[python] Bump version: 0.6.4 → 0.6.5

changed the error msg for query with wrong vector dim thus need this
change to pass the nodejs tests.

.bumpversion.cfg

@@ -1,5 +1,5 @@
 [bumpversion]
-current_version = 0.4.8
+current_version = 0.4.13
 commit = True
 message = Bump version: {current_version} → {new_version}
 tag = True
@@ -9,4 +9,4 @@ tag_name = v{new_version}
 
 [bumpversion:file:rust/ffi/node/Cargo.toml]
 
-[bumpversion:file:rust/vectordb/Cargo.toml]
+[bumpversion:file:rust/lancedb/Cargo.toml]

.cargo/config.toml

@@ -25,6 +25,7 @@ rustflags = [
     "-Dclippy::dbg_macro",
     # not too much we can do to avoid multiple crate versions
     "-Aclippy::multiple-crate-versions",
+    "-Aclippy::wildcard_dependencies",
 ]
 
 [target.x86_64-unknown-linux-gnu]
@@ -32,3 +33,8 @@ rustflags = ["-C", "target-cpu=haswell", "-C", "target-feature=+avx2,+fma,+f16c"
 
 [target.aarch64-apple-darwin]
 rustflags = ["-C", "target-cpu=apple-m1", "-C", "target-feature=+neon,+fp16,+fhm,+dotprod"]
+
+# Not all Windows systems have the C runtime installed, so this avoids library
+# not found errors on systems that are missing it.
+[target.x86_64-pc-windows-msvc]
+rustflags = ["-Ctarget-feature=+crt-static"]

.github/workflows/build_linux_wheel/action.yml

@@ -0,0 +1,58 @@
+# We create a composite action to be re-used both for testing and for releasing
+name: build-linux-wheel
+description: "Build a manylinux wheel for lance"
+inputs:
+  python-minor-version:
+    description: "8, 9, 10, 11, 12"
+    required: true
+  args:
+    description: "--release"
+    required: false
+    default: ""
+  arm-build:
+    description: "Build for arm64 instead of x86_64"
+    # Note: this does *not* mean the host is arm64, since we might be cross-compiling.
+    required: false
+    default: "false"
+runs:
+  using: "composite"
+  steps:
+    - name: CONFIRM ARM BUILD
+      shell: bash
+      run: |
+        echo "ARM BUILD: ${{ inputs.arm-build }}"
+    - name: Build x86_64 Manylinux wheel
+      if: ${{ inputs.arm-build == 'false' }}
+      uses: PyO3/maturin-action@v1
+      with:
+        command: build
+        working-directory: python
+        target: x86_64-unknown-linux-gnu
+        manylinux: "2_17"
+        args: ${{ inputs.args }}
+        before-script-linux: |
+          set -e
+          yum install -y openssl-devel \
+            && curl -L https://github.com/protocolbuffers/protobuf/releases/download/v24.4/protoc-24.4-linux-$(uname -m).zip > /tmp/protoc.zip \
+            && unzip /tmp/protoc.zip -d /usr/local \
+            && rm /tmp/protoc.zip
+    - name: Build Arm Manylinux Wheel
+      if: ${{ inputs.arm-build == 'true' }}
+      uses: PyO3/maturin-action@v1
+      with:
+        command: build
+        working-directory: python
+        target: aarch64-unknown-linux-gnu
+        manylinux: "2_24"
+        args: ${{ inputs.args }}
+        before-script-linux: |
+          set -e
+          apt install -y unzip
+          if [ $(uname -m) = "x86_64" ]; then
+            PROTOC_ARCH="x86_64"
+          else
+            PROTOC_ARCH="aarch_64"
+          fi
+          curl -L https://github.com/protocolbuffers/protobuf/releases/download/v24.4/protoc-24.4-linux-$PROTOC_ARCH.zip > /tmp/protoc.zip \
+            && unzip /tmp/protoc.zip -d /usr/local \
+            && rm /tmp/protoc.zip

.github/workflows/build_mac_wheel/action.yml

@@ -0,0 +1,25 @@
+# We create a composite action to be re-used both for testing and for releasing
+name: build_wheel
+description: "Build a lance wheel"
+inputs:
+  python-minor-version:
+    description: "8, 9, 10, 11"
+    required: true
+  args:
+    description: "--release"
+    required: false
+    default: ""
+runs:
+  using: "composite"
+  steps:
+    - name: Install macos dependency
+      shell: bash
+      run: |
+        brew install protobuf
+    - name: Build wheel
+      uses: PyO3/maturin-action@v1
+      with:
+        command: build
+        args: ${{ inputs.args }}
+        working-directory: python
+        interpreter: 3.${{ inputs.python-minor-version }}

.github/workflows/build_windows_wheel/action.yml

@@ -0,0 +1,33 @@
+# We create a composite action to be re-used both for testing and for releasing
+name: build_wheel
+description: "Build a lance wheel"
+inputs:
+  python-minor-version:
+    description: "8, 9, 10, 11"
+    required: true
+  args:
+    description: "--release"
+    required: false
+    default: ""
+runs:
+  using: "composite"
+  steps:
+    - name: Install Protoc v21.12
+      working-directory: C:\
+      run: |
+        New-Item -Path 'C:\protoc' -ItemType Directory
+        Set-Location C:\protoc
+        Invoke-WebRequest https://github.com/protocolbuffers/protobuf/releases/download/v21.12/protoc-21.12-win64.zip -OutFile C:\protoc\protoc.zip
+        7z x protoc.zip
+        Add-Content $env:GITHUB_PATH "C:\protoc\bin"
+      shell: powershell
+    - name: Build wheel
+      uses: PyO3/maturin-action@v1
+      with:
+        command: build
+        args: ${{ inputs.args }}
+        working-directory: python
+    - uses: actions/upload-artifact@v3
+      with:
+        name: windows-wheels
+        path: python\target\wheels

.github/workflows/cargo-publish.yml

@@ -26,4 +26,4 @@ jobs:
           sudo apt install -y protobuf-compiler libssl-dev
       - name: Publish the package
         run: |
-          cargo publish -p vectordb --all-features --token ${{ secrets.CARGO_REGISTRY_TOKEN }}
+          cargo publish -p lancedb --all-features --token ${{ secrets.CARGO_REGISTRY_TOKEN }}

.github/workflows/docs.yml

@@ -24,10 +24,14 @@ jobs:
     environment:
       name: github-pages
       url: ${{ steps.deployment.outputs.page_url }}
-    runs-on: ubuntu-22.04
+    runs-on: buildjet-8vcpu-ubuntu-2204
     steps:
       - name: Checkout
         uses: actions/checkout@v4
+      - name: Install dependecies needed for ubuntu
+        run: |
+          sudo apt install -y protobuf-compiler libssl-dev
+          rustup update && rustup default        
       - name: Set up Python
         uses: actions/setup-python@v5
         with:

.github/workflows/docs_test.yml

@@ -24,16 +24,22 @@ env:
 jobs:
   test-python:
     name: Test doc python code
-    runs-on: "ubuntu-latest"
+    runs-on: "buildjet-8vcpu-ubuntu-2204"
     steps:
     - name: Checkout
       uses: actions/checkout@v4
+    - name: Install dependecies needed for ubuntu
+      run: |
+        sudo apt install -y protobuf-compiler libssl-dev
+        rustup update && rustup default
     - name: Set up Python
       uses: actions/setup-python@v5
       with:
         python-version: 3.11
         cache: "pip"
         cache-dependency-path: "docs/test/requirements.txt"
+    - name: Rust cache
+      uses: swatinem/rust-cache@v2        
     - name: Build Python
       working-directory: docs/test
       run:
@@ -48,8 +54,8 @@ jobs:
         for d in *; do cd "$d"; echo "$d".py; python "$d".py; cd ..; done
   test-node:
     name: Test doc nodejs code
-    runs-on: "ubuntu-latest"
+    runs-on: "buildjet-8vcpu-ubuntu-2204"
-    timeout-minutes: 45
+    timeout-minutes: 60
     strategy:
       fail-fast: false
     steps:
@@ -65,6 +71,7 @@ jobs:
     - name: Install dependecies needed for ubuntu
       run: |
         sudo apt install -y protobuf-compiler libssl-dev
+        rustup update && rustup default
     - name: Rust cache
       uses: swatinem/rust-cache@v2
     - name: Install node dependencies

.github/workflows/node.yml

@@ -24,27 +24,6 @@ env:
   RUST_BACKTRACE: "1"
 
 jobs:
-  lint:
-    name: Lint
-    runs-on: ubuntu-22.04
-    defaults:
-      run:
-        shell: bash
-        working-directory: node
-    steps:
-    - uses: actions/checkout@v4
-      with:
-        fetch-depth: 0
-        lfs: true
-    - uses: actions/setup-node@v3
-      with:
-        node-version: 20
-        cache: 'npm'
-        cache-dependency-path: node/package-lock.json
-    - name: Lint
-      run: |
-        npm ci
-        npm run lint
   linux:
     name: Linux (Node ${{ matrix.node-version }})
     timeout-minutes: 30

.github/workflows/nodejs.yml

@@ -49,6 +49,7 @@ jobs:
         cargo clippy --all --all-features -- -D warnings
         npm ci
         npm run lint
+        npm run chkformat
   linux:
     name: Linux (NodeJS ${{ matrix.node-version }})
     timeout-minutes: 30
@@ -111,4 +112,3 @@ jobs:
     - name: Test
       run: |
         npm run test
-

.github/workflows/npm-publish.yml

@@ -80,10 +80,25 @@ jobs:
           - arch: x86_64
             runner: ubuntu-latest
           - arch: aarch64
-            runner: buildjet-4vcpu-ubuntu-2204-arm
+            # For successful fat LTO builds, we need a large runner to avoid OOM errors.
+            runner: buildjet-16vcpu-ubuntu-2204-arm
     steps:
       - name: Checkout
         uses: actions/checkout@v4
+      # Buildjet aarch64 runners have only 1.5 GB RAM per core, vs 3.5 GB per core for
+      # x86_64 runners. To avoid OOM errors on ARM, we create a swap file.
+      - name: Configure aarch64 build
+        if: ${{ matrix.config.arch == 'aarch64' }}
+        run: |
+          free -h
+          sudo fallocate -l 16G /swapfile
+          sudo chmod 600 /swapfile
+          sudo mkswap /swapfile
+          sudo swapon /swapfile
+          echo "/swapfile swap swap defaults 0 0" >> sudo /etc/fstab
+          # print info
+          swapon --show
+          free -h
       - name: Build Linux Artifacts
         run: |
           bash ci/build_linux_artifacts.sh ${{ matrix.config.arch }}

.github/workflows/pypi-publish.yml

@@ -2,30 +2,91 @@ name: PyPI Publish
 
 on:
   release:
-    types: [ published ]
+    types: [published]
 
 jobs:
-  publish:
+  linux:
-    runs-on: ubuntu-latest
+    timeout-minutes: 60
-    # Only runs on tags that matches the python-make-release action
+    strategy:
-    if: startsWith(github.ref, 'refs/tags/python-v')
+      matrix:
-    defaults:
+        python-minor-version: ["8"]
-      run:
+        platform:
-        shell: bash
+          - x86_64
-        working-directory: python
+          - aarch64
+    runs-on: "ubuntu-22.04"
     steps:
       - uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+          lfs: true
       - name: Set up Python
-        uses: actions/setup-python@v5
+        uses: actions/setup-python@v4
         with:
-          python-version: "3.8"
+          python-version: 3.${{ matrix.python-minor-version }}
-      - name: Build distribution
+      - uses: ./.github/workflows/build_linux_wheel
-        run: |
-          ls -la
-          pip install wheel setuptools --upgrade
-          python setup.py sdist bdist_wheel
-      - name: Publish
-        uses: pypa/gh-action-pypi-publish@v1.8.5
         with:
-          password: ${{ secrets.LANCEDB_PYPI_API_TOKEN }}
+          python-minor-version: ${{ matrix.python-minor-version }}
-          packages-dir: python/dist
+          args: "--release --strip"
+          arm-build: ${{ matrix.platform == 'aarch64' }}
+      - uses: ./.github/workflows/upload_wheel
+        with:
+          token: ${{ secrets.LANCEDB_PYPI_API_TOKEN }}
+          repo: "pypi"
+  mac:
+    timeout-minutes: 60
+    runs-on: ${{ matrix.config.runner }}
+    strategy:
+      matrix:
+        python-minor-version: ["8"]
+        config:
+          - target: x86_64-apple-darwin
+            runner: macos-13
+          - target: aarch64-apple-darwin
+            runner: macos-14
+    env:
+      MACOSX_DEPLOYMENT_TARGET: 10.15
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          ref: ${{ inputs.ref }}
+          fetch-depth: 0
+          lfs: true
+      - name: Set up Python
+        uses: actions/setup-python@v4
+        with:
+          python-version: 3.12
+      - uses: ./.github/workflows/build_mac_wheel
+        with:
+          python-minor-version: ${{ matrix.python-minor-version }}
+          args: "--release --strip --target ${{ matrix.config.target }}"
+      - uses: ./.github/workflows/upload_wheel
+        with:
+          python-minor-version: ${{ matrix.python-minor-version }}
+          token: ${{ secrets.LANCEDB_PYPI_API_TOKEN }}
+          repo: "pypi"
+  windows:
+    timeout-minutes: 60
+    runs-on: windows-latest
+    strategy:
+      matrix:
+        python-minor-version: ["8"]
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          ref: ${{ inputs.ref }}
+          fetch-depth: 0
+          lfs: true
+      - name: Set up Python
+        uses: actions/setup-python@v4
+        with:
+          python-version: 3.${{ matrix.python-minor-version }}
+      - uses: ./.github/workflows/build_windows_wheel
+        with:
+          python-minor-version: ${{ matrix.python-minor-version }}
+          args: "--release --strip"
+          vcpkg_token: ${{ secrets.VCPKG_GITHUB_PACKAGES }}
+      - uses: ./.github/workflows/upload_wheel
+        with:
+          python-minor-version: ${{ matrix.python-minor-version }}
+          token: ${{ secrets.LANCEDB_PYPI_API_TOKEN }}
+          repo: "pypi"

.github/workflows/python.yml

@@ -14,49 +14,133 @@ concurrency:
   cancel-in-progress: true
 
 jobs:
-  linux:
+  lint:
+    name: "Lint"
     timeout-minutes: 30
-    strategy:
-      matrix:
-        python-minor-version: [ "8", "11" ]
     runs-on: "ubuntu-22.04"
     defaults:
       run:
         shell: bash
         working-directory: python
     steps:
-    - uses: actions/checkout@v4
+      - uses: actions/checkout@v4
-      with:
+        with:
-        fetch-depth: 0
+          fetch-depth: 0
-        lfs: true
+          lfs: true
-    - name: Set up Python
+      - name: Set up Python
-      uses: actions/setup-python@v5
+        uses: actions/setup-python@v5
-      with:
+        with:
-        python-version: 3.${{ matrix.python-minor-version }}
+          python-version: "3.11"
-    - name: Install lancedb
+      - name: Install ruff
-      run: |
+        run: |
-        pip install -e .[tests]
+          pip install ruff==0.2.2
-        pip install tantivy@git+https://github.com/quickwit-oss/tantivy-py#164adc87e1a033117001cf70e38c82a53014d985
+      - name: Format check
-        pip install pytest pytest-mock ruff
+        run: ruff format --check .
-    - name: Format check
+      - name: Lint
-      run: ruff format --check .
+        run: ruff .
-    - name: Lint
+  doctest:
-      run: ruff .
+    name: "Doctest"
-    - name: Run tests
+    timeout-minutes: 30
-      run: pytest -m "not slow" -x -v --durations=30 tests
+    runs-on: "ubuntu-22.04"
-    - name: doctest
+    defaults:
-      run: pytest --doctest-modules lancedb
+      run:
+        shell: bash
+        working-directory: python
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+          lfs: true
+      - name: Set up Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: "3.11"
+          cache: "pip"
+      - name: Install protobuf
+        run: |
+          sudo apt update
+          sudo apt install -y protobuf-compiler
+      - uses: Swatinem/rust-cache@v2
+        with:
+          workspaces: python
+      - name: Install
+        run: |
+          pip install -e .[tests,dev,embeddings]
+          pip install tantivy
+          pip install mlx
+      - name: Doctest
+        run: pytest --doctest-modules python/lancedb
+  linux:
+    name: "Linux: python-3.${{ matrix.python-minor-version }}"
+    timeout-minutes: 30
+    strategy:
+      matrix:
+        python-minor-version: ["8", "11"]
+    runs-on: "ubuntu-22.04"
+    defaults:
+      run:
+        shell: bash
+        working-directory: python
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+          lfs: true
+      - name: Install protobuf
+        run: |
+          sudo apt update
+          sudo apt install -y protobuf-compiler
+      - name: Set up Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: 3.${{ matrix.python-minor-version }}
+      - uses: Swatinem/rust-cache@v2
+        with:
+          workspaces: python
+      - uses: ./.github/workflows/build_linux_wheel
+      - uses: ./.github/workflows/run_tests
+      # Make sure wheels are not included in the Rust cache
+      - name: Delete wheels
+        run: rm -rf target/wheels
   platform:
-    name: "Platform: ${{ matrix.config.name }}"
+    name: "Mac: ${{ matrix.config.name }}"
     timeout-minutes: 30
     strategy:
       matrix:
         config:
-          - name: x86 Mac
+          - name: x86
             runner: macos-13
-          - name: Arm Mac
+          - name: Arm
             runner: macos-14
-          - name: x86 Windows
+    runs-on: "${{ matrix.config.runner }}"
+    defaults:
+      run:
+        shell: bash
+        working-directory: python
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+          lfs: true
+      - name: Set up Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: "3.11"
+      - uses: Swatinem/rust-cache@v2
+        with:
+          workspaces: python
+      - uses: ./.github/workflows/build_mac_wheel
+      - uses: ./.github/workflows/run_tests
+      # Make sure wheels are not included in the Rust cache
+      - name: Delete wheels
+        run: rm -rf target/wheels
+  windows:
+    name: "Windows: ${{ matrix.config.name }}"
+    timeout-minutes: 30
+    strategy:
+      matrix:
+        config:
+          - name: x86
             runner: windows-latest
     runs-on: "${{ matrix.config.runner }}"
     defaults:
@@ -64,21 +148,22 @@ jobs:
         shell: bash
         working-directory: python
     steps:
-    - uses: actions/checkout@v4
+      - uses: actions/checkout@v4
-      with:
+        with:
-        fetch-depth: 0
+          fetch-depth: 0
-        lfs: true
+          lfs: true
-    - name: Set up Python
+      - name: Set up Python
-      uses: actions/setup-python@v5
+        uses: actions/setup-python@v5
-      with:
+        with:
-        python-version: "3.11"
+          python-version: "3.11"
-    - name: Install lancedb
+      - uses: Swatinem/rust-cache@v2
-      run: |
+        with:
-        pip install -e .[tests]
+          workspaces: python
-        pip install tantivy@git+https://github.com/quickwit-oss/tantivy-py#164adc87e1a033117001cf70e38c82a53014d985
+      - uses: ./.github/workflows/build_windows_wheel
-        pip install pytest pytest-mock
+      - uses: ./.github/workflows/run_tests
-    - name: Run tests
+      # Make sure wheels are not included in the Rust cache
-      run: pytest -m "not slow" -x -v --durations=30 tests
+      - name: Delete wheels
+        run: rm -rf target/wheels
   pydantic1x:
     timeout-minutes: 30
     runs-on: "ubuntu-22.04"
@@ -87,21 +172,22 @@ jobs:
         shell: bash
         working-directory: python
     steps:
-    - uses: actions/checkout@v4
+      - uses: actions/checkout@v4
-      with:
+        with:
-        fetch-depth: 0
+          fetch-depth: 0
-        lfs: true
+          lfs: true
-    - name: Set up Python
+      - name: Install dependencies
-      uses: actions/setup-python@v5
+        run: |
-      with:
+          sudo apt update
-        python-version: 3.9
+          sudo apt install -y protobuf-compiler
-    - name: Install lancedb
+      - name: Set up Python
-      run: |
+        uses: actions/setup-python@v5
-        pip install "pydantic<2"
+        with:
-        pip install -e .[tests]
+          python-version: 3.9
-        pip install tantivy@git+https://github.com/quickwit-oss/tantivy-py#164adc87e1a033117001cf70e38c82a53014d985
+      - name: Install lancedb
-        pip install pytest pytest-mock
+        run: |
-    - name: Run tests
+          pip install "pydantic<2"
-      run: pytest -m "not slow" -x -v --durations=30 tests
+          pip install -e .[tests]
-    - name: doctest
+          pip install tantivy
-      run: pytest --doctest-modules lancedb
+      - name: Run tests
+        run: pytest -m "not slow" -x -v --durations=30 python/tests

.github/workflows/run_tests/action.yml

@@ -0,0 +1,17 @@
+name: run-tests
+
+description: "Install lance wheel and run unit tests"
+inputs:
+  python-minor-version:
+    required: true
+    description: "8 9 10 11 12"
+runs:
+  using: "composite"
+  steps:
+    - name: Install lancedb
+      shell: bash
+      run: |
+        pip3 install $(ls target/wheels/lancedb-*.whl)[tests,dev]
+    - name: pytest
+      shell: bash
+      run: pytest -m "not slow" -x -v --durations=30 python/python/tests

.github/workflows/rust.yml

@@ -119,3 +119,4 @@ jobs:
           $env:VCPKG_ROOT = $env:VCPKG_INSTALLATION_ROOT
           cargo build
           cargo test
+  

.github/workflows/upload_wheel/action.yml

@@ -0,0 +1,29 @@
+name: upload-wheel
+
+description: "Upload wheels to Pypi"
+inputs:
+  os:
+    required: true
+    description: "ubuntu-22.04 or macos-13"
+  repo:
+    required: false
+    description: "pypi or testpypi"
+    default: "pypi"
+  token:
+    required: true
+    description: "release token for the repo"
+
+runs:
+  using: "composite"
+  steps:
+    - name: Install dependencies
+      shell: bash
+      run: |
+        python -m pip install --upgrade pip
+        pip install twine
+    - name: Publish wheel
+      env:
+        TWINE_USERNAME: __token__
+        TWINE_PASSWORD: ${{ inputs.token }}
+      shell: bash
+      run: twine upload --repository ${{ inputs.repo }} target/wheels/lancedb-*.whl

.gitignore

@@ -22,6 +22,11 @@ python/dist
 
 **/.hypothesis
 
+# Compiled Dynamic libraries
+*.so
+*.dylib
+*.dll
+
 ## Javascript
 *.node
 **/node_modules
@@ -34,4 +39,6 @@ dist
 ## Rust
 target
 
+**/sccache.log
+
 Cargo.lock

.pre-commit-config.yaml

@@ -5,17 +5,14 @@ repos:
     -   id: check-yaml
     -   id: end-of-file-fixer
     -   id: trailing-whitespace
--   repo: https://github.com/psf/black
-    rev: 22.12.0
-    hooks:
-    -   id: black
 -   repo: https://github.com/astral-sh/ruff-pre-commit
     # Ruff version.
-    rev: v0.0.277
+    rev: v0.2.2
     hooks:
     - id: ruff
--   repo: https://github.com/pycqa/isort
+- repo: https://github.com/pre-commit/mirrors-prettier
-    rev: 5.12.0
+  rev: v3.1.0
-    hooks:
+  hooks:
-    - id: isort
+    - id: prettier
-      name: isort (python)
+      files: "nodejs/.*"
+      exclude: nodejs/lancedb/native.d.ts|nodejs/dist/.*

Cargo.toml

@@ -1,5 +1,5 @@
 [workspace]
-members = ["rust/ffi/node", "rust/vectordb", "nodejs"]
+members = ["rust/ffi/node", "rust/lancedb", "nodejs", "python"]
 # Python package needs to be built by maturin.
 exclude = ["python"]
 resolver = "2"
@@ -14,10 +14,10 @@ keywords = ["lancedb", "lance", "database", "vector", "search"]
 categories = ["database-implementations"]
 
 [workspace.dependencies]
-lance = { "version" = "=0.9.15", "features" = ["dynamodb"] }
+lance = { "version" = "=0.10.5", "features" = ["dynamodb"] }
-lance-index = { "version" = "=0.9.15" }
+lance-index = { "version" = "=0.10.5" }
-lance-linalg = { "version" = "=0.9.15" }
+lance-linalg = { "version" = "=0.10.5" }
-lance-testing = { "version" = "=0.9.15" }
+lance-testing = { "version" = "=0.10.5" }
 # Note that this one does not include pyarrow
 arrow = { version = "50.0", optional = false }
 arrow-array = "50.0"
@@ -28,13 +28,14 @@ arrow-schema = "50.0"
 arrow-arith = "50.0"
 arrow-cast = "50.0"
 async-trait = "0"
-chrono = "0.4.23"
+chrono = "0.4.35"
 half = { "version" = "=2.3.1", default-features = false, features = [
     "num-traits",
 ] }
 futures = "0"
 log = "0.4"
 object_store = "0.9.0"
+pin-project = "1.0.7"
 snafu = "0.7.4"
 url = "2"
 num-traits = "0.2"

ci/build_linux_artifacts.sh

@@ -13,7 +13,9 @@ docker build \
     .
 popd
 
+# We turn on memory swap to avoid OOM killer
 docker run \
     -v $(pwd):/io -w /io \
+    --memory-swap=-1 \
     lancedb-node-manylinux \
     bash ci/manylinux_node/build.sh $ARCH

dockerfiles/Dockerfile

@@ -0,0 +1,27 @@
+#Simple base dockerfile that supports basic dependencies required to run lance with FTS and Hybrid Search
+#Usage docker build -t lancedb:latest -f Dockerfile .
+FROM python:3.10-slim-buster
+
+# Install Rust
+RUN apt-get update && apt-get install -y curl build-essential && \
+  curl https://sh.rustup.rs -sSf | sh -s -- -y
+
+# Set the environment variable for Rust
+ENV PATH="/root/.cargo/bin:${PATH}"
+
+# Install protobuf compiler
+RUN apt-get install -y protobuf-compiler && \
+  apt-get clean && \
+  rm -rf /var/lib/apt/lists/*
+
+RUN apt-get -y update &&\
+  apt-get -y upgrade && \
+  apt-get -y install git
+
+
+# Verify installations
+RUN python --version && \
+  rustc --version && \
+  protoc --version
+
+RUN pip install tantivy lancedb

docs/mkdocs.yml

@@ -27,7 +27,6 @@ theme:
     - content.tabs.link
     - content.action.edit
     - toc.follow
-    # - toc.integrate
     - navigation.top
     - navigation.tabs
     - navigation.tabs.sticky
@@ -57,6 +56,16 @@ plugins:
           - https://arrow.apache.org/docs/objects.inv
           - https://pandas.pydata.org/docs/objects.inv
 - mkdocs-jupyter
+- ultralytics:
+    verbose: True
+    enabled: True
+    default_image: "assets/lancedb_and_lance.png" # Default image for all pages
+    add_image: True # Automatically add meta image
+    add_keywords: True # Add page keywords in the header tag
+    add_share_buttons: True # Add social share buttons
+    add_authors: False # Display page authors
+    add_desc: False
+    add_dates: False
 
 markdown_extensions:
 - admonition
@@ -90,16 +99,18 @@ nav:
     - Building an ANN index: ann_indexes.md
     - Vector Search: search.md
     - Full-text search: fts.md
-    - Hybrid search: hybrid_search.md
+    - Hybrid search:
+      - Overview: hybrid_search/hybrid_search.md
+      - Comparing Rerankers: hybrid_search/eval.md
+      - Airbnb financial data example: notebooks/hybrid_search.ipynb
     - Filtering: sql.md
     - Versioning & Reproducibility: notebooks/reproducibility.ipynb
     - Configuring Storage: guides/storage.md
   - 🧬 Managing embeddings:
     - Overview: embeddings/index.md
-    - Explicit management: embeddings/embedding_explicit.md
+    - Embedding functions: embeddings/embedding_functions.md
-    - Implicit management: embeddings/embedding_functions.md
+    - Available models: embeddings/default_embedding_functions.md
-    - Available Functions: embeddings/default_embedding_functions.md
+    - User-defined embedding functions: embeddings/custom_embedding_function.md
-    - Custom Embedding Functions: embeddings/api.md
     - "Example: Multi-lingual semantic search": notebooks/multi_lingual_example.ipynb
     - "Example: MultiModal CLIP Embeddings": notebooks/DisappearingEmbeddingFunction.ipynb
   - 🔌 Integrations:
@@ -128,12 +139,14 @@ nav:
       - Serverless Website Chatbot: examples/serverless_website_chatbot.md
       - YouTube Transcript Search: examples/youtube_transcript_bot_with_nodejs.md
       - TransformersJS Embedding Search: examples/transformerjs_embedding_search_nodejs.md
+    - 🦀 Rust:
+      - Overview: examples/examples_rust.md
   - 🔧 CLI & Config: cli_config.md
   - 💭 FAQs: faq.md
   - ⚙️ API reference:
     - 🐍 Python: python/python.md
     - 👾 JavaScript: javascript/modules.md
-    - 🦀 Rust: https://docs.rs/vectordb/latest/vectordb/
+    - 🦀 Rust: https://docs.rs/lancedb/latest/lancedb/
   - ☁️ LanceDB Cloud:
     - Overview: cloud/index.md
     - API reference:
@@ -152,16 +165,18 @@ nav:
     - Building an ANN index: ann_indexes.md
     - Vector Search: search.md
     - Full-text search: fts.md
-    - Hybrid search: hybrid_search.md
+    - Hybrid search:
+      - Overview: hybrid_search/hybrid_search.md
+      - Comparing Rerankers: hybrid_search/eval.md
+      - Airbnb financial data example: notebooks/hybrid_search.ipynb
     - Filtering: sql.md
     - Versioning & Reproducibility: notebooks/reproducibility.ipynb
     - Configuring Storage: guides/storage.md
 - Managing Embeddings:
   - Overview: embeddings/index.md
-  - Explicit management: embeddings/embedding_explicit.md
+  - Embedding functions: embeddings/embedding_functions.md
-  - Implicit management: embeddings/embedding_functions.md
+  - Available models: embeddings/default_embedding_functions.md
-  - Available Functions: embeddings/default_embedding_functions.md
+  - User-defined embedding functions: embeddings/custom_embedding_function.md
-  - Custom Embedding Functions: embeddings/api.md
   - "Example: Multi-lingual semantic search": notebooks/multi_lingual_example.ipynb
   - "Example: MultiModal CLIP Embeddings": notebooks/DisappearingEmbeddingFunction.ipynb
 - Integrations:
@@ -175,21 +190,21 @@ nav:
   - Pydantic: python/pydantic.md
   - Voxel51: integrations/voxel51.md
   - PromptTools: integrations/prompttools.md
-- Python examples:
+- Examples:
   - examples/index.md
   - YouTube Transcript Search: notebooks/youtube_transcript_search.ipynb
   - Documentation QA Bot using LangChain: notebooks/code_qa_bot.ipynb
   - Multimodal search using CLIP: notebooks/multimodal_search.ipynb
   - Serverless QA Bot with S3 and Lambda: examples/serverless_lancedb_with_s3_and_lambda.md
   - Serverless QA Bot with Modal: examples/serverless_qa_bot_with_modal_and_langchain.md
-- Javascript examples:
+  - YouTube Transcript Search (JS): examples/youtube_transcript_bot_with_nodejs.md
-  - Overview: examples/examples_js.md
-  - YouTube Transcript Search: examples/youtube_transcript_bot_with_nodejs.md
   - Serverless Chatbot from any website: examples/serverless_website_chatbot.md
   - TransformersJS Embedding Search: examples/transformerjs_embedding_search_nodejs.md
 - API reference:
+  - Overview: api_reference.md
   - Python: python/python.md
   - Javascript: javascript/modules.md
+  - Rust: https://docs.rs/lancedb/latest/lancedb/index.html
 - LanceDB Cloud:
     - Overview: cloud/index.md
     - API reference:

docs/requirements.txt

@@ -2,4 +2,5 @@ mkdocs==1.5.3
 mkdocs-jupyter==0.24.1
 mkdocs-material==9.5.3
 mkdocstrings[python]==0.20.0
-pydantic
+pydantic
+mkdocs-ultralytics-plugin==0.0.44

docs/src/ann_indexes.md

@@ -7,20 +7,11 @@ for brute-force scanning of the entire vector space.
 A vector index is faster but less accurate than exhaustive search (kNN or flat search).
 LanceDB provides many parameters to fine-tune the index's size, the speed of queries, and the accuracy of results.
 
-Currently, LanceDB does _not_ automatically create the ANN index.
+## Disk-based Index
-LanceDB has optimized code for kNN as well. For many use-cases, datasets under 100K vectors won't require index creation at all.
-If you can live with <100ms latency, skipping index creation is a simpler workflow while guaranteeing 100% recall.
 
-In the future we will look to automatically create and configure the ANN index as data comes in.
+Lance provides an `IVF_PQ` disk-based index. It uses **Inverted File Index (IVF)** to first divide
-
+the dataset into `N` partitions, and then applies **Product Quantization** to compress vectors in each partition.
-## Types of Index
+See the [indexing](concepts/index_ivfpq.md) concepts guide for more information on how this works.
-
-Lance can support multiple index types, the most widely used one is `IVF_PQ`.
-
-- `IVF_PQ`: use **Inverted File Index (IVF)** to first divide the dataset into `N` partitions,
-  and then use **Product Quantization** to compress vectors in each partition.
-- `DiskANN` (**Experimental**): organize the vector as a on-disk graph, where the vertices approximately
-  represent the nearest neighbors of each vector.
 
 ## Creating an IVF_PQ Index
 
@@ -28,39 +19,61 @@ Lance supports `IVF_PQ` index type by default.
 
 === "Python"
 
-     Creating indexes is done via the [create_index](https://lancedb.github.io/lancedb/python/#lancedb.table.LanceTable.create_index) method.
+    Creating indexes is done via the [create_index](https://lancedb.github.io/lancedb/python/#lancedb.table.LanceTable.create_index) method.
 
-     ```python
+    ```python
-     import lancedb
+    import lancedb
-     import numpy as np
+    import numpy as np
-     uri = "data/sample-lancedb"
+    uri = "data/sample-lancedb"
-     db = lancedb.connect(uri)
+    db = lancedb.connect(uri)
 
-     # Create 10,000 sample vectors
+    # Create 10,000 sample vectors
-     data = [{"vector": row, "item": f"item {i}"}
+    data = [{"vector": row, "item": f"item {i}"}
         for i, row in enumerate(np.random.random((10_000, 1536)).astype('float32'))]
 
-     # Add the vectors to a table
+    # Add the vectors to a table
-     tbl = db.create_table("my_vectors", data=data)
+    tbl = db.create_table("my_vectors", data=data)
 
-     # Create and train the index - you need to have enough data in the table for an effective training step
+    # Create and train the index - you need to have enough data in the table for an effective training step
-     tbl.create_index(num_partitions=256, num_sub_vectors=96)
+    tbl.create_index(num_partitions=256, num_sub_vectors=96)
-     ```
+    ```
 
 === "Typescript"
 
-     ```typescript
+    ```typescript
-     --8<--- "docs/src/ann_indexes.ts:import"
+    --8<--- "docs/src/ann_indexes.ts:import"
 
-     --8<-- "docs/src/ann_indexes.ts:ingest"
+    --8<-- "docs/src/ann_indexes.ts:ingest"
-     ```
+    ```
 
-- **metric** (default: "L2"): The distance metric to use. By default it uses euclidean distance "`L2`".
+=== "Rust"
+
+    ```rust
+    --8<-- "rust/lancedb/examples/ivf_pq.rs:create_index"
+    ```
+
+    IVF_PQ index parameters are more fully defined in the [crate docs](https://docs.rs/lancedb/latest/lancedb/index/vector/struct.IvfPqIndexBuilder.html).
+
+The following IVF_PQ paramters can be specified:
+
+- **distance_type**: The distance metric to use. By default it uses euclidean distance "`L2`".
   We also support "cosine" and "dot" distance as well.
-- **num_partitions** (default: 256): The number of partitions of the index.
+- **num_partitions**: The number of partitions in the index. The default is the square root
-- **num_sub_vectors** (default: 96): The number of sub-vectors (M) that will be created during Product Quantization (PQ).
+  of the number of rows.
-  For D dimensional vector, it will be divided into `M` of `D/M` sub-vectors, each of which is presented by
+
-  a single PQ code.
+!!! note
+
+    In the synchronous python SDK and node's `vectordb` the default is 256. This default has
+    changed in the asynchronous python SDK and node's `lancedb`.
+
+- **num_sub_vectors**: The number of sub-vectors (M) that will be created during Product Quantization (PQ).
+  For D dimensional vector, it will be divided into `M` subvectors with dimension `D/M`, each of which is replaced by
+  a single PQ code. The default is the dimension of the vector divided by 16.
+
+!!! note
+
+    In the synchronous python SDK and node's `vectordb` the default is currently 96. This default has
+    changed in the asynchronous python SDK and node's `lancedb`.
 
 <figure markdown>
   ![IVF PQ](./assets/ivf_pq.png)
@@ -88,7 +101,7 @@ You can specify the GPU device to train IVF partitions via
      )
      ```
 
-=== "Macos"
+=== "MacOS"
 
      <!-- skip-test -->
      ```python
@@ -100,7 +113,7 @@ You can specify the GPU device to train IVF partitions via
      )
      ```
 
-Trouble shootings:
+Troubleshooting:
 
 If you see `AssertionError: Torch not compiled with CUDA enabled`, you need to [install
 PyTorch with CUDA support](https://pytorch.org/get-started/locally/).
@@ -123,25 +136,33 @@ There are a couple of parameters that can be used to fine-tune the search:
 
 === "Python"
 
-     ```python
+    ```python
-     tbl.search(np.random.random((1536))) \
+    tbl.search(np.random.random((1536))) \
-         .limit(2) \
+        .limit(2) \
-         .nprobes(20) \
+        .nprobes(20) \
-         .refine_factor(10) \
+        .refine_factor(10) \
-         .to_pandas()
+        .to_pandas()
-     ```
+    ```
 
-     ```text
+    ```text
                                               vector       item       _distance
-     0  [0.44949695, 0.8444449, 0.06281311, 0.23338133...  item 1141  103.575333
+    0  [0.44949695, 0.8444449, 0.06281311, 0.23338133...  item 1141  103.575333
-     1  [0.48587373, 0.269207, 0.15095535, 0.65531915,...  item 3953  108.393867
+    1  [0.48587373, 0.269207, 0.15095535, 0.65531915,...  item 3953  108.393867
-     ```
+    ```
 
 === "Typescript"
 
-     ```typescript
+    ```typescript
-     --8<-- "docs/src/ann_indexes.ts:search1"
+    --8<-- "docs/src/ann_indexes.ts:search1"
-     ```
+    ```
+
+=== "Rust"
+
+    ```rust
+    --8<-- "rust/lancedb/examples/ivf_pq.rs:search1"
+    ```
+
+    Vector search options are more fully defined in the [crate docs](https://docs.rs/lancedb/latest/lancedb/query/struct.Query.html#method.nearest_to).
 
 The search will return the data requested in addition to the distance of each item.
 
@@ -187,13 +208,21 @@ You can select the columns returned by the query using a select clause.
 
 ## FAQ
 
+### Why do I need to manually create an index?
+
+Currently, LanceDB does _not_ automatically create the ANN index.
+LanceDB is well-optimized for kNN (exhaustive search) via a disk-based index. For many use-cases,
+datasets of the order of ~100K vectors don't require index creation. If you can live with up to
+100ms latency, skipping index creation is a simpler workflow while guaranteeing 100% recall.
+
 ### When is it necessary to create an ANN vector index?
 
-`LanceDB` has manually-tuned SIMD code for computing vector distances.
+`LanceDB` comes out-of-the-box with highly optimized SIMD code for computing vector similarity.
-In our benchmarks, computing 100K pairs of 1K dimension vectors takes **less than 20ms**.
+In our benchmarks, computing distances for 100K pairs of 1K dimension vectors takes **less than 20ms**.
-For small datasets (< 100K rows) or applications that can accept 100ms latency, vector indices are usually not necessary.
+We observe that for small datasets (~100K rows) or for applications that can accept 100ms latency,
+vector indices are usually not necessary.
 
-For large-scale or higher dimension vectors, it is beneficial to create vector index.
+For large-scale or higher dimension vectors, it can beneficial to create vector index for performance.
 
