chore: update lance dependency to v12.0.0-beta.5 (#4089)

Updates the Rust workspace and Java lance-core dependency to Lance
v12.0.0-beta.5. Includes minimal Rust 1.97 Clippy compatibility fixes
required by validation. Lance tag:
https://github.com/lance-format/lance/releases/tag/v12.0.0-beta.5

---------

Co-authored-by: Jack Ye <yezhaoqin@gmail.com>

.bumpversion.toml

@@ -1,5 +1,5 @@
 [tool.bumpversion]
-current_version = "0.38.0-beta.10"
+current_version = "0.38.0-beta.12"
 parse = """(?x)
     (?P<major>0|[1-9]\\d*)\\.
     (?P<minor>0|[1-9]\\d*)\\.

.github/dependabot.yml

@@ -44,3 +44,27 @@ updates:
       python-deps:
         patterns:
           - "*"
+
+  # The npm ecosystem covers pnpm lockfiles. There are two separate installs:
+  # the bindings themselves and the examples, which have their own lockfile.
+  # As with cargo and pip above, only bump the lockfile — the version ranges
+  # in package.json are our consumers' constraints, not ours.
+  - package-ecosystem: npm
+    directory: /nodejs
+    schedule:
+      interval: weekly
+    versioning-strategy: lockfile-only
+    groups:
+      nodejs-deps:
+        patterns:
+          - "*"
+
+  - package-ecosystem: npm
+    directory: /nodejs/examples
+    schedule:
+      interval: weekly
+    versioning-strategy: lockfile-only
+    groups:
+      nodejs-examples-deps:
+        patterns:
+          - "*"

.github/workflows/dev.yml

@@ -29,12 +29,14 @@ jobs:
     steps:
       - uses: actions/setup-node@v6
         with:
-          node-version: "18"
+          node-version: "24"
+      - uses: pnpm/action-setup@v6
+        with:
+          version: 11.1.1
       # These rules are disabled because Github will always ensure there
       # is a blank line between the title and the body and Github will
       # word wrap the description field to ensure a reasonable max line
       # length.
-      - run: npm install @commitlint/config-conventional
       - run: >
           echo 'module.exports = {
             "rules": {
@@ -43,7 +45,11 @@ jobs:
               "body-leading-blank": [0, "always"]
             }
           }' > .commitlintrc.js
-      - run: npx commitlint --extends @commitlint/config-conventional --verbose <<< $COMMIT_MSG
+      - run: >
+          pnpm dlx
+          --package @commitlint/cli@21.2.2
+          --package @commitlint/config-conventional@21.2.2
+          commitlint --extends @commitlint/config-conventional --verbose <<< $COMMIT_MSG
         env:
           COMMIT_MSG: >
             ${{ github.event.pull_request.title }}
@@ -54,7 +60,7 @@ jobs:
         with:
           script: |
             const message = `**ACTION NEEDED**
-              
+
               Lance follows the [Conventional Commits specification](https://www.conventionalcommits.org/en/v1.0.0/) for release automation.
 
               The PR title and description are used as the merge commit message.\

.github/workflows/docs-link-check.yml

@@ -56,7 +56,7 @@ jobs:
         uses: lycheeverse/lychee-action@e7477775783ea5526144ba13e8db5eec57747ce8 # v2.9.0
         with:
           # Restricted to http(s) on purpose. Much of docs/src is generated
-          # API reference (the js/ tree comes from `npm run docs` in nodejs)
+          # API reference (the js/ tree comes from `pnpm run docs` in nodejs)
           # and the hand-written pages use mkdocstrings cross-references and
           # nav-relative paths that only resolve in the site mkdocs builds,
           # not in this checkout, so relative links would be reported as

.github/workflows/docs.yml

@@ -55,9 +55,7 @@ jobs:
       - name: Set up node
         uses: actions/setup-node@v6
         with:
-          node-version: 20
-          cache: 'npm'
-          cache-dependency-path: docs/package-lock.json
+          node-version: 24
       - name: Install node dependencies
         working-directory: nodejs
         run: |

.github/workflows/nodejs.yml

@@ -47,9 +47,8 @@ jobs:
         version: 11.1.1
     - uses: actions/setup-node@v6
       with:
-        # pnpm 11 requires Node >= 22.13; use 24 since 22 hits EOL
-        # in October. The library itself still supports Node >= 18
-        # (see test matrix below).
+        # Build on a supported LTS; the matrix job below covers every
+        # Node version the library claims to support.
         node-version: 24
         cache: 'pnpm'
         cache-dependency-path: nodejs/pnpm-lock.yaml
@@ -84,7 +83,7 @@ jobs:
     timeout-minutes: 30
     strategy:
       matrix:
-        node-version: [ "18", "20" ]
+        node-version: [ "22", "24", "26" ]
     runs-on: "ubuntu-22.04"
     defaults:
       run:
@@ -101,9 +100,9 @@ jobs:
     - uses: actions/setup-node@v6
       name: Setup Node.js 24 for build
       with:
-        # pnpm 11 requires Node >= 22.13; use 24 since 22 hits EOL
-        # in October. Build/install runs on Node 24; tests run on the
-        # matrix version below using direct jest invocation.
+        # Build and install once on a fixed version so the generated docs
+        # are identical across matrix legs; the tests below then run on each
+        # supported Node version.
         node-version: 24
         cache: 'pnpm'
         cache-dependency-path: nodejs/pnpm-lock.yaml
@@ -152,9 +151,9 @@ jobs:
         S3_TEST: "1"
         # Newer @smithy/core uses dynamic ESM imports.
         NODE_OPTIONS: "--experimental-vm-modules"
-      # Invoke jest directly because pnpm 11 itself requires Node 22+
-      # while the matrix tests on older Node versions.
-      run: npx jest --verbose
+      # Invoke the installed jest binary directly; the pnpm shim is set up
+      # against the build-phase Node, not the version selected above.
+      run: node_modules/.bin/jest --verbose
     - name: Test examples
       working-directory: ./
       env:
@@ -164,7 +163,7 @@ jobs:
       run: |
         python ci/mock_openai.py &
         cd nodejs/examples
-        npx jest --testEnvironment jest-environment-node-single-context --verbose
+        node_modules/.bin/jest --testEnvironment jest-environment-node-single-context --verbose
   macos:
     timeout-minutes: 30
     # macos-15 ships a newer linker; the older macos-14 linker fails to insert
@@ -185,8 +184,7 @@ jobs:
         version: 11.1.1
     - uses: actions/setup-node@v6
       with:
-        # pnpm 11 requires Node >= 22.13; use 24 since 22 hits EOL
-        # in October.
+        # pnpm 11 requires Node >= 22.13.
         node-version: 24
         cache: 'pnpm'
         cache-dependency-path: nodejs/pnpm-lock.yaml

.github/workflows/npm-publish.yml

@@ -40,40 +40,31 @@ jobs:
           - target: aarch64-apple-darwin
             host: macos-latest
             features: fp16kernels
+            # Fat LTO was ~111 of this job's ~113 minutes.
+            lto: thin
+            codegen_units: 16
             pre_build: |-
               brew install protobuf
-              # Fat LTO (the workspace default in .cargo/config.toml) is
-              # single-threaded and is the peak-memory step of the build. On
-              # this runner it accounted for ~111 of the job's ~113 minutes,
-              # making it the critical path of the entire publish pipeline.
-              # ThinLTO parallelizes it across the runner's cores, for a few
-              # percent of runtime performance.
-              export CARGO_PROFILE_RELEASE_LTO=thin
-              export CARGO_PROFILE_RELEASE_CODEGEN_UNITS=16
           - target: x86_64-pc-windows-msvc
             host: windows-2025
             features: ","
+            # The lower peak also keeps this on the standard 4-core runner.
+            lto: thin
+            codegen_units: 16
             pre_build: |-
               choco install --no-progress protoc ninja nasm
               tail -n 1000 /c/ProgramData/chocolatey/logs/chocolatey.log
               # There is an issue where choco doesn't add nasm to the path
               export PATH="$PATH:/c/Program Files/NASM"
               nasm -v
-              # See the ThinLTO note on aarch64-apple-darwin above. Keeping
-              # peak memory down is also what lets this run on the standard
-              # 4-core runner: the 8-core larger runner was only needed to
-              # stop fat LTO from OOMing rustc-LLVM.
-              export CARGO_PROFILE_RELEASE_LTO=thin
-              export CARGO_PROFILE_RELEASE_CODEGEN_UNITS=16
           - target: aarch64-pc-windows-msvc
             host: windows-2025
             features: ","
+            lto: thin
+            codegen_units: 16
             pre_build: |-
                 choco install --no-progress protoc
                 rustup target add aarch64-pc-windows-msvc
-                # See the ThinLTO note on aarch64-apple-darwin above.
-                export CARGO_PROFILE_RELEASE_LTO=thin
-                export CARGO_PROFILE_RELEASE_CODEGEN_UNITS=16
           - target: x86_64-unknown-linux-gnu
             host: ubuntu-latest
             features: fp16kernels
@@ -103,6 +94,14 @@ jobs:
             # https://github.com/napi-rs/napi-rs/blob/main/debian-aarch64.Dockerfile
             docker: ghcr.io/napi-rs/napi-rs/nodejs-rust:lts-debian-aarch64
             features: "fp16kernels"
+            # Fat LTO OOM-killed rustc every nightly; even with lld it peaked
+            # at 31391 MiB of the runner's 32 GiB.
+            lto: thin
+            codegen_units: 16
+            # arm64 Linux links through GNU `ld` where x86_64 defaults to
+            # `rust-lld`, which is why only arm64 OOM'd. lld cut the largest
+            # linker process 7.0 -> 4.0 GiB (lancedb/sophon#7313).
+            linker: /tmp/aarch64-lld-clang
             pre_build: |-
               set -e &&
               apt-get update &&
@@ -112,9 +111,30 @@ jobs:
               # AT_HWCAP2 (added in Linux 3.17). Define it for aws-lc-sys.
               export CFLAGS="$CFLAGS -DAT_HWCAP2=26" &&
               rustup target add aarch64-unknown-linux-gnu
+              # Not `&&`-chained: in dash, errexit does not fire for a
+              # non-final command in an `&&` list, so failures were ignored.
+              #
+              # A wrapper rather than `-C link-arg` because the per-target
+              # rustflags variable does not reach every unit that links, while
+              # the linker variable does. `clang` because GCC silently ignores
+              # `-fuse-ld=lld` unless built with lld support. Two echoes
+              # because printf's newline escape gets rewritten to `;` between
+              # here and the container.
+              echo '#!/bin/sh' > /tmp/aarch64-lld-clang
+              echo 'exec clang --target=aarch64-unknown-linux-gnu --sysroot=/usr/aarch64-unknown-linux-gnu/aarch64-unknown-linux-gnu/sysroot --gcc-toolchain=/usr/aarch64-unknown-linux-gnu -fuse-ld=lld "$@"' >> /tmp/aarch64-lld-clang
+              chmod 0755 /tmp/aarch64-lld-clang
+              # Fail now, not at the cdylib link ~30 minutes later. Linking at
+              # all also proves lld resolved; clang errors out when it cannot.
+              echo 'int main(void){return 0;}' > /tmp/probe.c
+              /tmp/aarch64-lld-clang /tmp/probe.c -o /tmp/probe
+              readelf -h /tmp/probe | grep AArch64
           - target: aarch64-unknown-linux-musl
             host: ubuntu-2404-8x-x64
             features: ","
+            # Fat LTO took the whole runner down. lld cannot help: it died
+            # inside rustc's LLVM, before any linker was spawned.
+            lto: thin
+            codegen_units: 16
             pre_build: |-
               set -e &&
               sudo apt-get update &&
@@ -123,6 +143,19 @@ jobs:
               export EXTRA_ARGS="-x"
     name: build - ${{ matrix.settings.target }}
     runs-on: ${{ matrix.settings.host }}
+    # On the job, not exported from `pre_build`: `Swatinem/rust-cache` hashes
+    # `CARGO_*` into its cache key before any step runs, so a step-local export
+    # leaves the key unchanged while cargo still rebuilds cold. The ThinLTO
+    # legs had been doing that every run.
+    #
+    # Not `RUSTFLAGS`: setting it, even to "", discards every config-file
+    # rustflag, silently dropping .cargo/config.toml's `target-cpu` and
+    # `target-feature` from the published binaries.
+    env:
+      CARGO_PROFILE_RELEASE_LTO: ${{ matrix.settings.lto || 'fat' }}
+      CARGO_PROFILE_RELEASE_CODEGEN_UNITS: ${{ matrix.settings.codegen_units || '1' }}
+      # Empty elsewhere: a per-target variable is only read for that triple.
+      CARGO_TARGET_AARCH64_UNKNOWN_LINUX_GNU_LINKER: ${{ matrix.settings.linker }}
     defaults:
       run:
         working-directory: nodejs
@@ -135,8 +168,7 @@ jobs:
       - name: Setup node
         uses: actions/setup-node@v6
         with:
-          # pnpm 11 requires Node >= 22.13; use 24 since 22 hits EOL
-          # in October.
+          # pnpm 11 requires Node >= 22.13.
           node-version: 24
           cache: pnpm
           cache-dependency-path: nodejs/pnpm-lock.yaml
@@ -169,19 +201,15 @@ jobs:
           # creating ref). The nightly cadence also keeps entries inside
           # GitHub's 7-day eviction window, which a tag-only trigger would not.
           save-if: ${{ github.ref == 'refs/heads/main' }}
-      # Docker builds can use rust-cache too. `target/` already lives on the
-      # host because the whole workspace is bind-mounted into the container, and
-      # rust-cache's prune and save run host-side, so they can manage it -- which
-      # is what keeps the entry to dependency artifacts rather than a multi-GB
-      # copy of everything.
+      # Docker builds can use rust-cache too: the workspace is bind-mounted, so
+      # `target/` lives on the host and rust-cache's prune keeps the entry
+      # small.
       #
       # Two differences from the native builds. The container's CARGO_HOME is
-      # bind-mounted from `.cargo-cache` rather than the host's ~/.cargo, so that
-      # has to be cached explicitly. And the key is derived from the *host* rustc
-      # version, which is not the compiler that produced these artifacts; that is
-      # safe because cargo fingerprints the real compiler and rebuilds on a
-      # mismatch, it just means a base-image toolchain bump costs one cold build
-      # instead of invalidating the key.
+      # bind-mounted from `.cargo-cache` rather than ~/.cargo, so that is cached
+      # explicitly. And the key uses the *host* rustc version, not the compiler
+      # that built these artifacts -- safe, since cargo fingerprints the real
+      # one; a base-image bump just costs one cold build.
       - name: Cache cargo (docker builds)
         uses: Swatinem/rust-cache@v2
         if: ${{ matrix.settings.docker }}
@@ -210,14 +238,19 @@ jobs:
           # cache step above saves. Previously the registry mounts pointed at
           # `.cargo/...`, a path nothing cached, so the container re-downloaded
           # the whole crate registry on every run.
+          #
+          # `docker run` inherits nothing; `-e NAME` carries the job's `env:` in.
           options: "--user 0:0 -v ${{ github.workspace }}/.cargo-cache/git/db:/usr/local/cargo/git/db \
                     -v ${{ github.workspace }}/.cargo-cache/registry/cache:/usr/local/cargo/registry/cache \
                     -v ${{ github.workspace }}/.cargo-cache/registry/index:/usr/local/cargo/registry/index \
+                    -e CARGO_PROFILE_RELEASE_LTO \
+                    -e CARGO_PROFILE_RELEASE_CODEGEN_UNITS \
+                    -e CARGO_TARGET_AARCH64_UNKNOWN_LINUX_GNU_LINKER \
                     -v ${{ github.workspace }}:/build -w /build/nodejs"
           run: |
             set -e
             ${{ matrix.settings.pre_build }}
-            npx napi build --platform --release \
+            node_modules/.bin/napi build --platform --release \
               --features ${{ matrix.settings.features }} \
               --target ${{ matrix.settings.target }} \
               --dts ../lancedb/native.d.ts \
@@ -237,7 +270,7 @@ jobs:
       - name: Build
         run: |
           ${{ matrix.settings.pre_build }}
-          npx napi build --platform --release \
+          node_modules/.bin/napi build --platform --release \
               --features ${{ matrix.settings.features }} \
               --target ${{ matrix.settings.target }} \
               --dts ../lancedb/native.d.ts \
@@ -256,6 +289,18 @@ jobs:
         if: always()
         run: df -h
         shell: bash
+      - name: Report peak memory
+        if: always() && runner.os == 'Linux'
+        shell: bash
+        run: |
+          peak=$(find /sys/fs/cgroup -name memory.peak -readable \
+                 -exec cat {} + 2>/dev/null | sort -n | tail -1)
+          if [ -n "$peak" ]; then
+            echo "peak memory: $((peak / 1024 / 1024)) MiB"
+          else
+            echo "peak memory: unavailable (no readable cgroup v2 memory.peak)"
+          fi
+          free -g || true
       - name: Upload artifact
         uses: actions/upload-artifact@v7
         with:
@@ -293,7 +338,7 @@ jobs:
           - target: aarch64-unknown-linux-gnu
             host: ubuntu-2404-8x-arm64
         node:
-          - '20'
+          - '22'
     runs-on: ${{ matrix.settings.host }}
     defaults:
       run:
@@ -339,9 +384,9 @@ jobs:
       - name: Move built files
         run: cp dist/native.d.ts dist/native.js dist/*.node lancedb/
       - name: Test bindings
-        # Invoke jest directly because pnpm 11 itself requires Node 22+
-        # while the matrix tests on older Node versions.
-        run: npx jest --verbose
+        # Invoke the installed jest binary directly; the pnpm shim is set up
+        # against the install-phase Node, not the version selected above.
+        run: node_modules/.bin/jest --verbose
   publish:
     name: Publish
     runs-on: ubuntu-latest

.github/workflows/update_package_lock_run.yml

@@ -1,22 +0,0 @@
-name: Update package-lock.json
-
-on:
-  workflow_dispatch:
-
-permissions:
-  contents: read
-
-jobs:
-  publish:
-    runs-on: ubuntu-latest
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v6
-        with:
-          ref: main
-          persist-credentials: false
-          fetch-depth: 0
-          lfs: true
-      - uses: ./.github/workflows/update_package_lock
-        with:
-          github_token: ${{ secrets.LANCEDB_RELEASE_TOKEN }}

.github/workflows/update_package_lock_run_nodejs.yml

@@ -1,22 +0,0 @@
-name: Update NodeJs package-lock.json
-
-on:
-  workflow_dispatch:
-
-permissions:
-  contents: read
-
-jobs:
-  publish:
-    runs-on: ubuntu-latest
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v6
-        with:
-          ref: main
-          persist-credentials: false
-          fetch-depth: 0
-          lfs: true
-      - uses: ./.github/workflows/update_package_lock_nodejs
-        with:
-          github_token: ${{ secrets.LANCEDB_RELEASE_TOKEN }}

.pre-commit-config.yaml

@@ -20,7 +20,10 @@ repos:
     hooks:
       - id: local-biome-check
         name: biome check
-        entry: npx @biomejs/biome@1.8.3 check --config-path nodejs/biome.json nodejs/
+        # Use the biome from nodejs/package.json rather than a separately
+        # pinned one: the two drifted apart and disagreed on formatting, so
+        # this hook rejected code that `pnpm lint` accepted.
+        entry: nodejs/node_modules/.bin/biome check --config-path nodejs/biome.json nodejs/
         language: system
         types: [text]
         files: "nodejs/.*"

AGENTS.md

@@ -38,7 +38,7 @@ Before committing changes, run formatting for every language you touched. At min
 * Rust changes: run `cargo fmt --all`.
 * Python changes: run `ruff format .` and `ruff check .` from the repository root,
   and run targeted tests through `cd python && uv run ...`.
-* TypeScript changes: run the relevant `npm`/`pnpm` lint, format, build, and docs commands in `nodejs`.
+* TypeScript changes: run the relevant `pnpm` lint, format, build, and docs commands in `nodejs`.
 
 Before creating a PR, the exact value passed to `gh pr create --title` must follow
 Conventional Commits, such as `fix: support nested field paths in native index creation`
@@ -101,12 +101,12 @@ Python bindings changes:
 TypeScript bindings changes:
 
 1. Add napi-rs method binding on `Table` in `nodejs/src/table.rs`.
-2. Run `npm run build` to generate TypeScript definitions.
+2. Run `pnpm build` to generate TypeScript definitions.
 3. Add typescript method on abstract class `Table` in `nodejs/src/table.ts`.
 4. Add concrete method on `LocalTable` class in `nodejs/src/native_table.ts`.
     * Note: despite the name, this class is also used for remote tables.
 5. Add test in `nodejs/__test__/table.test.ts`.
-6. Run `npm run docs` to generate TypeScript documentation.
+6. Run `pnpm run docs` to generate TypeScript documentation.
 
