Renamed functions to l2_norm

README.md

@@ -777,6 +777,7 @@ Function | Description | Added
 cosine_distance(halfvec, halfvec) → double precision | cosine distance | unreleased
 inner_product(halfvec, halfvec) → double precision | inner product | unreleased
 l2_distance(halfvec, halfvec) → double precision | Euclidean distance | unreleased
+l2_norm(halfvec) → double precision | Euclidean norm | unreleased
 l1_distance(halfvec, halfvec) → double precision | taxicab distance | unreleased
 quantize_binary(halfvec) → bit | quantize | unreleased
 subvector(halfvec, integer, integer) → halfvec | subvector | unreleased
@@ -818,6 +819,7 @@ Function | Description | Added
 cosine_distance(sparsevec, sparsevec) → double precision | cosine distance | unreleased
 inner_product(sparsevec, sparsevec) → double precision | inner product | unreleased
 l2_distance(sparsevec, sparsevec) → double precision | Euclidean distance | unreleased
+l2_norm(sparsevec) → double precision | Euclidean norm | unreleased
 l1_distance(sparsevec, sparsevec) → double precision | taxicab distance | unreleased
 
 ## Installation Notes - Linux and Mac

sql/vector--0.6.2--0.7.0.sql

@@ -71,8 +71,8 @@ CREATE FUNCTION cosine_distance(halfvec, halfvec) RETURNS float8
 CREATE FUNCTION l1_distance(halfvec, halfvec) RETURNS float8
 	AS 'MODULE_PATHNAME', 'halfvec_l1_distance' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
 
-CREATE FUNCTION halfvec_norm(halfvec) RETURNS float8
+CREATE FUNCTION l2_norm(halfvec) RETURNS float8
-	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
+	AS 'MODULE_PATHNAME', 'halfvec_l2_norm' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
 
 CREATE FUNCTION quantize_binary(halfvec) RETURNS bit
 	AS 'MODULE_PATHNAME', 'halfvec_quantize_binary' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
@@ -151,7 +151,7 @@ CREATE OPERATOR CLASS halfvec_cosine_ops
 	FOR TYPE halfvec USING hnsw AS
 	OPERATOR 1 <=> (halfvec, halfvec) FOR ORDER BY float_ops,
 	FUNCTION 1 halfvec_negative_inner_product(halfvec, halfvec),
-	FUNCTION 2 halfvec_norm(halfvec);
+	FUNCTION 2 l2_norm(halfvec);
 
 CREATE FUNCTION halfvec_to_vector(halfvec, integer, boolean) RETURNS vector
 	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
@@ -173,6 +173,9 @@ CREATE FUNCTION sparsevec_in(cstring, oid, integer) RETURNS sparsevec
 CREATE FUNCTION sparsevec_out(sparsevec) RETURNS cstring
 	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
 
+CREATE FUNCTION sparsevec_typmod_in(cstring[]) RETURNS integer
+	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
+
 CREATE FUNCTION sparsevec_recv(internal, oid, integer) RETURNS sparsevec
 	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
 
@@ -182,6 +185,7 @@ CREATE FUNCTION sparsevec_send(sparsevec) RETURNS bytea
 CREATE TYPE sparsevec (
 	INPUT     = sparsevec_in,
 	OUTPUT    = sparsevec_out,
+	TYPMOD_IN = sparsevec_typmod_in,
 	RECEIVE   = sparsevec_recv,
 	SEND      = sparsevec_send,
 	STORAGE   = external
@@ -196,8 +200,8 @@ CREATE FUNCTION inner_product(sparsevec, sparsevec) RETURNS float8
 CREATE FUNCTION cosine_distance(sparsevec, sparsevec) RETURNS float8
 	AS 'MODULE_PATHNAME', 'sparsevec_cosine_distance' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
 
-CREATE FUNCTION sparsevec_norm(sparsevec) RETURNS float8
+CREATE FUNCTION l2_norm(sparsevec) RETURNS float8
-	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
+	AS 'MODULE_PATHNAME', 'sparsevec_l2_norm' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
 
 CREATE FUNCTION sparsevec_l2_squared_distance(sparsevec, sparsevec) RETURNS float8
 	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
@@ -252,4 +256,4 @@ CREATE OPERATOR CLASS sparsevec_cosine_ops
 	FOR TYPE sparsevec USING hnsw AS
 	OPERATOR 1 <=> (sparsevec, sparsevec) FOR ORDER BY float_ops,
 	FUNCTION 1 sparsevec_negative_inner_product(sparsevec, sparsevec),
-	FUNCTION 2 sparsevec_norm(sparsevec);
+	FUNCTION 2 l2_norm(sparsevec);

sql/vector.sql

@@ -364,8 +364,8 @@ CREATE FUNCTION cosine_distance(halfvec, halfvec) RETURNS float8
 CREATE FUNCTION l1_distance(halfvec, halfvec) RETURNS float8
 	AS 'MODULE_PATHNAME', 'halfvec_l1_distance' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
 
-CREATE FUNCTION halfvec_norm(halfvec) RETURNS float8
+CREATE FUNCTION l2_norm(halfvec) RETURNS float8
-	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
+	AS 'MODULE_PATHNAME', 'halfvec_l2_norm' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
 
 CREATE FUNCTION quantize_binary(halfvec) RETURNS bit
 	AS 'MODULE_PATHNAME', 'halfvec_quantize_binary' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
@@ -454,7 +454,7 @@ CREATE OPERATOR CLASS halfvec_cosine_ops
 	FOR TYPE halfvec USING hnsw AS
 	OPERATOR 1 <=> (halfvec, halfvec) FOR ORDER BY float_ops,
 	FUNCTION 1 halfvec_negative_inner_product(halfvec, halfvec),
-	FUNCTION 2 halfvec_norm(halfvec);
+	FUNCTION 2 l2_norm(halfvec);
 
 -- extension casts
 
@@ -480,6 +480,9 @@ CREATE FUNCTION sparsevec_in(cstring, oid, integer) RETURNS sparsevec
 CREATE FUNCTION sparsevec_out(sparsevec) RETURNS cstring
 	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
 
+CREATE FUNCTION sparsevec_typmod_in(cstring[]) RETURNS integer
+	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
+
 CREATE FUNCTION sparsevec_recv(internal, oid, integer) RETURNS sparsevec
 	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
 
@@ -489,6 +492,7 @@ CREATE FUNCTION sparsevec_send(sparsevec) RETURNS bytea
 CREATE TYPE sparsevec (
 	INPUT     = sparsevec_in,
 	OUTPUT    = sparsevec_out,
+	TYPMOD_IN = sparsevec_typmod_in,
 	RECEIVE   = sparsevec_recv,
 	SEND      = sparsevec_send,
 	STORAGE   = external
@@ -505,8 +509,8 @@ CREATE FUNCTION inner_product(sparsevec, sparsevec) RETURNS float8
 CREATE FUNCTION cosine_distance(sparsevec, sparsevec) RETURNS float8
 	AS 'MODULE_PATHNAME', 'sparsevec_cosine_distance' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
 