 ### How big is my index, and how many memory will it take?
 

docs/src/api_reference.md

@@ -0,0 +1,7 @@
+# API Reference
+
+The API reference for the LanceDB client SDKs are available at the following locations:
+
+- [Python](python/python.md)
+- [JavaScript](javascript/modules.md)
+- [Rust](https://docs.rs/lancedb/latest/lancedb/index.html)

docs/src/basic.md

@@ -3,7 +3,7 @@
 !!! info "LanceDB can be run in a number of ways:"
 
     * Embedded within an existing backend (like your Django, Flask, Node.js or FastAPI application)
-    * Connected to directly from a client application like a Jupyter notebook for analytical workloads
+    * Directly from a client application like a Jupyter notebook for analytical workloads
     * Deployed as a remote serverless database
 
 ![](assets/lancedb_embedded_explanation.png)
@@ -24,13 +24,11 @@
 
 === "Rust"
 
-    !!! warning "Rust SDK is experimental, might introduce breaking changes in the near future"
-
     ```shell
-    cargo add vectordb
+    cargo add lancedb
     ```
 
-    !!! info "To use the vectordb create, you first need to install protobuf."
+    !!! info "To use the lancedb create, you first need to install protobuf."
 
     === "macOS"
 
@@ -44,9 +42,9 @@
         sudo apt install -y protobuf-compiler libssl-dev
         ```
 
-    !!! info "Please also make sure you're using the same version of Arrow as in the [vectordb crate](https://github.com/lancedb/lancedb/blob/main/Cargo.toml)"
+    !!! info "Please also make sure you're using the same version of Arrow as in the [lancedb crate](https://github.com/lancedb/lancedb/blob/main/Cargo.toml)"
 
-## How to connect to a database
+## Connect to a database
 
 === "Python"
 
@@ -69,17 +67,23 @@
     ```rust
     #[tokio::main]
     async fn main() -> Result<()> {
-        --8<-- "rust/vectordb/examples/simple.rs:connect"
+        --8<-- "rust/lancedb/examples/simple.rs:connect"
     }
     ```
 
-    !!! info "See [examples/simple.rs](https://github.com/lancedb/lancedb/tree/main/rust/vectordb/examples/simple.rs) for a full working example."
+    !!! info "See [examples/simple.rs](https://github.com/lancedb/lancedb/tree/main/rust/lancedb/examples/simple.rs) for a full working example."
 
 LanceDB will create the directory if it doesn't exist (including parent directories).
 
 If you need a reminder of the uri, you can call `db.uri()`.
 
-## How to create a table
+## Create a table
+
+### Create a table from initial data
+
+If you have data to insert into the table at creation time, you can simultaneously create a
+table and insert the data into it.  The schema of the data will be used as the schema of the
+table.
 
 === "Python"
 
@@ -115,20 +119,27 @@ If you need a reminder of the uri, you can call `db.uri()`.
 === "Rust"
 
     ```rust
-    use arrow_schema::{DataType, Schema, Field};
+    --8<-- "rust/lancedb/examples/simple.rs:create_table"
-    use arrow_array::{RecordBatch, RecordBatchIterator};
-
-    --8<-- "rust/vectordb/examples/simple.rs:create_table"
     ```
 
-    If the table already exists, LanceDB will raise an error by default.
+    If the table already exists, LanceDB will raise an error by default.  See
+    [the mode option](https://docs.rs/lancedb/latest/lancedb/connection/struct.CreateTableBuilder.html#method.mode)
+    for details on how to overwrite (or open) existing tables instead.
 
-!!! info "Under the hood, LanceDB is converting the input data into an Apache Arrow table and persisting it to disk in [Lance format](https://www.github.com/lancedb/lance)."
+    !!! Providing table records in Rust
 
-### Creating an empty table
+        The Rust SDK currently expects data to be provided as an Arrow
+        [RecordBatchReader](https://docs.rs/arrow-array/latest/arrow_array/trait.RecordBatchReader.html)
+        Support for additional formats (such as serde or polars) is on the roadmap.
+
+!!! info "Under the hood, LanceDB reads in the Apache Arrow data and persists it to disk using the [Lance format](https://www.github.com/lancedb/lance)."
+
+### Create an empty table
 
 Sometimes you may not have the data to insert into the table at creation time.
-In this case, you can create an empty table and specify the schema.
+In this case, you can create an empty table and specify the schema, so that you can add
+data to the table at a later time (as long as it conforms to the schema).  This is
+similar to a `CREATE TABLE` statement in SQL.
 
 === "Python"
 
@@ -147,12 +158,12 @@ In this case, you can create an empty table and specify the schema.
 === "Rust"
 
     ```rust
-    --8<-- "rust/vectordb/examples/simple.rs:create_empty_table"
+    --8<-- "rust/lancedb/examples/simple.rs:create_empty_table"
     ```
 
-## How to open an existing table
+## Open an existing table
 
-Once created, you can open a table using the following code:
+Once created, you can open a table as follows:
 
 === "Python"
 
@@ -169,7 +180,7 @@ Once created, you can open a table using the following code:
 === "Rust"
 
     ```rust
-    --8<-- "rust/vectordb/examples/simple.rs:open_with_existing_file"
+    --8<-- "rust/lancedb/examples/simple.rs:open_existing_tbl"
     ```
 
 If you forget the name of your table, you can always get a listing of all table names:
@@ -189,12 +200,12 @@ If you forget the name of your table, you can always get a listing of all table
 === "Rust"
 
     ```rust
-    --8<-- "rust/vectordb/examples/simple.rs:list_names"
+    --8<-- "rust/lancedb/examples/simple.rs:list_names"
     ```
 
-## How to add data to a table
+## Add data to a table
 
-After a table has been created, you can always add more data to it using
+After a table has been created, you can always add more data to it as follows:
 
 === "Python"
 
@@ -219,12 +230,12 @@ After a table has been created, you can always add more data to it using
 === "Rust"
 
     ```rust
-    --8<-- "rust/vectordb/examples/simple.rs:add"
+    --8<-- "rust/lancedb/examples/simple.rs:add"
     ```
 
-## How to search for (approximate) nearest neighbors
+## Search for nearest neighbors
 
-Once you've embedded the query, you can find its nearest neighbors using the following code:
+Once you've embedded the query, you can find its nearest neighbors as follows:
 
 === "Python"
 
@@ -245,11 +256,20 @@ Once you've embedded the query, you can find its nearest neighbors using the fol
     ```rust
     use futures::TryStreamExt;
 
-    --8<-- "rust/vectordb/examples/simple.rs:search"
+    --8<-- "rust/lancedb/examples/simple.rs:search"
     ```
 
+    !!! Query vectors in Rust
+        Rust does not yet support automatic execution of embedding functions.  You will need to
+        calculate embeddings yourself.  Support for this is on the roadmap and can be tracked at
+        https://github.com/lancedb/lancedb/issues/994
+
+        Query vectors can be provided as Arrow arrays or a Vec/slice of Rust floats.
+        Support for additional formats (e.g. `polars::series::Series`) is on the roadmap.
+
 By default, LanceDB runs a brute-force scan over dataset to find the K nearest neighbours (KNN).
 For tables with more than 50K vectors, creating an ANN index is recommended to speed up search performance.
+LanceDB allows you to create an ANN index on a table as follows:
 
 === "Python"
 
@@ -266,12 +286,17 @@ For tables with more than 50K vectors, creating an ANN index is recommended to s
 === "Rust"
 
     ```rust
-     --8<-- "rust/vectordb/examples/simple.rs:create_index"
+     --8<-- "rust/lancedb/examples/simple.rs:create_index"
     ```
 
-Check [Approximate Nearest Neighbor (ANN) Indexes](/ann_indices.md) section for more details.
+!!! note "Why do I need to create an index manually?"
+    LanceDB does not automatically create the ANN index for two reasons. The first is that it's optimized
+    for really fast retrievals via a disk-based index, and the second is that data and query workloads can
+    be very diverse, so there's no one-size-fits-all index configuration. LanceDB provides many parameters
+    to fine-tune index size, query latency and accuracy. See the section on
+    [ANN indexes](ann_indexes.md) for more details.
 
-## How to delete rows from a table
+## Delete rows from a table
 
 Use the `delete()` method on tables to delete rows from a table. To choose
 which rows to delete, provide a filter that matches on the metadata columns.
@@ -292,12 +317,13 @@ This can delete any number of rows that match the filter.
 === "Rust"
 
     ```rust
-    --8<-- "rust/vectordb/examples/simple.rs:delete"
+    --8<-- "rust/lancedb/examples/simple.rs:delete"
     ```
 
 The deletion predicate is a SQL expression that supports the same expressions
-as the `where()` clause on a search. They can be as simple or complex as needed.
+as the `where()` clause (`only_if()` in Rust) on a search. They can be as
-To see what expressions are supported, see the [SQL filters](sql.md) section.
+simple or complex as needed.  To see what expressions are supported, see the
+[SQL filters](sql.md) section.
 
 === "Python"
 
@@ -307,7 +333,11 @@ To see what expressions are supported, see the [SQL filters](sql.md) section.
 
       Read more: [vectordb.Table.delete](javascript/interfaces/Table.md#delete)
 
-## How to remove a table
+=== "Rust"
+
+      Read more: [lancedb::Table::delete](https://docs.rs/lancedb/latest/lancedb/table/struct.Table.html#method.delete)
+
+## Drop a table
 
 Use the `drop_table()` method on the database to remove a table.
 
@@ -333,7 +363,7 @@ Use the `drop_table()` method on the database to remove a table.
 === "Rust"
 
     ```rust
-    --8<-- "rust/vectordb/examples/simple.rs:drop_table"
+    --8<-- "rust/lancedb/examples/simple.rs:drop_table"
     ```
 
 !!! note "Bundling `vectordb` apps with Webpack"

docs/src/concepts/index_ivfpq.md

@@ -31,7 +31,7 @@ As an example, consider starting with 128-dimensional vector consisting of 32-bi
 
 While PQ helps with reducing the size of the index, IVF primarily addresses search performance. The primary purpose of an inverted file index is to facilitate rapid and effective nearest neighbor search by narrowing down the search space.
 
-In IVF, the PQ vector space is divided into *Voronoi cells*, which are essentially partitions that consist of all the points in the space that are within a threshold distance of the given region's seed point. These seed points are used to create an inverted index that correlates each centroid with a list of vectors in the space, allowing a search to be restricted to just a subset of vectors in the index.
+In IVF, the PQ vector space is divided into *Voronoi cells*, which are essentially partitions that consist of all the points in the space that are within a threshold distance of the given region's seed point. These seed points are initialized by running K-means over the stored vectors. The centroids of K-means turn into the seed points which then each define a region. These regions are then are used to create an inverted index that correlates each centroid with a list of vectors in the space, allowing a search to be restricted to just a subset of vectors in the index.
 
 ![](../assets/ivfpq_ivf_desc.webp)
 
@@ -81,24 +81,4 @@ The above query will perform a search on the table `tbl` using the given query v
 * `to_pandas()`: Convert the results to a pandas DataFrame
 
 And there you have it! You now understand what an IVF-PQ index is, and how to create and query it in LanceDB.
-
+To see how to create an IVF-PQ index in LanceDB, take a look at the [ANN indexes](../ann_indexes.md) section.
-
-## FAQ
-
-### When is it necessary to create a vector index?
-
-LanceDB has manually-tuned SIMD code for computing vector distances. In our benchmarks, computing 100K pairs of 1K dimension vectors takes **<20ms**. For small datasets (<100K rows) or applications that can accept up to 100ms latency, vector indices are usually not necessary.
-
-For large-scale or higher dimension vectors, it is beneficial to create vector index.
-
-### How big is my index, and how much memory will it take?
-
-In LanceDB, all vector indices are disk-based, meaning that when responding to a vector query, only the relevant pages from the index file are loaded from disk and cached in memory. Additionally, each sub-vector is usually encoded into 1 byte PQ code.
-
-For example, with 1024-dimension vectors, if we choose `num_sub_vectors = 64`, each sub-vector has `1024 / 64 = 16` float32 numbers. Product quantization can lead to approximately `16 * sizeof(float32) / 1 = 64` times of space reduction.
-
-### How to choose `num_partitions` and `num_sub_vectors` for IVF_PQ index?
-
-`num_partitions` is used to decide how many partitions the first level IVF index uses. Higher number of partitions could lead to more efficient I/O during queries and better accuracy, but it takes much more time to train. On SIFT-1M dataset, our benchmark shows that keeping each partition 1K-4K rows lead to a good latency/recall.
-
-`num_sub_vectors` specifies how many PQ short codes to generate on each vector. Because PQ is a lossy compression of the original vector, a higher `num_sub_vectors` usually results in less space distortion, and thus yields better accuracy. However, a higher `num_sub_vectors` also causes heavier I/O and more PQ computation, and thus, higher latency. `dimension / num_sub_vectors` should be a multiple of 8 for optimum SIMD efficiency.

docs/src/embeddings/custom_embedding_function.md

@@ -17,6 +17,7 @@ Let's implement `SentenceTransformerEmbeddings` class. All you need to do is imp
 
 ```python
 from lancedb.embeddings import register
+from lancedb.util import attempt_import_or_raise
 
 @register("sentence-transformers")
 class SentenceTransformerEmbeddings(TextEmbeddingFunction):
@@ -81,7 +82,7 @@ class OpenClipEmbeddings(EmbeddingFunction):
 
     def __init__(self, *args, **kwargs):
         super().__init__(*args, **kwargs)
-        open_clip = self.safe_import("open_clip", "open-clip") # EmbeddingFunction util to import external libs and raise if not found
+        open_clip = attempt_import_or_raise("open_clip", "open-clip") # EmbeddingFunction util to import external libs and raise if not found
         model, _, preprocess = open_clip.create_model_and_transforms(
             self.name, pretrained=self.pretrained
         )
@@ -109,14 +110,14 @@ class OpenClipEmbeddings(EmbeddingFunction):
         if isinstance(query, str):
             return [self.generate_text_embeddings(query)]
         else:
-            PIL = self.safe_import("PIL", "pillow")
+            PIL = attempt_import_or_raise("PIL", "pillow")
             if isinstance(query, PIL.Image.Image):
                 return [self.generate_image_embedding(query)]
             else:
                 raise TypeError("OpenClip supports str or PIL Image as query")
 
     def generate_text_embeddings(self, text: str) -> np.ndarray:
-        torch = self.safe_import("torch")
+        torch = attempt_import_or_raise("torch")
         text = self.sanitize_input(text)
         text = self._tokenizer(text)
         text.to(self.device)
@@ -175,7 +176,7 @@ class OpenClipEmbeddings(EmbeddingFunction):
             The image to embed. If the image is a str, it is treated as a uri.
             If the image is bytes, it is treated as the raw image bytes.
         """
-        torch = self.safe_import("torch")
+        torch = attempt_import_or_raise("torch")
         # TODO handle retry and errors for https
         image = self._to_pil(image)
         image = self._preprocess(image).unsqueeze(0)
@@ -183,7 +184,7 @@ class OpenClipEmbeddings(EmbeddingFunction):
             return self._encode_and_normalize_image(image)
 
     def _to_pil(self, image: Union[str, bytes]):
-        PIL = self.safe_import("PIL", "pillow")
+        PIL = attempt_import_or_raise("PIL", "pillow")
         if isinstance(image, bytes):
             return PIL.Image.open(io.BytesIO(image))
         if isinstance(image, PIL.Image.Image):

docs/src/embeddings/default_embedding_functions.md

@@ -9,6 +9,9 @@ Contains the text embedding functions registered by default.
 ### Sentence transformers
 Allows you to set parameters when registering a `sentence-transformers` object.
 
+!!! info
+    Sentence transformer embeddings are normalized by default. It is recommended to use normalized embeddings for similarity search.
+
 | Parameter | Type | Default Value | Description |
 |---|---|---|---|
 | `name` | `str` | `all-MiniLM-L6-v2` | The name of the model |
@@ -44,6 +47,7 @@ LanceDB registers the OpenAI embeddings function in the registry by default, as
 | Parameter | Type | Default Value | Description |
 |---|---|---|---|
 | `name` | `str` | `"text-embedding-ada-002"` | The name of the model. |
+| `dim` | `int` |  Model default   | For OpenAI's newer text-embedding-3 model, we can specify a dimensionality that is smaller than the 1536 size. This feature supports it |
 
 
 ```python
@@ -172,7 +176,8 @@ Supported Embedding modelIDs are:
 * `cohere.embed-english-v3`
 * `cohere.embed-multilingual-v3`
 
-Supported paramters (to be passed in `create` method) are:
+Supported parameters (to be passed in `create` method) are:
+
 | Parameter | Type | Default Value | Description |
 |---|---|---|---|
 | **name** | str | "amazon.titan-embed-text-v1" | The model ID of the bedrock model to use. Supported base models for Text Embeddings: amazon.titan-embed-text-v1, cohere.embed-english-v3, cohere.embed-multilingual-v3 |
@@ -219,7 +224,6 @@ This embedding function supports ingesting images as both bytes and urls. You ca
 !!! info
     LanceDB supports ingesting images directly from accessible links.
 
-
 ```python
 
 db = lancedb.connect(tmp_path)
@@ -285,4 +289,67 @@ print(actual.label)
 
 ```
 
+### Imagebind embeddings
+We have support for [imagebind](https://github.com/facebookresearch/ImageBind) model embeddings. You can download our version of the packaged model via - `pip install imagebind-packaged==0.1.2`.
+
+This function is registered as `imagebind` and supports Audio, Video and Text modalities(extending to Thermal,Depth,IMU data):
+
+| Parameter | Type | Default Value | Description |
+|---|---|---|---|
+| `name` | `str` | `"imagebind_huge"` | Name of the model. |
+| `device` | `str` | `"cpu"` | The device to run the model on. Can be `"cpu"` or `"gpu"`. |
+| `normalize` | `bool` | `False` | set to `True` to normalize your inputs before model ingestion. |
+
+Below is an example demonstrating how the API works:
+
+```python
+db = lancedb.connect(tmp_path)
+registry = EmbeddingFunctionRegistry.get_instance()
+func = registry.get("imagebind").create()
+
+class ImageBindModel(LanceModel):
+    text: str
+    image_uri: str = func.SourceField()
+    audio_path: str
+    vector: Vector(func.ndims()) = func.VectorField()
+
+# add locally accessible image paths
+text_list=["A dog.", "A car", "A bird"]
+image_paths=[".assets/dog_image.jpg", ".assets/car_image.jpg", ".assets/bird_image.jpg"]
+audio_paths=[".assets/dog_audio.wav", ".assets/car_audio.wav", ".assets/bird_audio.wav"]
+
+# Load data
+inputs = [
+    {"text": a, "audio_path": b, "image_uri": c}
+    for a, b, c in zip(text_list, audio_paths, image_paths)
+]
+
+#create table and add data
+table = db.create_table("img_bind", schema=ImageBindModel)
+table.add(inputs)
+```
+
+Now, we can search using any modality:
+
+#### image search
+```python
+query_image = "./assets/dog_image2.jpg" #download an image and enter that path here
+actual = table.search(query_image).limit(1).to_pydantic(ImageBindModel)[0]
+print(actual.text == "dog")
+```
+#### audio search
+
+```python
+query_audio = "./assets/car_audio2.wav" #download an audio clip and enter path here
+actual = table.search(query_audio).limit(1).to_pydantic(ImageBindModel)[0]
+print(actual.text == "car")
+```
+#### Text search
+You can add any input query and fetch the result as follows:
+```python
+query = "an animal which flies and tweets" 
+actual = table.search(query).limit(1).to_pydantic(ImageBindModel)[0]
+print(actual.text == "bird")
+```
+
 If you have any questions about the embeddings API, supported models, or see a relevant model missing, please raise an issue [on GitHub](https://github.com/lancedb/lancedb/issues).

docs/src/embeddings/embedding_explicit.md

@@ -1,141 +0,0 @@
-In this workflow, you define your own embedding function and pass it as a callable to LanceDB, invoking it in your code to generate the embeddings. Let's look at some examples.
-
-### Hugging Face
-
-!!! note
-    Currently, the Hugging Face method is only supported in the Python SDK.
-
-=== "Python"
-    The most popular open source option is to use the [sentence-transformers](https://www.sbert.net/) 
-    library, which can be installed via pip.
-
-    ```bash
-    pip install sentence-transformers
-    ```
-
-    The example below shows how to use the `paraphrase-albert-small-v2` model to generate embeddings 
-    for a given document.
-
-    ```python
-    from sentence_transformers import SentenceTransformer
-
-    name="paraphrase-albert-small-v2"
-    model = SentenceTransformer(name)
-
-    # used for both training and querying
-    def embed_func(batch):
-        return [model.encode(sentence) for sentence in batch]
-    ```
-
-### OpenAI
-
-Another popular alternative is to use an external API like OpenAI's [embeddings API](https://platform.openai.com/docs/guides/embeddings/what-are-embeddings).
-
-=== "Python"
-      ```python
-        import openai
-        import os
-
-        # Configuring the environment variable OPENAI_API_KEY
-        if "OPENAI_API_KEY" not in os.environ:
-        # OR set the key here as a variable
-        openai.api_key = "sk-..."
-
-        # verify that the API key is working
-        assert len(openai.Model.list()["data"]) > 0
-
-        def embed_func(c):
-            rs = openai.Embedding.create(input=c, engine="text-embedding-ada-002")
-            return [record["embedding"] for record in rs["data"]]
-      ```
-
-=== "JavaScript"
-      ```javascript
-        const lancedb = require("vectordb");
-
-        // You need to provide an OpenAI API key
-        const apiKey = "sk-..."
-        // The embedding function will create embeddings for the 'text' column
-        const embedding = new lancedb.OpenAIEmbeddingFunction('text', apiKey)
-      ```
-
-## Applying an embedding function to data
-
-=== "Python"
-    Using an embedding function, you can apply it to raw data
-    to generate embeddings for each record.
-
-    Say you have a pandas DataFrame with a `text` column that you want embedded,
-    you can use the `with_embeddings` function to generate embeddings and add them to 
-    an existing table.
-
-    ```python
-     import pandas as pd
-     from lancedb.embeddings import with_embeddings
-
-     df = pd.DataFrame(
-        [
-            {"text": "pepperoni"},
-            {"text": "pineapple"}
-        ]
-    )
-     data = with_embeddings(embed_func, df)
-
-     # The output is used to create / append to a table
-     # db.create_table("my_table", data=data)
-    ```
-
-    If your data is in a different column, you can specify the `column` kwarg to `with_embeddings`.
-
-    By default, LanceDB calls the function with batches of 1000 rows. This can be configured
-    using the `batch_size` parameter to `with_embeddings`.
-
-    LanceDB automatically wraps the function with retry and rate-limit logic to ensure the OpenAI
-    API call is reliable.
-
-=== "JavaScript"
-    Using an embedding function, you can apply it to raw data
-    to generate embeddings for each record.
-
-    Simply pass the embedding function created above and LanceDB will use it to generate
-    embeddings for your data.
-
-    ```javascript
-    const db = await lancedb.connect("data/sample-lancedb");
-    const data = [
-    { text: "pepperoni"},
-    { text: "pineapple"}
-    ]
-
-    const table = await db.createTable("vectors", data, embedding)
-    ```
-
-## Querying using an embedding function
-
-!!! warning
-    At query time, you **must** use the same embedding function you used to vectorize your data.
-    If you use a different embedding function, the embeddings will not reside in the same vector
-    space and the results will be nonsensical.
-
-=== "Python"
-     ```python
-     query = "What's the best pizza topping?"
-     query_vector = embed_func([query])[0]
-     results = (
-        tbl.search(query_vector)
-        .limit(10)
-        .to_pandas()
-     )
-     ```
-
-     The above snippet returns a pandas DataFrame with the 10 closest vectors to the query.
-
-=== "JavaScript"
-     ```javascript
-      const results = await table
-        .search("What's the best pizza topping?")
-        .limit(10)
-        .execute()
-     ```
-
-     The above snippet returns an array of records with the top 10 nearest neighbors to the query.

docs/src/embeddings/embedding_functions.md

@@ -3,61 +3,126 @@ Representing multi-modal data as vector embeddings is becoming a standard practi
 For this purpose, LanceDB introduces an **embedding functions API**, that allow you simply set up once, during the configuration stage of your project. After this, the table remembers it, effectively making the embedding functions *disappear in the background* so you don't have to worry about manually passing callables, and instead, simply focus on the rest of your data engineering pipeline.
 
 !!! warning
-    Using the implicit embeddings management approach means that you can forget about the manually passing around embedding
+    Using the embedding function registry means that you don't have to explicitly generate the embeddings yourself. 
-    functions in your code, as long as you don't intend to change it at a later time. If your embedding function changes,
+    However, if your embedding function changes, you'll have to re-configure your table with the new embedding function 
-    you'll have to re-configure your table with the new embedding function and regenerate the embeddings.
+    and regenerate the embeddings. In the future, we plan to support the ability to change the embedding function via
+    table metadata and have LanceDB automatically take care of regenerating the embeddings.
+
 
 ## 1. Define the embedding function
-We have some pre-defined embedding functions in the global registry, with more coming soon. Here's let's an implementation of CLIP as example.
-```
-registry = EmbeddingFunctionRegistry.get_instance()
-clip = registry.get("open-clip").create()
 
-```
+=== "Python"
-You can also define your own embedding function by implementing the `EmbeddingFunction` abstract base interface. It subclasses Pydantic Model which can be utilized to write complex schemas simply as we'll see next!
+    In the LanceDB python SDK, we define a global embedding function registry with
+    many different embedding models and even more coming soon. 
+    Here's let's an implementation of CLIP as example.
+
+    ```python
+    from lancedb.embeddings import get_registry
+
+    registry = get_registry()
+    clip = registry.get("open-clip").create()
+    ```
+
+    You can also define your own embedding function by implementing the `EmbeddingFunction` 
+    abstract base interface. It subclasses Pydantic Model which can be utilized to write complex schemas simply as we'll see next!
+
+=== "JavaScript""
+    In the TypeScript SDK, the choices are more limited. For now, only the OpenAI
+    embedding function is available.
+
+    ```javascript
+    const lancedb = require("vectordb");
+
+    // You need to provide an OpenAI API key
+    const apiKey = "sk-..."
+    // The embedding function will create embeddings for the 'text' column
+    const embedding = new lancedb.OpenAIEmbeddingFunction('text', apiKey)
+    ```
 
 ## 2. Define the data model or schema
-The embedding function defined above abstracts away all the details about the models and dimensions required to define the schema. You can simply set a field as **source** or **vector** column. Here's how:
 
-```python
+=== "Python"
-class Pets(LanceModel):
+    The embedding function defined above abstracts away all the details about the models and dimensions required to define the schema. You can simply set a field as **source** or **vector** column. Here's how:
-    vector: Vector(clip.ndims) = clip.VectorField()
-    image_uri: str = clip.SourceField()
-```
 
-`VectorField` tells LanceDB to use the clip embedding function to generate query embeddings for the `vector` column and `SourceField` ensures that when adding data, we automatically use the specified embedding function to encode `image_uri`.
+    ```python
+    class Pets(LanceModel):
+        vector: Vector(clip.ndims) = clip.VectorField()
+        image_uri: str = clip.SourceField()
+    ```
 
-## 3. Create LanceDB table
+    `VectorField` tells LanceDB to use the clip embedding function to generate query embeddings for the `vector` column and `SourceField` ensures that when adding data, we automatically use the specified embedding function to encode `image_uri`.
-Now that we have chosen/defined our embedding function and the schema, we can create the table:
 
-```python
+=== "JavaScript"
-db = lancedb.connect("~/lancedb")
-table = db.create_table("pets", schema=Pets)
 
-```
+    For the TypeScript SDK, a schema can be inferred from input data, or an explicit
+    Arrow schema can be provided.
 
-That's it! We've provided all the information needed to embed the source and query inputs. We can now forget about the model and dimension details and start to build our VectorDB pipeline.
+## 3. Create table and add data
 
-## 4. Ingest lots of data and query your table
+Now that we have chosen/defined our embedding function and the schema, 
-Any new or incoming data can just be added and it'll be vectorized automatically.
+we can create the table and ingest data without needing to explicitly generate
+the embeddings at all:
 
-```python
+=== "Python"
-table.add([{"image_uri": u} for u in uris])
+    ```python
-```
+    db = lancedb.connect("~/lancedb")
+    table = db.create_table("pets", schema=Pets)
 
-Our OpenCLIP query embedding function supports querying via both text and images:
+    table.add([{"image_uri": u} for u in uris])
+    ```
 
-```python
+=== "JavaScript"
-result = table.search("dog")
-```
 
-Let's query an image:
+    ```javascript
+    const db = await lancedb.connect("data/sample-lancedb");
+    const data = [
+    { text: "pepperoni"},
+    { text: "pineapple"}
+    ]
 
-```python
+    const table = await db.createTable("vectors", data, embedding)
-p = Path("path/to/images/samoyed_100.jpg")
+    ```
-query_image = Image.open(p)
+
-table.search(query_image)
+## 4. Querying your table
-```
+Not only can you forget about the embeddings during ingestion, you also don't
+need to worry about it when you query the table:
+
+=== "Python"
+
+    Our OpenCLIP query embedding function supports querying via both text and images:
+
+    ```python
+    results = (
+        table.search("dog")
+        .limit(10)
+        .to_pandas()
+    )
+    ```
+
+    Or we can search using an image:
+
+    ```python
+    p = Path("path/to/images/samoyed_100.jpg")
+    query_image = Image.open(p)
+    results = (
+        table.search(query_image)
+        .limit(10)
+        .to_pandas()
+    )
+    ```
+
+    Both of the above snippet returns a pandas DataFrame with the 10 closest vectors to the query.
+
+=== "JavaScript"
+
+    ```javascript
+    const results = await table
+    .search("What's the best pizza topping?")
+    .limit(10)
+    .execute()
+    ```    
+    
+    The above snippet returns an array of records with the top 10 nearest neighbors to the query.
 
 ---
 
@@ -100,4 +165,5 @@ rs[2].image
 
 ![](../assets/dog_clip_output.png)
 
-Now that you have the basic idea about implicit management via embedding functions, let's dive deeper into a [custom API](./api.md) that you can use to implement your own embedding functions.
+Now that you have the basic idea about LanceDB embedding functions and the embedding function registry,
+let's dive deeper into defining your own [custom functions](./custom_embedding_function.md).

docs/src/embeddings/index.md

@@ -1,8 +1,14 @@
-Due to the nature of vector embeddings, they can be used to represent any kind of data, from text to images to audio. This makes them a very powerful tool for machine learning practitioners. However, there's no one-size-fits-all solution for generating embeddings - there are many different libraries and APIs (both commercial and open source) that can be used to generate embeddings from structured/unstructured data.
+Due to the nature of vector embeddings, they can be used to represent any kind of data, from text to images to audio. 
+This makes them a very powerful tool for machine learning practitioners. 
+However, there's no one-size-fits-all solution for generating embeddings - there are many different libraries and APIs 
+(both commercial and open source) that can be used to generate embeddings from structured/unstructured data.
 
-LanceDB supports 2 methods of vectorizing your raw data into embeddings.
+LanceDB supports 3 methods of working with embeddings.
 
-1. **Explicit**: By manually calling LanceDB's `with_embedding` function to vectorize your data via an `embed_func` of your choice
+1. You can manually generate embeddings for the data and queries. This is done outside of LanceDB.
-2. **Implicit**: Allow LanceDB to embed the data and queries in the background as they come in, by using the table's `EmbeddingRegistry` information
+2. You can use the built-in [embedding functions](./embedding_functions.md) to embed the data and queries in the background.
+3. For python users, you can define your own [custom embedding function](./custom_embedding_function.md)
+   that extends the default embedding functions.
 
-See the [explicit](embedding_explicit.md) and [implicit](embedding_functions.md) embedding sections for more details.
+For python users, there is also a legacy [with_embeddings API](./legacy.md).
+It is retained for compatibility and will be removed in a future version.

docs/src/embeddings/legacy.md

@@ -0,0 +1,99 @@
+The legacy `with_embeddings` API is for Python only and is deprecated.
+
+### Hugging Face
+
+The most popular open source option is to use the [sentence-transformers](https://www.sbert.net/) 
+library, which can be installed via pip.
+
+```bash
+pip install sentence-transformers
+```
+
+The example below shows how to use the `paraphrase-albert-small-v2` model to generate embeddings 
+for a given document.
+
+```python
+from sentence_transformers import SentenceTransformer
+
+name="paraphrase-albert-small-v2"
+model = SentenceTransformer(name)
+
+# used for both training and querying
+def embed_func(batch):
+    return [model.encode(sentence) for sentence in batch]
+```
+
+
+### OpenAI
+
+Another popular alternative is to use an external API like OpenAI's [embeddings API](https://platform.openai.com/docs/guides/embeddings/what-are-embeddings).
+
+```python
+import openai
+import os
+
+# Configuring the environment variable OPENAI_API_KEY
+if "OPENAI_API_KEY" not in os.environ:
+# OR set the key here as a variable
+openai.api_key = "sk-..."
+
+client = openai.OpenAI()
+
+def embed_func(c):    
+    rs = client.embeddings.create(input=c, model="text-embedding-ada-002")
+    return [record.embedding for record in rs["data"]]
+```
+
+
+## Applying an embedding function to data
+
+Using an embedding function, you can apply it to raw data
+to generate embeddings for each record.
+
+Say you have a pandas DataFrame with a `text` column that you want embedded,
+you can use the `with_embeddings` function to generate embeddings and add them to 
+an existing table.
+
+```python
+    import pandas as pd
+    from lancedb.embeddings import with_embeddings
+
+    df = pd.DataFrame(
+        [
+            {"text": "pepperoni"},
+            {"text": "pineapple"}
+        ]
+    )
+    data = with_embeddings(embed_func, df)
+
+    # The output is used to create / append to a table
+    tbl = db.create_table("my_table", data=data)
+```
+
+If your data is in a different column, you can specify the `column` kwarg to `with_embeddings`.
+
+By default, LanceDB calls the function with batches of 1000 rows. This can be configured
+using the `batch_size` parameter to `with_embeddings`.
+
+LanceDB automatically wraps the function with retry and rate-limit logic to ensure the OpenAI
+API call is reliable.
+
+## Querying using an embedding function
+
+!!! warning
+    At query time, you **must** use the same embedding function you used to vectorize your data.
+    If you use a different embedding function, the embeddings will not reside in the same vector
+    space and the results will be nonsensical.
+
+=== "Python"
+     ```python
+     query = "What's the best pizza topping?"
+     query_vector = embed_func([query])[0]
+     results = (
+        tbl.search(query_vector)
+        .limit(10)
+        .to_pandas()
+     )
+     ```
+
+     The above snippet returns a pandas DataFrame with the 10 closest vectors to the query.

docs/src/examples/examples_rust.md

@@ -0,0 +1,3 @@
+# Examples: Rust
+
+Our Rust SDK is now stable. Examples are coming soon.

docs/src/examples/image_embeddings_roboflow.md

@@ -43,7 +43,7 @@ pip install lancedb
 We also need to install a specific commit of `tantivy`, a dependency of the LanceDB full text search engine we will use later in this guide:
 
 ```
-pip install tantivy@git+https://github.com/quickwit-oss/tantivy-py#164adc87e1a033117001cf70e38c82a53014d985
+pip install tantivy
 ```
 
 Create a new Python file and add the following code:

docs/src/examples/index.md

@@ -2,10 +2,11 @@
 
 ## Recipes and example code
 
-LanceDB provides language APIs, allowing you to embed a database in your language of choice. We currently provide Python and Javascript APIs, with the Rust API and examples actively being worked on and will be available soon.
+LanceDB provides language APIs, allowing you to embed a database in your language of choice.
 
 * 🐍 [Python](examples_python.md) examples
-* 👾 [JavaScript](exampled_js.md) examples
+* 👾 [JavaScript](examples_js.md) examples
+* 🦀 Rust examples (coming soon)
 
 ## Applications powered by LanceDB
 

docs/src/examples/modal_langchain.py

@@ -1,6 +1,5 @@
 import pickle
 import re
-import sys
 import zipfile
 from pathlib import Path
 

docs/src/faq.md

@@ -16,7 +16,7 @@ As we mention in our talk titled “[Lance, a modern columnar data format](https
 
 ### Why build in Rust? 🦀
 
-We believe that the Rust ecosystem has attained mainstream maturity and that Rust will form the underpinnings of large parts of the data and ML landscape in a few years. Performance, latency and reliability are paramount to a vector DB, and building in Rust allows us to iterate and release updates more rapidly due to Rust’s safety guarantees. Both Lance (the data format) and LanceDB (the database) are written entirely in Rust. We also provide Python and JavaScript client libraries to interact with the database. Our Rust API is a little rough around the edges right now, but is fast becoming on par with the Python and JS APIs.
+We believe that the Rust ecosystem has attained mainstream maturity and that Rust will form the underpinnings of large parts of the data and ML landscape in a few years. Performance, latency and reliability are paramount to a vector DB, and building in Rust allows us to iterate and release updates more rapidly due to Rust’s safety guarantees. Both Lance (the data format) and LanceDB (the database) are written entirely in Rust. We also provide Python, JavaScript, and Rust client libraries to interact with the database.
 
 ### What is the difference between LanceDB OSS and LanceDB Cloud?
 
@@ -40,11 +40,11 @@ LanceDB and its underlying data format, Lance, are built to scale to really larg
 
 No. LanceDB is blazing fast (due to its disk-based index) for even brute force kNN search, within reason. In our benchmarks, computing 100K pairs of 1000-dimension vectors takes less than 20ms. For small datasets of ~100K records or applications that can accept ~100ms latency, an ANN index is usually not necessary.
 
-For large-scale (>1M) or higher dimension vectors, it is beneficial to create an ANN index.
+For large-scale (>1M) or higher dimension vectors, it is beneficial to create an ANN index. See the [ANN indexes](ann_indexes.md) section for more details.
 
 ### Does LanceDB support full-text search?
 
-Yes, LanceDB supports full-text search (FTS) via [Tantivy](https://github.com/quickwit-oss/tantivy). Our current FTS integration is Python-only, and our goal is to push it down to the Rust level in future versions to enable much more powerful search capabilities available to our Python, JavaScript and Rust clients.
+Yes, LanceDB supports full-text search (FTS) via [Tantivy](https://github.com/quickwit-oss/tantivy). Our current FTS integration is Python-only, and our goal is to push it down to the Rust level in future versions to enable much more powerful search capabilities available to our Python, JavaScript and Rust clients.  Follow along in the [Github issue](https://github.com/lancedb/lance/issues/1195)
 
 ### How can I speed up data inserts?
 

docs/src/fts.md

@@ -1,6 +1,6 @@
 # Full-text search
 
-LanceDB provides support for full-text search via [Tantivy](https://github.com/quickwit-oss/tantivy) (currently Python only), allowing you to incorporate keyword-based search (based on BM25) in your retrieval solutions. Our goal is to push the FTS integration down to the Rust level in the future, so that it's available for JavaScript users as well.
+LanceDB provides support for full-text search via [Tantivy](https://github.com/quickwit-oss/tantivy) (currently Python only), allowing you to incorporate keyword-based search (based on BM25) in your retrieval solutions. Our goal is to push the FTS integration down to the Rust level in the future, so that it's available for Rust and JavaScript users as well.  Follow along at [this Github issue](https://github.com/lancedb/lance/issues/1195)
 
 A hybrid search solution combining vector and full-text search is also on the way.
 
@@ -75,25 +75,74 @@ applied on top of the full text search results. This can be invoked via the fami
 table.search("puppy").limit(10).where("meta='foo'").to_list()
 ```
 
-## Syntax
+## Sorting
 
-For full-text search you can perform either a phrase query like "the old man and the sea", 
+You can pre-sort the documents by specifying `ordering_field_names` when
-or a structured search query like "(Old AND Man) AND Sea".
+creating the full-text search index. Once pre-sorted, you can then specify
-Double quotes are used to disambiguate.
+`ordering_field_name` while searching to return results sorted by the given
+field. For example, 
 
-For example: 
+```
+table.create_fts_index(["text_field"], ordering_field_names=["sort_by_field"])
 
-If you intended "they could have been dogs OR cats" as a phrase query, this actually
+(table.search("terms", ordering_field_name="sort_by_field")
-raises a syntax error since `OR` is a recognized operator. If you make `or` lower case,
+ .limit(20)
-this avoids the syntax error. However, it is cumbersome to have to remember what will
+ .to_list())
-conflict with the query syntax. Instead, if you search using 
+```
-`table.search('"they could have been dogs OR cats"')`, then the syntax checker avoids
-checking inside the quotes.
 
+!!! note
+    If you wish to specify an ordering field at query time, you must also
+    have specified it during indexing time. Otherwise at query time, an
+    error will be raised that looks like `ValueError: The field does not exist: xxx`
+
+!!! note
+    The fields to sort on must be of typed unsigned integer, or else you will see 
+    an error during indexing that looks like 
+    `TypeError: argument 'value': 'float' object cannot be interpreted as an integer`.
+
+!!! note
+    You can specify multiple fields for ordering at indexing time.
+    But at query time only one ordering field is supported.
+
+
+## Phrase queries vs. terms queries
+
+For full-text search you can specify either a **phrase** query like `"the old man and the sea"`,
+or a **terms** search query like `"(Old AND Man) AND Sea"`. For more details on the terms
+query syntax, see Tantivy's [query parser rules](https://docs.rs/tantivy/latest/tantivy/query/struct.QueryParser.html).
+
+!!! tip "Note"
+    The query parser will raise an exception on queries that are ambiguous. For example, in the query `they could have been dogs OR cats`, `OR` is capitalized so it's considered a keyword query operator. But it's ambiguous how the left part should be treated. So if you submit this search query as is, you'll get `Syntax Error: they could have been dogs OR cats`.
+
+    ```py
+    # This raises a syntax error
+    table.search("they could have been dogs OR cats")
+    ```
+
+    On the other hand, lowercasing `OR` to `or` will work, because there are no capitalized logical operators and
+    the query is treated as a phrase query.
+
+    ```py
+    # This works!
+    table.search("they could have been dogs or cats")
+    ```
+
+It can be cumbersome to have to remember what will cause a syntax error depending on the type of
+query you want to perform. To make this simpler, when you want to perform a phrase query, you can
+enforce it in one of two ways:
+
+1. Place the double-quoted query inside single quotes. For example, `table.search('"they could have been dogs OR cats"')` is treated as
+a phrase query.
+2. Explicitly declare the `phrase_query()` method. This is useful when you have a phrase query that
+itself contains double quotes. For example, `table.search('the cats OR dogs were not really "pets" at all').phrase_query()`
+is treated as a phrase query.
+
+In general, a query that's declared as a phrase query will be wrapped in double quotes during parsing, with nested
+double quotes replaced by single quotes.
 
 ## Configurations
 
-By default, LanceDB configures a 1GB heap size limit for creating the index. You can 
+By default, LanceDB configures a 1GB heap size limit for creating the index. You can
 reduce this if running on a smaller node, or increase this for faster performance while
 indexing a larger corpus.
 
@@ -109,7 +158,6 @@ table.create_fts_index(["text1", "text2"], writer_heap_size=heap, replace=True)
    If you add data after FTS index creation, it won't be reflected
    in search results until you do a full reindex.
 
-2. We currently only support local filesystem paths for the FTS index. 
+2. We currently only support local filesystem paths for the FTS index.
    This is a tantivy limitation. We've implemented an object store plugin
    but there's no way in tantivy-py to specify to use it.
-

docs/src/guides/tables.md

@@ -168,151 +168,151 @@ This guide will show how to create tables, insert data into them, and update the
     --8<-- "docs/src/basic_legacy.ts:create_f16_table"
     ```
 
-    ### From Pydantic Models
+### From Pydantic Models
 
-    When you create an empty table without data, you must specify the table schema.
+When you create an empty table without data, you must specify the table schema.
-    LanceDB supports creating tables by specifying a PyArrow schema or a specialized
+LanceDB supports creating tables by specifying a PyArrow schema or a specialized
-    Pydantic model called `LanceModel`.
+Pydantic model called `LanceModel`.
 
-    For example, the following Content model specifies a table with 5 columns:
+For example, the following Content model specifies a table with 5 columns:
-    `movie_id`, `vector`, `genres`, `title`, and `imdb_id`. When you create a table, you can
+`movie_id`, `vector`, `genres`, `title`, and `imdb_id`. When you create a table, you can
-    pass the class as the value of the `schema` parameter to `create_table`.
+pass the class as the value of the `schema` parameter to `create_table`.
-    The `vector` column is a `Vector` type, which is a specialized Pydantic type that
+The `vector` column is a `Vector` type, which is a specialized Pydantic type that
-    can be configured with the vector dimensions. It is also important to note that
+can be configured with the vector dimensions. It is also important to note that
-    LanceDB only understands subclasses of `lancedb.pydantic.LanceModel`
+LanceDB only understands subclasses of `lancedb.pydantic.LanceModel`
-    (which itself derives from `pydantic.BaseModel`).
+(which itself derives from `pydantic.BaseModel`).
 
-    ```python
+```python
-    from lancedb.pydantic import Vector, LanceModel
+from lancedb.pydantic import Vector, LanceModel
 
-    class Content(LanceModel):
+class Content(LanceModel):
-        movie_id: int
+    movie_id: int
-        vector: Vector(128)
+    vector: Vector(128)
-        genres: str
+    genres: str
-        title: str
+    title: str
-        imdb_id: int
+    imdb_id: int
 
-        @property
+    @property
-        def imdb_url(self) -> str:
+    def imdb_url(self) -> str:
-            return f"https://www.imdb.com/title/tt{self.imdb_id}"
+        return f"https://www.imdb.com/title/tt{self.imdb_id}"
 
-    import pyarrow as pa
+import pyarrow as pa
-    db = lancedb.connect("~/.lancedb")
+db = lancedb.connect("~/.lancedb")
-    table_name = "movielens_small"
+table_name = "movielens_small"
-    table = db.create_table(table_name, schema=Content)
+table = db.create_table(table_name, schema=Content)
-    ```
+```
 
-    #### Nested schemas
+#### Nested schemas
 
-    Sometimes your data model may contain nested objects.
+Sometimes your data model may contain nested objects.
-    For example, you may want to store the document string
+For example, you may want to store the document string
-    and the document soure name as a nested Document object:
+and the document soure name as a nested Document object:
 
-    ```python
+```python
-    class Document(BaseModel):
+class Document(BaseModel):
-        content: str
+    content: str
-        source: str
+    source: str
-    ```
+```
 
-    This can be used as the type of a LanceDB table column:
+This can be used as the type of a LanceDB table column:
 
-    ```python
+```python
-    class NestedSchema(LanceModel):
+class NestedSchema(LanceModel):
-        id: str
+    id: str
-        vector: Vector(1536)
+    vector: Vector(1536)
-        document: Document
+    document: Document
 
-    tbl = db.create_table("nested_table", schema=NestedSchema, mode="overwrite")
+tbl = db.create_table("nested_table", schema=NestedSchema, mode="overwrite")
-    ```
+```
 
-    This creates a struct column called "document" that has two subfields 
+This creates a struct column called "document" that has two subfields 
-    called "content" and "source":
+called "content" and "source":
 
-    ```
+```
-    In [28]: tbl.schema
+In [28]: tbl.schema
-    Out[28]:
+Out[28]:
-    id: string not null
+id: string not null
-    vector: fixed_size_list<item: float>[1536] not null
+vector: fixed_size_list<item: float>[1536] not null
-        child 0, item: float
+    child 0, item: float
-    document: struct<content: string not null, source: string not null> not null
+document: struct<content: string not null, source: string not null> not null
-        child 0, content: string not null
+    child 0, content: string not null
-        child 1, source: string not null    
+    child 1, source: string not null    
-    ```
+```
 
-    #### Validators
+#### Validators
 
-    Note that neither Pydantic nor PyArrow automatically validates that input data
+Note that neither Pydantic nor PyArrow automatically validates that input data
-    is of the correct timezone, but this is easy to add as a custom field validator:
+is of the correct timezone, but this is easy to add as a custom field validator:
 
-    ```python
+```python
-    from datetime import datetime
+from datetime import datetime
-    from zoneinfo import ZoneInfo
+from zoneinfo import ZoneInfo
 
-    from lancedb.pydantic import LanceModel
+from lancedb.pydantic import LanceModel
-    from pydantic import Field, field_validator, ValidationError, ValidationInfo
+from pydantic import Field, field_validator, ValidationError, ValidationInfo
 
-    tzname = "America/New_York"
+tzname = "America/New_York"
-    tz = ZoneInfo(tzname)
+tz = ZoneInfo(tzname)
 
-    class TestModel(LanceModel):
+class TestModel(LanceModel):
-        dt_with_tz: datetime = Field(json_schema_extra={"tz": tzname})
+    dt_with_tz: datetime = Field(json_schema_extra={"tz": tzname})
 
-        @field_validator('dt_with_tz')
+    @field_validator('dt_with_tz')
-        @classmethod
+    @classmethod
-        def tz_must_match(cls, dt: datetime) -> datetime:
+    def tz_must_match(cls, dt: datetime) -> datetime:
-            assert dt.tzinfo == tz
+        assert dt.tzinfo == tz
-            return dt        
+        return dt        
 
-    ok = TestModel(dt_with_tz=datetime.now(tz))
+ok = TestModel(dt_with_tz=datetime.now(tz))
 
-    try:
+try:
-        TestModel(dt_with_tz=datetime.now(ZoneInfo("Asia/Shanghai")))
+    TestModel(dt_with_tz=datetime.now(ZoneInfo("Asia/Shanghai")))
-        assert 0 == 1, "this should raise ValidationError"
+    assert 0 == 1, "this should raise ValidationError"
-    except ValidationError:
+except ValidationError:
-        print("A ValidationError was raised.")
+    print("A ValidationError was raised.")
-        pass
+    pass
-    ```
+```
 
-    When you run this code it should print "A ValidationError was raised."
+When you run this code it should print "A ValidationError was raised."
 
