Version bump to 0.4.4 [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
@@ -17,7 +19,7 @@ jobs:
           - postgres: 12
             os: ubuntu-20.04
           - postgres: 11
-            os: ubuntu-18.04
+            os: ubuntu-20.04
     steps:
       - uses: actions/checkout@v3
       - uses: ankane/setup-postgres@v1
@@ -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() }}
@@ -54,6 +60,7 @@ jobs:
           wget -q https://github.com/postgres/postgres/archive/refs/tags/REL_14_5.tar.gz
           tar xf REL_14_5.tar.gz
       - run: make prove_installcheck PROVE_FLAGS="-I ./postgres-REL_14_5/src/test/perl" PERL5LIB="/Users/runner/perl5/lib/perl5"
+      - run: make clean && /usr/local/opt/llvm@15/bin/scan-build --status-bugs make
   windows:
     runs-on: windows-latest
     if: ${{ !startsWith(github.ref_name, 'mac') }}
@@ -70,3 +77,22 @@ jobs:
           nmake /NOLOGO /F Makefile.win clean && ^
           nmake /NOLOGO /F Makefile.win uninstall
         shell: cmd
+  i386:
+    runs-on: ubuntu-latest
+    container:
+      image: debian:11
+      options: --platform linux/386
+    steps:
+      - run: apt-get update && apt-get install -y build-essential git libipc-run-perl postgresql-13 postgresql-server-dev-13 sudo
+      - run: service postgresql start
+      - run: |
+          git clone https://github.com/${{ 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 +1,26 @@
+## 0.4.4 (2023-06-12)
+
+- Improved error message for malformed vector literal
+- Fixed segmentation fault with text input
+- Fixed consecutive delimiters with text input
+
+## 0.4.3 (2023-06-10)
+
+- Improved cost estimation
+- Improved support for spaces with text input
+- Fixed infinite and NaN values with binary input
+- 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

Dockerfile

@@ -1,9 +1,11 @@
-FROM postgres:15
+ARG PG_MAJOR=15
+FROM postgres:$PG_MAJOR
+ARG PG_MAJOR
 
 COPY . /tmp/pgvector
 
 RUN apt-get update && \
-		apt-get install -y --no-install-recommends build-essential postgresql-server-dev-15 && \
+		apt-get install -y --no-install-recommends build-essential postgresql-server-dev-$PG_MAJOR && \
 		cd /tmp/pgvector && \
 		make clean && \
 		make OPTFLAGS="" && \
@@ -11,6 +13,6 @@ RUN apt-get update && \
 		mkdir /usr/share/doc/pgvector && \
 		cp LICENSE README.md /usr/share/doc/pgvector && \
 		rm -r /tmp/pgvector && \
-		apt-get remove -y build-essential postgresql-server-dev-15 && \
+		apt-get remove -y build-essential postgresql-server-dev-$PG_MAJOR && \
 		apt-get autoremove -y && \
 		rm -rf /var/lib/apt/lists/*

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": "0.4.4",
 	"maintainer": [
 		"Andrew Kane <andrew@ankane.org>"
 	],
@@ -20,7 +20,7 @@
 		"vector": {
 			"file": "sql/vector.sql",
 			"docfile": "README.md",
-			"version": "0.4.1",
+			"version": "0.4.4",
 			"abstract": "Open-source vector similarity search for Postgres"
 		}
 	},

Makefile

@@ -1,5 +1,5 @@
 EXTENSION = vector
-EXTVERSION = 0.4.1
+EXTVERSION = 0.4.4
 
 MODULE_big = vector
 DATA = $(wildcard sql/*--*.sql)
@@ -14,10 +14,16 @@ OPTFLAGS = -march=native
 # Mac ARM doesn't support -march=native
 ifeq ($(shell uname -s), Darwin)
 	ifeq ($(shell uname -p), arm)
+		# no difference with -march=armv8.5-a
 		OPTFLAGS =
 	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
@@ -62,3 +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 = 0.4.4
 
 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

@@ -2,13 +2,13 @@
 
 Open-source vector similarity search for Postgres
 
-```sql
-CREATE TABLE items (embedding vector(3));
-CREATE INDEX ON items USING ivfflat (embedding vector_l2_ops);
-SELECT * FROM items ORDER BY embedding <-> '[1,2,3]' LIMIT 5;
-```
+Supports
 
-Supports L2 distance, inner product, and cosine distance
+- 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)
 