-CREATE FUNCTION sparsevec_norm(sparsevec) RETURNS float8
+CREATE FUNCTION l2_norm(sparsevec) RETURNS float8
-	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
+	AS 'MODULE_PATHNAME', 'sparsevec_l2_norm' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
 
 -- sparsevec private functions
 
@@ -571,4 +575,4 @@ CREATE OPERATOR CLASS sparsevec_cosine_ops
 	FOR TYPE sparsevec USING hnsw AS
 	OPERATOR 1 <=> (sparsevec, sparsevec) FOR ORDER BY float_ops,
 	FUNCTION 1 sparsevec_negative_inner_product(sparsevec, sparsevec),
-	FUNCTION 2 sparsevec_norm(sparsevec);
+	FUNCTION 2 l2_norm(sparsevec);

src/halfvec.c

@@ -950,9 +950,9 @@ halfvec_l1_distance(PG_FUNCTION_ARGS)
 /*
  * Get the L2 norm of a half vector
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(halfvec_norm);
+PGDLLEXPORT PG_FUNCTION_INFO_V1(halfvec_l2_norm);
 Datum
-halfvec_norm(PG_FUNCTION_ARGS)
+halfvec_l2_norm(PG_FUNCTION_ARGS)
 {
 	HalfVector *a = PG_GETARG_HALFVEC_P(0);
 	half	   *ax = a->x;

src/hnswbuild.c

@@ -681,6 +681,8 @@ GetMaxDimensions(HnswType type)
 		maxDimensions *= 2;
 	else if (type == HNSW_TYPE_BIT)
 		maxDimensions *= 32;
+	else if (type == HNSW_TYPE_SPARSEVEC)
+		maxDimensions = INT_MAX;
 
 	return maxDimensions;
 }
@@ -691,6 +693,8 @@ GetMaxDimensions(HnswType type)
 static void
 InitBuildState(HnswBuildState * buildstate, Relation heap, Relation index, IndexInfo *indexInfo, ForkNumber forkNum)
 {
+	int			maxDimensions;
+
 	buildstate->heap = heap;
 	buildstate->index = index;
 	buildstate->indexInfo = indexInfo;
@@ -701,17 +705,14 @@ InitBuildState(HnswBuildState * buildstate, Relation heap, Relation index, Index
 	buildstate->efConstruction = HnswGetEfConstruction(index);
 	buildstate->dimensions = TupleDescAttr(index->rd_att, 0)->atttypmod;
 
+	maxDimensions = GetMaxDimensions(buildstate->type);
 
-	if (buildstate->type != HNSW_TYPE_SPARSEVEC)
+	/* Require column to have dimensions to be indexed */
-	{
+	if (buildstate->dimensions < 0)
-		int			maxDimensions = GetMaxDimensions(buildstate->type);
+		elog(ERROR, "column does not have dimensions");
 
-		if (buildstate->dimensions < 0)
+	if (buildstate->dimensions > maxDimensions)
-			elog(ERROR, "column does not have dimensions");
+		elog(ERROR, "column cannot have more than %d dimensions for hnsw index", maxDimensions);
-
-		if (buildstate->dimensions > maxDimensions)
-			elog(ERROR, "column cannot have more than %d dimensions for hnsw index", maxDimensions);
-	}
 
 	if (buildstate->efConstruction < 2 * buildstate->m)
 		elog(ERROR, "ef_construction must be greater than or equal to 2 * m");

src/hnswutils.c

@@ -229,7 +229,7 @@ HnswNormValue(FmgrInfo *procinfo, Oid collation, Datum *value, HnswType type)
 		else if (type == HNSW_TYPE_SPARSEVEC)
 		{
 			SparseVector *v = DatumGetSparseVector(*value);
-			SparseVector *result = InitSparseVector(v->nnz);
+			SparseVector *result = InitSparseVector(v->dim, v->nnz);
 			float	   *vx = SPARSEVEC_VALUES(v);
 			float	   *rx = SPARSEVEC_VALUES(result);
 

src/ivfbuild.c

@@ -57,7 +57,7 @@ AddSample(Datum *values, IvfflatBuildState * buildstate)
 	 */
 	if (buildstate->kmeansnormprocinfo != NULL)
 	{
-		if (!IvfflatNormValue(buildstate->kmeansnormprocinfo, buildstate->collation, &value, buildstate->type))
+		if (!IvfflatNormValue(buildstate->kmeansnormprocinfo, buildstate->collation, &value))
 			return;
 	}
 
@@ -153,7 +153,7 @@ AddTupleToSort(Relation index, ItemPointer tid, Datum *values, IvfflatBuildState
 	/* Normalize if needed */
 	if (buildstate->normprocinfo != NULL)
 	{
-		if (!IvfflatNormValue(buildstate->normprocinfo, buildstate->collation, &value, buildstate->type))
+		if (!IvfflatNormValue(buildstate->normprocinfo, buildstate->collation, &value))
 			return;
 	}
 
@@ -312,39 +312,25 @@ InsertTuples(Relation index, IvfflatBuildState * buildstate, ForkNumber forkNum)
 	}
 }
 
-/*
- * Get max dimensions
- */
-static int
-GetMaxDimensions(IvfflatType type)
-{
-	return IVFFLAT_MAX_DIM;
-}
-
 /*
  * Initialize the build state
  */
 static void
 InitBuildState(IvfflatBuildState * buildstate, Relation heap, Relation index, IndexInfo *indexInfo)
 {
-	int			maxDimensions;
-
 	buildstate->heap = heap;
 	buildstate->index = index;
 	buildstate->indexInfo = indexInfo;
-	buildstate->type = IvfflatGetType(index);
 
 	buildstate->lists = IvfflatGetLists(index);
 	buildstate->dimensions = TupleDescAttr(index->rd_att, 0)->atttypmod;
 
-	maxDimensions = GetMaxDimensions(buildstate->type);
-
 	/* Require column to have dimensions to be indexed */
 	if (buildstate->dimensions < 0)
 		elog(ERROR, "column does not have dimensions");
 
-	if (buildstate->dimensions > maxDimensions)
+	if (buildstate->dimensions > IVFFLAT_MAX_DIM)
-		elog(ERROR, "column cannot have more than %d dimensions for ivfflat index", maxDimensions);
+		elog(ERROR, "column cannot have more than %d dimensions for ivfflat index", IVFFLAT_MAX_DIM);
 
 	buildstate->reltuples = 0;
 	buildstate->indtuples = 0;

src/ivfflat.h

@@ -43,11 +43,6 @@
 #define IVFFLAT_MAX_LISTS		32768
 #define IVFFLAT_DEFAULT_PROBES	1
 