-    #### Pydantic custom types
+#### Pydantic custom types
 
-    LanceDB does NOT yet support converting pydantic custom types. If this is something you need,
+LanceDB does NOT yet support converting pydantic custom types. If this is something you need,
-    please file a feature request on the [LanceDB Github repo](https://github.com/lancedb/lancedb/issues/new).
+please file a feature request on the [LanceDB Github repo](https://github.com/lancedb/lancedb/issues/new).
 
-    ### Using Iterators / Writing Large Datasets
+### Using Iterators / Writing Large Datasets
 
-    It is recommended to use iterators to add large datasets in batches when creating your table in one go. This does not create multiple versions of your dataset unlike manually adding batches using `table.add()`
+It is recommended to use iterators to add large datasets in batches when creating your table in one go. This does not create multiple versions of your dataset unlike manually adding batches using `table.add()`
 
-    LanceDB additionally supports PyArrow's `RecordBatch` Iterators or other generators producing supported data types.
+LanceDB additionally supports PyArrow's `RecordBatch` Iterators or other generators producing supported data types.
 
-    Here's an example using using `RecordBatch` iterator for creating tables.
+Here's an example using using `RecordBatch` iterator for creating tables.
 
-    ```python
+```python
-    import pyarrow as pa
+import pyarrow as pa
 
-    def make_batches():
+def make_batches():
-        for i in range(5):
+    for i in range(5):
-            yield pa.RecordBatch.from_arrays(
+        yield pa.RecordBatch.from_arrays(
-                [
+            [
-                    pa.array([[3.1, 4.1, 5.1, 6.1], [5.9, 26.5, 4.7, 32.8]],
+                pa.array([[3.1, 4.1, 5.1, 6.1], [5.9, 26.5, 4.7, 32.8]],
-                            pa.list_(pa.float32(), 4)),
+                        pa.list_(pa.float32(), 4)),
-                    pa.array(["foo", "bar"]),
+                pa.array(["foo", "bar"]),
-                    pa.array([10.0, 20.0]),
+                pa.array([10.0, 20.0]),
-                ],
+            ],
-                ["vector", "item", "price"],
+            ["vector", "item", "price"],
-            )
+        )
 
-    schema = pa.schema([
+schema = pa.schema([
-        pa.field("vector", pa.list_(pa.float32(), 4)),
+    pa.field("vector", pa.list_(pa.float32(), 4)),
-        pa.field("item", pa.utf8()),
+    pa.field("item", pa.utf8()),
-        pa.field("price", pa.float32()),
+    pa.field("price", pa.float32()),
-    ])
+])
 
-    db.create_table("batched_tale", make_batches(), schema=schema)
+db.create_table("batched_tale", make_batches(), schema=schema)
-    ```
+```
 
-    You can also use iterators of other types like Pandas DataFrame or Pylists directly in the above example.
+You can also use iterators of other types like Pandas DataFrame or Pylists directly in the above example.
 
 ## Open existing tables
 
@@ -636,6 +636,70 @@ The `values` parameter is used to provide the new values for the columns as lite
 
     When rows are updated, they are moved out of the index. The row will still show up in ANN queries, but the query will not be as fast as it would be if the row was in the index. If you update a large proportion of rows, consider rebuilding the index afterwards.
 
+## Consistency
+
+In LanceDB OSS, users can set the `read_consistency_interval` parameter on connections to achieve different levels of read consistency. This parameter determines how frequently the database synchronizes with the underlying storage system to check for updates made by other processes. If another process updates a table, the database will not see the changes until the next synchronization.
+
+There are three possible settings for `read_consistency_interval`:
+
+1. **Unset (default)**: The database does not check for updates to tables made by other processes. This provides the best query performance, but means that clients may not see the most up-to-date data. This setting is suitable for applications where the data does not change during the lifetime of the table reference.
+2. **Zero seconds (Strong consistency)**: The database checks for updates on every read. This provides the strongest consistency guarantees, ensuring that all clients see the latest committed data. However, it has the most overhead. This setting is suitable when consistency matters more than having high QPS.
+3. **Custom interval (Eventual consistency)**: The database checks for updates at a custom interval, such as every 5 seconds. This provides eventual consistency, allowing for some lag between write and read operations. Performance wise, this is a middle ground between strong consistency and no consistency check. This setting is suitable for applications where immediate consistency is not critical, but clients should see updated data eventually.
+
+!!! tip "Consistency in LanceDB Cloud"
+
+    This is only tune-able in LanceDB OSS. In LanceDB Cloud, readers are always eventually consistent.
+
+=== "Python"
+    
+    To set strong consistency, use `timedelta(0)`:
+
+    ```python
+    from datetime import timedelta
+    db = lancedb.connect("./.lancedb",. read_consistency_interval=timedelta(0))
+    table = db.open_table("my_table")
+    ```
+
+    For eventual consistency, use a custom `timedelta`:
+
+    ```python
+    from datetime import timedelta
+    db = lancedb.connect("./.lancedb", read_consistency_interval=timedelta(seconds=5))
+    table = db.open_table("my_table")
+    ```
+
+    By default, a `Table` will never check for updates from other writers. To manually check for updates you can use `checkout_latest`:
+
+    ```python
+    db = lancedb.connect("./.lancedb")
+    table = db.open_table("my_table")
+    
+    # (Other writes happen to my_table from another process)
+
+    # Check for updates
+    table.checkout_latest()
+    ```
+
+=== "JavaScript/Typescript"
+
+    To set strong consistency, use `0`:
+
+    ```javascript
+    const db = await lancedb.connect({ uri: "./.lancedb", readConsistencyInterval: 0 });
+    const table = await db.openTable("my_table");
+    ```
+
+    For eventual consistency, specify the update interval as seconds:
+
+    ```javascript
+    const db = await lancedb.connect({ uri: "./.lancedb", readConsistencyInterval: 5 });
+    const table = await db.openTable("my_table");
+    ```
+
+<!-- Node doesn't yet support the version time travel: https://github.com/lancedb/lancedb/issues/1007 
+    Once it does, we can show manual consistency check for Node as well.
+-->
+
 ## What's next?
 
 Learn the best practices on creating an ANN index and getting the most out of it.

docs/src/hybrid_search/eval.md

@@ -0,0 +1,49 @@
+# Hybrid Search
+
+Hybrid Search is a broad (often misused) term. It can mean anything from combining multiple methods for searching, to applying ranking methods to better sort the results. In this blog, we use the definition of "hybrid search" to mean using a combination of keyword-based and vector search.
+
+## The challenge of (re)ranking search results
+Once you have a group of the most relevant search results from multiple search sources, you'd likely standardize the score and rank them accordingly. This process can also be seen as another independent step - reranking.
+There are two approaches for reranking search results from multiple sources.
+* <b>Score-based</b>: Calculate final relevance scores based on a weighted linear combination of individual search algorithm scores. Example - Weighted linear combination of semantic search & keyword-based search results.
+* <b>Relevance-based</b>: Discards the existing scores and calculates the relevance of each search result - query pair. Example - Cross Encoder models
+
+Even though there are many strategies for reranking search results, none works for all cases. Moreover, evaluating them itself is a challenge. Also, reranking can be dataset, application specific so it's hard to generalize.
+
+### Example evaluation of hybrid search with Reranking
+
+Here's some evaluation numbers from experiment comparing these re-rankers on about 800 queries. It is modified version of an evaluation script from [llama-index](https://github.com/run-llama/finetune-embedding/blob/main/evaluate.ipynb) that measures hit-rate at top-k.
+
+<b> With OpenAI ada2 embedding </b>
+
+Vector Search baseline - `0.64`
+
+| Reranker | Top-3 | Top-5 | Top-10 |
+| --- | --- | --- | --- |
+| Linear Combination | `0.73` | `0.74` | `0.85` |
+| Cross Encoder | `0.71` | `0.70` | `0.77` |
+| Cohere | `0.81` | `0.81` | `0.85` |
+| ColBERT | `0.68` | `0.68` | `0.73` |
+
+<p>
+<img src="https://github.com/AyushExel/assets/assets/15766192/d57b1780-ef27-414c-a5c3-73bee7808a45">
+</p>
+
+<b> With OpenAI embedding-v3-small </b>
+
+Vector Search baseline - `0.59`
+
+| Reranker | Top-3 | Top-5 | Top-10 |
+| --- | --- | --- | --- |
+| Linear Combination | `0.68` | `0.70` | `0.84` |
+| Cross Encoder | `0.72` | `0.72` | `0.79` |
+| Cohere | `0.79` | `0.79` | `0.84` |
+| ColBERT | `0.70` | `0.70` | `0.76` |
+
+<p>
+<img src="https://github.com/AyushExel/assets/assets/15766192/259adfd2-6ec6-4df6-a77d-1456598970dd">
+</p>
+
+### Conclusion
+
+The results show that the reranking methods are able to improve the search results. However, the improvement is not consistent across all rerankers. The choice of reranker depends on the dataset and the application. It is also important to note that the reranking methods are not a replacement for the search methods. They are complementary and should be used together to get the best results. The speed to recall tradeoff is also an important factor to consider when choosing the reranker.

docs/src/hybrid_search/hybrid_search.md

@@ -1,6 +1,6 @@
 # Hybrid Search
 
-LanceDB supports both semantic and keyword-based search. In real world applications, it is often useful to combine these two approaches to get the best best results. For example, you may want to search for a document that is semantically similar to a query document, but also contains a specific keyword. This is an example of *hybrid search*, a search algorithm that combines multiple search techniques.
+LanceDB supports both semantic and keyword-based search (also termed full-text search, or FTS). In real world applications, it is often useful to combine these two approaches to get the best best results. For example, you may want to search for a document that is semantically similar to a query document, but also contains a specific keyword. This is an example of *hybrid search*, a search algorithm that combines multiple search techniques.
 
 ## Hybrid search in LanceDB
 You can perform hybrid search in LanceDB by combining the results of semantic and full-text search via a reranking algorithm of your choice. LanceDB provides multiple rerankers out of the box. However, you can always write a custom reranker if your use case need more sophisticated logic .
@@ -69,7 +69,7 @@ reranker = LinearCombinationReranker(weight=0.3) # Use 0.3 as the weight for vec
 results = table.search("rebel", query_type="hybrid").rerank(reranker=reranker).to_pandas()
 ```
 
-Arguments
+### Arguments
 ----------------
 * `weight`: `float`, default `0.7`:
     The weight to use for the semantic search score. The weight for the full-text search score is `1 - weights`.
@@ -91,9 +91,9 @@ reranker = CohereReranker()
 results = table.search("vampire weekend", query_type="hybrid").rerank(reranker=reranker).to_pandas()
 ```
 
-Arguments
+### Arguments
 ----------------
-* `model_name`` : str, default `"rerank-english-v2.0"``
+* `model_name` : str, default `"rerank-english-v2.0"`
         The name of the cross encoder model to use. Available cohere models are:
         - rerank-english-v2.0
         - rerank-multilingual-v2.0
@@ -117,7 +117,7 @@ results = table.search("harmony hall", query_type="hybrid").rerank(reranker=rera
 ```
 
 
-Arguments
+### Arguments
 ----------------
 * `model` : str, default `"cross-encoder/ms-marco-TinyBERT-L-6"`
         The name of the cross encoder model to use. Available cross encoder models can be found [here](https://www.sbert.net/docs/pretrained_cross-encoders.html)
@@ -143,7 +143,7 @@ reranker = ColbertReranker()
 results = table.search("harmony hall", query_type="hybrid").rerank(reranker=reranker).to_pandas()
 ```
 
-Arguments
+### Arguments
 ----------------
 * `model_name` : `str`, default `"colbert-ir/colbertv2.0"`
         The name of the cross encoder model to use.
@@ -162,7 +162,8 @@ This reranker uses the OpenAI API to combine the results of semantic and full-te
     This prompts chat model to rerank results which is not a dedicated reranker model. This should be treated as experimental.
 
 !!! Tip
-    You might run out of token limit so set the search `limits` based on your token limit.
+    - You might run out of token limit so set the search `limits` based on your token limit.
+    - It is recommended to use gpt-4-turbo-preview, the default model, older models might lead to undesired behaviour
 
 ```python
 from lancedb.rerankers import OpenaiReranker
@@ -172,15 +173,15 @@ reranker = OpenaiReranker()
 results = table.search("harmony hall", query_type="hybrid").rerank(reranker=reranker).to_pandas()
 ```
 
-Arguments
+### Arguments
 ----------------
-`model_name` : `str`, default `"gpt-3.5-turbo-1106"`
+* `model_name` : `str`, default `"gpt-4-turbo-preview"`
     The name of the cross encoder model to use.
-`column` : `str`, default `"text"`
+* `column` : `str`, default `"text"`
     The name of the column to use as input to the cross encoder model.
-`return_score` : `str`, default `"relevance"`
+* `return_score` : `str`, default `"relevance"`
     options are "relevance" or "all". Only "relevance" is supported for now.
-`api_key` : `str`, default `None`
+* `api_key` : `str`, default `None`
     The API key to use. If None, will use the OPENAI_API_KEY environment variable.
 
 
@@ -212,24 +213,30 @@ class MyReranker(Reranker):
 
 ```
 
-You can also accept additional arguments like a filter along with fts and vector search results
+### Example of a Custom Reranker
+For the sake of simplicity let's build custom reranker that just enchances the Cohere Reranker by accepting a filter query, and accept other CohereReranker params as kwags.
 
 ```python
 
-from lancedb.rerankers import Reranker
+from typing import List, Union
-import pyarrow as pa
+import pandas as pd
+from lancedb.rerankers import CohereReranker
 
-class MyReranker(Reranker):
+class MofidifiedCohereReranker(CohereReranker):
-    ...
+    def __init__(self, filters: Union[str, List[str]], **kwargs):
-    
+        super().__init__(**kwargs)
-    def rerank_hybrid(self, query: str, vector_results: pa.Table, fts_results: pa.Table, filter: str):
+        filters = filters if isinstance(filters, list) else [filters]
-        # Use the built-in merging function
+        self.filters = filters
-        combined_result = self.merge_results(vector_results, fts_results)
-        
-        # Do something with the combined results & filter
-        # ...
 
-        # Return the combined results
+    def rerank_hybrid(self, query: str, vector_results: pa.Table, fts_results: pa.Table)-> pa.Table:
-        return combined_result
+        combined_result = super().rerank_hybrid(query, vector_results, fts_results)
+        df = combined_result.to_pandas()
+        for filter in self.filters:
+            df = df.query("not text.str.contains(@filter)")
+
+        return pa.Table.from_pandas(df)
 
 ```
+
+!!! tip
+    The `vector_results` and `fts_results` are pyarrow tables. You can convert them to pandas dataframes using `to_pandas()` method and perform any operations you want. After you are done, you can convert the dataframe back to pyarrow table using `pa.Table.from_pandas()` method and return it.

docs/src/index.md

@@ -28,7 +28,7 @@ LanceDB **Cloud** is a SaaS (software-as-a-service) solution that runs serverles
 
 * Fast production-scale vector similarity, full-text & hybrid search and a SQL query interface (via [DataFusion](https://github.com/apache/arrow-datafusion))
 
-* Native Python and Javascript/Typescript support
+* Python, Javascript/Typescript, and Rust support
 
 * Store, query & manage multi-modal data (text, images, videos, point clouds, etc.), not just the embeddings and metadata
 
@@ -54,3 +54,4 @@ The following pages go deeper into the internal of LanceDB and how to use it.
 * [Ecosystem Integrations](integrations/index.md): Integrate LanceDB with other tools in the data ecosystem
 * [Python API Reference](python/python.md): Python OSS and Cloud API references
 * [JavaScript API Reference](javascript/modules.md): JavaScript OSS and Cloud API references
+* [Rust API Reference](https://docs.rs/lancedb/latest/lancedb/index.html): Rust API reference

docs/src/integrations/index.md

@@ -13,7 +13,7 @@ Get started using these examples and quick links.
 | Integrations | |
 |---|---:|
 | <h3> LlamaIndex </h3>LlamaIndex is a simple, flexible data framework for connecting custom data sources to large language models. Llama index integrates with LanceDB as the serverless VectorDB. <h3>[Lean More](https://gpt-index.readthedocs.io/en/latest/examples/vector_stores/LanceDBIndexDemo.html) </h3> |<img src="../assets/llama-index.jpg" alt="image" width="150" height="auto">|
-| <h3>Langchain</h3>Langchain allows building applications with LLMs through composability <h3>[Lean More](https://python.langchain.com/en/latest/modules/indexes/vectorstores/examples/lancedb.html) | <img src="../assets/langchain.png" alt="image" width="150" height="auto">|
+| <h3>Langchain</h3>Langchain allows building applications with LLMs through composability <h3>[Lean More](https://python.langchain.com/docs/integrations/vectorstores/lancedb) | <img src="../assets/langchain.png" alt="image" width="150" height="auto">|
 | <h3>Langchain TS</h3> Javascript bindings for Langchain. It integrates with LanceDB's serverless vectordb allowing you to build powerful AI applications through composibility using only serverless functions. <h3>[Learn More]( https://js.langchain.com/docs/modules/data_connection/vectorstores/integrations/lancedb) | <img src="../assets/langchain.png" alt="image" width="150" height="auto">|
 | <h3>Voxel51</h3>  It is an open source toolkit that enables you to build better computer vision workflows by improving the quality of your datasets and delivering insights about your models.<h3>[Learn More](./voxel51.md) | <img src="../assets/voxel.gif" alt="image" width="150" height="auto">|
 | <h3>PromptTools</h3>  Offers a set of free, open-source tools for testing and experimenting with models, prompts, and configurations. The core idea is to enable developers to evaluate prompts using familiar interfaces like code and notebooks. You can use it to experiment with different configurations of LanceDB, and test how LanceDB integrates with the LLM of your choice.<h3>[Learn More](./prompttools.md) | <img src="../assets/prompttools.jpeg" alt="image" width="150" height="auto">|

docs/src/notebooks/DisappearingEmbeddingFunction.ipynb

@@ -290,7 +290,7 @@
         "from lancedb.pydantic import LanceModel, Vector\n",
         "\n",
         "class Pets(LanceModel):\n",
-        "    vector: Vector(clip.ndims) = clip.VectorField()\n",
+        "    vector: Vector(clip.ndims()) = clip.VectorField()\n",
         "    image_uri: str = clip.SourceField()\n",
         "\n",
         "    @property\n",
@@ -360,7 +360,7 @@
         "    table = db.create_table(\"pets\", schema=Pets)\n",
         "    # use a sampling of 1000 images\n",
         "    p = Path(\"~/Downloads/images\").expanduser()\n",
-        "    uris = [str(f) for f in p.iterdir()]\n",
+        "    uris = [str(f) for f in p.glob(\"*.jpg\")]\n",
         "    uris = sample(uris, 1000)\n",
         "    table.add(pd.DataFrame({\"image_uri\": uris}))"
       ]
@@ -543,7 +543,7 @@
       ],
       "source": [
         "from PIL import Image\n",
-        "p = Path(\"/Users/changshe/Downloads/images/samoyed_100.jpg\")\n",
+        "p = Path(\"~/Downloads/images/samoyed_100.jpg\").expanduser()\n",
         "query_image = Image.open(p)\n",
         "query_image"
       ]

docs/src/notebooks/diffusiondb/datagen.py

@@ -23,10 +23,8 @@ from multiprocessing import Pool
 import lance
 import pyarrow as pa
 from datasets import load_dataset
-from PIL import Image
 from transformers import CLIPModel, CLIPProcessor, CLIPTokenizerFast
 
-import lancedb
 
 MODEL_ID = "openai/clip-vit-base-patch32"
 

docs/src/python/duckdb.md

@@ -1,6 +1,9 @@
 # DuckDB
 
-LanceDB is very well-integrated with [DuckDB](https://duckdb.org/), an in-process SQL OLAP database. This integration is done via [Arrow](https://duckdb.org/docs/guides/python/sql_on_arrow) .
+In Python, LanceDB tables can also be queried with [DuckDB](https://duckdb.org/), an in-process SQL OLAP database. This means you can write complex SQL queries to analyze your data in LanceDB.
+
+This integration is done via [Apache Arrow](https://duckdb.org/docs/guides/python/sql_on_arrow), which provides zero-copy data sharing between LanceDB and DuckDB. DuckDB is capable of passing down column selections and basic filters to LanceDB, reducing the amount of data that needs to be scanned to perform your query. Finally, the integration allows streaming data from LanceDB tables, allowing you to aggregate tables that won't fit into memory. All of this uses the same mechanism described in DuckDB's blog post *[DuckDB quacks Arrow](https://duckdb.org/2021/12/03/duck-arrow.html)*.
+
 
 We can demonstrate this by first installing `duckdb` and `lancedb`.
 
@@ -19,14 +22,15 @@ data = [
     {"vector": [5.9, 26.5], "item": "bar", "price": 20.0}
 ]
 table = db.create_table("pd_table", data=data)
-arrow_table = table.to_arrow()
 ```
 
-DuckDB can directly query the `pyarrow.Table` object:
+To query the table, first call `to_lance` to convert the table to a "dataset", which is an object that can be queried by DuckDB. Then all you need to do is reference that dataset by the same name in your SQL query.
 
 ```python
 import duckdb
 
+arrow_table = table.to_lance()
+
 duckdb.query("SELECT * FROM arrow_table")
 ```
 

docs/src/python/python.md

@@ -24,6 +24,12 @@ pip install lancedb
 
 ::: lancedb.query.LanceQueryBuilder
 
+::: lancedb.query.LanceVectorQueryBuilder
+
+::: lancedb.query.LanceFtsQueryBuilder
+
+::: lancedb.query.LanceHybridQueryBuilder
+
 ## Embeddings
 
 ::: lancedb.embeddings.registry.EmbeddingFunctionRegistry
@@ -62,10 +68,22 @@ pip install lancedb
 
 ## Integrations
 
-### Pydantic
+## Pydantic
 
 ::: lancedb.pydantic.pydantic_to_schema
 
 ::: lancedb.pydantic.vector
 
 ::: lancedb.pydantic.LanceModel
+
+## Reranking
+
+::: lancedb.rerankers.linear_combination.LinearCombinationReranker
+
+::: lancedb.rerankers.cohere.CohereReranker
+
+::: lancedb.rerankers.colbert.ColbertReranker
+
+::: lancedb.rerankers.cross_encoder.CrossEncoderReranker
+
+::: lancedb.rerankers.openai.OpenaiReranker

docs/src/search.md

@@ -22,7 +22,7 @@ Currently, LanceDB supports the following metrics:
 ## Exhaustive search (kNN)
 
 If you do not create a vector index, LanceDB exhaustively scans the _entire_ vector space
-and compute the distance to every vector in order to find the exact nearest neighbors. This is effectively a kNN search.
+and computes the distance to every vector in order to find the exact nearest neighbors. This is effectively a kNN search.
 
