Include vectors with zero norm in indexes for cosine distance

The HnswInitElement() allocate an element with not initialize value
filed, which may has garbage that lead HnswFreeElement() free an
invalid pointer.

README.md

@@ -269,6 +269,8 @@ Specify HNSW parameters
 CREATE INDEX ON items USING hnsw (embedding vector_l2_ops) WITH (m = 16, ef_construction = 64);
 ```
 
+A higher value of `ef_construction` provides better recall at the cost of index build time / insert speed.
+
 ### Query Options
 
 Specify the size of the dynamic candidate list for search (40 by default)
@@ -678,16 +680,6 @@ SELECT extversion FROM pg_extension WHERE extname = 'vector';
 
 ## Upgrade Notes
 
-### 0.6.0
-
-If upgrading with Postgres < 13, remove this line from `sql/vector--0.5.1--0.6.0.sql`:
-
-```sql
-ALTER TYPE vector SET (STORAGE = external);
-```
-
-Then run `make install` and `ALTER EXTENSION vector UPDATE;`.
-
 ### 0.4.0
 
 If upgrading with Postgres < 13, remove this line from `sql/vector--0.3.2--0.4.0.sql`:

sql/vector--0.5.1--0.6.0.sql

@@ -1,5 +0,0 @@
--- complain if script is sourced in psql, rather than via CREATE EXTENSION
-\echo Use "ALTER EXTENSION vector UPDATE TO '0.6.0'" to load this file. \quit
-
--- remove this single line for Postgres < 13
-ALTER TYPE vector SET (STORAGE = external);

sql/vector.sql

@@ -26,7 +26,7 @@ CREATE TYPE vector (
 	TYPMOD_IN = vector_typmod_in,
 	RECEIVE   = vector_recv,
 	SEND      = vector_send,
-	STORAGE   = external
+	STORAGE   = extended
 );
 
 -- functions

src/hnsw.h

@@ -316,7 +316,7 @@ typedef struct HnswVacuumState
 int			HnswGetM(Relation index);
 int			HnswGetEfConstruction(Relation index);
 FmgrInfo   *HnswOptionalProcInfo(Relation index, uint16 procnum);
-bool		HnswNormValue(FmgrInfo *procinfo, Oid collation, Datum *value, Vector * result);
+void		HnswNormValue(FmgrInfo *procinfo, Oid collation, Datum *value, Vector * result);
 void		HnswCommitBuffer(Buffer buf, GenericXLogState *state);
 Buffer		HnswNewBuffer(Relation index, ForkNumber forkNum);
 void		HnswInitPage(Buffer buf, Page page);

src/hnswbuild.c

@@ -307,10 +307,7 @@ InsertTuple(Relation index, Datum *values, HnswElement element, HnswBuildState *
 
 	/* Normalize if needed */
 	if (buildstate->normprocinfo != NULL)
-	{
-		if (!HnswNormValue(buildstate->normprocinfo, collation, &value, buildstate->normvec))
-			return false;
-	}
+		HnswNormValue(buildstate->normprocinfo, collation, &value, buildstate->normvec);
 
 	/* Copy value to element so accessible outside of memory context */
 	oldCtx = MemoryContextSwitchTo(outerCtx);

src/hnswinsert.c

@@ -500,10 +500,7 @@ HnswInsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heap_ti
 	/* Normalize if needed */
 	normprocinfo = HnswOptionalProcInfo(index, HNSW_NORM_PROC);
 	if (normprocinfo != NULL)
-	{
-		if (!HnswNormValue(normprocinfo, collation, &value, NULL))
-			return false;
-	}
+		HnswNormValue(normprocinfo, collation, &value, NULL);
 
 	/*
 	 * Get a shared lock. This allows vacuum to ensure no in-flight inserts

src/hnswutils.c

@@ -55,7 +55,7 @@ HnswOptionalProcInfo(Relation index, uint16 procnum)
  * The caller needs to free the pointer stored in value
  * if it's different than the original value
  */
-bool
+void
 HnswNormValue(FmgrInfo *procinfo, Oid collation, Datum *value, Vector * result)
 {
 	double		norm = DatumGetFloat8(FunctionCall1Coll(procinfo, collation, *value));
@@ -71,11 +71,7 @@ HnswNormValue(FmgrInfo *procinfo, Oid collation, Datum *value, Vector * result)
 			result->x[i] = v->x[i] / norm;
 
 		*value = PointerGetDatum(result);
-
-		return true;
 	}
-
-	return false;
 }
 
 /*
@@ -177,6 +173,8 @@ HnswInitElement(ItemPointer heaptid, int m, double ml, int maxLevel)
 
 	HnswInitNeighbors(element, m);
 
+	element->value = PointerGetDatum(NULL);
+
 	return element;
 }
 

src/ivfbuild.c

@@ -177,10 +177,7 @@ AddTupleToSort(Relation index, ItemPointer tid, Datum *values, IvfflatBuildState
 
 	/* Normalize if needed */
 	if (buildstate->normprocinfo != NULL)
-	{
-		if (!IvfflatNormValue(buildstate->normprocinfo, buildstate->collation, &value, buildstate->normvec))
-			return;
-	}
+		IvfflatNormValue(buildstate->normprocinfo, buildstate->collation, &value, buildstate->normvec);
 
 	/* Find the list that minimizes the distance */
 	for (int i = 0; i < centers->length; i++)

src/ivfinsert.c

@@ -83,10 +83,7 @@ InsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heap_tid, R
 	/* Normalize if needed */
 	normprocinfo = IvfflatOptionalProcInfo(index, IVFFLAT_NORM_PROC);
 	if (normprocinfo != NULL)
-	{
-		if (!IvfflatNormValue(normprocinfo, index->rd_indcollation[0], &value, NULL))
-			return;
-	}
+		IvfflatNormValue(normprocinfo, index->rd_indcollation[0], &value, NULL);
 
 	/* Find the insert page - sets the page and list info */
 	FindInsertPage(index, values, &insertPage, &listInfo);

test/expected/hnsw_cosine.out

@@ -9,18 +9,19 @@ SELECT * FROM t ORDER BY val <=> '[3,3,3]';
  [1,1,1]
  [1,2,3]
  [1,2,4]
-(3 rows)
+ [0,0,0]
+(4 rows)
 
 SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> '[0,0,0]') t2;
  count 
 -------
-     3
+     4
 (1 row)
 
 SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> (SELECT NULL::vector)) t2;
  count 
 -------
-     3
+     4
 (1 row)
 
 DROP TABLE t;

test/expected/ivfflat_cosine.out

@@ -9,18 +9,19 @@ SELECT * FROM t ORDER BY val <=> '[3,3,3]';
  [1,1,1]
  [1,2,3]
  [1,2,4]
-(3 rows)
+ [0,0,0]
+(4 rows)
 
 SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> '[0,0,0]') t2;
  count 
 -------
-     3
+     4
 (1 row)
 
 SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> (SELECT NULL::vector)) t2;
  count 
 -------
-     3
+     4
 (1 row)
 
 DROP TABLE t;