Removed dimensions from sparsevec

.github/workflows/build.yml

@@ -123,6 +123,6 @@ jobs:
       - uses: ankane/setup-postgres-valgrind@v1
         with:
           postgres-version: 16
-      - run: make OPTFLAGS=""
+      - run: make
       - run: sudo --preserve-env=PG_CONFIG make install
       - run: make installcheck

README.md

@@ -419,103 +419,6 @@ Use [partitioning](https://www.postgresql.org/docs/current/ddl-partitioning.html
 CREATE TABLE items (embedding vector(3), category_id int) PARTITION BY LIST(category_id);
 ```
 
-## Half Vectors
-
-*Unreleased*
-
-Use the `halfvec` type to store half-precision vectors
-
-```sql
-CREATE TABLE items (id bigserial PRIMARY KEY, embedding halfvec(3));
-```
-
-## Half Indexing
-
-*Unreleased*
-
-Index vectors at half precision for smaller indexes and faster build times
-
-```sql
-CREATE INDEX ON items USING hnsw ((embedding::halfvec(3)) halfvec_l2_ops);
-```
-
-Get the nearest neighbors
-
-```sql
-SELECT * FROM items ORDER BY embedding::halfvec(3) <-> '[1,2,3]' LIMIT 5;
-```
-
-## Binary Vectors
-
-Use the `bit` type to store binary vectors ([example](https://github.com/pgvector/pgvector-python/blob/master/examples/hash_image_search.py))
-
-```sql
-CREATE TABLE items (id bigserial PRIMARY KEY, embedding bit(3));
-INSERT INTO items (embedding) VALUES ('000'), ('111');
-```
-
-Get the nearest neighbors by Hamming distance
-
-```sql
-SELECT * FROM items ORDER BY bit_count(embedding # '101') LIMIT 5;
-```
-
-Or (unreleased)
-
-```sql
-SELECT * FROM items ORDER BY embedding <~> '101' LIMIT 5;
-```
-
-Also supports Jaccard distance (`<%>`)
-
-## Binary Quantization
-
-*Unreleased*
-
-Use expression indexing for binary quantization
-
-```sql
-CREATE INDEX ON items USING hnsw ((quantize_binary(embedding)::bit(3)) bit_hamming_ops);
-```
-
-Get the nearest neighbors by Hamming distance
-
-```sql
-SELECT * FROM items ORDER BY quantize_binary(embedding)::bit(3) <~> quantize_binary('[1,-2,3]') LIMIT 5;
-```
-
-Re-rank by the original vectors for better recall
-
-```sql
-SELECT * FROM (
-    SELECT * FROM items ORDER BY quantize_binary(embedding)::bit(3) <~> quantize_binary('[1,-2,3]') LIMIT 20
-) ORDER BY embedding <=> '[1,-2,3]' LIMIT 5;
-```
-
-## Sparse Vectors
-
-*Unreleased*
-
-Use the `sparsevec` type to store sparse vectors
-
-```sql
-CREATE TABLE items (id bigserial PRIMARY KEY, embedding sparsevec(5));
-```
-
-Insert vectors
-
-```sql
-INSERT INTO items (embedding) VALUES ('{1:1,3:2,5:3}/5'), ('{1:4,3:5,5:6}/5');
-```
-
-Note: The format is `{index1:value1,index2:value2,...}/dimensions` and indices start at 1 like SQL arrays
-
-Get the nearest neighbors by L2 distance
-
-```sql
-SELECT * FROM items ORDER BY embedding <-> '{1:3,3:1,5:2}/5' LIMIT 5;
-```
-
 ## Hybrid Search
 
 Use together with Postgres [full-text search](https://www.postgresql.org/docs/current/textsearch-intro.html) for hybrid search.
@@ -732,6 +635,18 @@ and query with:
 SELECT * FROM items ORDER BY embedding::vector(3) <-> '[3,1,2]' LIMIT 5;
 ```
 
+#### Are binary vectors supported?
+
+You can store binary vectors and perform exact nearest neighbor search by Hamming distance in Postgres without an extension ([example](https://github.com/pgvector/pgvector-python/blob/master/examples/hash_image_search.py)).
+
+```tsql
+CREATE TABLE items (id bigserial PRIMARY KEY, embedding bit(3));
+INSERT INTO items (embedding) VALUES (B'000'), (B'111');
+SELECT * FROM items ORDER BY bit_count(embedding # B'101') LIMIT 5;
+```
+
+Indexing is not currently supported.
+
 #### Do indexes need to fit into memory?
 
 No, but like other index types, you’ll likely see better performance if they do. You can get the size of an index with:
@@ -886,7 +801,7 @@ jaccard_distance(bit, bit) → double precision | Jaccard distance | unreleased
 
 ### Sparsevec Type
 
-Each sparse vector takes `8 * non-zero elements + 16` bytes of storage. Each element is a single-precision floating-point number, and all elements must be finite (no `NaN`, `Infinity` or `-Infinity`). Sparse vectors can have up to 16,000 non-zero elements.
+Each sparse vector takes `8 * non-zero elements + 16` bytes of storage. Each element is a single-precision floating-point number, and all elements must be finite (no `NaN`, `Infinity` or `-Infinity`). Sparse vectors can have up to 100,000 dimensions.
 