 <!-- Setup Code
 ```python
@@ -85,7 +85,7 @@ To perform scalable vector retrieval with acceptable latencies, it's common to b
 While the exhaustive search is guaranteed to always return 100% recall, the approximate nature of
 an ANN search means that using an index often involves a trade-off between recall and latency.
 
-See the [IVF_PQ index](./concepts/index_ivfpq.md.md) for a deeper description of how `IVF_PQ`
+See the [IVF_PQ index](./concepts/index_ivfpq.md) for a deeper description of how `IVF_PQ`
 indexes work in LanceDB.
 
 ## Output search results
@@ -184,4 +184,3 @@ Let's create a LanceDB table with a nested schema:
 
     Note that in this case the extra `_distance` field is discarded since
     it's not part of the LanceSchema.
-

docs/test/md_testing.py

@@ -14,7 +14,7 @@ excluded_globs = [
     "../src/concepts/*.md",
     "../src/ann_indexes.md",
     "../src/basic.md",
-    "../src/hybrid_search.md",
+    "../src/hybrid_search/hybrid_search.md",
 ]
 
 python_prefix = "py"

node/.eslintrc.js

@@ -13,5 +13,10 @@ module.exports = {
   },
   rules: {
     "@typescript-eslint/method-signature-style": "off",
+    "@typescript-eslint/quotes": "off",
+    "@typescript-eslint/semi": "off",
+    "@typescript-eslint/explicit-function-return-type": "off",
+    "@typescript-eslint/space-before-function-paren": "off",
+    "@typescript-eslint/indent": "off",
   }
 }

node/package-lock.json

@@ -1,12 +1,12 @@
 {
   "name": "vectordb",
-  "version": "0.4.8",
+  "version": "0.4.13",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "vectordb",
-      "version": "0.4.8",
+      "version": "0.4.13",
       "cpu": [
         "x64",
         "arm64"
@@ -18,9 +18,7 @@
         "win32"
       ],
       "dependencies": {
-        "@apache-arrow/ts": "^14.0.2",
         "@neon-rs/load": "^0.0.74",
-        "apache-arrow": "^14.0.2",
         "axios": "^1.4.0"
       },
       "devDependencies": {
@@ -33,6 +31,7 @@
         "@types/temp": "^0.9.1",
         "@types/uuid": "^9.0.3",
         "@typescript-eslint/eslint-plugin": "^5.59.1",
+        "apache-arrow-old": "npm:apache-arrow@13.0.0",
         "cargo-cp-artifact": "^0.1",
         "chai": "^4.3.7",
         "chai-as-promised": "^7.1.1",
@@ -53,11 +52,15 @@
         "uuid": "^9.0.0"
       },
       "optionalDependencies": {
-        "@lancedb/vectordb-darwin-arm64": "0.4.8",
+        "@lancedb/vectordb-darwin-arm64": "0.4.13",
-        "@lancedb/vectordb-darwin-x64": "0.4.8",
+        "@lancedb/vectordb-darwin-x64": "0.4.13",
-        "@lancedb/vectordb-linux-arm64-gnu": "0.4.8",
+        "@lancedb/vectordb-linux-arm64-gnu": "0.4.13",
-        "@lancedb/vectordb-linux-x64-gnu": "0.4.8",
+        "@lancedb/vectordb-linux-x64-gnu": "0.4.13",
-        "@lancedb/vectordb-win32-x64-msvc": "0.4.8"
+        "@lancedb/vectordb-win32-x64-msvc": "0.4.13"
+      },
+      "peerDependencies": {
+        "@apache-arrow/ts": "^14.0.2",
+        "apache-arrow": "^14.0.2"
       }
     },
     "node_modules/@75lb/deep-merge": {
@@ -93,6 +96,7 @@
       "version": "14.0.2",
       "resolved": "https://registry.npmjs.org/@apache-arrow/ts/-/ts-14.0.2.tgz",
       "integrity": "sha512-CtwAvLkK0CZv7xsYeCo91ml6PvlfzAmAJZkRYuz2GNBwfYufj5SVi0iuSMwIMkcU/szVwvLdzORSLa5PlF/2ug==",
+      "peer": true,
       "dependencies": {
         "@types/command-line-args": "5.2.0",
         "@types/command-line-usage": "5.0.2",
@@ -109,7 +113,8 @@
     "node_modules/@apache-arrow/ts/node_modules/@types/node": {
       "version": "20.3.0",
       "resolved": "https://registry.npmjs.org/@types/node/-/node-20.3.0.tgz",
-      "integrity": "sha512-cumHmIAf6On83X7yP+LrsEyUOf/YlociZelmpRYaGFydoaPdxdt80MAbu6vWerQT2COCp2nPvHdsbD7tHn/YlQ=="
+      "integrity": "sha512-cumHmIAf6On83X7yP+LrsEyUOf/YlociZelmpRYaGFydoaPdxdt80MAbu6vWerQT2COCp2nPvHdsbD7tHn/YlQ==",
+      "peer": true
     },
     "node_modules/@cargo-messages/android-arm-eabi": {
       "version": "0.0.160",
@@ -329,9 +334,9 @@
       }
     },
     "node_modules/@lancedb/vectordb-darwin-arm64": {
-      "version": "0.4.8",
+      "version": "0.4.13",
-      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-darwin-arm64/-/vectordb-darwin-arm64-0.4.8.tgz",
+      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-darwin-arm64/-/vectordb-darwin-arm64-0.4.13.tgz",
-      "integrity": "sha512-FpnJaw7KmNdD/FtOw9AcmPL5P+L04AcnfPj9ZyEjN8iCwB/qaOGYgdfBv+EbEtfHIsqA12q/1BRduu9KdB6BIA==",
+      "integrity": "sha512-JfroNCG8yKIU931Y+x8d0Fp8C9DHUSC5j+CjI+e5err7rTWtie4j3JbsXlWAnPFaFEOg0Xk3BWkSikCvhPGJGg==",
       "cpu": [
         "arm64"
       ],
@@ -341,9 +346,9 @@
       ]
     },
     "node_modules/@lancedb/vectordb-darwin-x64": {
-      "version": "0.4.8",
+      "version": "0.4.13",
-      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-darwin-x64/-/vectordb-darwin-x64-0.4.8.tgz",
+      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-darwin-x64/-/vectordb-darwin-x64-0.4.13.tgz",
-      "integrity": "sha512-RafOEYyZIgphp8wPGuVLFaTc8aAqo0NCO1LQMx0mB0xV96vrdo0Mooivs+dYN3RFfSHtTKPw9O1Jc957Vp1TLg==",
+      "integrity": "sha512-dG6IMvfpHpnHdbJ0UffzJ7cZfMiC02MjIi6YJzgx+hKz2UNXWNBIfTvvhqli85mZsGRXL1OYDdYv0K1YzNjXlA==",
       "cpu": [
         "x64"
       ],
@@ -353,9 +358,9 @@
       ]
     },
     "node_modules/@lancedb/vectordb-linux-arm64-gnu": {
-      "version": "0.4.8",
+      "version": "0.4.13",
-      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-linux-arm64-gnu/-/vectordb-linux-arm64-gnu-0.4.8.tgz",
+      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-linux-arm64-gnu/-/vectordb-linux-arm64-gnu-0.4.13.tgz",
-      "integrity": "sha512-WlbYNfj4+v1hBHUluF+hnlG/A0ZaQFdXBTGDfHQniL11o+n3emWm4ujP5nSAoQHXjSH9DaOTGr/N4Mc9Xe+luw==",
+      "integrity": "sha512-BRR1VzaMviXby7qmLm0axNZM8eUZF3ZqfvnDKdVRpC3LaRueD6pMXHuC2IUKaFkn7xktf+8BlDZb6foFNEj8bQ==",
       "cpu": [
         "arm64"
       ],
@@ -365,9 +370,9 @@
       ]
     },
     "node_modules/@lancedb/vectordb-linux-x64-gnu": {
-      "version": "0.4.8",
+      "version": "0.4.13",
-      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-linux-x64-gnu/-/vectordb-linux-x64-gnu-0.4.8.tgz",
+      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-linux-x64-gnu/-/vectordb-linux-x64-gnu-0.4.13.tgz",
-      "integrity": "sha512-z+qFJrDqnNEv4JcwYDyt51PHmWjuM/XaOlSjpBnyyuUImeY+QcwctMuyXt8+Q4zhuqQR1AhLKrMwCU+YmMfk5g==",
+      "integrity": "sha512-WnekZ7ZMlria+NODZ6aBCljCFQSe2bBNUS9ZpyFl/Y1vHduSQPuBxM6V7vp2QubC0daq/rifgjDob89DF+x3xw==",
       "cpu": [
         "x64"
       ],
@@ -377,9 +382,9 @@
       ]
     },
     "node_modules/@lancedb/vectordb-win32-x64-msvc": {
-      "version": "0.4.8",
+      "version": "0.4.13",
-      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-win32-x64-msvc/-/vectordb-win32-x64-msvc-0.4.8.tgz",
+      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-win32-x64-msvc/-/vectordb-win32-x64-msvc-0.4.13.tgz",
-      "integrity": "sha512-VjUryVvEA04r0j4lU9pJy84cmjuQm1GhBzbPc8kwbn5voT4A6BPglrlNsU0Zc+j8Fbjyvauzw2lMEcMsF4F0rw==",
+      "integrity": "sha512-3NDpMWBL2ksDHXAraXhowiLqQcNWM5bdbeHwze4+InYMD54hyQ2ODNc+4usxp63Nya9biVnFS27yXULqkzIEqQ==",
       "cpu": [
         "x64"
       ],
@@ -948,6 +953,7 @@
       "version": "14.0.2",
       "resolved": "https://registry.npmjs.org/apache-arrow/-/apache-arrow-14.0.2.tgz",
       "integrity": "sha512-EBO2xJN36/XoY81nhLcwCJgFwkboDZeyNQ+OPsG7bCoQjc2BT0aTyH/MR6SrL+LirSNz+cYqjGRlupMMlP1aEg==",
+      "peer": true,
       "dependencies": {
         "@types/command-line-args": "5.2.0",
         "@types/command-line-usage": "5.0.2",
@@ -964,10 +970,39 @@
         "arrow2csv": "bin/arrow2csv.js"
       }
     },
+    "node_modules/apache-arrow-old": {
+      "name": "apache-arrow",
+      "version": "13.0.0",
+      "resolved": "https://registry.npmjs.org/apache-arrow/-/apache-arrow-13.0.0.tgz",
+      "integrity": "sha512-3gvCX0GDawWz6KFNC28p65U+zGh/LZ6ZNKWNu74N6CQlKzxeoWHpi4CgEQsgRSEMuyrIIXi1Ea2syja7dwcHvw==",
+      "dev": true,
+      "dependencies": {
+        "@types/command-line-args": "5.2.0",
+        "@types/command-line-usage": "5.0.2",
+        "@types/node": "20.3.0",
+        "@types/pad-left": "2.1.1",
+        "command-line-args": "5.2.1",
+        "command-line-usage": "7.0.1",
+        "flatbuffers": "23.5.26",
+        "json-bignum": "^0.0.3",
+        "pad-left": "^2.1.0",
+        "tslib": "^2.5.3"
+      },
+      "bin": {
+        "arrow2csv": "bin/arrow2csv.js"
+      }
+    },
+    "node_modules/apache-arrow-old/node_modules/@types/node": {
+      "version": "20.3.0",
+      "resolved": "https://registry.npmjs.org/@types/node/-/node-20.3.0.tgz",
+      "integrity": "sha512-cumHmIAf6On83X7yP+LrsEyUOf/YlociZelmpRYaGFydoaPdxdt80MAbu6vWerQT2COCp2nPvHdsbD7tHn/YlQ==",
+      "dev": true
+    },
     "node_modules/apache-arrow/node_modules/@types/node": {
       "version": "20.3.0",
       "resolved": "https://registry.npmjs.org/@types/node/-/node-20.3.0.tgz",
-      "integrity": "sha512-cumHmIAf6On83X7yP+LrsEyUOf/YlociZelmpRYaGFydoaPdxdt80MAbu6vWerQT2COCp2nPvHdsbD7tHn/YlQ=="
+      "integrity": "sha512-cumHmIAf6On83X7yP+LrsEyUOf/YlociZelmpRYaGFydoaPdxdt80MAbu6vWerQT2COCp2nPvHdsbD7tHn/YlQ==",
+      "peer": true
     },
     "node_modules/arg": {
       "version": "4.1.3",

node/package.json

@@ -1,12 +1,12 @@
 {
   "name": "vectordb",
-  "version": "0.4.8",
+  "version": "0.4.13",
   "description": " Serverless, low-latency vector database for AI applications",
   "main": "dist/index.js",
   "types": "dist/index.d.ts",
   "scripts": {
     "tsc": "tsc -b",
-    "build": "npm run tsc && cargo-cp-artifact --artifact cdylib vectordb-node index.node -- cargo build --message-format=json",
+    "build": "npm run tsc && cargo-cp-artifact --artifact cdylib lancedb-node index.node -- cargo build --message-format=json",
     "build-release": "npm run build -- --release",
     "test": "npm run tsc && mocha -recursive dist/test",
     "integration-test": "npm run tsc && mocha -recursive dist/integration_test",
@@ -41,6 +41,7 @@
     "@types/temp": "^0.9.1",
     "@types/uuid": "^9.0.3",
     "@typescript-eslint/eslint-plugin": "^5.59.1",
+    "apache-arrow-old": "npm:apache-arrow@13.0.0",
     "cargo-cp-artifact": "^0.1",
     "chai": "^4.3.7",
     "chai-as-promised": "^7.1.1",
@@ -61,11 +62,13 @@
     "uuid": "^9.0.0"
   },
   "dependencies": {
-    "@apache-arrow/ts": "^14.0.2",
     "@neon-rs/load": "^0.0.74",
-    "apache-arrow": "^14.0.2",
     "axios": "^1.4.0"
   },
+  "peerDependencies": {
+    "@apache-arrow/ts": "^14.0.2",
+    "apache-arrow": "^14.0.2"
+  },
   "os": [
     "darwin",
     "linux",
@@ -85,10 +88,10 @@
     }
   },
   "optionalDependencies": {
-    "@lancedb/vectordb-darwin-arm64": "0.4.8",
+    "@lancedb/vectordb-darwin-arm64": "0.4.13",
-    "@lancedb/vectordb-darwin-x64": "0.4.8",
+    "@lancedb/vectordb-darwin-x64": "0.4.13",
-    "@lancedb/vectordb-linux-arm64-gnu": "0.4.8",
+    "@lancedb/vectordb-linux-arm64-gnu": "0.4.13",
-    "@lancedb/vectordb-linux-x64-gnu": "0.4.8",
+    "@lancedb/vectordb-linux-x64-gnu": "0.4.13",
-    "@lancedb/vectordb-win32-x64-msvc": "0.4.8"
+    "@lancedb/vectordb-win32-x64-msvc": "0.4.13"
   }
 }

node/src/arrow.ts

@@ -14,26 +14,30 @@
 
 import {
   Field,
-  type FixedSizeListBuilder,
-  Float32,
   makeBuilder,
   RecordBatchFileWriter,
   Utf8,
   type Vector,
   FixedSizeList,
   vectorFromArray,
-  type Schema,
+  Schema,
   Table as ArrowTable,
   RecordBatchStreamWriter,
   List,
-  Float64,
   RecordBatch,
   makeData,
   Struct,
-  type Float
+  Float,
+  DataType,
+  Binary,
+  Float32
 } from 'apache-arrow'
 import { type EmbeddingFunction } from './index'
+import { sanitizeSchema } from './sanitize'
 
+/*
+ * Options to control how a column should be converted to a vector array
+ */
 export class VectorColumnOptions {
   /** Vector column type. */
   type: Float = new Float32()
@@ -45,14 +49,50 @@ export class VectorColumnOptions {
 
 /** Options to control the makeArrowTable call. */
 export class MakeArrowTableOptions {
-  /** Provided schema. */
+  /*
+   * Schema of the data.
+   *
+   * If this is not provided then the data type will be inferred from the
+   * JS type.  Integer numbers will become int64, floating point numbers
+   * will become float64 and arrays will become variable sized lists with
+   * the data type inferred from the first element in the array.
+   *
+   * The schema must be specified if there are no records (e.g. to make
+   * an empty table)
+   */
   schema?: Schema
 
-  /** Vector columns */
+  /*
+   * Mapping from vector column name to expected type
+   *
+   * Lance expects vector columns to be fixed size list arrays (i.e. tensors)
+   * However, `makeArrowTable` will not infer this by default (it creates
+   * variable size list arrays).  This field can be used to indicate that a column
+   * should be treated as a vector column and converted to a fixed size list.
+   *
+   * The keys should be the names of the vector columns.  The value specifies the
+   * expected data type of the vector columns.
+   *
+   * If `schema` is provided then this field is ignored.
+   *
+   * By default, the column named "vector" will be assumed to be a float32
+   * vector column.
+   */
   vectorColumns: Record<string, VectorColumnOptions> = {
     vector: new VectorColumnOptions()
   }
 
+  /**
+   * If true then string columns will be encoded with dictionary encoding
+   *
+   * Set this to true if your string columns tend to repeat the same values
+   * often.  For more precise control use the `schema` property to specify the
+   * data type for individual columns.
+   *
+   * If `schema` is provided then this property is ignored.
+   */
+  dictionaryEncodeStrings: boolean = false
+
   constructor (values?: Partial<MakeArrowTableOptions>) {
     Object.assign(this, values)
   }
@@ -62,8 +102,29 @@ export class MakeArrowTableOptions {
  * An enhanced version of the {@link makeTable} function from Apache Arrow
  * that supports nested fields and embeddings columns.
  *
+ * This function converts an array of Record<String, any> (row-major JS objects)
+ * to an Arrow Table (a columnar structure)
+ *
  * Note that it currently does not support nulls.
  *
+ * If a schema is provided then it will be used to determine the resulting array
+ * types.  Fields will also be reordered to fit the order defined by the schema.
+ *
+ * If a schema is not provided then the types will be inferred and the field order
+ * will be controlled by the order of properties in the first record.
+ *
+ * If the input is empty then a schema must be provided to create an empty table.
+ *
+ * When a schema is not specified then data types will be inferred.  The inference
+ * rules are as follows:
+ *
+ *  - boolean => Bool
+ *  - number => Float64
+ *  - String => Utf8
+ *  - Buffer => Binary
+ *  - Record<String, any> => Struct
+ *  - Array<any> => List
+ *
  * @param data input data
  * @param options options to control the makeArrowTable call.
  *
@@ -86,8 +147,10 @@ export class MakeArrowTableOptions {
  *  ], { schema });
  * ```
  *
- * It guesses the vector columns if the schema is not provided. For example,
+ * By default it assumes that the column named `vector` is a vector column
- * by default it assumes that the column named `vector` is a vector column.
+ * and it will be converted into a fixed size list array of type float32.
+ * The `vectorColumns` option can be used to support other vector column
+ * names and data types.
  *
  * ```ts
  *
@@ -134,211 +197,322 @@ export function makeArrowTable (
   data: Array<Record<string, any>>,
   options?: Partial<MakeArrowTableOptions>
 ): ArrowTable {
-  if (data.length === 0) {
+  if (data.length === 0 && (options?.schema === undefined || options?.schema === null)) {
-    throw new Error('At least one record needs to be provided')
+    throw new Error('At least one record or a schema needs to be provided')
   }
+
   const opt = new MakeArrowTableOptions(options !== undefined ? options : {})
+  if (opt.schema !== undefined && opt.schema !== null) {
+    opt.schema = sanitizeSchema(opt.schema)
+  }
   const columns: Record<string, Vector> = {}
   // TODO: sample dataset to find missing columns
-  const columnNames = Object.keys(data[0])
+  // Prefer the field ordering of the schema, if present
+  const columnNames = ((opt.schema) != null) ? (opt.schema.names as string[]) : Object.keys(data[0])
   for (const colName of columnNames) {
-    const values = data.map((datum) => datum[colName])
+    if (data.length !== 0 && !Object.prototype.hasOwnProperty.call(data[0], colName)) {
-    let vector: Vector
+      // The field is present in the schema, but not in the data, skip it
+      continue
+    }
+    // Extract a single column from the records (transpose from row-major to col-major)
+    let values = data.map((datum) => datum[colName])
 
+    // By default (type === undefined) arrow will infer the type from the JS type
+    let type
     if (opt.schema !== undefined) {
-      // Explicit schema is provided, highest priority
+      // If there is a schema provided, then use that for the type instead
-      vector = vectorFromArray(
+      type = opt.schema?.fields.filter((f) => f.name === colName)[0]?.type
-        values,
+      if (DataType.isInt(type) && type.bitWidth === 64) {
-        opt.schema?.fields.filter((f) => f.name === colName)[0]?.type
+        // wrap in BigInt to avoid bug: https://github.com/apache/arrow/issues/40051
-      )
+        values = values.map((v) => {
+          if (v === null) {
+            return v
+          }
+          return BigInt(v)
+        })
+      }
     } else {
+      // Otherwise, check to see if this column is one of the vector columns
+      // defined by opt.vectorColumns and, if so, use the fixed size list type
       const vectorColumnOptions = opt.vectorColumns[colName]
       if (vectorColumnOptions !== undefined) {
-        const fslType = new FixedSizeList(
+        type = newVectorType(values[0].length, vectorColumnOptions.type)
-          values[0].length,
-          new Field('item', vectorColumnOptions.type, false)
-        )
-        vector = vectorFromArray(values, fslType)
-      } else {
-        // Normal case
-        vector = vectorFromArray(values)
       }
     }
-    columns[colName] = vector
+
+    try {
+      // Convert an Array of JS values to an arrow vector
+      columns[colName] = makeVector(values, type, opt.dictionaryEncodeStrings)
+    } catch (error: unknown) {
+      // eslint-disable-next-line @typescript-eslint/restrict-template-expressions
+      throw Error(`Could not convert column "${colName}" to Arrow: ${error}`)
+    }
   }
 
-  return new ArrowTable(columns)
+  if (opt.schema != null) {
+    // `new ArrowTable(columns)` infers a schema which may sometimes have
+    // incorrect nullability (it assumes nullable=true if there are 0 rows)
+    //
+    // `new ArrowTable(schema, columns)` will also fail because it will create a
+    // batch with an inferred schema and then complain that the batch schema
+    // does not match the provided schema.
+    //
+    // To work around this we first create a table with the wrong schema and
+    // then patch the schema of the batches so we can use
+    // `new ArrowTable(schema, batches)` which does not do any schema inference
+    const firstTable = new ArrowTable(columns)
+    // eslint-disable-next-line @typescript-eslint/no-non-null-assertion
+    const batchesFixed = firstTable.batches.map(batch => new RecordBatch(opt.schema!, batch.data))
+    return new ArrowTable(opt.schema, batchesFixed)
+  } else {
+    return new ArrowTable(columns)
+  }
 }
 
-// Converts an Array of records into an Arrow Table, optionally applying an embeddings function to it.
+/**
+ * Create an empty Arrow table with the provided schema
+ */
+export function makeEmptyTable (schema: Schema): ArrowTable {
+  return makeArrowTable([], { schema })
+}
+
+// Helper function to convert Array<Array<any>> to a variable sized list array
+function makeListVector (lists: any[][]): Vector<any> {
+  if (lists.length === 0 || lists[0].length === 0) {
+    throw Error('Cannot infer list vector from empty array or empty list')
+  }
+  const sampleList = lists[0]
+  let inferredType
+  try {
+    const sampleVector = makeVector(sampleList)
+    inferredType = sampleVector.type
+  } catch (error: unknown) {
+    // eslint-disable-next-line @typescript-eslint/restrict-template-expressions
+    throw Error(`Cannot infer list vector.  Cannot infer inner type: ${error}`)
+  }
+
+  const listBuilder = makeBuilder({
+    type: new List(new Field('item', inferredType, true))
+  })
+  for (const list of lists) {
+    listBuilder.append(list)
+  }
+  return listBuilder.finish().toVector()
+}
+
+// Helper function to convert an Array of JS values to an Arrow Vector
+function makeVector (values: any[], type?: DataType, stringAsDictionary?: boolean): Vector<any> {
+  if (type !== undefined) {
+    // No need for inference, let Arrow create it
+    return vectorFromArray(values, type)
+  }
+  if (values.length === 0) {
+    throw Error('makeVector requires at least one value or the type must be specfied')
+  }
+  const sampleValue = values.find(val => val !== null && val !== undefined)
+  if (sampleValue === undefined) {
+    throw Error('makeVector cannot infer the type if all values are null or undefined')
+  }
+  if (Array.isArray(sampleValue)) {
+    // Default Arrow inference doesn't handle list types
+    return makeListVector(values)
+  } else if (Buffer.isBuffer(sampleValue)) {
+    // Default Arrow inference doesn't handle Buffer
+    return vectorFromArray(values, new Binary())
+  } else if (!(stringAsDictionary ?? false) && (typeof sampleValue === 'string' || sampleValue instanceof String)) {
+    // If the type is string then don't use Arrow's default inference unless dictionaries are requested
+    // because it will always use dictionary encoding for strings
+    return vectorFromArray(values, new Utf8())
+  } else {
+    // Convert a JS array of values to an arrow vector
+    return vectorFromArray(values)
+  }
+}
+
+async function applyEmbeddings<T> (table: ArrowTable, embeddings?: EmbeddingFunction<T>, schema?: Schema): Promise<ArrowTable> {
+  if (embeddings == null) {
+    return table
+  }
+  if (schema !== undefined && schema !== null) {
+    schema = sanitizeSchema(schema)
+  }
+
+  // Convert from ArrowTable to Record<String, Vector>
+  const colEntries = [...Array(table.numCols).keys()].map((_, idx) => {
+    const name = table.schema.fields[idx].name
+    // eslint-disable-next-line @typescript-eslint/no-non-null-assertion
+    const vec = table.getChildAt(idx)!
+    return [name, vec]
+  })
+  const newColumns = Object.fromEntries(colEntries)
+
+  const sourceColumn = newColumns[embeddings.sourceColumn]
+  const destColumn = embeddings.destColumn ?? 'vector'
+  const innerDestType = embeddings.embeddingDataType ?? new Float32()
+  if (sourceColumn === undefined) {
+    throw new Error(`Cannot apply embedding function because the source column '${embeddings.sourceColumn}' was not present in the data`)
+  }
+
+  if (table.numRows === 0) {
+    if (Object.prototype.hasOwnProperty.call(newColumns, destColumn)) {
+      // We have an empty table and it already has the embedding column so no work needs to be done
+      // Note: we don't return an error like we did below because this is a common occurrence.  For example,
+      // if we call convertToTable with 0 records and a schema that includes the embedding
+      return table
+    }
+    if (embeddings.embeddingDimension !== undefined) {
+      const destType = newVectorType(embeddings.embeddingDimension, innerDestType)
+      newColumns[destColumn] = makeVector([], destType)
+    } else if (schema != null) {
+      const destField = schema.fields.find(f => f.name === destColumn)
+      if (destField != null) {
+        newColumns[destColumn] = makeVector([], destField.type)
+      } else {
+        throw new Error(`Attempt to apply embeddings to an empty table failed because schema was missing embedding column '${destColumn}'`)
+      }
+    } else {
+      throw new Error('Attempt to apply embeddings to an empty table when the embeddings function does not specify `embeddingDimension`')
+    }
+  } else {
+    if (Object.prototype.hasOwnProperty.call(newColumns, destColumn)) {
+      throw new Error(`Attempt to apply embeddings to table failed because column ${destColumn} already existed`)
+    }
+    if (table.batches.length > 1) {
+      throw new Error('Internal error: `makeArrowTable` unexpectedly created a table with more than one batch')
+    }
+    const values = sourceColumn.toArray()
+    const vectors = await embeddings.embed(values as T[])
+    if (vectors.length !== values.length) {
+      throw new Error('Embedding function did not return an embedding for each input element')
+    }
+    const destType = newVectorType(vectors[0].length, innerDestType)
+    newColumns[destColumn] = makeVector(vectors, destType)
+  }
+
+  const newTable = new ArrowTable(newColumns)
+  if (schema != null) {
+    if (schema.fields.find(f => f.name === destColumn) === undefined) {
+      throw new Error(`When using embedding functions and specifying a schema the schema should include the embedding column but the column ${destColumn} was missing`)
+    }
+    return alignTable(newTable, schema)
+  }
+  return newTable
+}
+
+/*
+ * Convert an Array of records into an Arrow Table, optionally applying an
+ * embeddings function to it.
+ *
+ * This function calls `makeArrowTable` first to create the Arrow Table.
+ * Any provided `makeTableOptions` (e.g. a schema) will be passed on to
+ * that call.
+ *
+ * The embedding function will be passed a column of values (based on the
+ * `sourceColumn` of the embedding function) and expects to receive back
+ * number[][] which will be converted into a fixed size list column.  By
+ * default this will be a fixed size list of Float32 but that can be
+ * customized by the `embeddingDataType` property of the embedding function.
+ *
+ * If a schema is provided in `makeTableOptions` then it should include the
+ * embedding columns.  If no schema is provded then embedding columns will
+ * be placed at the end of the table, after all of the input columns.
+ */
 export async function convertToTable<T> (
   data: Array<Record<string, unknown>>,
-  embeddings?: EmbeddingFunction<T>
+  embeddings?: EmbeddingFunction<T>,
+  makeTableOptions?: Partial<MakeArrowTableOptions>
 ): Promise<ArrowTable> {
-  if (data.length === 0) {
+  const table = makeArrowTable(data, makeTableOptions)
-    throw new Error('At least one record needs to be provided')
+  return await applyEmbeddings(table, embeddings, makeTableOptions?.schema)
-  }
-
-  const columns = Object.keys(data[0])
-  const records: Record<string, Vector> = {}
-
-  for (const columnsKey of columns) {
-    if (columnsKey === 'vector') {
-      const vectorSize = (data[0].vector as any[]).length
-      const listBuilder = newVectorBuilder(vectorSize)
-      for (const datum of data) {
-        if ((datum[columnsKey] as any[]).length !== vectorSize) {
-          throw new Error(`Invalid vector size, expected ${vectorSize}`)
-        }
-
-        listBuilder.append(datum[columnsKey])
-      }
-      records[columnsKey] = listBuilder.finish().toVector()
-    } else {
-      const values = []
-      for (const datum of data) {
-        values.push(datum[columnsKey])
-      }
-
-      if (columnsKey === embeddings?.sourceColumn) {
-        const vectors = await embeddings.embed(values as T[])
-        records.vector = vectorFromArray(
-          vectors,
-          newVectorType(vectors[0].length)
-        )
-      }
-
-      if (typeof values[0] === 'string') {
-        // `vectorFromArray` converts strings into dictionary vectors, forcing it back to a string column
-        records[columnsKey] = vectorFromArray(values, new Utf8())
-      } else if (Array.isArray(values[0])) {
-        const elementType = getElementType(values[0])
-        let innerType
-        if (elementType === 'string') {
-          innerType = new Utf8()
-        } else if (elementType === 'number') {
-          innerType = new Float64()
-        } else {
-          // TODO: pass in schema if it exists, else keep going to the next element
-          throw new Error(`Unsupported array element type ${elementType}`)
-        }
-        const listBuilder = makeBuilder({
-          type: new List(new Field('item', innerType, true))
-        })
-        for (const value of values) {
-          listBuilder.append(value)
-        }
-        records[columnsKey] = listBuilder.finish().toVector()
-      } else {
-        // TODO if this is a struct field then recursively align the subfields
-        records[columnsKey] = vectorFromArray(values)
-      }
-    }
-  }
-
-  return new ArrowTable(records)
-}
-
-function getElementType (arr: any[]): string {
-  if (arr.length === 0) {
-    return 'undefined'
-  }
-
-  return typeof arr[0]
-}
-
-// Creates a new Arrow ListBuilder that stores a Vector column
-function newVectorBuilder (dim: number): FixedSizeListBuilder<Float32> {
-  return makeBuilder({
-    type: newVectorType(dim)
-  })
 }
 
 // Creates the Arrow Type for a Vector column with dimension `dim`
-function newVectorType (dim: number): FixedSizeList<Float32> {
+function newVectorType <T extends Float> (dim: number, innerType: T): FixedSizeList<T> {
   // Somewhere we always default to have the elements nullable, so we need to set it to true
   // otherwise we often get schema mismatches because the stored data always has schema with nullable elements
-  const children = new Field<Float32>('item', new Float32(), true)
+  const children = new Field<T>('item', innerType, true)
   return new FixedSizeList(dim, children)
 }
 
-// Converts an Array of records into Arrow IPC format
+/**
+ * Serialize an Array of records into a buffer using the Arrow IPC File serialization
+ *
+ * This function will call `convertToTable` and pass on `embeddings` and `schema`
+ *
+ * `schema` is required if data is empty
+ */
 export async function fromRecordsToBuffer<T> (
   data: Array<Record<string, unknown>>,
   embeddings?: EmbeddingFunction<T>,
   schema?: Schema
 ): Promise<Buffer> {
-  let table = await convertToTable(data, embeddings)
+  if (schema !== undefined && schema !== null) {
-  if (schema !== undefined) {
+    schema = sanitizeSchema(schema)
-    table = alignTable(table, schema)
   }
+  const table = await convertToTable(data, embeddings, { schema })
   const writer = RecordBatchFileWriter.writeAll(table)
   return Buffer.from(await writer.toUint8Array())
 }
 
-// Converts an Array of records into Arrow IPC stream format
+/**
+ * Serialize an Array of records into a buffer using the Arrow IPC Stream serialization
+ *
+ * This function will call `convertToTable` and pass on `embeddings` and `schema`
+ *
+ * `schema` is required if data is empty
+ */
 export async function fromRecordsToStreamBuffer<T> (
   data: Array<Record<string, unknown>>,
   embeddings?: EmbeddingFunction<T>,
   schema?: Schema
 ): Promise<Buffer> {
-  let table = await convertToTable(data, embeddings)
+  if (schema !== null && schema !== undefined) {
-  if (schema !== undefined) {
+    schema = sanitizeSchema(schema)
-    table = alignTable(table, schema)
   }
+  const table = await convertToTable(data, embeddings, { schema })
   const writer = RecordBatchStreamWriter.writeAll(table)
   return Buffer.from(await writer.toUint8Array())
 }
 
-// Converts an Arrow Table into Arrow IPC format
+/**
+ * Serialize an Arrow Table into a buffer using the Arrow IPC File serialization
+ *
+ * This function will apply `embeddings` to the table in a manner similar to
+ * `convertToTable`.
+ *
+ * `schema` is required if the table is empty
+ */
 export async function fromTableToBuffer<T> (
   table: ArrowTable,
   embeddings?: EmbeddingFunction<T>,
   schema?: Schema
 ): Promise<Buffer> {
-  if (embeddings !== undefined) {
+  if (schema !== null && schema !== undefined) {
-    const source = table.getChild(embeddings.sourceColumn)
+    schema = sanitizeSchema(schema)
-
-    if (source === null) {
-      throw new Error(
-        `The embedding source column ${embeddings.sourceColumn} was not found in the Arrow Table`
-      )
-    }
-
-    const vectors = await embeddings.embed(source.toArray() as T[])
-    const column = vectorFromArray(vectors, newVectorType(vectors[0].length))
-    table = table.assign(new ArrowTable({ vector: column }))
   }
-  if (schema !== undefined) {
+  const tableWithEmbeddings = await applyEmbeddings(table, embeddings, schema)
-    table = alignTable(table, schema)
+  const writer = RecordBatchFileWriter.writeAll(tableWithEmbeddings)
-  }
-  const writer = RecordBatchFileWriter.writeAll(table)
   return Buffer.from(await writer.toUint8Array())
 }
 
-// Converts an Arrow Table into Arrow IPC stream format
+/**
+ * Serialize an Arrow Table into a buffer using the Arrow IPC Stream serialization
+ *
+ * This function will apply `embeddings` to the table in a manner similar to
+ * `convertToTable`.
+ *
+ * `schema` is required if the table is empty
+ */
 export async function fromTableToStreamBuffer<T> (
   table: ArrowTable,
   embeddings?: EmbeddingFunction<T>,
   schema?: Schema
 ): Promise<Buffer> {
-  if (embeddings !== undefined) {
+  if (schema !== null && schema !== undefined) {
-    const source = table.getChild(embeddings.sourceColumn)
+    schema = sanitizeSchema(schema)
-
-    if (source === null) {
-      throw new Error(
-        `The embedding source column ${embeddings.sourceColumn} was not found in the Arrow Table`
-      )
-    }
-
-    const vectors = await embeddings.embed(source.toArray() as T[])
-    const column = vectorFromArray(vectors, newVectorType(vectors[0].length))
-    table = table.assign(new ArrowTable({ vector: column }))
   }
-  if (schema !== undefined) {
+  const tableWithEmbeddings = await applyEmbeddings(table, embeddings, schema)
-    table = alignTable(table, schema)
+  const writer = RecordBatchStreamWriter.writeAll(tableWithEmbeddings)
-  }
-  const writer = RecordBatchStreamWriter.writeAll(table)
   return Buffer.from(await writer.toUint8Array())
 }
 
@@ -373,5 +547,5 @@ function alignTable (table: ArrowTable, schema: Schema): ArrowTable {
 
 // Creates an empty Arrow Table
 export function createEmptyTable (schema: Schema): ArrowTable {
-  return new ArrowTable(schema)
+  return new ArrowTable(sanitizeSchema(schema))
 }

node/src/embedding/embedding_function.ts

@@ -12,18 +12,53 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+import { type Float } from 'apache-arrow'
+
 /**
  * An embedding function that automatically creates vector representation for a given column.
  */
 export interface EmbeddingFunction<T> {
   /**
-     * The name of the column that will be used as input for the Embedding Function.
+   * The name of the column that will be used as input for the Embedding Function.
-     */
+   */
   sourceColumn: string
 
   /**
-     * Creates a vector representation for the given values.
+   * The data type of the embedding
-     */
+   *
+   * The embedding function should return `number`.  This will be converted into
+   * an Arrow float array.  By default this will be Float32 but this property can
+   * be used to control the conversion.
+   */
+  embeddingDataType?: Float
+
+  /**
+   * The dimension of the embedding
+   *
+   * This is optional, normally this can be determined by looking at the results of
+   * `embed`.  If this is not specified, and there is an attempt to apply the embedding
+   * to an empty table, then that process will fail.
+   */
+  embeddingDimension?: number
+
+  /**
+   * The name of the column that will contain the embedding
+   *
+   * By default this is "vector"
+   */
+  destColumn?: string
+
+  /**
+   * Should the source column be excluded from the resulting table
+   *
+   * By default the source column is included.  Set this to true and
+   * only the embedding will be stored.
+   */
+  excludeSource?: boolean
+
+  /**
+   * Creates a vector representation for the given values.
+   */
   embed: (data: T[]) => Promise<number[][]>
 }
 

node/src/index.ts

@@ -42,14 +42,17 @@ const {
   tableCompactFiles,
   tableListIndices,
   tableIndexStats,
-  tableSchema
+  tableSchema,
+  tableAddColumns,
+  tableAlterColumns,
+  tableDropColumns
   // eslint-disable-next-line @typescript-eslint/no-var-requires
 } = require('../native.js')
 
 export { Query }
 export type { EmbeddingFunction }
 export { OpenAIEmbeddingFunction } from './embedding/openai'
-export { makeArrowTable, type MakeArrowTableOptions } from './arrow'
+export { convertToTable, makeArrowTable, type MakeArrowTableOptions } from './arrow'
 
 const defaultAwsRegion = 'us-west-2'
 
@@ -96,6 +99,19 @@ export interface ConnectionOptions {
    * This is useful for local testing.
    */
   hostOverride?: string
+
+  /**
+   * (For LanceDB OSS only): The interval, in seconds, at which to check for
+   * updates to the table from other processes. If None, then consistency is not
+   * checked. For performance reasons, this is the default. For strong
+   * consistency, set this to zero seconds. Then every read will check for
+   * updates from other processes. As a compromise, you can set this to a
+   * non-zero value for eventual consistency. If more than that interval
+   * has passed since the last check, then the table will be checked for updates.
+   * Note: this consistency only applies to read operations. Write operations are
+   * always consistent.
+   */
+  readConsistencyInterval?: number
 }
 
 function getAwsArgs (opts: ConnectionOptions): any[] {
@@ -160,16 +176,21 @@ export async function connect (
     opts = { uri: arg }
   } else {
     // opts = { uri: arg.uri, awsCredentials = arg.awsCredentials }
-    opts = Object.assign(
+    const keys = Object.keys(arg)
-      {
+    if (keys.length === 1 && keys[0] === 'uri' && typeof arg.uri === 'string') {
-        uri: '',
+      opts = { uri: arg.uri }
-        awsCredentials: undefined,
+    } else {
-        awsRegion: defaultAwsRegion,
+      opts = Object.assign(
-        apiKey: undefined,
+        {
-        region: defaultAwsRegion
+          uri: '',
-      },
+          awsCredentials: undefined,
-      arg
+          awsRegion: defaultAwsRegion,
-    )
+          apiKey: undefined,
+          region: defaultAwsRegion
+        },
+        arg
+      )
+    }
   }
 
   if (opts.uri.startsWith('db://')) {
@@ -181,7 +202,8 @@ export async function connect (
     opts.awsCredentials?.accessKeyId,
     opts.awsCredentials?.secretKey,
     opts.awsCredentials?.sessionToken,
-    opts.awsRegion
+    opts.awsRegion,
+    opts.readConsistencyInterval
   )
   return new LocalConnection(db, opts)
 }
@@ -324,6 +346,7 @@ export interface Table<T = number[]> {
    *
    * @param column The column to index
    * @param replace If false, fail if an index already exists on the column
+   * it is always set to true for remote connections
    *
    * Scalar indices, like vector indices, can be used to speed up scans.  A scalar
    * index can speed up scans that contain filter expressions on the indexed column.
@@ -367,7 +390,7 @@ export interface Table<T = number[]> {
    * await table.createScalarIndex('my_col')
    * ```
    */
-  createScalarIndex: (column: string, replace: boolean) => Promise<void>
+  createScalarIndex: (column: string, replace?: boolean) => Promise<void>
 
   /**
    * Returns the number of rows in this table.
@@ -486,6 +509,59 @@ export interface Table<T = number[]> {
   filter(value: string): Query<T>
 
   schema: Promise<Schema>
+
+  // TODO: Support BatchUDF
+  /**
+   * Add new columns with defined values.
+   *
+   * @param newColumnTransforms pairs of column names and the SQL expression to use
+   *                            to calculate the value of the new column. These
+   *                            expressions will be evaluated for each row in the
+   *                            table, and can reference existing columns in the table.
+   */
+  addColumns(newColumnTransforms: Array<{ name: string, valueSql: string }>): Promise<void>
+
+  /**
+   * Alter the name or nullability of columns.
+   *
+   * @param columnAlterations One or more alterations to apply to columns.
+   */
+  alterColumns(columnAlterations: ColumnAlteration[]): Promise<void>
+
+  /**
+   * Drop one or more columns from the dataset
+   *
+   * This is a metadata-only operation and does not remove the data from the
+   * underlying storage. In order to remove the data, you must subsequently
+   * call ``compact_files`` to rewrite the data without the removed columns and
+   * then call ``cleanup_files`` to remove the old files.
+   *
+   * @param columnNames The names of the columns to drop. These can be nested
+   *                    column references (e.g. "a.b.c") or top-level column
+   *                    names (e.g. "a").
+   */
+  dropColumns(columnNames: string[]): Promise<void>
+}
+
+/**
+ * A definition of a column alteration. The alteration changes the column at
+ * `path` to have the new name `name`, to be nullable if `nullable` is true,
+ * and to have the data type `data_type`. At least one of `rename` or `nullable`
+ * must be provided.
+ */
+export interface ColumnAlteration {
+  /**
+   * The path to the column to alter. This is a dot-separated path to the column.
+   * If it is a top-level column then it is just the name of the column. If it is
+   * a nested column then it is the path to the column, e.g. "a.b.c" for a column
+   * `c` nested inside a column `b` nested inside a column `a`.
+   */
+  path: string
+  rename?: string
+  /**
+   * Set the new nullability. Note that a nullable column cannot be made non-nullable.
+   */
+  nullable?: boolean
 }
 
 export interface UpdateArgs {
@@ -844,7 +920,10 @@ export class LocalTable<T = number[]> implements Table<T> {
       })
   }
 
-  async createScalarIndex (column: string, replace: boolean): Promise<void> {
+  async createScalarIndex (column: string, replace?: boolean): Promise<void> {
+    if (replace === undefined) {
+      replace = true
+    }
     return tableCreateScalarIndex.call(this._tbl, column, replace)
   }
 
@@ -1014,6 +1093,18 @@ export class LocalTable<T = number[]> implements Table<T> {
       return false
     }
   }
+
+  async addColumns (newColumnTransforms: Array<{ name: string, valueSql: string }>): Promise<void> {
+    return tableAddColumns.call(this._tbl, newColumnTransforms)
+  }
+
+  async alterColumns (columnAlterations: ColumnAlteration[]): Promise<void> {
+    return tableAlterColumns.call(this._tbl, columnAlterations)
+  }
+
+  async dropColumns (columnNames: string[]): Promise<void> {
+    return tableDropColumns.call(this._tbl, columnNames)
+  }
 }
 
 export interface CleanupStats {

node/src/remote/index.ts

@@ -25,7 +25,8 @@ import {
   type UpdateArgs,
   type UpdateSqlArgs,
   makeArrowTable,
-  type MergeInsertArgs
+  type MergeInsertArgs,
+  type ColumnAlteration
 } from '../index'
 import { Query } from '../query'
 
@@ -396,7 +397,7 @@ export class RemoteTable<T = number[]> implements Table<T> {
     }
 
     const column = indexParams.column ?? 'vector'
-    const indexType = 'vector' // only vector index is supported for remote connections
+    const indexType = 'vector'
     const metricType = indexParams.metric_type ?? 'L2'
     const indexCacheSize = indexParams.index_cache_size ?? null
 
@@ -419,8 +420,25 @@ export class RemoteTable<T = number[]> implements Table<T> {
     }
   }
 
-  async createScalarIndex (column: string, replace: boolean): Promise<void> {
+  async createScalarIndex (column: string): Promise<void> {
-    throw new Error('Not implemented')
+    const indexType = 'scalar'
+
+    const data = {
+      column,
+      index_type: indexType,
+      replace: true
+    }
+    const res = await this._client.post(
+      `/v1/table/${this._name}/create_scalar_index/`,
+      data
+    )
+    if (res.status !== 200) {
+      throw new Error(
+        `Server Error, status: ${res.status}, ` +
+          // eslint-disable-next-line @typescript-eslint/restrict-template-expressions
+          `message: ${res.statusText}: ${res.data}`
+      )
+    }
   }
 
   async countRows (): Promise<number> {
@@ -474,4 +492,16 @@ export class RemoteTable<T = number[]> implements Table<T> {
       numUnindexedRows: results.data.num_unindexed_rows
     }
   }
+
+  async addColumns (newColumnTransforms: Array<{ name: string, valueSql: string }>): Promise<void> {
+    throw new Error('Add columns is not yet supported in LanceDB Cloud.')
+  }
+
+  async alterColumns (columnAlterations: ColumnAlteration[]): Promise<void> {
+    throw new Error('Alter columns is not yet supported in LanceDB Cloud.')
+  }
+
+  async dropColumns (columnNames: string[]): Promise<void> {
+    throw new Error('Drop columns is not yet supported in LanceDB Cloud.')
+  }
 }

node/src/sanitize.ts

@@ -0,0 +1,508 @@
+// Copyright 2023 LanceDB Developers.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// The utilities in this file help sanitize data from the user's arrow
+// library into the types expected by vectordb's arrow library.  Node
+// generally allows for mulitple versions of the same library (and sometimes
+// even multiple copies of the same version) to be installed at the same
+// time.  However, arrow-js uses instanceof which expected that the input
+// comes from the exact same library instance.  This is not always the case
+// and so we must sanitize the input to ensure that it is compatible.
+
+import {
+  Field,
+  Utf8,
+  FixedSizeBinary,
+  FixedSizeList,
+  Schema,
+  List,
+  Struct,
+  Float,
+  Bool,
+  Date_,
+  Decimal,
+  DataType,
+  Dictionary,
+  Binary,
+  Float32,
+  Interval,
+  Map_,
+  Duration,
+  Union,
+  Time,
+  Timestamp,
+  Type,
+  Null,
+  Int,
+  type Precision,
+  type DateUnit,
+  Int8,
+  Int16,
+  Int32,
+  Int64,
+  Uint8,
+  Uint16,
+  Uint32,
+  Uint64,
+  Float16,
+  Float64,
+  DateDay,
+  DateMillisecond,
+  DenseUnion,
+  SparseUnion,
+  TimeNanosecond,
+  TimeMicrosecond,
+  TimeMillisecond,
+  TimeSecond,
+  TimestampNanosecond,
+  TimestampMicrosecond,
+  TimestampMillisecond,
+  TimestampSecond,
+  IntervalDayTime,
+  IntervalYearMonth,
+  DurationNanosecond,
+  DurationMicrosecond,
+  DurationMillisecond,
+  DurationSecond,
+} from "apache-arrow";
+import type { IntBitWidth, TimeBitWidth } from "apache-arrow/type";
+
+function sanitizeMetadata(
+  metadataLike?: unknown,
+): Map<string, string> | undefined {
+  if (metadataLike === undefined || metadataLike === null) {
+    return undefined;
+  }
+  if (!(metadataLike instanceof Map)) {
+    throw Error("Expected metadata, if present, to be a Map<string, string>");
+  }
+  for (const item of metadataLike) {
+    if (!(typeof item[0] === "string" || !(typeof item[1] === "string"))) {
+      throw Error(
+        "Expected metadata, if present, to be a Map<string, string> but it had non-string keys or values",
+      );
+    }
+  }
+  return metadataLike as Map<string, string>;
+}
+
+function sanitizeInt(typeLike: object) {
+  if (
+    !("bitWidth" in typeLike) ||
+    typeof typeLike.bitWidth !== "number" ||
+    !("isSigned" in typeLike) ||
+    typeof typeLike.isSigned !== "boolean"
+  ) {
+    throw Error(
+      "Expected an Int Type to have a `bitWidth` and `isSigned` property",
+    );
+  }
+  return new Int(typeLike.isSigned, typeLike.bitWidth as IntBitWidth);
+}
+
+function sanitizeFloat(typeLike: object) {
+  if (!("precision" in typeLike) || typeof typeLike.precision !== "number") {
+    throw Error("Expected a Float Type to have a `precision` property");
+  }
+  return new Float(typeLike.precision as Precision);
+}
+
+function sanitizeDecimal(typeLike: object) {
+  if (
+    !("scale" in typeLike) ||
+    typeof typeLike.scale !== "number" ||
+    !("precision" in typeLike) ||
+    typeof typeLike.precision !== "number" ||
+    !("bitWidth" in typeLike) ||
+    typeof typeLike.bitWidth !== "number"
+  ) {
+    throw Error(
+      "Expected a Decimal Type to have `scale`, `precision`, and `bitWidth` properties",
+    );
+  }
+  return new Decimal(typeLike.scale, typeLike.precision, typeLike.bitWidth);
+}
+
+function sanitizeDate(typeLike: object) {
+  if (!("unit" in typeLike) || typeof typeLike.unit !== "number") {
+    throw Error("Expected a Date type to have a `unit` property");
+  }
+  return new Date_(typeLike.unit as DateUnit);
+}
+
+function sanitizeTime(typeLike: object) {
+  if (
+    !("unit" in typeLike) ||
+    typeof typeLike.unit !== "number" ||
+    !("bitWidth" in typeLike) ||
+    typeof typeLike.bitWidth !== "number"
+  ) {
+    throw Error(
+      "Expected a Time type to have `unit` and `bitWidth` properties",
+    );
+  }
+  return new Time(typeLike.unit, typeLike.bitWidth as TimeBitWidth);
+}
+
+function sanitizeTimestamp(typeLike: object) {
+  if (!("unit" in typeLike) || typeof typeLike.unit !== "number") {
+    throw Error("Expected a Timestamp type to have a `unit` property");
+  }
+  let timezone = null;
+  if ("timezone" in typeLike && typeof typeLike.timezone === "string") {
+    timezone = typeLike.timezone;
+  }
+  return new Timestamp(typeLike.unit, timezone);
+}
+
+function sanitizeTypedTimestamp(
+  typeLike: object,
+  Datatype:
+    | typeof TimestampNanosecond
+    | typeof TimestampMicrosecond
+    | typeof TimestampMillisecond
+    | typeof TimestampSecond,
+) {
+  let timezone = null;
+  if ("timezone" in typeLike && typeof typeLike.timezone === "string") {
+    timezone = typeLike.timezone;
+  }
+  return new Datatype(timezone);
+}
+
+function sanitizeInterval(typeLike: object) {
+  if (!("unit" in typeLike) || typeof typeLike.unit !== "number") {
+    throw Error("Expected an Interval type to have a `unit` property");
+  }
+  return new Interval(typeLike.unit);
+}
+
+function sanitizeList(typeLike: object) {
+  if (!("children" in typeLike) || !Array.isArray(typeLike.children)) {
+    throw Error(
+      "Expected a List type to have an array-like `children` property",
+    );
+  }
+  if (typeLike.children.length !== 1) {
+    throw Error("Expected a List type to have exactly one child");
+  }
+  return new List(sanitizeField(typeLike.children[0]));
+}
+
+function sanitizeStruct(typeLike: object) {
+  if (!("children" in typeLike) || !Array.isArray(typeLike.children)) {
+    throw Error(
+      "Expected a Struct type to have an array-like `children` property",
+    );
+  }
+  return new Struct(typeLike.children.map((child) => sanitizeField(child)));
+}
+
+function sanitizeUnion(typeLike: object) {
+  if (
+    !("typeIds" in typeLike) ||
+    !("mode" in typeLike) ||
+    typeof typeLike.mode !== "number"
+  ) {
+    throw Error(
+      "Expected a Union type to have `typeIds` and `mode` properties",
+    );
+  }
+  if (!("children" in typeLike) || !Array.isArray(typeLike.children)) {
+    throw Error(
+      "Expected a Union type to have an array-like `children` property",
+    );
+  }
+
+  return new Union(
+    typeLike.mode,
+    typeLike.typeIds as any,
+    typeLike.children.map((child) => sanitizeField(child)),
+  );
+}
+
+function sanitizeTypedUnion(
+  typeLike: object,
+  UnionType: typeof DenseUnion | typeof SparseUnion,
+) {
+  if (!("typeIds" in typeLike)) {
+    throw Error(
+      "Expected a DenseUnion/SparseUnion type to have a `typeIds` property",
+    );
+  }
+  if (!("children" in typeLike) || !Array.isArray(typeLike.children)) {
+    throw Error(
+      "Expected a DenseUnion/SparseUnion type to have an array-like `children` property",
+    );
+  }
+
+  return new UnionType(
+    typeLike.typeIds as any,
+    typeLike.children.map((child) => sanitizeField(child)),
+  );
+}
+
+function sanitizeFixedSizeBinary(typeLike: object) {
+  if (!("byteWidth" in typeLike) || typeof typeLike.byteWidth !== "number") {
+    throw Error(
+      "Expected a FixedSizeBinary type to have a `byteWidth` property",
+    );
+  }
+  return new FixedSizeBinary(typeLike.byteWidth);
+}
+
+function sanitizeFixedSizeList(typeLike: object) {
+  if (!("listSize" in typeLike) || typeof typeLike.listSize !== "number") {
+    throw Error("Expected a FixedSizeList type to have a `listSize` property");
+  }
+  if (!("children" in typeLike) || !Array.isArray(typeLike.children)) {
+    throw Error(
+      "Expected a FixedSizeList type to have an array-like `children` property",
+    );
+  }
+  if (typeLike.children.length !== 1) {
+    throw Error("Expected a FixedSizeList type to have exactly one child");
+  }
+  return new FixedSizeList(
+    typeLike.listSize,
+    sanitizeField(typeLike.children[0]),
+  );
+}
+
+function sanitizeMap(typeLike: object) {
+  if (!("children" in typeLike) || !Array.isArray(typeLike.children)) {
+    throw Error(
+      "Expected a Map type to have an array-like `children` property",
+    );
+  }
+  if (!("keysSorted" in typeLike) || typeof typeLike.keysSorted !== "boolean") {
+    throw Error("Expected a Map type to have a `keysSorted` property");
+  }
+  return new Map_(
+    typeLike.children.map((field) => sanitizeField(field)) as any,
+    typeLike.keysSorted,
+  );
+}
+
+function sanitizeDuration(typeLike: object) {
+  if (!("unit" in typeLike) || typeof typeLike.unit !== "number") {
+    throw Error("Expected a Duration type to have a `unit` property");
+  }
+  return new Duration(typeLike.unit);
+}
+
+function sanitizeDictionary(typeLike: object) {
+  if (!("id" in typeLike) || typeof typeLike.id !== "number") {
+    throw Error("Expected a Dictionary type to have an `id` property");
+  }
+  if (!("indices" in typeLike) || typeof typeLike.indices !== "object") {
+    throw Error("Expected a Dictionary type to have an `indices` property");
+  }
+  if (!("dictionary" in typeLike) || typeof typeLike.dictionary !== "object") {
+    throw Error("Expected a Dictionary type to have an `dictionary` property");
+  }
+  if (!("isOrdered" in typeLike) || typeof typeLike.isOrdered !== "boolean") {
+    throw Error("Expected a Dictionary type to have an `isOrdered` property");
+  }
+  return new Dictionary(
+    sanitizeType(typeLike.dictionary),
+    sanitizeType(typeLike.indices) as any,
+    typeLike.id,
+    typeLike.isOrdered,
+  );
+}
+
+function sanitizeType(typeLike: unknown): DataType<any> {
+  if (typeof typeLike !== "object" || typeLike === null) {
+    throw Error("Expected a Type but object was null/undefined");
+  }
+  if (!("typeId" in typeLike) || !(typeof typeLike.typeId !== "function")) {
+    throw Error("Expected a Type to have a typeId function");
+  }
+  let typeId: Type;
+  if (typeof typeLike.typeId === "function") {
+    typeId = (typeLike.typeId as () => unknown)() as Type;
+  } else if (typeof typeLike.typeId === "number") {
+    typeId = typeLike.typeId as Type;
+  } else {
+    throw Error("Type's typeId property was not a function or number");
+  }
+
+  switch (typeId) {
+    case Type.NONE:
+      throw Error("Received a Type with a typeId of NONE");
+    case Type.Null:
+      return new Null();
+    case Type.Int:
+      return sanitizeInt(typeLike);
+    case Type.Float:
+      return sanitizeFloat(typeLike);
+    case Type.Binary:
+      return new Binary();
+    case Type.Utf8:
+      return new Utf8();
+    case Type.Bool:
+      return new Bool();
+    case Type.Decimal:
+      return sanitizeDecimal(typeLike);
+    case Type.Date:
+      return sanitizeDate(typeLike);
+    case Type.Time:
+      return sanitizeTime(typeLike);
+    case Type.Timestamp:
+      return sanitizeTimestamp(typeLike);
+    case Type.Interval:
+      return sanitizeInterval(typeLike);
+    case Type.List:
+      return sanitizeList(typeLike);
+    case Type.Struct:
+      return sanitizeStruct(typeLike);
+    case Type.Union:
+      return sanitizeUnion(typeLike);
+    case Type.FixedSizeBinary:
+      return sanitizeFixedSizeBinary(typeLike);
+    case Type.FixedSizeList:
+      return sanitizeFixedSizeList(typeLike);
+    case Type.Map:
+      return sanitizeMap(typeLike);
+    case Type.Duration:
+      return sanitizeDuration(typeLike);
+    case Type.Dictionary:
+      return sanitizeDictionary(typeLike);
+    case Type.Int8:
+      return new Int8();
+    case Type.Int16:
+      return new Int16();
+    case Type.Int32:
+      return new Int32();
+    case Type.Int64:
+      return new Int64();
+    case Type.Uint8:
+      return new Uint8();
+    case Type.Uint16:
+      return new Uint16();
+    case Type.Uint32:
+      return new Uint32();
+    case Type.Uint64:
+      return new Uint64();
+    case Type.Float16:
+      return new Float16();
+    case Type.Float32:
+      return new Float32();
+    case Type.Float64:
+      return new Float64();
+    case Type.DateMillisecond:
+      return new DateMillisecond();
+    case Type.DateDay:
+      return new DateDay();
+    case Type.TimeNanosecond:
+      return new TimeNanosecond();
+    case Type.TimeMicrosecond:
+      return new TimeMicrosecond();
+    case Type.TimeMillisecond:
+      return new TimeMillisecond();
+    case Type.TimeSecond:
+      return new TimeSecond();
+    case Type.TimestampNanosecond:
+      return sanitizeTypedTimestamp(typeLike, TimestampNanosecond);
+    case Type.TimestampMicrosecond:
+      return sanitizeTypedTimestamp(typeLike, TimestampMicrosecond);
+    case Type.TimestampMillisecond:
+      return sanitizeTypedTimestamp(typeLike, TimestampMillisecond);
+    case Type.TimestampSecond:
+      return sanitizeTypedTimestamp(typeLike, TimestampSecond);
+    case Type.DenseUnion:
+      return sanitizeTypedUnion(typeLike, DenseUnion);
+    case Type.SparseUnion:
+      return sanitizeTypedUnion(typeLike, SparseUnion);
+    case Type.IntervalDayTime:
+      return new IntervalDayTime();
+    case Type.IntervalYearMonth:
+      return new IntervalYearMonth();
+    case Type.DurationNanosecond:
+      return new DurationNanosecond();
+    case Type.DurationMicrosecond:
+      return new DurationMicrosecond();
+    case Type.DurationMillisecond:
+      return new DurationMillisecond();
+    case Type.DurationSecond:
+      return new DurationSecond();
+  }
+}
+
+function sanitizeField(fieldLike: unknown): Field {
+  if (fieldLike instanceof Field) {
+    return fieldLike;
+  }
+  if (typeof fieldLike !== "object" || fieldLike === null) {
+    throw Error("Expected a Field but object was null/undefined");
+  }
+  if (
+    !("type" in fieldLike) ||
+    !("name" in fieldLike) ||
+    !("nullable" in fieldLike)
+  ) {
+    throw Error(
+      "The field passed in is missing a `type`/`name`/`nullable` property",
+    );
+  }
+  const type = sanitizeType(fieldLike.type);
+  const name = fieldLike.name;
+  if (!(typeof name === "string")) {
+    throw Error("The field passed in had a non-string `name` property");
+  }
+  const nullable = fieldLike.nullable;
+  if (!(typeof nullable === "boolean")) {
+    throw Error("The field passed in had a non-boolean `nullable` property");
+  }
+  let metadata;
+  if ("metadata" in fieldLike) {
+    metadata = sanitizeMetadata(fieldLike.metadata);
+  }
+  return new Field(name, type, nullable, metadata);
+}
+
+/**
+ * Convert something schemaLike into a Schema instance
+ *
+ * This method is often needed even when the caller is using a Schema
+ * instance because they might be using a different instance of apache-arrow
+ * than lancedb is using.
+ */
+export function sanitizeSchema(schemaLike: unknown): Schema {
+  if (schemaLike instanceof Schema) {
+    return schemaLike;
+  }
+  if (typeof schemaLike !== "object" || schemaLike === null) {
+    throw Error("Expected a Schema but object was null/undefined");
+  }
+  if (!("fields" in schemaLike)) {
+    throw Error(
+      "The schema passed in does not appear to be a schema (no 'fields' property)",
+    );
+  }
+  let metadata;
+  if ("metadata" in schemaLike) {
+    metadata = sanitizeMetadata(schemaLike.metadata);
+  }
+  if (!Array.isArray(schemaLike.fields)) {
+    throw Error(
+      "The schema passed in had a 'fields' property but it was not an array",
+    );
+  }
+  const sanitizedFields = schemaLike.fields.map((field) =>
+    sanitizeField(field),
+  );
+  return new Schema(sanitizedFields, metadata);
+}

node/src/test/arrow.test.ts

@@ -13,9 +13,10 @@
 // limitations under the License.
 
 import { describe } from 'mocha'
-import { assert } from 'chai'
+import { assert, expect, use as chaiUse } from 'chai'
+import * as chaiAsPromised from 'chai-as-promised'
 
-import { fromTableToBuffer, makeArrowTable } from '../arrow'
+import { convertToTable, fromTableToBuffer, makeArrowTable, makeEmptyTable } from '../arrow'
 import {
   Field,
   FixedSizeList,
@@ -24,21 +25,91 @@ import {
   Int32,
   tableFromIPC,
   Schema,
-  Float64
+  Float64,
+  type Table,
+  Binary,
+  Bool,
+  Utf8,
+  Struct,
+  List,
+  DataType,
+  Dictionary,
+  Int64,
+  MetadataVersion
 } from 'apache-arrow'
+import {
+  Dictionary as OldDictionary,
+  Field as OldField,
+  FixedSizeList as OldFixedSizeList,
+  Float32 as OldFloat32,
+  Int32 as OldInt32,
+  Struct as OldStruct,
+  Schema as OldSchema,
+  TimestampNanosecond as OldTimestampNanosecond,
+  Utf8 as OldUtf8
+} from 'apache-arrow-old'
+import { type EmbeddingFunction } from '../embedding/embedding_function'
 
-describe('Apache Arrow tables', function () {
+chaiUse(chaiAsPromised)
-  it('customized schema', async function () {
+
+function sampleRecords (): Array<Record<string, any>> {
+  return [
+    {
+      binary: Buffer.alloc(5),
+      boolean: false,
+      number: 7,
+      string: 'hello',
+      struct: { x: 0, y: 0 },
+      list: ['anime', 'action', 'comedy']
+    }
+  ]
+}
+
+// Helper method to verify various ways to create a table
+async function checkTableCreation (tableCreationMethod: (records: any, recordsReversed: any, schema: Schema) => Promise<Table>): Promise<void> {
+  const records = sampleRecords()
+  const recordsReversed = [{
+    list: ['anime', 'action', 'comedy'],
+    struct: { x: 0, y: 0 },
+    string: 'hello',
+    number: 7,
+    boolean: false,
+    binary: Buffer.alloc(5)
+  }]
+  const schema = new Schema([
+    new Field('binary', new Binary(), false),
+    new Field('boolean', new Bool(), false),
+    new Field('number', new Float64(), false),
+    new Field('string', new Utf8(), false),
+    new Field('struct', new Struct([
+      new Field('x', new Float64(), false),
+      new Field('y', new Float64(), false)
+    ])),
+    new Field('list', new List(new Field('item', new Utf8(), false)), false)
+  ])
+
+  const table = await tableCreationMethod(records, recordsReversed, schema)
+  schema.fields.forEach((field, idx) => {
+    const actualField = table.schema.fields[idx]
+    assert.isFalse(actualField.nullable)
+    assert.equal(table.getChild(field.name)?.type.toString(), field.type.toString())
+    assert.equal(table.getChildAt(idx)?.type.toString(), field.type.toString())
+  })
+}
+
+describe('The function makeArrowTable', function () {
+  it('will use data types from a provided schema instead of inference', async function () {
     const schema = new Schema([
       new Field('a', new Int32()),
       new Field('b', new Float32()),
-      new Field('c', new FixedSizeList(3, new Field('item', new Float16())))
+      new Field('c', new FixedSizeList(3, new Field('item', new Float16()))),
+      new Field('d', new Int64())
     ])
     const table = makeArrowTable(
       [
-        { a: 1, b: 2, c: [1, 2, 3] },
+        { a: 1, b: 2, c: [1, 2, 3], d: 9 },
-        { a: 4, b: 5, c: [4, 5, 6] },
+        { a: 4, b: 5, c: [4, 5, 6], d: 10 },
-        { a: 7, b: 8, c: [7, 8, 9] }
+        { a: 7, b: 8, c: [7, 8, 9], d: null }
       ],
       { schema }
     )
@@ -52,13 +123,13 @@ describe('Apache Arrow tables', function () {
     assert.deepEqual(actualSchema, schema)
   })
 
-  it('default vector column', async function () {
+  it('will assume the column `vector` is FixedSizeList<Float32> by default', async function () {
     const schema = new Schema([
       new Field('a', new Float64()),
       new Field('b', new Float64()),
       new Field(
         'vector',
-        new FixedSizeList(3, new Field('item', new Float32()))
+        new FixedSizeList(3, new Field('item', new Float32(), true))
       )
     ])
     const table = makeArrowTable([
@@ -76,12 +147,12 @@ describe('Apache Arrow tables', function () {
     assert.deepEqual(actualSchema, schema)
   })
 
-  it('2 vector columns', async function () {
+  it('can support multiple vector columns', async function () {
     const schema = new Schema([
       new Field('a', new Float64()),
       new Field('b', new Float64()),
-      new Field('vec1', new FixedSizeList(3, new Field('item', new Float16()))),
+      new Field('vec1', new FixedSizeList(3, new Field('item', new Float16(), true))),
-      new Field('vec2', new FixedSizeList(3, new Field('item', new Float16())))
+      new Field('vec2', new FixedSizeList(3, new Field('item', new Float16(), true)))
     ])
     const table = makeArrowTable(
       [
@@ -105,4 +176,185 @@ describe('Apache Arrow tables', function () {
     const actualSchema = actual.schema
     assert.deepEqual(actualSchema, schema)
   })
+
+  it('will allow different vector column types', async function () {
+    const table = makeArrowTable(
+      [
+        { fp16: [1], fp32: [1], fp64: [1] }
+      ],
+      {
+        vectorColumns: {
+          fp16: { type: new Float16() },
+          fp32: { type: new Float32() },
+          fp64: { type: new Float64() }
+        }
+      }
+    )
+
+    assert.equal(table.getChild('fp16')?.type.children[0].type.toString(), new Float16().toString())
+    assert.equal(table.getChild('fp32')?.type.children[0].type.toString(), new Float32().toString())
+    assert.equal(table.getChild('fp64')?.type.children[0].type.toString(), new Float64().toString())
+  })
+
+  it('will use dictionary encoded strings if asked', async function () {
+    const table = makeArrowTable([{ str: 'hello' }])
+    assert.isTrue(DataType.isUtf8(table.getChild('str')?.type))
+
+    const tableWithDict = makeArrowTable([{ str: 'hello' }], { dictionaryEncodeStrings: true })
+    assert.isTrue(DataType.isDictionary(tableWithDict.getChild('str')?.type))
+
+    const schema = new Schema([
+      new Field('str', new Dictionary(new Utf8(), new Int32()))
+    ])
+
+    const tableWithDict2 = makeArrowTable([{ str: 'hello' }], { schema })
+    assert.isTrue(DataType.isDictionary(tableWithDict2.getChild('str')?.type))
+  })
+
+  it('will infer data types correctly', async function () {
+    await checkTableCreation(async (records) => makeArrowTable(records))
+  })
+
+  it('will allow a schema to be provided', async function () {
+    await checkTableCreation(async (records, _, schema) => makeArrowTable(records, { schema }))
+  })
+
+  it('will use the field order of any provided schema', async function () {
+    await checkTableCreation(async (_, recordsReversed, schema) => makeArrowTable(recordsReversed, { schema }))
+  })
+
+  it('will make an empty table', async function () {
+    await checkTableCreation(async (_, __, schema) => makeArrowTable([], { schema }))
+  })
+})
+
+class DummyEmbedding implements EmbeddingFunction<string> {
+  public readonly sourceColumn = 'string'
+  public readonly embeddingDimension = 2
+  public readonly embeddingDataType = new Float16()
+
+  async embed (data: string[]): Promise<number[][]> {
+    return data.map(
+      () => [0.0, 0.0]
+    )
+  }
+}
+
+class DummyEmbeddingWithNoDimension implements EmbeddingFunction<string> {
+  public readonly sourceColumn = 'string'
+
+  async embed (data: string[]): Promise<number[][]> {
+    return data.map(
+      () => [0.0, 0.0]
+    )
+  }
+}
+
+describe('convertToTable', function () {
+  it('will infer data types correctly', async function () {
+    await checkTableCreation(async (records) => await convertToTable(records))
+  })
+
+  it('will allow a schema to be provided', async function () {
+    await checkTableCreation(async (records, _, schema) => await convertToTable(records, undefined, { schema }))
+  })
+
+  it('will use the field order of any provided schema', async function () {
+    await checkTableCreation(async (_, recordsReversed, schema) => await convertToTable(recordsReversed, undefined, { schema }))
+  })
+
+  it('will make an empty table', async function () {
+    await checkTableCreation(async (_, __, schema) => await convertToTable([], undefined, { schema }))
+  })
+
+  it('will apply embeddings', async function () {
+    const records = sampleRecords()
+    const table = await convertToTable(records, new DummyEmbedding())
+    assert.isTrue(DataType.isFixedSizeList(table.getChild('vector')?.type))
+    assert.equal(table.getChild('vector')?.type.children[0].type.toString(), new Float16().toString())
+  })
+
+  it('will fail if missing the embedding source column', async function () {
+    return await expect(convertToTable([{ id: 1 }], new DummyEmbedding())).to.be.rejectedWith("'string' was not present")
+  })
+
+  it('use embeddingDimension if embedding missing from table', async function () {
+    const schema = new Schema([
+      new Field('string', new Utf8(), false)
+    ])
+    // Simulate getting an empty Arrow table (minus embedding) from some other source
+    // In other words, we aren't starting with records
+    const table = makeEmptyTable(schema)
+
+    // If the embedding specifies the dimension we are fine
+    await fromTableToBuffer(table, new DummyEmbedding())
+
+    // We can also supply a schema and should be ok
+    const schemaWithEmbedding = new Schema([
+      new Field('string', new Utf8(), false),
+      new Field('vector', new FixedSizeList(2, new Field('item', new Float16(), false)), false)
+    ])
+    await fromTableToBuffer(table, new DummyEmbeddingWithNoDimension(), schemaWithEmbedding)
+
+    // Otherwise we will get an error
+    return await expect(fromTableToBuffer(table, new DummyEmbeddingWithNoDimension())).to.be.rejectedWith('does not specify `embeddingDimension`')
+  })
+
+  it('will apply embeddings to an empty table', async function () {
+    const schema = new Schema([
+      new Field('string', new Utf8(), false),
+      new Field('vector', new FixedSizeList(2, new Field('item', new Float16(), false)), false)
+    ])
+    const table = await convertToTable([], new DummyEmbedding(), { schema })
+    assert.isTrue(DataType.isFixedSizeList(table.getChild('vector')?.type))
+    assert.equal(table.getChild('vector')?.type.children[0].type.toString(), new Float16().toString())
+  })
+
+  it('will complain if embeddings present but schema missing embedding column', async function () {
+    const schema = new Schema([
+      new Field('string', new Utf8(), false)
+    ])
+    return await expect(convertToTable([], new DummyEmbedding(), { schema })).to.be.rejectedWith('column vector was missing')
+  })
+
+  it('will provide a nice error if run twice', async function () {
+    const records = sampleRecords()
+    const table = await convertToTable(records, new DummyEmbedding())
+    // fromTableToBuffer will try and apply the embeddings again
+    return await expect(fromTableToBuffer(table, new DummyEmbedding())).to.be.rejectedWith('already existed')
+  })
+})
+
+describe('makeEmptyTable', function () {
+  it('will make an empty table', async function () {
+    await checkTableCreation(async (_, __, schema) => makeEmptyTable(schema))
+  })
+})
+
+describe('when using two versions of arrow', function () {
+  it('can still import data', async function() {
+    const schema = new OldSchema([
+      new OldField('id', new OldInt32()),
+      new OldField('vector', new OldFixedSizeList(1024, new OldField("item", new OldFloat32(), true))),
+      new OldField('struct', new OldStruct([
+        new OldField('nested', new OldDictionary(new OldUtf8(), new OldInt32(), 1, true)),
+        new OldField('ts_with_tz', new OldTimestampNanosecond("some_tz")),
+        new OldField('ts_no_tz', new OldTimestampNanosecond(null))
+      ]))
+    ]) as any
+    // We use arrow version 13 to emulate a "foreign arrow" and this version doesn't have metadataVersion
+    // In theory, this wouldn't matter.  We don't rely on that property.  However, it causes deepEqual to
+    // fail so we patch it back in
+    schema.metadataVersion = MetadataVersion.V5
+    const table = makeArrowTable(
+      [],
+      { schema }
+    )
+
+    const buf = await fromTableToBuffer(table)
+    assert.isAbove(buf.byteLength, 0)
+    const actual = tableFromIPC(buf)
+    const actualSchema = actual.schema
+    assert.deepEqual(actualSchema, schema)
+  })
 })

node/src/test/test.ts

@@ -37,8 +37,10 @@ import {
   Utf8,
   Table as ArrowTable,
   vectorFromArray,
+  Float64,
   Float32,
-  Float16
+  Float16,
+  Int64
 } from 'apache-arrow'
 
 const expect = chai.expect
@@ -126,6 +128,11 @@ describe('LanceDB client', function () {
       assertResults(results)
       results = await table.where('id % 2 = 0').execute()
       assertResults(results)
+
+      // Should reject a bad filter
+      await expect(table.filter('id % 2 = 0 AND').execute()).to.be.rejectedWith(
+        /.*sql parser error: Expected an expression:, found: EOF.*/
+      )
     })
 
     it('uses a filter / where clause', async function () {
@@ -196,7 +203,7 @@ describe('LanceDB client', function () {
       const table = await con.openTable('vectors')
       const results = await table
         .search([0.1, 0.1])
-        .select(['is_active'])
+        .select(['is_active', 'vector'])
         .execute()
       assert.equal(results.length, 2)
       // vector and _distance are always returned
@@ -281,7 +288,8 @@ describe('LanceDB client', function () {
 
     it('create a table from an Arrow Table', async function () {
       const dir = await track().mkdir('lancejs')
-      const con = await lancedb.connect(dir)
+      // Also test the connect function with an object
+      const con = await lancedb.connect({ uri: dir })
 
       const i32s = new Int32Array(new Array<number>(10))
       const i32 = makeVector(i32s)
@@ -743,11 +751,11 @@ describe('LanceDB client', function () {
         num_sub_vectors: 2
       })
       await expect(createIndex).to.be.rejectedWith(
-        /VectorIndex requires the column data type to be fixed size list of float32s/
+        "index cannot be created on the column `name` which has data type Utf8"
       )
     })
 
-    it('it should fail when the column is not a vector', async function () {
+    it('it should fail when num_partitions is invalid', async function () {
       const uri = await createTestDB(32, 300)
       const con = await lancedb.connect(uri)
       const table = await con.openTable('vectors')
@@ -1057,3 +1065,63 @@ describe('Compact and cleanup', function () {
     assert.equal(await table.countRows(), 3)
   })
 })
+
+describe('schema evolution', function () {
+  // Create a new sample table
+  it('can add a new column to the schema', async function () {
+    const dir = await track().mkdir('lancejs')
+    const con = await lancedb.connect(dir)
+    const table = await con.createTable('vectors', [
+      { id: 1n, vector: [0.1, 0.2] }
+    ])
+
+    await table.addColumns([{ name: 'price', valueSql: 'cast(10.0 as float)' }])
+
+    const expectedSchema = new Schema([
+      new Field('id', new Int64()),
+      new Field('vector', new FixedSizeList(2, new Field('item', new Float32(), true))),
+      new Field('price', new Float32())
+    ])
+    expect(await table.schema).to.deep.equal(expectedSchema)
+  })
+
+  it('can alter the columns in the schema', async function () {
+    const dir = await track().mkdir('lancejs')
+    const con = await lancedb.connect(dir)
+    const schema = new Schema([
+      new Field('id', new Int64(), false),
+      new Field('vector', new FixedSizeList(2, new Field('item', new Float32(), true))),
+      new Field('price', new Float64(), false)
+    ])
+    const table = await con.createTable('vectors', [
+      { id: 1n, vector: [0.1, 0.2], price: 10.0 }
+    ])
+    expect(await table.schema).to.deep.equal(schema)
+
+    await table.alterColumns([
+      { path: 'id', rename: 'new_id' },
+      { path: 'price', nullable: true }
+    ])
+
+    const expectedSchema = new Schema([
+      new Field('new_id', new Int64(), false),
+      new Field('vector', new FixedSizeList(2, new Field('item', new Float32(), true))),
+      new Field('price', new Float64(), true)
+    ])
+    expect(await table.schema).to.deep.equal(expectedSchema)
+  })
+
+  it('can drop a column from the schema', async function () {
+    const dir = await track().mkdir('lancejs')
+    const con = await lancedb.connect(dir)
+    const table = await con.createTable('vectors', [
+      { id: 1n, vector: [0.1, 0.2] }
+    ])
+    await table.dropColumns(['vector'])
+
+    const expectedSchema = new Schema([
+      new Field('id', new Int64(), false)
+    ])
+    expect(await table.schema).to.deep.equal(expectedSchema)
+  })
+})

node/tsconfig.json

@@ -9,6 +9,6 @@
     "declaration": true,
     "outDir": "./dist",
     "strict": true,
-    // "esModuleInterop": true,
+    "sourceMap": true,
   }
 }

nodejs/.eslintignore

@@ -0,0 +1,3 @@
+**/dist/**/*
+**/native.js
+**/native.d.ts

nodejs/.eslintrc.js

@@ -1,22 +0,0 @@
-module.exports = {
-  env: {
-    browser: true,
-    es2021: true,
-  },
-  extends: [
-    "eslint:recommended",
-    "plugin:@typescript-eslint/recommended-type-checked",
-    "plugin:@typescript-eslint/stylistic-type-checked",
-  ],
-  overrides: [],
-  parserOptions: {
-    project: "./tsconfig.json",
-    ecmaVersion: "latest",
-    sourceType: "module",
-  },
-  rules: {
-    "@typescript-eslint/method-signature-style": "off",
-    "@typescript-eslint/no-explicit-any": "off",
-  },
-  ignorePatterns: ["node_modules/", "dist/", "build/", "vectordb/native.*"],
-};

nodejs/.prettierignore

@@ -0,0 +1 @@
+.eslintignore

nodejs/Cargo.toml

@@ -1,5 +1,5 @@
 [package]
-name = "vectordb-nodejs"
+name = "lancedb-nodejs"
 edition.workspace = true
 version = "0.0.0"
 license.workspace = true
@@ -14,14 +14,15 @@ crate-type = ["cdylib"]
 [dependencies]
 arrow-ipc.workspace = true
 futures.workspace = true
-lance-linalg.workspace = true
+lancedb = { path = "../rust/lancedb" }
-lance.workspace = true
-vectordb = { path = "../rust/vectordb" }
 napi = { version = "2.15", default-features = false, features = [
     "napi7",
-    "async"
+    "async",
 ] }
 napi-derive = "2"
 
+# Prevent dynamic linking of lzma, which comes from datafusion
+lzma-sys = { version = "*", features = ["static"] }
+
 [build-dependencies]
 napi-build = "2.1"

nodejs/README.md

@@ -2,7 +2,6 @@
 
 It will replace the NodeJS SDK when it is ready.
 
-
 ## Development
 