 ## Python API reference
 

Cargo.lock

@@ -1597,9 +1597,9 @@ checksum = "613afe47fcd5fac7ccf1db93babcb082c5994d996f20b8b159f2ad1658eb5724"
 
 [[package]]
 name = "chacha20"
-version = "0.10.0"
+version = "0.10.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "6f8d983286843e49675a4b7a2d174efe136dc93a18d69130dd18198a6c167601"
+checksum = "65c35e4b699c7e15ccbe7ee35c005e4fc0a278d22238a2857e6ce2dadeda1b06"
 dependencies = [
  "cfg-if 1.0.4",
  "cpufeatures 0.3.0",
@@ -3455,8 +3455,8 @@ checksum = "42703706b716c37f96a77aea830392ad231f44c9e9a67872fa5548707e11b11c"
 
 [[package]]
 name = "fsst"
-version = "11.0.0-beta.22"
-source = "git+https://github.com/lance-format/lance.git?tag=v11.0.0-beta.22#ea3cb4d799c468232735e9bcb43959487aca5c20"
+version = "12.0.0-beta.5"
+source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.5#556637791d0048c2b4f1342dd84b67c8bbd65259"
 dependencies = [
  "arrow-array",
  "rand 0.9.5",
@@ -4815,8 +4815,8 @@ checksum = "e037a2e1d8d5fdbd49b16a4ea09d5d6401c1f29eca5ff29d03d3824dba16256a"
 
 [[package]]
 name = "lance"
-version = "11.0.0-beta.22"
-source = "git+https://github.com/lance-format/lance.git?tag=v11.0.0-beta.22#ea3cb4d799c468232735e9bcb43959487aca5c20"
+version = "12.0.0-beta.5"
+source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.5#556637791d0048c2b4f1342dd84b67c8bbd65259"
 dependencies = [
  "arc-swap",
  "arrow",
@@ -4888,8 +4888,8 @@ dependencies = [
 
 [[package]]
 name = "lance-arrow"
-version = "11.0.0-beta.22"
-source = "git+https://github.com/lance-format/lance.git?tag=v11.0.0-beta.22#ea3cb4d799c468232735e9bcb43959487aca5c20"
+version = "12.0.0-beta.5"
+source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.5#556637791d0048c2b4f1342dd84b67c8bbd65259"
 dependencies = [
  "arrow-array",
  "arrow-buffer",
@@ -4911,7 +4911,7 @@ dependencies = [
 [[package]]
 name = "lance-arrow-scalar"
 version = "58.0.0"
-source = "git+https://github.com/lance-format/lance.git?tag=v11.0.0-beta.22#ea3cb4d799c468232735e9bcb43959487aca5c20"
+source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.5#556637791d0048c2b4f1342dd84b67c8bbd65259"
 dependencies = [
  "arrow-array",
  "arrow-buffer",
@@ -4925,7 +4925,7 @@ dependencies = [
 [[package]]
 name = "lance-arrow-stats"
 version = "58.0.0"
-source = "git+https://github.com/lance-format/lance.git?tag=v11.0.0-beta.22#ea3cb4d799c468232735e9bcb43959487aca5c20"
+source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.5#556637791d0048c2b4f1342dd84b67c8bbd65259"
 dependencies = [
  "arrow-array",
  "arrow-schema",
@@ -4934,8 +4934,8 @@ dependencies = [
 
 [[package]]
 name = "lance-bitpacking"
-version = "11.0.0-beta.22"
-source = "git+https://github.com/lance-format/lance.git?tag=v11.0.0-beta.22#ea3cb4d799c468232735e9bcb43959487aca5c20"
+version = "12.0.0-beta.5"
+source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.5#556637791d0048c2b4f1342dd84b67c8bbd65259"
 dependencies = [
  "arrayref",
  "crunchy",
@@ -4945,8 +4945,8 @@ dependencies = [
 
 [[package]]
 name = "lance-core"
-version = "11.0.0-beta.22"
-source = "git+https://github.com/lance-format/lance.git?tag=v11.0.0-beta.22#ea3cb4d799c468232735e9bcb43959487aca5c20"
+version = "12.0.0-beta.5"
+source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.5#556637791d0048c2b4f1342dd84b67c8bbd65259"
 dependencies = [
  "arrow-array",
  "arrow-buffer",
@@ -4983,8 +4983,8 @@ dependencies = [
 
 [[package]]
 name = "lance-datafusion"
-version = "11.0.0-beta.22"
-source = "git+https://github.com/lance-format/lance.git?tag=v11.0.0-beta.22#ea3cb4d799c468232735e9bcb43959487aca5c20"
+version = "12.0.0-beta.5"
+source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.5#556637791d0048c2b4f1342dd84b67c8bbd65259"
 dependencies = [
  "arrow",
  "arrow-array",
@@ -5013,8 +5013,8 @@ dependencies = [
 
 [[package]]
 name = "lance-datagen"
-version = "11.0.0-beta.22"
-source = "git+https://github.com/lance-format/lance.git?tag=v11.0.0-beta.22#ea3cb4d799c468232735e9bcb43959487aca5c20"
+version = "12.0.0-beta.5"
+source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.5#556637791d0048c2b4f1342dd84b67c8bbd65259"
 dependencies = [
  "arrow",
  "arrow-array",
@@ -5031,8 +5031,8 @@ dependencies = [
 
 [[package]]
 name = "lance-derive"
-version = "11.0.0-beta.22"
-source = "git+https://github.com/lance-format/lance.git?tag=v11.0.0-beta.22#ea3cb4d799c468232735e9bcb43959487aca5c20"
+version = "12.0.0-beta.5"
+source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.5#556637791d0048c2b4f1342dd84b67c8bbd65259"
 dependencies = [
  "proc-macro2",
  "quote",
@@ -5041,8 +5041,8 @@ dependencies = [
 
 [[package]]
 name = "lance-encoding"
-version = "11.0.0-beta.22"
-source = "git+https://github.com/lance-format/lance.git?tag=v11.0.0-beta.22#ea3cb4d799c468232735e9bcb43959487aca5c20"
+version = "12.0.0-beta.5"
+source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.5#556637791d0048c2b4f1342dd84b67c8bbd65259"
 dependencies = [
  "arrow-arith",
  "arrow-array",
@@ -5075,8 +5075,8 @@ dependencies = [
 
 [[package]]
 name = "lance-file"
-version = "11.0.0-beta.22"
-source = "git+https://github.com/lance-format/lance.git?tag=v11.0.0-beta.22#ea3cb4d799c468232735e9bcb43959487aca5c20"
+version = "12.0.0-beta.5"
+source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.5#556637791d0048c2b4f1342dd84b67c8bbd65259"
 dependencies = [
  "arrow-arith",
  "arrow-array",
@@ -5107,8 +5107,8 @@ dependencies = [
 
 [[package]]
 name = "lance-index"
-version = "11.0.0-beta.22"
-source = "git+https://github.com/lance-format/lance.git?tag=v11.0.0-beta.22#ea3cb4d799c468232735e9bcb43959487aca5c20"
+version = "12.0.0-beta.5"
+source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.5#556637791d0048c2b4f1342dd84b67c8bbd65259"
 dependencies = [
  "arc-swap",
  "arrow",
@@ -5172,8 +5172,8 @@ dependencies = [
 
 [[package]]
 name = "lance-index-core"
-version = "11.0.0-beta.22"
-source = "git+https://github.com/lance-format/lance.git?tag=v11.0.0-beta.22#ea3cb4d799c468232735e9bcb43959487aca5c20"
+version = "12.0.0-beta.5"
+source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.5#556637791d0048c2b4f1342dd84b67c8bbd65259"
 dependencies = [
  "arrow-array",
  "arrow-schema",
@@ -5195,8 +5195,8 @@ dependencies = [
 
 [[package]]
 name = "lance-io"
-version = "11.0.0-beta.22"
-source = "git+https://github.com/lance-format/lance.git?tag=v11.0.0-beta.22#ea3cb4d799c468232735e9bcb43959487aca5c20"
+version = "12.0.0-beta.5"
+source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.5#556637791d0048c2b4f1342dd84b67c8bbd65259"
 dependencies = [
  "arrow",
  "arrow-array",
@@ -5236,8 +5236,8 @@ dependencies = [
 
 [[package]]
 name = "lance-linalg"
-version = "11.0.0-beta.22"
-source = "git+https://github.com/lance-format/lance.git?tag=v11.0.0-beta.22#ea3cb4d799c468232735e9bcb43959487aca5c20"
+version = "12.0.0-beta.5"
+source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.5#556637791d0048c2b4f1342dd84b67c8bbd65259"
 dependencies = [
  "arrow-array",
  "arrow-schema",
@@ -5251,8 +5251,8 @@ dependencies = [
 
 [[package]]
 name = "lance-namespace"
-version = "11.0.0-beta.22"
-source = "git+https://github.com/lance-format/lance.git?tag=v11.0.0-beta.22#ea3cb4d799c468232735e9bcb43959487aca5c20"
+version = "12.0.0-beta.5"
+source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.5#556637791d0048c2b4f1342dd84b67c8bbd65259"
 dependencies = [
  "arrow",
  "async-trait",
@@ -5264,8 +5264,8 @@ dependencies = [
 
 [[package]]
 name = "lance-namespace-impls"
-version = "11.0.0-beta.22"
-source = "git+https://github.com/lance-format/lance.git?tag=v11.0.0-beta.22#ea3cb4d799c468232735e9bcb43959487aca5c20"
+version = "12.0.0-beta.5"
+source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.5#556637791d0048c2b4f1342dd84b67c8bbd65259"
 dependencies = [
  "arrow",
  "arrow-ipc",
@@ -5318,8 +5318,8 @@ dependencies = [
 
 [[package]]
 name = "lance-select"
-version = "11.0.0-beta.22"
-source = "git+https://github.com/lance-format/lance.git?tag=v11.0.0-beta.22#ea3cb4d799c468232735e9bcb43959487aca5c20"
+version = "12.0.0-beta.5"
+source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.5#556637791d0048c2b4f1342dd84b67c8bbd65259"
 dependencies = [
  "arrow-array",
  "arrow-buffer",
@@ -5333,8 +5333,8 @@ dependencies = [
 
 [[package]]
 name = "lance-table"
-version = "11.0.0-beta.22"
-source = "git+https://github.com/lance-format/lance.git?tag=v11.0.0-beta.22#ea3cb4d799c468232735e9bcb43959487aca5c20"
+version = "12.0.0-beta.5"
+source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.5#556637791d0048c2b4f1342dd84b67c8bbd65259"
 dependencies = [
  "arrow",
  "arrow-array",
@@ -5374,8 +5374,8 @@ dependencies = [
 
 [[package]]
 name = "lance-testing"
-version = "11.0.0-beta.22"
-source = "git+https://github.com/lance-format/lance.git?tag=v11.0.0-beta.22#ea3cb4d799c468232735e9bcb43959487aca5c20"
+version = "12.0.0-beta.5"
+source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.5#556637791d0048c2b4f1342dd84b67c8bbd65259"
 dependencies = [
  "arrow-array",
  "arrow-schema",
@@ -5388,8 +5388,8 @@ dependencies = [
 
 [[package]]
 name = "lance-tokenizer"
-version = "11.0.0-beta.22"
-source = "git+https://github.com/lance-format/lance.git?tag=v11.0.0-beta.22#ea3cb4d799c468232735e9bcb43959487aca5c20"
+version = "12.0.0-beta.5"
+source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.5#556637791d0048c2b4f1342dd84b67c8bbd65259"
 dependencies = [
  "frostem",
  "icu_segmenter",
@@ -5402,7 +5402,7 @@ dependencies = [
 
 [[package]]
 name = "lancedb"
-version = "0.38.0-beta.9"
+version = "0.38.0-beta.12"
 dependencies = [
  "ahash",
  "anyhow",
@@ -5490,7 +5490,7 @@ dependencies = [
 
 [[package]]
 name = "lancedb-nodejs"
-version = "0.38.0-beta.9"
+version = "0.38.0-beta.12"
 dependencies = [
  "arrow-array",
  "arrow-buffer",
@@ -5515,7 +5515,7 @@ dependencies = [
 
 [[package]]
 name = "lancedb-python"
-version = "0.38.0-beta.9"
+version = "0.38.0-beta.12"
 dependencies = [
  "arrow",
  "async-trait",

Cargo.toml

@@ -13,20 +13,20 @@ categories = ["database-implementations"]
 rust-version = "1.91.0"
 
 [workspace.dependencies]
-lance = { "version" = "=11.0.0-beta.22", default-features = false, "tag" = "v11.0.0-beta.22", "git" = "https://github.com/lance-format/lance.git" }
-lance-core = { "version" = "=11.0.0-beta.22", "tag" = "v11.0.0-beta.22", "git" = "https://github.com/lance-format/lance.git" }
-lance-datagen = { "version" = "=11.0.0-beta.22", "tag" = "v11.0.0-beta.22", "git" = "https://github.com/lance-format/lance.git" }
-lance-file = { "version" = "=11.0.0-beta.22", "tag" = "v11.0.0-beta.22", "git" = "https://github.com/lance-format/lance.git" }
-lance-io = { "version" = "=11.0.0-beta.22", default-features = false, "tag" = "v11.0.0-beta.22", "git" = "https://github.com/lance-format/lance.git" }
-lance-index = { "version" = "=11.0.0-beta.22", "tag" = "v11.0.0-beta.22", "git" = "https://github.com/lance-format/lance.git" }
-lance-linalg = { "version" = "=11.0.0-beta.22", "tag" = "v11.0.0-beta.22", "git" = "https://github.com/lance-format/lance.git" }
-lance-namespace = { "version" = "=11.0.0-beta.22", "tag" = "v11.0.0-beta.22", "git" = "https://github.com/lance-format/lance.git" }
-lance-namespace-impls = { "version" = "=11.0.0-beta.22", default-features = false, "tag" = "v11.0.0-beta.22", "git" = "https://github.com/lance-format/lance.git" }
-lance-table = { "version" = "=11.0.0-beta.22", "tag" = "v11.0.0-beta.22", "git" = "https://github.com/lance-format/lance.git" }
-lance-testing = { "version" = "=11.0.0-beta.22", "tag" = "v11.0.0-beta.22", "git" = "https://github.com/lance-format/lance.git" }
-lance-datafusion = { "version" = "=11.0.0-beta.22", "tag" = "v11.0.0-beta.22", "git" = "https://github.com/lance-format/lance.git" }
-lance-encoding = { "version" = "=11.0.0-beta.22", "tag" = "v11.0.0-beta.22", "git" = "https://github.com/lance-format/lance.git" }
-lance-arrow = { "version" = "=11.0.0-beta.22", "tag" = "v11.0.0-beta.22", "git" = "https://github.com/lance-format/lance.git" }
+lance = { "version" = "=12.0.0-beta.5", default-features = false, "tag" = "v12.0.0-beta.5", "git" = "https://github.com/lance-format/lance.git" }
+lance-core = { "version" = "=12.0.0-beta.5", "tag" = "v12.0.0-beta.5", "git" = "https://github.com/lance-format/lance.git" }
+lance-datagen = { "version" = "=12.0.0-beta.5", "tag" = "v12.0.0-beta.5", "git" = "https://github.com/lance-format/lance.git" }
+lance-file = { "version" = "=12.0.0-beta.5", "tag" = "v12.0.0-beta.5", "git" = "https://github.com/lance-format/lance.git" }
+lance-io = { "version" = "=12.0.0-beta.5", default-features = false, "tag" = "v12.0.0-beta.5", "git" = "https://github.com/lance-format/lance.git" }
+lance-index = { "version" = "=12.0.0-beta.5", "tag" = "v12.0.0-beta.5", "git" = "https://github.com/lance-format/lance.git" }
+lance-linalg = { "version" = "=12.0.0-beta.5", "tag" = "v12.0.0-beta.5", "git" = "https://github.com/lance-format/lance.git" }
+lance-namespace = { "version" = "=12.0.0-beta.5", "tag" = "v12.0.0-beta.5", "git" = "https://github.com/lance-format/lance.git" }
+lance-namespace-impls = { "version" = "=12.0.0-beta.5", default-features = false, "tag" = "v12.0.0-beta.5", "git" = "https://github.com/lance-format/lance.git" }
+lance-table = { "version" = "=12.0.0-beta.5", "tag" = "v12.0.0-beta.5", "git" = "https://github.com/lance-format/lance.git" }
+lance-testing = { "version" = "=12.0.0-beta.5", "tag" = "v12.0.0-beta.5", "git" = "https://github.com/lance-format/lance.git" }
+lance-datafusion = { "version" = "=12.0.0-beta.5", "tag" = "v12.0.0-beta.5", "git" = "https://github.com/lance-format/lance.git" }
+lance-encoding = { "version" = "=12.0.0-beta.5", "tag" = "v12.0.0-beta.5", "git" = "https://github.com/lance-format/lance.git" }
+lance-arrow = { "version" = "=12.0.0-beta.5", "tag" = "v12.0.0-beta.5", "git" = "https://github.com/lance-format/lance.git" }
 lancedb = { path = "rust/lancedb", default-features = false }
 ahash = "0.8"
 # Note that this one does not include pyarrow

Makefile

@@ -5,5 +5,5 @@ licenses:
 	cd python && cargo about generate ../about.hbs -o RUST_THIRD_PARTY_LICENSES.html -c ../about.toml
 	cd python && uv sync --all-extras && uv tool run pip-licenses --python .venv/bin/python --format=markdown --with-urls --output-file=PYTHON_THIRD_PARTY_LICENSES.md
 	cd nodejs && cargo about generate ../about.hbs -o RUST_THIRD_PARTY_LICENSES.html -c ../about.toml
-	cd nodejs && npx license-checker --markdown --out NODEJS_THIRD_PARTY_LICENSES.md
+	cd nodejs && pnpm dlx license-checker@25 --markdown --out NODEJS_THIRD_PARTY_LICENSES.md
 	cd java && ./mvnw license:aggregate-add-third-party -q

ci/update_lockfiles.sh

@@ -12,16 +12,12 @@ done
 # This updates the lockfile without building
 cargo metadata --quiet > /dev/null
 
-pushd nodejs || exit 1
-npm install --package-lock-only --silent
-popd
-
 if git diff --quiet --exit-code; then
   echo "No lockfile changes to commit; skipping amend."
 elif $AMEND; then
-  git add Cargo.lock nodejs/package-lock.json
+  git add Cargo.lock
   git commit --amend --no-edit
 else
-  git add Cargo.lock nodejs/package-lock.json
+  git add Cargo.lock
   git commit -m "Update lockfiles"
 fi

deny.toml

@@ -131,18 +131,13 @@ allow = [
     "BSD-3-Clause",
     "ISC",
     "Unicode-3.0",
-    "Unicode-DFS-2016",
     "Zlib",
     "CC0-1.0",
     "MPL-2.0",
     "BSL-1.0",
-    "OpenSSL",
     # 0BSD ("BSD Zero Clause") is effectively public domain — no attribution
     # required. Pulled in by `mock_instant`.
     "0BSD",
-    # bzip2-1.0.6 is the permissive upstream bzip2 license (BSD-like). Pulled
-    # in by `libbz2-rs-sys`, the pure-Rust bzip2 implementation.
-    "bzip2-1.0.6",
     # CDLA-Permissive-2.0 is a permissive data license used by `webpki-roots`
     # for the Mozilla CA root bundle. Data-only, distribution-compatible.
     "CDLA-Permissive-2.0",
@@ -150,12 +145,7 @@ allow = [
 confidence-threshold = 0.8
 # Per-crate license exceptions: allow a license for a specific crate only,
 # rather than globally via the `allow` list above.
-exceptions = [
-    # CDDL-1.0 (copyleft) is pulled in only as a dev/profiling dependency via
-    # `inferno` -> `pprof` -> `lance-testing`; it is a test dependency that we
-    # do not distribute, so scope the allowance to `inferno` alone.
-    { allow = ["CDDL-1.0"], crate = "inferno" },
-]
+exceptions = []
 # Crates whose license cannot be determined from Cargo metadata but whose
 # license we've manually confirmed from upstream. Keep this list minimal.
 [[licenses.clarify]]

docs/README.md

@@ -47,22 +47,24 @@ pytest -vv python/tests/docs
 
 ### Checking typescript examples
 
-The `@lancedb/lancedb` package must be built before running the tests:
+The examples depend on `@lancedb/lancedb` at `file:../dist`, so the package must be
+built before running the tests. This uses pnpm; see the
+[Typescript contributing guide](../nodejs/CONTRIBUTING.md) for the toolchain setup.
 
 ```shell
 pushd nodejs
-npm ci
-npm run build
+pnpm install
+pnpm build
 popd
 ```
 
-Then you can run the examples by going to the `nodejs/examples` directory and
-running the tests like a normal npm package:
+Then you can run the examples by going to the `nodejs/examples` directory, which is a
+separate pnpm package with its own lockfile:
 
 ```shell
 pushd nodejs/examples
-npm ci
-npm test
+pnpm install
+pnpm test
 popd
 ```
 
@@ -84,6 +86,7 @@ The new files should be checked into the repository.
 
 ```shell
 pushd nodejs
-npm run docs
+# `pnpm docs` would invoke pnpm's built-in `docs` command, not the script.
+pnpm run docs
 popd
 ```

docs/package-lock.json

@@ -1,135 +0,0 @@
-{
-  "name": "lancedb-docs-test",
-  "version": "1.0.0",
-  "lockfileVersion": 3,
-  "requires": true,
-  "packages": {
-    "": {
-      "name": "lancedb-docs-test",
-      "version": "1.0.0",
-      "license": "Apache 2",
-      "dependencies": {
-        "apache-arrow": "file:../node/node_modules/apache-arrow",
-        "vectordb": "file:../node"
-      },
-      "devDependencies": {
-        "@types/node": "^20.11.8",
-        "typescript": "^5.3.3"
-      }
-    },
-    "../node": {
-      "name": "vectordb",
-      "version": "0.21.2-beta.0",
-      "cpu": [
-        "x64",
-        "arm64"
-      ],
-      "license": "Apache-2.0",
-      "os": [
-        "darwin",
-        "linux",
-        "win32"
-      ],
-      "dependencies": {
-        "@neon-rs/load": "^0.0.74",
-        "axios": "^1.4.0"
-      },
-      "devDependencies": {
-        "@neon-rs/cli": "^0.0.160",
-        "@types/chai": "^4.3.4",
-        "@types/chai-as-promised": "^7.1.5",
-        "@types/mocha": "^10.0.1",
-        "@types/node": "^18.16.2",
-        "@types/sinon": "^10.0.15",
-        "@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",
-        "eslint": "^8.39.0",
-        "eslint-config-standard-with-typescript": "^34.0.1",
-        "eslint-plugin-import": "^2.26.0",
-        "eslint-plugin-n": "^15.7.0",
-        "eslint-plugin-promise": "^6.1.1",
-        "mocha": "^10.2.0",
-        "openai": "^4.24.1",
-        "sinon": "^15.1.0",
-        "temp": "^0.9.4",
-        "ts-node": "^10.9.1",
-        "ts-node-dev": "^2.0.0",
-        "typedoc": "^0.24.7",
-        "typedoc-plugin-markdown": "^3.15.3",
-        "typescript": "^5.1.0",
-        "uuid": "^9.0.0"
-      },
-      "optionalDependencies": {
-        "@lancedb/vectordb-darwin-arm64": "0.21.2-beta.0",
-        "@lancedb/vectordb-darwin-x64": "0.21.2-beta.0",
-        "@lancedb/vectordb-linux-arm64-gnu": "0.21.2-beta.0",
-        "@lancedb/vectordb-linux-x64-gnu": "0.21.2-beta.0",
-        "@lancedb/vectordb-win32-x64-msvc": "0.21.2-beta.0"
-      },
-      "peerDependencies": {
-        "@apache-arrow/ts": "^14.0.2",
-        "apache-arrow": "^14.0.2"
-      }
-    },
-    "../node/node_modules/apache-arrow": {
-      "version": "14.0.2",
-      "license": "Apache-2.0",
-      "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/@types/node": {
-      "version": "20.11.8",
-      "resolved": "https://registry.npmjs.org/@types/node/-/node-20.11.8.tgz",
-      "integrity": "sha512-i7omyekpPTNdv4Jb/Rgqg0RU8YqLcNsI12quKSDkRXNfx7Wxdm6HhK1awT3xTgEkgxPn3bvnSpiEAc7a7Lpyow==",
-      "dev": true,
-      "dependencies": {
-        "undici-types": "~5.26.4"
-      }
-    },
-    "node_modules/apache-arrow": {
-      "resolved": "../node/node_modules/apache-arrow",
-      "link": true
-    },
-    "node_modules/typescript": {
-      "version": "5.3.3",
-      "resolved": "https://registry.npmjs.org/typescript/-/typescript-5.3.3.tgz",
-      "integrity": "sha512-pXWcraxM0uxAS+tN0AG/BF2TyqmHO014Z070UsJ+pFvYuRSq8KH8DmWpnbXe0pEPDHXZV3FcAbJkijJ5oNEnWw==",
-      "dev": true,
-      "bin": {
-        "tsc": "bin/tsc",
-        "tsserver": "bin/tsserver"
-      },
-      "engines": {
-        "node": ">=14.17"
-      }
-    },
-    "node_modules/undici-types": {
-      "version": "5.26.5",
-      "resolved": "https://registry.npmjs.org/undici-types/-/undici-types-5.26.5.tgz",
-      "integrity": "sha512-JlCMO+ehdEIKqlFxk6IfVoAUVmgz7cU7zD/h9XZ0qzeosSHmUJVOzSQvvYSYWXkFXC+IfLKSIffhv0sVZup6pA==",
-      "dev": true
-    },
-    "node_modules/vectordb": {
-      "resolved": "../node",
-      "link": true
-    }
-  }
-}