@@ -17,53 +17,101 @@ Supports L2 distance, inner product, and cosine distance
 Compile and install the extension (supports Postgres 11+)
 
 ```sh
-git clone --branch v0.4.1 https://github.com/pgvector/pgvector.git
+cd /tmp
+git clone --branch v0.4.4 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
+See the [installation notes](#installation-notes) if you run into issues
 
-```sql
-CREATE EXTENSION vector;
-```
-
-You can also install it with [Docker](#docker), [Homebrew](#homebrew), [PGXN](#pgxn), 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 (embedding vector(3));
+CREATE TABLE items (id bigserial PRIMARY KEY, embedding vector(3));
 ```
 
-Insert values
+Insert vectors
 
 ```sql
-INSERT INTO items VALUES ('[1,2,3]'), ('[4,5,6]');
+INSERT INTO items (embedding) VALUES ('[1,2,3]'), ('[4,5,6]');
 ```
 
-Get the nearest neighbor by L2 distance
+Get the nearest neighbors by L2 distance
 
 ```sql
-SELECT * FROM items ORDER BY embedding <-> '[3,1,2]' LIMIT 1;
+SELECT * FROM items ORDER BY embedding <-> '[3,1,2]' LIMIT 5;
 ```
 
 Also supports inner product (`<#>`) and cosine distance (`<=>`)
 
 Note: `<#>` returns the negative inner product since Postgres only supports `ASC` order index scans on operators
 
-## Querying
+## Storing
 
-Use a `SELECT` clause to get the distance
+Create a new table with a vector column
 
 ```sql
-SELECT embedding <-> '[3,1,2]' AS distance FROM items;
+CREATE TABLE items (id bigserial PRIMARY KEY, embedding vector(3));
 ```
 
-Use a `WHERE` clause to get rows within a certain distance
+Or add a vector column to an existing table
+
+```sql
+ALTER TABLE items ADD COLUMN embedding vector(3);
+```
+
+Insert vectors
+
+```sql
+INSERT INTO items (embedding) VALUES ('[1,2,3]'), ('[4,5,6]');
+```
+
+Upsert vectors
+
+```sql
+INSERT INTO items (id, embedding) VALUES (1, '[1,2,3]'), (2, '[4,5,6]')
+    ON CONFLICT (id) DO UPDATE SET embedding = EXCLUDED.embedding;
+```
+
+Update vectors
+
+```sql
+UPDATE items SET embedding = '[1,2,3]' WHERE id = 1;
+```
+
+Delete vectors
+
+```sql
+DELETE FROM items WHERE id = 1;
+```
+
+## Querying
+
+Get the nearest neighbors to a vector
+
+```sql
+SELECT * FROM items ORDER BY embedding <-> '[3,1,2]' LIMIT 5;
+```
+
+Get the nearest neighbors to a row
+
+```sql
+SELECT * FROM items WHERE id != 1 ORDER BY embedding <-> (SELECT embedding FROM items WHERE id = 1) LIMIT 5;
+```
+
+Get rows within a certain distance
 
 ```sql
 SELECT * FROM items WHERE embedding <-> '[3,1,2]' < 5;
@@ -71,61 +119,89 @@ SELECT * FROM items WHERE embedding <-> '[3,1,2]' < 5;
 
 Note: Combine with `ORDER BY` and `LIMIT` to use an index
 
-Get the average of vectors
+#### Distances
+
+Get the distance
+
+```sql
+SELECT embedding <-> '[3,1,2]' AS distance FROM items;
+```
+
+For inner product, multiply by -1 (since `<#>` returns the negative inner product)
+
+```tsql
+SELECT (embedding <#> '[3,1,2]') * -1 AS inner_product FROM items;
+```
+
+For cosine similarity, use 1 - cosine distance
+
+```sql
+SELECT 1 - (embedding <=> '[3,1,2]') AS cosine_similarity FROM items;
+```
+
+#### Aggregates
+
+Average vectors
 
