Version bump to 0.4.3 [skip ci]

The compiler was complaining about a missing include due to the
addition of NaN/inf checks in 482a5f8b. Adding the include
silences the warnings.

.github/workflows/build.yml

@@ -8,6 +8,8 @@ jobs:
       fail-fast: false
       matrix:
         include:
+          - postgres: 16
+            os: ubuntu-22.04
           - postgres: 15
             os: ubuntu-22.04
           - postgres: 14
@@ -25,6 +27,8 @@ jobs:
           postgres-version: ${{ matrix.postgres }}
           dev-files: true
       - run: make
+        env:
+          PG_CFLAGS: -Wall -Wextra -Werror -Wno-unused-parameter -Wno-sign-compare
       - run: |
           export PG_CONFIG=`which pg_config`
           sudo --preserve-env=PG_CONFIG make install
@@ -44,6 +48,8 @@ jobs:
         with:
           postgres-version: 14
       - run: make
+        env:
+          PG_CFLAGS: -Wall -Wextra -Werror -Wno-unused-parameter
       - run: make install
       - run: make installcheck
       - if: ${{ failure() }}
@@ -70,3 +76,22 @@ jobs:
           nmake /NOLOGO /F Makefile.win clean && ^
           nmake /NOLOGO /F Makefile.win uninstall
         shell: cmd
+  i386:
+    runs-on: ubuntu-latest
+    container:
+      image: debian:11
+      options: --platform linux/386
+    steps:
+      - run: apt-get update && apt-get install -y build-essential git libipc-run-perl postgresql-13 postgresql-server-dev-13 sudo
+      - run: service postgresql start
+      - run: |
+          git clone https://github.com/${{ github.repository }}.git pgvector
+          cd pgvector
+          git checkout ${{ github.ref }}
+          make
+          make install
+          chown -R postgres .
+          sudo -u postgres make installcheck
+          sudo -u postgres make prove_installcheck
+        env:
+          PG_CFLAGS: -Wall -Wextra -Werror -Wno-unused-parameter -Wno-sign-compare

CHANGELOG.md

@@ -1,3 +1,20 @@
+## 0.4.3 (2023-06-10)
+
+- Improved cost estimation
+- Improved support for spaces with text representation
+- Fixed infinite and NaN values with binary representation
+- Fixed infinite values with vector addition and subtraction
+- Fixed infinite values with list centers
+- Fixed compilation error when `float8` is pass by reference
+- Fixed compilation error on PowerPC
+- Fixed segmentation fault with index creation on i386
+
+## 0.4.2 (2023-05-13)
+
+- Added notice when index created with little data
+- Fixed dimensions check for some direct function calls
+- Fixed installation error with Postgres 12.0-12.2
+
 ## 0.4.1 (2023-03-21)
 
 - Improved performance of cosine distance

LICENSE

@@ -1,4 +1,4 @@
-Portions Copyright (c) 1996-2022, PostgreSQL Global Development Group
+Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group
 
 Portions Copyright (c) 1994, The Regents of the University of California
 

META.json

@@ -2,7 +2,7 @@
 	"name": "vector",
 	"abstract": "Open-source vector similarity search for Postgres",
 	"description": "Supports L2 distance, inner product, and cosine distance",
-	"version": "0.4.1",
+	"version": "0.4.3",
 	"maintainer": [
 		"Andrew Kane <andrew@ankane.org>"
 	],
@@ -20,7 +20,7 @@
 		"vector": {
 			"file": "sql/vector.sql",
 			"docfile": "README.md",
-			"version": "0.4.1",
+			"version": "0.4.3",
 			"abstract": "Open-source vector similarity search for Postgres"
 		}
 	},

Makefile

@@ -1,5 +1,5 @@
 EXTENSION = vector
-EXTVERSION = 0.4.1
+EXTVERSION = 0.4.3
 
 MODULE_big = vector
 DATA = $(wildcard sql/*--*.sql)
@@ -19,6 +19,11 @@ ifeq ($(shell uname -s), Darwin)
 	endif
 endif
 
+# PowerPC doesn't support -march=native
+ifneq ($(filter ppc64%, $(shell uname -m)), )
+	OPTFLAGS =
+endif
+
 # For auto-vectorization:
 # - GCC (needs -ftree-vectorize OR -O3) - https://gcc.gnu.org/projects/tree-ssa/vectorization.html
 # - Clang (could use pragma instead) - https://llvm.org/docs/Vectorizers.html
@@ -63,3 +68,9 @@ dist:
 
 docker:
 	docker build --pull --no-cache --platform linux/amd64 -t ankane/pgvector:latest .
+
+.PHONY: docker-release
+
+docker-release:
+	docker buildx build --push --pull --no-cache --platform linux/amd64,linux/arm64 -t ankane/pgvector:latest .
+	docker buildx build --push --platform linux/amd64,linux/arm64 -t ankane/pgvector:v$(EXTVERSION) .

