expect lower == upper, so only compare lower */
+	result = DirectFunctionCall2(byteacmp,
+								 PointerGetDatum(arg1.lower),
+								 PointerGetDatum(arg2.lower));
+
+	GBT_FREE_IF_COPY(key1, x);
+	GBT_FREE_IF_COPY(key2, y);
+
+	return DatumGetInt32(result);
+}
+
 static bool
 gbt_bitgt(const void *a, const void *b, Oid collation, FmgrInfo *flinfo)
 {
@@ -207,3 +231,25 @@ gbt_bit_penalty(PG_FUNCTION_ARGS)
 	PG_RETURN_POINTER(gbt_var_penalty(result, o, n, PG_GET_COLLATION(),
 									  &tinfo, fcinfo->flinfo));
 }
+
+Datum
+gbt_bit_sortsupport(PG_FUNCTION_ARGS)
+{
+	SortSupport ssup = (SortSupport) PG_GETARG_POINTER(0);
+
+	ssup->comparator = gbt_bit_ssup_cmp;
+	ssup->ssup_extra = NULL;
+
+	PG_RETURN_VOID();
+}
+
+Datum
+gbt_varbit_sortsupport(PG_FUNCTION_ARGS)
+{
+	SortSupport ssup = (SortSupport) PG_GETARG_POINTER(0);
+
+	ssup->comparator = gbt_bit_ssup_cmp;
+	ssup->ssup_extra = NULL;
+
+	PG_RETURN_VOID();
+}

contrib/btree_gist/btree_bool.c

@@ -5,6 +5,7 @@
 
 #include "btree_gist.h"
 #include "btree_utils_num.h"
+#include "utils/sortsupport.h"
 
 typedef struct boolkey
 {
@@ -22,6 +23,17 @@ PG_FUNCTION_INFO_V1(gbt_bool_picksplit);
 PG_FUNCTION_INFO_V1(gbt_bool_consistent);
 PG_FUNCTION_INFO_V1(gbt_bool_penalty);
 PG_FUNCTION_INFO_V1(gbt_bool_same);
+PG_FUNCTION_INFO_V1(gbt_bool_sortsupport);
+
+static int
+gbt_bool_ssup_cmp(Datum x, Datum y, SortSupport ssup)
+{
+	boolKEY    *arg1 = (boolKEY *) DatumGetPointer(x);
+	boolKEY    *arg2 = (boolKEY *) DatumGetPointer(y);
+
+	/* for leaf items we expect lower == upper, so only compare lower */
+	return (int32) arg1->lower - (int32) arg2->lower;
+}
 
 static bool
 gbt_boolgt(const void *a, const void *b, FmgrInfo *flinfo)
@@ -166,3 +178,14 @@ gbt_bool_same(PG_FUNCTION_ARGS)
 	*result = gbt_num_same((void *) b1, (void *) b2, &tinfo, fcinfo->flinfo);
 	PG_RETURN_POINTER(result);
 }
+
+Datum
+gbt_bool_sortsupport(PG_FUNCTION_ARGS)
+{
+	SortSupport ssup = (SortSupport) PG_GETARG_POINTER(0);
+
+	ssup->comparator = gbt_bool_ssup_cmp;
+	ssup->ssup_extra = NULL;
+
+	PG_RETURN_VOID();
+}

contrib/btree_gist/btree_bytea.c

@@ -6,6 +6,7 @@
 #include "btree_gist.h"
 #include "btree_utils_var.h"
 #include "utils/fmgrprotos.h"
+#include "utils/sortsupport.h"
 
 
 /*
@@ -17,7 +18,41 @@ PG_FUNCTION_INFO_V1(gbt_bytea_picksplit);
 PG_FUNCTION_INFO_V1(gbt_bytea_consistent);
 PG_FUNCTION_INFO_V1(gbt_bytea_penalty);
 PG_FUNCTION_INFO_V1(gbt_bytea_same);
+PG_FUNCTION_INFO_V1(gbt_bytea_sortsupport);
 
+/* sortsupport support */
+
+static int
+gbt_bytea_ssup_cmp(Datum x, Datum y, SortSupport ssup)
+{
+	GBT_VARKEY *key1 = PG_DETOAST_DATUM(x);
+	GBT_VARKEY *key2 = PG_DETOAST_DATUM(y);
+
+	GBT_VARKEY_R xkey = gbt_var_key_readable(key1);
+	GBT_VARKEY_R ykey = gbt_var_key_readable(key2);
+	Datum		result;
+
+	/* for leaf items we expect lower == upper, so only compare lower */
+	result = DirectFunctionCall2(byteacmp,
+								 PointerGetDatum(xkey.lower),
+								 PointerGetDatum(ykey.lower));
+
+	GBT_FREE_IF_COPY(key1, x);
+	GBT_FREE_IF_COPY(key2, y);
+
+	return DatumGetInt32(result);
+}
+
+Datum
+gbt_bytea_sortsupport(PG_FUNCTION_ARGS)
+{
+	SortSupport ssup = (SortSupport) PG_GETARG_POINTER(0);
+
+	ssup->comparator = gbt_bytea_ssup_cmp;
+	ssup->ssup_extra = NULL;
+
+	PG_RETURN_VOID();
+}
 