 ### Sparsevec Operators
 

sql/vector--0.6.2--0.7.0.sql

@@ -173,9 +173,6 @@ 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;
 
@@ -185,7 +182,6 @@ 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

sql/vector.sql

@@ -480,9 +480,6 @@ 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;
 
@@ -492,7 +489,6 @@ 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

src/halfvec.c

@@ -22,10 +22,6 @@
 #define TYPALIGN_INT 'i'
 #endif
 
-#ifdef F16C_SUPPORT
-#include <immintrin.h>
-#endif
-
 /*
  * Check if half is NaN
  */
@@ -103,10 +99,7 @@ pq_sendhalf(StringInfo buf, half h)
 float
 HalfToFloat4(half num)
 {
-#if defined(F16C_SUPPORT) && !defined(_MSC_VER)
-	/* TODO Use instrinsics for Windows */
-	return _cvtsh_ss(num);
-#elif defined(FLT16_SUPPORT)
+#ifdef FLT16_SUPPORT
 	return (float) num;
 #else
 	/* TODO Improve performance */
@@ -137,7 +130,7 @@ HalfToFloat4(half num)
 	/* Sign */
 	result = (bin & 0x8000) << 16;
 
-	if (unlikely(exponent == 31))
+	if (exponent == 31)
 	{
 		if (mantissa == 0)
 		{
@@ -148,9 +141,10 @@ HalfToFloat4(half num)
 		{
 			/* NaN */
 			result |= 0x7FC00000;
+			result |= mantissa << 13;
 		}
 	}
-	else if (unlikely(exponent == 0))
+	else if (exponent == 0)
 	{
 		/* Subnormal */
 		if (mantissa != 0)
@@ -170,16 +164,16 @@ HalfToFloat4(half num)
 			}
 
 			result |= (exponent + 127) << 23;
+			result |= mantissa << 13;
 		}
 	}
 	else
 	{
 		/* Normal */
 		result |= (exponent - 15 + 127) << 23;
+		result |= mantissa << 13;
 	}
 
-	result |= mantissa << 13;
-
 	swapfloat.i = result;
 	return swapfloat.f;
 #endif
@@ -191,10 +185,7 @@ HalfToFloat4(half num)
 half
 Float4ToHalfUnchecked(float num)
 {
-#if defined(F16C_SUPPORT) && !defined(_MSC_VER)
-	/* TODO Use instrinsics for Windows */
-	return _cvtss_sh(num, 0);
-#elif defined(FLT16_SUPPORT)
+#ifdef FLT16_SUPPORT
 	return (_Float16) num;
 #else
 	/* TODO Improve performance */
@@ -444,8 +435,6 @@ halfvec_in(PG_FUNCTION_ARGS)
 
 	while (pt != NULL && *stringEnd != ']')
 	{
-		float		val;
-
 		if (dim == HALFVEC_MAX_DIM)
 			ereport(ERROR,
 					(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
@@ -461,24 +450,15 @@ halfvec_in(PG_FUNCTION_ARGS)
 					 errmsg("invalid input syntax for type halfvec: \"%s\"", lit)));
 
 		/* Use strtof like float4in to avoid a double-rounding problem */
-		errno = 0;
-		val = strtof(pt, &stringEnd);
+		x[dim] = Float4ToHalf(strtof(pt, &stringEnd));
+		CheckElement(x[dim]);
+		dim++;
 
 		if (stringEnd == pt)
 			ereport(ERROR,
 					(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
 					 errmsg("invalid input syntax for type halfvec: \"%s\"", lit)));
 
-		x[dim] = Float4ToHalfUnchecked(val);
-
-		if ((errno == ERANGE && (isinf(val) || val == 0)) || (HalfIsInf(x[dim]) && !isinf(val)) || (HalfIsZero(x[dim]) && val != 0))
-			ereport(ERROR,
-					(errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
-					 errmsg("\"%s\" is out of range for type halfvec", pt)));
-
-		CheckElement(x[dim]);
-		dim++;
-
 		while (halfvec_isspace(*stringEnd))
 			stringEnd++;
 
@@ -802,56 +782,6 @@ vector_to_halfvec(PG_FUNCTION_ARGS)
 	PG_RETURN_POINTER(result);
 }
 
-/*
- * Get the L2 squared distance between half vectors
- */
-static double
-l2_distance_squared_internal(HalfVector * a, HalfVector * b)
-{
-	half	   *ax = a->x;
-	half	   *bx = b->x;
-	float		distance = 0.0;
-
-#if defined(F16C_SUPPORT) && defined(__FMA__)
-	int			i;
-	float		s[8];
-	int			count = (a->dim / 8) * 8;
-	__m256		dist = _mm256_setzero_ps();
-
-	for (i = 0; i < count; i += 8)
-	{
-		__m128i		axi = _mm_loadu_si128((__m128i *) (ax + i));
-		__m128i		bxi = _mm_loadu_si128((__m128i *) (bx + i));
-		__m256		axs = _mm256_cvtph_ps(axi);
-		__m256		bxs = _mm256_cvtph_ps(bxi);
-		__m256		diff = _mm256_sub_ps(axs, bxs);
-
-		dist = _mm256_fmadd_ps(diff, diff, dist);
-	}
-
-	_mm256_storeu_ps(s, dist);
-
-	distance = s[0] + s[1] + s[2] + s[3] + s[4] + s[5] + s[6] + s[7];
-
-	for (; i < a->dim; i++)
-	{
-		float		diff = HalfToFloat4(ax[i]) - HalfToFloat4(bx[i]);
-
-		distance += diff * diff;
-	}
-#else
-	/* Auto-vectorized */
-	for (int i = 0; i < a->dim; i++)
-	{
-		float		diff = HalfToFloat4(ax[i]) - HalfToFloat4(bx[i]);
-
-		distance += diff * diff;
-	}
-#endif
-
-	return (double) distance;
-}
-
 /*
  * Get the L2 distance between half vectors
  */
@@ -861,10 +791,21 @@ halfvec_l2_distance(PG_FUNCTION_ARGS)
 {
 	HalfVector *a = PG_GETARG_HALFVEC_P(0);
 	HalfVector *b = PG_GETARG_HALFVEC_P(1);
+	half	   *ax = a->x;
+	half	   *bx = b->x;
+	float		distance = 0.0;
 