Makefile.win

@@ -1,5 +1,5 @@
 EXTENSION = vector
-EXTVERSION = 0.4.1
+EXTVERSION = 0.4.3
 
 OBJS = src\ivfbuild.obj src\ivfflat.obj src\ivfinsert.obj src\ivfkmeans.obj src\ivfscan.obj src\ivfutils.obj src\ivfvacuum.obj src\vector.obj
 

README.md

@@ -2,7 +2,13 @@
 
 Open-source vector similarity search for Postgres
 
-Supports exact and approximate nearest neighbor search for L2 distance, inner product, and cosine distance
+Supports
+
+- exact and approximate nearest neighbor search
+- L2 distance, inner product, and cosine distance
+- any [language](#languages) with a Postgres client
+
+Plus [ACID](https://en.wikipedia.org/wiki/ACID) compliance, point-in-time recovery, JOINs, and all of the other [great features](https://www.postgresql.org/about/) of Postgres
 
 [![Build Status](https://github.com/pgvector/pgvector/workflows/build/badge.svg?branch=master)](https://github.com/pgvector/pgvector/actions)
 
@@ -12,31 +18,31 @@ Compile and install the extension (supports Postgres 11+)
 
 ```sh
 cd /tmp
-git clone --branch v0.4.1 https://github.com/pgvector/pgvector.git
+git clone --branch v0.4.3 https://github.com/pgvector/pgvector.git
 cd pgvector
 make
 make install # may need sudo
 ```
 
-Then load it in databases where you want to use it
-
-```sql
-CREATE EXTENSION vector;
-```
-
 See the [installation notes](#installation-notes) if you run into issues
 
-You can also install it with [Docker](#docker), [Homebrew](#homebrew), [PGXN](#pgxn), [Yum](#yum), or [conda-forge](#conda-forge)
+You can also install it with [Docker](#docker), [Homebrew](#homebrew), [PGXN](#pgxn), [APT](#apt), [Yum](#yum), or [conda-forge](#conda-forge), and it comes preinstalled with [Postgres.app](#postgresapp) and many [hosted providers](#hosted-postgres)
 
 ## Getting Started
 
+Enable the extension (do this once in each database where you want to use it)
+
+```tsql
+CREATE EXTENSION vector;
+```
+
 Create a vector column with 3 dimensions
 
 ```sql
 CREATE TABLE items (id bigserial PRIMARY KEY, embedding vector(3));
 ```
 
-Insert values
+Insert vectors
 
 ```sql
 INSERT INTO items (embedding) VALUES ('[1,2,3]'), ('[4,5,6]');
@@ -123,8 +129,8 @@ SELECT embedding <-> '[3,1,2]' AS distance FROM items;
 
 For inner product, multiply by -1 (since `<#>` returns the negative inner product)
 
-```sql
-SELECT -1 * (embedding <#> '[3,1,2]') AS inner_product FROM items;
+```tsql
+SELECT (embedding <#> '[3,1,2]') * -1 AS inner_product FROM items;
 ```
 
 For cosine similarity, use 1 - cosine distance
@@ -133,7 +139,7 @@ For cosine similarity, use 1 - cosine distance
 SELECT 1 - (embedding <=> '[3,1,2]') AS cosine_similarity FROM items;
 ```
 
-#### Averaging
+#### Aggregates
 
 Average vectors
 
@@ -153,15 +159,11 @@ By default, pgvector performs exact nearest neighbor search, which provides perf
 
 You can add an index to use approximate nearest neighbor search, which trades some recall for performance. Unlike typical indexes, you will see different results for queries after adding an approximate index.
 
-Two keys to achieving good recall are:
+Three keys to achieving good recall are:
 
 1. Create the index *after* the table has some data
-2. Choose an appropriate number of lists (lower is better for recall, higher is better for speed)
-
-A good place to start is:
-
-- `rows / 1000` for up to 1M rows
-- `sqrt(rows)` for over 1M rows
+2. Choose an appropriate number of lists - a good place to start is `rows / 1000` for up to 1M rows and `sqrt(rows)` for over 1M rows
+3. When querying, specify an appropriate number of [probes](#query-options) (higher is better for recall, lower is better for speed) - a good place to start is `lists / 10` for up to 1M rows and `sqrt(lists)` for over 1M rows
 
 Add an index for each distance function you want to use.
 
@@ -190,7 +192,7 @@ Vectors with up to 2,000 dimensions can be indexed.
 Specify the number of probes (1 by default)
 
 ```sql
-SET ivfflat.probes = 1;
+SET ivfflat.probes = 10;
 ```
 
 A higher value provides better recall at the cost of speed, and it can be set to the number of lists for exact nearest neighbor search (at which point the planner won’t use the index)
@@ -199,7 +201,7 @@ Use `SET LOCAL` inside a transaction to set it for a single query
 
 ```sql
 BEGIN;
-SET LOCAL ivfflat.probes = 1;
+SET LOCAL ivfflat.probes = 10;
 SELECT ...
 COMMIT;
 ```
@@ -221,26 +223,42 @@ The phases are:
 
 Note: `tuples_done` and `tuples_total` are only populated during the `loading tuples` phase
 
-### Partial Indexes
+### Filtering
 
-Consider [partial indexes](https://www.postgresql.org/docs/current/indexes-partial.html) for queries with a `WHERE` clause
+There are a few ways to index nearest neighbor queries with a `WHERE` clause
 
 ```sql
 SELECT * FROM items WHERE category_id = 123 ORDER BY embedding <-> '[3,1,2]' LIMIT 5;
 ```
 
-can be indexed with:
+Create an index on one [or more](https://www.postgresql.org/docs/current/indexes-multicolumn.html) of the `WHERE` columns for exact search
 
 ```sql
-CREATE INDEX ON items USING ivfflat (embedding vector_l2_ops) WITH (lists = 100) WHERE (category_id = 123);
+CREATE INDEX ON items (category_id);
 ```
 
-To index many different values of `category_id`, consider [partitioning](https://www.postgresql.org/docs/current/ddl-partitioning.html) on `category_id`.
+Or a [partial index](https://www.postgresql.org/docs/current/indexes-partial.html) on the vector column for approximate search
+
+```sql
+CREATE INDEX ON items USING ivfflat (embedding vector_l2_ops) WITH (lists = 100)
+    WHERE (category_id = 123);
+```
+
+Use [partitioning](https://www.postgresql.org/docs/current/ddl-partitioning.html) for approximate search on many different values of the `WHERE` columns
 