-typedef enum IvfflatType
-{
-	IVFFLAT_TYPE_VECTOR
-}			IvfflatType;
-
 /* Build phases */
 /* PROGRESS_CREATEIDX_SUBPHASE_INITIALIZE is 1 */
 #define PROGRESS_IVFFLAT_PHASE_KMEANS	2
@@ -158,7 +153,6 @@ typedef struct IvfflatBuildState
 	Relation	heap;
 	Relation	index;
 	IndexInfo  *indexInfo;
-	IvfflatType type;
 
 	/* Settings */
 	int			dimensions;
@@ -272,8 +266,7 @@ void		VectorArrayFree(VectorArray arr);
 void		PrintVectorArray(char *msg, VectorArray arr);
 void		IvfflatKmeans(Relation index, VectorArray samples, VectorArray centers);
 FmgrInfo   *IvfflatOptionalProcInfo(Relation index, uint16 procnum);
-IvfflatType IvfflatGetType(Relation index);
+bool		IvfflatNormValue(FmgrInfo *procinfo, Oid collation, Datum *value);
-bool		IvfflatNormValue(FmgrInfo *procinfo, Oid collation, Datum *value, IvfflatType type);
 int			IvfflatGetLists(Relation index);
 void		IvfflatGetMetaPageInfo(Relation index, int *lists, int *dimensions);
 void		IvfflatUpdateList(Relation index, ListInfo listInfo, BlockNumber insertPage, BlockNumber originalInsertPage, BlockNumber startPage, ForkNumber forkNum);

src/ivfinsert.c

@@ -85,7 +85,7 @@ InsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heap_tid, R
 	normprocinfo = IvfflatOptionalProcInfo(index, IVFFLAT_NORM_PROC);
 	if (normprocinfo != NULL)
 	{
-		if (!IvfflatNormValue(normprocinfo, index->rd_indcollation[0], &value, IvfflatGetType(index)))
+		if (!IvfflatNormValue(normprocinfo, index->rd_indcollation[0], &value))
 			return;
 	}
 

src/ivfscan.c

@@ -268,7 +268,6 @@ ivfflatgettuple(IndexScanDesc scan, ScanDirection dir)
 	if (so->first)
 	{
 		Datum		value;
-		IvfflatType type = IvfflatGetType(scan->indexRelation);
 
 		/* Count index scan for stats */
 		pgstat_count_index_scan(scan->indexRelation);
@@ -283,12 +282,7 @@ ivfflatgettuple(IndexScanDesc scan, ScanDirection dir)
 			elog(ERROR, "non-MVCC snapshots are not supported with ivfflat");
 
 		if (scan->orderByData->sk_flags & SK_ISNULL)
-		{
+			value = PointerGetDatum(InitVector(so->dimensions));
-			if (type == IVFFLAT_TYPE_VECTOR)
-				value = PointerGetDatum(InitVector(so->dimensions));
-			else
-				elog(ERROR, "Unsupported type");
-		}
 		else
 		{
 			value = scan->orderByData->sk_argument;
@@ -299,7 +293,7 @@ ivfflatgettuple(IndexScanDesc scan, ScanDirection dir)
 
 			/* Fine if normalization fails */
 			if (so->normprocinfo != NULL)
-				IvfflatNormValue(so->normprocinfo, so->collation, &value, type);
+				IvfflatNormValue(so->normprocinfo, so->collation, &value);
 		}
 
 		IvfflatBench("GetScanLists", GetScanLists(scan, value));

src/ivfutils.c

@@ -66,15 +66,6 @@ IvfflatOptionalProcInfo(Relation index, uint16 procnum)
 	return index_getprocinfo(index, 1, procnum);
 }
 
-/*
- * Get type
- */
-IvfflatType
-IvfflatGetType(Relation index)
-{
-	return IVFFLAT_TYPE_VECTOR;
-}
-
 /*
  * Divide by the norm
  *
@@ -84,24 +75,19 @@ IvfflatGetType(Relation index)
  * if it's different than the original value
  */
 bool
-IvfflatNormValue(FmgrInfo *procinfo, Oid collation, Datum *value, IvfflatType type)
+IvfflatNormValue(FmgrInfo *procinfo, Oid collation, Datum *value)
 {
 	double		norm = DatumGetFloat8(FunctionCall1Coll(procinfo, collation, *value));
 
 	if (norm > 0)
 	{
-		if (type == IVFFLAT_TYPE_VECTOR)
+		Vector	   *v = DatumGetVector(*value);
-		{
+		Vector	   *result = InitVector(v->dim);
-			Vector	   *v = DatumGetVector(*value);
-			Vector	   *result = InitVector(v->dim);
 
-			for (int i = 0; i < v->dim; i++)
+		for (int i = 0; i < v->dim; i++)
-				result->x[i] = v->x[i] / norm;
+			result->x[i] = v->x[i] / norm;
 
-			*value = PointerGetDatum(result);
+		*value = PointerGetDatum(result);
-		}
-		else
-			elog(ERROR, "Unsupported type");
 
 		return true;
 	}

src/sparsevec.c

@@ -18,28 +18,69 @@
 #include "utils/builtins.h"
 #endif
 
+/*
+ * Ensure same dimensions
+ */
+static inline void
+CheckDims(SparseVector * a, SparseVector * b)
+{
+	if (a->dim != b->dim)
+		ereport(ERROR,
+				(errcode(ERRCODE_DATA_EXCEPTION),
+				 errmsg("different sparsevec dimensions %d and %d", a->dim, b->dim)));
+}
+
+/*
+ * Ensure expected dimensions
+ */
+static inline void
+CheckExpectedDim(int32 typmod, int dim)
+{
+	if (typmod != -1 && typmod != dim)
+		ereport(ERROR,
+				(errcode(ERRCODE_DATA_EXCEPTION),
+				 errmsg("expected %d dimensions, not %d", typmod, dim)));
+}
+
+/*
+ * Ensure valid dimensions
+ */
+static inline void
+CheckDim(int dim)
+{
+	if (dim < 1)
+		ereport(ERROR,
+				(errcode(ERRCODE_DATA_EXCEPTION),
+				 errmsg("sparsevec must have at least 1 dimension")));
+
+	if (dim > SPARSEVEC_MAX_DIM)
+		ereport(ERROR,
+				(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
+				 errmsg("sparsevec cannot have more than %d dimensions", SPARSEVEC_MAX_DIM)));
+}
+
 /*
  * Ensure valid nnz
  */
 static inline void
-CheckNnz(int nnz)
+CheckNnz(int nnz, int dim)
 {
 	if (nnz < 0)
 		ereport(ERROR,
 				(errcode(ERRCODE_DATA_EXCEPTION),
 				 errmsg("sparsevec must have at least one element")));
 