 ```sh
@@ -10,9 +9,35 @@ npm run build
 npm t
 ```
 
-Generating docs
+### Running lint / format
 
+LanceDb uses eslint for linting. VSCode does not need any plugins to use eslint. However, it
+may need some additional configuration. Make sure that eslint.experimental.useFlatConfig is
+set to true. Also, if your vscode root folder is the repo root then you will need to set
+the eslint.workingDirectories to ["nodejs"]. To manually lint your code you can run:
+
+```sh
+npm run lint
 ```
+
+LanceDb uses prettier for formatting. If you are using VSCode you will need to install the
+"Prettier - Code formatter" extension. You should then configure it to be the default formatter
+for typescript and you should enable format on save. To manually check your code's format you
+can run:
+
+```sh
+npm run chkformat
+```
+
+If you need to manually format your code you can run:
+
+```sh
+npx prettier --write .
+```
+
+### Generating docs
+
+```sh
 npm run docs
 
 cd ../docs

nodejs/__test__/arrow.test.ts

@@ -12,7 +12,12 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-import { makeArrowTable, toBuffer } from "../vectordb/arrow";
+import {
+  convertToTable,
+  fromTableToBuffer,
+  makeArrowTable,
+  makeEmptyTable,
+} from "../dist/arrow";
 import {
   Field,
   FixedSizeList,
@@ -22,85 +27,444 @@ import {
   tableFromIPC,
   Schema,
   Float64,
+  type Table,
+  Binary,
+  Bool,
+  Utf8,
+  Struct,
+  List,
+  DataType,
+  Dictionary,
+  Int64,
+  Float,
+  Precision,
+  MetadataVersion,
 } from "apache-arrow";
+import {
+  Dictionary as OldDictionary,
+  Field as OldField,
+  FixedSizeList as OldFixedSizeList,
+  Float32 as OldFloat32,
+  Int32 as OldInt32,
+  Struct as OldStruct,
+  Schema as OldSchema,
+  TimestampNanosecond as OldTimestampNanosecond,
+  Utf8 as OldUtf8,
+} from "apache-arrow-old";
+import { type EmbeddingFunction } from "../dist/embedding/embedding_function";
 
-test("customized schema", function () {
+// eslint-disable-next-line @typescript-eslint/no-explicit-any
-  const schema = new Schema([
+function sampleRecords(): Array<Record<string, any>> {
-    new Field("a", new Int32(), true),
+  return [
-    new Field("b", new Float32(), true),
-    new Field(
-      "c",
-      new FixedSizeList(3, new Field("item", new Float16())),
-      true
-    ),
-  ]);
-  const table = makeArrowTable(
-    [
-      { a: 1, b: 2, c: [1, 2, 3] },
-      { a: 4, b: 5, c: [4, 5, 6] },
-      { a: 7, b: 8, c: [7, 8, 9] },
-    ],
-    { schema }
-  );
-
-  expect(table.schema.toString()).toEqual(schema.toString());
-
-  const buf = toBuffer(table);
-  expect(buf.byteLength).toBeGreaterThan(0);
-
-  const actual = tableFromIPC(buf);
-  expect(actual.numRows).toBe(3);
-  const actualSchema = actual.schema;
-  expect(actualSchema.toString()).toStrictEqual(schema.toString());
-});
-
-test("default vector column", function () {
-  const schema = new Schema([
-    new Field("a", new Float64(), true),
-    new Field("b", new Float64(), true),
-    new Field("vector", new FixedSizeList(3, new Field("item", new Float32()))),
-  ]);
-  const table = makeArrowTable([
-    { a: 1, b: 2, vector: [1, 2, 3] },
-    { a: 4, b: 5, vector: [4, 5, 6] },
-    { a: 7, b: 8, vector: [7, 8, 9] },
-  ]);
-
-  const buf = toBuffer(table);
-  expect(buf.byteLength).toBeGreaterThan(0);
-
-  const actual = tableFromIPC(buf);
-  expect(actual.numRows).toBe(3);
-  const actualSchema = actual.schema;
-  expect(actualSchema.toString()).toEqual(actualSchema.toString());
-});
-
-test("2 vector columns", function () {
-  const schema = new Schema([
-    new Field("a", new Float64()),
-    new Field("b", new Float64()),
-    new Field("vec1", new FixedSizeList(3, new Field("item", new Float16()))),
-    new Field("vec2", new FixedSizeList(3, new Field("item", new Float16()))),
-  ]);
-  const table = makeArrowTable(
-    [
-      { a: 1, b: 2, vec1: [1, 2, 3], vec2: [2, 4, 6] },
-      { a: 4, b: 5, vec1: [4, 5, 6], vec2: [8, 10, 12] },
-      { a: 7, b: 8, vec1: [7, 8, 9], vec2: [14, 16, 18] },
-    ],
     {
-      vectorColumns: {
+      binary: Buffer.alloc(5),
-        vec1: { type: new Float16() },
+      boolean: false,
-        vec2: { type: new Float16() },
+      number: 7,
-      },
+      string: "hello",
+      struct: { x: 0, y: 0 },
+      list: ["anime", "action", "comedy"],
+    },
+  ];
+}
+
+// Helper method to verify various ways to create a table
+async function checkTableCreation(
+  tableCreationMethod: (
+    records: Record<string, unknown>[],
+    recordsReversed: Record<string, unknown>[],
+    schema: Schema,
+  ) => Promise<Table>,
+  infersTypes: boolean,
+): Promise<void> {
+  const records = sampleRecords();
+  const recordsReversed = [
+    {
+      list: ["anime", "action", "comedy"],
+      struct: { x: 0, y: 0 },
+      string: "hello",
+      number: 7,
+      boolean: false,
+      binary: Buffer.alloc(5),
+    },
+  ];
+  const schema = new Schema([
+    new Field("binary", new Binary(), false),
+    new Field("boolean", new Bool(), false),
+    new Field("number", new Float64(), false),
+    new Field("string", new Utf8(), false),
+    new Field(
+      "struct",
+      new Struct([
+        new Field("x", new Float64(), false),
+        new Field("y", new Float64(), false),
+      ]),
+    ),
+    new Field("list", new List(new Field("item", new Utf8(), false)), false),
+  ]);
+
+  const table = await tableCreationMethod(records, recordsReversed, schema);
+  schema.fields.forEach((field, idx) => {
+    const actualField = table.schema.fields[idx];
+    // Type inference always assumes nullable=true
+    if (infersTypes) {
+      expect(actualField.nullable).toBe(true);
+    } else {
+      expect(actualField.nullable).toBe(false);
     }
-  );
+    expect(table.getChild(field.name)?.type.toString()).toEqual(
+      field.type.toString(),
+    );
+    expect(table.getChildAt(idx)?.type.toString()).toEqual(
+      field.type.toString(),
+    );
+  });
+}
 
-  const buf = toBuffer(table);
+describe("The function makeArrowTable", function () {
-  expect(buf.byteLength).toBeGreaterThan(0);
+  it("will use data types from a provided schema instead of inference", async function () {
+    const schema = new Schema([
+      new Field("a", new Int32()),
+      new Field("b", new Float32()),
+      new Field("c", new FixedSizeList(3, new Field("item", new Float16()))),
+      new Field("d", new Int64()),
+    ]);
+    const table = makeArrowTable(
+      [
+        { a: 1, b: 2, c: [1, 2, 3], d: 9 },
+        { a: 4, b: 5, c: [4, 5, 6], d: 10 },
+        { a: 7, b: 8, c: [7, 8, 9], d: null },
+      ],
+      { schema },
+    );
 
-  const actual = tableFromIPC(buf);
+    const buf = await fromTableToBuffer(table);
-  expect(actual.numRows).toBe(3);
+    expect(buf.byteLength).toBeGreaterThan(0);
-  const actualSchema = actual.schema;
+
-  expect(actualSchema.toString()).toEqual(schema.toString());
+    const actual = tableFromIPC(buf);
+    expect(actual.numRows).toBe(3);
+    const actualSchema = actual.schema;
+    expect(actualSchema).toEqual(schema);
+  });
+
+  it("will assume the column `vector` is FixedSizeList<Float32> by default", async function () {
+    const schema = new Schema([
+      new Field("a", new Float(Precision.DOUBLE), true),
+      new Field("b", new Float(Precision.DOUBLE), true),
+      new Field(
+        "vector",
+        new FixedSizeList(
+          3,
+          new Field("item", new Float(Precision.SINGLE), true),
+        ),
+        true,
+      ),
+    ]);
+    const table = makeArrowTable([
+      { a: 1, b: 2, vector: [1, 2, 3] },
+      { a: 4, b: 5, vector: [4, 5, 6] },
+      { a: 7, b: 8, vector: [7, 8, 9] },
+    ]);
+
+    const buf = await fromTableToBuffer(table);
+    expect(buf.byteLength).toBeGreaterThan(0);
+
+    const actual = tableFromIPC(buf);
+    expect(actual.numRows).toBe(3);
+    const actualSchema = actual.schema;
+    expect(actualSchema).toEqual(schema);
+  });
+
+  it("can support multiple vector columns", async function () {
+    const schema = new Schema([
+      new Field("a", new Float(Precision.DOUBLE), true),
+      new Field("b", new Float(Precision.DOUBLE), true),
+      new Field(
+        "vec1",
+        new FixedSizeList(3, new Field("item", new Float16(), true)),
+        true,
+      ),
+      new Field(
+        "vec2",
+        new FixedSizeList(3, new Field("item", new Float16(), true)),
+        true,
+      ),
+    ]);
+    const table = makeArrowTable(
+      [
+        { a: 1, b: 2, vec1: [1, 2, 3], vec2: [2, 4, 6] },
+        { a: 4, b: 5, vec1: [4, 5, 6], vec2: [8, 10, 12] },
+        { a: 7, b: 8, vec1: [7, 8, 9], vec2: [14, 16, 18] },
+      ],
+      {
+        vectorColumns: {
+          vec1: { type: new Float16() },
+          vec2: { type: new Float16() },
+        },
+      },
+    );
+
+    const buf = await fromTableToBuffer(table);
+    expect(buf.byteLength).toBeGreaterThan(0);
+
+    const actual = tableFromIPC(buf);
+    expect(actual.numRows).toBe(3);
+    const actualSchema = actual.schema;
+    expect(actualSchema).toEqual(schema);
+  });
+
+  it("will allow different vector column types", async function () {
+    const table = makeArrowTable([{ fp16: [1], fp32: [1], fp64: [1] }], {
+      vectorColumns: {
+        fp16: { type: new Float16() },
+        fp32: { type: new Float32() },
+        fp64: { type: new Float64() },
+      },
+    });
+
+    expect(table.getChild("fp16")?.type.children[0].type.toString()).toEqual(
+      new Float16().toString(),
+    );
+    expect(table.getChild("fp32")?.type.children[0].type.toString()).toEqual(
+      new Float32().toString(),
+    );
+    expect(table.getChild("fp64")?.type.children[0].type.toString()).toEqual(
+      new Float64().toString(),
+    );
+  });
+
+  it("will use dictionary encoded strings if asked", async function () {
+    const table = makeArrowTable([{ str: "hello" }]);
+    expect(DataType.isUtf8(table.getChild("str")?.type)).toBe(true);
+
+    const tableWithDict = makeArrowTable([{ str: "hello" }], {
+      dictionaryEncodeStrings: true,
+    });
+    expect(DataType.isDictionary(tableWithDict.getChild("str")?.type)).toBe(
+      true,
+    );
+
+    const schema = new Schema([
+      new Field("str", new Dictionary(new Utf8(), new Int32())),
+    ]);
+
+    const tableWithDict2 = makeArrowTable([{ str: "hello" }], { schema });
+    expect(DataType.isDictionary(tableWithDict2.getChild("str")?.type)).toBe(
+      true,
+    );
+  });
+
+  it("will infer data types correctly", async function () {
+    await checkTableCreation(async (records) => makeArrowTable(records), true);
+  });
+
+  it("will allow a schema to be provided", async function () {
+    await checkTableCreation(
+      async (records, _, schema) => makeArrowTable(records, { schema }),
+      false,
+    );
+  });
+
+  it("will use the field order of any provided schema", async function () {
+    await checkTableCreation(
+      async (_, recordsReversed, schema) =>
+        makeArrowTable(recordsReversed, { schema }),
+      false,
+    );
+  });
+
+  it("will make an empty table", async function () {
+    await checkTableCreation(
+      async (_, __, schema) => makeArrowTable([], { schema }),
+      false,
+    );
+  });
+});
+
+class DummyEmbedding implements EmbeddingFunction<string> {
+  public readonly sourceColumn = "string";
+  public readonly embeddingDimension = 2;
+  public readonly embeddingDataType = new Float16();
+
+  async embed(data: string[]): Promise<number[][]> {
+    return data.map(() => [0.0, 0.0]);
+  }
+}
+
+class DummyEmbeddingWithNoDimension implements EmbeddingFunction<string> {
+  public readonly sourceColumn = "string";
+
+  async embed(data: string[]): Promise<number[][]> {
+    return data.map(() => [0.0, 0.0]);
+  }
+}
+
+describe("convertToTable", function () {
+  it("will infer data types correctly", async function () {
+    await checkTableCreation(
+      async (records) => await convertToTable(records),
+      true,
+    );
+  });
+
+  it("will allow a schema to be provided", async function () {
+    await checkTableCreation(
+      async (records, _, schema) =>
+        await convertToTable(records, undefined, { schema }),
+      false,
+    );
+  });
+
+  it("will use the field order of any provided schema", async function () {
+    await checkTableCreation(
+      async (_, recordsReversed, schema) =>
+        await convertToTable(recordsReversed, undefined, { schema }),
+      false,
+    );
+  });
+
+  it("will make an empty table", async function () {
+    await checkTableCreation(
+      async (_, __, schema) => await convertToTable([], undefined, { schema }),
+      false,
+    );
+  });
+
+  it("will apply embeddings", async function () {
+    const records = sampleRecords();
+    const table = await convertToTable(records, new DummyEmbedding());
+    expect(DataType.isFixedSizeList(table.getChild("vector")?.type)).toBe(true);
+    expect(table.getChild("vector")?.type.children[0].type.toString()).toEqual(
+      new Float16().toString(),
+    );
+  });
+
+  it("will fail if missing the embedding source column", async function () {
+    await expect(
+      convertToTable([{ id: 1 }], new DummyEmbedding()),
+    ).rejects.toThrow("'string' was not present");
+  });
+
+  it("use embeddingDimension if embedding missing from table", async function () {
+    const schema = new Schema([new Field("string", new Utf8(), false)]);
+    // Simulate getting an empty Arrow table (minus embedding) from some other source
+    // In other words, we aren't starting with records
+    const table = makeEmptyTable(schema);
+
+    // If the embedding specifies the dimension we are fine
+    await fromTableToBuffer(table, new DummyEmbedding());
+
+    // We can also supply a schema and should be ok
+    const schemaWithEmbedding = new Schema([
+      new Field("string", new Utf8(), false),
+      new Field(
+        "vector",
+        new FixedSizeList(2, new Field("item", new Float16(), false)),
+        false,
+      ),
+    ]);
+    await fromTableToBuffer(
+      table,
+      new DummyEmbeddingWithNoDimension(),
+      schemaWithEmbedding,
+    );
+
+    // Otherwise we will get an error
+    await expect(
+      fromTableToBuffer(table, new DummyEmbeddingWithNoDimension()),
+    ).rejects.toThrow("does not specify `embeddingDimension`");
+  });
+
+  it("will apply embeddings to an empty table", async function () {
+    const schema = new Schema([
+      new Field("string", new Utf8(), false),
+      new Field(
+        "vector",
+        new FixedSizeList(2, new Field("item", new Float16(), false)),
+        false,
+      ),
+    ]);
+    const table = await convertToTable([], new DummyEmbedding(), { schema });
+    expect(DataType.isFixedSizeList(table.getChild("vector")?.type)).toBe(true);
+    expect(table.getChild("vector")?.type.children[0].type.toString()).toEqual(
+      new Float16().toString(),
+    );
+  });
+
+  it("will complain if embeddings present but schema missing embedding column", async function () {
+    const schema = new Schema([new Field("string", new Utf8(), false)]);
+    await expect(
+      convertToTable([], new DummyEmbedding(), { schema }),
+    ).rejects.toThrow("column vector was missing");
+  });
+
+  it("will provide a nice error if run twice", async function () {
+    const records = sampleRecords();
+    const table = await convertToTable(records, new DummyEmbedding());
+    // fromTableToBuffer will try and apply the embeddings again
+    await expect(
+      fromTableToBuffer(table, new DummyEmbedding()),
+    ).rejects.toThrow("already existed");
+  });
+});
+
+describe("makeEmptyTable", function () {
+  it("will make an empty table", async function () {
+    await checkTableCreation(
+      async (_, __, schema) => makeEmptyTable(schema),
+      false,
+    );
+  });
+});
+
+describe("when using two versions of arrow", function () {
+  it("can still import data", async function () {
+    const schema = new OldSchema([
+      new OldField("id", new OldInt32()),
+      new OldField(
+        "vector",
+        new OldFixedSizeList(
+          1024,
+          new OldField("item", new OldFloat32(), true),
+        ),
+      ),
+      new OldField(
+        "struct",
+        new OldStruct([
+          new OldField(
+            "nested",
+            new OldDictionary(new OldUtf8(), new OldInt32(), 1, true),
+          ),
+          new OldField("ts_with_tz", new OldTimestampNanosecond("some_tz")),
+          new OldField("ts_no_tz", new OldTimestampNanosecond(null)),
+        ]),
+      ),
+      // eslint-disable-next-line @typescript-eslint/no-explicit-any
+    ]) as any;
+    schema.metadataVersion = MetadataVersion.V5;
+    const table = makeArrowTable([], { schema });
+
+    const buf = await fromTableToBuffer(table);
+    expect(buf.byteLength).toBeGreaterThan(0);
+    const actual = tableFromIPC(buf);
+    const actualSchema = actual.schema;
+    expect(actualSchema.fields.length).toBe(3);
+
+    // Deep equality gets hung up on some very minor unimportant differences
+    // between arrow version 13 and 15 which isn't really what we're testing for
+    // and so we do our own comparison that just checks name/type/nullability
+    function compareFields(lhs: Field, rhs: Field) {
+      expect(lhs.name).toEqual(rhs.name);
+      expect(lhs.nullable).toEqual(rhs.nullable);
+      expect(lhs.typeId).toEqual(rhs.typeId);
+      if ("children" in lhs.type && lhs.type.children !== null) {
+        const lhsChildren = lhs.type.children as Field[];
+        lhsChildren.forEach((child: Field, idx) => {
+          compareFields(child, rhs.type.children[idx]);
+        });
+      }
+    }
+    actualSchema.fields.forEach((field, idx) => {
+      compareFields(field, actualSchema.fields[idx]);
+    });
+  });
 });

nodejs/__test__/connection.test.ts

@@ -0,0 +1,88 @@
+// Copyright 2024 Lance Developers.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import * as tmp from "tmp";
+
+import { Connection, connect } from "../dist/index.js";
+
+describe("when connecting", () => {
+  let tmpDir: tmp.DirResult;
+  beforeEach(() => (tmpDir = tmp.dirSync({ unsafeCleanup: true })));
+  afterEach(() => tmpDir.removeCallback());
+
+  it("should connect", async () => {
+    const db = await connect(tmpDir.name);
+    expect(db.display()).toBe(
+      `NativeDatabase(uri=${tmpDir.name}, read_consistency_interval=None)`,
+    );
+  });
+
+  it("should allow read consistency interval to be specified", async () => {
+    const db = await connect(tmpDir.name, { readConsistencyInterval: 5 });
+    expect(db.display()).toBe(
+      `NativeDatabase(uri=${tmpDir.name}, read_consistency_interval=5s)`,
+    );
+  });
+});
+
+describe("given a connection", () => {
+  let tmpDir: tmp.DirResult;
+  let db: Connection;
+  beforeEach(async () => {
+    tmpDir = tmp.dirSync({ unsafeCleanup: true });
+    db = await connect(tmpDir.name);
+  });
+  afterEach(() => tmpDir.removeCallback());
+
+  it("should raise an error if opening a non-existent table", async () => {
+    await expect(db.openTable("non-existent")).rejects.toThrow("was not found");
+  });
+
+  it("should raise an error if any operation is tried after it is closed", async () => {
+    expect(db.isOpen()).toBe(true);
+    await db.close();
+    expect(db.isOpen()).toBe(false);
+    await expect(db.tableNames()).rejects.toThrow("Connection is closed");
+  });
+
+  it("should fail if creating table twice, unless overwrite is true", async () => {
+    let tbl = await db.createTable("test", [{ id: 1 }, { id: 2 }]);
+    await expect(tbl.countRows()).resolves.toBe(2);
+    await expect(
+      db.createTable("test", [{ id: 1 }, { id: 2 }]),
+    ).rejects.toThrow();
+    tbl = await db.createTable("test", [{ id: 3 }], { mode: "overwrite" });
+    await expect(tbl.countRows()).resolves.toBe(1);
+  });
+
+  it("should respect limit and page token when listing tables", async () => {
+    const db = await connect(tmpDir.name);
+
+    await db.createTable("b", [{ id: 1 }]);
+    await db.createTable("a", [{ id: 1 }]);
+    await db.createTable("c", [{ id: 1 }]);
+
+    let tables = await db.tableNames();
+    expect(tables).toEqual(["a", "b", "c"]);
+
+    tables = await db.tableNames({ limit: 1 });
+    expect(tables).toEqual(["a"]);
+
+    tables = await db.tableNames({ limit: 1, startAfter: "a" });
+    expect(tables).toEqual(["b"]);
+
+    tables = await db.tableNames({ startAfter: "a" });
+    expect(tables).toEqual(["b", "c"]);
+  });
+});

nodejs/__test__/index.test.ts

@@ -1,34 +0,0 @@
-// Copyright 2024 Lance Developers.
-//
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-//     http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-
-import * as os from "os";
-import * as path from "path";
-import * as fs from "fs";
-
-import { Schema, Field, Float64 } from "apache-arrow";
-import { connect } from "../dist/index.js";
-
-test("open database", async () => {
-  const tmpDir = fs.mkdtempSync(path.join(os.tmpdir(), "test-open"));
-
-  const db = await connect(tmpDir);
-  let tableNames = await db.tableNames();
-  expect(tableNames).toStrictEqual([]);
-
-  const tbl = await db.createTable("test", [{ id: 1 }, { id: 2 }]);
-  expect(await db.tableNames()).toStrictEqual(["test"]);
-
-  const schema = tbl.schema;
-  expect(schema).toEqual(new Schema([new Field("id", new Float64(), true)]));
-});

nodejs/__test__/table.test.ts

@@ -12,27 +12,91 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-import * as os from "os";
-import * as path from "path";
 import * as fs from "fs";
+import * as path from "path";
+import * as tmp from "tmp";
 
-import { connect } from "../dist";
+import { Table, connect } from "../dist";
-import { Schema, Field, Float32, Int32, FixedSizeList } from "apache-arrow";
+import {
+  Schema,
+  Field,
+  Float32,
+  Int32,
+  FixedSizeList,
+  Int64,
+  Float64,
+} from "apache-arrow";
 import { makeArrowTable } from "../dist/arrow";
+import { Index } from "../dist/indices";
 