 ```sql
 SELECT AVG(embedding) FROM items;
 ```
 
+Average groups of vectors
+
+```sql
+SELECT category_id, AVG(embedding) FROM items GROUP BY category_id;
+```
+
 ## Indexing
 
-Speed up queries with an approximate index. Add an index for each distance function you want to use.
+By default, pgvector performs exact nearest neighbor search, which provides perfect recall.
+
+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.
+
+Three keys to achieving good recall are:
+
+1. Create the index *after* the table has some data
+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 `sqrt(lists)`
+
+Add an index for each distance function you want to use.
 
 L2 distance
 
-```sql
-CREATE INDEX ON items USING ivfflat (embedding vector_l2_ops);
-```
-
-Inner product
-
-```sql
-CREATE INDEX ON items USING ivfflat (embedding vector_ip_ops);
-```
-
-Cosine distance
-
-```sql
-CREATE INDEX ON items USING ivfflat (embedding vector_cosine_ops);
-```
-
-Indexes should be created after the table has some data for optimal clustering. Also, unlike typical indexes which only affect performance, you may see different results for queries after adding an approximate index. Vectors with up to 2,000 dimensions can be indexed.
-
-### Index Options
-
-Specify the number of inverted lists (100 by default)
-
 ```sql
 CREATE INDEX ON items USING ivfflat (embedding vector_l2_ops) WITH (lists = 100);
 ```
 
-A [good place to start](https://github.com/facebookresearch/faiss/issues/112) is `4 * sqrt(rows)`
+Inner product
+
+```sql
+CREATE INDEX ON items USING ivfflat (embedding vector_ip_ops) WITH (lists = 100);
+```
+
+Cosine distance
+
+```sql
+CREATE INDEX ON items USING ivfflat (embedding vector_cosine_ops) WITH (lists = 100);
+```
+
+Vectors with up to 2,000 dimensions can be indexed.
 
 ### Query Options
 
 Specify the number of probes (1 by default)
 
 ```sql
-SET ivfflat.probes = 1;
+SET ivfflat.probes = 10;
 ```
 
-A higher value improves 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)
+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)
 
 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;
 ```
@@ -147,46 +223,72 @@ 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) 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.
+
+```sql
+EXPLAIN ANALYZE SELECT * FROM items ORDER BY embedding <-> '[3,1,2]' LIMIT 5;
+```
+
+### Exact Search
+
 To speed up queries without an index, increase `max_parallel_workers_per_gather`.
 
 ```sql
 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.
+
+```tsql
+SELECT * FROM items ORDER BY embedding <#> '[3,1,2]' LIMIT 5;
+```
+
+### Approximate Search
+
 To speed up queries with an index, increase the number of inverted lists (at the expense of recall).
 
 ```sql
 CREATE INDEX ON items USING ivfflat (embedding vector_l2_ops) WITH (lists = 1000);
 ```
 
-Use `EXPLAIN ANALYZE` to debug performance.
-
-```sql
-EXPLAIN ANALYZE SELECT * FROM items ORDER BY embedding <-> '[3,1,2]' LIMIT 1;
-```
-
 ## Languages
 
 Use pgvector from any language with a Postgres client. You can even generate and store vectors in one language and query them in another.
@@ -195,17 +297,21 @@ 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)
 Node.js | [pgvector-node](https://github.com/pgvector/pgvector-node)
+Perl | [pgvector-perl](https://github.com/pgvector/pgvector-perl)
 PHP | [pgvector-php](https://github.com/pgvector/pgvector-php)
 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
 
@@ -219,10 +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
 
@@ -256,6 +363,46 @@ Function | Description
 --- | ---
 avg(vector) → vector | arithmetic mean
 
+## Installation Notes
+
+### Postgres Location
+
+If your machine has multiple Postgres installations, specify the path to [pg_config](https://www.postgresql.org/docs/current/app-pgconfig.html) with:
+
+```sh
+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). If `sudo` is needed for `make install`, use:
+
+```sh
+sudo --preserve-env=PG_CONFIG make install
+```
+
+### Missing Header
+
+If compilation fails with `fatal error: postgres.h: No such file or directory`, make sure Postgres development files are installed on the server.
+
+For Ubuntu and Debian, use:
+
+```sh
+sudo apt install postgresql-server-dev-15
+```
+
+Note: Replace `15` with your Postgres server version
+
+### Windows
+
+Support for Windows is currently experimental. Use `nmake` to build:
+
+```cmd
+set "PGROOT=C:\Program Files\PostgreSQL\15"
+git clone --branch v0.4.4 https://github.com/pgvector/pgvector.git
+cd pgvector
+nmake /F Makefile.win
+nmake /F Makefile.win install
+```
+
 ## Additional Installation Methods
 
 ### Docker
@@ -271,9 +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 v0.4.4 https://github.com/pgvector/pgvector.git
 cd pgvector
-docker build -t pgvector .
+docker build --build-arg PG_MAJOR=15 -t myuser/pgvector .
 ```
 