-	if (nnz > SPARSEVEC_MAX_NNZ)
+	if (nnz > dim)
 		ereport(ERROR,
 				(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
-				 errmsg("sparsevec cannot have more elements than non-zero elements")));
+				 errmsg("sparsevec cannot have more elements than dimensions")));
 }
 
 /*
  * Ensure valid index
  */
 static inline void
-CheckIndex(int32 *indices, int i)
+CheckIndex(int32 *indices, int i, int dim)
 {
 	int32		index = indices[i];
 
@@ -48,6 +89,11 @@ CheckIndex(int32 *indices, int i)
 				(errcode(ERRCODE_DATA_EXCEPTION),
 				 errmsg("index must not be negative")));
 
+	if (index >= dim)
+		ereport(ERROR,
+				(errcode(ERRCODE_DATA_EXCEPTION),
+				 errmsg("index must be less than dimensions")));
+
 	if (i > 0)
 	{
 		if (index < indices[i - 1])
@@ -83,7 +129,7 @@ CheckElement(float value)
  * Allocate and initialize a new sparse vector
  */
 SparseVector *
-InitSparseVector(int nnz)
+InitSparseVector(int dim, int nnz)
 {
 	SparseVector *result;
 	int			size;
@@ -91,6 +137,7 @@ InitSparseVector(int nnz)
 	size = SPARSEVEC_SIZE(nnz);
 	result = (SparseVector *) palloc0(size);
 	SET_VARSIZE(result, size);
+	result->dim = dim;
 	result->nnz = nnz;
 
 	return result;
@@ -120,6 +167,8 @@ Datum
 sparsevec_in(PG_FUNCTION_ARGS)
 {
 	char	   *lit = PG_GETARG_CSTRING(0);
+	int32		typmod = PG_GETARG_INT32(2);
+	int			dim;
 	char	   *pt;
 	char	   *stringEnd;
 	SparseVector *result;
@@ -141,11 +190,6 @@ sparsevec_in(PG_FUNCTION_ARGS)
 		pt++;
 	}
 
-	if (maxNnz > SPARSEVEC_MAX_NNZ)
-		ereport(ERROR,
-				(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
-				 errmsg("sparsevec cannot have more than %d non-zero elements", SPARSEVEC_MAX_NNZ)));
-
 	indices = palloc(maxNnz * sizeof(int32));
 	values = palloc(maxNnz * sizeof(float));
 
@@ -253,6 +297,24 @@ sparsevec_in(PG_FUNCTION_ARGS)
 
 	stringEnd++;
 
+	if (*stringEnd != '/')
+		ereport(ERROR,
+				(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
+				 errmsg("malformed sparsevec literal: \"%s\"", lit),
+				 errdetail("Unexpected end of input.")));
+
+	stringEnd++;
+
+	/* Use similar logic as int2vectorin */
+	errno = 0;
+	pt = stringEnd;
+	dim = strtol(pt, &stringEnd, 10);
+
+	if (stringEnd == pt)
+		ereport(ERROR,
+				(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
+				 errmsg("invalid input syntax for type sparsevec: \"%s\"", lit)));
+
 	/* Only whitespace is allowed after the closing brace */
 	while (sparsevec_isspace(*stringEnd))
 		stringEnd++;
@@ -261,18 +323,21 @@ sparsevec_in(PG_FUNCTION_ARGS)
 		ereport(ERROR,
 				(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
 				 errmsg("malformed sparsevec literal: \"%s\"", lit),
-				 errdetail("Junk after closing right brace.")));
+				 errdetail("Junk after closing.")));
 
 	pfree(litcopy);
 
-	result = InitSparseVector(nnz);
+	CheckDim(dim);
+	CheckExpectedDim(typmod, dim);
+
+	result = InitSparseVector(dim, nnz);
 	rvalues = SPARSEVEC_VALUES(result);
 	for (int i = 0; i < nnz; i++)
 	{
 		result->indices[i] = indices[i];
 		rvalues[i] = values[i];
 
-		CheckIndex(result->indices, i);
+		CheckIndex(result->indices, i, dim);
 		CheckElement(rvalues[i]);
 	}
 
@@ -317,9 +382,11 @@ sparsevec_out(PG_FUNCTION_ARGS)
 	 *
 	 * nnz - 1 bytes for ,
 	 *
-	 * 3 bytes for {, }, and \0
+	 * 10 bytes for dimensions
+	 *
+	 * 4 bytes for {, }, /, and \0
 	 */
-	buf = (char *) palloc((11 + FLOAT_SHORTEST_DECIMAL_LEN) * sparsevec->nnz + 2);
+	buf = (char *) palloc((11 + FLOAT_SHORTEST_DECIMAL_LEN) * sparsevec->nnz + 13);
 	ptr = buf;
 
 	AppendChar(ptr, '{');
@@ -335,12 +402,45 @@ sparsevec_out(PG_FUNCTION_ARGS)
 	}
 
 	AppendChar(ptr, '}');
+	AppendChar(ptr, '/');
+	AppendInt(ptr, sparsevec->dim);
 	*ptr = '\0';
 
 	PG_FREE_IF_COPY(sparsevec, 0);
 	PG_RETURN_CSTRING(buf);
 }
 
+/*
+ * Convert type modifier
+ */
+PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec_typmod_in);
+Datum
+sparsevec_typmod_in(PG_FUNCTION_ARGS)
+{
+	ArrayType  *ta = PG_GETARG_ARRAYTYPE_P(0);
+	int32	   *tl;
+	int			n;
+
+	tl = ArrayGetIntegerTypmods(ta, &n);
+
+	if (n != 1)
+		ereport(ERROR,
+				(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
+				 errmsg("invalid type modifier")));
+
+	if (*tl < 1)
+		ereport(ERROR,
+				(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
+				 errmsg("dimensions for type sparsevec must be at least 1")));
+
+	if (*tl > SPARSEVEC_MAX_DIM)
+		ereport(ERROR,
+				(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
+				 errmsg("dimensions for type sparsevec cannot exceed %d", SPARSEVEC_MAX_DIM)));
+
+	PG_RETURN_INT32(*tl);
+}
+
 /*
  * Convert external binary representation to internal representation
  */
@@ -349,30 +449,33 @@ Datum
 sparsevec_recv(PG_FUNCTION_ARGS)
 {
 	StringInfo	buf = (StringInfo) PG_GETARG_POINTER(0);
+	int32		typmod = PG_GETARG_INT32(2);
 	SparseVector *result;
+	int32		dim;
 	int32		nnz;
 	int32		unused;
-	int32		unused2;
 	float	   *values;
 