-describe("Test creating index", () => {
+describe("Given a table", () => {
-  let tmpDir: string;
+  let tmpDir: tmp.DirResult;
+  let table: Table;
+  const schema = new Schema([new Field("id", new Float64(), true)]);
+  beforeEach(async () => {
+    tmpDir = tmp.dirSync({ unsafeCleanup: true });
+    const conn = await connect(tmpDir.name);
+    table = await conn.createEmptyTable("some_table", schema);
+  });
+  afterEach(() => tmpDir.removeCallback());
+
+  it("be displayable", async () => {
+    expect(table.display()).toMatch(
+      /NativeTable\(some_table, uri=.*, read_consistency_interval=None\)/,
+    );
+    table.close();
+    expect(table.display()).toBe("ClosedTable(some_table)");
+  });
+
+  it("should let me add data", async () => {
+    await table.add([{ id: 1 }, { id: 2 }]);
+    await table.add([{ id: 1 }]);
+    await expect(table.countRows()).resolves.toBe(3);
+  });
+
+  it("should overwrite data if asked", async () => {
+    await table.add([{ id: 1 }, { id: 2 }]);
+    await table.add([{ id: 1 }], { mode: "overwrite" });
+    await expect(table.countRows()).resolves.toBe(1);
+  });
+
+  it("should let me close the table", async () => {
+    expect(table.isOpen()).toBe(true);
+    table.close();
+    expect(table.isOpen()).toBe(false);
+    expect(table.countRows()).rejects.toThrow("Table some_table is closed");
+  });
+
+  it("should let me update values", async () => {
+    await table.add([{ id: 1 }]);
+    expect(await table.countRows("id == 1")).toBe(1);
+    expect(await table.countRows("id == 7")).toBe(0);
+    await table.update({ id: "7" });
+    expect(await table.countRows("id == 1")).toBe(0);
+    expect(await table.countRows("id == 7")).toBe(1);
+    await table.add([{ id: 2 }]);
+    // Test Map as input
+    await table.update(new Map(Object.entries({ id: "10" })), {
+      where: "id % 2 == 0",
+    });
+    expect(await table.countRows("id == 2")).toBe(0);
+    expect(await table.countRows("id == 7")).toBe(1);
+    expect(await table.countRows("id == 10")).toBe(1);
+  });
+});
+
+describe("When creating an index", () => {
+  let tmpDir: tmp.DirResult;
   const schema = new Schema([
     new Field("id", new Int32(), true),
     new Field("vec", new FixedSizeList(32, new Field("item", new Float32()))),
   ]);
+  let tbl: Table;
+  let queryVec: number[];
 
-  beforeEach(() => {
+  beforeEach(async () => {
-    tmpDir = fs.mkdtempSync(path.join(os.tmpdir(), "index-"));
+    tmpDir = tmp.dirSync({ unsafeCleanup: true });
-  });
+    const db = await connect(tmpDir.name);
-
-  test("create vector index with no column", async () => {
-    const db = await connect(tmpDir);
     const data = makeArrowTable(
       Array(300)
         .fill(1)
@@ -44,57 +108,90 @@ describe("Test creating index", () => {
         })),
       {
         schema,
-      }
+      },
     );
-    const tbl = await db.createTable("test", data);
+    queryVec = data.toArray()[5].vec.toJSON();
-    await tbl.createIndex().build();
+    tbl = await db.createTable("test", data);
+  });
+  afterEach(() => tmpDir.removeCallback());
+
+  it("should create a vector index on vector columns", async () => {
+    await tbl.createIndex("vec");
 
     // check index directory
-    const indexDir = path.join(tmpDir, "test.lance", "_indices");
+    const indexDir = path.join(tmpDir.name, "test.lance", "_indices");
     expect(fs.readdirSync(indexDir)).toHaveLength(1);
-    // TODO: check index type.
+    const indices = await tbl.listIndices();
+    expect(indices.length).toBe(1);
+    expect(indices[0]).toEqual({
+      indexType: "IvfPq",
+      columns: ["vec"],
+    });
 
     // Search without specifying the column
-    let query_vector = data.toArray()[5].vec.toJSON();
+    let rst = await tbl
-    let rst = await tbl.query().nearestTo(query_vector).limit(2).toArrow();
+      .query()
+      .limit(2)
+      .nearestTo(queryVec)
+      .distanceType("DoT")
+      .toArrow();
+    expect(rst.numRows).toBe(2);
+
+    // Search using `vectorSearch`
+    rst = await tbl.vectorSearch(queryVec).limit(2).toArrow();
     expect(rst.numRows).toBe(2);
 
     // Search with specifying the column
-    let rst2 = await tbl.search(query_vector, "vec").limit(2).toArrow();
+    const rst2 = await tbl
+      .query()
+      .limit(2)
+      .nearestTo(queryVec)
+      .column("vec")
+      .toArrow();
     expect(rst2.numRows).toBe(2);
     expect(rst.toString()).toEqual(rst2.toString());
   });
 
-  test("no vector column available", async () => {
+  it("should allow parameters to be specified", async () => {
-    const db = await connect(tmpDir);
+    await tbl.createIndex("vec", {
-    const tbl = await db.createTable(
+      config: Index.ivfPq({
-      "no_vec",
+        numPartitions: 10,
-      makeArrowTable([
+      }),
-        { id: 1, val: 2 },
+    });
-        { id: 2, val: 3 },
-      ])
-    );
-    await expect(tbl.createIndex().build()).rejects.toThrow(
-      "No vector column found"
-    );
 
-    await tbl.createIndex("val").build();
+    // TODO: Verify parameters when we can load index config as part of list indices
-    const indexDir = path.join(tmpDir, "no_vec.lance", "_indices");
+  });
+
+  it("should allow me to replace (or not) an existing index", async () => {
+    await tbl.createIndex("id");
+    // Default is replace=true
+    await tbl.createIndex("id");
+    await expect(tbl.createIndex("id", { replace: false })).rejects.toThrow(
+      "already exists",
+    );
+    await tbl.createIndex("id", { replace: true });
+  });
+
+  test("should create a scalar index on scalar columns", async () => {
+    await tbl.createIndex("id");
+    const indexDir = path.join(tmpDir.name, "test.lance", "_indices");
     expect(fs.readdirSync(indexDir)).toHaveLength(1);
 
-    for await (const r of tbl.query().filter("id > 1").select(["id"])) {
+    for await (const r of tbl.query().where("id > 1").select(["id"])) {
-      expect(r.numRows).toBe(1);
+      expect(r.numRows).toBe(298);
     }
   });
 
+  // TODO: Move this test to the query API test (making sure we can reject queries
+  // when the dimension is incorrect)
   test("two columns with different dimensions", async () => {
-    const db = await connect(tmpDir);
+    const db = await connect(tmpDir.name);
     const schema = new Schema([
       new Field("id", new Int32(), true),
       new Field("vec", new FixedSizeList(32, new Field("item", new Float32()))),
       new Field(
         "vec2",
-        new FixedSizeList(64, new Field("item", new Float32()))
+        new FixedSizeList(64, new Field("item", new Float32())),
       ),
     ]);
     const tbl = await db.createTable(
@@ -111,73 +208,214 @@ describe("Test creating index", () => {
               .fill(1)
               .map(() => Math.random()),
           })),
-        { schema }
+        { schema },
-      )
+      ),
     );
 
     // Only build index over v1
-    await expect(tbl.createIndex().build()).rejects.toThrow(
+    await tbl.createIndex("vec", {
-      /.*More than one vector columns found.*/
+      config: Index.ivfPq({ numPartitions: 2, numSubVectors: 2 }),
-    );
+    });
-    tbl
-      .createIndex("vec")
-      .ivf_pq({ num_partitions: 2, num_sub_vectors: 2 })
-      .build();
 
     const rst = await tbl
       .query()
+      .limit(2)
       .nearestTo(
         Array(32)
           .fill(1)
-          .map(() => Math.random())
+          .map(() => Math.random()),
       )
-      .limit(2)
       .toArrow();
     expect(rst.numRows).toBe(2);
 
     // Search with specifying the column
     await expect(
       tbl
-        .search(
+        .query()
+        .limit(2)
+        .nearestTo(
           Array(64)
             .fill(1)
             .map(() => Math.random()),
-          "vec"
         )
-        .limit(2)
+        .column("vec")
-        .toArrow()
+        .toArrow(),
-    ).rejects.toThrow(/.*does not match the dimension.*/);
+    ).rejects.toThrow(/.* query dim=64, expected vector dim=32.*/);
 
     const query64 = Array(64)
       .fill(1)
       .map(() => Math.random());
-    const rst64_1 = await tbl.query().nearestTo(query64).limit(2).toArrow();
+    const rst64Query = await tbl.query().limit(2).nearestTo(query64).toArrow();
-    const rst64_2 = await tbl.search(query64, "vec2").limit(2).toArrow();
+    const rst64Search = await tbl
-    expect(rst64_1.toString()).toEqual(rst64_2.toString());
+      .query()
-    expect(rst64_1.numRows).toBe(2);
+      .limit(2)
-  });
+      .nearestTo(query64)
-
+      .column("vec2")
-  test("create scalar index", async () => {
+      .toArrow();
-    const db = await connect(tmpDir);
+    expect(rst64Query.toString()).toEqual(rst64Search.toString());
-    const data = makeArrowTable(
+    expect(rst64Query.numRows).toBe(2);
-      Array(300)
+  });
-        .fill(1)
+});
-        .map((_, i) => ({
+
-          id: i,
+describe("Read consistency interval", () => {
-          vec: Array(32)
+  let tmpDir: tmp.DirResult;
-            .fill(1)
+  beforeEach(() => {
-            .map(() => Math.random()),
+    tmpDir = tmp.dirSync({ unsafeCleanup: true });
-        })),
+  });
-      {
+  afterEach(() => tmpDir.removeCallback());
-        schema,
+
-      }
+  // const intervals = [undefined, 0, 0.1];
-    );
+  const intervals = [0];
-    const tbl = await db.createTable("test", data);
+  test.each(intervals)("read consistency interval %p", async (interval) => {
-    await tbl.createIndex("id").build();
+    const db = await connect(tmpDir.name);
-
+    const table = await db.createTable("my_table", [{ id: 1 }]);
-    // check index directory
+
-    const indexDir = path.join(tmpDir, "test.lance", "_indices");
+    const db2 = await connect(tmpDir.name, {
-    expect(fs.readdirSync(indexDir)).toHaveLength(1);
+      readConsistencyInterval: interval,
-    // TODO: check index type.
+    });
+    const table2 = await db2.openTable("my_table");
+    expect(await table2.countRows()).toEqual(await table.countRows());
+
+    await table.add([{ id: 2 }]);
+
+    if (interval === undefined) {
+      expect(await table2.countRows()).toEqual(1);
+      // TODO: once we implement time travel we can uncomment this part of the test.
+      // await table2.checkout_latest();
+      // expect(await table2.countRows()).toEqual(2);
+    } else if (interval === 0) {
+      expect(await table2.countRows()).toEqual(2);
+    } else {
+      // interval == 0.1
+      expect(await table2.countRows()).toEqual(1);
+      await new Promise((r) => setTimeout(r, 100));
+      expect(await table2.countRows()).toEqual(2);
+    }
+  });
+});
+
+describe("schema evolution", function () {
+  let tmpDir: tmp.DirResult;
+  beforeEach(() => {
+    tmpDir = tmp.dirSync({ unsafeCleanup: true });
+  });
+  afterEach(() => {
+    tmpDir.removeCallback();
+  });
+
+  // Create a new sample table
+  it("can add a new column to the schema", async function () {
+    const con = await connect(tmpDir.name);
+    const table = await con.createTable("vectors", [
+      { id: 1n, vector: [0.1, 0.2] },
+    ]);
+
+    await table.addColumns([
+      { name: "price", valueSql: "cast(10.0 as float)" },
+    ]);
+
+    const expectedSchema = new Schema([
+      new Field("id", new Int64(), true),
+      new Field(
+        "vector",
+        new FixedSizeList(2, new Field("item", new Float32(), true)),
+        true,
+      ),
+      new Field("price", new Float32(), false),
+    ]);
+    expect(await table.schema()).toEqual(expectedSchema);
+  });
+
+  it("can alter the columns in the schema", async function () {
+    const con = await connect(tmpDir.name);
+    const schema = new Schema([
+      new Field("id", new Int64(), true),
+      new Field(
+        "vector",
+        new FixedSizeList(2, new Field("item", new Float32(), true)),
+        true,
+      ),
+      new Field("price", new Float64(), false),
+    ]);
+    const table = await con.createTable("vectors", [
+      { id: 1n, vector: [0.1, 0.2] },
+    ]);
+    // Can create a non-nullable column only through addColumns at the moment.
+    await table.addColumns([
+      { name: "price", valueSql: "cast(10.0 as double)" },
+    ]);
+    expect(await table.schema()).toEqual(schema);
+
+    await table.alterColumns([
+      { path: "id", rename: "new_id" },
+      { path: "price", nullable: true },
+    ]);
+
+    const expectedSchema = new Schema([
+      new Field("new_id", new Int64(), true),
+      new Field(
+        "vector",
+        new FixedSizeList(2, new Field("item", new Float32(), true)),
+        true,
+      ),
+      new Field("price", new Float64(), true),
+    ]);
+    expect(await table.schema()).toEqual(expectedSchema);
+  });
+
+  it("can drop a column from the schema", async function () {
+    const con = await connect(tmpDir.name);
+    const table = await con.createTable("vectors", [
+      { id: 1n, vector: [0.1, 0.2] },
+    ]);
+    await table.dropColumns(["vector"]);
+
+    const expectedSchema = new Schema([new Field("id", new Int64(), true)]);
+    expect(await table.schema()).toEqual(expectedSchema);
+  });
+});
+
+describe("when dealing with versioning", () => {
+  let tmpDir: tmp.DirResult;
+  beforeEach(() => {
+    tmpDir = tmp.dirSync({ unsafeCleanup: true });
+  });
+  afterEach(() => {
+    tmpDir.removeCallback();
+  });
+
+  it("can travel in time", async () => {
+    // Setup
+    const con = await connect(tmpDir.name);
+    const table = await con.createTable("vectors", [
+      { id: 1n, vector: [0.1, 0.2] },
+    ]);
+    const version = await table.version();
+    await table.add([{ id: 2n, vector: [0.1, 0.2] }]);
+    expect(await table.countRows()).toBe(2);
+    // Make sure we can rewind
+    await table.checkout(version);
+    expect(await table.countRows()).toBe(1);
+    // Can't add data in time travel mode
+    await expect(table.add([{ id: 3n, vector: [0.1, 0.2] }])).rejects.toThrow(
+      "table cannot be modified when a specific version is checked out",
+    );
+    // Can go back to normal mode
+    await table.checkoutLatest();
+    expect(await table.countRows()).toBe(2);
+    // Should be able to add data again
+    await table.add([{ id: 2n, vector: [0.1, 0.2] }]);
+    expect(await table.countRows()).toBe(3);
+    // Now checkout and restore
+    await table.checkout(version);
+    await table.restore();
+    expect(await table.countRows()).toBe(1);
+    // Should be able to add data
+    await table.add([{ id: 2n, vector: [0.1, 0.2] }]);
+    expect(await table.countRows()).toBe(2);
+    // Can't use restore if not checked out
+    await expect(table.restore()).rejects.toThrow(
+      "checkout before running restore",
+    );
   });
 });

nodejs/__test__/tsconfig.json

@@ -0,0 +1,10 @@
+{
+  "extends": "../tsconfig.json",
+  "compilerOptions": {
+    "outDir": "./dist/spec",
+    "module": "commonjs",
+    "target": "es2022",
+    "types": ["jest", "node"]
+  },
+  "include": ["**/*"]
+}

nodejs/eslint.config.js

@@ -0,0 +1,28 @@
+/* eslint-disable @typescript-eslint/naming-convention */
+// @ts-check
+
+const eslint = require("@eslint/js");
+const tseslint = require("typescript-eslint");
+const eslintConfigPrettier = require("eslint-config-prettier");
+const jsdoc = require("eslint-plugin-jsdoc");
+
+module.exports = tseslint.config(
+  eslint.configs.recommended,
+  jsdoc.configs["flat/recommended"],
+  eslintConfigPrettier,
+  ...tseslint.configs.recommended,
+  {
+    rules: {
+      "@typescript-eslint/naming-convention": "error",
+      "jsdoc/require-returns": "off",
+      "jsdoc/require-param": "off",
+      "jsdoc/require-jsdoc": [
+        "error",
+        {
+          publicOnly: true,
+        },
+      ],
+    },
+    plugins: jsdoc,
+  },
+);

nodejs/jest.config.js

@@ -1,7 +1,7 @@
 /** @type {import('ts-jest').JestConfigWithTsJest} */
 module.exports = {
-  preset: 'ts-jest',
+  preset: "ts-jest",
-  testEnvironment: 'node',
+  testEnvironment: "node",
   moduleDirectories: ["node_modules", "./dist"],
   moduleFileExtensions: ["js", "ts"],
 };

nodejs/lancedb/arrow.ts

@@ -0,0 +1,647 @@
+// Copyright 2023 Lance Developers.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import {
+  Field,
+  makeBuilder,
+  RecordBatchFileWriter,
+  Utf8,
+  type Vector,
+  FixedSizeList,
+  vectorFromArray,
+  type Schema,
+  Table as ArrowTable,
+  RecordBatchStreamWriter,
+  List,
+  RecordBatch,
+  makeData,
+  Struct,
+  type Float,
+  DataType,
+  Binary,
+  Float32,
+  type makeTable,
+} from "apache-arrow";
+import { type EmbeddingFunction } from "./embedding/embedding_function";
+import { sanitizeSchema } from "./sanitize";
+
+/** Data type accepted by NodeJS SDK */
+export type Data = Record<string, unknown>[] | ArrowTable;
+
+/*
+ * Options to control how a column should be converted to a vector array
+ */
+export class VectorColumnOptions {
+  /** Vector column type. */
+  type: Float = new Float32();
+
+  constructor(values?: Partial<VectorColumnOptions>) {
+    Object.assign(this, values);
+  }
+}
+
+/** Options to control the makeArrowTable call. */
+export class MakeArrowTableOptions {
+  /*
+   * Schema of the data.
+   *
+   * If this is not provided then the data type will be inferred from the
+   * JS type.  Integer numbers will become int64, floating point numbers
+   * will become float64 and arrays will become variable sized lists with
+   * the data type inferred from the first element in the array.
+   *
+   * The schema must be specified if there are no records (e.g. to make
+   * an empty table)
+   */
+  schema?: Schema;
+
+  /*
+   * Mapping from vector column name to expected type
+   *
+   * Lance expects vector columns to be fixed size list arrays (i.e. tensors)
+   * However, `makeArrowTable` will not infer this by default (it creates
+   * variable size list arrays).  This field can be used to indicate that a column
+   * should be treated as a vector column and converted to a fixed size list.
+   *
+   * The keys should be the names of the vector columns.  The value specifies the
+   * expected data type of the vector columns.
+   *
+   * If `schema` is provided then this field is ignored.
+   *
+   * By default, the column named "vector" will be assumed to be a float32
+   * vector column.
+   */
+  vectorColumns: Record<string, VectorColumnOptions> = {
+    vector: new VectorColumnOptions(),
+  };
+
+  /**
+   * If true then string columns will be encoded with dictionary encoding
+   *
+   * Set this to true if your string columns tend to repeat the same values
+   * often.  For more precise control use the `schema` property to specify the
+   * data type for individual columns.
+   *
+   * If `schema` is provided then this property is ignored.
+   */
+  dictionaryEncodeStrings: boolean = false;
+
+  constructor(values?: Partial<MakeArrowTableOptions>) {
+    Object.assign(this, values);
+  }
+}
+
+/**
+ * An enhanced version of the {@link makeTable} function from Apache Arrow
+ * that supports nested fields and embeddings columns.
+ *
+ * This function converts an array of Record<String, any> (row-major JS objects)
+ * to an Arrow Table (a columnar structure)
+ *
+ * Note that it currently does not support nulls.
+ *
+ * If a schema is provided then it will be used to determine the resulting array
+ * types.  Fields will also be reordered to fit the order defined by the schema.
+ *
+ * If a schema is not provided then the types will be inferred and the field order
+ * will be controlled by the order of properties in the first record.  If a type
+ * is inferred it will always be nullable.
+ *
+ * If the input is empty then a schema must be provided to create an empty table.
+ *
+ * When a schema is not specified then data types will be inferred.  The inference
+ * rules are as follows:
+ *
+ *  - boolean => Bool
+ *  - number => Float64
+ *  - String => Utf8
+ *  - Buffer => Binary
+ *  - Record<String, any> => Struct
+ *  - Array<any> => List
+ * @example
+ * import { fromTableToBuffer, makeArrowTable } from "../arrow";
+ * import { Field, FixedSizeList, Float16, Float32, Int32, Schema } from "apache-arrow";
+ *
+ * const schema = new Schema([
+ *   new Field("a", new Int32()),
+ *   new Field("b", new Float32()),
+ *   new Field("c", new FixedSizeList(3, new Field("item", new Float16()))),
+ *  ]);
+ *  const table = makeArrowTable([
+ *    { a: 1, b: 2, c: [1, 2, 3] },
+ *    { a: 4, b: 5, c: [4, 5, 6] },
+ *    { a: 7, b: 8, c: [7, 8, 9] },
+ *  ], { schema });
+ * ```
+ *
+ * By default it assumes that the column named `vector` is a vector column
+ * and it will be converted into a fixed size list array of type float32.
+ * The `vectorColumns` option can be used to support other vector column
+ * names and data types.
+ *
+ * ```ts
+ *
+ * const schema = new Schema([
+    new Field("a", new Float64()),
+    new Field("b", new Float64()),
+    new Field(
+      "vector",
+      new FixedSizeList(3, new Field("item", new Float32()))
+    ),
+  ]);
+  const table = makeArrowTable([
+    { a: 1, b: 2, vector: [1, 2, 3] },
+    { a: 4, b: 5, vector: [4, 5, 6] },
+    { a: 7, b: 8, vector: [7, 8, 9] },
+  ]);
+  assert.deepEqual(table.schema, schema);
+ * ```
+ *
+ * You can specify the vector column types and names using the options as well
+ *
+ * ```typescript
+ *
+ * const schema = new Schema([
+    new Field('a', new Float64()),
+    new Field('b', new Float64()),
+    new Field('vec1', new FixedSizeList(3, new Field('item', new Float16()))),
+    new Field('vec2', new FixedSizeList(3, new Field('item', new Float16())))
+  ]);
+ * const table = makeArrowTable([
+    { a: 1, b: 2, vec1: [1, 2, 3], vec2: [2, 4, 6] },
+    { a: 4, b: 5, vec1: [4, 5, 6], vec2: [8, 10, 12] },
+    { a: 7, b: 8, vec1: [7, 8, 9], vec2: [14, 16, 18] }
+  ], {
+    vectorColumns: {
+      vec1: { type: new Float16() },
+      vec2: { type: new Float16() }
+    }
+  }
+ * assert.deepEqual(table.schema, schema)
+ * ```
+ */
+export function makeArrowTable(
+  data: Array<Record<string, unknown>>,
+  options?: Partial<MakeArrowTableOptions>,
+): ArrowTable {
+  if (
+    data.length === 0 &&
+    (options?.schema === undefined || options?.schema === null)
+  ) {
+    throw new Error("At least one record or a schema needs to be provided");
+  }
+
+  const opt = new MakeArrowTableOptions(options !== undefined ? options : {});
+  if (opt.schema !== undefined && opt.schema !== null) {
+    opt.schema = sanitizeSchema(opt.schema);
+  }
+  const columns: Record<string, Vector> = {};
+  // TODO: sample dataset to find missing columns
+  // Prefer the field ordering of the schema, if present
+  const columnNames =
+    opt.schema != null ? (opt.schema.names as string[]) : Object.keys(data[0]);
+  for (const colName of columnNames) {
+    if (
+      data.length !== 0 &&
+      !Object.prototype.hasOwnProperty.call(data[0], colName)
+    ) {
+      // The field is present in the schema, but not in the data, skip it
+      continue;
+    }
+    // Extract a single column from the records (transpose from row-major to col-major)
+    let values = data.map((datum) => datum[colName]);
+
+    // By default (type === undefined) arrow will infer the type from the JS type
+    let type;
+    if (opt.schema !== undefined) {
+      // If there is a schema provided, then use that for the type instead
+      type = opt.schema?.fields.filter((f) => f.name === colName)[0]?.type;
+      if (DataType.isInt(type) && type.bitWidth === 64) {
+        // wrap in BigInt to avoid bug: https://github.com/apache/arrow/issues/40051
+        values = values.map((v) => {
+          if (v === null) {
+            return v;
+          }
+          if (typeof v === "bigint") {
+            return v;
+          }
+          if (typeof v === "number") {
+            return BigInt(v);
+          }
+          throw new Error(
+            `Expected BigInt or number for column ${colName}, got ${typeof v}`,
+          );
+        });
+      }
+    } else {
+      // Otherwise, check to see if this column is one of the vector columns
+      // defined by opt.vectorColumns and, if so, use the fixed size list type
+      const vectorColumnOptions = opt.vectorColumns[colName];
+      if (vectorColumnOptions !== undefined) {
+        const firstNonNullValue = values.find((v) => v !== null);
+        if (Array.isArray(firstNonNullValue)) {
+          type = newVectorType(
+            firstNonNullValue.length,
+            vectorColumnOptions.type,
+          );
+        } else {
+          throw new Error(
+            `Column ${colName} is expected to be a vector column but first non-null value is not an array.  Could not determine size of vector column`,
+          );
+        }
+      }
+    }
+
+    try {
+      // Convert an Array of JS values to an arrow vector
+      columns[colName] = makeVector(values, type, opt.dictionaryEncodeStrings);
+    } catch (error: unknown) {
+      // eslint-disable-next-line @typescript-eslint/restrict-template-expressions
+      throw Error(`Could not convert column "${colName}" to Arrow: ${error}`);
+    }
+  }
+
+  if (opt.schema != null) {
+    // `new ArrowTable(columns)` infers a schema which may sometimes have
+    // incorrect nullability (it assumes nullable=true always)
+    //
+    // `new ArrowTable(schema, columns)` will also fail because it will create a
+    // batch with an inferred schema and then complain that the batch schema
+    // does not match the provided schema.
+    //
+    // To work around this we first create a table with the wrong schema and
+    // then patch the schema of the batches so we can use
+    // `new ArrowTable(schema, batches)` which does not do any schema inference
+    const firstTable = new ArrowTable(columns);
+    // eslint-disable-next-line @typescript-eslint/no-non-null-assertion
+    const batchesFixed = firstTable.batches.map(
+      (batch) => new RecordBatch(opt.schema!, batch.data),
+    );
+    return new ArrowTable(opt.schema, batchesFixed);
+  } else {
+    return new ArrowTable(columns);
+  }
+}
+
+/**
+ * Create an empty Arrow table with the provided schema
+ */
+export function makeEmptyTable(schema: Schema): ArrowTable {
+  return makeArrowTable([], { schema });
+}
+
+/**
+ * Helper function to convert Array<Array<any>> to a variable sized list array
+ */
+// @ts-expect-error (Vector<unknown> is not assignable to Vector<any>)
+function makeListVector(lists: unknown[][]): Vector<unknown> {
+  if (lists.length === 0 || lists[0].length === 0) {
+    throw Error("Cannot infer list vector from empty array or empty list");
+  }
+  const sampleList = lists[0];
+  // eslint-disable-next-line @typescript-eslint/no-explicit-any
+  let inferredType: any;
+  try {
+    const sampleVector = makeVector(sampleList);
+    inferredType = sampleVector.type;
+  } catch (error: unknown) {
+    // eslint-disable-next-line @typescript-eslint/restrict-template-expressions
+    throw Error(`Cannot infer list vector.  Cannot infer inner type: ${error}`);
+  }
+
+  const listBuilder = makeBuilder({
+    type: new List(new Field("item", inferredType, true)),
+  });
+  for (const list of lists) {
+    listBuilder.append(list);
+  }
+  return listBuilder.finish().toVector();
+}
+
+/** Helper function to convert an Array of JS values to an Arrow Vector */
+function makeVector(
+  values: unknown[],
+  type?: DataType,
+  stringAsDictionary?: boolean,
+  // eslint-disable-next-line @typescript-eslint/no-explicit-any
+): Vector<any> {
+  if (type !== undefined) {
+    // No need for inference, let Arrow create it
+    return vectorFromArray(values, type);
+  }
+  if (values.length === 0) {
+    throw Error(
+      "makeVector requires at least one value or the type must be specfied",
+    );
+  }
+  const sampleValue = values.find((val) => val !== null && val !== undefined);
+  if (sampleValue === undefined) {
+    throw Error(
+      "makeVector cannot infer the type if all values are null or undefined",
+    );
+  }
+  if (Array.isArray(sampleValue)) {
+    // Default Arrow inference doesn't handle list types
+    return makeListVector(values as unknown[][]);
+  } else if (Buffer.isBuffer(sampleValue)) {
+    // Default Arrow inference doesn't handle Buffer
+    return vectorFromArray(values, new Binary());
+  } else if (
+    !(stringAsDictionary ?? false) &&
+    (typeof sampleValue === "string" || sampleValue instanceof String)
+  ) {
+    // If the type is string then don't use Arrow's default inference unless dictionaries are requested
+    // because it will always use dictionary encoding for strings
+    return vectorFromArray(values, new Utf8());
+  } else {
+    // Convert a JS array of values to an arrow vector
+    return vectorFromArray(values);
+  }
+}
+
+/** Helper function to apply embeddings to an input table */
+async function applyEmbeddings<T>(
+  table: ArrowTable,
+  embeddings?: EmbeddingFunction<T>,
+  schema?: Schema,
+): Promise<ArrowTable> {
+  if (embeddings == null) {
+    return table;
+  }
+
+  if (schema !== undefined && schema !== null) {
+    schema = sanitizeSchema(schema);
+  }
+
+  // Convert from ArrowTable to Record<String, Vector>
+  const colEntries = [...Array(table.numCols).keys()].map((_, idx) => {
+    const name = table.schema.fields[idx].name;
+    // eslint-disable-next-line @typescript-eslint/no-non-null-assertion
+    const vec = table.getChildAt(idx)!;
+    return [name, vec];
+  });
+  const newColumns = Object.fromEntries(colEntries);
+
+  const sourceColumn = newColumns[embeddings.sourceColumn];
+  const destColumn = embeddings.destColumn ?? "vector";
+  const innerDestType = embeddings.embeddingDataType ?? new Float32();
+  if (sourceColumn === undefined) {
+    throw new Error(
+      `Cannot apply embedding function because the source column '${embeddings.sourceColumn}' was not present in the data`,
+    );
+  }
+
+  if (table.numRows === 0) {
+    if (Object.prototype.hasOwnProperty.call(newColumns, destColumn)) {
+      // We have an empty table and it already has the embedding column so no work needs to be done
+      // Note: we don't return an error like we did below because this is a common occurrence.  For example,
+      // if we call convertToTable with 0 records and a schema that includes the embedding
+      return table;
+    }
+    if (embeddings.embeddingDimension !== undefined) {
+      const destType = newVectorType(
+        embeddings.embeddingDimension,
+        innerDestType,
+      );
+      newColumns[destColumn] = makeVector([], destType);
+    } else if (schema != null) {
+      const destField = schema.fields.find((f) => f.name === destColumn);
+      if (destField != null) {
+        newColumns[destColumn] = makeVector([], destField.type);
+      } else {
+        throw new Error(
+          `Attempt to apply embeddings to an empty table failed because schema was missing embedding column '${destColumn}'`,
+        );
+      }
+    } else {
+      throw new Error(
+        "Attempt to apply embeddings to an empty table when the embeddings function does not specify `embeddingDimension`",
+      );
+    }
+  } else {
+    if (Object.prototype.hasOwnProperty.call(newColumns, destColumn)) {
+      throw new Error(
+        `Attempt to apply embeddings to table failed because column ${destColumn} already existed`,
+      );
+    }
+    if (table.batches.length > 1) {
+      throw new Error(
+        "Internal error: `makeArrowTable` unexpectedly created a table with more than one batch",
+      );
+    }
+    const values = sourceColumn.toArray();
+    const vectors = await embeddings.embed(values as T[]);
+    if (vectors.length !== values.length) {
+      throw new Error(
+        "Embedding function did not return an embedding for each input element",
+      );
+    }
+    const destType = newVectorType(vectors[0].length, innerDestType);
+    newColumns[destColumn] = makeVector(vectors, destType);
+  }
+
+  const newTable = new ArrowTable(newColumns);
+  if (schema != null) {
+    if (schema.fields.find((f) => f.name === destColumn) === undefined) {
+      throw new Error(
+        `When using embedding functions and specifying a schema the schema should include the embedding column but the column ${destColumn} was missing`,
+      );
+    }
+    return alignTable(newTable, schema);
+  }
+  return newTable;
+}
+
+/**
+ * Convert an Array of records into an Arrow Table, optionally applying an
+ * embeddings function to it.
+ *
+ * This function calls `makeArrowTable` first to create the Arrow Table.
+ * Any provided `makeTableOptions` (e.g. a schema) will be passed on to
+ * that call.
+ *
+ * The embedding function will be passed a column of values (based on the
+ * `sourceColumn` of the embedding function) and expects to receive back
+ * number[][] which will be converted into a fixed size list column.  By
+ * default this will be a fixed size list of Float32 but that can be
+ * customized by the `embeddingDataType` property of the embedding function.
+ *
+ * If a schema is provided in `makeTableOptions` then it should include the
+ * embedding columns.  If no schema is provded then embedding columns will
+ * be placed at the end of the table, after all of the input columns.
+ */
+export async function convertToTable<T>(
+  data: Array<Record<string, unknown>>,
+  embeddings?: EmbeddingFunction<T>,
+  makeTableOptions?: Partial<MakeArrowTableOptions>,
+): Promise<ArrowTable> {
+  const table = makeArrowTable(data, makeTableOptions);
+  return await applyEmbeddings(table, embeddings, makeTableOptions?.schema);
+}
+
+/** Creates the Arrow Type for a Vector column with dimension `dim` */
+function newVectorType<T extends Float>(
+  dim: number,
+  innerType: T,
+): FixedSizeList<T> {
+  // in Lance we always default to have the elements nullable, so we need to set it to true
+  // otherwise we often get schema mismatches because the stored data always has schema with nullable elements
+  const children = new Field<T>("item", innerType, true);
+  return new FixedSizeList(dim, children);
+}
+
+/**
+ * Serialize an Array of records into a buffer using the Arrow IPC File serialization
+ *
+ * This function will call `convertToTable` and pass on `embeddings` and `schema`
+ *
+ * `schema` is required if data is empty
+ */
+export async function fromRecordsToBuffer<T>(
+  data: Array<Record<string, unknown>>,
+  embeddings?: EmbeddingFunction<T>,
+  schema?: Schema,
+): Promise<Buffer> {
+  if (schema !== undefined && schema !== null) {
+    schema = sanitizeSchema(schema);
+  }
+  const table = await convertToTable(data, embeddings, { schema });
+  const writer = RecordBatchFileWriter.writeAll(table);
+  return Buffer.from(await writer.toUint8Array());
+}
+
+/**
+ * Serialize an Array of records into a buffer using the Arrow IPC Stream serialization
+ *
+ * This function will call `convertToTable` and pass on `embeddings` and `schema`
+ *
+ * `schema` is required if data is empty
+ */
+export async function fromRecordsToStreamBuffer<T>(
+  data: Array<Record<string, unknown>>,
+  embeddings?: EmbeddingFunction<T>,
+  schema?: Schema,
+): Promise<Buffer> {
+  if (schema !== undefined && schema !== null) {
+    schema = sanitizeSchema(schema);
+  }
+  const table = await convertToTable(data, embeddings, { schema });
+  const writer = RecordBatchStreamWriter.writeAll(table);
+  return Buffer.from(await writer.toUint8Array());
+}
+
+/**
+ * Serialize an Arrow Table into a buffer using the Arrow IPC File serialization
+ *
+ * This function will apply `embeddings` to the table in a manner similar to
+ * `convertToTable`.
+ *
+ * `schema` is required if the table is empty
+ */
+export async function fromTableToBuffer<T>(
+  table: ArrowTable,
+  embeddings?: EmbeddingFunction<T>,
+  schema?: Schema,
+): Promise<Buffer> {
+  if (schema !== undefined && schema !== null) {
+    schema = sanitizeSchema(schema);
+  }
+  const tableWithEmbeddings = await applyEmbeddings(table, embeddings, schema);
+  const writer = RecordBatchFileWriter.writeAll(tableWithEmbeddings);
+  return Buffer.from(await writer.toUint8Array());
+}
+
+/**
+ * Serialize an Arrow Table into a buffer using the Arrow IPC File serialization
+ *
+ * This function will apply `embeddings` to the table in a manner similar to
+ * `convertToTable`.
+ *
+ * `schema` is required if the table is empty
+ */
+export async function fromDataToBuffer<T>(
+  data: Data,
+  embeddings?: EmbeddingFunction<T>,
+  schema?: Schema,
+): Promise<Buffer> {
+  if (schema !== undefined && schema !== null) {
+    schema = sanitizeSchema(schema);
+  }
+  if (data instanceof ArrowTable) {
+    return fromTableToBuffer(data, embeddings, schema);
+  } else {
+    const table = await convertToTable(data);
+    return fromTableToBuffer(table, embeddings, schema);
+  }
+}
+
+/**
+ * Serialize an Arrow Table into a buffer using the Arrow IPC Stream serialization
+ *
+ * This function will apply `embeddings` to the table in a manner similar to
+ * `convertToTable`.
+ *
+ * `schema` is required if the table is empty
+ */
+export async function fromTableToStreamBuffer<T>(
+  table: ArrowTable,
+  embeddings?: EmbeddingFunction<T>,
+  schema?: Schema,
+): Promise<Buffer> {
+  const tableWithEmbeddings = await applyEmbeddings(table, embeddings, schema);
+  const writer = RecordBatchStreamWriter.writeAll(tableWithEmbeddings);
+  return Buffer.from(await writer.toUint8Array());
+}
+
+/**
+ * Reorder the columns in `batch` so that they agree with the field order in `schema`
+ */
+function alignBatch(batch: RecordBatch, schema: Schema): RecordBatch {
+  const alignedChildren = [];
+  for (const field of schema.fields) {
+    const indexInBatch = batch.schema.fields?.findIndex(
+      (f) => f.name === field.name,
+    );
+    if (indexInBatch < 0) {
+      throw new Error(
+        `The column ${field.name} was not found in the Arrow Table`,
+      );
+    }
+    alignedChildren.push(batch.data.children[indexInBatch]);
+  }
+  const newData = makeData({
+    type: new Struct(schema.fields),
+    length: batch.numRows,
+    nullCount: batch.nullCount,
+    children: alignedChildren,
+  });
+  return new RecordBatch(schema, newData);
+}
+
+/**
+ * Reorder the columns in `table` so that they agree with the field order in `schema`
+ */
+function alignTable(table: ArrowTable, schema: Schema): ArrowTable {
+  const alignedBatches = table.batches.map((batch) =>
+    alignBatch(batch, schema),
+  );
+  return new ArrowTable(schema, alignedBatches);
+}
+
+/**
+ * Create an empty table with the given schema
+ */
+export function createEmptyTable(schema: Schema): ArrowTable {
+  return new ArrowTable(sanitizeSchema(schema));
+}

nodejs/lancedb/connection.ts

@@ -0,0 +1,176 @@
+// Copyright 2024 Lance Developers.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import { fromTableToBuffer, makeArrowTable, makeEmptyTable } from "./arrow";
+import { Connection as LanceDbConnection } from "./native";
+import { Table } from "./table";
+import { Table as ArrowTable, Schema } from "apache-arrow";
+
+export interface CreateTableOptions {
+  /**
+   * The mode to use when creating the table.
+   *
+   * If this is set to "create" and the table already exists then either
+   * an error will be thrown or, if existOk is true, then nothing will
+   * happen.  Any provided data will be ignored.
+   *
+   * If this is set to "overwrite" then any existing table will be replaced.
+   */
+  mode: "create" | "overwrite";
+  /**
+   * If this is true and the table already exists and the mode is "create"
+   * then no error will be raised.
+   */
+  existOk: boolean;
+}
+
+export interface TableNamesOptions {
+  /**
+   * If present, only return names that come lexicographically after the
+   * supplied value.
+   *
+   * This can be combined with limit to implement pagination by setting this to
+   * the last table name from the previous page.
+   */
+  startAfter?: string;
+  /** An optional limit to the number of results to return. */
+  limit?: number;
+}
+
+/**
+ * A LanceDB Connection that allows you to open tables and create new ones.
+ *
+ * Connection could be local against filesystem or remote against a server.
+ *
+ * A Connection is intended to be a long lived object and may hold open
+ * resources such as HTTP connection pools.  This is generally fine and
+ * a single connection should be shared if it is going to be used many
+ * times. However, if you are finished with a connection, you may call
+ * close to eagerly free these resources.  Any call to a Connection
+ * method after it has been closed will result in an error.
+ *
+ * Closing a connection is optional.  Connections will automatically
+ * be closed when they are garbage collected.
+ *
+ * Any created tables are independent and will continue to work even if
+ * the underlying connection has been closed.
+ */
+export class Connection {
+  readonly inner: LanceDbConnection;
+
+  constructor(inner: LanceDbConnection) {
+    this.inner = inner;
+  }
+
+  /** Return true if the connection has not been closed */
+  isOpen(): boolean {
+    return this.inner.isOpen();
+  }
+
+  /**
+   * Close the connection, releasing any underlying resources.
+   *
+   * It is safe to call this method multiple times.
+   *
+   * Any attempt to use the connection after it is closed will result in an error.
+   */
+  close(): void {
+    this.inner.close();
+  }
+
+  /** Return a brief description of the connection */
+  display(): string {
+    return this.inner.display();
+  }
+
+  /**
+   * List all the table names in this database.
+   *
+   * Tables will be returned in lexicographical order.
+   * @param {Partial<TableNamesOptions>} options - options to control the
+   * paging / start point
+   */
+  async tableNames(options?: Partial<TableNamesOptions>): Promise<string[]> {
+    return this.inner.tableNames(options?.startAfter, options?.limit);
+  }
+
+  /**
+   * Open a table in the database.
+   * @param {string} name - The name of the table
+   */
+  async openTable(name: string): Promise<Table> {
+    const innerTable = await this.inner.openTable(name);
+    return new Table(innerTable);
+  }
+
+  /**
+   * Creates a new Table and initialize it with new data.
+   * @param {string} name - The name of the table.
+   * @param {Record<string, unknown>[] | ArrowTable} data - Non-empty Array of Records
+   * to be inserted into the table
+   */
+  async createTable(
+    name: string,
+    data: Record<string, unknown>[] | ArrowTable,
+    options?: Partial<CreateTableOptions>,
+  ): Promise<Table> {
+    let mode: string = options?.mode ?? "create";
+    const existOk = options?.existOk ?? false;
+
+    if (mode === "create" && existOk) {
+      mode = "exist_ok";
+    }
+
+    let table: ArrowTable;
+    if (data instanceof ArrowTable) {
+      table = data;
+    } else {
+      table = makeArrowTable(data);
+    }
+    const buf = await fromTableToBuffer(table);
+    const innerTable = await this.inner.createTable(name, buf, mode);
+    return new Table(innerTable);
+  }
+
+  /**
+   * Creates a new empty Table
+   * @param {string} name - The name of the table.
+   * @param {Schema} schema - The schema of the table
+   */
+  async createEmptyTable(
+    name: string,
+    schema: Schema,
+    options?: Partial<CreateTableOptions>,
+  ): Promise<Table> {
+    let mode: string = options?.mode ?? "create";
+    const existOk = options?.existOk ?? false;
+
+    if (mode === "create" && existOk) {
+      mode = "exist_ok";
+    }
+
+    const table = makeEmptyTable(schema);
+    const buf = await fromTableToBuffer(table);
+    const innerTable = await this.inner.createEmptyTable(name, buf, mode);
+    return new Table(innerTable);
+  }
+
+  /**
+   * Drop an existing table.
+   * @param {string} name The name of the table to drop.
+   */
+  async dropTable(name: string): Promise<void> {
+    return this.inner.dropTable(name);
+  }
+}

nodejs/lancedb/embedding/embedding_function.ts

@@ -0,0 +1,78 @@
+// Copyright 2023 Lance Developers.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import { type Float } from "apache-arrow";
+
+/**
+ * An embedding function that automatically creates vector representation for a given column.
+ */
+export interface EmbeddingFunction<T> {
+  /**
+   * The name of the column that will be used as input for the Embedding Function.
+   */
+  sourceColumn: string;
+
+  /**
+   * The data type of the embedding
+   *
+   * The embedding function should return `number`.  This will be converted into
+   * an Arrow float array.  By default this will be Float32 but this property can
+   * be used to control the conversion.
+   */
+  embeddingDataType?: Float;
+
+  /**
+   * The dimension of the embedding
+   *
+   * This is optional, normally this can be determined by looking at the results of
+   * `embed`.  If this is not specified, and there is an attempt to apply the embedding
+   * to an empty table, then that process will fail.
+   */
+  embeddingDimension?: number;
+
+  /**
+   * The name of the column that will contain the embedding
+   *
+   * By default this is "vector"
+   */
+  destColumn?: string;
+
+  /**
+   * Should the source column be excluded from the resulting table
+   *
+   * By default the source column is included.  Set this to true and
+   * only the embedding will be stored.
+   */
+  excludeSource?: boolean;
+
+  /**
+   * Creates a vector representation for the given values.
+   */
+  embed: (data: T[]) => Promise<number[][]>;
+}
+
+/** Test if the input seems to be an embedding function */
+export function isEmbeddingFunction<T>(
+  value: unknown,
+): value is EmbeddingFunction<T> {
+  if (typeof value !== "object" || value === null) {
+    return false;
+  }
+  if (!("sourceColumn" in value) || !("embed" in value)) {
+    return false;
+  }
+  return (
+    typeof value.sourceColumn === "string" && typeof value.embed === "function"
+  );
+}

nodejs/lancedb/embedding/openai.ts

@@ -0,0 +1,62 @@
+// Copyright 2023 Lance Developers.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import { type EmbeddingFunction } from "./embedding_function";
+import type OpenAI from "openai";
+
+export class OpenAIEmbeddingFunction implements EmbeddingFunction<string> {
+  private readonly _openai: OpenAI;
+  private readonly _modelName: string;
+
+  constructor(
+    sourceColumn: string,
+    openAIKey: string,
+    modelName: string = "text-embedding-ada-002",
+  ) {
+    /**
+     * @type {import("openai").default}
+     */
+    // eslint-disable-next-line @typescript-eslint/naming-convention
+    let Openai;
+    try {
+      // eslint-disable-next-line @typescript-eslint/no-var-requires
+      Openai = require("openai");
+    } catch {
+      throw new Error("please install openai@^4.24.1 using npm install openai");
+    }
+
+    this.sourceColumn = sourceColumn;
+    const configuration = {
+      apiKey: openAIKey,
+    };
+
+    this._openai = new Openai(configuration);
+    this._modelName = modelName;
+  }
+
+  async embed(data: string[]): Promise<number[][]> {
+    const response = await this._openai.embeddings.create({
+      model: this._modelName,
+      input: data,
+    });
+
+    const embeddings: number[][] = [];
+    for (let i = 0; i < response.data.length; i++) {
+      embeddings.push(response.data[i].embedding);
+    }
+    return embeddings;
+  }
+
+  sourceColumn: string;
+}

nodejs/lancedb/index.ts

@@ -13,18 +13,14 @@
 // limitations under the License.
 