 ### Homebrew
@@ -281,9 +428,11 @@ docker build -t pgvector .
 With Homebrew Postgres, you can use:
 
 ```sh
-brew install pgvector/brew/pgvector
+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:
@@ -292,6 +441,28 @@ 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:
+
+```sh
+sudo yum install pgvector_15
+# or
+sudo dnf install pgvector_15
+```
+
+Note: Replace `15` with your Postgres server version
+
 ### conda-forge
 
 With Conda Postgres, install from [conda-forge](https://anaconda.org/conda-forge/pgvector) with:
@@ -302,16 +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)
-- DigitalOcean Managed Databases - vote or comment on [this page](https://ideas.digitalocean.com/app-framework-services/p/pgvector-extension-for-postgresql)
-- 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/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

sql/vector--0.4.3--0.4.4.sql

@@ -0,0 +1,2 @@
+-- complain if script is sourced in psql, rather than via CREATE EXTENSION
+\echo Use "ALTER EXTENSION vector UPDATE TO '0.4.4'" 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));
 
@@ -558,6 +568,21 @@ PrintKmeansMetrics(IvfflatBuildState * buildstate)
 }
 #endif
 
+/*
+ * Scan table for tuples to index
+ */
+static void
+ScanTable(IvfflatBuildState * buildstate)
+{
+#if PG_VERSION_NUM >= 120000
+	buildstate->reltuples = table_index_build_scan(buildstate->heap, buildstate->index, buildstate->indexInfo,
+												   true, true, BuildCallback, (void *) buildstate, NULL);
+#else
+	buildstate->reltuples = IndexBuildHeapScan(buildstate->heap, buildstate->index, buildstate->indexInfo,
+											   true, BuildCallback, (void *) buildstate, NULL);
+#endif
+}
+
 /*
  * Create entry pages
  */
@@ -565,7 +590,7 @@ static void
 CreateEntryPages(IvfflatBuildState * buildstate, ForkNumber forkNum)
 {
 	AttrNumber	attNums[] = {1};
-	Oid			sortOperators[] = {Float8LessOperator};
+	Oid			sortOperators[] = {Int4LessOperator};
 	Oid			sortCollations[] = {InvalidOid};
 	bool		nullsFirstFlags[] = {false};
 
@@ -575,25 +600,17 @@ CreateEntryPages(IvfflatBuildState * buildstate, ForkNumber forkNum)
 
 	/* Add tuples to sort */
 	if (buildstate->heap != NULL)
-	{
-#if PG_VERSION_NUM >= 120000
-		buildstate->reltuples = table_index_build_scan(buildstate->heap, buildstate->index, buildstate->indexInfo,
-													   true, true, BuildCallback, (void *) buildstate, NULL);
-#else
-		buildstate->reltuples = IndexBuildHeapScan(buildstate->heap, buildstate->index, buildstate->indexInfo,
-												   true, BuildCallback, (void *) buildstate, NULL);
-#endif
-	}
+		IvfflatBench("assign tuples", ScanTable(buildstate));
 
 	/* Sort */
-	tuplesort_performsort(buildstate->sortstate);
+	IvfflatBench("sort tuples", tuplesort_performsort(buildstate->sortstate));
 
 #ifdef IVFFLAT_KMEANS_DEBUG
 	PrintKmeansMetrics(buildstate);
 #endif
 
 	/* Insert */
-	InsertTuples(buildstate->index, buildstate, forkNum);
+	IvfflatBench("load tuples", InsertTuples(buildstate->index, buildstate, forkNum));
 	tuplesort_end(buildstate->sortstate);
 }
 
@@ -611,7 +628,7 @@ BuildIndex(Relation heap, Relation index, IndexInfo *indexInfo,
 	/* Create pages */
 	CreateMetaPage(index, buildstate->dimensions, buildstate->lists, forkNum);
 	CreateListPages(index, buildstate->centers, buildstate->dimensions, buildstate->lists, forkNum, &buildstate->listInfo);
-	IvfflatBench("CreateEntryPages", CreateEntryPages(buildstate, forkNum));
+	CreateEntryPages(buildstate, forkNum);
 