+	dim = pq_getmsgint(buf, sizeof(int32));
 	nnz = pq_getmsgint(buf, sizeof(int32));
 	unused = pq_getmsgint(buf, sizeof(int32));
-	unused2 = pq_getmsgint(buf, sizeof(int32));
 
-	CheckNnz(nnz);
+	CheckDim(dim);
+	CheckNnz(nnz, dim);
+	CheckExpectedDim(typmod, dim);
 
-	if (unused != 0 || unused2 != 0)
+	if (unused != 0)
 		ereport(ERROR,
 				(errcode(ERRCODE_DATA_EXCEPTION),
 				 errmsg("expected unused to be 0, not %d", unused)));
 
-	result = InitSparseVector(nnz);
+	result = InitSparseVector(dim, nnz);
 	values = SPARSEVEC_VALUES(result);
 
 	for (int i = 0; i < nnz; i++)
 	{
 		result->indices[i] = pq_getmsgint(buf, sizeof(int32));
-		CheckIndex(result->indices, i);
+		CheckIndex(result->indices, i, dim);
 	}
 
 	for (int i = 0; i < nnz; i++)
@@ -396,9 +499,9 @@ sparsevec_send(PG_FUNCTION_ARGS)
 	StringInfoData buf;
 
 	pq_begintypsend(&buf);
+	pq_sendint(&buf, svec->dim, sizeof(int32));
 	pq_sendint(&buf, svec->nnz, sizeof(int32));
 	pq_sendint(&buf, svec->unused, sizeof(int32));
-	pq_sendint(&buf, svec->unused2, sizeof(int32));
 	for (int i = 0; i < svec->nnz; i++)
 		pq_sendint(&buf, svec->indices[i], sizeof(int32));
 	for (int i = 0; i < svec->nnz; i++)
@@ -409,12 +512,16 @@ sparsevec_send(PG_FUNCTION_ARGS)
 
 /*
  * Convert sparse vector to sparse vector
+ * This is needed to check the type modifier
  */
 PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec);
 Datum
 sparsevec(PG_FUNCTION_ARGS)
 {
 	SparseVector *svec = PG_GETARG_SPARSEVEC_P(0);
+	int32		typmod = PG_GETARG_INT32(1);
+
+	CheckExpectedDim(typmod, svec->dim);
 
 	PG_RETURN_POINTER(svec);
 }
@@ -427,20 +534,23 @@ Datum
 vector_to_sparsevec(PG_FUNCTION_ARGS)
 {
 	Vector	   *vec = PG_GETARG_VECTOR_P(0);
+	int32		typmod = PG_GETARG_INT32(1);
 	SparseVector *result;
 	int			dim = vec->dim;
 	int			nnz = 0;
 	float	   *values;
 	int			j = 0;
 
+	CheckDim(dim);
+	CheckExpectedDim(typmod, dim);
+
 	for (int i = 0; i < dim; i++)
 	{
 		if (vec->x[i] != 0)
 			nnz++;
 	}
 
-	CheckNnz(nnz);
+	result = InitSparseVector(dim, nnz);
-	result = InitSparseVector(nnz);
 	values = SPARSEVEC_VALUES(result);
 	for (int i = 0; i < dim; i++)
 	{
@@ -517,6 +627,8 @@ sparsevec_l2_distance(PG_FUNCTION_ARGS)
 	SparseVector *a = PG_GETARG_SPARSEVEC_P(0);
 	SparseVector *b = PG_GETARG_SPARSEVEC_P(1);
 
+	CheckDims(a, b);
+
 	PG_RETURN_FLOAT8(sqrt(l2_distance_squared_internal(a, b)));
 }
 
@@ -531,6 +643,8 @@ sparsevec_l2_squared_distance(PG_FUNCTION_ARGS)
 	SparseVector *a = PG_GETARG_SPARSEVEC_P(0);
 	SparseVector *b = PG_GETARG_SPARSEVEC_P(1);
 
+	CheckDims(a, b);
+
 	PG_RETURN_FLOAT8(l2_distance_squared_internal(a, b));
 }
 
@@ -580,6 +694,8 @@ sparsevec_inner_product(PG_FUNCTION_ARGS)
 	SparseVector *a = PG_GETARG_SPARSEVEC_P(0);
 	SparseVector *b = PG_GETARG_SPARSEVEC_P(1);
 
+	CheckDims(a, b);
+
 	PG_RETURN_FLOAT8(inner_product_internal(a, b));
 }
 
@@ -593,6 +709,8 @@ sparsevec_negative_inner_product(PG_FUNCTION_ARGS)
 	SparseVector *a = PG_GETARG_SPARSEVEC_P(0);
 	SparseVector *b = PG_GETARG_SPARSEVEC_P(1);
 
+	CheckDims(a, b);
+
 	PG_RETURN_FLOAT8(-inner_product_internal(a, b));
 }
 
@@ -611,6 +729,8 @@ sparsevec_cosine_distance(PG_FUNCTION_ARGS)
 	float		normb = 0.0;
 	double		similarity;
 
+	CheckDims(a, b);
+
 	similarity = inner_product_internal(a, b);
 
 	/* Auto-vectorized */
@@ -642,9 +762,9 @@ sparsevec_cosine_distance(PG_FUNCTION_ARGS)
 /*
  * Get the L2 norm of a sparse vector
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec_norm);
+PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec_l2_norm);
 Datum
-sparsevec_norm(PG_FUNCTION_ARGS)
+sparsevec_l2_norm(PG_FUNCTION_ARGS)
 {
 	SparseVector *a = PG_GETARG_SPARSEVEC_P(0);
 	float	   *ax = SPARSEVEC_VALUES(a);

src/sparsevec.h

@@ -1,7 +1,7 @@
 #ifndef SPARSEVEC_H
 #define SPARSEVEC_H
 
-#define SPARSEVEC_MAX_NNZ 100000
+#define SPARSEVEC_MAX_DIM 100000
 
 /* Ensure values are aligned */
 #define SPARSEVEC_SIZE(_nnz)		(offsetof(SparseVector, indices) + MAXALIGN((_nnz) * sizeof(int32)) + (_nnz * sizeof(float)))
@@ -13,12 +13,12 @@
 typedef struct SparseVector
 {
 	int32		vl_len_;		/* varlena header (do not touch directly!) */
+	int32		dim;			/* number of dimensions */
 	int32		nnz;
 	int32		unused;
-	int32		unused2;
 	int32		indices[FLEXIBLE_ARRAY_MEMBER];
 }			SparseVector;
 
-SparseVector *InitSparseVector(int nnz);
+SparseVector *InitSparseVector(int dim, int nnz);
 