 ```sql
 CREATE TABLE items (embedding vector(3), category_id int) PARTITION BY LIST(category_id);
 ```
 
+## Hybrid Search
+
+Use together with Postgres [full-text search](https://www.postgresql.org/docs/current/textsearch-intro.html) for hybrid search ([Python example](https://github.com/pgvector/pgvector-python/blob/master/examples/hybrid_search.py)).
+
+```sql
+SELECT id, content FROM items, to_tsquery('hello & search') query
+    WHERE textsearch @@ query ORDER BY ts_rank_cd(textsearch, query) DESC LIMIT 5;
+```
+
 ## Performance
 
 Use `EXPLAIN ANALYZE` to debug performance.
@@ -259,7 +277,7 @@ SET max_parallel_workers_per_gather = 4;
 
 If vectors are normalized to length 1 (like [OpenAI embeddings](https://platform.openai.com/docs/guides/embeddings/which-distance-function-should-i-use)), use inner product for best performance.
 
-```sql
+```tsql
 SELECT * FROM items ORDER BY embedding <#> '[3,1,2]' LIMIT 5;
 ```
 
@@ -279,8 +297,10 @@ Language | Libraries / Examples
 --- | ---
 C++ | [pgvector-cpp](https://github.com/pgvector/pgvector-cpp)
 C# | [pgvector-dotnet](https://github.com/pgvector/pgvector-dotnet)
+Crystal | [pgvector-crystal](https://github.com/pgvector/pgvector-crystal)
 Elixir | [pgvector-elixir](https://github.com/pgvector/pgvector-elixir)
 Go | [pgvector-go](https://github.com/pgvector/pgvector-go)
+Haskell | [pgvector-haskell](https://github.com/pgvector/pgvector-haskell)
 Java, Scala | [pgvector-java](https://github.com/pgvector/pgvector-java)
 Julia | [pgvector-julia](https://github.com/pgvector/pgvector-julia)
 Lua | [pgvector-lua](https://github.com/pgvector/pgvector-lua)
@@ -291,6 +311,7 @@ Python | [pgvector-python](https://github.com/pgvector/pgvector-python)
 R | [pgvector-r](https://github.com/pgvector/pgvector-r)
 Ruby | [pgvector-ruby](https://github.com/pgvector/pgvector-ruby), [Neighbor](https://github.com/ankane/neighbor)
 Rust | [pgvector-rust](https://github.com/pgvector/pgvector-rust)
+Swift | [pgvector-swift](https://github.com/pgvector/pgvector-swift)
 
 ## Frequently Asked Questions
 
@@ -304,10 +325,11 @@ Yes, pgvector uses the write-ahead log (WAL), which allows for replication and p
 
 #### What if I want to index vectors with more than 2,000 dimensions?
 
-Two things you can try are:
+You’ll need to use [dimensionality reduction](https://en.wikipedia.org/wiki/Dimensionality_reduction) at the moment.
 
-1. use dimensionality reduction
-2. compile Postgres with a larger block size (`./configure --with-blocksize=32`) and edit the limit in `src/ivfflat.h`
+#### Why am I seeing less results after adding an index?
+
+The index was likely created with too little data for the number of lists. Drop the index until the table has more data.
 
 ## Reference
 
@@ -351,7 +373,11 @@ If your machine has multiple Postgres installations, specify the path to [pg_con
 export PG_CONFIG=/Applications/Postgres.app/Contents/Versions/latest/bin/pg_config
 ```
 
-Then re-run the installation instructions (run `make clean` before `make` if needed)
+Then re-run the installation instructions (run `make clean` before `make` if needed). If `sudo` is needed for `make install`, use:
+
+```sh
+sudo --preserve-env=PG_CONFIG make install
+```
 
 ### Missing Header
 
@@ -360,7 +386,7 @@ If compilation fails with `fatal error: postgres.h: No such file or directory`,
 For Ubuntu and Debian, use:
 
 ```sh
-sudo apt-get install postgresql-server-dev-15
+sudo apt install postgresql-server-dev-15
 ```
 
 Note: Replace `15` with your Postgres server version
@@ -371,7 +397,7 @@ Support for Windows is currently experimental. Use `nmake` to build:
 
 ```cmd
 set "PGROOT=C:\Program Files\PostgreSQL\15"
-git clone --branch v0.4.1 https://github.com/pgvector/pgvector.git
+git clone --branch v0.4.3 https://github.com/pgvector/pgvector.git
 cd pgvector
 nmake /F Makefile.win
 nmake /F Makefile.win install
@@ -392,9 +418,9 @@ This adds pgvector to the [Postgres image](https://hub.docker.com/_/postgres) (r
 You can also build the image manually:
 
 ```sh
-git clone --branch v0.4.1 https://github.com/pgvector/pgvector.git
+git clone --branch v0.4.3 https://github.com/pgvector/pgvector.git
 cd pgvector
-docker build -t pgvector .
+docker build --build-arg PG_MAJOR=15 -t myuser/pgvector .
 ```
 
 ### Homebrew
@@ -405,6 +431,8 @@ With Homebrew Postgres, you can use:
 brew install pgvector
 ```
 
+Note: This only adds it to the `postgresql@14` formula
+
 ### PGXN
 
 Install from the [PostgreSQL Extension Network](https://pgxn.org/dist/vector) with:
@@ -413,6 +441,16 @@ Install from the [PostgreSQL Extension Network](https://pgxn.org/dist/vector) wi
 pgxn install vector
 ```
 