 	CheckDims(a, b);
 
-	PG_RETURN_FLOAT8(sqrt(l2_distance_squared_internal(a, b)));
+	/* Auto-vectorized */
+	for (int i = 0; i < a->dim; i++)
+	{
+		float		diff = HalfToFloat4(ax[i]) - HalfToFloat4(bx[i]);
+
+		distance += diff * diff;
+	}
+
+	PG_RETURN_FLOAT8(sqrt((double) distance));
 }
 
 /*
@@ -876,51 +817,21 @@ halfvec_l2_squared_distance(PG_FUNCTION_ARGS)
 {
 	HalfVector *a = PG_GETARG_HALFVEC_P(0);
 	HalfVector *b = PG_GETARG_HALFVEC_P(1);
-
-	CheckDims(a, b);
-
-	PG_RETURN_FLOAT8(l2_distance_squared_internal(a, b));
-}
-
-/*
- * Get the inner product of two half vectors
- */
-static double
-inner_product_internal(HalfVector * a, HalfVector * b)
-{
 	half	   *ax = a->x;
 	half	   *bx = b->x;
 	float		distance = 0.0;
 
-#if defined(F16C_SUPPORT) && defined(__FMA__)
-	int			i;
-	float		s[8];
-	int			count = (a->dim / 8) * 8;
-	__m256		dist = _mm256_setzero_ps();
+	CheckDims(a, b);
 
-	for (i = 0; i < count; i += 8)
-	{
-		__m128i		axi = _mm_loadu_si128((__m128i *) (ax + i));
-		__m128i		bxi = _mm_loadu_si128((__m128i *) (bx + i));
-		__m256		axs = _mm256_cvtph_ps(axi);
-		__m256		bxs = _mm256_cvtph_ps(bxi);
-
-		dist = _mm256_fmadd_ps(axs, bxs, dist);
-	}
-
-	_mm256_storeu_ps(s, dist);
-
-	distance = s[0] + s[1] + s[2] + s[3] + s[4] + s[5] + s[6] + s[7];
-
-	for (; i < a->dim; i++)
-		distance += HalfToFloat4(ax[i]) * HalfToFloat4(bx[i]);
-#else
 	/* Auto-vectorized */
 	for (int i = 0; i < a->dim; i++)
-		distance += HalfToFloat4(ax[i]) * HalfToFloat4(bx[i]);
-#endif
+	{
+		float		diff = HalfToFloat4(ax[i]) - HalfToFloat4(bx[i]);
 
-	return (double) distance;
+		distance += diff * diff;
+	}
+
+	PG_RETURN_FLOAT8((double) distance);
 }
 
 /*
@@ -932,10 +843,17 @@ halfvec_inner_product(PG_FUNCTION_ARGS)
 {
 	HalfVector *a = PG_GETARG_HALFVEC_P(0);
 	HalfVector *b = PG_GETARG_HALFVEC_P(1);
+	half	   *ax = a->x;
+	half	   *bx = b->x;
+	float		distance = 0.0;
 
 	CheckDims(a, b);
 
-	PG_RETURN_FLOAT8(inner_product_internal(a, b));
+	/* Auto-vectorized */
+	for (int i = 0; i < a->dim; i++)
+		distance += HalfToFloat4(ax[i]) * HalfToFloat4(bx[i]);
+
+	PG_RETURN_FLOAT8((double) distance);
 }
 
 /*
@@ -947,10 +865,17 @@ halfvec_negative_inner_product(PG_FUNCTION_ARGS)
 {
 	HalfVector *a = PG_GETARG_HALFVEC_P(0);
 	HalfVector *b = PG_GETARG_HALFVEC_P(1);
+	half	   *ax = a->x;
+	half	   *bx = b->x;
+	float		distance = 0.0;
 
 	CheckDims(a, b);
 
-	PG_RETURN_FLOAT8(-inner_product_internal(a, b));
+	/* Auto-vectorized */
+	for (int i = 0; i < a->dim; i++)
+		distance += HalfToFloat4(ax[i]) * HalfToFloat4(bx[i]);
+
+	PG_RETURN_FLOAT8((double) distance * -1);
 }
 
 /*

src/halfvec.h

@@ -7,10 +7,7 @@
 
 #include "vector.h"
 
-/* F16C has better performance than _Float16 (on x86-64) */
-#if defined(__F16C__) || defined(_MSC_VER)
-#define F16C_SUPPORT
-#elif defined(__FLT16_MAX__)
+#ifdef __FLT16_MAX__
 #define FLT16_SUPPORT
 #endif
 
@@ -18,6 +15,7 @@
 #define half _Float16
 #define HALF_MAX FLT16_MAX
 #else
+/* TODO #pragma message("")? */
 #define half uint16
 #define HALF_MAX 65504
 #endif

src/hnswbuild.c

@@ -681,8 +681,6 @@ GetMaxDimensions(HnswType type)
 		maxDimensions *= 2;
 	else if (type == HNSW_TYPE_BIT)
 		maxDimensions *= 32;
-	else if (type == HNSW_TYPE_SPARSEVEC)
-		maxDimensions = INT_MAX;
 
 	return maxDimensions;
 }
@@ -693,8 +691,6 @@ 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;
@@ -705,14 +701,17 @@ InitBuildState(HnswBuildState * buildstate, Relation heap, Relation index, Index
 	buildstate->efConstruction = HnswGetEfConstruction(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->type != HNSW_TYPE_SPARSEVEC)
+	{
+		int			maxDimensions = GetMaxDimensions(buildstate->type);
 
-	if (buildstate->dimensions > maxDimensions)
-		elog(ERROR, "column cannot have more than %d dimensions for hnsw index", maxDimensions);
+		if (buildstate->dimensions < 0)
+			elog(ERROR, "column does not have dimensions");
+
+		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/hnswscan.c