 #endif

src/vector.c

@@ -1236,19 +1236,11 @@ sparsevec_to_vector(PG_FUNCTION_ARGS)
 	SparseVector *svec = PG_GETARG_SPARSEVEC_P(0);
 	int32		typmod = PG_GETARG_INT32(1);
 	Vector	   *result;
-	int			dim;
+	int			dim = svec->dim;
 	float	   *values = SPARSEVEC_VALUES(svec);
-	int			maxIndex = svec->nnz == 0 ? -1 : svec->indices[svec->nnz - 1];
-
-	if (typmod == -1)
-		dim = maxIndex + 1;
-	else
-		dim = typmod;
 
 	CheckDim(dim);
-
+	CheckExpectedDim(typmod, dim);
-	if (dim < maxIndex + 1)
-		elog(ERROR, "Vector must have at least %d dimensions", maxIndex + 1);
 
 	result = InitVector(dim);
 	for (int i = 0; i < svec->nnz; i++)

test/expected/hnsw_sparsevec_cosine.out

@@ -1,17 +1,17 @@
 SET enable_seqscan = off;
-CREATE TABLE t (val sparsevec);
+CREATE TABLE t (val sparsevec(3));
-INSERT INTO t (val) VALUES ('{}'), ('{0:1,1:2,2:3}'), ('{0:1,1:1,2:1}'), (NULL);
+INSERT INTO t (val) VALUES ('{}/3'), ('{0:1,1:2,2:3}/3'), ('{0:1,1:1,2:1}/3'), (NULL);
 CREATE INDEX ON t USING hnsw (val sparsevec_cosine_ops);
-INSERT INTO t (val) VALUES ('{0:1,1:2,2:4}');
+INSERT INTO t (val) VALUES ('{0:1,1:2,2:4}/3');
-SELECT * FROM t ORDER BY val <=> '{0:3,1:3,2:3}';
+SELECT * FROM t ORDER BY val <=> '{0:3,1:3,2:3}/3';
-      val      
+       val       
----------------
+-----------------
- {0:1,1:1,2:1}
+ {0:1,1:1,2:1}/3
- {0:1,1:2,2:3}
+ {0:1,1:2,2:3}/3
- {0:1,1:2,2:4}
+ {0:1,1:2,2:4}/3
 (3 rows)
 
-SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> '{}') t2;
+SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> '{}/3') t2;
  count 
 -------
      3

test/expected/hnsw_sparsevec_ip.out

@@ -1,15 +1,15 @@
 SET enable_seqscan = off;
-CREATE TABLE t (val sparsevec);
+CREATE TABLE t (val sparsevec(3));
-INSERT INTO t (val) VALUES ('{}'), ('{0:1,1:2,2:3}'), ('{0:1,1:1,2:1}'), (NULL);
+INSERT INTO t (val) VALUES ('{}/3'), ('{0:1,1:2,2:3}/3'), ('{0:1,1:1,2:1}/3'), (NULL);
 CREATE INDEX ON t USING hnsw (val sparsevec_ip_ops);
-INSERT INTO t (val) VALUES ('{0:1,1:2,2:4}');
+INSERT INTO t (val) VALUES ('{0:1,1:2,2:4}/3');
-SELECT * FROM t ORDER BY val <#> '{0:3,1:3,2:3}';
+SELECT * FROM t ORDER BY val <#> '{0:3,1:3,2:3}/3';
-      val      
+       val       
----------------
+-----------------
- {0:1,1:2,2:4}
+ {0:1,1:2,2:4}/3
- {0:1,1:2,2:3}
+ {0:1,1:2,2:3}/3
- {0:1,1:1,2:1}
+ {0:1,1:1,2:1}/3
- {}
+ {}/3
 (4 rows)
 
 SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <#> (SELECT NULL::sparsevec)) t2;

test/expected/hnsw_sparsevec_l2.out

@@ -1,15 +1,15 @@
 SET enable_seqscan = off;
-CREATE TABLE t (val sparsevec);
+CREATE TABLE t (val sparsevec(3));
-INSERT INTO t (val) VALUES ('{}'), ('{0:1,1:2,2:3}'), ('{0:1,1:1,2:1}'), (NULL);
+INSERT INTO t (val) VALUES ('{}/3'), ('{0:1,1:2,2:3}/3'), ('{0:1,1:1,2:1}/3'), (NULL);
 CREATE INDEX ON t USING hnsw (val sparsevec_l2_ops);
-INSERT INTO t (val) VALUES ('{0:1,1:2,2:4}');
+INSERT INTO t (val) VALUES ('{0:1,1:2,2:4}/3');
-SELECT * FROM t ORDER BY val <-> '{0:3,1:3,2:3}';
+SELECT * FROM t ORDER BY val <-> '{0:3,1:3,2:3}/3';
-      val      
+       val       
----------------
+-----------------
- {0:1,1:2,2:3}
+ {0:1,1:2,2:3}/3
- {0:1,1:2,2:4}
+ {0:1,1:2,2:4}/3
- {0:1,1:1,2:1}
+ {0:1,1:1,2:1}/3
- {}
+ {}/3
 (4 rows)
 
 SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <-> (SELECT NULL::sparsevec)) t2;
@@ -25,14 +25,14 @@ SELECT COUNT(*) FROM t;
 (1 row)
 
 TRUNCATE t;
-SELECT * FROM t ORDER BY val <-> '{0:3,1:3,2:3}';
+SELECT * FROM t ORDER BY val <-> '{0:3,1:3,2:3}/3';
  val 
 -----
 (0 rows)
 
 DROP TABLE t;
 -- TODO move
-CREATE TABLE t (val sparsevec);
+CREATE TABLE t (val sparsevec(1001));
 INSERT INTO t (val) VALUES (array_fill(1, ARRAY[1001])::vector::sparsevec);
 CREATE INDEX ON t USING hnsw (val sparsevec_l2_ops);
 ERROR:  sparsevec cannot have more than 1000 non-zero elements for hnsw index

test/expected/sparsevec_functions.out

@@ -1,60 +1,62 @@
-SELECT l2_distance('{}'::sparsevec, '{0:3,1:4}');
+SELECT l2_distance('{}/2'::sparsevec, '{0:3,1:4}/2');
  l2_distance 
 -------------
            5
 (1 row)
 
-SELECT l2_distance('{}'::sparsevec, '{1:1}');
+SELECT l2_distance('{}/2'::sparsevec, '{1:1}/2');
  l2_distance 
 -------------
            1
 (1 row)
 
-SELECT '{}'::sparsevec <-> '{0:3,1:4}';
+SELECT '{}/2'::sparsevec <-> '{0:3,1:4}/2';
  ?column? 
 ----------
         5
 (1 row)
 
-SELECT inner_product('{0:1,1:2}'::sparsevec, '{0:2,1:4}');
+SELECT inner_product('{0:1,1:2}/2'::sparsevec, '{0:2,1:4}/2');
  inner_product 
 ---------------
             10
 (1 row)
 