 	FreeBuildState(buildstate);
 }

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

@@ -111,6 +111,7 @@ GetScanItems(IndexScanDesc scan, Datum value)
 	Datum		datum;
 	bool		isnull;
 	TupleDesc	tupdesc = RelationGetDescr(scan->indexRelation);
+	double		tuples = 0;
 
 #if PG_VERSION_NUM >= 120000
 	TupleTableSlot *slot = MakeSingleTupleTableSlot(so->tupdesc, &TTSOpsVirtual);
@@ -159,6 +160,8 @@ GetScanItems(IndexScanDesc scan, Datum value)
 				ExecStoreVirtualTuple(slot);
 
 				tuplesort_puttupleslot(so->sortstate, slot);
+
+				tuples++;
 			}
 
 			searchPage = IvfflatPageGetOpaque(page)->nextblkno;
@@ -167,6 +170,15 @@ 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 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

@@ -42,7 +42,7 @@ CheckDims(Vector * a, Vector * b)
 }
 
 /*
- * Ensure expected dimension
+ * Ensure expected dimensions
  */
 static inline void
 CheckExpectedDim(int32 typmod, int dim)
@@ -53,7 +53,9 @@ CheckExpectedDim(int32 typmod, int dim)
 				 errmsg("expected %d dimensions, not %d", typmod, dim)));
 }
 
-
+/*
+ * Ensure valid dimensions
+ */
 static inline void
 CheckDim(int dim)
 {
@@ -79,13 +81,28 @@ CheckElement(float value)
 				(errcode(ERRCODE_DATA_EXCEPTION),
 				 errmsg("NaN not allowed in vector")));
 
-
 	if (isinf(value))
 		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;
+}
+
 /*
  * Check state array
  */
@@ -100,7 +117,7 @@ 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)
 {
@@ -110,30 +127,6 @@ float_overflow_error(void)
 }
 #endif
 
-/*
- * Print vector - useful for debugging
- */
-void
-PrintVector(char *msg, Vector * vector)
-{
-	StringInfoData buf;
-	int			dim = vector->dim;
-	int			i;
-
-	initStringInfo(&buf);
-
-	appendStringInfoChar(&buf, '[');
-	for (i = 0; i < dim; i++)
-	{
-		if (i > 0)
-			appendStringInfoString(&buf, ",");
-		appendStringInfoString(&buf, float8out_internal(vector->x[i]));
-	}
-	appendStringInfoChar(&buf, ']');
-
-	elog(INFO, "%s = %s", msg, buf.data);
-}
-
 /*
  * Convert textual representation to internal representation
  */
@@ -149,11 +142,15 @@ vector_in(PG_FUNCTION_ARGS)
 	char	   *pt;
 	char	   *stringEnd;
 	Vector	   *result;
+	char	   *lit = pstrdup(str);
+
+	while (vector_isspace(*str))
+		str++;
 
 	if (*str != '[')
 		ereport(ERROR,
 				(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
-				 errmsg("malformed vector literal: \"%s\"", str),
+				 errmsg("malformed vector literal: \"%s\"", lit),
 				 errdetail("Vector contents must start with \"[\".")));
 
 	str++;
@@ -167,6 +164,15 @@ vector_in(PG_FUNCTION_ARGS)
 					(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
 					 errmsg("vector cannot have more than %d dimensions", VECTOR_MAX_DIM)));
 
+		while (vector_isspace(*pt))
+			pt++;
+
+		/* Check for empty string like float4in */
+		if (*pt == '\0')
+			ereport(ERROR,
+					(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
+					 errmsg("invalid input syntax for type vector: \"%s\"", lit)));
+
 		/* Use strtof like float4in to avoid a double-rounding problem */
 		x[dim] = strtof(pt, &stringEnd);
 		CheckElement(x[dim]);
@@ -175,33 +181,53 @@ vector_in(PG_FUNCTION_ARGS)
 		if (stringEnd == pt)
 			ereport(ERROR,
 					(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
-					 errmsg("invalid input syntax for type vector: \"%s\"", pt)));
+					 errmsg("invalid input syntax for type vector: \"%s\"", lit)));
+
+		while (vector_isspace(*stringEnd))
+			stringEnd++;
 
 		if (*stringEnd != '\0' && *stringEnd != ']')
 			ereport(ERROR,