 /* define for comparison */
 

contrib/btree_gist/btree_cash.c

@@ -7,6 +7,7 @@
 #include "btree_utils_num.h"
 #include "common/int.h"
 #include "utils/cash.h"
+#include "utils/sortsupport.h"
 
 typedef struct
 {
@@ -25,6 +26,24 @@ PG_FUNCTION_INFO_V1(gbt_cash_consistent);
 PG_FUNCTION_INFO_V1(gbt_cash_distance);
 PG_FUNCTION_INFO_V1(gbt_cash_penalty);
 PG_FUNCTION_INFO_V1(gbt_cash_same);
+PG_FUNCTION_INFO_V1(gbt_cash_sortsupport);
+
+extern Datum cash_cmp(PG_FUNCTION_ARGS);
+
+static int
+gbt_cash_ssup_cmp(Datum x, Datum y, SortSupport ssup)
+{
+	cashKEY    *arg1 = (cashKEY *) DatumGetPointer(x);
+	cashKEY    *arg2 = (cashKEY *) DatumGetPointer(y);
+
+	/* for leaf items we expect lower == upper, so only compare lower */
+	if (arg1->lower > arg2->lower)
+		return 1;
+	else if (arg1->lower < arg2->lower)
+		return -1;
+	else
+		return 0;
+}
 
 static bool
 gbt_cashgt(const void *a, const void *b, FmgrInfo *flinfo)
@@ -215,3 +234,14 @@ gbt_cash_same(PG_FUNCTION_ARGS)
 	*result = gbt_num_same((void *) b1, (void *) b2, &tinfo, fcinfo->flinfo);
 	PG_RETURN_POINTER(result);
 }
+
+Datum
+gbt_cash_sortsupport(PG_FUNCTION_ARGS)
+{
+	SortSupport ssup = (SortSupport) PG_GETARG_POINTER(0);
+
+	ssup->comparator = gbt_cash_ssup_cmp;
+	ssup->ssup_extra = NULL;
+
+	PG_RETURN_VOID();
+}

contrib/btree_gist/btree_date.c

@@ -7,6 +7,7 @@
 #include "btree_utils_num.h"
 #include "utils/fmgrprotos.h"
 #include "utils/date.h"
+#include "utils/sortsupport.h"
 
 typedef struct
 {
@@ -25,6 +26,32 @@ PG_FUNCTION_INFO_V1(gbt_date_consistent);
 PG_FUNCTION_INFO_V1(gbt_date_distance);
 PG_FUNCTION_INFO_V1(gbt_date_penalty);
 PG_FUNCTION_INFO_V1(gbt_date_same);
+PG_FUNCTION_INFO_V1(gbt_date_sortsupport);
+
+/* sortsupport functions */
+
+static int
+gbt_date_ssup_cmp(Datum x, Datum y, SortSupport ssup)
+{
+	dateKEY    *akey = (dateKEY *) DatumGetPointer(x);
+	dateKEY    *bkey = (dateKEY *) DatumGetPointer(y);
+
+	/* for leaf items we expect lower == upper, so only compare lower */
+	return DatumGetInt32(DirectFunctionCall2(date_cmp,
+											 DateADTGetDatum(akey->lower),
+											 DateADTGetDatum(bkey->lower)));
+}
+
+Datum
+gbt_date_sortsupport(PG_FUNCTION_ARGS)
+{
+	SortSupport ssup = (SortSupport) PG_GETARG_POINTER(0);
+
+	ssup->comparator = gbt_date_ssup_cmp;
+	ssup->ssup_extra = NULL;
+
+	PG_RETURN_VOID();
+}
 
 static bool
 gbt_dategt(const void *a, const void *b, FmgrInfo *flinfo)

contrib/btree_gist/btree_enum.c

@@ -7,6 +7,8 @@
 #include "btree_utils_num.h"
 #include "fmgr.h"
 #include "utils/fmgrprotos.h"
+#include "utils/fmgroids.h"
+#include "utils/sortsupport.h"
 
 /* enums are really Oids, so we just use the same structure */
 
@@ -26,8 +28,23 @@ PG_FUNCTION_INFO_V1(gbt_enum_picksplit);
 PG_FUNCTION_INFO_V1(gbt_enum_consistent);
 PG_FUNCTION_INFO_V1(gbt_enum_penalty);
 PG_FUNCTION_INFO_V1(gbt_enum_same);
+PG_FUNCTION_INFO_V1(gbt_enum_sortsupport);
 