-SELECT sparsevec_negative_inner_product('{0:1,1:2}', '{0:2,1:4}');
+SELECT sparsevec_negative_inner_product('{0:1,1:2}/2', '{0:2,1:4}/2');
  sparsevec_negative_inner_product 
 ----------------------------------
                               -10
 (1 row)
 
-SELECT cosine_distance('{0:1,1:2}'::sparsevec, '{0:2,1:4}');
+SELECT cosine_distance('{0:1,1:2}/2'::sparsevec, '{0:2,1:4}/2');
  cosine_distance 
 -----------------
                0
 (1 row)
 
-SELECT cosine_distance('{0:1,1:2}'::sparsevec, '{}');
+SELECT cosine_distance('{0:1,1:2}/2'::sparsevec, '{}/2');
  cosine_distance 
 -----------------
              NaN
 (1 row)
 
-SELECT cosine_distance('{0:1,1:1}'::sparsevec, '{0:-1,1:-1}');
+SELECT cosine_distance('{0:1,1:1}/2'::sparsevec, '{0:-1,1:-1}/2');
  cosine_distance 
 -----------------
                2
 (1 row)
 
-SELECT cosine_distance('{0:1}'::sparsevec, '{1:2}');
+SELECT cosine_distance('{0:1}/2'::sparsevec, '{1:2}/2');
  cosine_distance 
 -----------------
                1
 (1 row)
 
-SELECT cosine_distance('{}'::sparsevec, '{}');
+SELECT cosine_distance('{}/1'::sparsevec, '{}/1');
  cosine_distance 
 -----------------
              NaN
 (1 row)
 
+SELECT cosine_distance('{0:1}/2'::sparsevec, '{0:1}/3');
+ERROR:  different sparsevec dimensions 2 and 3

test/expected/sparsevec_input.out

@@ -1,64 +1,62 @@
-SELECT '{0:1.5,2:3.5}'::sparsevec;
+SELECT '{0:1.5,2:3.5}/5'::sparsevec;
-   sparsevec   
+    sparsevec    
----------------
+-----------------
- {0:1.5,2:3.5}
+ {0:1.5,2:3.5}/5
 (1 row)
 
-SELECT '{0:1.5,2:3.5}'::sparsevec::vector;
+SELECT '{0:1.5,2:3.5}/5'::sparsevec::vector;
-   vector    
--------------
- [1.5,0,3.5]
-(1 row)
-
-SELECT '{0:1.5,2:3.5}'::sparsevec::vector(5);
      vector      
 -----------------
  [1.5,0,3.5,0,0]
 (1 row)
 
-SELECT '{0:1.5,2:3.5}'::sparsevec::vector(4);
+SELECT '{0:1.5,2:3.5}/5'::sparsevec::vector(5);
-    vector     
+     vector      
----------------
+-----------------
- [1.5,0,3.5,0]
+ [1.5,0,3.5,0,0]
 (1 row)
 
-SELECT '{0:1.5,2:3.5}'::sparsevec::vector(2);
+SELECT '{0:1.5,2:3.5}/5'::sparsevec::vector(4);
-ERROR:  Vector must have at least 3 dimensions
+ERROR:  expected 4 dimensions, not 5
 SELECT '[0,1.5,0,3.5,0]'::vector::sparsevec;
-   sparsevec   
+    sparsevec    
----------------
+-----------------
- {1:1.5,3:3.5}
+ {1:1.5,3:3.5}/5
 (1 row)
 
-SELECT '{0:0,1:1,2:0}'::sparsevec;
+SELECT '{0:0,1:1,2:0}/3'::sparsevec;
  sparsevec 
 -----------
- {1:1}
+ {1:1}/3
 (1 row)
 
-SELECT '{1:1,0:1}'::sparsevec;
+SELECT '{1:1,0:1}/2'::sparsevec;
 ERROR:  indexes must be in ascending order
-LINE 1: SELECT '{1:1,0:1}'::sparsevec;
+LINE 1: SELECT '{1:1,0:1}/2'::sparsevec;
                ^
-SELECT '{}'::sparsevec;
+SELECT '{}/5'::sparsevec;
  sparsevec 
 -----------
- {}
+ {}/5
 (1 row)
 
-SELECT '{}'::sparsevec::vector;
+SELECT '{}/-1'::sparsevec;
-ERROR:  vector must have at least 1 dimension
+ERROR:  sparsevec must have at least 1 dimension
-SELECT '{-1:1}'::sparsevec;
+LINE 1: SELECT '{}/-1'::sparsevec;
+               ^
+SELECT '{}/100001'::sparsevec;
+ERROR:  sparsevec cannot have more than 100000 dimensions
+LINE 1: SELECT '{}/100001'::sparsevec;
+               ^
+SELECT '{}/16001'::sparsevec::vector;
+ERROR:  vector cannot have more than 16000 dimensions
+SELECT '{-1:1}/1'::sparsevec;
 ERROR:  index "-1" is out of range for type sparsevec
-LINE 1: SELECT '{-1:1}'::sparsevec;
+LINE 1: SELECT '{-1:1}/1'::sparsevec;
                ^
-SELECT '{1:1}'::sparsevec;
+SELECT '{1:1}/1'::sparsevec;
- sparsevec 
+ERROR:  index must be less than dimensions
------------
+LINE 1: SELECT '{1:1}/1'::sparsevec;
- {1:1}
+               ^
-(1 row)
+SELECT '{}/1'::sparsevec(2);
-
+ERROR:  expected 2 dimensions, not 1
-SELECT '{}'::sparsevec(2);
-ERROR:  type modifier is not allowed for type "sparsevec"
-LINE 1: SELECT '{}'::sparsevec(2);
-                     ^

test/sql/hnsw_sparsevec_cosine.sql

@@ -1,13 +1,13 @@
 SET enable_seqscan = off;
 
-CREATE TABLE t (val sparsevec);
+CREATE TABLE t (val sparsevec(3));
-INSERT INTO t (val) VALUES ('{}'), ('{0:1,1:2,2:3}'), ('{0:1,1:1,2:1}'), (NULL);
+INSERT INTO t (val) VALUES ('{}/3'), ('{0:1,1:2,2:3}/3'), ('{0:1,1:1,2:1}/3'), (NULL);
 CREATE INDEX ON t USING hnsw (val sparsevec_cosine_ops);
 
-INSERT INTO t (val) VALUES ('{0:1,1:2,2:4}');
+INSERT INTO t (val) VALUES ('{0:1,1:2,2:4}/3');
 
-SELECT * FROM t ORDER BY val <=> '{0:3,1:3,2:3}';
+SELECT * FROM t ORDER BY val <=> '{0:3,1:3,2:3}/3';
-SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> '{}') t2;
+SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> '{}/3') t2;
 SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> (SELECT NULL::sparsevec)) t2;
 