 					(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
-					 errmsg("invalid input syntax for type vector: \"%s\"", pt)));
+					 errmsg("invalid input syntax for type vector: \"%s\"", lit)));
 
 		pt = strtok(NULL, ",");
 	}
 
-	if (*stringEnd != ']')
+	if (stringEnd == NULL || *stringEnd != ']')
 		ereport(ERROR,
 				(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
-				 errmsg("malformed vector literal"),
+				 errmsg("malformed vector literal: \"%s\"", lit),
 				 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"),
+				 errmsg("malformed vector literal: \"%s\"", lit),
 				 errdetail("Junk after closing right brace.")));
 
+	/* Ensure no consecutive delimiters since strtok skips */
+	for (pt = lit + 1; *pt != '\0'; pt++)
+	{
+		if (pt[-1] == ',' && *pt == ',')
+			ereport(ERROR,
+					(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
+					 errmsg("malformed vector literal: \"%s\"", lit)));
+	}
+
 	if (dim < 1)
 		ereport(ERROR,
 				(errcode(ERRCODE_DATA_EXCEPTION),
 				 errmsg("vector must have at least 1 dimension")));
 
+	pfree(lit);
+
 	CheckExpectedDim(typmod, dim);
 
 	result = InitVector(dim);
@@ -272,6 +298,18 @@ vector_out(PG_FUNCTION_ARGS)
 	PG_RETURN_CSTRING(buf);
 }
 
+/*
+ * Print vector - useful for debugging
+ */
+void
+PrintVector(char *msg, Vector * vector)
+{
+	char	   *out = DatumGetPointer(DirectFunctionCall1(vector_out, PointerGetDatum(vector)));
+
+	elog(INFO, "%s = %s", msg, out);
+	pfree(out);
+}
+
 /*
  * Convert type modifier
  */
@@ -330,7 +368,10 @@ vector_recv(PG_FUNCTION_ARGS)
 
 	result = InitVector(dim);
 	for (i = 0; i < dim; i++)
+	{
 		result->x[i] = pq_getmsgfloat4(buf);
+		CheckElement(result->x[i]);
+	}
 
 	PG_RETURN_POINTER(result);
 }
@@ -396,10 +437,8 @@ 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 +448,7 @@ 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)
@@ -436,17 +476,19 @@ 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);
 }
@@ -467,6 +509,7 @@ l2_distance(PG_FUNCTION_ARGS)
 
 	CheckDims(a, b);
 
+	/* Auto-vectorized */
 	for (int i = 0; i < a->dim; i++)
 	{
 		diff = ax[i] - bx[i];
@@ -493,6 +536,7 @@ vector_l2_squared_distance(PG_FUNCTION_ARGS)
 
 	CheckDims(a, b);
 
+	/* Auto-vectorized */
 	for (int i = 0; i < a->dim; i++)
 	{
 		diff = ax[i] - bx[i];
@@ -517,6 +561,7 @@ inner_product(PG_FUNCTION_ARGS)
 
 	CheckDims(a, b);
 
+	/* Auto-vectorized */
 	for (int i = 0; i < a->dim; i++)
 		distance += ax[i] * bx[i];
 
@@ -538,6 +583,7 @@ vector_negative_inner_product(PG_FUNCTION_ARGS)
 
 	CheckDims(a, b);
 
+	/* Auto-vectorized */
 	for (int i = 0; i < a->dim; i++)
 		distance += ax[i] * bx[i];
 
@@ -561,6 +607,7 @@ cosine_distance(PG_FUNCTION_ARGS)
 
 	CheckDims(a, b);
 
+	/* Auto-vectorized */
 	for (int i = 0; i < a->dim; i++)
 	{
 		distance += ax[i] * bx[i];
@@ -587,6 +634,7 @@ vector_spherical_distance(PG_FUNCTION_ARGS)
 
 	CheckDims(a, b);
 
+	/* Auto-vectorized */
 	for (int i = 0; i < a->dim; i++)
 		distance += a->x[i] * b->x[i];
 
@@ -622,6 +670,7 @@ vector_norm(PG_FUNCTION_ARGS)
 	float	   *ax = a->x;
 	double		norm = 0.0;
 
+	/* Auto-vectorized */
 	for (int i = 0; i < a->dim; i++)
 		norm += ax[i] * ax[i];
 
@@ -646,9 +695,18 @@ vector_add(PG_FUNCTION_ARGS)
 
 	result = InitVector(a->dim);
 	rx = result->x;
+
+	/* Auto-vectorized */
 	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]))
+			float_overflow_error();
+	}
+
 	PG_RETURN_POINTER(result);
 }
 
@@ -670,9 +728,18 @@ vector_sub(PG_FUNCTION_ARGS)
 