docs/package.json

@@ -1,20 +0,0 @@
-{
-  "name": "lancedb-docs-test",
-  "version": "1.0.0",
-  "description": "auto-generated tests from doc",
-  "author": "dev@lancedb.com",
-  "license": "Apache 2",
-  "dependencies": {
-    "apache-arrow": "file:../node/node_modules/apache-arrow",
-    "vectordb": "file:../node"
-  },
-  "scripts": {
-    "build": "tsc -b && cd ../node && npm run build-release",
-    "example": "npm run build && node",
-    "test": "npm run build && ls dist/*.js | xargs -n 1 node"
-  },
-  "devDependencies": {
-    "@types/node": "^20.11.8",
-    "typescript": "^5.3.3"
-  }
-}

docs/src/java/java.md

@@ -14,7 +14,7 @@ Add the following dependency to your `pom.xml`:
 <dependency>
     <groupId>com.lancedb</groupId>
     <artifactId>lancedb-core</artifactId>
-    <version>0.38.0-beta.10</version>
+    <version>0.38.0-beta.12</version>
 </dependency>
 ```
 

docs/src/js/classes/AutoQuery.md

@@ -0,0 +1,518 @@
+[**@lancedb/lancedb**](../README.md) • **Docs**
+
+***
+
+[@lancedb/lancedb](../globals.md) / AutoQuery
+
+# Class: AutoQuery
+
+A builder for automatic string searches.
+
+Automatic search determines whether to use full-text or vector search from
+the table revision selected for each execution. This builder exposes the
+common operations supported by both query families.
+
+## Extends
+
+- `StandardQueryBase`<`NativeQuery` \| `NativeVectorQuery`>
+
+## Properties
+
+### inner
+
+```ts
+protected inner: Query | VectorQuery | Promise<Query | VectorQuery>;
+```
+
+#### Inherited from
+
+`StandardQueryBase.inner`
+
+## Methods
+
+### analyzePlan()
+
+```ts
+analyzePlan(distributedMetrics?): Promise<string>
+```
+
+Executes the query and returns the physical query plan annotated with runtime metrics.
+
+This is useful for debugging and performance analysis, as it shows how the query was executed
+and includes metrics such as elapsed time, rows processed, and I/O statistics.
+
+#### Parameters
+
+* **distributedMetrics?**: [`AnalyzePlanDistributedMetrics`](../type-aliases/AnalyzePlanDistributedMetrics.md)
+    How distributed worker metrics are displayed for remote query plans.
+    Defaults to `"aggregate"`.
+
+#### Returns
+
+`Promise`&lt;`string`&gt;
+
+A query execution plan with runtime metrics for each step.
+
+#### Example
+
+```ts
+import * as lancedb from "@lancedb/lancedb"
+
+const db = await lancedb.connect("./.lancedb");
+const table = await db.createTable("my_table", [
+  { vector: [1.1, 0.9], id: "1" },
+]);
+
+const plan = await table.query().nearestTo([0.5, 0.2]).analyzePlan();
+
+Example output (with runtime metrics inlined):
+AnalyzeExec verbose=true, metrics=[]
+ ProjectionExec: expr=[id@3 as id, vector@0 as vector, _distance@2 as _distance], metrics=[output_rows=1, elapsed_compute=3.292µs]
+  Take: columns="vector, _rowid, _distance, (id)", metrics=[output_rows=1, elapsed_compute=66.001µs, batches_processed=1, bytes_read=8, iops=1, requests=1]
+   CoalesceBatchesExec: target_batch_size=1024, metrics=[output_rows=1, elapsed_compute=3.333µs]
+    GlobalLimitExec: skip=0, fetch=10, metrics=[output_rows=1, elapsed_compute=167ns]
+     FilterExec: _distance@2 IS NOT NULL, metrics=[output_rows=1, elapsed_compute=8.542µs]
+      SortExec: TopK(fetch=10), expr=[_distance@2 ASC NULLS LAST], metrics=[output_rows=1, elapsed_compute=63.25µs, row_replacements=1]
+       KNNVectorDistance: metric=l2, metrics=[output_rows=1, elapsed_compute=114.333µs, output_batches=1]
+        LanceScan: uri=/path/to/data, projection=[vector], row_id=true, row_addr=false, ordered=false, metrics=[output_rows=1, elapsed_compute=103.626µs, bytes_read=549, iops=2, requests=2]
+```
+
+#### Inherited from
+
+`StandardQueryBase.analyzePlan`
+
+***
+
+### execute()
+
+```ts
+protected execute(options?): AsyncGenerator<RecordBatch<any>, void, unknown>
+```
+
+Execute the query and return the results as an
+
+#### Parameters
+
+* **options?**: `Partial`&lt;[`QueryExecutionOptions`](../interfaces/QueryExecutionOptions.md)&gt;
+
+#### Returns
+
+`AsyncGenerator`&lt;`RecordBatch`&lt;`any`&gt;, `void`, `unknown`&gt;
+
+#### See
+
+ - AsyncIterator
+of
+ - 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)
+
+#### Inherited from
+
+`StandardQueryBase.execute`
+
+***
+
+### explainPlan()
+
+```ts
+explainPlan(verbose): Promise<string>
+```
+
+Generates an explanation of the query execution plan.
+
+#### Parameters
+
+* **verbose**: `boolean` = `false`
+    If true, provides a more detailed explanation. Defaults to false.
+
+#### Returns
+
+`Promise`&lt;`string`&gt;
+
+A Promise that resolves to a string containing the query execution plan explanation.
+
+#### Example
+
+```ts
+import * as lancedb from "@lancedb/lancedb"
+const db = await lancedb.connect("./.lancedb");
+const table = await db.createTable("my_table", [
+  { vector: [1.1, 0.9], id: "1" },
+]);
+const plan = await table.query().nearestTo([0.5, 0.2]).explainPlan();
+```
+
+#### Inherited from
+
+`StandardQueryBase.explainPlan`
+
+***
+
+### fastSearch()
+
+```ts
+fastSearch(): this
+```
+
+Skip searching un-indexed data. This can make search faster, but will miss
+any data that is not yet indexed.
+
+Use [Table#optimize](Table.md#optimize) to index all un-indexed data.
+
+#### Returns
+
+`this`
+
+#### Inherited from
+
+`StandardQueryBase.fastSearch`
+
+***
+
+### ~~filter()~~
+
+```ts
+filter(predicate): this
+```
+
+A filter statement to be applied to this query.
+
+#### Parameters
+
+* **predicate**: `string`
+
+#### Returns
+
+`this`
+
+#### See
+
+where
+
+#### Deprecated
+
+Use `where` instead
+
+#### Inherited from
+
+`StandardQueryBase.filter`
+
+***
+
+### fullTextSearch()
+
+```ts
+fullTextSearch(query, options?): this
+```
+
+#### Parameters
+
+* **query**: `string` \| [`FullTextQuery`](../interfaces/FullTextQuery.md)
+
+* **options?**: `Partial`&lt;[`FullTextSearchOptions`](../interfaces/FullTextSearchOptions.md)&gt;
+
+#### Returns
+
+`this`
+
+#### Inherited from
+
+`StandardQueryBase.fullTextSearch`
+
+***
+
+### limit()
+
+```ts
+limit(limit): this
+```
+
+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.
+
+#### Parameters
+
+* **limit**: `number`
+
+#### Returns
+
+`this`
+
+#### Inherited from
+
+`StandardQueryBase.limit`
+
+***
+
+### offset()
+
+```ts
+offset(offset): this
+```
+
+Set the number of rows to skip before returning results.
+
+This is useful for pagination.
+
+#### Parameters
+
+* **offset**: `number`
+
+#### Returns
+
+`this`
+
+#### Inherited from
+
+`StandardQueryBase.offset`
+
+***
+
+### orderBy()
+
+```ts
+orderBy(ordering): this
+```
+
+Sort the results by the specified column(s).
+
+#### Parameters
+
+* **ordering**: [`ColumnOrdering`](../interfaces/ColumnOrdering.md) \| [`ColumnOrdering`](../interfaces/ColumnOrdering.md)[]
+
+#### Returns
+
+`this`
+
+This query builder.
+
+#### Inherited from
+
+`StandardQueryBase.orderBy`
+
+***
+
+### outputSchema()
+
+```ts
+outputSchema(): Promise<Schema<any>>
+```
+
+Returns the schema of the output that will be returned by this query.
+
+This can be used to inspect the types and names of the columns that will be
+returned by the query before executing it.
+
+#### Returns
+
+`Promise`&lt;`Schema`&lt;`any`&gt;&gt;
+
+An Arrow Schema describing the output columns.
+
+#### Inherited from
+
+`StandardQueryBase.outputSchema`
+
+***
+
+### select()
+
+```ts
+select(columns): this
+```
+
+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:
+
+#### Parameters
+
+* **columns**: `string` \| `string`[] \| `Record`&lt;`string`, `string`&gt; \| `Map`&lt;`string`, `string`&gt;
+
+#### Returns
+
+`this`
+
+#### Example
+
+```ts
+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.
+```
+
+#### Inherited from
+
+`StandardQueryBase.select`
+
+***
+
+### toArray()
+
+```ts
+toArray(options?): Promise<any[]>
+```
+
+Collect the results as an array of objects.
+
+#### Parameters
+
+* **options?**: `Partial`&lt;[`QueryExecutionOptions`](../interfaces/QueryExecutionOptions.md)&gt;
+
+#### Returns
+
+`Promise`&lt;`any`[]&gt;
+
+#### Inherited from
+
+`StandardQueryBase.toArray`
+
+***
+
+### toArrow()
+
+```ts
+toArrow(options?): Promise<Table<any>>
+```
+
+Collect the results as an Arrow
+
+#### Parameters
+
+* **options?**: `Partial`&lt;[`QueryExecutionOptions`](../interfaces/QueryExecutionOptions.md)&gt;
+
+#### Returns
+
+`Promise`&lt;`Table`&lt;`any`&gt;&gt;
+
+#### See
+
+ArrowTable.
+
+#### Inherited from
+
+`StandardQueryBase.toArrow`
+
+***
+
+### useLsm()
+
+```ts
+useLsm(enable): this
+```
+
+Control MemWAL read routing for this query.
+
+By default (unset), when the table carries a MemWAL write spec (see
+[Table#setLsmWriteSpec](Table.md#setlsmwritespec)), reads are routed through the LSM scanner so
+they also return data written via the `mergeInsert` LSM path that has not yet
+been compacted into the base table (the active/frozen in-memory memtables and
+the flushed generations), deduplicated by primary key; a table without a spec
+reads the base table.
+
+#### Parameters
+
+* **enable**: `boolean`
+    `true` forces the LSM scanner and errors if the table has no
+    MemWAL write spec. `false` bypasses the MemWAL and reads the base table only,
+    even when a spec is present.
+    Note: the LSM scanner does not support every query shape (e.g. reranking,
+    hybrid search, `orderBy`). On a MemWAL table those shapes error unless
+    `useLsm(false)` is set, because a base-only read would silently exclude
+    un-compacted MemWAL data.
+
+#### Returns
+
+`this`
+
+#### Inherited from
+
+`StandardQueryBase.useLsm`
+
+***
+
+### where()
+
+```ts
+where(predicate): this
+```
+
+A filter statement to be applied to this query.
+
+The filter should be supplied as an SQL query string.  For example:
+
+#### Parameters
+
+* **predicate**: `string`
+
+#### Returns
+
+`this`
+
+#### Example
+
+```ts
+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).
+
+Calling this multiple times combines the filters with a logical AND rather
+than replacing the previous filter.
+```
+
+#### Inherited from
+
+`StandardQueryBase.where`
+
+***
+
+### withRowId()
+
+```ts
+withRowId(): this
+```
+
+Whether to return the row id in the results.
+
+This column can be used to match results between different queries. For
+example, to match results from a full text search and a vector search in
+order to perform hybrid search.
+
+#### Returns
+
+`this`
+
+#### Inherited from
+
+`StandardQueryBase.withRowId`

docs/src/js/classes/Table.md

@@ -942,7 +942,7 @@ Get the schema of the table.
 abstract search(
    query,
    queryType?,
-   ftsColumns?): Query | VectorQuery
+   ftsColumns?): Query | VectorQuery | AutoQuery
 ```
 
 Create a search query to find the nearest neighbors
@@ -964,7 +964,7 @@ of the given query
 
 #### Returns
 
-[`Query`](Query.md) \| [`VectorQuery`](VectorQuery.md)
+[`Query`](Query.md) \| [`VectorQuery`](VectorQuery.md) \| [`AutoQuery`](AutoQuery.md)
 
 ***
 
@@ -1292,6 +1292,18 @@ abstract updateFieldMetadata(updates): Promise<UpdateFieldMetadataResult>
 
 Update per-field (column) metadata.
 
+The following keys are treated specially, by convention, and should be
+used when appropriate:
+
+- `lancedb:description`: for a human-readable description of a field.
+- `lancedb:tag:<name>`: for a user-defined key-value tag, where the suffix
+  names the tag category; e.g. `lancedb:tag:model: "clip"`.
+- `lancedb:logical-column`: for a column grouping; e.g. `feature_v1` and
+  `feature_v2` might be in the same logical column.
+- `lancedb:status`: for status options (`production`, `candidate`,
+  `deprecated`, `archived`) to designate the current life cycle state of
+  this column.
+
 #### Parameters
 
 * **updates**: [`FieldMetadataUpdate`](../interfaces/FieldMetadataUpdate.md)[]

docs/src/js/globals.md

@@ -18,6 +18,7 @@
 
 ## Classes
 
+- [AutoQuery](classes/AutoQuery.md)
 - [BooleanQuery](classes/BooleanQuery.md)
 - [BoostQuery](classes/BoostQuery.md)
 - [BranchContents](classes/BranchContents.md)

docs/src/js/interfaces/FieldMetadataUpdate.md

@@ -17,7 +17,8 @@ metadata: Record<string, null | string>;
 ```
 
 Metadata key/value pairs. Merged into the field's existing metadata by
-default; a value of `null` deletes that key.
+default; a value of `null` deletes that key. See
+[Table.updateFieldMetadata](../classes/Table.md#updatefieldmetadata) for the conventional `lancedb:*` keys.
 
 ***
 

docs/src/js/namespaces/embedding/functions/getRegistry.md

@@ -10,16 +10,12 @@
 function getRegistry(): EmbeddingFunctionRegistry
 ```
 
-Utility function to get the global instance of the registry
+Get the global embedding function registry.
+
+LanceDB built-in providers are initialized when this public API is first
+used, so importing the root package does not change automatic search
+selection for tables without embedding metadata.
 
 ## Returns
 
 [`EmbeddingFunctionRegistry`](../classes/EmbeddingFunctionRegistry.md)
-
-`EmbeddingFunctionRegistry` The global instance of the registry
-
-## Example
-
-```ts
-const registry = getRegistry();
-const openai = registry.get("openai").create();

docs/src/python/python.md

@@ -159,6 +159,8 @@ and combined with [BooleanQuery][lancedb.query.BooleanQuery].
 
 ::: lancedb.query.FullTextOperator
 
+::: lancedb.query.DocumentGranularity
+
 ::: lancedb.query.Occur
 
 ## Embeddings
@@ -221,9 +223,13 @@ tokens = list(
 Blob columns store large binary values out of line so they can be read lazily
 instead of being materialized with the rest of the row.
 
-::: lancedb.blob
+`lancedb.BlobType` is `lance.blob.BlobType` when pylance is installed. Without
+pylance, LanceDB uses a matching `lance.blob.v2` extension type so blob columns
+still work. Queries return descriptors. Call
+[`fetch_blob_files`][lancedb.table.Table.fetch_blob_files] for lazy reads or
+[`fetch_blobs`][lancedb.table.Table.fetch_blobs] for eager bytes.
 
-::: lancedb.BlobType
+::: lancedb.blob
 
 ::: lancedb._blob.BlobFile
     options:

docs/tsconfig.json

@@ -1,17 +0,0 @@
-{
-  "include": [
-    "src/*.ts",
-  ],
-  "compilerOptions": {
-    "target": "es2022",
-    "module": "nodenext",
-    "declaration": true,
-    "outDir": "./dist",
-    "strict": true,
-    "allowJs": true,
-    "resolveJsonModule": true,
-  },
-  "exclude": [
-    "./dist/*",
-  ]
-}

java/lancedb-core/pom.xml

@@ -8,7 +8,7 @@
     <parent>
       <groupId>com.lancedb</groupId>
       <artifactId>lancedb-parent</artifactId>
-      <version>0.38.0-beta.10</version>
+      <version>0.38.0-beta.12</version>
       <relativePath>../pom.xml</relativePath>
     </parent>
 

java/pom.xml

@@ -6,7 +6,7 @@
 
     <groupId>com.lancedb</groupId>
     <artifactId>lancedb-parent</artifactId>
-    <version>0.38.0-beta.10</version>
+    <version>0.38.0-beta.12</version>
     <packaging>pom</packaging>
     <name>${project.artifactId}</name>
     <description>LanceDB Java SDK Parent POM</description>
@@ -28,7 +28,7 @@
     <properties>
         <project.build.sourceEncoding>UTF-8</project.build.sourceEncoding>
         <arrow.version>15.0.0</arrow.version>
-        <lance-core.version>11.0.0-beta.22</lance-core.version>
+        <lance-core.version>12.0.0-beta.5</lance-core.version>
         <spotless.skip>false</spotless.skip>
         <spotless.version>2.30.0</spotless.version>
         <spotless.java.googlejavaformat.version>1.7</spotless.java.googlejavaformat.version>

nodejs/Cargo.toml

@@ -1,7 +1,7 @@
 [package]
 name = "lancedb-nodejs"
 edition.workspace = true
-version = "0.38.0-beta.10"
+version = "0.38.0-beta.12"
 publish = false
 license.workspace = true
 description.workspace = true

nodejs/__test__/arrow.test.ts

@@ -1,11 +1,16 @@
 // SPDX-License-Identifier: Apache-2.0
 // SPDX-FileCopyrightText: Copyright The LanceDB Authors
+import * as fs from "node:fs";
+import * as vm from "node:vm";
 import * as arrow15 from "apache-arrow-15";
 import * as arrow16 from "apache-arrow-16";
 import * as arrow17 from "apache-arrow-17";
 import * as arrow18 from "apache-arrow-18";
 
 import {
+  Field as CurrentField,
+  LargeBinary as CurrentLargeBinary,
+  Schema as CurrentSchema,
   Vector as CurrentVector,
   convertToTable,
   tableFromIPC as currentTableFromIPC,
@@ -36,6 +41,59 @@ function sampleRecords(): Array<Record<string, any>> {
     },
   ];
 }
+
+it("serializes an Arrow Table created in another JavaScript realm", async () => {
+  const context = vm.createContext({
+    TextDecoder,
+    TextEncoder,
+    console,
+    setTimeout,
+    clearTimeout,
+  });
+  vm.runInContext(
+    fs.readFileSync(
+      require.resolve("apache-arrow-15/Arrow.es2015.min"),
+      "utf8",
+    ),
+    context,
+  );
+  const foreignTable: unknown = vm.runInContext(
+    "Arrow.tableFromArrays({ id: new Int32Array([1, 2, 3]), text: ['foo', 'bar', 'baz'] })",
+    context,
+  );
+
+  const foreignMetadata = (
+    foreignTable as { schema: { metadata: Map<string, string> } }
+  ).schema.metadata;
+  expect(foreignMetadata).not.toBeInstanceOf(Map);
+
+  const buf = await fromDataToBuffer(
+    foreignTable as Parameters<typeof fromDataToBuffer>[0],
+  );
+  const actual = currentTableFromIPC(buf);
+
+  expect(actual.numRows).toBe(3);
+  expect(actual.getChild("id")?.toJSON()).toEqual([1, 2, 3]);
+  expect(actual.getChild("text")?.toJSON()).toEqual(["foo", "bar", "baz"]);
+});
+
+it("preserves field metadata from a provided schema", async function () {
+  const jsonMetadata = new Map([["ARROW:extension:name", "lance.json"]]);
+  const schema = new CurrentSchema([
+    new CurrentField("meta", new CurrentLargeBinary(), true, jsonMetadata),
+  ]);
+
+  const table = makeArrowTable(
+    [{ meta: Buffer.from(JSON.stringify({ source: "test" })) }],
+    { schema },
+  );
+
+  expect(table.schema.fields[0].metadata).toEqual(jsonMetadata);
+
+  const roundTripped = currentTableFromIPC(await fromTableToBuffer(table));
+  expect(roundTripped.schema.fields[0].metadata).toEqual(jsonMetadata);
+});
+
 describe.each([arrow15, arrow16, arrow17, arrow18])(
   "Arrow",
   (