@@ -158,10 +158,6 @@ hnswgettuple(IndexScanDesc scan, ScanDirection dir)
 		UnlockPage(scan->indexRelation, HNSW_SCAN_LOCK, ShareLock);
 
 		so->first = false;
-
-#if defined(HNSW_MEMORY) && PG_VERSION_NUM >= 130000
-		elog(INFO, "memory: %zu MB", MemoryContextMemAllocated(so->tmpCtx, false) / (1024 * 1024));
-#endif
 	}
 
 	while (list_length(so->w) > 0)

src/hnswutils.c

@@ -163,9 +163,9 @@ HnswGetType(Relation index)
 	Oid			typid = TupleDescAttr(index->rd_att, 0)->atttypid;
 	HeapTuple	tuple;
 	Form_pg_type type;
-	HnswType	result;
+	int			result;
 
-	if (typid == BITOID)
+	if (typid == BITOID || typid == VARBITOID)
 		return HNSW_TYPE_BIT;
 
 	tuple = SearchSysCache1(TYPEOID, ObjectIdGetDatum(typid));
@@ -180,10 +180,7 @@ HnswGetType(Relation index)
 	else if (strcmp(NameStr(type->typname), "sparsevec") == 0)
 		result = HNSW_TYPE_SPARSEVEC;
 	else
-	{
-		ReleaseSysCache(tuple);
-		elog(ERROR, "type not supported for hnsw index");
-	}
+		elog(ERROR, "Unsupported type");
 
 	ReleaseSysCache(tuple);
 
@@ -232,7 +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->dim, v->nnz);
+			SparseVector *result = InitSparseVector(v->nnz);
 			float	   *vx = SPARSEVEC_VALUES(v);
 			float	   *rx = SPARSEVEC_VALUES(result);
 

src/sparsevec.c

@@ -18,86 +18,35 @@
 #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, int dim)
+CheckNnz(int nnz)
 {
 	if (nnz < 0)
 		ereport(ERROR,
 				(errcode(ERRCODE_DATA_EXCEPTION),
-				 errmsg("sparsevec cannot have negative number of elements")));
+				 errmsg("sparsevec must have at least one element")));
 
 	if (nnz > SPARSEVEC_MAX_NNZ)
 		ereport(ERROR,
 				(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
-				 errmsg("sparsevec cannot have more than %d non-zero elements", SPARSEVEC_MAX_NNZ)));
-
-	if (nnz > dim)
-		ereport(ERROR,
-				(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
-				 errmsg("sparsevec cannot have more elements than dimensions")));
+				 errmsg("sparsevec cannot have more elements than non-zero elements")));
 }
 
 /*
  * Ensure valid index
  */
 static inline void
-CheckIndex(int32 *indices, int i, int dim)
+CheckIndex(int32 *indices, int i)
 {
 	int32		index = indices[i];
 
-	if (index < 1)
+	if (index < 0)
 		ereport(ERROR,
 				(errcode(ERRCODE_DATA_EXCEPTION),
-				 errmsg("index must be greater than zero")));
-
-	if (index > dim)
-		ereport(ERROR,
-				(errcode(ERRCODE_DATA_EXCEPTION),
-				 errmsg("index must be less than or equal to dimensions")));
+				 errmsg("index must not be negative")));
 
 	if (i > 0)
 	{
@@ -134,7 +83,7 @@ CheckElement(float value)
  * Allocate and initialize a new sparse vector
  */
 SparseVector *
-InitSparseVector(int dim, int nnz)
+InitSparseVector(int nnz)
 {
 	SparseVector *result;
 	int			size;
@@ -142,7 +91,6 @@ InitSparseVector(int dim, int nnz)
 	size = SPARSEVEC_SIZE(nnz);
 	result = (SparseVector *) palloc0(size);
 	SET_VARSIZE(result, size);
-	result->dim = dim;
 	result->nnz = nnz;
 
 	return result;
@@ -172,8 +120,6 @@ 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;
@@ -245,7 +191,7 @@ sparsevec_in(PG_FUNCTION_ARGS)
 					(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
 					 errmsg("invalid input syntax for type sparsevec: \"%s\"", lit)));
 
-		if (errno == ERANGE || index < 1 || index > INT_MAX)
+		if (errno == ERANGE || index < 0 || index > INT_MAX)
 			ereport(ERROR,
 					(errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
 					 errmsg("index \"%ld\" is out of range for type sparsevec", index)));
@@ -307,24 +253,6 @@ 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++;
@@ -333,21 +261,18 @@ sparsevec_in(PG_FUNCTION_ARGS)
 		ereport(ERROR,
 				(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
 				 errmsg("malformed sparsevec literal: \"%s\"", lit),
-				 errdetail("Junk after closing.")));
+				 errdetail("Junk after closing right brace.")));
 
 	pfree(litcopy);
 
-	CheckDim(dim);
-	CheckExpectedDim(typmod, dim);
-
-	result = InitSparseVector(dim, nnz);
+	result = InitSparseVector(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, dim);
+		CheckIndex(result->indices, i);
 		CheckElement(rvalues[i]);
 	}
 
@@ -392,11 +317,9 @@ sparsevec_out(PG_FUNCTION_ARGS)
 	 *
 	 * nnz - 1 bytes for ,
 	 *
-	 * 10 bytes for dimensions
-	 *
-	 * 4 bytes for {, }, /, and \0
+	 * 3 bytes for {, }, and \0
 	 */
-	buf = (char *) palloc((11 + FLOAT_SHORTEST_DECIMAL_LEN) * sparsevec->nnz + 13);
+	buf = (char *) palloc((11 + FLOAT_SHORTEST_DECIMAL_LEN) * sparsevec->nnz + 2);
 	ptr = buf;
 