+### APT
+
+Debian and Ubuntu packages are available from the [PostgreSQL APT Repository](https://wiki.postgresql.org/wiki/Apt). Follow the [setup instructions](https://wiki.postgresql.org/wiki/Apt#Quickstart) and run:
+
+```sh
+sudo apt install postgresql-15-pgvector
+```
+
+Note: Replace `15` with your Postgres server version
+
 ### Yum
 
 RPM packages are available from the [PostgreSQL Yum Repository](https://yum.postgresql.org/). Follow the [setup instructions](https://www.postgresql.org/download/linux/redhat/) for your distribution and run:
@@ -435,17 +473,19 @@ conda install -c conda-forge pgvector
 
 This method is [community-maintained](https://github.com/conda-forge/pgvector-feedstock) by [@mmcauliffe](https://github.com/mmcauliffe)
 
+### Postgres.app
+
+Download the [latest release](https://postgresapp.com/downloads.html) with Postgres 15+.
+
 ## Hosted Postgres
 
 pgvector is available on [these providers](https://github.com/pgvector/pgvector/issues/54).
 
 To request a new extension on other providers:
 
-- Amazon RDS - follow the instructions on [this page](https://aws.amazon.com/rds/postgresql/faqs/)
 - Google Cloud SQL - vote or comment on [this page](https://issuetracker.google.com/issues/265172065)
-- Azure Database - vote or comment on [this page](https://feedback.azure.com/d365community/idea/7b423322-6189-ed11-a81b-000d3ae49307)
-- DigitalOcean Managed Databases - vote or comment on [this page](https://ideas.digitalocean.com/app-framework-services/p/pgvector-extension-for-postgresql)
-- Render - vote or comment on [this page](https://feedback.render.com/features/p/add-pgvector-extension-to-postgresql)
+- DigitalOcean Managed Databases - vote or comment on [this page](https://ideas.digitalocean.com/managed-database/p/pgvector-extension-for-postgresql)
+- Heroku Postgres - vote or comment on [this page](https://github.com/heroku/roadmap/issues/156)
 
 ## Upgrading
 

sql/vector--0.4.1--0.4.2.sql

@@ -0,0 +1,2 @@
+-- complain if script is sourced in psql, rather than via CREATE EXTENSION
+\echo Use "ALTER EXTENSION vector UPDATE TO '0.4.2'" to load this file. \quit

sql/vector--0.4.2--0.4.3.sql

@@ -0,0 +1,2 @@
+-- complain if script is sourced in psql, rather than via CREATE EXTENSION
+\echo Use "ALTER EXTENSION vector UPDATE TO '0.4.3'" to load this file. \quit

src/ivfbuild.c

@@ -147,7 +147,7 @@ AddTupleToSort(Relation index, ItemPointer tid, Datum *values, IvfflatBuildState
 {
 	double		distance;
 	double		minDistance = DBL_MAX;
-	int			closestCenter = -1;
+	int			closestCenter = 0;
 	VectorArray centers = buildstate->centers;
 	TupleTableSlot *slot = buildstate->slot;
 	int			i;
@@ -431,8 +431,18 @@ ComputeCenters(IvfflatBuildState * buildstate)
 	/* TODO Ensure within maintenance_work_mem */
 	buildstate->samples = VectorArrayInit(numSamples, buildstate->dimensions);
 	if (buildstate->heap != NULL)
+	{
 		SampleRows(buildstate);
 
+		if (buildstate->samples->length < buildstate->lists)
+		{
+			ereport(NOTICE,
+					(errmsg("ivfflat index created with little data"),
+					 errdetail("This will cause low recall."),
+					 errhint("Drop the index until the table has more data.")));
+		}
+	}
+
 	/* Calculate centers */
 	IvfflatBench("k-means", IvfflatKmeans(buildstate->index, buildstate->samples, buildstate->centers));
 
@@ -565,7 +575,7 @@ static void
 CreateEntryPages(IvfflatBuildState * buildstate, ForkNumber forkNum)
 {
 	AttrNumber	attNums[] = {1};
-	Oid			sortOperators[] = {Float8LessOperator};
+	Oid			sortOperators[] = {Int4LessOperator};
 	Oid			sortCollations[] = {InvalidOid};
 	bool		nullsFirstFlags[] = {false};
 

src/ivfflat.c

@@ -7,6 +7,7 @@
 #include "ivfflat.h"
 #include "utils/guc.h"
 #include "utils/selfuncs.h"
+#include "utils/spccache.h"
 
 #if PG_VERSION_NUM >= 120000
 #include "commands/progress.h"
@@ -63,13 +64,13 @@ ivfflatbuildphasename(int64 phasenum)
 static void
 ivfflatcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
 					Cost *indexStartupCost, Cost *indexTotalCost,
-					Selectivity *indexSelectivity, double *indexCorrelation
-					,double *indexPages
-)
+					Selectivity *indexSelectivity, double *indexCorrelation,
+					double *indexPages)
 {
 	GenericCosts costs;
 	int			lists;
 	double		ratio;
+	double		spc_seq_page_cost;
 	Relation	indexRel;
 #if PG_VERSION_NUM < 120000
 	List	   *qinfos;
@@ -88,6 +89,22 @@ ivfflatcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
 
 	MemSet(&costs, 0, sizeof(costs));
 