nodejs/__test__/embedding.test.ts

@@ -187,6 +187,58 @@ describe("embedding functions", () => {
     const vector0 = JSON.parse(JSON.stringify(arr[0].vector));
     expect(vector0).toEqual([1, 2, 3]);
   });
+  it("should append multiple Python embeddings with the same alias", async () => {
+    @register("python-mock")
+    // biome-ignore lint/correctness/noUnusedVariables: the decorator registers this class
+    class MockEmbeddingFunction extends EmbeddingFunction<string> {
+      ndims() {
+        return 3;
+      }
+      embeddingDataType(): Float {
+        return new Float32();
+      }
+      async computeQueryEmbeddings(_data: string) {
+        return [1, 2, 3];
+      }
+      async computeSourceEmbeddings(data: string[]) {
+        return data.map((value) =>
+          value === "hello world" ? [1, 2, 3] : [4, 5, 6],
+        );
+      }
+    }
+
+    const metadata = new Map([
+      [
+        "embedding_functions",
+        '[{"source_column":"text1","vector_column":"vector1","name":"python-mock","model":{}},{"source_column":"text2","vector_column":"vector2","name":"python-mock","model":{}}]',
+      ],
+    ]);
+    const schema = new Schema(
+      [
+        new Field("text1", new Utf8(), true),
+        new Field("text2", new Utf8(), true),
+        new Field(
+          "vector1",
+          new FixedSizeList(3, new Field("item", new Float32(), true)),
+          true,
+        ),
+        new Field(
+          "vector2",
+          new FixedSizeList(3, new Field("item", new Float32(), true)),
+          true,
+        ),
+      ],
+      metadata,
+    );
+
+    const db = await connect(tmpDir.name);
+    const table = await db.createEmptyTable("test", schema);
+    await table.add([{ text1: "hello world", text2: "goodbye world" }]);
+
+    const rows = await table.query().toArray();
+    expect(JSON.parse(JSON.stringify(rows[0].vector1))).toEqual([1, 2, 3]);
+    expect(JSON.parse(JSON.stringify(rows[0].vector2))).toEqual([4, 5, 6]);
+  });
 
   it("should append generated vectors to a non-nullable schema", async () => {
     @register("non_nullable_schema_test")

nodejs/__test__/embedding_registry.test.ts

@@ -0,0 +1,95 @@
+// SPDX-License-Identifier: Apache-2.0
+// SPDX-FileCopyrightText: Copyright The LanceDB Authors
+
+import { execFileSync } from "node:child_process";
+import { resolve } from "node:path";
+
+import type { OpenAIEmbeddingFunction } from "../lancedb/embedding/openai";
+import type { EmbeddingFunctionRegistry } from "../lancedb/embedding/registry";
+
+type EmbeddingModule = typeof import("../lancedb/embedding");
+type OpenAIModule = typeof import("../lancedb/embedding/openai");
+type RegistryModule = typeof import("../lancedb/embedding/registry");
+
+describe("embedding function registry", () => {
+  const registries: EmbeddingFunctionRegistry[] = [];
+
+  afterEach(() => {
+    for (const registry of registries) {
+      registry.reset();
+    }
+    registries.length = 0;
+  });
+
+  it("defers built-in providers until the public registry API is used", () => {
+    jest.isolateModules(() => {
+      const embedding = require("../lancedb/embedding") as EmbeddingModule;
+      const { getRegistry: getInternalRegistry } =
+        require("../lancedb/embedding/registry") as RegistryModule;
+      const registry = getInternalRegistry();
+      registries.push(registry);
+
+      expect(registry.length()).toBe(0);
+      expect(embedding.getRegistry()).toBe(registry);
+      expect(registry.get("openai")).toBeDefined();
+      expect(registry.get("huggingface")).toBeDefined();
+    });
+  });
+
+  it("preserves automatic FTS search in a fresh process", () => {
+    execFileSync(
+      process.execPath,
+      [resolve(__dirname, "fixtures", "auto_fts_search.cjs")],
+      { stdio: "pipe" },
+    );
+  });
+
+  it("shares registrations across duplicated provider module graphs", () => {
+    let registeringRegistry: EmbeddingFunctionRegistry | undefined;
+    let latestOpenAIConstructor: typeof OpenAIEmbeddingFunction | undefined;
+
+    jest.isolateModules(() => {
+      require("../lancedb/embedding/openai");
+      const { getRegistry } =
+        require("../lancedb/embedding/registry") as RegistryModule;
+      registeringRegistry = getRegistry();
+      registries.push(registeringRegistry);
+      expect(registeringRegistry.get("openai")).toBeDefined();
+    });
+
+    expect(() => {
+      jest.isolateModules(() => {
+        const { OpenAIEmbeddingFunction } =
+          require("../lancedb/embedding/openai") as OpenAIModule;
+        latestOpenAIConstructor = OpenAIEmbeddingFunction;
+        const { getRegistry } =
+          require("../lancedb/embedding/registry") as RegistryModule;
+        registries.push(getRegistry());
+      });
+    }).not.toThrow();
+
+    const previousApiKey = process.env.OPENAI_API_KEY;
+    process.env.OPENAI_API_KEY = "test";
+    try {
+      const latestOpenAI = registeringRegistry!
+        .get<OpenAIEmbeddingFunction>("openai")!
+        .create();
+      expect(latestOpenAI).toBeInstanceOf(latestOpenAIConstructor!);
+    } finally {
+      if (previousApiKey === undefined) {
+        delete process.env.OPENAI_API_KEY;
+      } else {
+        process.env.OPENAI_API_KEY = previousApiKey;
+      }
+    }
+
+    jest.isolateModules(() => {
+      const { getRegistry } =
+        require("../lancedb/embedding") as EmbeddingModule;
+      const publicRegistry = getRegistry();
+      registries.push(publicRegistry);
+      expect(publicRegistry).toBe(registeringRegistry);
+      expect(publicRegistry.get("openai")).toBeDefined();
+    });
+  });
+});

nodejs/__test__/fixtures/auto_fts_search.cjs

@@ -0,0 +1,33 @@
+// SPDX-License-Identifier: Apache-2.0
+// SPDX-FileCopyrightText: Copyright The LanceDB Authors
+
+const assert = require("node:assert/strict");
+const tmp = require("tmp");
+const { connect, embedding, Index } = require("../../dist");
+const { getRegistry } = require("../../dist/embedding/registry");
+
+async function main() {
+  assert.equal(typeof embedding.getRegistry, "function");
+  assert.equal(getRegistry().length(), 0);
+  assert.equal(embedding.getRegistry(), getRegistry());
+  assert.equal(getRegistry().length(), 2);
+
+  const dir = tmp.dirSync({ unsafeCleanup: true });
+  let db;
+  try {
+    db = await connect(dir.name);
+    const table = await db.createTable("docs", [{ text: "hello world" }]);
+    await table.createIndex("text", { config: Index.fts() });
+
+    const rows = await table.search("hello").toArray();
+    assert.equal(rows[0].text, "hello world");
+  } finally {
+    db?.close();
+    dir.removeCallback();
+  }
+}
+
+main().catch((error) => {
+  console.error(error);
+  process.exitCode = 1;
+});

nodejs/__test__/package.test.ts

@@ -5,8 +5,8 @@ import packageJson = require("../package.json");
 
 describe("package metadata", () => {
   it("requires Node.js type declarations compatible with the runtime", () => {
-    expect(packageJson.engines.node).toBe(">= 18");
-    expect(packageJson.peerDependencies["@types/node"]).toBe(">=18");
+    expect(packageJson.engines.node).toBe(">= 22");
+    expect(packageJson.peerDependencies["@types/node"]).toBe(">=22");
     expect(packageJson.peerDependenciesMeta["@types/node"]).toEqual({
       optional: true,
     });

nodejs/__test__/remote.test.ts

@@ -3,6 +3,7 @@
 
 import * as http from "http";
 import { RequestListener } from "http";
+import packageJson = require("../package.json");
 import {
   ClientConfig,
   Connection,
@@ -70,7 +71,13 @@ async function withMockDatabase(
   try {
     await callback(db);
   } finally {
-    server.close();
+    // `close()` alone leaves the port bound until keep-alive sockets drain, so
+    // a single failing test would cascade into EADDRINUSE for every test after
+    // it. Destroy the connections and wait for the port to actually be free.
+    await new Promise<void>((resolve) => {
+      server.closeAllConnections();
+      server.close(() => resolve());
+    });
   }
 }
 
@@ -131,7 +138,7 @@ describe("remote connection", () => {
       (req, res) => {
         expect(req.headers["x-api-key"]).toEqual("fake");
         expect(req.headers["user-agent"]).toEqual(
-          `LanceDB-Node-Client/${process.env.npm_package_version}`,
+          `LanceDB-Node-Client/${packageJson.version}`,
         );
 
         const body = JSON.stringify({ tables: [] });

nodejs/__test__/table.test.ts

@@ -11,10 +11,13 @@ import * as arrow17 from "apache-arrow-17";
 import * as arrow18 from "apache-arrow-18";
 
 import {
+  AutoQuery,
   Connection,
   MatchQuery,
   PhraseQuery,
+  Query,
   Table,
+  VectorQuery,
   connect,
   tokenize,
 } from "../lancedb";
@@ -682,6 +685,56 @@ describe.each([arrow15, arrow16, arrow17, arrow18])(
   },
 );
 
+// https://github.com/lancedb/lancedb/issues/1963
+it("should query documents with LangChain PDF metadata", async () => {
+  const tmpDir = tmp.dirSync({ unsafeCleanup: true });
+  try {
+    const db = await connect(tmpDir.name);
+    const documents = [
+      {
+        text: "first page",
+        vector: [1, 0],
+        source: "first.pdf",
+        loc: { pageNumber: 1, lines: { from: 1, to: 12 } },
+        pdf: {
+          version: "1.10.100",
+          info: {
+            format: "PDF 1.7",
+            producer: "pdf.js",
+            creator: "Writer",
+          },
+          totalPages: 2,
+        },
+      },
+      {
+        text: "second page",
+        vector: [0, 1],
+        source: "second.pdf",
+        loc: { pageNumber: 2, lines: { from: 13, to: 24 } },
+        pdf: {
+          version: "1.10.100",
+          info: {
+            format: "PDF 1.7",
+            producer: "pdf.js",
+            creator: "Writer",
+          },
+          totalPages: 2,
+        },
+      },
+    ];
+    const documentsTable = await db.createTable("documents", documents);
+
+    const results = await documentsTable.query().toArray();
+
+    expect(results).toHaveLength(2);
+    expect(results[0].source).toBe("first.pdf");
+    expect(results[0].pdf.info.producer).toBe("pdf.js");
+    expect(results[1].loc.pageNumber).toBe(2);
+  } finally {
+    tmpDir.removeCallback();
+  }
+});
+
 describe("merge insert", () => {
   let tmpDir: tmp.DirResult;
   let table: Table;
@@ -1777,6 +1830,194 @@ describe("Read consistency interval", () => {
   });
 });
 
+describe("automatic search schema consistency", () => {
+  let tmpDir: tmp.DirResult;
+
+  class SchemaRefreshEmbedding extends EmbeddingFunction<string> {
+    ndims() {
+      return 2;
+    }
+
+    embeddingDataType() {
+      return new Float32();
+    }
+
+    async computeSourceEmbeddings(data: string[]) {
+      return data.map((value) => [value.length, 1]);
+    }
+
+    async computeQueryEmbeddings(value: string) {
+      return [value.length, 1];
+    }
+  }
+
+  function embeddingSchema() {
+    const func = new SchemaRefreshEmbedding();
+    return LanceSchema({
+      text: func.sourceField(new Utf8()),
+      vector: func.vectorField(),
+    });
+  }
+
+  beforeEach(() => {
+    getRegistry().reset();
+    register("schema-refresh")(SchemaRefreshEmbedding);
+    tmpDir = tmp.dirSync({ unsafeCleanup: true });
+  });
+
+  afterEach(() => {
+    getRegistry().reset();
+    tmpDir.removeCallback();
+  });
+
+  it("uses the schema refreshed from another connection", async () => {
+    const first = await connect(tmpDir.name, { readConsistencyInterval: 0 });
+    const second = await connect(tmpDir.name, { readConsistencyInterval: 0 });
+
+    try {
+      const stale = await first.createTable("docs", [{ text: "before" }], {
+        schema: embeddingSchema(),
+      });
+      const replacement = await second.createTable(
+        "docs",
+        [{ text: "after hello" }],
+        { mode: "overwrite" },
+      );
+      await replacement.createIndex("text", { config: Index.fts() });
+
+      const search = stale.search("hello");
+      expect(search).toBeInstanceOf(AutoQuery);
+      expect(search).not.toBeInstanceOf(Query);
+      expect(search).not.toBeInstanceOf(VectorQuery);
+      expect("nprobes" in search).toBe(false);
+
+      const rows = await search.toArray();
+      expect(rows[0].text).toBe("after hello");
+      expect((await stale.schema()).metadata.has("embedding_functions")).toBe(
+        false,
+      );
+    } finally {
+      first.close();
+      second.close();
+    }
+  });
+
+  it("tracks embedding metadata across checkout and restore", async () => {
+    const first = await connect(tmpDir.name, { readConsistencyInterval: 0 });
+    const second = await connect(tmpDir.name, { readConsistencyInterval: 0 });
+
+    try {
+      await first.createTable("docs", [{ text: "before" }], {
+        schema: embeddingSchema(),
+      });
+      const table = await second.createTable(
+        "docs",
+        [{ text: "after hello" }],
+        { mode: "overwrite" },
+      );
+      await table.createIndex("text", { config: Index.fts() });
+
+      await table.checkout(1);
+      expect((await table.search("before").toArray())[0].text).toBe("before");
+
+      await table.checkoutLatest();
+      expect((await table.search("hello").toArray())[0].text).toBe(
+        "after hello",
+      );
+
+      await table.checkout(1);
+      await table.restore();
+      expect((await table.search("before").toArray())[0].text).toBe("before");
+    } finally {
+      first.close();
+      second.close();
+    }
+  });
+
+  it("pins automatic search while computing an embedding", async () => {
+    let markStarted!: () => void;
+    let releaseEmbedding!: () => void;
+    const started = new Promise<void>((resolve) => {
+      markStarted = resolve;
+    });
+    const released = new Promise<void>((resolve) => {
+      releaseEmbedding = resolve;
+    });
+
+    class BlockingEmbedding extends SchemaRefreshEmbedding {
+      async computeQueryEmbeddings(value: string) {
+        markStarted();
+        await released;
+        return [value.length, 1];
+      }
+    }
+
+    register("schema-refresh-blocking")(BlockingEmbedding);
+    const func = new BlockingEmbedding();
+    const schema = LanceSchema({
+      text: func.sourceField(new Utf8()),
+      vector: func.vectorField(),
+    });
+    const first = await connect(tmpDir.name, { readConsistencyInterval: 0 });
+    const second = await connect(tmpDir.name, { readConsistencyInterval: 0 });
+
+    try {
+      const table = await first.createTable(
+        "docs",
+        [{ text: "hello before" }],
+        { schema },
+      );
+      const pending = table.search("hello").toArray();
+      await started;
+
+      const replacement = await second.createTable(
+        "docs",
+        [{ text: "hello after" }],
+        { mode: "overwrite" },
+      );
+      await replacement.createIndex("text", { config: Index.fts() });
+      releaseEmbedding();
+
+      expect((await pending)[0].text).toBe("hello before");
+    } finally {
+      releaseEmbedding();
+      first.close();
+      second.close();
+    }
+  });
+
+  it("refreshes a reused automatic search for every execution", async () => {
+    const first = await connect(tmpDir.name, { readConsistencyInterval: 0 });
+    const second = await connect(tmpDir.name, { readConsistencyInterval: 0 });
+
+    try {
+      const table = await first.createTable("docs", [
+        { text: "hello before", marker: "before" },
+      ]);
+      await table.createIndex("text", { config: Index.fts() });
+      const search = table.search("hello").select(["text"]);
+
+      const before = (await search.toArray())[0];
+      expect(before.text).toBe("hello before");
+      expect(before.marker).toBeUndefined();
+
+      const replacement = await second.createTable(
+        "docs",
+        [{ text: "hello after", marker: "after" }],
+        { mode: "overwrite" },
+      );
+      await replacement.createIndex("text", { config: Index.fts() });
+
+      const after = (await search.toArray())[0];
+      expect(after.text).toBe("hello after");
+      expect(after.marker).toBeUndefined();
+    } finally {
+      first.close();
+      second.close();
+    }
+  });
+});
+
 describe("schema evolution", function () {
   let tmpDir: tmp.DirResult;
   beforeEach(() => {
@@ -2344,7 +2585,24 @@ describe.each([arrow15, arrow16, arrow17, arrow18])(
       );
     });
 
-    test("full text search if no embedding function provided", async () => {
+    test("full text search if only an unrelated embedding function is registered", async () => {
+      register("unused")(
+        class extends EmbeddingFunction<string> {
+          ndims() {
+            return 3;
+          }
+          embeddingDataType() {
+            return new Float32();
+          }
+          async computeQueryEmbeddings(_data: string) {
+            return [1, 2, 3];
+          }
+          async computeSourceEmbeddings(data: string[]) {
+            return data.map(() => [1, 2, 3]);
+          }
+        },
+      );
+
       const db = await connect(tmpDir.name);
       const data = [
         { text: "hello world", vector: [0.1, 0.2, 0.3] },
@@ -2366,6 +2624,306 @@ describe.each([arrow15, arrow16, arrow17, arrow18])(
       expect(results2[0].text).toBe(data[1].text);
     });
 
+    test("auto search stays consistent with the active revision", async () => {
+      let initCalls = 0;
+      let queryCalls = 0;
+      let markStarted!: () => void;
+      const started = new Promise<void>((resolve) => {
+        markStarted = resolve;
+      });
+      let releaseEmbedding!: () => void;
+      const embeddingReleased = new Promise<void>((resolve) => {
+        releaseEmbedding = resolve;
+      });
+
+      @register("refresh-test")
+      class TestEmbedding extends EmbeddingFunction<string> {
+        async init() {
+          initCalls += 1;
+        }
+        ndims() {
+          return 1;
+        }
+        embeddingDataType() {
+          return new arrow.Float32();
+        }
+        async computeQueryEmbeddings(value: string) {
+          queryCalls += 1;
+          if (value === "blocked") {
+            markStarted();
+            await embeddingReleased;
+          }
+          return value === "greetings" ? [0.1] : [0.2];
+        }
+        async computeSourceEmbeddings(values: string[]) {
+          return values.map((value) =>
+            value === "hello world" ? [0.1] : [0.2],
+          );
+        }
+      }
+
+      const writer = await connect(tmpDir.name);
+      await writer.createTable("test", [{ text: "plain", vector: [0.0] }]);
+      const reader = await connect(tmpDir.name, {
+        readConsistencyInterval: 0,
+      });
+      const tracked = await reader.openTable("test");
+      type SnapshotCountingNative = {
+        querySnapshot: () => Promise<unknown>;
+      };
+      const native = (tracked as unknown as { inner: SnapshotCountingNative })
+        .inner;
+      const querySnapshot = native.querySnapshot.bind(native);
+      let snapshotCalls = 0;
+      native.querySnapshot = async () => {
+        snapshotCalls += 1;
+        return await querySnapshot();
+      };
+      const autoQuery = tracked.search("greetings").select(["text"]).limit(1);
+
+      const func = new TestEmbedding();
+      const schema = LanceSchema({
+        text: func.sourceField(new arrow.Utf8()),
+        vector: func.vectorField(),
+      });
+      const data = [{ text: "hello world" }, { text: "goodbye world" }];
+      await writer.createTable("test", data, { mode: "overwrite", schema });
+      const baselineInitCalls = initCalls;
+
+      expect(
+        (await tracked.schema()).metadata.get("embedding_functions"),
+      ).toBeDefined();
+      const results = await autoQuery.toArray();
+      expect(results[0].text).toBe(data[0].text);
+      expect(initCalls).toBe(baselineInitCalls + 1);
+      expect(queryCalls).toBe(1);
+      expect(snapshotCalls).toBe(1);
+
+      const repeatedResults = await autoQuery.toArray();
+      expect(repeatedResults[0].text).toBe(data[0].text);
+      expect(initCalls).toBe(baselineInitCalls + 1);
+      expect(queryCalls).toBe(1);
+      expect(snapshotCalls).toBe(2);
+
+      const pending = tracked
+        .search("blocked")
+        .select(["text"])
+        .limit(1)
+        .toArray();
+      await started;
+
+      const ftsData = [
+        { text: "greetings from full text", vector: [0.0] },
+        { text: "blocked from full text", vector: [0.0] },
+      ];
+      const ftsTable = await writer.createTable("test", ftsData, {
+        mode: "overwrite",
+      });
+      await ftsTable.createIndex("text", { config: Index.fts() });
+      releaseEmbedding();
+
+      const pendingResults = await pending;
+      expect(pendingResults[0].text).toBe(data[1].text);
+
+      expect(
+        (await tracked.schema()).metadata.get("embedding_functions"),
+      ).toBeUndefined();
+      const ftsResults = await autoQuery.toArray();
+      expect(ftsResults[0].text).toBe(ftsData[0].text);
+    });
+
+    test("auto search keeps newer preparation during a revision race", async () => {
+      let aCalls = 0;
+      let bCalls = 0;
+      let markAStarted!: () => void;
+      const aStarted = new Promise<void>((resolve) => {
+        markAStarted = resolve;
+      });
+      let releaseA!: () => void;
+      const aReleased = new Promise<void>((resolve) => {
+        releaseA = resolve;
+      });
+      let markBStarted!: () => void;
+      const bStarted = new Promise<void>((resolve) => {
+        markBStarted = resolve;
+      });
+      let releaseB!: () => void;
+      const bReleased = new Promise<void>((resolve) => {
+        releaseB = resolve;
+      });
+
+      @register("race-a")
+      class EmbeddingA extends EmbeddingFunction<string> {
+        ndims() {
+          return 1;
+        }
+        embeddingDataType() {
+          return new arrow.Float32();
+        }
+        async computeQueryEmbeddings() {
+          aCalls += 1;
+          markAStarted();
+          await aReleased;
+          return [0.1];
+        }
+        async computeSourceEmbeddings(values: string[]) {
+          return values.map(() => [0.1]);
+        }
+      }
+
+      @register("race-b")
+      class EmbeddingB extends EmbeddingFunction<string> {
+        ndims() {
+          return 1;
+        }
+        embeddingDataType() {
+          return new arrow.Float32();
+        }
+        async computeQueryEmbeddings() {
+          bCalls += 1;
+          markBStarted();
+          await bReleased;
+          return [0.2];
+        }
+        async computeSourceEmbeddings(values: string[]) {
+          return values.map(() => [0.2]);
+        }
+      }
+
+      const writer = await connect(tmpDir.name);
+      const embeddingA = new EmbeddingA();
+      const schemaA = LanceSchema({
+        text: embeddingA.sourceField(new arrow.Utf8()),
+        vector: embeddingA.vectorField(),
+      });
+      await writer.createTable("race", [{ text: "revision a" }], {
+        schema: schemaA,
+      });
+      const reader = await connect(tmpDir.name, {
+        readConsistencyInterval: 0,
+      });
+      const tracked = await reader.openTable("race");
+      const query = tracked.search("query");
+
+      const first = query.toArray();
+      await aStarted;
+
+      const embeddingB = new EmbeddingB();
+      const schemaB = LanceSchema({
+        text: embeddingB.sourceField(new arrow.Utf8()),
+        vector: embeddingB.vectorField(),
+      });
+      await writer.createTable("race", [{ text: "revision b" }], {
+        mode: "overwrite",
+        schema: schemaB,
+      });
+      const second = query.toArray();
+      await bStarted;
+
+      releaseA();
+      releaseB();
+      await Promise.all([first, second]);
+      expect(aCalls).toBe(1);
+      expect(bCalls).toBe(1);
+    });
+
+    test("stale FTS routing keeps newer vector preparation", async () => {
+      let vectorCalls = 0;
+      let markVectorStarted!: () => void;
+      const vectorStarted = new Promise<void>((resolve) => {
+        markVectorStarted = resolve;
+      });
+      let releaseVector!: () => void;
+      const vectorReleased = new Promise<void>((resolve) => {
+        releaseVector = resolve;
+      });
+
+      @register("stale-fts-race")
+      class RaceEmbedding extends EmbeddingFunction<string> {
+        ndims() {
+          return 1;
+        }
+        embeddingDataType() {
+          return new arrow.Float32();
+        }
+        async computeQueryEmbeddings() {
+          vectorCalls += 1;
+          markVectorStarted();
+          await vectorReleased;
+          return [0.1];
+        }
+        async computeSourceEmbeddings(values: string[]) {
+          return values.map(() => [0.1]);
+        }
+      }
+
+      const writer = await connect(tmpDir.name);
+      const ftsTable = await writer.createTable("stale_fts", [
+        { text: "hello", vector: [0.0] },
+      ]);
+      await ftsTable.createIndex("text", { config: Index.fts() });
+
+      const reader = await connect(tmpDir.name, {
+        readConsistencyInterval: 0,
+      });
+      const tracked = await reader.openTable("stale_fts");
+      type Snapshot = {
+        schema: () => Promise<Buffer>;
+      };
+      type NativeWithSnapshot = {
+        querySnapshot: () => Promise<Snapshot>;
+      };
+      const native = (tracked as unknown as { inner: NativeWithSnapshot })
+        .inner;
+      const querySnapshot = native.querySnapshot.bind(native);
+      let snapshotCalls = 0;
+      let markStaleSchemaStarted!: () => void;
+      const staleSchemaStarted = new Promise<void>((resolve) => {
+        markStaleSchemaStarted = resolve;
+      });
+      let releaseStaleSchema!: () => void;
+      const staleSchemaReleased = new Promise<void>((resolve) => {
+        releaseStaleSchema = resolve;
+      });
+      native.querySnapshot = async () => {
+        const snapshot = await querySnapshot();
+        snapshotCalls += 1;
+        if (snapshotCalls === 1) {
+          const schema = snapshot.schema.bind(snapshot);
+          snapshot.schema = async () => {
+            markStaleSchemaStarted();
+            await staleSchemaReleased;
+            return await schema();
+          };
+        }
+        return snapshot;
+      };
+
+      const query = tracked.search("hello");
+      const staleFtsExecution = query.toArray();
+      await staleSchemaStarted;
+
+      const embedding = new RaceEmbedding();
+      const vectorSchema = LanceSchema({
+        text: embedding.sourceField(new arrow.Utf8()),
+        vector: embedding.vectorField(),
+      });
+      await writer.createTable("stale_fts", [{ text: "hello" }], {
+        mode: "overwrite",
+        schema: vectorSchema,
+      });
+
+      const vectorExecution = query.toArray();
+      await vectorStarted;
+      releaseStaleSchema();
+      await staleFtsExecution;
+      releaseVector();
+      await vectorExecution;
+
+      await query.toArray();
+      expect(vectorCalls).toBe(1);
+    });
+
     test("tokenizes FTS queries by column or index name", async () => {
       const db = await connect(tmpDir.name);
       const data = [
@@ -2916,6 +3474,30 @@ describe("column name options", () => {
     expect(results[1].query_index).toBe(1);
   });
 