 	result = InitVector(a->dim);
 	rx = result->x;
+
+	/* Auto-vectorized */
 	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]))
+			float_overflow_error();
+	}
+
 	PG_RETURN_POINTER(result);
 }
 
@@ -818,12 +885,12 @@ 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
 	{
@@ -831,10 +898,11 @@ vector_accum(PG_FUNCTION_ARGS)
 		{
 			double		v = statevalues[i + 1] + x[i];
 
+			/* Check for overflow */
 			if (isinf(v))
 				float_overflow_error();
 
-			statedatums[i + 1] = Float8GetDatumFast(v);
+			statedatums[i + 1] = Float8GetDatum(v);
 		}
 	}
 
@@ -879,7 +947,7 @@ 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)
 	{
@@ -887,7 +955,7 @@ 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
 	{
@@ -899,14 +967,15 @@ vector_combine(PG_FUNCTION_ARGS)
 		{
 			double		v = statevalues1[i] + statevalues2[i];
 
+			/* Check for overflow */
 			if (isinf(v))
 				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,
@@ -929,7 +998,6 @@ vector_avg(PG_FUNCTION_ARGS)
 	float8		n;
 	uint16		dim;
 	Vector	   *result;
-	float		v;
 
 	/* Check array before using */
 	statevalues = CheckStateArray(statearray, "vector_avg");
@@ -941,12 +1009,12 @@ 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;
+		CheckElement(result->x[i]);
 	}
 
 	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 
 -------------
@@ -22,10 +26,28 @@ SELECT round(vector_norm('[1,1]')::numeric, 5);
  1.41421
 (1 row)
 
-SELECT round(l2_distance('[1,2]', '[0,0]')::numeric, 5);
-  round  
----------
- 2.23607
+SELECT vector_norm('[3,4]');
+ vector_norm 
+-------------
+           5
+(1 row)
+
+SELECT vector_norm('[0,1]');
+ vector_norm 
+-------------
+           1
+(1 row)
+
+SELECT l2_distance('[0,0]', '[3,4]');
+ l2_distance 
+-------------
+           5
+(1 row)
+
+SELECT l2_distance('[0,0]', '[0,1]');
+ l2_distance 
+-------------
+           1
 (1 row)
 
 SELECT l2_distance('[1,2]', '[3]');
@@ -38,10 +60,10 @@ SELECT inner_product('[1,2]', '[3,4]');
 
 SELECT inner_product('[1,2]', '[3]');
 ERROR:  different vector dimensions 2 and 1
-SELECT round(cosine_distance('[1,2]', '[2,4]')::numeric, 5);
-  round  
----------
- 0.00000
+SELECT cosine_distance('[1,2]', '[2,4]');
+ cosine_distance 
+-----------------
+               0
 (1 row)
 
 SELECT cosine_distance('[1,2]', '[0,0]');
@@ -84,3 +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

@@ -4,10 +4,22 @@ SELECT '[1,2,3]'::vector;
  [1,2,3]
 (1 row)
 
-SELECT '[-1,2,3]'::vector;
-  vector  
-----------
- [-1,2,3]
+SELECT '[-1,-2,-3]'::vector;
+   vector   
+------------
+ [-1,-2,-3]
+(1 row)
+
+SELECT '[1.,2.,3.]'::vector;
+ vector  
+---------
+ [1,2,3]
+(1 row)
+
+SELECT ' [ 1,  2 ,    3  ] '::vector;
+ vector  
+---------
+ [1,2,3]
 (1 row)
 
 SELECT '[1.23456]'::vector;
@@ -17,7 +29,7 @@ SELECT '[1.23456]'::vector;
 (1 row)
 
 SELECT '[hello,1]'::vector;
-ERROR:  invalid input syntax for type vector: "hello"
+ERROR:  invalid input syntax for type vector: "[hello,1]"
 LINE 1: SELECT '[hello,1]'::vector;
                ^
 SELECT '[NaN,1]'::vector;
@@ -32,13 +44,35 @@ SELECT '[-Infinity,1]'::vector;
 ERROR:  infinite value not allowed in vector
 LINE 1: SELECT '[-Infinity,1]'::vector;
                ^
+SELECT '[1.5e38,-1.5e38]'::vector;
+       vector       
+--------------------
+ [1.5e+38,-1.5e+38]
+(1 row)
+
+SELECT '[1.5e+38,-1.5e+38]'::vector;
+       vector       
+--------------------
+ [1.5e+38,-1.5e+38]
+(1 row)
+
+SELECT '[1.5e-38,-1.5e-38]'::vector;
+       vector       
+--------------------