 import { Connection } from "./connection";
-import { Connection as NativeConnection, ConnectionOptions } from "./native.js";
+import {
-
+  Connection as LanceDbConnection,
-export {
   ConnectionOptions,
-  WriteOptions,
-  Query,
-  MetricType,
 } from "./native.js";
-export { Connection } from "./connection";
+
-export { Table } from "./table";
+export { ConnectionOptions, WriteOptions, Query } from "./native.js";
-export { Data } from "./arrow";
+export { Connection, CreateTableOptions } from "./connection";
-export { IvfPQOptions, IndexBuilder } from "./indexer";
+export { Table, AddDataOptions } from "./table";
 
 /**
  * Connect to a LanceDB instance at the given URI.
@@ -34,31 +30,15 @@ export { IvfPQOptions, IndexBuilder } from "./indexer";
  * - `/path/to/database` - local database
  * - `s3://bucket/path/to/database` or `gs://bucket/path/to/database` - database on cloud storage
  * - `db://host:port` - remote database (LanceDB cloud)
- *
+ * @param {string} uri - The uri of the database. If the database uri starts
- * @param uri The uri of the database. If the database uri starts with `db://` then it connects to a remote database.
+ * with `db://` then it connects to a remote database.
- *
  * @see {@link ConnectionOptions} for more details on the URI format.
  */
-export async function connect(uri: string): Promise<Connection>;
 export async function connect(
-  opts: Partial<ConnectionOptions>
+  uri: string,
-): Promise<Connection>;
+  opts?: Partial<ConnectionOptions>,
-export async function connect(
-  args: string | Partial<ConnectionOptions>
 ): Promise<Connection> {
-  let opts: ConnectionOptions;
+  opts = opts ?? {};
-  if (typeof args === "string") {
+  const nativeConn = await LanceDbConnection.new(uri, opts);
-    opts = { uri: args };
-  } else {
-    opts = Object.assign(
-      {
-        uri: "",
-        apiKey: "",
-        hostOverride: "",
-      },
-      args
-    );
-  }
-  const nativeConn = await NativeConnection.new(opts.uri);
   return new Connection(nativeConn);
 }

nodejs/lancedb/indices.ts

@@ -0,0 +1,203 @@
+// Copyright 2024 Lance Developers.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import { Index as LanceDbIndex } from "./native";
+
+/**
+ * Options to create an `IVF_PQ` index
+ */
+export interface IvfPqOptions {
+  /**
+   * The number of IVF partitions to create.
+   *
+   * This value should generally scale with the number of rows in the dataset.
+   * By default the number of partitions is the square root of the number of
+   * rows.
+   *
+   * If this value is too large then the first part of the search (picking the
+   * right partition) will be slow.  If this value is too small then the second
+   * part of the search (searching within a partition) will be slow.
+   */
+  numPartitions?: number;
+
+  /**
+   * Number of sub-vectors of PQ.
+   *
+   * This value controls how much the vector is compressed during the quantization step.
+   * The more sub vectors there are the less the vector is compressed.  The default is
+   * the dimension of the vector divided by 16.  If the dimension is not evenly divisible
+   * by 16 we use the dimension divded by 8.
+   *
+   * The above two cases are highly preferred.  Having 8 or 16 values per subvector allows
+   * us to use efficient SIMD instructions.
+   *
+   * If the dimension is not visible by 8 then we use 1 subvector.  This is not ideal and
+   * will likely result in poor performance.
+   */
+  numSubVectors?: number;
+
+  /**
+   * Distance type to use to build the index.
+   *
+   * Default value is "l2".
+   *
+   * This is used when training the index to calculate the IVF partitions
+   * (vectors are grouped in partitions with similar vectors according to this
+   * distance type) and to calculate a subvector's code during quantization.
+   *
+   * The distance type used to train an index MUST match the distance type used
+   * to search the index.  Failure to do so will yield inaccurate results.
+   *
+   * The following distance types are available:
+   *
+   * "l2" - Euclidean distance. This is a very common distance metric that
+   * accounts for both magnitude and direction when determining the distance
+   * between vectors. L2 distance has a range of [0, ∞).
+   *
+   * "cosine" - Cosine distance.  Cosine distance is a distance metric
+   * calculated from the cosine similarity between two vectors. Cosine
+   * similarity is a measure of similarity between two non-zero vectors of an
+   * inner product space. It is defined to equal the cosine of the angle
+   * between them.  Unlike L2, the cosine distance is not affected by the
+   * magnitude of the vectors.  Cosine distance has a range of [0, 2].
+   *
+   * Note: the cosine distance is undefined when one (or both) of the vectors
+   * are all zeros (there is no direction).  These vectors are invalid and may
+   * never be returned from a vector search.
+   *
+   * "dot" - Dot product. Dot distance is the dot product of two vectors. Dot
+   * distance has a range of (-∞, ∞). If the vectors are normalized (i.e. their
+   * L2 norm is 1), then dot distance is equivalent to the cosine distance.
+   */
+  distanceType?: "l2" | "cosine" | "dot";
+
+  /**
+   * Max iteration to train IVF kmeans.
+   *
+   * When training an IVF PQ index we use kmeans to calculate the partitions.  This parameter
+   * controls how many iterations of kmeans to run.
+   *
+   * Increasing this might improve the quality of the index but in most cases these extra
+   * iterations have diminishing returns.
+   *
+   * The default value is 50.
+   */
+  maxIterations?: number;
+
+  /**
+   * The number of vectors, per partition, to sample when training IVF kmeans.
+   *
+   * When an IVF PQ index is trained, we need to calculate partitions.  These are groups
+   * of vectors that are similar to each other.  To do this we use an algorithm called kmeans.
+   *
+   * Running kmeans on a large dataset can be slow.  To speed this up we run kmeans on a
+   * random sample of the data.  This parameter controls the size of the sample.  The total
+   * number of vectors used to train the index is `sample_rate * num_partitions`.
+   *
+   * Increasing this value might improve the quality of the index but in most cases the
+   * default should be sufficient.
+   *
+   * The default value is 256.
+   */
+  sampleRate?: number;
+}
+
+export class Index {
+  private readonly inner: LanceDbIndex;
+  private constructor(inner: LanceDbIndex) {
+    this.inner = inner;
+  }
+
+  /**
+   * Create an IvfPq index
+   *
+   * This index stores a compressed (quantized) copy of every vector.  These vectors
+   * are grouped into partitions of similar vectors.  Each partition keeps track of
+   * a centroid which is the average value of all vectors in the group.
+   *
+   * During a query the centroids are compared with the query vector to find the closest
+   * partitions.  The compressed vectors in these partitions are then searched to find
+   * the closest vectors.
+   *
+   * The compression scheme is called product quantization.  Each vector is divided into
+   * subvectors and then each subvector is quantized into a small number of bits.  the
+   * parameters `num_bits` and `num_subvectors` control this process, providing a tradeoff
+   * between index size (and thus search speed) and index accuracy.
+   *
+   * The partitioning process is called IVF and the `num_partitions` parameter controls how
+   * many groups to create.
+   *
+   * Note that training an IVF PQ index on a large dataset is a slow operation and
+   * currently is also a memory intensive operation.
+   */
+  static ivfPq(options?: Partial<IvfPqOptions>) {
+    return new Index(
+      LanceDbIndex.ivfPq(
+        options?.distanceType,
+        options?.numPartitions,
+        options?.numSubVectors,
+        options?.maxIterations,
+        options?.sampleRate,
+      ),
+    );
+  }
+
+  /**
+   * Create a btree index
+   *
+   * A btree index is an index on a scalar columns.  The index stores a copy of the column
+   * in sorted order.  A header entry is created for each block of rows (currently the
+   * block size is fixed at 4096).  These header entries are stored in a separate
+   * cacheable structure (a btree).  To search for data the header is used to determine
+   * which blocks need to be read from disk.
+   *
+   * For example, a btree index in a table with 1Bi rows requires sizeof(Scalar) * 256Ki
+   * bytes of memory and will generally need to read sizeof(Scalar) * 4096 bytes to find
+   * the correct row ids.
+   *
+   * This index is good for scalar columns with mostly distinct values and does best when
+   * the query is highly selective.
+   *
+   * The btree index does not currently have any parameters though parameters such as the
+   * block size may be added in the future.
+   */
+  static btree() {
+    return new Index(LanceDbIndex.btree());
+  }
+}
+
+export interface IndexOptions {
+  /**
+   * Advanced index configuration
+   *
+   * This option allows you to specify a specfic index to create and also
+   * allows you to pass in configuration for training the index.
+   *
+   * See the static methods on Index for details on the various index types.
+   *
+   * If this is not supplied then column data type(s) and column statistics
+   * will be used to determine the most useful kind of index to create.
+   */
+  config?: Index;
+  /**
+   * Whether to replace the existing index
+   *
+   * If this is false, and another index already exists on the same columns
+   * and the same name, then an error will be returned.  This is true even if
+   * that index is out of date.
+   *
+   * The default is true
+   */
+  replace?: boolean;
+}

nodejs/lancedb/native.d.ts

@@ -0,0 +1,147 @@
+/* tslint:disable */
+/* eslint-disable */
+
+/* auto-generated by NAPI-RS */
+
+/** A description of an index currently configured on a column */
+export interface IndexConfig {
+  /** The type of the index */
+  indexType: string
+  /**
+   * The columns in the index
+   *
+   * Currently this is always an array of size 1.  In the future there may
+   * be more columns to represent composite indices.
+   */
+  columns: Array<string>
+}
+/**
+ *  A definition of a column alteration. The alteration changes the column at
+ * `path` to have the new name `name`, to be nullable if `nullable` is true,
+ * and to have the data type `data_type`. At least one of `rename` or `nullable`
+ * must be provided.
+ */
+export interface ColumnAlteration {
+  /**
+   * The path to the column to alter. This is a dot-separated path to the column.
+   * If it is a top-level column then it is just the name of the column. If it is
+   * a nested column then it is the path to the column, e.g. "a.b.c" for a column
+   * `c` nested inside a column `b` nested inside a column `a`.
+   */
+  path: string
+  /**
+   * The new name of the column. If not provided then the name will not be changed.
+   * This must be distinct from the names of all other columns in the table.
+   */
+  rename?: string
+  /** Set the new nullability. Note that a nullable column cannot be made non-nullable. */
+  nullable?: boolean
+}
+/** A definition of a new column to add to a table. */
+export interface AddColumnsSql {
+  /** The name of the new column. */
+  name: string
+  /**
+   * The values to populate the new column with, as a SQL expression.
+   * The expression can reference other columns in the table.
+   */
+  valueSql: string
+}
+export interface ConnectionOptions {
+  apiKey?: string
+  hostOverride?: string
+  /**
+   * (For LanceDB OSS only): The interval, in seconds, at which to check for
+   * updates to the table from other processes. If None, then consistency is not
+   * checked. For performance reasons, this is the default. For strong
+   * consistency, set this to zero seconds. Then every read will check for
+   * updates from other processes. As a compromise, you can set this to a
+   * non-zero value for eventual consistency. If more than that interval
+   * has passed since the last check, then the table will be checked for updates.
+   * Note: this consistency only applies to read operations. Write operations are
+   * always consistent.
+   */
+  readConsistencyInterval?: number
+}
+/** Write mode for writing a table. */
+export const enum WriteMode {
+  Create = 'Create',
+  Append = 'Append',
+  Overwrite = 'Overwrite'
+}
+/** Write options when creating a Table. */
+export interface WriteOptions {
+  mode?: WriteMode
+}
+export function connect(uri: string, options: ConnectionOptions): Promise<Connection>
+export class Connection {
+  /** Create a new Connection instance from the given URI. */
+  static new(uri: string, options: ConnectionOptions): Promise<Connection>
+  display(): string
+  isOpen(): boolean
+  close(): void
+  /** List all tables in the dataset. */
+  tableNames(startAfter?: string | undefined | null, limit?: number | undefined | null): Promise<Array<string>>
+  /**
+   * Create table from a Apache Arrow IPC (file) buffer.
+   *
+   * Parameters:
+   * - name: The name of the table.
+   * - buf: The buffer containing the IPC file.
+   *
+   */
+  createTable(name: string, buf: Buffer, mode: string): Promise<Table>
+  createEmptyTable(name: string, schemaBuf: Buffer, mode: string): Promise<Table>
+  openTable(name: string): Promise<Table>
+  /** Drop table with the name. Or raise an error if the table does not exist. */
+  dropTable(name: string): Promise<void>
+}
+export class Index {
+  static ivfPq(distanceType?: string | undefined | null, numPartitions?: number | undefined | null, numSubVectors?: number | undefined | null, maxIterations?: number | undefined | null, sampleRate?: number | undefined | null): Index
+  static btree(): Index
+}
+/** Typescript-style Async Iterator over RecordBatches  */
+export class RecordBatchIterator {
+  next(): Promise<Buffer | null>
+}
+export class Query {
+  onlyIf(predicate: string): void
+  select(columns: Array<[string, string]>): void
+  limit(limit: number): void
+  nearestTo(vector: Float32Array): VectorQuery
+  execute(): Promise<RecordBatchIterator>
+}
+export class VectorQuery {
+  column(column: string): void
+  distanceType(distanceType: string): void
+  postfilter(): void
+  refineFactor(refineFactor: number): void
+  nprobes(nprobe: number): void
+  bypassVectorIndex(): void
+  onlyIf(predicate: string): void
+  select(columns: Array<[string, string]>): void
+  limit(limit: number): void
+  execute(): Promise<RecordBatchIterator>
+}
+export class Table {
+  display(): string
+  isOpen(): boolean
+  close(): void
+  /** Return Schema as empty Arrow IPC file. */
+  schema(): Promise<Buffer>
+  add(buf: Buffer, mode: string): Promise<void>
+  countRows(filter?: string | undefined | null): Promise<number>
+  delete(predicate: string): Promise<void>
+  createIndex(index: Index | undefined | null, column: string, replace?: boolean | undefined | null): Promise<void>
+  update(onlyIf: string | undefined | null, columns: Array<[string, string]>): Promise<void>
+  query(): Query
+  vectorSearch(vector: Float32Array): VectorQuery
+  addColumns(transforms: Array<AddColumnsSql>): Promise<void>
+  alterColumns(alterations: Array<ColumnAlteration>): Promise<void>
+  dropColumns(columns: Array<string>): Promise<void>
+  version(): Promise<number>
+  checkout(version: number): Promise<void>
+  checkoutLatest(): Promise<void>
+  restore(): Promise<void>
+  listIndices(): Promise<Array<IndexConfig>>
+}

nodejs/lancedb/native.js

@@ -0,0 +1,329 @@
+/* tslint:disable */
+/* eslint-disable */
+/* prettier-ignore */
+
+/* auto-generated by NAPI-RS */
+
+const { existsSync, readFileSync } = require('fs')
+const { join } = require("path");
+
+const { platform, arch } = process;
+
+let nativeBinding = null;
+let localFileExisted = false;
+let loadError = null;
+
+function isMusl() {
+  // For Node 10
+  if (!process.report || typeof process.report.getReport !== "function") {
+    try {
+      const lddPath = require("child_process")
+        .execSync("which ldd")
+        .toString()
+        .trim();
+      return readFileSync(lddPath, "utf8").includes("musl");
+    } catch (e) {
+      return true;
+    }
+  } else {
+    const { glibcVersionRuntime } = process.report.getReport().header;
+    return !glibcVersionRuntime;
+  }
+}
+
+switch (platform) {
+  case "android":
+    switch (arch) {
+      case "arm64":
+        localFileExisted = existsSync(
+          join(__dirname, "lancedb-nodejs.android-arm64.node"),
+        );
+        try {
+          if (localFileExisted) {
+            nativeBinding = require("./lancedb-nodejs.android-arm64.node");
+          } else {
+            nativeBinding = require("lancedb-android-arm64");
+          }
+        } catch (e) {
+          loadError = e;
+        }
+        break;
+      case "arm":
+        localFileExisted = existsSync(
+          join(__dirname, "lancedb-nodejs.android-arm-eabi.node"),
+        );
+        try {
+          if (localFileExisted) {
+            nativeBinding = require("./lancedb-nodejs.android-arm-eabi.node");
+          } else {
+            nativeBinding = require("lancedb-android-arm-eabi");
+          }
+        } catch (e) {
+          loadError = e;
+        }
+        break;
+      default:
+        throw new Error(`Unsupported architecture on Android ${arch}`);
+    }
+    break;
+  case "win32":
+    switch (arch) {
+      case "x64":
+        localFileExisted = existsSync(
+          join(__dirname, "lancedb-nodejs.win32-x64-msvc.node"),
+        );
+        try {
+          if (localFileExisted) {
+            nativeBinding = require("./lancedb-nodejs.win32-x64-msvc.node");
+          } else {
+            nativeBinding = require("lancedb-win32-x64-msvc");
+          }
+        } catch (e) {
+          loadError = e;
+        }
+        break;
+      case "ia32":
+        localFileExisted = existsSync(
+          join(__dirname, "lancedb-nodejs.win32-ia32-msvc.node"),
+        );
+        try {
+          if (localFileExisted) {
+            nativeBinding = require("./lancedb-nodejs.win32-ia32-msvc.node");
+          } else {
+            nativeBinding = require("lancedb-win32-ia32-msvc");
+          }
+        } catch (e) {
+          loadError = e;
+        }
+        break;
+      case "arm64":
+        localFileExisted = existsSync(
+          join(__dirname, "lancedb-nodejs.win32-arm64-msvc.node"),
+        );
+        try {
+          if (localFileExisted) {
+            nativeBinding = require("./lancedb-nodejs.win32-arm64-msvc.node");
+          } else {
+            nativeBinding = require("lancedb-win32-arm64-msvc");
+          }
+        } catch (e) {
+          loadError = e;
+        }
+        break;
+      default:
+        throw new Error(`Unsupported architecture on Windows: ${arch}`);
+    }
+    break;
+  case "darwin":
+    localFileExisted = existsSync(
+      join(__dirname, "lancedb-nodejs.darwin-universal.node"),
+    );
+    try {
+      if (localFileExisted) {
+        nativeBinding = require("./lancedb-nodejs.darwin-universal.node");
+      } else {
+        nativeBinding = require("lancedb-darwin-universal");
+      }
+      break;
+    } catch {}
+    switch (arch) {
+      case "x64":
+        localFileExisted = existsSync(
+          join(__dirname, "lancedb-nodejs.darwin-x64.node"),
+        );
+        try {
+          if (localFileExisted) {
+            nativeBinding = require("./lancedb-nodejs.darwin-x64.node");
+          } else {
+            nativeBinding = require("lancedb-darwin-x64");
+          }
+        } catch (e) {
+          loadError = e;
+        }
+        break;
+      case "arm64":
+        localFileExisted = existsSync(
+          join(__dirname, "lancedb-nodejs.darwin-arm64.node"),
+        );
+        try {
+          if (localFileExisted) {
+            nativeBinding = require("./lancedb-nodejs.darwin-arm64.node");
+          } else {
+            nativeBinding = require("lancedb-darwin-arm64");
+          }
+        } catch (e) {
+          loadError = e;
+        }
+        break;
+      default:
+        throw new Error(`Unsupported architecture on macOS: ${arch}`);
+    }
+    break;
+  case "freebsd":
+    if (arch !== "x64") {
+      throw new Error(`Unsupported architecture on FreeBSD: ${arch}`);
+    }
+    localFileExisted = existsSync(
+      join(__dirname, "lancedb-nodejs.freebsd-x64.node"),
+    );
+    try {
+      if (localFileExisted) {
+        nativeBinding = require("./lancedb-nodejs.freebsd-x64.node");
+      } else {
+        nativeBinding = require("lancedb-freebsd-x64");
+      }
+    } catch (e) {
+      loadError = e;
+    }
+    break;
+  case "linux":
+    switch (arch) {
+      case "x64":
+        if (isMusl()) {
+          localFileExisted = existsSync(
+            join(__dirname, "lancedb-nodejs.linux-x64-musl.node"),
+          );
+          try {
+            if (localFileExisted) {
+              nativeBinding = require("./lancedb-nodejs.linux-x64-musl.node");
+            } else {
+              nativeBinding = require("lancedb-linux-x64-musl");
+            }
+          } catch (e) {
+            loadError = e;
+          }
+        } else {
+          localFileExisted = existsSync(
+            join(__dirname, "lancedb-nodejs.linux-x64-gnu.node"),
+          );
+          try {
+            if (localFileExisted) {
+              nativeBinding = require("./lancedb-nodejs.linux-x64-gnu.node");
+            } else {
+              nativeBinding = require("lancedb-linux-x64-gnu");
+            }
+          } catch (e) {
+            loadError = e;
+          }
+        }
+        break;
+      case "arm64":
+        if (isMusl()) {
+          localFileExisted = existsSync(
+            join(__dirname, "lancedb-nodejs.linux-arm64-musl.node"),
+          );
+          try {
+            if (localFileExisted) {
+              nativeBinding = require("./lancedb-nodejs.linux-arm64-musl.node");
+            } else {
+              nativeBinding = require("lancedb-linux-arm64-musl");
+            }
+          } catch (e) {
+            loadError = e;
+          }
+        } else {
+          localFileExisted = existsSync(
+            join(__dirname, "lancedb-nodejs.linux-arm64-gnu.node"),
+          );
+          try {
+            if (localFileExisted) {
+              nativeBinding = require("./lancedb-nodejs.linux-arm64-gnu.node");
+            } else {
+              nativeBinding = require("lancedb-linux-arm64-gnu");
+            }
+          } catch (e) {
+            loadError = e;
+          }
+        }
+        break;
+      case "arm":
+        localFileExisted = existsSync(
+          join(__dirname, "lancedb-nodejs.linux-arm-gnueabihf.node"),
+        );
+        try {
+          if (localFileExisted) {
+            nativeBinding = require("./lancedb-nodejs.linux-arm-gnueabihf.node");
+          } else {
+            nativeBinding = require("lancedb-linux-arm-gnueabihf");
+          }
+        } catch (e) {
+          loadError = e;
+        }
+        break;
+      case "riscv64":
+        if (isMusl()) {
+          localFileExisted = existsSync(
+            join(__dirname, "lancedb-nodejs.linux-riscv64-musl.node"),
+          );
+          try {
+            if (localFileExisted) {
+              nativeBinding = require("./lancedb-nodejs.linux-riscv64-musl.node");
+            } else {
+              nativeBinding = require("lancedb-linux-riscv64-musl");
+            }
+          } catch (e) {
+            loadError = e;
+          }
+        } else {
+          localFileExisted = existsSync(
+            join(__dirname, "lancedb-nodejs.linux-riscv64-gnu.node"),
+          );
+          try {
+            if (localFileExisted) {
+              nativeBinding = require("./lancedb-nodejs.linux-riscv64-gnu.node");
+            } else {
+              nativeBinding = require("lancedb-linux-riscv64-gnu");
+            }
+          } catch (e) {
+            loadError = e;
+          }
+        }
+        break;
+      case "s390x":
+        localFileExisted = existsSync(
+          join(__dirname, "lancedb-nodejs.linux-s390x-gnu.node"),
+        );
+        try {
+          if (localFileExisted) {
+            nativeBinding = require("./lancedb-nodejs.linux-s390x-gnu.node");
+          } else {
+            nativeBinding = require("lancedb-linux-s390x-gnu");
+          }
+        } catch (e) {
+          loadError = e;
+        }
+        break;
+      default:
+        throw new Error(`Unsupported architecture on Linux: ${arch}`);
+    }
+    break;
+  default:
+    throw new Error(`Unsupported OS: ${platform}, architecture: ${arch}`);
+}
+
+if (!nativeBinding) {
+  if (loadError) {
+    throw loadError;
+  }
+  throw new Error(`Failed to load native binding`);
+}
+
+const {
+  Connection,
+  Index,
+  RecordBatchIterator,
+  Query,
+  VectorQuery,
+  Table,
+  WriteMode,
+  connect,
+} = nativeBinding;
+
+module.exports.Connection = Connection;
+module.exports.Index = Index;
+module.exports.RecordBatchIterator = RecordBatchIterator;
+module.exports.Query = Query;
+module.exports.VectorQuery = VectorQuery;
+module.exports.Table = Table;
+module.exports.WriteMode = WriteMode;
+module.exports.connect = connect;

nodejs/lancedb/query.ts

@@ -0,0 +1,375 @@
+// Copyright 2024 Lance Developers.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import { RecordBatch, tableFromIPC, Table as ArrowTable } from "apache-arrow";
+import {
+  RecordBatchIterator as NativeBatchIterator,
+  Query as NativeQuery,
+  Table as NativeTable,
+  VectorQuery as NativeVectorQuery,
+} from "./native";
+import { type IvfPqOptions } from "./indices";
+class RecordBatchIterator implements AsyncIterator<RecordBatch> {
+  private promisedInner?: Promise<NativeBatchIterator>;
+  private inner?: NativeBatchIterator;
+
+  constructor(promise?: Promise<NativeBatchIterator>) {
+    // TODO: check promise reliably so we dont need to pass two arguments.
+    this.promisedInner = promise;
+  }
+
+  // eslint-disable-next-line @typescript-eslint/no-explicit-any
+  async next(): Promise<IteratorResult<RecordBatch<any>>> {
+    if (this.inner === undefined) {
+      this.inner = await this.promisedInner;
+    }
+    if (this.inner === undefined) {
+      throw new Error("Invalid iterator state state");
+    }
+    const n = await this.inner.next();
+    if (n == null) {
+      return Promise.resolve({ done: true, value: null });
+    }
+    const tbl = tableFromIPC(n);
+    if (tbl.batches.length != 1) {
+      throw new Error("Expected only one batch");
+    }
+    return Promise.resolve({ done: false, value: tbl.batches[0] });
+  }
+}
+/* eslint-enable */
+
+/** Common methods supported by all query types */
+export class QueryBase<
+  NativeQueryType extends NativeQuery | NativeVectorQuery,
+  QueryType,
+> implements AsyncIterable<RecordBatch>
+{
+  protected constructor(protected inner: NativeQueryType) {}
+
+  /**
+   * A filter statement to be applied to this query.
+   *
+   * The filter should be supplied as an SQL query string.  For example:
+   * @example
+   * x > 10
+   * y > 0 AND y < 100
+   * x > 5 OR y = 'test'
+   *
+   * Filtering performance can often be improved by creating a scalar index
+   * on the filter column(s).
+   */
+  where(predicate: string): QueryType {
+    this.inner.onlyIf(predicate);
+    return this as unknown as QueryType;
+  }
+
+  /**
+   * Return only the specified columns.
+   *
+   * By default a query will return all columns from the table.  However, this can have
+   * a very significant impact on latency.  LanceDb stores data in a columnar fashion.  This
+   * means we can finely tune our I/O to select exactly the columns we need.
+   *
+   * As a best practice you should always limit queries to the columns that you need.  If you
+   * pass in an array of column names then only those columns will be returned.
+   *
+   * You can also use this method to create new "dynamic" columns based on your existing columns.
+   * For example, you may not care about "a" or "b" but instead simply want "a + b".  This is often
+   * seen in the SELECT clause of an SQL query (e.g. `SELECT a+b FROM my_table`).
+   *
+   * To create dynamic columns you can pass in a Map<string, string>.  A column will be returned
+   * for each entry in the map.  The key provides the name of the column.  The value is
+   * an SQL string used to specify how the column is calculated.
+   *
+   * For example, an SQL query might state `SELECT a + b AS combined, c`.  The equivalent
+   * input to this method would be:
+   * @example
+   * new Map([["combined", "a + b"], ["c", "c"]])
+   *
+   * Columns will always be returned in the order given, even if that order is different than
+   * the order used when adding the data.
+   *
+   * Note that you can pass in a `Record<string, string>` (e.g. an object literal). This method
+   * uses `Object.entries` which should preserve the insertion order of the object.  However,
+   * object insertion order is easy to get wrong and `Map` is more foolproof.
+   */
+  select(
+    columns: string[] | Map<string, string> | Record<string, string>,
+  ): QueryType {
+    let columnTuples: [string, string][];
+    if (Array.isArray(columns)) {
+      columnTuples = columns.map((c) => [c, c]);
+    } else if (columns instanceof Map) {
+      columnTuples = Array.from(columns.entries());
+    } else {
+      columnTuples = Object.entries(columns);
+    }
+    this.inner.select(columnTuples);
+    return this as unknown as QueryType;
+  }
+
+  /**
+   * Set the maximum number of results to return.
+   *
+   * By default, a plain search has no limit.  If this method is not
+   * called then every valid row from the table will be returned.
+   */
+  limit(limit: number): QueryType {
+    this.inner.limit(limit);
+    return this as unknown as QueryType;
+  }
+
+  protected nativeExecute(): Promise<NativeBatchIterator> {
+    return this.inner.execute();
+  }
+
+  /**
+   * Execute the query and return the results as an @see {@link AsyncIterator}
+   * of @see {@link RecordBatch}.
+   *
+   * By default, LanceDb will use many threads to calculate results and, when
+   * the result set is large, multiple batches will be processed at one time.
+   * This readahead is limited however and backpressure will be applied if this
+   * stream is consumed slowly (this constrains the maximum memory used by a
+   * single query)
+   *
+   */
+  protected execute(): RecordBatchIterator {
+    return new RecordBatchIterator(this.nativeExecute());
+  }
+
+  // eslint-disable-next-line @typescript-eslint/no-explicit-any
+  [Symbol.asyncIterator](): AsyncIterator<RecordBatch<any>> {
+    const promise = this.nativeExecute();
+    return new RecordBatchIterator(promise);
+  }
+
+  /** Collect the results as an Arrow @see {@link ArrowTable}. */
+  async toArrow(): Promise<ArrowTable> {
+    const batches = [];
+    for await (const batch of this) {
+      batches.push(batch);
+    }
+    return new ArrowTable(batches);
+  }
+
+  /** Collect the results as an array of objects. */
+  async toArray(): Promise<unknown[]> {
+    const tbl = await this.toArrow();
+    // eslint-disable-next-line @typescript-eslint/no-unsafe-return
+    return tbl.toArray();
+  }
+}
+
+/**
+ * An interface for a query that can be executed
+ *
+ * Supported by all query types
+ */
+export interface ExecutableQuery {}
+
+/**
+ * A builder used to construct a vector search
+ *
+ * This builder can be reused to execute the query many times.
+ */
+export class VectorQuery extends QueryBase<NativeVectorQuery, VectorQuery> {
+  constructor(inner: NativeVectorQuery) {
+    super(inner);
+  }
+
+  /**
+   * Set the number of partitions to search (probe)
+   *
+   * This argument is only used when the vector column has an IVF PQ index.
+   * If there is no index then this value is ignored.
+   *
+   * The IVF stage of IVF PQ divides the input into partitions (clusters) of
+   * related values.
+   *
+   * The partition whose centroids are closest to the query vector will be
+   * exhaustiely searched to find matches.  This parameter controls how many
+   * partitions should be searched.
+   *
+   * Increasing this value will increase the recall of your query but will
+   * also increase the latency of your query.  The default value is 20.  This
+   * default is good for many cases but the best value to use will depend on
+   * your data and the recall that you need to achieve.
+   *
+   * For best results we recommend tuning this parameter with a benchmark against
+   * your actual data to find the smallest possible value that will still give
+   * you the desired recall.
+   */
+  nprobes(nprobes: number): VectorQuery {
+    this.inner.nprobes(nprobes);
+    return this;
+  }
+
+  /**
+   * Set the vector column to query
+   *
+   * This controls which column is compared to the query vector supplied in
+   * the call to @see {@link Query#nearestTo}
+   *
+   * This parameter must be specified if the table has more than one column
+   * whose data type is a fixed-size-list of floats.
+   */
+  column(column: string): VectorQuery {
+    this.inner.column(column);
+    return this;
+  }
+
+  /**
+   * Set the distance metric to use
+   *
+   * When performing a vector search we try and find the "nearest" vectors according
+   * to some kind of distance metric.  This parameter controls which distance metric to
+   * use.  See @see {@link IvfPqOptions.distanceType} for more details on the different
+   * distance metrics available.
+   *
+   * Note: if there is a vector index then the distance type used MUST match the distance
+   * type used to train the vector index.  If this is not done then the results will be
+   * invalid.
+   *
+   * By default "l2" is used.
+   */
+  distanceType(distanceType: string): VectorQuery {
+    this.inner.distanceType(distanceType);
+    return this;
+  }
+
+  /**
+   * A multiplier to control how many additional rows are taken during the refine step
+   *
+   * This argument is only used when the vector column has an IVF PQ index.
+   * If there is no index then this value is ignored.
+   *
+   * An IVF PQ index stores compressed (quantized) values.  They query vector is compared
+   * against these values and, since they are compressed, the comparison is inaccurate.
+   *
+   * This parameter can be used to refine the results.  It can improve both improve recall
+   * and correct the ordering of the nearest results.
+   *
+   * To refine results LanceDb will first perform an ANN search to find the nearest
+   * `limit` * `refine_factor` results.  In other words, if `refine_factor` is 3 and
+   * `limit` is the default (10) then the first 30 results will be selected.  LanceDb
+   * then fetches the full, uncompressed, values for these 30 results.  The results are
+   * then reordered by the true distance and only the nearest 10 are kept.
+   *
+   * Note: there is a difference between calling this method with a value of 1 and never
+   * calling this method at all.  Calling this method with any value will have an impact
+   * on your search latency.  When you call this method with a `refine_factor` of 1 then
+   * LanceDb still needs to fetch the full, uncompressed, values so that it can potentially
+   * reorder the results.
+   *
+   * Note: if this method is NOT called then the distances returned in the _distance column
+   * will be approximate distances based on the comparison of the quantized query vector
+   * and the quantized result vectors.  This can be considerably different than the true
+   * distance between the query vector and the actual uncompressed vector.
+   */
+  refineFactor(refineFactor: number): VectorQuery {
+    this.inner.refineFactor(refineFactor);
+    return this;
+  }
+
+  /**
+   * If this is called then filtering will happen after the vector search instead of
+   * before.
+   *
+   * By default filtering will be performed before the vector search.  This is how
+   * filtering is typically understood to work.  This prefilter step does add some
+   * additional latency.  Creating a scalar index on the filter column(s) can
+   * often improve this latency.  However, sometimes a filter is too complex or scalar
+   * indices cannot be applied to the column.  In these cases postfiltering can be
+   * used instead of prefiltering to improve latency.
+   *
+   * Post filtering applies the filter to the results of the vector search.  This means
+   * we only run the filter on a much smaller set of data.  However, it can cause the
+   * query to return fewer than `limit` results (or even no results) if none of the nearest
+   * results match the filter.
+   *
+   * Post filtering happens during the "refine stage" (described in more detail in
+   * @see {@link VectorQuery#refineFactor}).  This means that setting a higher refine
+   * factor can often help restore some of the results lost by post filtering.
+   */
+  postfilter(): VectorQuery {
+    this.inner.postfilter();
+    return this;
+  }
+
+  /**
+   * If this is called then any vector index is skipped
+   *
+   * An exhaustive (flat) search will be performed.  The query vector will
+   * be compared to every vector in the table.  At high scales this can be
+   * expensive.  However, this is often still useful.  For example, skipping
+   * the vector index can give you ground truth results which you can use to
+   * calculate your recall to select an appropriate value for nprobes.
+   */
+  bypassVectorIndex(): VectorQuery {
+    this.inner.bypassVectorIndex();
+    return this;
+  }
+}
+
+/** A builder for LanceDB queries. */
+export class Query extends QueryBase<NativeQuery, Query> {
+  constructor(tbl: NativeTable) {
+    super(tbl.query());
+  }
+
+  /**
+   * Find the nearest vectors to the given query vector.
+   *
+   * This converts the query from a plain query to a vector query.
+   *
+   * This method will attempt to convert the input to the query vector
+   * expected by the embedding model.  If the input cannot be converted
+   * then an error will be thrown.
+   *
+   * By default, there is no embedding model, and the input should be
+   * an array-like object of numbers (something that can be used as input
+   * to Float32Array.from)
+   *
+   * If there is only one vector column (a column whose data type is a
+   * fixed size list of floats) then the column does not need to be specified.
+   * If there is more than one vector column you must use
+   * @see {@link VectorQuery#column}  to specify which column you would like
+   * to compare with.
+   *
+   * If no index has been created on the vector column then a vector query
+   * will perform a distance comparison between the query vector and every
+   * vector in the database and then sort the results.  This is sometimes
+   * called a "flat search"
+   *
+   * For small databases, with a few hundred thousand vectors or less, this can
+   * be reasonably fast.  In larger databases you should create a vector index
+   * on the column.  If there is a vector index then an "approximate" nearest
+   * neighbor search (frequently called an ANN search) will be performed.  This
+   * search is much faster, but the results will be approximate.
+   *
+   * The query can be further parameterized using the returned builder.  There
+   * are various ANN search parameters that will let you fine tune your recall
+   * accuracy vs search latency.
+   *
+   * Vector searches always have a `limit`.  If `limit` has not been called then
+   * a default `limit` of 10 will be used.  @see {@link Query#limit}
+   */
+  nearestTo(vector: unknown): VectorQuery {
+    // eslint-disable-next-line @typescript-eslint/no-explicit-any
+    const vectorQuery = this.inner.nearestTo(Float32Array.from(vector as any));
+    return new VectorQuery(vectorQuery);
+  }
+}