+	indexRel = index_open(path->indexinfo->indexoid, NoLock);
+	lists = IvfflatGetLists(indexRel);
+	index_close(indexRel, NoLock);
+
+	/* Get the ratio of lists that we need to visit */
+	ratio = ((double) ivfflat_probes) / lists;
+	if (ratio > 1.0)
+		ratio = 1.0;
+
+	/*
+	 * This gives us the subset of tuples to visit. This value is passed into
+	 * the generic cost estimator to determine the number of pages to visit
+	 * during the index scan.
+	 */
+	costs.numIndexTuples = path->indexinfo->tuples * ratio;
+
 #if PG_VERSION_NUM >= 120000
 	genericcostestimate(root, path, loop_count, &costs);
 #else
@@ -95,17 +112,31 @@ ivfflatcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
 	genericcostestimate(root, path, loop_count, qinfos, &costs);
 #endif
 
-	indexRel = index_open(path->indexinfo->indexoid, NoLock);
-	lists = IvfflatGetLists(indexRel);
-	index_close(indexRel, NoLock);
+	get_tablespace_page_costs(path->indexinfo->reltablespace, NULL, &spc_seq_page_cost);
 
-	ratio = ((double) ivfflat_probes) / lists;
-	if (ratio > 1)
-		ratio = 1;
+	/* Adjust cost if needed since TOAST not included in seq scan cost */
+	if (costs.numIndexPages > path->indexinfo->rel->pages && ratio < 0.5)
+	{
+		/* Change all page cost from random to sequential */
+		costs.indexTotalCost -= costs.numIndexPages * (costs.spc_random_page_cost - spc_seq_page_cost);
 
-	costs.indexTotalCost *= ratio;
+		/* Remove cost of extra pages */
+		costs.indexTotalCost -= (costs.numIndexPages - path->indexinfo->rel->pages) * spc_seq_page_cost;
+	}
+	else
+	{
+		/* Change some page cost from random to sequential */
+		costs.indexTotalCost -= 0.5 * costs.numIndexPages * (costs.spc_random_page_cost - spc_seq_page_cost);
+	}
 
-	/* Startup cost and total cost are same */
+	/*
+	 * If the list selectivity is lower than what is returned from the generic
+	 * cost estimator, use that.
+	 */
+	if (ratio < costs.indexSelectivity)
+		costs.indexSelectivity = ratio;
+
+	/* Use total cost since most work happens before first tuple is returned */
 	*indexStartupCost = costs.indexTotalCost;
 	*indexTotalCost = costs.indexTotalCost;
 	*indexSelectivity = costs.indexSelectivity;

src/ivfinsert.c

@@ -23,6 +23,10 @@ FindInsertPage(Relation rel, Datum *values, BlockNumber *insertPage, ListInfo *
 	OffsetNumber offno;
 	OffsetNumber maxoffno;
 
+	/* Avoid compiler warning */
+	listInfo->blkno = nextblkno;
+	listInfo->offno = FirstOffsetNumber;
+
 	procinfo = index_getprocinfo(rel, 1, IVFFLAT_DISTANCE_PROC);
 	collation = rel->rd_indcollation[0];
 
@@ -39,7 +43,7 @@ FindInsertPage(Relation rel, Datum *values, BlockNumber *insertPage, ListInfo *
 			list = (IvfflatList) PageGetItem(cpage, PageGetItemId(cpage, offno));
 			distance = DatumGetFloat8(FunctionCall2Coll(procinfo, collation, values[0], PointerGetDatum(&list->center)));
 
-			if (distance < minDistance)
+			if (distance < minDistance || !BlockNumberIsValid(*insertPage))
 			{
 				*insertPage = list->insertPage;
 				listInfo->blkno = nextblkno;

src/ivfkmeans.c

@@ -1,6 +1,7 @@
 #include "postgres.h"
 
 #include <float.h>
+#include <math.h>
 
 #include "ivfflat.h"
 #include "miscadmin.h"
@@ -211,7 +212,7 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers)
 
 	/* Check memory requirements */
 	/* Add one to error message to ceil */
-	if (totalSize / 1024 > maintenance_work_mem)
+	if (totalSize > (Size) maintenance_work_mem * 1024L)
 		ereport(ERROR,
 				(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
 				 errmsg("memory required is %zu MB, maintenance_work_mem is %d MB",
@@ -251,7 +252,7 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers)
 	for (j = 0; j < numSamples; j++)
 	{
 		minDistance = DBL_MAX;
-		closestCenter = -1;
+		closestCenter = 0;
 
 		/* Find closest center */
 		for (k = 0; k < numCenters; k++)
@@ -398,6 +399,14 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers)
 
 			if (centerCounts[j] > 0)
 			{
+				/* Double avoids overflow, but requires more memory */
+				/* TODO Update bounds */
+				for (k = 0; k < dimensions; k++)
+				{
+					if (isinf(vec->x[k]))
+						vec->x[k] = vec->x[k] > 0 ? FLT_MAX : -FLT_MAX;
+				}
+
 				for (k = 0; k < dimensions; k++)
 					vec->x[k] /= centerCounts[j];
 			}
@@ -461,12 +470,29 @@ CheckCenters(Relation index, VectorArray centers)
 {
 	FmgrInfo   *normprocinfo;
 	Oid			collation;
+	Vector	   *vec;
 	int			i;
+	int			j;
 	double		norm;
 
 	if (centers->length != centers->maxlen)
 		elog(ERROR, "Not enough centers. Please report a bug.");
 