+  test("observes promised additional vectors while the query is pending", async () => {
+    const initialVector = new Promise<number[]>(() => undefined);
+    const query = table.query().nearestTo(initialVector);
+    const unhandled: unknown[] = [];
+    const onUnhandled = (reason: unknown) => unhandled.push(reason);
+    process.on("unhandledRejection", onUnhandled);
+
+    try {
+      query.addQueryVector(Promise.reject(new Error("extra vector failed")));
+      await new Promise<void>((resolve) => setImmediate(resolve));
+      expect(unhandled).toEqual([]);
+
+      const rejectedQuery = table
+        .query()
+        .nearestTo([0.1, 0.2])
+        .addQueryVector(Promise.reject(new Error("consumed vector failed")));
+      await expect(rejectedQuery.toArray()).rejects.toThrow(
+        "consumed vector failed",
+      );
+    } finally {
+      process.off("unhandledRejection", onUnhandled);
+    }
+  });
+
   test("index and search multivectors", async () => {
     const db = await connect(tmpDir.name);
     const data = [];
@@ -2979,6 +3561,27 @@ describe("when creating an empty table", () => {
     expect((actualSchema.fields[1].type as Float64).precision).toBe(2);
   });
 
+  it("can add and query JSON data", async () => {
+    const schema = new Schema([
+      new Field("id", new Int32(), true),
+      new Field(
+        "meta",
+        new Utf8(),
+        true,
+        new Map([["ARROW:extension:name", "arrow.json"]]),
+      ),
+    ]);
+    const table = await con.createEmptyTable("json", schema);
+    const meta = JSON.stringify({ x: 1 });
+
+    await table.add([{ id: 1, meta }]);
+
+    const rows = await table.query().toArray();
+    expect(rows).toHaveLength(1);
+    expect(rows[0].id).toBe(1);
+    expect(rows[0].meta).toBe(meta);
+  });
+
   it("can create an empty table from schema that specifies field types by name", async () => {
     const schemaLike = {
       fields: [

nodejs/examples/basic.test.ts

@@ -170,7 +170,7 @@ test("basic table examples", async () => {
     // --8<-- [end:create_index]
 
     // --8<-- [start:delete_rows]
-    await tbl.delete('item = "fizz"');
+    await tbl.delete("item = 'fizz'");
     // --8<-- [end:delete_rows]
 
     // --8<-- [start:drop_table]

nodejs/examples/package.json

@@ -8,7 +8,8 @@
     "//1": "--experimental-vm-modules is needed to run jest with sentence-transformers",
     "//2": "--testEnvironment is needed to run jest with sentence-transformers",
     "//3": "See: https://github.com/huggingface/transformers.js/issues/57",
-    "test": "node --experimental-vm-modules node_modules/.bin/jest --testEnvironment jest-environment-node-single-context --verbose",
+    "//4": "jest is invoked by its JS entry, not node_modules/.bin/jest: under pnpm that path is a shell shim, which `node` cannot execute",
+    "test": "node --experimental-vm-modules node_modules/jest/bin/jest.js --testEnvironment jest-environment-node-single-context --verbose",
     "lint": "biome check *.ts && biome format *.ts",
     "lint-ci": "biome ci .",
     "lint-fix": "biome check --write *.ts && pnpm format",

nodejs/lancedb/arrow.ts

@@ -72,8 +72,7 @@ export type FieldLike =
     };
 
 export type DataLike =
-  // biome-ignore lint/suspicious/noExplicitAny: <explanation>
-  | import("apache-arrow").Data<Struct<any>>
+  | import("apache-arrow").Data
   | {
       // biome-ignore lint/suspicious/noExplicitAny: <explanation>
       type: any;
@@ -82,6 +81,7 @@ export type DataLike =
       stride: number;
       nullable: boolean;
       children: DataLike[];
+      dictionary?: { data: readonly DataLike[] };
       get nullCount(): number;
       // biome-ignore lint/suspicious/noExplicitAny: <explanation>
       values: Buffers<any>[BufferType.DATA];

nodejs/lancedb/embedding/index.ts

@@ -4,7 +4,15 @@
 import { Field, Schema } from "../arrow";
 import { sanitizeType } from "../sanitize";
 import { EmbeddingFunction } from "./embedding_function";
-import { EmbeddingFunctionConfig, getRegistry } from "./registry";
+import {
+  EmbeddingFunctionConfig,
+  EmbeddingFunctionRegistry,
+  getRegistry as getGlobalRegistry,
+  registerBuiltIn,
+} from "./registry";
+
+type OpenAIModule = typeof import("./openai");
+type TransformersModule = typeof import("./transformers");
 
 export {
   FieldOptions,
@@ -14,7 +22,39 @@ export {
   EmbeddingFunctionConstructor,
 } from "./embedding_function";
 
-export * from "./registry";
+export {
+  EmbeddingFunctionRegistry,
+  parseEmbeddingMetadata,
+  register,
+} from "./registry";
+export type {
+  CreateReturnType,
+  EmbeddingFunctionConfig,
+  EmbeddingFunctionCreate,
+  EmbeddingMetadataEntry,
+  ResolvedEmbeddingFunctionConfig,
+} from "./registry";
+
+function initializeBuiltInProviders() {
+  const { OpenAIEmbeddingFunction } = require("./openai") as OpenAIModule;
+  const { TransformersEmbeddingFunction } =
+    require("./transformers") as TransformersModule;
+
+  registerBuiltIn("openai", OpenAIEmbeddingFunction);
+  registerBuiltIn("huggingface", TransformersEmbeddingFunction);
+}
+
+/**
+ * Get the global embedding function registry.
+ *
+ * LanceDB built-in providers are initialized when this public API is first
+ * used, so importing the root package does not change automatic search
+ * selection for tables without embedding metadata.
+ */
+export function getRegistry(): EmbeddingFunctionRegistry {
+  initializeBuiltInProviders();
+  return getGlobalRegistry();
+}
 
 /**
  * Create a schema with embedding functions.

nodejs/lancedb/embedding/openai.ts

@@ -5,14 +5,13 @@ import type OpenAI from "openai";
 import type { EmbeddingCreateParams } from "openai/resources/index";
 import { Float, Float32 } from "../arrow";
 import { EmbeddingFunction } from "./embedding_function";
-import { register } from "./registry";
+import { registerBuiltIn } from "./registry";
 
 export type OpenAIOptions = {
   apiKey: string;
   model: EmbeddingCreateParams["model"];
 };
 
-@register("openai")
 export class OpenAIEmbeddingFunction extends EmbeddingFunction<
   string,
   Partial<OpenAIOptions>
@@ -100,3 +99,5 @@ export class OpenAIEmbeddingFunction extends EmbeddingFunction<
     return response.data[0].embedding;
   }
 }
+
+registerBuiltIn("openai", OpenAIEmbeddingFunction);

nodejs/lancedb/embedding/registry.ts

@@ -7,6 +7,10 @@ import {
 } from "./embedding_function";
 import "reflect-metadata";
 
+const builtInFunctionsKey = Symbol.for(
+  "@lancedb/lancedb::embedding-built-in-functions::v1",
+);
+
 export type CreateReturnType<T> = T extends { init: () => Promise<void> }
   ? Promise<T>
   : T;
@@ -59,6 +63,15 @@ export class EmbeddingFunctionRegistry {
     };
   }
 
+  /** @ignore */
+  setBuiltIn<
+    T extends EmbeddingFunctionConstructor = EmbeddingFunctionConstructor,
+  >(name: string, ctor: T): T {
+    this.#functions.set(name, ctor);
+    Reflect.defineMetadata("lancedb::embedding::name", name, ctor);
+    return ctor;
+  }
+
   get<T extends EmbeddingFunction<unknown>>(
     name: string,
   ): EmbeddingFunctionCreate<T> | undefined;
@@ -96,6 +109,7 @@ export class EmbeddingFunctionRegistry {
    */
   reset(this: EmbeddingFunctionRegistry) {
     this.#functions.clear();
+    getBuiltInFunctions(this).clear();
   }
 
   /**
@@ -183,12 +197,56 @@ export class EmbeddingFunctionRegistry {
   }
 }
 
-const _REGISTRY = new EmbeddingFunctionRegistry();
+function getBuiltInFunctions(registry: EmbeddingFunctionRegistry): Set<string> {
+  const registryWithBuiltIns = registry as EmbeddingFunctionRegistry & {
+    [key: symbol]: Set<string> | undefined;
+  };
+  let builtInFunctions = registryWithBuiltIns[builtInFunctionsKey];
+  if (builtInFunctions === undefined) {
+    builtInFunctions = new Set<string>();
+    registryWithBuiltIns[builtInFunctionsKey] = builtInFunctions;
+  }
+  return builtInFunctions;
+}
+
+// Server bundlers can load the side-effect embedding entry points and the public
+// embedding API from separate module graphs. Keep their registry shared.
+const registryKey = Symbol.for(
+  "@lancedb/lancedb::embedding-function-registry::v1",
+);
+const registryGlobal = globalThis as typeof globalThis & {
+  [key: symbol]: EmbeddingFunctionRegistry | undefined;
+};
+
+function getGlobalRegistry(): EmbeddingFunctionRegistry {
+  const existingRegistry = registryGlobal[registryKey];
+  if (existingRegistry !== undefined) {
+    return existingRegistry;
+  }
+  const registry = new EmbeddingFunctionRegistry();
+  registryGlobal[registryKey] = registry;
+  return registry;
+}
+
+const _REGISTRY = getGlobalRegistry();
 
 export function register(name?: string) {
   return _REGISTRY.register(name);
 }
 
+/** @ignore */
+export function registerBuiltIn<
+  T extends EmbeddingFunctionConstructor = EmbeddingFunctionConstructor,
+>(name: string, ctor: T): T {
+  const builtInFunctions = getBuiltInFunctions(_REGISTRY);
+  if (builtInFunctions.has(name)) {
+    return _REGISTRY.setBuiltIn(name, ctor);
+  }
+  _REGISTRY.register(name)(ctor);
+  builtInFunctions.add(name);
+  return ctor;
+}
+
 /**
  * Utility function to get the global instance of the registry
  * @returns `EmbeddingFunctionRegistry` The global instance of the registry

nodejs/lancedb/embedding/transformers.ts

@@ -3,7 +3,7 @@
 
 import { Float, Float32 } from "../arrow";
 import { EmbeddingFunction } from "./embedding_function";
-import { register } from "./registry";
+import { registerBuiltIn } from "./registry";
 
 export type XenovaTransformerOptions = {
   /** The wasm compatible model to use */
@@ -31,7 +31,6 @@ export type XenovaTransformerOptions = {
   };
 };
 
-@register("huggingface")
 export class TransformersEmbeddingFunction extends EmbeddingFunction<
   string,
   Partial<XenovaTransformerOptions>
@@ -158,6 +157,8 @@ export class TransformersEmbeddingFunction extends EmbeddingFunction<
   }
 }
 
+registerBuiltIn("huggingface", TransformersEmbeddingFunction);
+
 const tensorDiv = (
   src: import("@huggingface/transformers").Tensor,
   divBy: number,

nodejs/lancedb/index.ts

@@ -103,6 +103,7 @@ export {
 } from "./native.js";
 
 export {
+  AutoQuery,
   ExecutableQuery,
   Query,
   QueryBase,

nodejs/lancedb/query.ts

@@ -100,6 +100,29 @@ export interface FullTextSearchOptions {
   columns?: string | string[];
 }
 
+function nearestToNative(
+  inner: NativeQuery,
+  vector: Awaited<IntoVector>,
+): NativeVectorQuery {
+  const raw = Array.isArray(vector) ? null : extractVectorBuffer(vector);
+  if (raw) {
+    return inner.nearestToRaw(raw.data, raw.dtype);
+  }
+  return inner.nearestTo(Float32Array.from(vector as number[]));
+}
+
+function addQueryVectorToNative(
+  inner: NativeVectorQuery,
+  vector: Awaited<IntoVector>,
+) {
+  const raw = Array.isArray(vector) ? null : extractVectorBuffer(vector);
+  if (raw) {
+    inner.addQueryVectorRaw(raw.data, raw.dtype);
+  } else {
+    inner.addQueryVector(Float32Array.from(vector as number[]));
+  }
+}
+
 /** Common methods supported by all query types
  *
  * @see {@link Query}
@@ -111,13 +134,15 @@ export class QueryBase<
   NativeQueryType extends NativeQuery | NativeVectorQuery | NativeTakeQuery,
 > implements AsyncIterable<RecordBatch>
 {
+  protected inner!: NativeQueryType | Promise<NativeQueryType>;
+
   /**
    * @hidden
    */
-  protected constructor(
-    protected inner: NativeQueryType | Promise<NativeQueryType>,
-  ) {
-    // intentionally empty
+  protected constructor(inner?: NativeQueryType | Promise<NativeQueryType>) {
+    if (inner !== undefined) {
+      this.inner = inner;
+    }
   }
 
   // call a function on the inner (either a promise or the actual object)
@@ -135,6 +160,15 @@ export class QueryBase<
     }
   }
 
+  /**
+   * Return the native query used by the next terminal operation.
+   *
+   * @hidden
+   */
+  protected async getInner(): Promise<NativeQueryType> {
+    return this.inner;
+  }
+
   /**
    * Return only the specified columns.
    *
@@ -207,16 +241,11 @@ export class QueryBase<
   /**
    * @hidden
    */
-  protected nativeExecute(
+  protected async nativeExecute(
     options?: Partial<QueryExecutionOptions>,
   ): Promise<NativeBatchIterator> {
-    if (this.inner instanceof Promise) {
-      return this.inner.then((inner) =>
-        inner.execute(options?.maxBatchLength, options?.timeoutMs),
-      );
-    } else {
-      return this.inner.execute(options?.maxBatchLength, options?.timeoutMs);
-    }
+    const inner = await this.getInner();
+    return inner.execute(options?.maxBatchLength, options?.timeoutMs);
   }
 
   /**
@@ -245,12 +274,7 @@ export class QueryBase<
   /** Collect the results as an Arrow @see {@link ArrowTable}. */
   async toArrow(options?: Partial<QueryExecutionOptions>): Promise<ArrowTable> {
     const batches = [];
-    let inner;
-    if (this.inner instanceof Promise) {
-      inner = await this.inner;
-    } else {
-      inner = this.inner;
-    }
+    const inner = await this.getInner();
     for await (const batch of new RecordBatchIterable(inner, options)) {
       batches.push(batch);
     }
@@ -279,11 +303,8 @@ export class QueryBase<
    * @returns A Promise that resolves to a string containing the query execution plan explanation.
    */
   async explainPlan(verbose = false): Promise<string> {
-    if (this.inner instanceof Promise) {
-      return this.inner.then((inner) => inner.explainPlan(verbose));
-    } else {
-      return this.inner.explainPlan(verbose);
-    }
+    const inner = await this.getInner();
+    return inner.explainPlan(verbose);
   }
 
   /**
@@ -321,13 +342,8 @@ export class QueryBase<
     distributedMetrics?: AnalyzePlanDistributedMetrics,
   ): Promise<string> {
     const distributedMetricsMode = distributedMetrics ?? "aggregate";
-    if (this.inner instanceof Promise) {
-      return this.inner.then((inner) =>
-        inner.analyzePlan(distributedMetricsMode),
-      );
-    } else {
-      return this.inner.analyzePlan(distributedMetricsMode);
-    }
+    const inner = await this.getInner();
+    return inner.analyzePlan(distributedMetricsMode);
   }
 
   /**
@@ -339,12 +355,8 @@ export class QueryBase<
    * @returns An Arrow Schema describing the output columns.
    */
   async outputSchema(): Promise<import("./arrow").Schema> {
-    let schemaBuffer: Buffer;
-    if (this.inner instanceof Promise) {
-      schemaBuffer = await this.inner.then((inner) => inner.outputSchema());
-    } else {
-      schemaBuffer = await this.inner.outputSchema();
-    }
+    const inner = await this.getInner();
+    const schemaBuffer = await inner.outputSchema();
     const schema = tableFromIPC(schemaBuffer).schema;
     return schema;
   }
@@ -356,7 +368,7 @@ export class StandardQueryBase<
   extends QueryBase<NativeQueryType>
   implements ExecutableQuery
 {
-  constructor(inner: NativeQueryType | Promise<NativeQueryType>) {
+  constructor(inner?: NativeQueryType | Promise<NativeQueryType>) {
     super(inner);
   }
 
@@ -510,6 +522,13 @@ export class VectorQuery extends StandardQueryBase<NativeVectorQuery> {
     super(inner);
   }
 
+  /**
+   * @hidden
+   */
+  protected doVectorCall(fn: (inner: NativeVectorQuery) => void) {
+    super.doCall(fn);
+  }
+
   /**
    * Set the number of partitions to search (probe)
    *
@@ -537,7 +556,7 @@ export class VectorQuery extends StandardQueryBase<NativeVectorQuery> {
    * the minimum and maximum to the same value.
    */
   nprobes(nprobes: number): VectorQuery {
-    super.doCall((inner) => inner.nprobes(nprobes));
+    this.doVectorCall((inner) => inner.nprobes(nprobes));
 
     return this;
   }
@@ -551,7 +570,7 @@ export class VectorQuery extends StandardQueryBase<NativeVectorQuery> {
    * but will also increase latency.
    */
   minimumNprobes(minimumNprobes: number): VectorQuery {
-    super.doCall((inner) => inner.minimumNprobes(minimumNprobes));
+    this.doVectorCall((inner) => inner.minimumNprobes(minimumNprobes));
     return this;
   }
 
@@ -565,7 +584,7 @@ export class VectorQuery extends StandardQueryBase<NativeVectorQuery> {
    * potential false negatives.
    */
   maximumNprobes(maximumNprobes: number): VectorQuery {
-    super.doCall((inner) => inner.maximumNprobes(maximumNprobes));
+    this.doVectorCall((inner) => inner.maximumNprobes(maximumNprobes));
     return this;
   }
 
@@ -578,7 +597,7 @@ export class VectorQuery extends StandardQueryBase<NativeVectorQuery> {
    * `undefined` means no lower or upper bound.
    */
   distanceRange(lowerBound?: number, upperBound?: number): VectorQuery {
-    super.doCall((inner) => inner.distanceRange(lowerBound, upperBound));
+    this.doVectorCall((inner) => inner.distanceRange(lowerBound, upperBound));
     return this;
   }
 
@@ -592,7 +611,7 @@ export class VectorQuery extends StandardQueryBase<NativeVectorQuery> {
    * also increase the latency of your query. The default value is 1.5*limit.
    */
   ef(ef: number): VectorQuery {
-    super.doCall((inner) => inner.ef(ef));
+    this.doVectorCall((inner) => inner.ef(ef));
     return this;
   }
 