nodejs/lancedb/sanitize.ts

@@ -0,0 +1,516 @@
+// Copyright 2023 LanceDB Developers.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// The utilities in this file help sanitize data from the user's arrow
+// library into the types expected by vectordb's arrow library.  Node
+// generally allows for mulitple versions of the same library (and sometimes
+// even multiple copies of the same version) to be installed at the same
+// time.  However, arrow-js uses instanceof which expected that the input
+// comes from the exact same library instance.  This is not always the case
+// and so we must sanitize the input to ensure that it is compatible.
+
+import {
+  Field,
+  Utf8,
+  FixedSizeBinary,
+  FixedSizeList,
+  Schema,
+  List,
+  Struct,
+  Float,
+  Bool,
+  Date_,
+  Decimal,
+  DataType,
+  Dictionary,
+  Binary,
+  Float32,
+  Interval,
+  Map_,
+  Duration,
+  Union,
+  Time,
+  Timestamp,
+  Type,
+  Null,
+  Int,
+  type Precision,
+  type DateUnit,
+  Int8,
+  Int16,
+  Int32,
+  Int64,
+  Uint8,
+  Uint16,
+  Uint32,
+  Uint64,
+  Float16,
+  Float64,
+  DateDay,
+  DateMillisecond,
+  DenseUnion,
+  SparseUnion,
+  TimeNanosecond,
+  TimeMicrosecond,
+  TimeMillisecond,
+  TimeSecond,
+  TimestampNanosecond,
+  TimestampMicrosecond,
+  TimestampMillisecond,
+  TimestampSecond,
+  IntervalDayTime,
+  IntervalYearMonth,
+  DurationNanosecond,
+  DurationMicrosecond,
+  DurationMillisecond,
+  DurationSecond,
+} from "apache-arrow";
+import type { IntBitWidth, TKeys, TimeBitWidth } from "apache-arrow/type";
+
+function sanitizeMetadata(
+  metadataLike?: unknown,
+): Map<string, string> | undefined {
+  if (metadataLike === undefined || metadataLike === null) {
+    return undefined;
+  }
+  if (!(metadataLike instanceof Map)) {
+    throw Error("Expected metadata, if present, to be a Map<string, string>");
+  }
+  for (const item of metadataLike) {
+    if (!(typeof item[0] === "string" || !(typeof item[1] === "string"))) {
+      throw Error(
+        "Expected metadata, if present, to be a Map<string, string> but it had non-string keys or values",
+      );
+    }
+  }
+  return metadataLike as Map<string, string>;
+}
+
+function sanitizeInt(typeLike: object) {
+  if (
+    !("bitWidth" in typeLike) ||
+    typeof typeLike.bitWidth !== "number" ||
+    !("isSigned" in typeLike) ||
+    typeof typeLike.isSigned !== "boolean"
+  ) {
+    throw Error(
+      "Expected an Int Type to have a `bitWidth` and `isSigned` property",
+    );
+  }
+  return new Int(typeLike.isSigned, typeLike.bitWidth as IntBitWidth);
+}
+
+function sanitizeFloat(typeLike: object) {
+  if (!("precision" in typeLike) || typeof typeLike.precision !== "number") {
+    throw Error("Expected a Float Type to have a `precision` property");
+  }
+  return new Float(typeLike.precision as Precision);
+}
+
+function sanitizeDecimal(typeLike: object) {
+  if (
+    !("scale" in typeLike) ||
+    typeof typeLike.scale !== "number" ||
+    !("precision" in typeLike) ||
+    typeof typeLike.precision !== "number" ||
+    !("bitWidth" in typeLike) ||
+    typeof typeLike.bitWidth !== "number"
+  ) {
+    throw Error(
+      "Expected a Decimal Type to have `scale`, `precision`, and `bitWidth` properties",
+    );
+  }
+  return new Decimal(typeLike.scale, typeLike.precision, typeLike.bitWidth);
+}
+
+function sanitizeDate(typeLike: object) {
+  if (!("unit" in typeLike) || typeof typeLike.unit !== "number") {
+    throw Error("Expected a Date type to have a `unit` property");
+  }
+  return new Date_(typeLike.unit as DateUnit);
+}
+
+function sanitizeTime(typeLike: object) {
+  if (
+    !("unit" in typeLike) ||
+    typeof typeLike.unit !== "number" ||
+    !("bitWidth" in typeLike) ||
+    typeof typeLike.bitWidth !== "number"
+  ) {
+    throw Error(
+      "Expected a Time type to have `unit` and `bitWidth` properties",
+    );
+  }
+  return new Time(typeLike.unit, typeLike.bitWidth as TimeBitWidth);
+}
+
+function sanitizeTimestamp(typeLike: object) {
+  if (!("unit" in typeLike) || typeof typeLike.unit !== "number") {
+    throw Error("Expected a Timestamp type to have a `unit` property");
+  }
+  let timezone = null;
+  if ("timezone" in typeLike && typeof typeLike.timezone === "string") {
+    timezone = typeLike.timezone;
+  }
+  return new Timestamp(typeLike.unit, timezone);
+}
+
+function sanitizeTypedTimestamp(
+  typeLike: object,
+  // eslint-disable-next-line @typescript-eslint/naming-convention
+  Datatype:
+    | typeof TimestampNanosecond
+    | typeof TimestampMicrosecond
+    | typeof TimestampMillisecond
+    | typeof TimestampSecond,
+) {
+  let timezone = null;
+  if ("timezone" in typeLike && typeof typeLike.timezone === "string") {
+    timezone = typeLike.timezone;
+  }
+  return new Datatype(timezone);
+}
+
+function sanitizeInterval(typeLike: object) {
+  if (!("unit" in typeLike) || typeof typeLike.unit !== "number") {
+    throw Error("Expected an Interval type to have a `unit` property");
+  }
+  return new Interval(typeLike.unit);
+}
+
+function sanitizeList(typeLike: object) {
+  if (!("children" in typeLike) || !Array.isArray(typeLike.children)) {
+    throw Error(
+      "Expected a List type to have an array-like `children` property",
+    );
+  }
+  if (typeLike.children.length !== 1) {
+    throw Error("Expected a List type to have exactly one child");
+  }
+  return new List(sanitizeField(typeLike.children[0]));
+}
+
+function sanitizeStruct(typeLike: object) {
+  if (!("children" in typeLike) || !Array.isArray(typeLike.children)) {
+    throw Error(
+      "Expected a Struct type to have an array-like `children` property",
+    );
+  }
+  return new Struct(typeLike.children.map((child) => sanitizeField(child)));
+}
+
+function sanitizeUnion(typeLike: object) {
+  if (
+    !("typeIds" in typeLike) ||
+    !("mode" in typeLike) ||
+    typeof typeLike.mode !== "number"
+  ) {
+    throw Error(
+      "Expected a Union type to have `typeIds` and `mode` properties",
+    );
+  }
+  if (!("children" in typeLike) || !Array.isArray(typeLike.children)) {
+    throw Error(
+      "Expected a Union type to have an array-like `children` property",
+    );
+  }
+
+  return new Union(
+    typeLike.mode,
+    // eslint-disable-next-line @typescript-eslint/no-explicit-any
+    typeLike.typeIds as any,
+    typeLike.children.map((child) => sanitizeField(child)),
+  );
+}
+
+function sanitizeTypedUnion(
+  typeLike: object,
+  // eslint-disable-next-line @typescript-eslint/naming-convention
+  UnionType: typeof DenseUnion | typeof SparseUnion,
+) {
+  if (!("typeIds" in typeLike)) {
+    throw Error(
+      "Expected a DenseUnion/SparseUnion type to have a `typeIds` property",
+    );
+  }
+  if (!("children" in typeLike) || !Array.isArray(typeLike.children)) {
+    throw Error(
+      "Expected a DenseUnion/SparseUnion type to have an array-like `children` property",
+    );
+  }
+
+  return new UnionType(
+    typeLike.typeIds as Int32Array | number[],
+    typeLike.children.map((child) => sanitizeField(child)),
+  );
+}
+
+function sanitizeFixedSizeBinary(typeLike: object) {
+  if (!("byteWidth" in typeLike) || typeof typeLike.byteWidth !== "number") {
+    throw Error(
+      "Expected a FixedSizeBinary type to have a `byteWidth` property",
+    );
+  }
+  return new FixedSizeBinary(typeLike.byteWidth);
+}
+
+function sanitizeFixedSizeList(typeLike: object) {
+  if (!("listSize" in typeLike) || typeof typeLike.listSize !== "number") {
+    throw Error("Expected a FixedSizeList type to have a `listSize` property");
+  }
+  if (!("children" in typeLike) || !Array.isArray(typeLike.children)) {
+    throw Error(
+      "Expected a FixedSizeList type to have an array-like `children` property",
+    );
+  }
+  if (typeLike.children.length !== 1) {
+    throw Error("Expected a FixedSizeList type to have exactly one child");
+  }
+  return new FixedSizeList(
+    typeLike.listSize,
+    sanitizeField(typeLike.children[0]),
+  );
+}
+
+function sanitizeMap(typeLike: object) {
+  if (!("children" in typeLike) || !Array.isArray(typeLike.children)) {
+    throw Error(
+      "Expected a Map type to have an array-like `children` property",
+    );
+  }
+  if (!("keysSorted" in typeLike) || typeof typeLike.keysSorted !== "boolean") {
+    throw Error("Expected a Map type to have a `keysSorted` property");
+  }
+
+  return new Map_(
+    // eslint-disable-next-line @typescript-eslint/no-explicit-any
+    typeLike.children.map((field) => sanitizeField(field)) as any,
+    typeLike.keysSorted,
+  );
+}
+
+function sanitizeDuration(typeLike: object) {
+  if (!("unit" in typeLike) || typeof typeLike.unit !== "number") {
+    throw Error("Expected a Duration type to have a `unit` property");
+  }
+  return new Duration(typeLike.unit);
+}
+
+function sanitizeDictionary(typeLike: object) {
+  if (!("id" in typeLike) || typeof typeLike.id !== "number") {
+    throw Error("Expected a Dictionary type to have an `id` property");
+  }
+  if (!("indices" in typeLike) || typeof typeLike.indices !== "object") {
+    throw Error("Expected a Dictionary type to have an `indices` property");
+  }
+  if (!("dictionary" in typeLike) || typeof typeLike.dictionary !== "object") {
+    throw Error("Expected a Dictionary type to have an `dictionary` property");
+  }
+  if (!("isOrdered" in typeLike) || typeof typeLike.isOrdered !== "boolean") {
+    throw Error("Expected a Dictionary type to have an `isOrdered` property");
+  }
+  return new Dictionary(
+    sanitizeType(typeLike.dictionary),
+    sanitizeType(typeLike.indices) as TKeys,
+    typeLike.id,
+    typeLike.isOrdered,
+  );
+}
+
+// eslint-disable-next-line @typescript-eslint/no-explicit-any
+function sanitizeType(typeLike: unknown): DataType<any> {
+  if (typeof typeLike !== "object" || typeLike === null) {
+    throw Error("Expected a Type but object was null/undefined");
+  }
+  if (!("typeId" in typeLike) || !(typeof typeLike.typeId !== "function")) {
+    throw Error("Expected a Type to have a typeId function");
+  }
+  let typeId: Type;
+  if (typeof typeLike.typeId === "function") {
+    typeId = (typeLike.typeId as () => unknown)() as Type;
+  } else if (typeof typeLike.typeId === "number") {
+    typeId = typeLike.typeId as Type;
+  } else {
+    throw Error("Type's typeId property was not a function or number");
+  }
+
+  switch (typeId) {
+    case Type.NONE:
+      throw Error("Received a Type with a typeId of NONE");
+    case Type.Null:
+      return new Null();
+    case Type.Int:
+      return sanitizeInt(typeLike);
+    case Type.Float:
+      return sanitizeFloat(typeLike);
+    case Type.Binary:
+      return new Binary();
+    case Type.Utf8:
+      return new Utf8();
+    case Type.Bool:
+      return new Bool();
+    case Type.Decimal:
+      return sanitizeDecimal(typeLike);
+    case Type.Date:
+      return sanitizeDate(typeLike);
+    case Type.Time:
+      return sanitizeTime(typeLike);
+    case Type.Timestamp:
+      return sanitizeTimestamp(typeLike);
+    case Type.Interval:
+      return sanitizeInterval(typeLike);
+    case Type.List:
+      return sanitizeList(typeLike);
+    case Type.Struct:
+      return sanitizeStruct(typeLike);
+    case Type.Union:
+      return sanitizeUnion(typeLike);
+    case Type.FixedSizeBinary:
+      return sanitizeFixedSizeBinary(typeLike);
+    case Type.FixedSizeList:
+      return sanitizeFixedSizeList(typeLike);
+    case Type.Map:
+      return sanitizeMap(typeLike);
+    case Type.Duration:
+      return sanitizeDuration(typeLike);
+    case Type.Dictionary:
+      return sanitizeDictionary(typeLike);
+    case Type.Int8:
+      return new Int8();
+    case Type.Int16:
+      return new Int16();
+    case Type.Int32:
+      return new Int32();
+    case Type.Int64:
+      return new Int64();
+    case Type.Uint8:
+      return new Uint8();
+    case Type.Uint16:
+      return new Uint16();
+    case Type.Uint32:
+      return new Uint32();
+    case Type.Uint64:
+      return new Uint64();
+    case Type.Float16:
+      return new Float16();
+    case Type.Float32:
+      return new Float32();
+    case Type.Float64:
+      return new Float64();
+    case Type.DateMillisecond:
+      return new DateMillisecond();
+    case Type.DateDay:
+      return new DateDay();
+    case Type.TimeNanosecond:
+      return new TimeNanosecond();
+    case Type.TimeMicrosecond:
+      return new TimeMicrosecond();
+    case Type.TimeMillisecond:
+      return new TimeMillisecond();
+    case Type.TimeSecond:
+      return new TimeSecond();
+    case Type.TimestampNanosecond:
+      return sanitizeTypedTimestamp(typeLike, TimestampNanosecond);
+    case Type.TimestampMicrosecond:
+      return sanitizeTypedTimestamp(typeLike, TimestampMicrosecond);
+    case Type.TimestampMillisecond:
+      return sanitizeTypedTimestamp(typeLike, TimestampMillisecond);
+    case Type.TimestampSecond:
+      return sanitizeTypedTimestamp(typeLike, TimestampSecond);
+    case Type.DenseUnion:
+      return sanitizeTypedUnion(typeLike, DenseUnion);
+    case Type.SparseUnion:
+      return sanitizeTypedUnion(typeLike, SparseUnion);
+    case Type.IntervalDayTime:
+      return new IntervalDayTime();
+    case Type.IntervalYearMonth:
+      return new IntervalYearMonth();
+    case Type.DurationNanosecond:
+      return new DurationNanosecond();
+    case Type.DurationMicrosecond:
+      return new DurationMicrosecond();
+    case Type.DurationMillisecond:
+      return new DurationMillisecond();
+    case Type.DurationSecond:
+      return new DurationSecond();
+    default:
+      throw new Error("Unrecoginized type id in schema: " + typeId);
+  }
+}
+
+function sanitizeField(fieldLike: unknown): Field {
+  if (fieldLike instanceof Field) {
+    return fieldLike;
+  }
+  if (typeof fieldLike !== "object" || fieldLike === null) {
+    throw Error("Expected a Field but object was null/undefined");
+  }
+  if (
+    !("type" in fieldLike) ||
+    !("name" in fieldLike) ||
+    !("nullable" in fieldLike)
+  ) {
+    throw Error(
+      "The field passed in is missing a `type`/`name`/`nullable` property",
+    );
+  }
+  const type = sanitizeType(fieldLike.type);
+  const name = fieldLike.name;
+  if (!(typeof name === "string")) {
+    throw Error("The field passed in had a non-string `name` property");
+  }
+  const nullable = fieldLike.nullable;
+  if (!(typeof nullable === "boolean")) {
+    throw Error("The field passed in had a non-boolean `nullable` property");
+  }
+  let metadata;
+  if ("metadata" in fieldLike) {
+    metadata = sanitizeMetadata(fieldLike.metadata);
+  }
+  return new Field(name, type, nullable, metadata);
+}
+
+/**
+ * Convert something schemaLike into a Schema instance
+ *
+ * This method is often needed even when the caller is using a Schema
+ * instance because they might be using a different instance of apache-arrow
+ * than lancedb is using.
+ */
+export function sanitizeSchema(schemaLike: unknown): Schema {
+  if (schemaLike instanceof Schema) {
+    return schemaLike;
+  }
+  if (typeof schemaLike !== "object" || schemaLike === null) {
+    throw Error("Expected a Schema but object was null/undefined");
+  }
+  if (!("fields" in schemaLike)) {
+    throw Error(
+      "The schema passed in does not appear to be a schema (no 'fields' property)",
+    );
+  }
+  let metadata;
+  if ("metadata" in schemaLike) {
+    metadata = sanitizeMetadata(schemaLike.metadata);
+  }
+  if (!Array.isArray(schemaLike.fields)) {
+    throw Error(
+      "The schema passed in had a 'fields' property but it was not an array",
+    );
+  }
+  const sanitizedFields = schemaLike.fields.map((field) =>
+    sanitizeField(field),
+  );
+  return new Schema(sanitizedFields, metadata);
+}

nodejs/lancedb/table.ts

@@ -0,0 +1,353 @@
+// Copyright 2024 Lance Developers.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import { Schema, tableFromIPC } from "apache-arrow";
+import {
+  AddColumnsSql,
+  ColumnAlteration,
+  IndexConfig,
+  Table as _NativeTable,
+} from "./native";
+import { Query, VectorQuery } from "./query";
+import { IndexOptions } from "./indices";
+import { Data, fromDataToBuffer } from "./arrow";
+
+export { IndexConfig } from "./native";
+/**
+ * Options for adding data to a table.
+ */
+export interface AddDataOptions {
+  /**
+   * If "append" (the default) then the new data will be added to the table
+   *
+   * If "overwrite" then the new data will replace the existing data in the table.
+   */
+  mode: "append" | "overwrite";
+}
+
+export interface UpdateOptions {
+  /**
+   * A filter that limits the scope of the update.
+   *
+   * This should be an SQL filter expression.
+   *
+   * Only rows that satisfy the expression will be updated.
+   *
+   * For example, this could be 'my_col == 0' to replace all instances
+   * of 0 in a column with some other default value.
+   */
+  where: string;
+}
+
+/**
+ * A Table is a collection of Records in a LanceDB Database.
+ *
+ * A Table object is expected to be long lived and reused for multiple operations.
+ * Table objects will cache a certain amount of index data in memory.  This cache
+ * will be freed when the Table is garbage collected.  To eagerly free the cache you
+ * can call the `close` method.  Once the Table is closed, it cannot be used for any
+ * further operations.
+ *
+ * Closing a table is optional.  It not closed, it will be closed when it is garbage
+ * collected.
+ */
+export class Table {
+  private readonly inner: _NativeTable;
+
+  /** Construct a Table. Internal use only. */
+  constructor(inner: _NativeTable) {
+    this.inner = inner;
+  }
+
+  /** Return true if the table has not been closed */
+  isOpen(): boolean {
+    return this.inner.isOpen();
+  }
+
+  /**
+   * Close the table, releasing any underlying resources.
+   *
+   * It is safe to call this method multiple times.
+   *
+   * Any attempt to use the table after it is closed will result in an error.
+   */
+  close(): void {
+    this.inner.close();
+  }
+
+  /** Return a brief description of the table */
+  display(): string {
+    return this.inner.display();
+  }
+
+  /** Get the schema of the table. */
+  async schema(): Promise<Schema> {
+    const schemaBuf = await this.inner.schema();
+    const tbl = tableFromIPC(schemaBuf);
+    return tbl.schema;
+  }
+
+  /**
+   * Insert records into this Table.
+   * @param {Data} data Records to be inserted into the Table
+   */
+  async add(data: Data, options?: Partial<AddDataOptions>): Promise<void> {
+    const mode = options?.mode ?? "append";
+
+    const buffer = await fromDataToBuffer(data);
+    await this.inner.add(buffer, mode);
+  }
+
+  /**
+   * Update existing records in the Table
+   *
+   * An update operation can be used to adjust existing values.  Use the
+   * returned builder to specify which columns to update.  The new value
+   * can be a literal value (e.g. replacing nulls with some default value)
+   * or an expression applied to the old value (e.g. incrementing a value)
+   *
+   * An optional condition can be specified (e.g. "only update if the old
+   * value is 0")
+   *
+   * Note: if your condition is something like "some_id_column == 7" and
+   * you are updating many rows (with different ids) then you will get
+   * better performance with a single [`merge_insert`] call instead of
+   * repeatedly calilng this method.
+   * @param {Map<string, string> | Record<string, string>} updates - the
+   * columns to update
+   *
+   * Keys in the map should specify the name of the column to update.
+   * Values in the map provide the new value of the column.  These can
+   * be SQL literal strings (e.g. "7" or "'foo'") or they can be expressions
+   * based on the row being updated (e.g. "my_col + 1")
+   * @param {Partial<UpdateOptions>} options - additional options to control
+   * the update behavior
+   */
+  async update(
+    updates: Map<string, string> | Record<string, string>,
+    options?: Partial<UpdateOptions>,
+  ) {
+    const onlyIf = options?.where;
+    let columns: [string, string][];
+    if (updates instanceof Map) {
+      columns = Array.from(updates.entries());
+    } else {
+      columns = Object.entries(updates);
+    }
+    await this.inner.update(onlyIf, columns);
+  }
+
+  /** Count the total number of rows in the dataset. */
+  async countRows(filter?: string): Promise<number> {
+    return await this.inner.countRows(filter);
+  }
+
+  /** Delete the rows that satisfy the predicate. */
+  async delete(predicate: string): Promise<void> {
+    await this.inner.delete(predicate);
+  }
+
+  /**
+   * Create an index to speed up queries.
+   *
+   * Indices can be created on vector columns or scalar columns.
+   * Indices on vector columns will speed up vector searches.
+   * Indices on scalar columns will speed up filtering (in both
+   * vector and non-vector searches)
+   * @example
+   * // If the column has a vector (fixed size list) data type then
+   * // an IvfPq vector index will be created.
+   * const table = await conn.openTable("my_table");
+   * await table.createIndex(["vector"]);
+   * @example
+   * // For advanced control over vector index creation you can specify
+   * // the index type and options.
+   * const table = await conn.openTable("my_table");
+   * await table.createIndex(["vector"], I)
+   *   .ivf_pq({ num_partitions: 128, num_sub_vectors: 16 })
+   *   .build();
+   * @example
+   * // Or create a Scalar index
+   * await table.createIndex("my_float_col").build();
+   */
+  async createIndex(column: string, options?: Partial<IndexOptions>) {
+    // Bit of a hack to get around the fact that TS has no package-scope.
+    // eslint-disable-next-line @typescript-eslint/no-explicit-any
+    const nativeIndex = (options?.config as any)?.inner;
+    await this.inner.createIndex(nativeIndex, column, options?.replace);
+  }
+
+  /**
+   * Create a {@link Query} Builder.
+   *
+   * Queries allow you to search your existing data.  By default the query will
+   * return all the data in the table in no particular order.  The builder
+   * returned by this method can be used to control the query using filtering,
+   * vector similarity, sorting, and more.
+   *
+   * Note: By default, all columns are returned.  For best performance, you should
+   * only fetch the columns you need.  See [`Query::select_with_projection`] for
+   * more details.
+   *
+   * When appropriate, various indices and statistics based pruning will be used to
+   * accelerate the query.
+   * @example
+   * // SQL-style filtering
+   * //
+   * // This query will return up to 1000 rows whose value in the `id` column
+   * // is greater than 5.  LanceDb supports a broad set of filtering functions.
+   * for await (const batch of table.query()
+   *                          .filter("id > 1").select(["id"]).limit(20)) {
+   *  console.log(batch);
+   * }
+   * @example
+   * // Vector Similarity Search
+   * //
+   * // This example will find the 10 rows whose value in the "vector" column are
+   * // closest to the query vector [1.0, 2.0, 3.0].  If an index has been created
+   * // on the "vector" column then this will perform an ANN search.
+   * //
+   * // The `refine_factor` and `nprobes` methods are used to control the recall /
+   * // latency tradeoff of the search.
+   * for await (const batch of table.query()
+   *                    .nearestTo([1, 2, 3])
+   *                    .refineFactor(5).nprobe(10)
+   *                    .limit(10)) {
+   *  console.log(batch);
+   * }
+   * @example
+   * // Scan the full dataset
+   * //
+   * // This query will return everything in the table in no particular order.
+   * for await (const batch of table.query()) {
+   *   console.log(batch);
+   * }
+   * @returns {Query} A builder that can be used to parameterize the query
+   */
+  query(): Query {
+    return new Query(this.inner);
+  }
+
+  /**
+   * Search the table with a given query vector.
+   *
+   * This is a convenience method for preparing a vector query and
+   * is the same thing as calling `nearestTo` on the builder returned
+   * by `query`.  @see {@link Query#nearestTo} for more details.
+   */
+  vectorSearch(vector: unknown): VectorQuery {
+    return this.query().nearestTo(vector);
+  }
+
+  // TODO: Support BatchUDF
+  /**
+   * Add new columns with defined values.
+   * @param {AddColumnsSql[]} newColumnTransforms pairs of column names and
+   * the SQL expression to use to calculate the value of the new column. These
+   * expressions will be evaluated for each row in the table, and can
+   * reference existing columns in the table.
+   */
+  async addColumns(newColumnTransforms: AddColumnsSql[]): Promise<void> {
+    await this.inner.addColumns(newColumnTransforms);
+  }
+
+  /**
+   * Alter the name or nullability of columns.
+   * @param {ColumnAlteration[]} columnAlterations One or more alterations to
+   * apply to columns.
+   */
+  async alterColumns(columnAlterations: ColumnAlteration[]): Promise<void> {
+    await this.inner.alterColumns(columnAlterations);
+  }
+
+  /**
+   * Drop one or more columns from the dataset
+   *
+   * This is a metadata-only operation and does not remove the data from the
+   * underlying storage. In order to remove the data, you must subsequently
+   * call ``compact_files`` to rewrite the data without the removed columns and
+   * then call ``cleanup_files`` to remove the old files.
+   * @param {string[]} columnNames The names of the columns to drop. These can
+   * be nested column references (e.g. "a.b.c") or top-level column names
+   * (e.g. "a").
+   */
+  async dropColumns(columnNames: string[]): Promise<void> {
+    await this.inner.dropColumns(columnNames);
+  }
+
+  /**
+   * Retrieve the version of the table
+   *
+   * LanceDb supports versioning.  Every operation that modifies the table increases
+   * version.  As long as a version hasn't been deleted you can `[Self::checkout]` that
+   * version to view the data at that point.  In addition, you can `[Self::restore]` the
+   * version to replace the current table with a previous version.
+   */
+  async version(): Promise<number> {
+    return await this.inner.version();
+  }
+
+  /**
+   * Checks out a specific version of the Table
+   *
+   * Any read operation on the table will now access the data at the checked out version.
+   * As a consequence, calling this method will disable any read consistency interval
+   * that was previously set.
+   *
+   * This is a read-only operation that turns the table into a sort of "view"
+   * or "detached head".  Other table instances will not be affected.  To make the change
+   * permanent you can use the `[Self::restore]` method.
+   *
+   * Any operation that modifies the table will fail while the table is in a checked
+   * out state.
+   *
+   * To return the table to a normal state use `[Self::checkout_latest]`
+   */
+  async checkout(version: number): Promise<void> {
+    await this.inner.checkout(version);
+  }
+
+  /**
+   * Ensures the table is pointing at the latest version
+   *
+   * This can be used to manually update a table when the read_consistency_interval is None
+   * It can also be used to undo a `[Self::checkout]` operation
+   */
+  async checkoutLatest(): Promise<void> {
+    await this.inner.checkoutLatest();
+  }
+
+  /**
+   * Restore the table to the currently checked out version
+   *
+   * This operation will fail if checkout has not been called previously
+   *
+   * This operation will overwrite the latest version of the table with a
+   * previous version.  Any changes made since the checked out version will
+   * no longer be visible.
+   *
+   * Once the operation concludes the table will no longer be in a checked
+   * out state and the read_consistency_interval, if any, will apply.
+   */
+  async restore(): Promise<void> {
+    await this.inner.restore();
+  }
+
+  /**
+   * List all indices that have been created with Self::create_index
+   */
+  async listIndices(): Promise<IndexConfig[]> {
+    return await this.inner.listIndices();
+  }
+}

nodejs/npm/darwin-arm64/README.md

@@ -1,3 +1,3 @@
-# `vectordb-darwin-arm64`
+# `lancedb-darwin-arm64`
 
-This is the **aarch64-apple-darwin** binary for `vectordb`
+This is the **aarch64-apple-darwin** binary for `lancedb`

nodejs/npm/darwin-arm64/package.json

@@ -1,5 +1,5 @@
 {
-  "name": "vectordb-darwin-arm64",
+  "name": "lancedb-darwin-arm64",
   "version": "0.4.3",
   "os": [
     "darwin"
@@ -7,12 +7,12 @@
   "cpu": [
     "arm64"
   ],
-  "main": "vectordb.darwin-arm64.node",
+  "main": "lancedb.darwin-arm64.node",
   "files": [
-    "vectordb.darwin-arm64.node"
+    "lancedb.darwin-arm64.node"
   ],
   "license": "MIT",
   "engines": {
     "node": ">= 18"
   }
-}
+}

nodejs/npm/darwin-x64/README.md

@@ -1,3 +1,3 @@
-# `vectordb-darwin-x64`
+# `lancedb-darwin-x64`
 
-This is the **x86_64-apple-darwin** binary for `vectordb`
+This is the **x86_64-apple-darwin** binary for `lancedb`

nodejs/npm/darwin-x64/package.json

@@ -1,5 +1,5 @@
 {
-  "name": "vectordb-darwin-x64",
+  "name": "lancedb-darwin-x64",
   "version": "0.4.3",
   "os": [
     "darwin"
@@ -7,12 +7,12 @@
   "cpu": [
     "x64"
   ],
-  "main": "vectordb.darwin-x64.node",
+  "main": "lancedb.darwin-x64.node",
   "files": [
-    "vectordb.darwin-x64.node"
+    "lancedb.darwin-x64.node"
   ],
   "license": "MIT",
   "engines": {
     "node": ">= 18"
   }
-}
+}

nodejs/npm/linux-arm64-gnu/README.md

@@ -1,3 +1,3 @@
-# `vectordb-linux-arm64-gnu`
+# `lancedb-linux-arm64-gnu`
 
-This is the **aarch64-unknown-linux-gnu** binary for `vectordb`
+This is the **aarch64-unknown-linux-gnu** binary for `lancedb`

nodejs/npm/linux-arm64-gnu/package.json

@@ -1,5 +1,5 @@
 {
-  "name": "vectordb-linux-arm64-gnu",
+  "name": "lancedb-linux-arm64-gnu",
   "version": "0.4.3",
   "os": [
     "linux"
@@ -7,9 +7,9 @@
   "cpu": [
     "arm64"
   ],
-  "main": "vectordb.linux-arm64-gnu.node",
+  "main": "lancedb.linux-arm64-gnu.node",
   "files": [
-    "vectordb.linux-arm64-gnu.node"
+    "lancedb.linux-arm64-gnu.node"
   ],
   "license": "MIT",
   "engines": {
@@ -18,4 +18,4 @@
   "libc": [
     "glibc"
   ]
-}
+}

nodejs/npm/linux-x64-gnu/README.md

@@ -1,3 +1,3 @@
-# `vectordb-linux-x64-gnu`
+# `lancedb-linux-x64-gnu`
 
-This is the **x86_64-unknown-linux-gnu** binary for `vectordb`
+This is the **x86_64-unknown-linux-gnu** binary for `lancedb`

nodejs/npm/linux-x64-gnu/package.json

@@ -1,5 +1,5 @@
 {
-  "name": "vectordb-linux-x64-gnu",
+  "name": "lancedb-linux-x64-gnu",
   "version": "0.4.3",
   "os": [
     "linux"
@@ -7,9 +7,9 @@
   "cpu": [
     "x64"
   ],
-  "main": "vectordb.linux-x64-gnu.node",
+  "main": "lancedb.linux-x64-gnu.node",
   "files": [
-    "vectordb.linux-x64-gnu.node"
+    "lancedb.linux-x64-gnu.node"
   ],
   "license": "MIT",
   "engines": {
@@ -18,4 +18,4 @@
   "libc": [
     "glibc"
   ]
-}
+}

nodejs/package.json

@@ -1,10 +1,10 @@
 {
-  "name": "vectordb",
+  "name": "lancedb",
   "version": "0.4.3",
   "main": "./dist/index.js",
   "types": "./dist/index.d.ts",
   "napi": {
-    "name": "vectordb-nodejs",
+    "name": "lancedb-nodejs",
     "triples": {
       "defaults": false,
       "additional": [
@@ -18,15 +18,22 @@
   "license": "Apache 2.0",
   "devDependencies": {
     "@napi-rs/cli": "^2.18.0",
-    "@types/jest": "^29.5.11",
+    "@types/jest": "^29.1.2",
+    "@types/tmp": "^0.2.6",
     "@typescript-eslint/eslint-plugin": "^6.19.0",
     "@typescript-eslint/parser": "^6.19.0",
-    "eslint": "^8.56.0",
+    "apache-arrow-old": "npm:apache-arrow@13.0.0",
+    "eslint": "^8.57.0",
+    "eslint-config-prettier": "^9.1.0",
+    "eslint-plugin-jsdoc": "^48.2.1",
     "jest": "^29.7.0",
+    "prettier": "^3.1.0",
+    "tmp": "^0.2.3",
     "ts-jest": "^29.1.2",
     "typedoc": "^0.25.7",
     "typedoc-plugin-markdown": "^3.17.1",
-    "typescript": "^5.3.3"
+    "typescript": "^5.3.3",
+    "typescript-eslint": "^7.1.0"
   },
   "ava": {
     "timeout": "3m"
@@ -45,23 +52,25 @@
   ],
   "scripts": {
     "artifacts": "napi artifacts",
-    "build:native": "napi build --platform --release --js vectordb/native.js --dts vectordb/native.d.ts dist/",
+    "build:native": "napi build --platform --release --js lancedb/native.js --dts lancedb/native.d.ts dist/",
-    "build:debug": "napi build --platform --dts ../vectordb/native.d.ts --js ../vectordb/native.js dist/",
+    "build:debug": "napi build --platform --dts ../lancedb/native.d.ts --js ../lancedb/native.js dist/",
     "build": "npm run build:debug && tsc -b",
-    "docs": "typedoc --plugin typedoc-plugin-markdown vectordb/index.ts",
+    "chkformat": "prettier . --check",
-    "lint": "eslint vectordb --ext .js,.ts",
+    "docs": "typedoc --plugin typedoc-plugin-markdown lancedb/index.ts",
+    "lint": "eslint lancedb && eslint __test__",
     "prepublishOnly": "napi prepublish -t npm",
-    "test": "npm run build && jest",
+    "test": "npm run build && jest --verbose",
     "universal": "napi universal",
     "version": "napi version"
   },
   "optionalDependencies": {
-    "vectordb-darwin-arm64": "0.4.3",
+    "lancedb-darwin-arm64": "0.4.3",
-    "vectordb-darwin-x64": "0.4.3",
+    "lancedb-darwin-x64": "0.4.3",
-    "vectordb-linux-arm64-gnu": "0.4.3",
+    "lancedb-linux-arm64-gnu": "0.4.3",
-    "vectordb-linux-x64-gnu": "0.4.3"
+    "lancedb-linux-x64-gnu": "0.4.3",
+    "openai": "^4.28.4"
   },
-  "dependencies": {
+  "peerDependencies": {
     "apache-arrow": "^15.0.0"
   }
 }

nodejs/src/connection.rs

@@ -12,37 +12,96 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-use std::sync::Arc;
-
 use napi::bindgen_prelude::*;
 use napi_derive::*;
 
 use crate::table::Table;
-use vectordb::connection::{Connection as LanceDBConnection, Database};
+use crate::ConnectionOptions;
-use vectordb::ipc::ipc_file_to_batches;
+use lancedb::connection::{ConnectBuilder, Connection as LanceDBConnection, CreateTableMode};
+use lancedb::ipc::{ipc_file_to_batches, ipc_file_to_schema};
 
 #[napi]
 pub struct Connection {
-    conn: Arc<dyn LanceDBConnection>,
+    inner: Option<LanceDBConnection>,
+}
+
+impl Connection {
+    pub(crate) fn inner_new(inner: LanceDBConnection) -> Self {
+        Self { inner: Some(inner) }
+    }
+
+    fn get_inner(&self) -> napi::Result<&LanceDBConnection> {
+        self.inner
+            .as_ref()
+            .ok_or_else(|| napi::Error::from_reason("Connection is closed"))
+    }
+}
+
+impl Connection {
+    fn parse_create_mode_str(mode: &str) -> napi::Result<CreateTableMode> {
+        match mode {
+            "create" => Ok(CreateTableMode::Create),
+            "overwrite" => Ok(CreateTableMode::Overwrite),
+            "exist_ok" => Ok(CreateTableMode::exist_ok(|builder| builder)),
+            _ => Err(napi::Error::from_reason(format!("Invalid mode {}", mode))),
+        }
+    }
 }
 
 #[napi]
 impl Connection {
     /// Create a new Connection instance from the given URI.
     #[napi(factory)]
-    pub async fn new(uri: String) -> napi::Result<Self> {
+    pub async fn new(uri: String, options: ConnectionOptions) -> napi::Result<Self> {
-        Ok(Self {
+        let mut builder = ConnectBuilder::new(&uri);
-            conn: Arc::new(Database::connect(&uri).await.map_err(|e| {
+        if let Some(api_key) = options.api_key {
-                napi::Error::from_reason(format!("Failed to connect to database: {}", e))
+            builder = builder.api_key(&api_key);
-            })?),
+        }
-        })
+        if let Some(host_override) = options.host_override {
+            builder = builder.host_override(&host_override);
+        }
+        if let Some(interval) = options.read_consistency_interval {
+            builder =
+                builder.read_consistency_interval(std::time::Duration::from_secs_f64(interval));
+        }
+        Ok(Self::inner_new(
+            builder
+                .execute()
+                .await
+                .map_err(|e| napi::Error::from_reason(format!("{}", e)))?,
+        ))
+    }
+
+    #[napi]
+    pub fn display(&self) -> napi::Result<String> {
+        Ok(self.get_inner()?.to_string())
+    }
+
+    #[napi]
+    pub fn is_open(&self) -> bool {
+        self.inner.is_some()
+    }
+
+    #[napi]
+    pub fn close(&mut self) {
+        self.inner.take();
     }
 
     /// List all tables in the dataset.
     #[napi]
-    pub async fn table_names(&self) -> napi::Result<Vec<String>> {
+    pub async fn table_names(
-        self.conn
+        &self,
-            .table_names()
+        start_after: Option<String>,
+        limit: Option<u32>,
+    ) -> napi::Result<Vec<String>> {
+        let mut op = self.get_inner()?.table_names();
+        if let Some(start_after) = start_after {
+            op = op.start_after(start_after);
+        }
+        if let Some(limit) = limit {
+            op = op.limit(limit);
+        }
+        op.execute()
             .await
             .map_err(|e| napi::Error::from_reason(format!("{}", e)))
     }
@@ -54,12 +113,41 @@ impl Connection {
     /// - buf: The buffer containing the IPC file.
     ///
     #[napi]
-    pub async fn create_table(&self, name: String, buf: Buffer) -> napi::Result<Table> {
+    pub async fn create_table(
+        &self,
+        name: String,
+        buf: Buffer,
+        mode: String,
+    ) -> napi::Result<Table> {
         let batches = ipc_file_to_batches(buf.to_vec())
             .map_err(|e| napi::Error::from_reason(format!("Failed to read IPC file: {}", e)))?;
+        let mode = Self::parse_create_mode_str(&mode)?;
         let tbl = self
-            .conn
+            .get_inner()?
-            .create_table(&name, Box::new(batches), None)
+            .create_table(&name, batches)
+            .mode(mode)
+            .execute()
+            .await
+            .map_err(|e| napi::Error::from_reason(format!("{}", e)))?;
+        Ok(Table::new(tbl))
+    }
+
+    #[napi]
+    pub async fn create_empty_table(
+        &self,
+        name: String,
+        schema_buf: Buffer,
+        mode: String,
+    ) -> napi::Result<Table> {
+        let schema = ipc_file_to_schema(schema_buf.to_vec()).map_err(|e| {
+            napi::Error::from_reason(format!("Failed to marshal schema from JS to Rust: {}", e))
+        })?;
+        let mode = Self::parse_create_mode_str(&mode)?;
+        let tbl = self
+            .get_inner()?
+            .create_empty_table(&name, schema)
+            .mode(mode)
+            .execute()
             .await
             .map_err(|e| napi::Error::from_reason(format!("{}", e)))?;
         Ok(Table::new(tbl))
@@ -68,8 +156,9 @@ impl Connection {
     #[napi]
     pub async fn open_table(&self, name: String) -> napi::Result<Table> {
         let tbl = self
-            .conn
+            .get_inner()?
             .open_table(&name)
+            .execute()
             .await
             .map_err(|e| napi::Error::from_reason(format!("{}", e)))?;
         Ok(Table::new(tbl))
@@ -78,7 +167,7 @@ impl Connection {
     /// Drop table with the name. Or raise an error if the table does not exist.
     #[napi]
     pub async fn drop_table(&self, name: String) -> napi::Result<()> {
-        self.conn
+        self.get_inner()?
             .drop_table(&name)
             .await
             .map_err(|e| napi::Error::from_reason(format!("{}", e)))

nodejs/src/error.rs

@@ -0,0 +1,12 @@
+pub type Result<T> = napi::Result<T>;
+
+pub trait NapiErrorExt<T> {
+    /// Convert to a napi error using from_reason(err.to_string())
+    fn default_error(self) -> Result<T>;
+}
+
+impl<T> NapiErrorExt<T> for std::result::Result<T, lancedb::Error> {
+    fn default_error(self) -> Result<T> {
+        self.map_err(|err| napi::Error::from_reason(err.to_string()))
+    }
+}

nodejs/src/index.rs

@@ -12,89 +12,68 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-use lance_linalg::distance::MetricType as LanceMetricType;
+use std::sync::Mutex;
+
+use lancedb::index::scalar::BTreeIndexBuilder;
+use lancedb::index::vector::IvfPqIndexBuilder;
+use lancedb::index::Index as LanceDbIndex;
 use napi_derive::napi;
 
+use crate::util::parse_distance_type;
+
 #[napi]
-pub enum IndexType {
+pub struct Index {
-    Scalar,
+    inner: Mutex<Option<LanceDbIndex>>,
-    IvfPq,
+}
+
+impl Index {
+    pub fn consume(&self) -> napi::Result<LanceDbIndex> {
+        self.inner
+            .lock()
+            .unwrap()
+            .take()
+            .ok_or(napi::Error::from_reason(
+                "attempt to use an index more than once",
+            ))
+    }
 }
 
 #[napi]
-pub enum MetricType {
+impl Index {
-    L2,
+    #[napi(factory)]
-    Cosine,
+    pub fn ivf_pq(
-    Dot,
+        distance_type: Option<String>,
-}
+        num_partitions: Option<u32>,
+        num_sub_vectors: Option<u32>,
+        max_iterations: Option<u32>,
+        sample_rate: Option<u32>,
+    ) -> napi::Result<Self> {
+        let mut ivf_pq_builder = IvfPqIndexBuilder::default();
+        if let Some(distance_type) = distance_type {
+            let distance_type = parse_distance_type(distance_type)?;
+            ivf_pq_builder = ivf_pq_builder.distance_type(distance_type);
+        }
+        if let Some(num_partitions) = num_partitions {
+            ivf_pq_builder = ivf_pq_builder.num_partitions(num_partitions);
+        }
+        if let Some(num_sub_vectors) = num_sub_vectors {
+            ivf_pq_builder = ivf_pq_builder.num_sub_vectors(num_sub_vectors);
+        }
+        if let Some(max_iterations) = max_iterations {
+            ivf_pq_builder = ivf_pq_builder.max_iterations(max_iterations);
+        }
+        if let Some(sample_rate) = sample_rate {
+            ivf_pq_builder = ivf_pq_builder.sample_rate(sample_rate);
+        }
+        Ok(Self {
+            inner: Mutex::new(Some(LanceDbIndex::IvfPq(ivf_pq_builder))),
+        })
+    }
 
-impl From<MetricType> for LanceMetricType {
+    #[napi(factory)]
-    fn from(metric: MetricType) -> Self {
+    pub fn btree() -> Self {
-        match metric {
+        Self {
-            MetricType::L2 => Self::L2,
+            inner: Mutex::new(Some(LanceDbIndex::BTree(BTreeIndexBuilder::default()))),
-            MetricType::Cosine => Self::Cosine,
-            MetricType::Dot => Self::Dot,
         }
     }
 }
-
-#[napi]
-pub struct IndexBuilder {
-    inner: vectordb::index::IndexBuilder,
-}
-
-#[napi]
-impl IndexBuilder {
-    pub fn new(tbl: &dyn vectordb::Table) -> Self {
-        let inner = tbl.create_index(&[]);
-        Self { inner }
-    }
-
-    #[napi]
-    pub unsafe fn replace(&mut self, v: bool) {
-        self.inner.replace(v);
-    }
-
-    #[napi]
-    pub unsafe fn column(&mut self, c: String) {
-        self.inner.columns(&[c.as_str()]);
-    }
-
-    #[napi]
-    pub unsafe fn name(&mut self, name: String) {
-        self.inner.name(name.as_str());
-    }
-
-    #[napi]
-    pub unsafe fn ivf_pq(
-        &mut self,
-        metric_type: Option<MetricType>,
-        num_partitions: Option<u32>,
-        num_sub_vectors: Option<u32>,
-        num_bits: Option<u32>,
-        max_iterations: Option<u32>,
-        sample_rate: Option<u32>,
-    ) {
-        self.inner.ivf_pq();
-        metric_type.map(|m| self.inner.metric_type(m.into()));
-        num_partitions.map(|p| self.inner.num_partitions(p));
-        num_sub_vectors.map(|s| self.inner.num_sub_vectors(s));
-        num_bits.map(|b| self.inner.num_bits(b));
-        max_iterations.map(|i| self.inner.max_iterations(i));
-        sample_rate.map(|s| self.inner.sample_rate(s));
-    }
-
-    #[napi]
-    pub unsafe fn scalar(&mut self) {
-        self.inner.scalar();
-    }
-
-    #[napi]
-    pub async fn build(&self) -> napi::Result<()> {
-        self.inner
-            .build()
-            .await
-            .map_err(|e| napi::Error::from_reason(format!("Failed to build index: {}", e)))?;
-        Ok(())
-    }
-}

nodejs/src/iterator.rs

@@ -13,20 +13,20 @@
 // limitations under the License.
 
 use futures::StreamExt;
-use lance::io::RecordBatchStream;
+use lancedb::arrow::SendableRecordBatchStream;
+use lancedb::ipc::batches_to_ipc_file;
 use napi::bindgen_prelude::*;
 use napi_derive::napi;
-use vectordb::ipc::batches_to_ipc_file;
 
 /** Typescript-style Async Iterator over RecordBatches */
 #[napi]
 pub struct RecordBatchIterator {
-    inner: Box<dyn RecordBatchStream + Unpin>,
+    inner: SendableRecordBatchStream,
 }
 
 #[napi]
 impl RecordBatchIterator {
-    pub(crate) fn new(inner: Box<dyn RecordBatchStream + Unpin>) -> Self {
+    pub(crate) fn new(inner: SendableRecordBatchStream) -> Self {
         Self { inner }
     }