 	AppendChar(ptr, '{');
@@ -412,45 +335,12 @@ 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
  */
@@ -459,33 +349,30 @@ 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));
 
-	CheckDim(dim);
-	CheckNnz(nnz, dim);
-	CheckExpectedDim(typmod, dim);
+	CheckNnz(nnz);
 
-	if (unused != 0)
+	if (unused != 0 || unused2 != 0)
 		ereport(ERROR,
 				(errcode(ERRCODE_DATA_EXCEPTION),
 				 errmsg("expected unused to be 0, not %d", unused)));
 
-	result = InitSparseVector(dim, nnz);
+	result = InitSparseVector(nnz);
 	values = SPARSEVEC_VALUES(result);
 
 	for (int i = 0; i < nnz; i++)
 	{
 		result->indices[i] = pq_getmsgint(buf, sizeof(int32));
-		CheckIndex(result->indices, i, dim);
+		CheckIndex(result->indices, i);
 	}
 
 	for (int i = 0; i < nnz; i++)
@@ -509,9 +396,9 @@ 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++)
@@ -522,16 +409,12 @@ 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);
 }
@@ -544,23 +427,20 @@ 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++;
 	}
 
-	result = InitSparseVector(dim, nnz);
+	CheckNnz(nnz);
+	result = InitSparseVector(nnz);
 	values = SPARSEVEC_VALUES(result);
 	for (int i = 0; i < dim; i++)
 	{
@@ -570,7 +450,7 @@ vector_to_sparsevec(PG_FUNCTION_ARGS)
 			if (j == nnz)
 				elog(ERROR, "safety check failed");
 
-			result->indices[j] = i + 1;
+			result->indices[j] = i;
 			values[j] = vec->x[i];
 			j++;
 		}
@@ -580,7 +460,7 @@ vector_to_sparsevec(PG_FUNCTION_ARGS)
 }
 
 /*
- * Get the L2 squared distance between sparse vectors
+ *  Get the L2 squared distance between sparse vectors
  */
 static double
 l2_distance_squared_internal(SparseVector * a, SparseVector * b)
@@ -637,8 +517,6 @@ 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)));
 }
 
@@ -653,8 +531,6 @@ 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));
 }
 
@@ -704,8 +580,6 @@ 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));
 }
 
@@ -719,8 +593,6 @@ 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));
 }
 
@@ -739,8 +611,6 @@ sparsevec_cosine_distance(PG_FUNCTION_ARGS)
 	float		normb = 0.0;
 	double		similarity;
 
-	CheckDims(a, b);
-
 	similarity = inner_product_internal(a, b);
 
 	/* Auto-vectorized */

src/sparsevec.h

@@ -1,8 +1,7 @@
 #ifndef SPARSEVEC_H
 #define SPARSEVEC_H
 
-#define SPARSEVEC_MAX_DIM 100000
-#define SPARSEVEC_MAX_NNZ 16000
+#define SPARSEVEC_MAX_NNZ 100000
 
 /* Ensure values are aligned */
 #define SPARSEVEC_SIZE(_nnz)		(offsetof(SparseVector, indices) + MAXALIGN((_nnz) * sizeof(int32)) + (_nnz * sizeof(float)))
@@ -14,12 +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 dim, int nnz);
+SparseVector *InitSparseVector(int nnz);
 
 #endif

src/vector.c

@@ -1236,15 +1236,23 @@ sparsevec_to_vector(PG_FUNCTION_ARGS)
 	SparseVector *svec = PG_GETARG_SPARSEVEC_P(0);
 	int32		typmod = PG_GETARG_INT32(1);
 	Vector	   *result;
-	int			dim = svec->dim;
+	int			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++)
-		result->x[svec->indices[i] - 1] = values[i];
+		result->x[svec->indices[i]] = values[i];
 
 	PG_RETURN_POINTER(result);
 }

test/expected/bit_functions.out

@@ -1,104 +1,64 @@
-SELECT hamming_distance('111', '111');
+SELECT hamming_distance(B'111', B'111');
  hamming_distance 
 ------------------
                 0
 (1 row)
 
-SELECT hamming_distance('111', '110');
+SELECT hamming_distance(B'111', B'110');
  hamming_distance 
 ------------------
                 1
 (1 row)
 
-SELECT hamming_distance('111', '100');
+SELECT hamming_distance(B'111', B'100');
  hamming_distance 
 ------------------
                 2
 (1 row)
 
-SELECT hamming_distance('111', '000');
+SELECT hamming_distance(B'111', B'000');
  hamming_distance 
 ------------------
                 3
 (1 row)
 
-SELECT hamming_distance('10101010101010101010', '01010101010101010101');
- hamming_distance 
-------------------
-               20
-(1 row)
-
-SELECT hamming_distance('', '');
- hamming_distance 
-------------------
-                0
-(1 row)
-
-SELECT hamming_distance('111', '00');
+SELECT hamming_distance(B'111', B'00');
 ERROR:  different bit lengths 3 and 2
-SELECT hamming_distance('111', '000'::varbit(4));
- hamming_distance 
-------------------
-                3
-(1 row)
-
-SELECT hamming_distance('111', '0000'::varbit(4));
-ERROR:  different bit lengths 3 and 4
-SELECT jaccard_distance('1111', '1111');
+SELECT jaccard_distance(B'1111', B'1111');
  jaccard_distance 
 ------------------
                 0
 (1 row)
 
-SELECT jaccard_distance('1111', '1110');
+SELECT jaccard_distance(B'1111', B'1110');
  jaccard_distance 
 ------------------
              0.25
 (1 row)
 
-SELECT jaccard_distance('1111', '1100');
+SELECT jaccard_distance(B'1111', B'1100');
  jaccard_distance 
 ------------------
               0.5
 (1 row)
 