+	/* Ensure no NaN or infinite values */
+	for (i = 0; i < centers->length; i++)
+	{
+		vec = VectorArrayGet(centers, i);
+
+		for (j = 0; j < vec->dim; j++)
+		{
+			if (isnan(vec->x[j]))
+				elog(ERROR, "NaN detected. Please report a bug.");
+
+			if (isinf(vec->x[j]))
+				elog(ERROR, "Infinite value detected. Please report a bug.");
+		}
+	}
+
 	/* Ensure no duplicate centers */
 	/* Fine to sort in-place */
 	qsort(centers->items, centers->length, VECTOR_SIZE(centers->dim), CompareVectors);

src/ivfscan.c

@@ -170,12 +170,14 @@ GetScanItems(IndexScanDesc scan, Datum value)
 		}
 	}
 
+	FreeAccessStrategy(bas);
+
 	/* TODO Scan more lists */
 	if (tuples < 100)
 		ereport(DEBUG1,
 				(errmsg("index scan found few tuples"),
-				 errdetail("index may have been created without data or lists is too high"),
-				 errhint("recreate the index and possibly decrease lists")));
+				 errdetail("Index may have been created with little data."),
+				 errhint("Recreate the index and possibly decrease lists.")));
 
 	tuplesort_performsort(so->sortstate);
 }

src/ivfvacuum.c

@@ -132,6 +132,8 @@ ivfflatbulkdelete(IndexVacuumInfo *info, IndexBulkDeleteResult *stats,
 		}
 	}
 
+	FreeAccessStrategy(bas);
+
 	return stats;
 }
 

src/vector.c

@@ -72,18 +72,43 @@ CheckDim(int dim)
  * Ensure finite elements
  */
 static inline void
-CheckElement(float value)
+CheckElementOrFree(float value, Vector * vec)
 {
 	if (isnan(value))
+	{
+		if (vec != NULL)
+			pfree(vec);
+
 		ereport(ERROR,
 				(errcode(ERRCODE_DATA_EXCEPTION),
 				 errmsg("NaN not allowed in vector")));
-
+	}
 
 	if (isinf(value))
+	{
+		if (vec != NULL)
+			pfree(vec);
+
 		ereport(ERROR,
 				(errcode(ERRCODE_DATA_EXCEPTION),
 				 errmsg("infinite value not allowed in vector")));
+	}
+}
+
+/*
+ * Check for whitespace, since array_isspace() is static
+ */
+static inline bool
+vector_isspace(char ch)
+{
+	if (ch == ' ' ||
+		ch == '\t' ||
+		ch == '\n' ||
+		ch == '\r' ||
+		ch == '\v' ||
+		ch == '\f')
+		return true;
+	return false;
 }
 
 /*
@@ -100,7 +125,7 @@ CheckStateArray(ArrayType *statearray, const char *caller)
 	return (float8 *) ARR_DATA_PTR(statearray);
 }
 
-#if PG_VERSION_NUM < 120000
+#if PG_VERSION_NUM < 120003
 static pg_noinline void
 float_overflow_error(void)
 {
@@ -150,6 +175,9 @@ vector_in(PG_FUNCTION_ARGS)
 	char	   *stringEnd;
 	Vector	   *result;
 
+	while (vector_isspace(*str))
+		str++;
+
 	if (*str != '[')
 		ereport(ERROR,
 				(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
@@ -169,7 +197,7 @@ vector_in(PG_FUNCTION_ARGS)
 
 		/* Use strtof like float4in to avoid a double-rounding problem */
 		x[dim] = strtof(pt, &stringEnd);
-		CheckElement(x[dim]);
+		CheckElementOrFree(x[dim], NULL);
 		dim++;
 
 		if (stringEnd == pt)
@@ -177,6 +205,9 @@ vector_in(PG_FUNCTION_ARGS)
 					(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
 					 errmsg("invalid input syntax for type vector: \"%s\"", pt)));
 
+		while (vector_isspace(*stringEnd))
+			stringEnd++;
+
 		if (*stringEnd != '\0' && *stringEnd != ']')
 			ereport(ERROR,
 					(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
@@ -191,7 +222,13 @@ vector_in(PG_FUNCTION_ARGS)
 				 errmsg("malformed vector literal"),
 				 errdetail("Unexpected end of input.")));
 
-	if (stringEnd[1] != '\0')
+	stringEnd++;
+
+	/* Only whitespace is allowed after the closing brace */
+	while (vector_isspace(*stringEnd))
+		stringEnd++;
+
+	if (*stringEnd != '\0')
 		ereport(ERROR,
 				(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
 				 errmsg("malformed vector literal"),
@@ -330,7 +367,10 @@ vector_recv(PG_FUNCTION_ARGS)
 
 	result = InitVector(dim);
 	for (i = 0; i < dim; i++)
+	{
 		result->x[i] = pq_getmsgfloat4(buf);
+		CheckElementOrFree(result->x[i], result);
+	}
 
 	PG_RETURN_POINTER(result);
 }
@@ -396,10 +436,8 @@ array_to_vector(PG_FUNCTION_ARGS)
 	get_typlenbyvalalign(ARR_ELEMTYPE(array), &typlen, &typbyval, &typalign);
 	deconstruct_array(array, ARR_ELEMTYPE(array), typlen, typbyval, typalign, &elemsp, &nullsp, &nelemsp);
 
-	if (typmod == -1)
-		CheckDim(nelemsp);
-	else
-		CheckExpectedDim(typmod, nelemsp);
+	CheckDim(nelemsp);
+	CheckExpectedDim(typmod, nelemsp);
 
 	result = InitVector(nelemsp);
 	for (i = 0; i < nelemsp; i++)
@@ -409,6 +447,7 @@ array_to_vector(PG_FUNCTION_ARGS)
 					(errcode(ERRCODE_NULL_VALUE_NOT_ALLOWED),
 					 errmsg("array must not containing NULLs")));
 