@@ -606,7 +625,7 @@ export class VectorQuery extends StandardQueryBase<NativeVectorQuery> {
    * whose data type is a fixed-size-list of floats.
    */
   column(column: string): VectorQuery {
-    super.doCall((inner) => inner.column(column));
+    this.doVectorCall((inner) => inner.column(column));
     return this;
   }
 
@@ -627,7 +646,7 @@ export class VectorQuery extends StandardQueryBase<NativeVectorQuery> {
   distanceType(
     distanceType: Required<IvfPqOptions>["distanceType"],
   ): VectorQuery {
-    super.doCall((inner) => inner.distanceType(distanceType));
+    this.doVectorCall((inner) => inner.distanceType(distanceType));
     return this;
   }
 
@@ -661,7 +680,7 @@ export class VectorQuery extends StandardQueryBase<NativeVectorQuery> {
    * distance between the query vector and the actual uncompressed vector.
    */
   refineFactor(refineFactor: number): VectorQuery {
-    super.doCall((inner) => inner.refineFactor(refineFactor));
+    this.doVectorCall((inner) => inner.refineFactor(refineFactor));
     return this;
   }
 
@@ -686,7 +705,7 @@ export class VectorQuery extends StandardQueryBase<NativeVectorQuery> {
    * factor can often help restore some of the results lost by post filtering.
    */
   postfilter(): VectorQuery {
-    super.doCall((inner) => inner.postfilter());
+    this.doVectorCall((inner) => inner.postfilter());
     return this;
   }
 
@@ -700,7 +719,7 @@ export class VectorQuery extends StandardQueryBase<NativeVectorQuery> {
    * calculate your recall to select an appropriate value for nprobes.
    */
   bypassVectorIndex(): VectorQuery {
-    super.doCall((inner) => inner.bypassVectorIndex());
+    this.doVectorCall((inner) => inner.bypassVectorIndex());
     return this;
   }
 
@@ -708,43 +727,39 @@ export class VectorQuery extends StandardQueryBase<NativeVectorQuery> {
    * Add a query vector to the search
    *
    * This method can be called multiple times to add multiple query vectors
-   * to the search. If multiple query vectors are added, then they will be searched
-   * in parallel, and the results will be concatenated. A column called `query_index`
-   * will be added to indicate the index of the query vector that produced the result.
-   *
-   * Performance wise, this is equivalent to running multiple queries concurrently.
+   * to the search. A column called `query_index` will be added to indicate the index
+   * of the query vector that produced the result. Flat searches share one table scan
+   * across the query vectors, avoiding the scan and memory amplification of running
+   * multiple queries concurrently. Indexed searches may still perform per-vector
+   * index work.
    */
   addQueryVector(vector: IntoVector): VectorQuery {
     if (vector instanceof Promise) {
+      // Observe the promise as soon as it is accepted. The existing native
+      // query may still be pending, and delaying observation until it resolves
+      // can otherwise surface a fast rejection as unhandled.
+      const settledVector = vector.then(
+        (value) => ({ status: "fulfilled" as const, value }),
+        (reason) => ({ status: "rejected" as const, reason }),
+      );
       const res = (async () => {
-        try {
-          const v = await vector;
-          // biome-ignore lint/suspicious/noExplicitAny: we need to get the `inner`, but js has no package scoping
-          const value: any = this.addQueryVector(v);
-          const inner = value.inner as
-            | NativeVectorQuery
-            | Promise<NativeVectorQuery>;
-          return inner;
-        } catch (e) {
-          return Promise.reject(e);
+        const inner = await this.getInner();
+        const outcome = await settledVector;
+        if (outcome.status === "rejected") {
+          throw outcome.reason;
         }
+        addQueryVectorToNative(inner, outcome.value);
+        return inner;
       })();
       return new VectorQuery(res);
     } else {
-      super.doCall((inner) => {
-        const raw = Array.isArray(vector) ? null : extractVectorBuffer(vector);
-        if (raw) {
-          inner.addQueryVectorRaw(raw.data, raw.dtype);
-        } else {
-          inner.addQueryVector(Float32Array.from(vector as number[]));
-        }
-      });
+      this.doVectorCall((inner) => addQueryVectorToNative(inner, vector));
       return this;
     }
   }
 
   rerank(reranker: Reranker): VectorQuery {
-    super.doCall((inner) =>
+    this.doVectorCall((inner) =>
       inner.rerank(async (args) => {
         const vecResults = await fromBufferToRecordBatch(args.vecResults);
         const ftsResults = await fromBufferToRecordBatch(args.ftsResults);
@@ -763,6 +778,71 @@ export class VectorQuery extends StandardQueryBase<NativeVectorQuery> {
   }
 }
 
+/**
+ * Create a string query whose vector/FTS routing is resolved against the active
+ * table schema when the query executes.
+ *
+ * @hidden
+ */
+export function createAutoQuery(
+  table: NativeTable,
+  query: string,
+  columns: string[] | null,
+  getVector: (metadata: string) => Promise<Awaited<IntoVector>>,
+): AutoQuery {
+  type RouteSnapshot = {
+    table: NativeTable;
+    embeddingMetadata: string | undefined;
+  };
+  type CachedPreparation = {
+    metadata: string;
+    vector: Promise<Awaited<IntoVector>>;
+  };
+
+  let cachedPreparation: CachedPreparation | undefined;
+
+  const snapshotRoute = async (): Promise<RouteSnapshot> => {
+    const snapshot = await table.querySnapshot();
+    const schema = tableFromIPC(await snapshot.schema()).schema;
+    return {
+      table: snapshot,
+      embeddingMetadata: schema.metadata.get("embedding_functions"),
+    };
+  };
+
+  const createInner = async (): Promise<NativeQuery | NativeVectorQuery> => {
+    const route = await snapshotRoute();
+    if (route.embeddingMetadata === undefined) {
+      const inner = route.table.query();
+      inner.fullTextSearch({ query, columns });
+      return inner;
+    }
+
+    const metadata = route.embeddingMetadata;
+    if (cachedPreparation?.metadata !== metadata) {
+      cachedPreparation = {
+        metadata,
+        vector: Promise.resolve().then(() => getVector(metadata)),
+      };
+    }
+
+    const preparation = cachedPreparation;
+    let vector: Awaited<IntoVector>;
+    try {
+      vector = await preparation.vector;
+    } catch (error) {
+      if (cachedPreparation === preparation) {
+        cachedPreparation = undefined;
+      }
+      throw error;
+    }
+
+    return nearestToNative(route.table.query(), vector);
+  };
+
+  return new AutoQuery(createInner);
+}
+
 /**
  * A query that returns a subset of the rows in the table.
  *
@@ -788,6 +868,51 @@ export class TakeQuery extends QueryBase<NativeTakeQuery> {
   }
 }
 
+/**
+ * A builder for automatic string searches.
+ *
+ * Automatic search determines whether to use full-text or vector search from
+ * the table revision selected for each execution. This builder exposes the
+ * common operations supported by both query families.
+ *
+ * @hideconstructor
+ */
+export class AutoQuery extends StandardQueryBase<
+  NativeQuery | NativeVectorQuery
+> {
+  private readonly calls: Array<
+    (inner: NativeQuery | NativeVectorQuery) => void
+  > = [];
+
+  /** @hidden */
+  constructor(
+    private readonly createInner: () => Promise<
+      NativeQuery | NativeVectorQuery
+    >,
+  ) {
+    super();
+  }
+
+  /** @hidden */
+  protected override doCall(
+    fn: (inner: NativeQuery | NativeVectorQuery) => void,
+  ) {
+    this.calls.push(fn);
+  }
+
+  /** @hidden */
+  protected override async getInner(): Promise<
+    NativeQuery | NativeVectorQuery
+  > {
+    const calls = [...this.calls];
+    const inner = await this.createInner();
+    for (const call of calls) {
+      call(inner);
+    }
+    return inner;
+  }
+}
+
 /** A builder for LanceDB queries.
  *
  * @see {@link Table#query}, {@link Table#search}
@@ -840,45 +965,19 @@ export class Query extends StandardQueryBase<NativeQuery> {
    * a default `limit` of 10 will be used.  @see {@link Query#limit}
    */
   nearestTo(vector: IntoVector): VectorQuery {
-    const callNearestTo = (
-      inner: NativeQuery,
-      resolved: Float32Array | Float64Array | Uint8Array | number[],
-    ): NativeVectorQuery => {
-      const raw = Array.isArray(resolved)
-        ? null
-        : extractVectorBuffer(resolved);
-      if (raw) {
-        return inner.nearestToRaw(raw.data, raw.dtype);
-      }
-      return inner.nearestTo(Float32Array.from(resolved as number[]));
-    };
-
-    if (this.inner instanceof Promise) {
-      const nativeQuery = this.inner.then(async (inner) => {
-        const resolved = vector instanceof Promise ? await vector : vector;
-        return callNearestTo(inner, resolved);
-      });
+    const inner = this.inner;
+    if (inner instanceof Promise) {
+      const nativeQuery = inner.then(async (resolvedInner) =>
+        nearestToNative(resolvedInner, await vector),
+      );
       return new VectorQuery(nativeQuery);
     }
     if (vector instanceof Promise) {
-      const res = (async () => {
-        try {
-          const v = await vector;
-          // biome-ignore lint/suspicious/noExplicitAny: we need to get the `inner`, but js has no package scoping
-          const value: any = this.nearestTo(v);
-          const inner = value.inner as
-            | NativeVectorQuery
-            | Promise<NativeVectorQuery>;
-          return inner;
-        } catch (e) {
-          return Promise.reject(e);
-        }
-      })();
-      return new VectorQuery(res);
-    } else {
-      const vectorQuery = callNearestTo(this.inner, vector);
-      return new VectorQuery(vectorQuery);
+      return new VectorQuery(
+        vector.then((resolvedVector) => nearestToNative(inner, resolvedVector)),
+      );
     }
+    return new VectorQuery(nearestToNative(inner, vector));
   }
 
   nearestToText(query: string | FullTextQuery, columns?: string[]): Query {

nodejs/lancedb/sanitize.ts

@@ -94,17 +94,24 @@ export function sanitizeMetadata(
   if (metadataLike === undefined || metadataLike === null) {
     return undefined;
   }
-  if (!(metadataLike instanceof Map)) {
+
+  let entries: IterableIterator<[unknown, unknown]>;
+  try {
+    entries = Map.prototype.entries.call(metadataLike);
+  } catch {
     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") {
+
+  const metadata = new Map<string, string>();
+  for (const [key, value] of entries) {
+    if (typeof key !== "string" || typeof value !== "string") {
       throw Error(
         "Expected metadata, if present, to be a Map<string, string> but it had non-string keys or values",
       );
     }
+    metadata.set(key, value);
   }
-  return metadataLike as Map<string, string>;
+  return metadata;
 }
 
 export function sanitizeInt(typeLike: object) {

nodejs/lancedb/schema.ts

@@ -406,10 +406,11 @@ function matchingFields(fields: Field[], tree: FieldTree): Field[] {
           field.name,
           new Struct(matchingFields(struct.children, value)),
           field.nullable,
+          field.metadata,
         ),
       );
     } else {
-      matches.push(new Field(field.name, value as DataType, field.nullable));
+      matches.push(field);
     }
   }
   return matches;

nodejs/lancedb/table.ts

@@ -43,10 +43,12 @@ import {
   Table as _NativeTable,
 } from "./native";
 import {
+  AutoQuery,
   FullTextQuery,
   Query,
   TakeQuery,
   VectorQuery,
+  createAutoQuery,
   instanceOfFullTextQuery,
 } from "./query";
 import { sanitizeType } from "./sanitize";
@@ -523,7 +525,7 @@ export abstract class Table {
     query: string | IntoVector | MultiVector | FullTextQuery,
     queryType?: string,
     ftsColumns?: string | string[],
-  ): VectorQuery | Query;
+  ): VectorQuery | Query | AutoQuery;
   /**
    * Search the table with a given query vector.
    *
@@ -628,6 +630,18 @@ export abstract class Table {
 
   /**
    * Update per-field (column) metadata.
+   *
+   * The following keys are treated specially, by convention, and should be
+   * used when appropriate:
+   *
+   * - `lancedb:description`: for a human-readable description of a field.
+   * - `lancedb:tag:<name>`: for a user-defined key-value tag, where the suffix
+   *   names the tag category; e.g. `lancedb:tag:model: "clip"`.
+   * - `lancedb:logical-column`: for a column grouping; e.g. `feature_v1` and
+   *   `feature_v2` might be in the same logical column.
+   * - `lancedb:status`: for status options (`production`, `candidate`,
+   *   `deprecated`, `archived`) to designate the current life cycle state of
+   *   this column.
    * @param {FieldMetadataUpdate[]} updates One or more per-field updates. Each
    * update's metadata is merged into the field's existing metadata by default;
    * a value of `null` deletes that key, and `replace: true` swaps the whole map.
@@ -975,10 +989,11 @@ export class LocalTable extends Table {
     return this.inner.display();
   }
 
-  private async getEmbeddingFunctions(): Promise<
-    Map<string, EmbeddingFunctionConfig>
-  > {
-    const schema = await this.schema();
+  private async getEmbeddingFunctions(
+    inner: _NativeTable = this.inner,
+  ): Promise<Map<string, EmbeddingFunctionConfig>> {
+    const schemaBuf = await inner.schema();
+    const schema = tableFromIPC(schemaBuf).schema;
     const registry = getRegistry();
     return registry.parseFunctions(schema.metadata);
   }
@@ -1160,7 +1175,7 @@ export class LocalTable extends Table {
     query: string | IntoVector | MultiVector | FullTextQuery,
     queryType: string = "auto",
     ftsColumns?: string | string[],
-  ): VectorQuery | Query {
+  ): VectorQuery | Query | AutoQuery {
     if (typeof query !== "string" && !instanceOfFullTextQuery(query)) {
       if (queryType === "fts") {
         throw new Error("Cannot perform full text search on a vector query");
@@ -1175,14 +1190,28 @@ export class LocalTable extends Table {
       });
     }
 
-    // The query type is auto or vector
-    // fall back to full text search if no embedding functions are defined and the query is a string
-    if (
-      queryType === "auto" &&
-      (getRegistry().length() === 0 || instanceOfFullTextQuery(query))
-    ) {
-      return this.query().fullTextSearch(query, {
-        columns: ftsColumns,
+    if (queryType === "auto") {
+      if (instanceOfFullTextQuery(query)) {
+        return this.query().fullTextSearch(query, {
+          columns: ftsColumns,
+        });
+      }
+
+      const columns =
+        typeof ftsColumns === "string" ? [ftsColumns] : (ftsColumns ?? null);
+      return createAutoQuery(this.inner, query, columns, async (metadata) => {
+        const functions = await getRegistry().parseFunctions(
+          new Map([["embedding_functions", metadata]]),
+        );
+        // TODO: Support multiple embedding functions
+        const embeddingFunc: EmbeddingFunctionConfig | undefined = functions
+          .values()
+          .next().value;
+        // The route only calls this callback when embedding metadata exists.
+        // parseFunctions either yields a provider or reports malformed metadata.
+        if (!embeddingFunc)
+          throw new Error("Invalid embedding function metadata");
+        return await embeddingFunc.function.computeQueryEmbeddings(query);
       });
     }
 
@@ -1538,7 +1567,8 @@ export interface FieldMetadataUpdate {
   path: string;
   /**
    * Metadata key/value pairs. Merged into the field's existing metadata by
-   * default; a value of `null` deletes that key.
+   * default; a value of `null` deletes that key. See
+   * {@link Table.updateFieldMetadata} for the conventional `lancedb:*` keys.
    */
   metadata: Record<string, string | null>;
   /** If true, replace the field's entire metadata map instead of merging. */

nodejs/npm/darwin-arm64/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-darwin-arm64",
-	"version": "0.38.0-beta.10",
+	"version": "0.38.0-beta.12",
 	"os": ["darwin"],
 	"cpu": ["arm64"],
 	"main": "lancedb.darwin-arm64.node",

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

@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-linux-arm64-gnu",
-	"version": "0.38.0-beta.10",
+	"version": "0.38.0-beta.12",
 	"os": ["linux"],
 	"cpu": ["arm64"],
 	"main": "lancedb.linux-arm64-gnu.node",

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

@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-linux-arm64-musl",
-	"version": "0.38.0-beta.10",
+	"version": "0.38.0-beta.12",
 	"os": ["linux"],
 	"cpu": ["arm64"],
 	"main": "lancedb.linux-arm64-musl.node",

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

@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-linux-x64-gnu",
-	"version": "0.38.0-beta.10",
+	"version": "0.38.0-beta.12",
 	"os": ["linux"],
 	"cpu": ["x64"],
 	"main": "lancedb.linux-x64-gnu.node",

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

@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-linux-x64-musl",
-	"version": "0.38.0-beta.10",
+	"version": "0.38.0-beta.12",
 	"os": ["linux"],
 	"cpu": ["x64"],
 	"main": "lancedb.linux-x64-musl.node",

nodejs/npm/win32-arm64-msvc/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@lancedb/lancedb-win32-arm64-msvc",
-  "version": "0.38.0-beta.10",
+  "version": "0.38.0-beta.12",
   "os": [
     "win32"
   ],

nodejs/npm/win32-x64-msvc/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-win32-x64-msvc",
-	"version": "0.38.0-beta.10",
+	"version": "0.38.0-beta.12",
 	"os": ["win32"],
 	"cpu": ["x64"],
 	"main": "lancedb.win32-x64-msvc.node",

nodejs/package.json

@@ -11,7 +11,7 @@
     "ann"
   ],
   "private": false,
-  "version": "0.38.0-beta.10",
+  "version": "0.38.0-beta.12",
   "main": "dist/index.js",
   "exports": {
     ".": "./dist/index.js",
@@ -67,7 +67,7 @@
     "timeout": "3m"
   },
   "engines": {
-    "node": ">= 18"
+    "node": ">= 22"
   },
   "packageManager": "pnpm@11.1.1",
   "cpu": ["x64", "arm64"],
@@ -101,7 +101,7 @@
     "openai": "4.29.2"
   },
   "peerDependencies": {
-    "@types/node": ">=18",
+    "@types/node": ">=22",
     "apache-arrow": ">=15.0.0 <=18.1.0"
   },
   "peerDependenciesMeta": {

nodejs/src/table.rs

@@ -278,6 +278,13 @@ impl Table {
         Ok(Query::new(self.inner_ref()?.query()))
     }
 
+    /// Return a read-only table handle pinned to the current query revision.
+    #[napi(catch_unwind)]
+    pub async fn query_snapshot(&self) -> napi::Result<Self> {
+        let snapshot = self.inner_ref()?.query_snapshot().await.default_error()?;
+        Ok(Self::new(snapshot))
+    }
+
     #[napi(catch_unwind)]
     pub fn take_offsets(&self, offsets: Vec<i64>) -> napi::Result<TakeQuery> {
         Ok(TakeQuery::new(
@@ -554,6 +561,12 @@ impl Table {
             .default_error()
     }
 
+    #[napi(catch_unwind)]
+    pub async fn checkout_current(&self) -> napi::Result<Self> {
+        let table = self.inner_ref()?.checkout_current().await.default_error()?;
+        Ok(Self::new(table))
+    }
+
     #[napi(catch_unwind)]
     pub async fn checkout(&self, version: i64) -> napi::Result<()> {
         self.inner_ref()?

python/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "lancedb-python"
-version = "0.38.0-beta.10"
+version = "0.38.0-beta.12"
 publish = false
 edition.workspace = true
 description = "Python bindings for LanceDB"

python/pyproject.toml

@@ -101,9 +101,12 @@ azure = ["adlfs>=2024.2.0"]
 [tool.maturin]
 python-source = "python"
 module-name = "lancedb._lancedb"
+# uv installs the project as an editable package before `uv run`, so keep that
+# bootstrap build consistent with `maturin develop`.
+editable-profile = "dev"
 
 [build-system]
-requires = ["maturin>=1.9.4"]
+requires = ["maturin>=1.10"]
 build-backend = "maturin"
 
 [tool.ruff.lint]

python/python/lancedb/__init__.py

@@ -6,7 +6,7 @@ import importlib.metadata
 import os
 from concurrent.futures import ThreadPoolExecutor
 from datetime import timedelta
-from typing import Dict, Optional, Union, Any, List, Iterable
+from typing import Dict, Optional, Union, Any, List, Iterable, TYPE_CHECKING
 
 __version__ = importlib.metadata.version("lancedb")
 
@@ -20,7 +20,7 @@ from .db import AsyncConnection, DBConnection, LanceDBConnection
 from .remote import ClientConfig
 from .remote.db import RemoteDBConnection
 from .expr import Expr, col, lit, func
-from .schema import blob, vector, BlobType
+from .schema import blob, vector
 from .job import AsyncJob, Job
 from .functions import (
     FunctionArtifactRequest as FunctionArtifactRequest,
@@ -49,6 +49,19 @@ from .namespace import (
 )
 
 
+if TYPE_CHECKING:
+    from lance.blob import BlobType as BlobType
+
+
+def __getattr__(name: str):
+    if name == "BlobType":
+        from .schema import BlobType
+
+        globals()["BlobType"] = BlobType
+        return BlobType
+    raise AttributeError(f"module {__name__!r} has no attribute {name!r}")
+
+
 def _check_s3_bucket_with_dots(
     uri: str, storage_options: Optional[Dict[str, str]]
 ) -> None:
@@ -179,6 +192,18 @@ def connect(
     ...     },
     ... )
 
+    For Azure Blob Storage, credentials can be passed directly without setting
+    environment variables:
+
+    >>> azure_storage_options = {
+    ...     "account_name": "some-account",
+    ...     "account_key": "some-key",
+    ... }
+    >>> db = lancedb.connect(  # doctest: +SKIP
+    ...     "az://my-container/my-database",
+    ...     storage_options=azure_storage_options,
+    ... )
+
     For tests and temporary data, use an in-memory database:
 
     >>> db = lancedb.connect("memory://")
@@ -465,6 +490,10 @@ async def connect_async(
     --------
 