-SELECT jaccard_distance('1111', '1000');
+SELECT jaccard_distance(B'1111', B'1000');
  jaccard_distance 
 ------------------
              0.75
 (1 row)
 
-SELECT jaccard_distance('1111', '0000');
+SELECT jaccard_distance(B'1111', B'0000');
  jaccard_distance 
 ------------------
                 1
 (1 row)
 
-SELECT jaccard_distance('1100', '1000');
+SELECT jaccard_distance(B'1100', B'1000');
  jaccard_distance 
 ------------------
               0.5
 (1 row)
 
-SELECT jaccard_distance('10101010101010101010', '01010101010101010101');
- jaccard_distance 
-------------------
-                1
-(1 row)
-
-SELECT jaccard_distance('', '');
- jaccard_distance 
-------------------
-                1
-(1 row)
-
-SELECT jaccard_distance('1111', '000');
+SELECT jaccard_distance(B'1111', B'000');
 ERROR:  different bit lengths 4 and 3
-SELECT jaccard_distance('1111', '0000'::varbit(5));
- jaccard_distance 
-------------------
-                1
-(1 row)
-
-SELECT jaccard_distance('1111', '00000'::varbit(5));
-ERROR:  different bit lengths 4 and 5

test/expected/halfvec_input.out

@@ -51,21 +51,17 @@ SELECT '[65519,-65519]'::halfvec;
 (1 row)
 
 SELECT '[65520,-65520]'::halfvec;
-ERROR:  "65520" is out of range for type halfvec
+ERROR:  value out of range: overflow
 LINE 1: SELECT '[65520,-65520]'::halfvec;
                ^
 SELECT '[1e-8,-1e-8]'::halfvec;
-ERROR:  "1e-8" is out of range for type halfvec
+ERROR:  value out of range: underflow
 LINE 1: SELECT '[1e-8,-1e-8]'::halfvec;
                ^
 SELECT '[4e38,1]'::halfvec;
-ERROR:  "4e38" is out of range for type halfvec
+ERROR:  infinite value not allowed in halfvec
 LINE 1: SELECT '[4e38,1]'::halfvec;
                ^
-SELECT '[1e-46,1]'::halfvec;
-ERROR:  "1e-46" is out of range for type halfvec
-LINE 1: SELECT '[1e-46,1]'::halfvec;
-               ^
 SELECT '[1,2,3'::halfvec;
 ERROR:  malformed halfvec literal: "[1,2,3"
 LINE 1: SELECT '[1,2,3'::halfvec;

test/expected/hnsw_bit_hamming.out

@@ -19,9 +19,3 @@ SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <~> (SELECT NULL::bit)) t2;
 (1 row)
 
 DROP TABLE t;
--- TODO move
-CREATE TABLE t (val varbit(3));
-CREATE INDEX ON t USING hnsw (val bit_hamming_ops);
-ERROR:  type not supported for hnsw index
-CREATE INDEX ON t USING hnsw ((val::bit(3)) bit_hamming_ops);
-DROP TABLE t;

test/expected/hnsw_sparsevec_cosine.out

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

test/expected/hnsw_sparsevec_ip.out

@@ -1,15 +1,15 @@
 SET enable_seqscan = off;
-CREATE TABLE t (val sparsevec(3));
-INSERT INTO t (val) VALUES ('{}/3'), ('{1:1,2:2,3:3}/3'), ('{1:1,2:1,3:1}/3'), (NULL);
+CREATE TABLE t (val sparsevec);
+INSERT INTO t (val) VALUES ('{}'), ('{0:1,1:2,2:3}'), ('{0:1,1:1,2:1}'), (NULL);
 CREATE INDEX ON t USING hnsw (val sparsevec_ip_ops);
-INSERT INTO t (val) VALUES ('{1:1,2:2,3:4}/3');
-SELECT * FROM t ORDER BY val <#> '{1:3,2:3,3:3}/3';
-       val       
------------------
- {1:1,2:2,3:4}/3
- {1:1,2:2,3:3}/3
- {1:1,2:1,3:1}/3
- {}/3
+INSERT INTO t (val) VALUES ('{0:1,1:2,2:4}');
+SELECT * FROM t ORDER BY val <#> '{0:3,1:3,2:3}';
+      val      
+---------------
+ {0:1,1:2,2:4}
+ {0:1,1:2,2:3}
+ {0:1,1:1,2:1}
+ {}
 (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(3));
-INSERT INTO t (val) VALUES ('{}/3'), ('{1:1,2:2,3:3}/3'), ('{1:1,2:1,3:1}/3'), (NULL);
+CREATE TABLE t (val sparsevec);
+INSERT INTO t (val) VALUES ('{}'), ('{0:1,1:2,2:3}'), ('{0:1,1:1,2:1}'), (NULL);
 CREATE INDEX ON t USING hnsw (val sparsevec_l2_ops);
-INSERT INTO t (val) VALUES ('{1:1,2:2,3:4}/3');
-SELECT * FROM t ORDER BY val <-> '{1:3,2:3,3:3}/3';
-       val       
------------------
- {1:1,2:2,3:3}/3
- {1:1,2:2,3:4}/3
- {1:1,2:1,3:1}/3
- {}/3
+INSERT INTO t (val) VALUES ('{0:1,1:2,2:4}');
+SELECT * FROM t ORDER BY val <-> '{0:3,1:3,2:3}';
+      val      
+---------------
+ {0:1,1:2,2:3}
+ {0:1,1:2,2:4}
+ {0:1,1:1,2:1}
+ {}
 (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 <-> '{1:3,2:3,3:3}/3';
+SELECT * FROM t ORDER BY val <-> '{0:3,1:3,2:3}';
  val 
 -----
 (0 rows)
 