+		/* TODO Move outside loop in 0.5.0 */
 		if (ARR_ELEMTYPE(array) == INT4OID)
 			result->x[i] = DatumGetInt32(elemsp[i]);
 		else if (ARR_ELEMTYPE(array) == FLOAT8OID)
@@ -422,7 +461,7 @@ array_to_vector(PG_FUNCTION_ARGS)
 					(errcode(ERRCODE_DATA_EXCEPTION),
 					 errmsg("unsupported array type")));
 
-		CheckElement(result->x[i]);
+		CheckElementOrFree(result->x[i], result);
 	}
 
 	PG_RETURN_POINTER(result);
@@ -436,17 +475,19 @@ Datum
 vector_to_float4(PG_FUNCTION_ARGS)
 {
 	Vector	   *vec = PG_GETARG_VECTOR_P(0);
-	Datum	   *d;
+	Datum	   *datums;
 	ArrayType  *result;
 	int			i;
 
-	d = (Datum *) palloc(sizeof(Datum) * vec->dim);
+	datums = (Datum *) palloc(sizeof(Datum) * vec->dim);
 
 	for (i = 0; i < vec->dim; i++)
-		d[i] = Float4GetDatum(vec->x[i]);
+		datums[i] = Float4GetDatum(vec->x[i]);
 
 	/* Use TYPALIGN_INT for float4 */
-	result = construct_array(d, vec->dim, FLOAT4OID, sizeof(float4), true, TYPALIGN_INT);
+	result = construct_array(datums, vec->dim, FLOAT4OID, sizeof(float4), true, TYPALIGN_INT);
+
+	pfree(datums);
 
 	PG_RETURN_POINTER(result);
 }
@@ -658,6 +699,16 @@ vector_add(PG_FUNCTION_ARGS)
 	for (int i = 0, imax = a->dim; i < imax; i++)
 		rx[i] = ax[i] + bx[i];
 
+	/* Check for overflow */
+	for (int i = 0, imax = a->dim; i < imax; i++)
+	{
+		if (isinf(rx[i]))
+		{
+			pfree(result);
+			float_overflow_error();
+		}
+	}
+
 	PG_RETURN_POINTER(result);
 }
 
@@ -684,6 +735,16 @@ vector_sub(PG_FUNCTION_ARGS)
 	for (int i = 0, imax = a->dim; i < imax; i++)
 		rx[i] = ax[i] - bx[i];
 
+	/* Check for overflow */
+	for (int i = 0, imax = a->dim; i < imax; i++)
+	{
+		if (isinf(rx[i]))
+		{
+			pfree(result);
+			float_overflow_error();
+		}
+	}
+
 	PG_RETURN_POINTER(result);
 }
 
@@ -829,12 +890,12 @@ vector_accum(PG_FUNCTION_ARGS)
 	n = statevalues[0] + 1.0;
 
 	statedatums = CreateStateDatums(dim);
-	statedatums[0] = Float8GetDatumFast(n);
+	statedatums[0] = Float8GetDatum(n);
 
 	if (newarr)
 	{
 		for (int i = 0; i < dim; i++)
-			statedatums[i + 1] = Float8GetDatumFast((double) x[i]);
+			statedatums[i + 1] = Float8GetDatum((double) x[i]);
 	}
 	else
 	{
@@ -842,10 +903,14 @@ vector_accum(PG_FUNCTION_ARGS)
 		{
 			double		v = statevalues[i + 1] + x[i];
 
+			/* Check for overflow */
 			if (isinf(v))
+			{
+				pfree(statedatums);
 				float_overflow_error();
+			}
 
-			statedatums[i + 1] = Float8GetDatumFast(v);
+			statedatums[i + 1] = Float8GetDatum(v);
 		}
 	}
 
@@ -890,7 +955,7 @@ vector_combine(PG_FUNCTION_ARGS)
 		dim = STATE_DIMS(statearray2);
 		statedatums = CreateStateDatums(dim);
 		for (int i = 1; i <= dim; i++)
-			statedatums[i] = Float8GetDatumFast(statevalues2[i]);
+			statedatums[i] = Float8GetDatum(statevalues2[i]);
 	}
 	else if (n2 == 0.0)
 	{
@@ -898,7 +963,7 @@ vector_combine(PG_FUNCTION_ARGS)
 		dim = STATE_DIMS(statearray1);
 		statedatums = CreateStateDatums(dim);
 		for (int i = 1; i <= dim; i++)
-			statedatums[i] = Float8GetDatumFast(statevalues1[i]);
+			statedatums[i] = Float8GetDatum(statevalues1[i]);
 	}
 	else
 	{
@@ -910,14 +975,18 @@ vector_combine(PG_FUNCTION_ARGS)
 		{
 			double		v = statevalues1[i] + statevalues2[i];
 
+			/* Check for overflow */
 			if (isinf(v))
+			{
+				pfree(statedatums);
 				float_overflow_error();
+			}
 
-			statedatums[i] = Float8GetDatumFast(v);
+			statedatums[i] = Float8GetDatum(v);
 		}
 	}
 
-	statedatums[0] = Float8GetDatumFast(n);
+	statedatums[0] = Float8GetDatum(n);
 
 	result = construct_array(statedatums, dim + 1,
 							 FLOAT8OID,
@@ -940,7 +1009,6 @@ vector_avg(PG_FUNCTION_ARGS)
 	float8		n;
 	uint16		dim;
 	Vector	   *result;
-	float		v;
 