     >>> import lancedb
+    >>> azure_storage_options = {
+    ...     "account_name": "some-account",
+    ...     "account_key": "some-key",
+    ... }
     >>> async def doctest_example():
     ...     # For a local directory, provide a path to the database
     ...     db = await lancedb.connect_async("~/.lancedb")
@@ -472,6 +501,11 @@ async def connect_async(
     ...     db = await lancedb.connect_async("s3://my-bucket/lancedb",
     ...                                      storage_options={
     ...                                          "aws_access_key_id": "***"})
+    ...     # Azure credentials can also be passed directly
+    ...     db = await lancedb.connect_async(
+    ...         "az://my-container/my-database",
+    ...         storage_options=azure_storage_options,
+    ...     )
     ...     # For tests and temporary data, use an in-memory database
     ...     db = await lancedb.connect_async("memory://")
     ...     # Connect to LanceDB cloud

python/python/lancedb/_blob.py

@@ -12,7 +12,7 @@ from typing import TYPE_CHECKING, Optional, Union
 import pyarrow as pa
 
 from .expr import Expr
-from .schema import blob_v2_column_paths
+from .schema import row_addressable_blob_v2_paths
 from .types import BlobMode, QueryProjection, QueryProjectionSpec
 
 if TYPE_CHECKING:
@@ -119,7 +119,7 @@ def blob_v2_projection_sources(
     schema: pa.Schema,
     projection: QueryProjection,
 ) -> dict[str, str]:
-    blob_columns = blob_v2_column_paths(schema)
+    blob_columns = row_addressable_blob_v2_paths(schema)
     if not blob_columns:
         return {}
     columns = set(blob_columns)
@@ -140,7 +140,9 @@ def v2_projection_needs_row_id(
 ) -> bool:
     if with_row_id:
         return False
-    return projection_includes_blob_column(projection, blob_v2_column_paths(schema))
+    return projection_includes_blob_column(
+        projection, row_addressable_blob_v2_paths(schema)
+    )
 
 
 def blob_auto_row_id_for_scan(
@@ -270,7 +272,8 @@ def _iter_projection_pairs(
             if isinstance(expr, str):
                 yield name, expr
             elif isinstance(expr, Expr):
-                yield name, expr.to_sql()
+                source = expr._column_name()
+                yield name, source if source is not None else expr.to_sql()
         return
     for column in projection:
         if isinstance(column, str):
@@ -280,7 +283,8 @@ def _iter_projection_pairs(
             if isinstance(expr, str):
                 yield name, expr
             elif isinstance(expr, Expr):
-                yield name, expr.to_sql()
+                source = expr._column_name()
+                yield name, source if source is not None else expr.to_sql()
 
 
 def _set_blob_column(tbl: pa.Table, output_name: str, blobs: pa.Array) -> pa.Table:

python/python/lancedb/_lancedb.pyi

@@ -87,6 +87,7 @@ class PyExpr:
     def contains(self, substr: "PyExpr") -> "PyExpr": ...
     def isin(self, values: List["PyExpr"]) -> "PyExpr": ...
     def cast(self, data_type: pa.DataType) -> "PyExpr": ...
+    def column_name(self) -> Optional[str]: ...
     def to_sql(self) -> str: ...
 
 def expr_col(name: str) -> PyExpr: ...
@@ -283,6 +284,7 @@ class Table:
         mode: Literal["append", "overwrite"],
         progress: Optional[Any] = None,
         write_parallelism: Optional[int] = None,
+        allow_external_blob_outside_bases: bool = False,
     ) -> AddResult: ...
     async def update(
         self, updates: Dict[str, str], where: Optional[str]
@@ -607,6 +609,7 @@ class PyQueryRequest:
     filter: Optional[Union[str, bytes]]
     full_text_search: Optional[FullTextQuery]
     select: Optional[Union[str, List[str]]]
+    select_source_columns: Optional[Dict[str, str]]
     fast_search: Optional[bool]
     with_row_id: Optional[bool]
     use_lsm: Optional[bool]

python/python/lancedb/expr.py

@@ -249,6 +249,10 @@ class Expr:
 
     # ── utilities ────────────────────────────────────────────────────────────
 
+    def _column_name(self) -> str | None:
+        """Return the source name when this is a bare column expression."""
+        return self._inner.column_name()
+
     def to_sql(self) -> str:
         """Render the expression as a SQL string (useful for debugging)."""
         return self._inner.to_sql()
@@ -312,7 +316,7 @@ def func(name: str, *args: ExprLike) -> Expr:
     --------
     >>> from lancedb.expr import col, func
     >>> func("lower", col("name"))
-    Expr(lower(name))
+    Expr(lower(`name`))
     """
     inner_args = [_coerce(a)._inner for a in args]
     return Expr(expr_func(name, inner_args))

python/python/lancedb/functions.py

@@ -222,6 +222,7 @@ class PythonEnvironmentSpec(_RemoteValue):
 
     kind: str
     packages: tuple[str, ...] = ()
+    channels: tuple[str, ...] = ()
     path: Optional[str] = None
     modules: tuple[str, ...] = ()
     image: Optional[str] = None
@@ -909,13 +910,25 @@ class UdfDefinition:
         pip: tuple[str, ...],
         env: Mapping[str, str],
         python_version: Optional[str],
+        conda: tuple[str, ...] = (),
+        conda_channels: tuple[str, ...] = (),
     ):
         function_name = name or function.__name__
         if not _FUNCTION_NAME.fullmatch(function_name):
             raise ValueError(f"invalid Function name: {function_name!r}")
-        packages = tuple(sorted(set(pip)))
+        if pip and conda:
+            raise ValueError("a Function environment is pip or conda, not both")
+        if conda_channels and not conda:
+            raise ValueError("conda_channels requires conda packages")
+        packages = tuple(sorted(set(conda if conda else pip)))
         if any(not package or package != package.strip() for package in packages):
-            raise ValueError("pip requirements must be non-empty and trimmed")
+            raise ValueError("package requirements must be non-empty and trimmed")
+        if conda:
+            environment_spec = PythonEnvironmentSpec(
+                kind="conda", packages=packages, channels=tuple(conda_channels)
+            )
+        else:
+            environment_spec = PythonEnvironmentSpec(kind="pip", packages=packages)
         environment = dict(env)
         if any(
             not isinstance(key, str) or not isinstance(value, str)
@@ -929,7 +942,7 @@ class UdfDefinition:
             kind="python",
             python_version=python_version
             or f"{sys.version_info.major}.{sys.version_info.minor}",
-            environment=PythonEnvironmentSpec(kind="pip", packages=packages),
+            environment=environment_spec,
             env=environment,
         )
         self._function = function
@@ -976,6 +989,8 @@ def udf(
     pip: tuple[str, ...] | list[str] = (),
     env: Optional[Mapping[str, str]] = None,
     python_version: Optional[str] = None,
+    conda: tuple[str, ...] | list[str] = (),
+    conda_channels: tuple[str, ...] | list[str] = (),
 ) -> Callable[[Callable[..., Any]], UdfDefinition]: ...
 
 
@@ -988,6 +1003,8 @@ def udf(
     pip: tuple[str, ...] | list[str] = (),
     env: Optional[Mapping[str, str]] = None,
     python_version: Optional[str] = None,
+    conda: tuple[str, ...] | list[str] = (),
+    conda_channels: tuple[str, ...] | list[str] = (),
 ):
     """Prepare a scalar Python callable for remote Function registration.
 
@@ -1010,6 +1027,10 @@ def udf(
         provided together with ``input_schema``.
     pip : sequence of str, optional
         Pip requirements for the remote environment.
+    conda : sequence of str, optional
+        Conda packages for the remote environment, instead of ``pip``.
+    conda_channels : sequence of str, optional
+        Conda channels in priority order; requires ``conda``.
     env : mapping of str to str, optional
         Environment variables included in the Function definition.
     python_version : str, optional
@@ -1049,6 +1070,8 @@ def udf(
             pip=tuple(pip),
             env={} if env is None else env,
             python_version=python_version,
+            conda=tuple(conda),
+            conda_channels=tuple(conda_channels),
         )
 
     if function is None:

python/python/lancedb/index.py

@@ -7,6 +7,7 @@ from typing import List, Literal, Optional
 from ._lancedb import (
     IndexConfig,
 )
+from .query import DocumentGranularity
 from .types import BaseTokenizerType
 
 lang_mapping = {
@@ -121,6 +122,11 @@ class FTS:
 
     >>> config = FTS(block_size=256)
 
+    Create an index that treats each deepest-list element as one document:
+
+    >>> from lancedb.query import DocumentGranularity
+    >>> config = FTS(document_granularity=DocumentGranularity.LIST_ELEMENT)
+
     Attributes
     ----------
     with_position : bool, default False
@@ -172,6 +178,11 @@ class FTS:
         roughly half of the available CPU cores. The effective value is
         limited by the available compute capacity. This build-only setting is
         not persisted with the index and does not apply to remote tables.
+    document_granularity : DocumentGranularity, default ROW
+        ``ROW`` treats the selected text in one table row as one document.
+        ``LIST_ELEMENT`` treats each element of the deepest list on the indexed
+        field path as one document and returns its physical coordinates in
+        ``_doc_index`` for matching queries.
 
     Notes
     -----
@@ -196,6 +207,7 @@ class FTS:
     custom_stop_words: Optional[List[str]] = None
     memory_limit: Optional[int] = None
     num_workers: Optional[int] = None
+    document_granularity: DocumentGranularity = DocumentGranularity.ROW
 
 
 @dataclass

python/python/lancedb/query.py

@@ -167,6 +167,12 @@ def _projection_to_scanner_kwargs(columns: QueryProjection) -> Dict[str, Any]:
     return {"columns": projection}
 
 
+def _query_request_projection(req: "PyQueryRequest") -> QueryProjection:
+    if req.select_source_columns is not None:
+        return req.select_source_columns
+    return req.select
+
+
 def _scanner_kwargs_for_query(
     query: Query,
     blob_mode: BlobMode,
@@ -375,6 +381,13 @@ class FullTextOperator(str, Enum):
     OR = "OR"
 
 
+class DocumentGranularity(str, Enum):
+    """The unit treated as one full-text-search document."""
+
+    ROW = "row"
+    LIST_ELEMENT = "list_element"
+
+
 class Occur(str, Enum):
     SHOULD = "SHOULD"
     MUST = "MUST"
@@ -478,6 +491,10 @@ class MatchQuery(FullTextQuery):
     prefix_length : int, optional
         The number of beginning characters being unchanged for fuzzy matching.
         This is useful to achieve prefix matching.
+    document_granularity : DocumentGranularity, optional
+        Explicitly select row or deepest-list-element documents. If omitted,
+        the indexed granularity is inferred. When both granularities are indexed
+        for the field, this must be specified. With no index, row granularity is used.
     """
 
     query: str
@@ -487,6 +504,9 @@ class MatchQuery(FullTextQuery):
     max_expansions: int = pydantic.Field(50, kw_only=True)
     operator: FullTextOperator = pydantic.Field(FullTextOperator.OR, kw_only=True)
     prefix_length: int = pydantic.Field(0, kw_only=True)
+    document_granularity: Optional[DocumentGranularity] = pydantic.Field(
+        None, kw_only=True
+    )
 
     def query_type(self) -> FullTextQueryType:
         return FullTextQueryType.MATCH
@@ -503,11 +523,20 @@ class PhraseQuery(FullTextQuery):
         The query string to match against.
     column : str
         The name of the column to match against.
+    slop : int, default 0
+        The maximum number of intervening positions permitted in the phrase.
+    document_granularity : DocumentGranularity, optional
+        Explicitly select row or deepest-list-element documents. If omitted,
+        the indexed granularity is inferred. When both granularities are indexed
+        for the field, this must be specified. With no index, row granularity is used.
     """
 
     query: str
     column: str
     slop: int = pydantic.Field(0, kw_only=True)
+    document_granularity: Optional[DocumentGranularity] = pydantic.Field(
+        None, kw_only=True
+    )
 
     def query_type(self) -> FullTextQueryType:
         return FullTextQueryType.MATCH_PHRASE
@@ -2776,15 +2805,16 @@ class AsyncQueryBase(object):
 
         req = self._inner.to_query_request()
         schema = await self._table.schema()
+        projection = _query_request_projection(req)
         self._blob_auto_row_id = blob_auto_row_id_for_scan(
             schema,
-            req.select,
+            projection,
             with_row_id=self._with_row_id,
         )
         if not self._blob_auto_row_id:
             self._blob_paths = ()
             return
-        self._blob_paths = tuple(blob_v2_projection_sources(schema, req.select).keys())
+        self._blob_paths = tuple(blob_v2_projection_sources(schema, projection).keys())
         self._inner.with_row_id()
 
     def select(self, columns: Union[List[str], dict[str, str]]) -> Self:
@@ -3378,9 +3408,10 @@ class AsyncQuery(AsyncStandardQuery):
         pass in multiple vectors. When multiple vectors are passed in, if the vector
         column is with multivector type, then the vectors will be treated as a single
         query. Or the vectors will be treated as multiple queries, this can be useful
-        if you want to find the nearest vectors to multiple query vectors.
-        This is not expected to be faster than making multiple queries concurrently;
-        it is just a convenience method. If multiple vectors are passed in then
+        if you want to find the nearest vectors to multiple query vectors. Flat
+        searches share one table scan across the query vectors, avoiding the scan
+        and memory amplification of making multiple queries concurrently. If
+        multiple vectors are passed in then
         an additional column `query_index` will be added to the results. This column
         will contain the index of the query vector that the result is nearest to.
         """
@@ -3509,8 +3540,8 @@ class AsyncFTSQuery(AsyncStandardQuery):
 
         Typically, a single vector is passed in as the query. However, you can also
         pass in multiple vectors.  This can be useful if you want to find the nearest
-        vectors to multiple query vectors. This is not expected to be faster than
-        making multiple queries concurrently; it is just a convenience method.
+        vectors to multiple query vectors. Flat searches share one table scan across
+        the query vectors instead of issuing concurrent full scans.
         If multiple vectors are passed in then an additional column `query_index`
         will be added to the results.  This column will contain the index of the
         query vector that the result is nearest to.
@@ -3870,14 +3901,15 @@ class AsyncHybridQuery(AsyncStandardQuery, AsyncVectorQueryBase):
         blob_paths: tuple[str, ...] = ()
         if self._table is not None:
             schema = await self._table.schema()
+            projection = _query_request_projection(req)
             blob_auto_row_id = blob_auto_row_id_for_scan(
                 schema,
-                req.select,
+                projection,
                 with_row_id=self._with_row_id,
             )
             if blob_auto_row_id:
                 blob_paths = tuple(
-                    blob_v2_projection_sources(schema, req.select).keys()
+                    blob_v2_projection_sources(schema, projection).keys()
                 )
         self._blob_auto_row_id = blob_auto_row_id
         self._blob_paths = blob_paths

python/python/lancedb/remote/table.py

@@ -36,6 +36,7 @@ from lancedb._lancedb import (
     UpdateResult,
 )
 from lancedb.embeddings.base import EmbeddingFunctionConfig
+from lancedb.expr import Expr
 from lancedb.index import (
     FTS,
     BTree,
@@ -61,11 +62,20 @@ from lancedb.table import _normalize_progress
 
 from ..query import (
     AnalyzePlanDistributedMetrics,
+    DocumentGranularity,
     LanceQueryBuilder,
     LanceTakeQueryBuilder,
     LanceVectorQueryBuilder,
 )
-from ..table import AsyncTable, BlobMode, Branches, IndexStatistics, Query, Table, Tags
+from ..table import (
+    AsyncTable,
+    BlobMode,
+    Branches,
+    IndexStatistics,
+    Query,
+    Table,
+    Tags,
+)
 from ..types import BaseTokenizerType
 
 
@@ -349,6 +359,7 @@ class RemoteTable(Table):
         ngram_max_length: int = 3,
         prefix_only: bool = False,
         block_size: int = 128,
+        document_granularity: DocumentGranularity = DocumentGranularity.ROW,
         name: Optional[str] = None,
     ):
         """Create a full-text search index on a column.
@@ -371,6 +382,7 @@ class RemoteTable(Table):
             ngram_max_length=ngram_max_length,
             prefix_only=prefix_only,
             block_size=block_size,
+            document_granularity=document_granularity,
         )
         LOOP.run(
             self._table.create_index(
@@ -610,6 +622,7 @@ class RemoteTable(Table):
         fill_value: float = 0.0,
         progress: Optional[Union[bool, Callable, Any]] = None,
         write_parallelism: Optional[int] = None,
+        allow_external_blob_outside_bases: bool = False,
     ) -> AddResult:
         """Add more data to the [Table][lancedb.table.Table].
 
@@ -642,6 +655,8 @@ class RemoteTable(Table):
             data in flight. Defaults to an estimate based on the data size,
             capped at the number of CPU cores. Lower this if bulk ingestion is
             using too much memory.
+        allow_external_blob_outside_bases: bool, default False
+            Not supported on LanceDB Cloud. Setting this raises.
 
         Returns
         -------
@@ -658,6 +673,7 @@ class RemoteTable(Table):
                     fill_value=fill_value,
                     progress=progress,
                     write_parallelism=write_parallelism,
+                    allow_external_blob_outside_bases=allow_external_blob_outside_bases,
                 )
             )
         finally:
@@ -856,7 +872,7 @@ class RemoteTable(Table):
 
     def update(
         self,
-        where: Optional[str] = None,
+        where: Optional[Union[str, Expr]] = None,
         values: Optional[dict] = None,
         *,
         values_sql: Optional[Dict[str, str]] = None,
@@ -867,9 +883,11 @@ class RemoteTable(Table):
 
         Parameters
         ----------
-        where: str, optional
-            The SQL where clause to use when updating rows. For example, 'x = 2'
-            or 'x IN (1, 2, 3)'. The filter must not be empty, or it will error.
+        where: str or [Expr][lancedb.expr.Expr], optional
+            The filter condition. Can be a SQL string or a type-safe
+            [Expr][lancedb.expr.Expr] built with [col][lancedb.expr.col] and
+            [lit][lancedb.expr.lit]. The filter must not be empty, or it will
+            error.
         values: dict, optional
             The values to update. The keys are the column names and the values
             are the values to set.

python/python/lancedb/schema.py

@@ -4,30 +4,34 @@
 
 """Schema helpers for Lance blob columns."""
 
+import importlib
+from typing import TYPE_CHECKING
+
 import pyarrow as pa
+import pyarrow.ipc
+
+if TYPE_CHECKING:
+    from lance.blob import BlobType as BlobType
 
 _BLOB_EXTENSION_NAME = "lance.blob.v2"
 _BLOB_V1_KEY = "lance-encoding:blob"
 _ARROW_EXT_NAME_KEY = "ARROW:extension:name"
+_BLOB_V2_STORAGE_TYPE = pa.struct(
+    [
+        pa.field("data", pa.large_binary(), nullable=True),
+        pa.field("uri", pa.utf8(), nullable=True),
+        pa.field("position", pa.uint64(), nullable=True),
+        pa.field("size", pa.uint64(), nullable=True),
+    ]
+)
+_resolved_blob_type = None
 
 
-class BlobType(pa.ExtensionType):
-    """PyArrow extension type for a Lance blob v2 column.
-
-    Queries return descriptors; call :meth:`~lancedb.table.Table.fetch_blob_files`
-    for lazy reads or :meth:`~lancedb.table.Table.fetch_blobs` for eager bytes.
-    """
+class _FallbackBlobType(pa.ExtensionType):
+    """lance.blob.v2 extension type used when pylance is not installed."""
 
     def __init__(self) -> None:
-        storage_type = pa.struct(
-            [
-                pa.field("data", pa.large_binary(), nullable=True),
-                pa.field("uri", pa.utf8(), nullable=True),
-                pa.field("position", pa.uint64(), nullable=True),
-                pa.field("size", pa.uint64(), nullable=True),
-            ]
-        )
-        super().__init__(storage_type, _BLOB_EXTENSION_NAME)
+        pa.ExtensionType.__init__(self, _BLOB_V2_STORAGE_TYPE, _BLOB_EXTENSION_NAME)
 
     def __arrow_ext_serialize__(self) -> bytes:
         return b""
@@ -35,23 +39,16 @@ class BlobType(pa.ExtensionType):
     @classmethod
     def __arrow_ext_deserialize__(
         cls, storage_type: pa.DataType, serialized: bytes
-    ) -> "BlobType":
+    ) -> "_FallbackBlobType":
         return cls()
 
     def __reduce__(self):
-        # Ensure pickle round-trips on older pyarrow (apache/arrow#35599).
         return type(self).__arrow_ext_deserialize__, (
             self.storage_type,
             self.__arrow_ext_serialize__(),
         )
 
 
-try:
-    pa.register_extension_type(BlobType())  # type: ignore[arg-type]
-except pa.ArrowKeyError:
-    pass
-
-
 def _metadata_value(metadata: dict, key: str):
     return metadata.get(key.encode()) or metadata.get(key)
 
@@ -92,43 +89,105 @@ def is_blob_like_field(field: pa.Field) -> bool:
     return is_blob_v2_field(field) or _metadata_marks_legacy_blob(field.metadata or {})
 
 
-def _collect_blob_paths(schema: pa.Schema, is_blob) -> list[str]:
-    paths: list[str] = []
+def _collect_blob_paths(schema: pa.Schema, is_blob) -> list[tuple[str, bool]]:
+    """Walk the schema and return (path, has_list_ancestor) for each blob field."""
+    paths: list[tuple[str, bool]] = []
 
-    def walk(fields, prefix: str) -> None:
+    def walk(fields, prefix: str, has_list_ancestor: bool) -> None:
         for field in fields:
             path = f"{prefix}.{field.name}" if prefix else field.name
             if is_blob(field):
-                paths.append(path)
+                paths.append((path, has_list_ancestor))
             elif pa.types.is_struct(field.type):
-                walk(field.type, path)
+                walk(field.type, path, has_list_ancestor)
             elif (
                 pa.types.is_list(field.type)
                 or pa.types.is_large_list(field.type)
                 or pa.types.is_fixed_size_list(field.type)
             ):
-                walk([field.type.value_field], path)
+                walk([field.type.value_field], path, True)
 
-    walk(schema, "")
+    walk(schema, "", False)
     return paths
 
 
 def blob_column_paths(schema: pa.Schema) -> list[str]:
     """Dotted paths of blob-like columns (v2 extension or legacy metadata)."""
-    return _collect_blob_paths(schema, is_blob_like_field)
+    return [path for path, _ in _collect_blob_paths(schema, is_blob_like_field)]
 
 
 def blob_v2_column_paths(schema: pa.Schema) -> list[str]:
-    return _collect_blob_paths(schema, is_blob_v2_field)
+    return [path for path, _ in _collect_blob_paths(schema, is_blob_v2_field)]
+
+
+def row_addressable_blob_v2_paths(schema: pa.Schema) -> list[str]:
+    """Blob v2 paths with one blob addressable by table row id.
+
+    ``fetch_blobs`` and the descriptor row-id ride-along address one blob per
+    row, so a blob inside a list container has no row-id slot and no fetch
+    path. Those columns still store and query as raw descriptors.
+    """
+    return [
+        path
+        for path, has_list_ancestor in _collect_blob_paths(schema, is_blob_v2_field)
+        if not has_list_ancestor
+    ]
 
 
 def schema_has_blob_field(schema: pa.Schema) -> bool:
     return bool(blob_column_paths(schema))
 