 
+static int
+gbt_enum_ssup_cmp(Datum x, Datum y, SortSupport ssup)
+{
+	oidKEY	   *arg1 = (oidKEY *) DatumGetPointer(x);
+	oidKEY	   *arg2 = (oidKEY *) DatumGetPointer(y);
+
+	/* for leaf items we expect lower == upper, so only compare lower */
+	return DatumGetInt32(CallerFInfoFunctionCall2(enum_cmp,
+												  ssup->ssup_extra,
+												  InvalidOid,
+												  arg1->lower,
+												  arg2->lower));
+}
+
 static bool
 gbt_enumgt(const void *a, const void *b, FmgrInfo *flinfo)
 {
@@ -183,3 +200,25 @@ gbt_enum_same(PG_FUNCTION_ARGS)
 	*result = gbt_num_same((void *) b1, (void *) b2, &tinfo, fcinfo->flinfo);
 	PG_RETURN_POINTER(result);
 }
+
+Datum
+gbt_enum_sortsupport(PG_FUNCTION_ARGS)
+{
+	SortSupport ssup = (SortSupport) PG_GETARG_POINTER(0);
+	FmgrInfo   *flinfo;
+
+	ssup->comparator = gbt_enum_ssup_cmp;
+
+	/*
+	 * Since gbt_enum_ssup_cmp() uses enum_cmp() like the rest of the
+	 * comparison functions, it also needs to pass flinfo when calling it. The
+	 * caller to a SortSupport comparison function doesn't provide an FmgrInfo
+	 * struct, so look it up now, save it in ssup_extra and use it in
+	 * gbt_enum_ssup_cmp() later.
+	 */
+	flinfo = MemoryContextAlloc(ssup->ssup_cxt, sizeof(FmgrInfo));
+	fmgr_info_cxt(F_ENUM_CMP, flinfo, ssup->ssup_cxt);
+	ssup->ssup_extra = flinfo;
+
+	PG_RETURN_VOID();
+}

contrib/btree_gist/btree_float4.c

@@ -5,6 +5,7 @@
 
 #include "btree_gist.h"
 #include "btree_utils_num.h"
+#include "utils/sortsupport.h"
 #include "utils/float.h"
 
 typedef struct float4key
@@ -24,6 +25,31 @@ PG_FUNCTION_INFO_V1(gbt_float4_consistent);
 PG_FUNCTION_INFO_V1(gbt_float4_distance);
 PG_FUNCTION_INFO_V1(gbt_float4_penalty);
 PG_FUNCTION_INFO_V1(gbt_float4_same);
+PG_FUNCTION_INFO_V1(gbt_float4_sortsupport);
+
+extern Datum btfloat4cmp(PG_FUNCTION_ARGS);
+
+/* sortsupport functions */
+static int
+gbt_float4_ssup_cmp(Datum x, Datum y, SortSupport ssup)
+{
+	float4KEY  *arg1 = (float4KEY *) DatumGetPointer(x);
+	float4KEY  *arg2 = (float4KEY *) DatumGetPointer(y);
+
+	/* for leaf items we expect lower == upper, so only compare lower */
+	return float4_cmp_internal(arg1->lower, arg2->lower);
+}
+
+Datum
+gbt_float4_sortsupport(PG_FUNCTION_ARGS)
+{
+	SortSupport ssup = (SortSupport) PG_GETARG_POINTER(0);
+
+	ssup->comparator = gbt_float4_ssup_cmp;
+	ssup->ssup_extra = NULL;
+
+	PG_RETURN_VOID();
+}
 
 static bool
 gbt_float4gt(const void *a, const void *b, FmgrInfo *flinfo)

contrib/btree_gist/btree_float8.c

@@ -5,6 +5,7 @@
 
 #include "btree_gist.h"
 #include "btree_utils_num.h"
+#include "utils/sortsupport.h"
 #include "utils/float.h"
 
 typedef struct float8key
@@ -24,7 +25,31 @@ PG_FUNCTION_INFO_V1(gbt_float8_consistent);
 PG_FUNCTION_INFO_V1(gbt_float8_distance);
 PG_FUNCTION_INFO_V1(gbt_float8_penalty);
 PG_FUNCTION_INFO_V1(gbt_float8_same);
+PG_FUNCTION_INFO_V1(gbt_float8_sortsupport);
 
+extern Datum btfloat8cmp(PG_FUNCTION_ARGS);
+
+/* sortsupport functions */
+static int
+gbt_float8_ssup_cmp(Datum x, Datum y, SortSupport ssup)
+{
+	float8KEY  *arg1 = (float8KEY *) DatumGetPointer(x);
+	float8KEY  *arg2 = (float8KEY *) DatumGetPointer(y);
+
+	/* for leaf items we expect lower == upper, so only compare lower */
+	return float8_cmp_internal(arg1->lower, arg2->lower);
+}
+
+Datum
+gbt_float8_sortsupport(PG_FUNCTION_ARGS)
+{
+	SortSupport ssup = (SortSupport) PG_GETARG_POINTER(0);
+
+	ssup->comparator = gbt_float8_ssup_cmp;
+	ssup->ssup_extra = NULL;
+
+	PG_RETURN_VOID();
+}
 
 static bool
 gbt_float8gt(const void *a, const void *b, FmgrInfo *flinfo)