 	/* Check array before using */
 	statevalues = CheckStateArray(statearray, "vector_avg");
@@ -952,12 +1020,12 @@ vector_avg(PG_FUNCTION_ARGS)
 
 	/* Create vector */
 	dim = STATE_DIMS(statearray);
+	CheckDim(dim);
 	result = InitVector(dim);
 	for (int i = 0; i < dim; i++)
 	{
-		v = statevalues[i + 1] / n;
-		CheckElement(v);
-		result->x[i] = v;
+		result->x[i] = statevalues[i + 1] / n;
+		CheckElementOrFree(result->x[i], result);
 	}
 
 	PG_RETURN_POINTER(result);

test/expected/cast.out

@@ -46,6 +46,8 @@ SELECT '[1,2,3]'::vector::real[];
 
 SELECT array_agg(n)::vector FROM generate_series(1, 16001) n;
 ERROR:  vector cannot have more than 16000 dimensions
+SELECT array_to_vector(array_agg(n), 16001, false) FROM generate_series(1, 16001) n;
+ERROR:  vector cannot have more than 16000 dimensions
 -- ensure no error
 SELECT ARRAY[1,2,3] = ARRAY[1,2,3];
  ?column? 

test/expected/functions.out

@@ -4,12 +4,16 @@ SELECT '[1,2,3]'::vector + '[4,5,6]';
  [5,7,9]
 (1 row)
 
+SELECT '[3e38]'::vector + '[3e38]';
+ERROR:  value out of range: overflow
 SELECT '[1,2,3]'::vector - '[4,5,6]';
   ?column?  
 ------------
  [-3,-3,-3]
 (1 row)
 
+SELECT '[-3e38]'::vector - '[3e38]';
+ERROR:  value out of range: overflow
 SELECT vector_dims('[1,2,3]');
  vector_dims 
 -------------
@@ -102,3 +106,5 @@ SELECT avg(v) FROM unnest(ARRAY[]::vector[]) v;
 
 SELECT avg(v) FROM unnest(ARRAY['[1,2]'::vector, '[3]']) v;
 ERROR:  expected 2 dimensions, not 1
+SELECT vector_avg(array_agg(n)) FROM generate_series(1, 16002) n;
+ERROR:  vector cannot have more than 16000 dimensions

test/expected/input.out

@@ -10,6 +10,12 @@ SELECT '[-1,2,3]'::vector;
  [-1,2,3]
 (1 row)
 
+SELECT ' [ 1,  2 ,    3  ] '::vector(3);
+ vector  
+---------
+ [1,2,3]
+(1 row)
+
 SELECT '[1.23456]'::vector;
   vector   
 -----------
@@ -32,6 +38,10 @@ SELECT '[-Infinity,1]'::vector;
 ERROR:  infinite value not allowed in vector
 LINE 1: SELECT '[-Infinity,1]'::vector;
                ^
+SELECT '[4e38,1]'::vector;
+ERROR:  infinite value not allowed in vector
+LINE 1: SELECT '[4e38,1]'::vector;
+               ^
 SELECT '[1,2,3'::vector;
 ERROR:  malformed vector literal
 LINE 1: SELECT '[1,2,3'::vector;

test/sql/cast.sql

@@ -10,6 +10,7 @@ SELECT '{-Infinity}'::real[]::vector;
 SELECT '{}'::real[]::vector;
 SELECT '[1,2,3]'::vector::real[];
 SELECT array_agg(n)::vector FROM generate_series(1, 16001) n;
+SELECT array_to_vector(array_agg(n), 16001, false) FROM generate_series(1, 16001) n;
 
 -- ensure no error
 SELECT ARRAY[1,2,3] = ARRAY[1,2,3];

test/sql/functions.sql

@@ -1,5 +1,7 @@
 SELECT '[1,2,3]'::vector + '[4,5,6]';
+SELECT '[3e38]'::vector + '[3e38]';
 SELECT '[1,2,3]'::vector - '[4,5,6]';
+SELECT '[-3e38]'::vector - '[3e38]';
 
 SELECT vector_dims('[1,2,3]');
 
@@ -24,3 +26,4 @@ SELECT avg(v) FROM unnest(ARRAY['[1,2,3]'::vector, '[3,5,7]']) v;
 SELECT avg(v) FROM unnest(ARRAY['[1,2,3]'::vector, '[3,5,7]', NULL]) v;
 SELECT avg(v) FROM unnest(ARRAY[]::vector[]) v;
 SELECT avg(v) FROM unnest(ARRAY['[1,2]'::vector, '[3]']) v;
+SELECT vector_avg(array_agg(n)) FROM generate_series(1, 16002) n;

test/sql/input.sql

@@ -1,10 +1,12 @@
 SELECT '[1,2,3]'::vector;
 SELECT '[-1,2,3]'::vector;
+SELECT ' [ 1,  2 ,    3  ] '::vector(3);
 SELECT '[1.23456]'::vector;
 SELECT '[hello,1]'::vector;
 SELECT '[NaN,1]'::vector;
 SELECT '[Infinity,1]'::vector;
 SELECT '[-Infinity,1]'::vector;
+SELECT '[4e38,1]'::vector;
 SELECT '[1,2,3'::vector;
 SELECT '[1,2,3]9'::vector;
 SELECT '1,2,3'::vector;

vector.control

@@ -1,4 +1,4 @@
 comment = 'vector data type and ivfflat access method'
-default_version = '0.4.1'
+default_version = '0.4.3'
 module_pathname = '$libdir/vector'
 relocatable = true