 DROP TABLE t;

test/sql/hnsw_sparsevec_ip.sql

@@ -1,12 +1,12 @@
 SET enable_seqscan = off;
 
-CREATE TABLE t (val sparsevec);
+CREATE TABLE t (val sparsevec(3));
-INSERT INTO t (val) VALUES ('{}'), ('{0:1,1:2,2:3}'), ('{0:1,1:1,2:1}'), (NULL);
+INSERT INTO t (val) VALUES ('{}/3'), ('{0:1,1:2,2:3}/3'), ('{0:1,1:1,2:1}/3'), (NULL);
 CREATE INDEX ON t USING hnsw (val sparsevec_ip_ops);
 
-INSERT INTO t (val) VALUES ('{0:1,1:2,2:4}');
+INSERT INTO t (val) VALUES ('{0:1,1:2,2:4}/3');
 
-SELECT * FROM t ORDER BY val <#> '{0:3,1:3,2:3}';
+SELECT * FROM t ORDER BY val <#> '{0:3,1:3,2:3}/3';
 SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <#> (SELECT NULL::sparsevec)) t2;
 
 DROP TABLE t;

test/sql/hnsw_sparsevec_l2.sql

@@ -1,22 +1,22 @@
 SET enable_seqscan = off;
 
-CREATE TABLE t (val sparsevec);
+CREATE TABLE t (val sparsevec(3));
-INSERT INTO t (val) VALUES ('{}'), ('{0:1,1:2,2:3}'), ('{0:1,1:1,2:1}'), (NULL);
+INSERT INTO t (val) VALUES ('{}/3'), ('{0:1,1:2,2:3}/3'), ('{0:1,1:1,2:1}/3'), (NULL);
 CREATE INDEX ON t USING hnsw (val sparsevec_l2_ops);
 
-INSERT INTO t (val) VALUES ('{0:1,1:2,2:4}');
+INSERT INTO t (val) VALUES ('{0:1,1:2,2:4}/3');
 
-SELECT * FROM t ORDER BY val <-> '{0:3,1:3,2:3}';
+SELECT * FROM t ORDER BY val <-> '{0:3,1:3,2:3}/3';
 SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <-> (SELECT NULL::sparsevec)) t2;
 SELECT COUNT(*) FROM t;
 
 TRUNCATE t;
-SELECT * FROM t ORDER BY val <-> '{0:3,1:3,2:3}';
+SELECT * FROM t ORDER BY val <-> '{0:3,1:3,2:3}/3';
 
 DROP TABLE t;
 
 -- TODO move
-CREATE TABLE t (val sparsevec);
+CREATE TABLE t (val sparsevec(1001));
 INSERT INTO t (val) VALUES (array_fill(1, ARRAY[1001])::vector::sparsevec);
 CREATE INDEX ON t USING hnsw (val sparsevec_l2_ops);
 TRUNCATE t;

test/sql/sparsevec_functions.sql

@@ -1,12 +1,13 @@
-SELECT l2_distance('{}'::sparsevec, '{0:3,1:4}');
+SELECT l2_distance('{}/2'::sparsevec, '{0:3,1:4}/2');
-SELECT l2_distance('{}'::sparsevec, '{1:1}');
+SELECT l2_distance('{}/2'::sparsevec, '{1:1}/2');
-SELECT '{}'::sparsevec <-> '{0:3,1:4}';
+SELECT '{}/2'::sparsevec <-> '{0:3,1:4}/2';
 
-SELECT inner_product('{0:1,1:2}'::sparsevec, '{0:2,1:4}');
+SELECT inner_product('{0:1,1:2}/2'::sparsevec, '{0:2,1:4}/2');
-SELECT sparsevec_negative_inner_product('{0:1,1:2}', '{0:2,1:4}');
+SELECT sparsevec_negative_inner_product('{0:1,1:2}/2', '{0:2,1:4}/2');
 
-SELECT cosine_distance('{0:1,1:2}'::sparsevec, '{0:2,1:4}');
+SELECT cosine_distance('{0:1,1:2}/2'::sparsevec, '{0:2,1:4}/2');
-SELECT cosine_distance('{0:1,1:2}'::sparsevec, '{}');
+SELECT cosine_distance('{0:1,1:2}/2'::sparsevec, '{}/2');
-SELECT cosine_distance('{0:1,1:1}'::sparsevec, '{0:-1,1:-1}');
+SELECT cosine_distance('{0:1,1:1}/2'::sparsevec, '{0:-1,1:-1}/2');
-SELECT cosine_distance('{0:1}'::sparsevec, '{1:2}');
+SELECT cosine_distance('{0:1}/2'::sparsevec, '{1:2}/2');
-SELECT cosine_distance('{}'::sparsevec, '{}');
+SELECT cosine_distance('{}/1'::sparsevec, '{}/1');
+SELECT cosine_distance('{0:1}/2'::sparsevec, '{0:1}/3');

test/sql/sparsevec_input.sql

@@ -1,18 +1,19 @@
-SELECT '{0:1.5,2:3.5}'::sparsevec;
+SELECT '{0:1.5,2:3.5}/5'::sparsevec;
-SELECT '{0:1.5,2:3.5}'::sparsevec::vector;
+SELECT '{0:1.5,2:3.5}/5'::sparsevec::vector;
-SELECT '{0:1.5,2:3.5}'::sparsevec::vector(5);
+SELECT '{0:1.5,2:3.5}/5'::sparsevec::vector(5);
-SELECT '{0:1.5,2:3.5}'::sparsevec::vector(4);
+SELECT '{0:1.5,2:3.5}/5'::sparsevec::vector(4);
-SELECT '{0:1.5,2:3.5}'::sparsevec::vector(2);
 SELECT '[0,1.5,0,3.5,0]'::vector::sparsevec;
 
-SELECT '{0:0,1:1,2:0}'::sparsevec;
+SELECT '{0:0,1:1,2:0}/3'::sparsevec;
 
-SELECT '{1:1,0:1}'::sparsevec;
+SELECT '{1:1,0:1}/2'::sparsevec;
 
-SELECT '{}'::sparsevec;
+SELECT '{}/5'::sparsevec;
-SELECT '{}'::sparsevec::vector;
+SELECT '{}/-1'::sparsevec;
+SELECT '{}/100001'::sparsevec;
+SELECT '{}/16001'::sparsevec::vector;
 
-SELECT '{-1:1}'::sparsevec;
+SELECT '{-1:1}/1'::sparsevec;
-SELECT '{1:1}'::sparsevec;
+SELECT '{1:1}/1'::sparsevec;
 
-SELECT '{}'::sparsevec(2);
+SELECT '{}/1'::sparsevec(2);