 DROP TABLE t;
 -- TODO move
-CREATE TABLE t (val sparsevec(1001));
+CREATE TABLE t (val sparsevec);
 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

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

test/expected/sparsevec_input.out

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

test/sql/bit_functions.sql

@@ -1,21 +1,13 @@
-SELECT hamming_distance('111', '111');
-SELECT hamming_distance('111', '110');
-SELECT hamming_distance('111', '100');
-SELECT hamming_distance('111', '000');
-SELECT hamming_distance('10101010101010101010', '01010101010101010101');
-SELECT hamming_distance('', '');
-SELECT hamming_distance('111', '00');
-SELECT hamming_distance('111', '000'::varbit(4));
-SELECT hamming_distance('111', '0000'::varbit(4));
+SELECT hamming_distance(B'111', B'111');
+SELECT hamming_distance(B'111', B'110');
+SELECT hamming_distance(B'111', B'100');
+SELECT hamming_distance(B'111', B'000');
+SELECT hamming_distance(B'111', B'00');
 
-SELECT jaccard_distance('1111', '1111');
-SELECT jaccard_distance('1111', '1110');
-SELECT jaccard_distance('1111', '1100');
-SELECT jaccard_distance('1111', '1000');
-SELECT jaccard_distance('1111', '0000');
-SELECT jaccard_distance('1100', '1000');
-SELECT jaccard_distance('10101010101010101010', '01010101010101010101');
-SELECT jaccard_distance('', '');
-SELECT jaccard_distance('1111', '000');
-SELECT jaccard_distance('1111', '0000'::varbit(5));
-SELECT jaccard_distance('1111', '00000'::varbit(5));
+SELECT jaccard_distance(B'1111', B'1111');
+SELECT jaccard_distance(B'1111', B'1110');
+SELECT jaccard_distance(B'1111', B'1100');
+SELECT jaccard_distance(B'1111', B'1000');
+SELECT jaccard_distance(B'1111', B'0000');
+SELECT jaccard_distance(B'1100', B'1000');
+SELECT jaccard_distance(B'1111', B'000');

test/sql/halfvec_input.sql

@@ -11,7 +11,6 @@ SELECT '[65519,-65519]'::halfvec;
 SELECT '[65520,-65520]'::halfvec;
 SELECT '[1e-8,-1e-8]'::halfvec;
 SELECT '[4e38,1]'::halfvec;
-SELECT '[1e-46,1]'::halfvec;
 SELECT '[1,2,3'::halfvec;
 SELECT '[1,2,3]9'::halfvec;
 SELECT '1,2,3'::halfvec;

test/sql/hnsw_bit_hamming.sql

@@ -10,9 +10,3 @@ SELECT * FROM t ORDER BY val <~> B'111';
 SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <~> (SELECT NULL::bit)) t2;
 
 DROP TABLE t;
-
--- TODO move
-CREATE TABLE t (val varbit(3));
-CREATE INDEX ON t USING hnsw (val bit_hamming_ops);
-CREATE INDEX ON t USING hnsw ((val::bit(3)) bit_hamming_ops);
-DROP TABLE t;

test/sql/hnsw_sparsevec_cosine.sql

@@ -1,13 +1,13 @@
 SET enable_seqscan = off;
 
-CREATE TABLE t (val sparsevec(3));
-INSERT INTO t (val) VALUES ('{}/3'), ('{1:1,2:2,3:3}/3'), ('{1:1,2:1,3:1}/3'), (NULL);
+CREATE TABLE t (val sparsevec);
+INSERT INTO t (val) VALUES ('{}'), ('{0:1,1:2,2:3}'), ('{0:1,1:1,2:1}'), (NULL);
 CREATE INDEX ON t USING hnsw (val sparsevec_cosine_ops);
 
-INSERT INTO t (val) VALUES ('{1:1,2:2,3:4}/3');
+INSERT INTO t (val) VALUES ('{0:1,1:2,2:4}');
 
-SELECT * FROM t ORDER BY val <=> '{1:3,2:3,3:3}/3';
-SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> '{}/3') t2;
+SELECT * FROM t ORDER BY val <=> '{0:3,1:3,2:3}';
+SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> '{}') 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(3));
-INSERT INTO t (val) VALUES ('{}/3'), ('{1:1,2:2,3:3}/3'), ('{1:1,2:1,3:1}/3'), (NULL);
+CREATE TABLE t (val sparsevec);
+INSERT INTO t (val) VALUES ('{}'), ('{0:1,1:2,2:3}'), ('{0:1,1:1,2:1}'), (NULL);
 CREATE INDEX ON t USING hnsw (val sparsevec_ip_ops);
 
-INSERT INTO t (val) VALUES ('{1:1,2:2,3:4}/3');
+INSERT INTO t (val) VALUES ('{0:1,1:2,2:4}');
 
-SELECT * FROM t ORDER BY val <#> '{1:3,2:3,3:3}/3';
+SELECT * FROM t ORDER BY val <#> '{0:3,1:3,2:3}';
 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(3));
-INSERT INTO t (val) VALUES ('{}/3'), ('{1:1,2:2,3:3}/3'), ('{1:1,2:1,3:1}/3'), (NULL);
+CREATE TABLE t (val sparsevec);
+INSERT INTO t (val) VALUES ('{}'), ('{0:1,1:2,2:3}'), ('{0:1,1:1,2:1}'), (NULL);
 CREATE INDEX ON t USING hnsw (val sparsevec_l2_ops);
 
-INSERT INTO t (val) VALUES ('{1:1,2:2,3:4}/3');
+INSERT INTO t (val) VALUES ('{0:1,1:2,2:4}');
 