+ [1.5e-38,-1.5e-38]
+(1 row)
+
+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
+ERROR:  malformed vector literal: "[1,2,3"
 LINE 1: SELECT '[1,2,3'::vector;
                ^
 DETAIL:  Unexpected end of input.
 SELECT '[1,2,3]9'::vector;
-ERROR:  malformed vector literal
+ERROR:  malformed vector literal: "[1,2,3]9"
 LINE 1: SELECT '[1,2,3]9'::vector;
                ^
 DETAIL:  Junk after closing right brace.
@@ -47,14 +81,36 @@ ERROR:  malformed vector literal: "1,2,3"
 LINE 1: SELECT '1,2,3'::vector;
                ^
 DETAIL:  Vector contents must start with "[".
+SELECT '['::vector;
+ERROR:  malformed vector literal: "["
+LINE 1: SELECT '['::vector;
+               ^
+DETAIL:  Unexpected end of input.
+SELECT '[,'::vector;
+ERROR:  malformed vector literal: "[,"
+LINE 1: SELECT '[,'::vector;
+               ^
+DETAIL:  Unexpected end of input.
 SELECT '[]'::vector;
 ERROR:  vector must have at least 1 dimension
 LINE 1: SELECT '[]'::vector;
                ^
 SELECT '[1,]'::vector;
-ERROR:  invalid input syntax for type vector: "]"
+ERROR:  invalid input syntax for type vector: "[1,]"
 LINE 1: SELECT '[1,]'::vector;
                ^
+SELECT '[1a]'::vector;
+ERROR:  invalid input syntax for type vector: "[1a]"
+LINE 1: SELECT '[1a]'::vector;
+               ^
+SELECT '[1,,3]'::vector;
+ERROR:  malformed vector literal: "[1,,3]"
+LINE 1: SELECT '[1,,3]'::vector;
+               ^
+SELECT '[1, ,3]'::vector;
+ERROR:  invalid input syntax for type vector: "[1, ,3]"
+LINE 1: SELECT '[1, ,3]'::vector;
+               ^
 SELECT '[1,2,3]'::vector(2);
 ERROR:  expected 2 dimensions, not 3
 SELECT unnest('{"[1,2,3]", "[4,5,6]"}'::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

@@ -1,16 +1,22 @@
 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]');
-SELECT round(vector_norm('[1,1]')::numeric, 5);
 
-SELECT round(l2_distance('[1,2]', '[0,0]')::numeric, 5);
+SELECT round(vector_norm('[1,1]')::numeric, 5);
+SELECT vector_norm('[3,4]');
+SELECT vector_norm('[0,1]');
+
+SELECT l2_distance('[0,0]', '[3,4]');
+SELECT l2_distance('[0,0]', '[0,1]');
 SELECT l2_distance('[1,2]', '[3]');
 
 SELECT inner_product('[1,2]', '[3,4]');
 SELECT inner_product('[1,2]', '[3]');
 
-SELECT round(cosine_distance('[1,2]', '[2,4]')::numeric, 5);
+SELECT cosine_distance('[1,2]', '[2,4]');
 SELECT cosine_distance('[1,2]', '[0,0]');
 SELECT cosine_distance('[1,1]', '[1,1]');
 SELECT cosine_distance('[1,1]', '[-1,-1]');
@@ -20,3 +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

@@ -1,15 +1,26 @@
 SELECT '[1,2,3]'::vector;
-SELECT '[-1,2,3]'::vector;
+SELECT '[-1,-2,-3]'::vector;
+SELECT '[1.,2.,3.]'::vector;
+SELECT ' [ 1,  2 ,    3  ] '::vector;
 SELECT '[1.23456]'::vector;
 SELECT '[hello,1]'::vector;
 SELECT '[NaN,1]'::vector;
 SELECT '[Infinity,1]'::vector;
 SELECT '[-Infinity,1]'::vector;
+SELECT '[1.5e38,-1.5e38]'::vector;
+SELECT '[1.5e+38,-1.5e+38]'::vector;
+SELECT '[1.5e-38,-1.5e-38]'::vector;
+SELECT '[4e38,1]'::vector;
 SELECT '[1,2,3'::vector;
 SELECT '[1,2,3]9'::vector;
 SELECT '1,2,3'::vector;
+SELECT '['::vector;
+SELECT '[,'::vector;
 SELECT '[]'::vector;
 SELECT '[1,]'::vector;
+SELECT '[1a]'::vector;
+SELECT '[1,,3]'::vector;
+SELECT '[1, ,3]'::vector;
 SELECT '[1,2,3]'::vector(2);
 
 SELECT unnest('{"[1,2,3]", "[4,5,6]"}'::vector[]);

vector.control

@@ -1,4 +1,4 @@
 comment = 'vector data type and ivfflat access method'
-default_version = '0.4.1'
+default_version = '0.4.4'
 module_pathname = '$libdir/vector'
 relocatable = true