-SELECT * FROM t ORDER BY val <-> '{1:3,2:3,3:3}/3';
+SELECT * FROM t ORDER BY val <-> '{0:3,1:3,2:3}';
 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 <-> '{1:3,2:3,3:3}/3';
+SELECT * FROM t ORDER BY val <-> '{0:3,1:3,2:3}';
 
 DROP TABLE t;
 
 -- TODO move
-CREATE TABLE t (val sparsevec(1001));
+CREATE TABLE t (val sparsevec);
 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

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

test/sql/sparsevec_input.sql

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

test/t/016_hnsw_vacuum_recall.pl

@@ -86,7 +86,7 @@ foreach (@queries)
 	push(@expected, $res);
 }
 
-test_recall(0.18, $limit, "before vacuum");
+test_recall(0.19, $limit, "before vacuum");
 test_recall(0.95, 100, "before vacuum");
 
 # TODO Test concurrent inserts with vacuum

test/t/020_hnsw_bit_build_recall.pl

@@ -99,7 +99,7 @@ for my $i (0 .. $#operators)
 	));
 
 	# Test approximate results
-	my $min = $operator eq "<\%>" ? 0.95 : 0.98;
+	my $min = $operator eq "<\%>" ? 0.96 : 0.98;
 	test_recall($min, $operator);
 
 	$node->safe_psql("postgres", "DROP INDEX idx;");

test/t/022_hnsw_sparsevec_build_recall.pl

@@ -1,128 +0,0 @@
-use strict;
-use warnings;
-use PostgresNode;
-use TestLib;
-use Test::More;
-
-my $node;
-my @queries = ();
-my @expected;
-my $limit = 20;
-
-sub test_recall
-{
-	my ($min, $operator) = @_;
-	my $correct = 0;
-	my $total = 0;
-
-	my $explain = $node->safe_psql("postgres", qq(
-		SET enable_seqscan = off;
-		EXPLAIN ANALYZE SELECT i FROM tst ORDER BY v $operator '$queries[0]' LIMIT $limit;
-	));
-	like($explain, qr/Index Scan/);
-
-	for my $i (0 .. $#queries)
-	{
-		my $actual = $node->safe_psql("postgres", qq(
-			SET enable_seqscan = off;
-			SELECT i FROM tst ORDER BY v $operator '$queries[$i]' LIMIT $limit;
-		));
-		my @actual_ids = split("\n", $actual);
-		my %actual_set = map { $_ => 1 } @actual_ids;
-
-		my @expected_ids = split("\n", $expected[$i]);
-
-		foreach (@expected_ids)
-		{
-			if (exists($actual_set{$_}))
-			{
-				$correct++;
-			}
-			$total++;
-		}
-	}
-
-	cmp_ok($correct / $total, ">=", $min, $operator);
-}
-
-# Initialize node
-$node = get_new_node('node');
-$node->init;
-$node->start;
-
-# Create table
-$node->safe_psql("postgres", "CREATE EXTENSION vector;");
-$node->safe_psql("postgres", "CREATE TABLE tst (i int4, v sparsevec(3));");
-$node->safe_psql("postgres",
-	"INSERT INTO tst SELECT i, ARRAY[random(), random(), random()]::vector::sparsevec FROM generate_series(1, 10000) i;"
-);
-
-# Generate queries
-for (1 .. 20)
-{
-	my $r1 = rand();
-	my $r2 = rand();
-	my $r3 = rand();
-	push(@queries, "{1:$r1,2:$r2,3:$r3}/3");
-}
-
-# Check each index type
-my @operators = ("<->", "<#>", "<=>");
-my @opclasses = ("sparsevec_l2_ops", "sparsevec_ip_ops", "sparsevec_cosine_ops");
-
-for my $i (0 .. $#operators)
-{
-	my $operator = $operators[$i];
-	my $opclass = $opclasses[$i];
-
-	# Get exact results
-	@expected = ();
-	foreach (@queries)
-	{
-		my $res = $node->safe_psql("postgres", "SELECT i FROM tst ORDER BY v $operator '$_' LIMIT $limit;");
-		push(@expected, $res);
-	}
-
-	# Build index serially
-	$node->safe_psql("postgres", qq(
-		SET max_parallel_maintenance_workers = 0;
-		CREATE INDEX idx ON tst USING hnsw (v $opclass);
-	));
-
-	# Test approximate results
-	my $min = $operator eq "<#>" ? 0.80 : 0.99;
-	test_recall($min, $operator);
-
-	$node->safe_psql("postgres", "DROP INDEX idx;");
-
-	# Build index in parallel in memory
-	my ($ret, $stdout, $stderr) = $node->psql("postgres", qq(
-		SET client_min_messages = DEBUG;
-		SET min_parallel_table_scan_size = 1;
-		CREATE INDEX idx ON tst USING hnsw (v $opclass);
-	));
-	is($ret, 0, $stderr);
-	like($stderr, qr/using \d+ parallel workers/);
-
-	# Test approximate results
-	test_recall($min, $operator);
-
-	$node->safe_psql("postgres", "DROP INDEX idx;");
-
-	# Build index in parallel on disk
-	# Set parallel_workers on table to use workers with low maintenance_work_mem
-	($ret, $stdout, $stderr) = $node->psql("postgres", qq(
-		ALTER TABLE tst SET (parallel_workers = 2);
-		SET client_min_messages = DEBUG;
-		SET maintenance_work_mem = '4MB';
-		CREATE INDEX idx ON tst USING hnsw (v $opclass);
-		ALTER TABLE tst RESET (parallel_workers);
-	));
-	is($ret, 0, $stderr);
-	like($stderr, qr/using \d+ parallel workers/);
-	like($stderr, qr/hnsw graph no longer fits into maintenance_work_mem/);
-
-	$node->safe_psql("postgres", "DROP INDEX idx;");
-}
-
-done_testing();