 
+def _deserialize_registered_type(extension_type: pa.ExtensionType) -> pa.DataType:
+    """Return the type Arrow reconstructs for this extension name."""
+    schema = pa.schema([pa.field("value", extension_type)])
+    restored = pa.ipc.read_schema(schema.serialize())
+    return restored.field("value").type
+
+
+def _resolve_blob_type():
+    """Return the BlobType class this process should use.
+
+    pylance's class when it owns the lance.blob.v2 registry entry,
+    otherwise LanceDB's fallback. A different registered class is an error.
+    """
+    global _resolved_blob_type
+    if _resolved_blob_type is not None:
+        return _resolved_blob_type
+    try:
+        blob_module = importlib.import_module("lance.blob")
+    except ModuleNotFoundError as err:
+        if err.name not in ("lance", "lance.blob"):
+            raise
+    else:
+        blob_type = getattr(blob_module, "BlobType", None)
+        if blob_type is not None:
+            registered_type = _deserialize_registered_type(blob_type())
+            if type(registered_type) is not blob_type:
+                registered_cls = type(registered_type)
+                raise ValueError(
+                    "lance.blob.v2 is already registered by "
+                    f"{registered_cls.__module__}.{registered_cls.__qualname__}"
+                )
+            _resolved_blob_type = blob_type
+            return blob_type
+    try:
+        pa.register_extension_type(_FallbackBlobType())  # type: ignore[arg-type]
+    except pa.ArrowKeyError as err:
+        raise ValueError(
+            "lance.blob.v2 is already registered by another extension class"
+        ) from err
+    _resolved_blob_type = _FallbackBlobType
+    return _resolved_blob_type
+
+
 def blob(name: str, nullable: bool = True) -> pa.Field:
-    """Create a Lance blob v2 column field."""
-    return pa.field(name, BlobType(), nullable=nullable)
+    """Create a Lance blob v2 column field.
+
+    When pylance is installed this is ``lance.blob.BlobType``.
+    """
+    blob_type = _resolve_blob_type()
+    return pa.field(name, blob_type(), nullable=nullable)
 
 
 def vector(dimension: int, value_type: pa.DataType = pa.float32()) -> pa.DataType:
@@ -155,3 +214,11 @@ def vector(dimension: int, value_type: pa.DataType = pa.float32()) -> pa.DataTyp
     ... ])
     """
     return pa.list_(value_type, dimension)
+
+
+def __getattr__(name: str):
+    if name == "BlobType":
+        blob_type = _resolve_blob_type()
+        globals()["BlobType"] = blob_type
+        return blob_type
+    raise AttributeError(f"module {__name__!r} has no attribute {name!r}")

python/python/lancedb/table.py

@@ -85,6 +85,7 @@ from .query import (
     AsyncQuery,
     AsyncTakeQuery,
     AsyncVectorQuery,
+    DocumentGranularity,
     FullTextQuery,
     LanceEmptyQueryBuilder,
     LanceFtsQueryBuilder,
@@ -103,7 +104,12 @@ from .util import (
     value_to_sql,
 )
 from .index import lang_mapping
-from .schema import blob_v2_column_paths, schema_has_blob_field
+from .schema import (
+    blob_v2_column_paths,
+    is_blob_v2_field,
+    row_addressable_blob_v2_paths,
+    schema_has_blob_field,
+)
 
 
 def _should_push_down_query_table(
@@ -425,6 +431,7 @@ def _cast_to_target_schema(
 
     def gen():
         for batch in reader:
+            batch = _coerce_blob_write_columns(batch, reordered_schema)
             # Table but not RecordBatch has cast.
             cast_batches = (
                 pa.Table.from_batches([batch]).cast(reordered_schema).to_batches()
@@ -437,6 +444,166 @@ def _cast_to_target_schema(
     return pa.RecordBatchReader.from_batches(reordered_schema, gen())
 
 
+def _coerce_blob_write_columns(
+    batch: pa.RecordBatch, target_schema: pa.Schema
+) -> pa.RecordBatch:
+    """Materialize blob storage structs before the stream leaves Python.
+
+    merge_insert requires its source reader to already match the table's
+    physical schema. Unlike add and insert, it does not pass through
+    LanceDB's Rust blob coercion, so preserving binary input here would
+    reach Lance as binary and fail the schema check.
+    """
+    columns = []
+    fields = []
+    changed = False
+    for field, column in zip(batch.schema, batch.columns):
+        target_field = target_schema.field(field.name)
+        coerced = _coerce_blob_value(column, target_field)
+        if coerced is not column:
+            column = coerced
+            field = pa.field(
+                field.name,
+                coerced.type,
+                field.nullable,
+                target_field.metadata,
+            )
+            changed = True
+        columns.append(column)
+        fields.append(field)
+    if not changed:
+        return batch
+    return pa.RecordBatch.from_arrays(
+        columns, schema=pa.schema(fields, metadata=batch.schema.metadata)
+    )
+
+
+def _coerce_blob_value(column: pa.Array, target_field: pa.Field) -> pa.Array:
+    if is_blob_v2_field(target_field) and _can_coerce_to_blob(column.type):
+        return _coerce_value_to_blob(column, target_field)
+
+    target_type = target_field.type
+    if pa.types.is_struct(target_type) and pa.types.is_struct(column.type):
+        children = []
+        fields = []
+        changed = False
+        for source_field in column.type:
+            source_column = column.field(source_field.name)
+            nested_target = next(
+                (field for field in target_type if field.name == source_field.name),
+                None,
+            )
+            if nested_target is None:
+                children.append(source_column)
+                fields.append(source_field)
+                continue
+            coerced = _coerce_blob_value(source_column, nested_target)
+            if coerced is not source_column:
+                changed = True
+            child_array, child_type = _physical_array_and_type(coerced)
+            children.append(child_array)
+            fields.append(
+                pa.field(
+                    source_field.name,
+                    child_type,
+                    source_field.nullable,
+                    nested_target.metadata,
+                )
+            )
+        if not changed:
+            return column
+        return pa.StructArray.from_arrays(
+            children,
+            fields=fields,
+            mask=column.is_null() if column.null_count else None,
+        )
+
+    if _is_list_like(target_type) and _is_list_like(column.type):
+        return _coerce_blob_list_values(column, target_type.value_field)
+
+    return column
+
+
+def _coerce_blob_list_values(
+    column: pa.Array, target_value_field: pa.Field
+) -> pa.Array:
+    """Coerce blob values inside a list column, preserving offsets and nulls.
+
+    Works on the raw child values window instead of ``pc.list_flatten`` because
+    flatten drops values spanned by null slots, which would misalign offsets.
+    """
+    mask = column.is_null() if column.null_count else None
+    if pa.types.is_fixed_size_list(column.type):
+        list_size = column.type.list_size
+        values = column.values.slice(column.offset * list_size, len(column) * list_size)
+        coerced = _coerce_blob_value(values, target_value_field)
+        if coerced is values:
+            return column
+        physical_values, _ = _physical_array_and_type(coerced)
+        return pa.FixedSizeListArray.from_arrays(physical_values, list_size, mask=mask)
+    offsets = column.offsets
+    first_offset = offsets[0].as_py()
+    values = column.values.slice(
+        first_offset,
+        offsets[-1].as_py() - first_offset,
+    )
+    coerced = _coerce_blob_value(values, target_value_field)
+    if coerced is values:
+        return column
+    physical_values, _ = _physical_array_and_type(coerced)
+    if first_offset:
+        offsets = pc.subtract(offsets, pa.scalar(first_offset, offsets.type))
+    if pa.types.is_large_list(column.type):
+        return pa.LargeListArray.from_arrays(offsets, physical_values, mask=mask)
+    return pa.ListArray.from_arrays(offsets, physical_values, mask=mask)
+
+
+def _coerce_value_to_blob(values: pa.Array, target_field: pa.Field) -> pa.Array:
+    if pa.types.is_null(values.type):
+        data = pa.nulls(len(values), type=pa.large_binary())
+    elif pa.types.is_large_binary(values.type):
+        data = values
+    else:
+        data = values.cast(pa.large_binary())
+    length = len(values)
+    storage_type = target_field.type
+    if isinstance(storage_type, pa.ExtensionType):
+        storage_type = storage_type.storage_type
+    storage_fields = list(storage_type)
+    children = []
+    for storage_field in storage_fields:
+        if storage_field.name == "data":
+            children.append(data)
+        else:
+            children.append(pa.nulls(length, type=storage_field.type))
+    storage = pa.StructArray.from_arrays(
+        children,
+        fields=storage_fields,
+        mask=values.is_null() if values.null_count else None,
+    )
+    if isinstance(target_field.type, pa.ExtensionType):
+        return pa.ExtensionArray.from_storage(target_field.type, storage)
+    return storage
+
+
+def _physical_array_and_type(array: pa.Array) -> tuple[pa.Array, pa.DataType]:
+    if isinstance(array.type, pa.ExtensionType):
+        return array.storage, array.type.storage_type
+    return array, array.type
+
+
+def _can_coerce_to_blob(data_type: pa.DataType) -> bool:
+    return _is_binary_like(data_type) or pa.types.is_null(data_type)
+
+
+def _is_binary_like(data_type: pa.DataType) -> bool:
+    return (
+        pa.types.is_binary(data_type)
+        or pa.types.is_large_binary(data_type)
+        or pa.types.is_binary_view(data_type)
+    )
+
+
 def _field_extension_name(field: pa.Field) -> Optional[str]:
     extension_name = getattr(field.type, "extension_name", None)
     if extension_name is not None:
@@ -463,63 +630,71 @@ def _align_field_types(
         target_field = next((f for f in target_fields if f.name == field.name), None)
         if target_field is None:
             raise ValueError(f"Field '{field.name}' not found in target schema")
-        # Preserve arrow.json input until it reaches Lance. LanceDB exposes stored
-        # JSON columns as lance.json (JSONB-backed LargeBinary), but casting the
-        # input to that storage type here merely relabels the raw JSON bytes as
-        # JSONB. Lance must see arrow.json so it can perform the JSONB encoding.
-        if (
-            _field_extension_name(field) == "arrow.json"
-            and _field_extension_name(target_field) == "lance.json"
-        ):
-            new_fields.append(field)
-            continue
-        if pa.types.is_struct(target_field.type):
-            if pa.types.is_struct(field.type):
-                new_type = pa.struct(
-                    _align_field_types(
-                        field.type.fields,
-                        target_field.type.fields,
-                    )
+        new_fields.append(_align_field(field, target_field))
+    return new_fields
+
+
+def _align_list_value_field(
+    value_field: pa.Field, target_value_field: pa.Field
+) -> pa.Field:
+    # A list has exactly one child, so the inferred child name ("item") aligns
+    # positionally and adopts the table's child name; pa.Table.cast renames it.
+    return _align_field(value_field, target_value_field).with_name(
+        target_value_field.name
+    )
+
+
+def _align_field(field: pa.Field, target_field: pa.Field) -> pa.Field:
+    # Preserve arrow.json input until it reaches Lance. LanceDB exposes stored
+    # JSON columns as lance.json (JSONB-backed LargeBinary), but casting the
+    # input to that storage type here merely relabels the raw JSON bytes as
+    # JSONB. Lance must see arrow.json so it can perform the JSONB encoding.
+    if (
+        _field_extension_name(field) == "arrow.json"
+        and _field_extension_name(target_field) == "lance.json"
+    ):
+        return field
+    if pa.types.is_struct(target_field.type):
+        if pa.types.is_struct(field.type):
+            new_type = pa.struct(
+                _align_field_types(
+                    field.type.fields,
+                    target_field.type.fields,
                 )
-            else:
-                new_type = target_field.type
-        elif pa.types.is_list(target_field.type):
-            if _is_list_like(field.type):
-                new_type = pa.list_(
-                    _align_field_types(
-                        [field.type.value_field],
-                        [target_field.type.value_field],
-                    )[0]
-                )
-            else:
-                new_type = target_field.type
-        elif pa.types.is_large_list(target_field.type):
-            if _is_list_like(field.type):
-                new_type = pa.large_list(
-                    _align_field_types(
-                        [field.type.value_field],
-                        [target_field.type.value_field],
-                    )[0]
-                )
-            else:
-                new_type = target_field.type
-        elif pa.types.is_fixed_size_list(target_field.type):
-            if _is_list_like(field.type):
-                new_type = pa.list_(
-                    _align_field_types(
-                        [field.type.value_field],
-                        [target_field.type.value_field],
-                    )[0],
-                    target_field.type.list_size,
-                )
-            else:
-                new_type = target_field.type
+            )
         else:
             new_type = target_field.type
-        new_fields.append(
-            pa.field(field.name, new_type, field.nullable, target_field.metadata)
-        )
-    return new_fields
+    elif pa.types.is_list(target_field.type):
+        if _is_list_like(field.type):
+            new_type = pa.list_(
+                _align_list_value_field(
+                    field.type.value_field, target_field.type.value_field
+                )
+            )
+        else:
+            new_type = target_field.type
+    elif pa.types.is_large_list(target_field.type):
+        if _is_list_like(field.type):
+            new_type = pa.large_list(
+                _align_list_value_field(
+                    field.type.value_field, target_field.type.value_field
+                )
+            )
+        else:
+            new_type = target_field.type
+    elif pa.types.is_fixed_size_list(target_field.type):
+        if _is_list_like(field.type):
+            new_type = pa.list_(
+                _align_list_value_field(
+                    field.type.value_field, target_field.type.value_field
+                ),
+                target_field.type.list_size,
+            )
+        else:
+            new_type = target_field.type
+    else:
+        new_type = target_field.type
+    return pa.field(field.name, new_type, field.nullable, target_field.metadata)
 
 
 def _infer_subschema(
@@ -588,7 +763,7 @@ def sanitize_create_table(
         schema = data.schema
     else:
         if schema is not None:
-            data = pa.Table.from_pylist([], schema)
+            data = pa.Table.from_batches([], schema=schema)
     if schema is None:
         if data is None:
             raise ValueError("Either data or schema must be provided")
@@ -1168,6 +1343,7 @@ class Table(ABC):
         ngram_max_length: int = 3,
         prefix_only: bool = False,
         block_size: int = 128,
+        document_granularity: DocumentGranularity = DocumentGranularity.ROW,
         wait_timeout: Optional[timedelta] = None,
         name: Optional[str] = None,
     ):
@@ -1246,6 +1422,11 @@ class Table(ABC):
             The number of documents per compressed posting block. Must be 128
             or 256. A value of 256 uses the experimental FTS V3 format and
             may introduce breaking changes.
+        document_granularity: DocumentGranularity, default ROW
+            ``ROW`` treats the selected text in one table row as one document.
+            ``LIST_ELEMENT`` treats each element of the deepest list on the field
+            path as one document and returns its physical coordinates in
+            ``_doc_index`` for matching queries.
         wait_timeout: timedelta, optional
             The timeout to wait if indexing is asynchronous.
         name: str, optional
@@ -1269,6 +1450,7 @@ class Table(ABC):
         fill_value: float = 0.0,
         progress: Optional[Union[bool, Callable, Any]] = None,
         write_parallelism: Optional[int] = None,
+        allow_external_blob_outside_bases: bool = False,
     ) -> AddResult:
         """Add more data to the [Table][lancedb.table.Table].
 
@@ -1320,6 +1502,10 @@ class Table(ABC):
             data in flight. Defaults to an estimate based on the data size,
             capped at the number of CPU cores. Lower this if bulk ingestion is
             using too much memory.
+        allow_external_blob_outside_bases: bool, default False
+            Store blob URIs that sit outside registered blob bases. The row
+            keeps a reference, so the object has to stay readable. Local
+            tables only.
 
         Returns
         -------
@@ -1732,7 +1918,7 @@ class Table(ABC):
     @abstractmethod
     def update(
         self,
-        where: Optional[str] = None,
+        where: Optional[Union[str, Expr]] = None,
         values: Optional[dict] = None,
         *,
         values_sql: Optional[Dict[str, str]] = None,
@@ -1747,9 +1933,11 @@ class Table(ABC):
 
         Parameters
         ----------
-        where: str, optional
-            The SQL where clause to use when updating rows. For example, 'x = 2'
-            or 'x IN (1, 2, 3)'. The filter must not be empty, or it will error.
+        where: str or [Expr][lancedb.expr.Expr], optional
+            The filter condition. Can be a SQL string or a type-safe
+            [Expr][lancedb.expr.Expr] built with [col][lancedb.expr.col] and
+            [lit][lancedb.expr.lit]. The filter must not be empty, or it will
+            error.
         values: dict, optional
             The values to update. The keys are the column names and the values
             are the values to set.
@@ -1767,6 +1955,7 @@ class Table(ABC):
         Examples
         --------
         >>> import lancedb
+        >>> from lancedb.expr import col
         >>> import pandas as pd
         >>> data = pd.DataFrame({"x": [1, 2, 3], "vector": [[1.0, 2], [3, 4], [5, 6]]})
         >>> db = lancedb.connect("./.lancedb")
@@ -1776,7 +1965,7 @@ class Table(ABC):
         0  1  [1.0, 2.0]
         1  2  [3.0, 4.0]
         2  3  [5.0, 6.0]
-        >>> table.update(where="x = 2", values={"vector": [10.0, 10]})
+        >>> table.update(where=col("x") == 2, values={"vector": [10.0, 10]})
         UpdateResult(rows_updated=1, version=2)
         >>> table.to_pandas()
            x        vector
@@ -1976,9 +2165,11 @@ class Table(ABC):
             Function columns are supported only on LanceDB Cloud and
             Enterprise.
         computed: Dict[str, str], optional
-            A map of column name to a SQL expression defining the column. The
-            column's type and inputs are derived from the expression, so no
-            data type is supplied.
+            A mapping from output column names to SQL expressions derives each
+            output field from its expression. A direct projection of a Blob v2
+            field inherits Blob v2 semantics; other expressions derive their
+            ordinary Arrow type. Mapping order is declaration and dependency
+            order.
 
             Unlike ``transforms``, the expression is stored rather than
             evaluated now: the column is committed with no values, and rows get
@@ -2115,12 +2306,25 @@ class Table(ABC):
         ----------
         updates : dict
             One or more dicts, each with:
+
             - "path": str — dot-path to the field (e.g. "embedding" or "a.b.c").
             - "metadata": dict[str, str | None] — keys to set; a value of ``None``
               deletes that key.
             - "replace": bool, optional — replace the field's whole metadata map
               instead of merging (default False).
 
+            The following keys are treated specially, by convention, and should
+            be used when appropriate:
+
+            - "lancedb:description": for a human-readable description of a field.
+            - ``"lancedb:tag:<name>"`` for a user-defined key-value tag, where the
+                suffix names the tag category; e.g. "lancedb:tag:model": "clip".
+            - "lancedb:logical-column" for a column grouping; e.g. "feature_v1"
+                and "feature_v2" might be in the same logical column.
+            - "lancedb:status" for status options ("production", "candidate",
+                "deprecated", "archived") to designate the current life cycle
+                state of this column.
+
         Returns
         -------
         UpdateFieldMetadataResult
@@ -2670,7 +2874,7 @@ class LanceTable(Table):
             arrow_tbl = self.to_arrow()
             if blob_mode == "descriptions":
                 arrow_tbl = strip_auto_row_ids(
-                    arrow_tbl, blob_v2_column_paths(self.schema)
+                    arrow_tbl, row_addressable_blob_v2_paths(self.schema)
                 )
             return arrow_tbl.to_pandas(**kwargs)
 
@@ -3268,6 +3472,7 @@ class LanceTable(Table):
         ngram_max_length: int = 3,
         prefix_only: bool = False,
         block_size: int = 128,
+        document_granularity: DocumentGranularity = DocumentGranularity.ROW,
         name: Optional[str] = None,
     ):
         """Create a full-text search index on a column.
@@ -3319,7 +3524,11 @@ class LanceTable(Table):
             tokenizer_configs = self.infer_tokenizer_configs(tokenizer_name)
             tokenizer_configs["custom_stop_words"] = custom_stop_words
 
-        config = FTS(block_size=block_size, **tokenizer_configs)
+        config = FTS(
+            block_size=block_size,
+            document_granularity=document_granularity,
+            **tokenizer_configs,
+        )
 
         try:
             LOOP.run(
@@ -3409,6 +3618,7 @@ class LanceTable(Table):
         fill_value: float = 0.0,
         progress: Optional[Union[bool, Callable, Any]] = None,
         write_parallelism: Optional[int] = None,
+        allow_external_blob_outside_bases: bool = False,
     ) -> AddResult:
         """Add data to the table.
         If vector columns are missing and the table
@@ -3436,6 +3646,9 @@ class LanceTable(Table):
             data in flight. Defaults to an estimate based on the data size,
             capped at the number of CPU cores. Lower this if bulk ingestion is
             using too much memory.
+        allow_external_blob_outside_bases: bool, default False
+            Allow blob URIs outside registered bases. See :meth:`Table.add`.
+            Local tables only.
 
         Returns
         -------
@@ -3452,6 +3665,7 @@ class LanceTable(Table):
                     fill_value=fill_value,
                     progress=progress,
                     write_parallelism=write_parallelism,
+                    allow_external_blob_outside_bases=allow_external_blob_outside_bases,
                 )
             )
         finally:
@@ -3806,7 +4020,7 @@ class LanceTable(Table):
 
     def update(
         self,
-        where: Optional[str] = None,
+        where: Optional[Union[str, Expr]] = None,
         values: Optional[dict] = None,
         *,
         values_sql: Optional[Dict[str, str]] = None,
@@ -3817,9 +4031,11 @@ class LanceTable(Table):
 
         Parameters
         ----------
-        where: str, optional
-            The SQL where clause to use when updating rows. For example, 'x = 2'
-            or 'x IN (1, 2, 3)'. The filter must not be empty, or it will error.
+        where: str or [Expr][lancedb.expr.Expr], optional
+            The filter condition. Can be a SQL string or a type-safe
+            [Expr][lancedb.expr.Expr] built with [col][lancedb.expr.col] and
+            [lit][lancedb.expr.lit]. The filter must not be empty, or it will
+            error.
         values: dict, optional
             The values to update. The keys are the column names and the values
             are the values to set.
@@ -3837,6 +4053,7 @@ class LanceTable(Table):
         Examples
         --------
         >>> import lancedb
+        >>> from lancedb.expr import col
         >>> import pandas as pd
         >>> data = pd.DataFrame({"x": [1, 2, 3], "vector": [[1.0, 2], [3, 4], [5, 6]]})
         >>> db = lancedb.connect("./.lancedb")
@@ -3846,7 +4063,7 @@ class LanceTable(Table):
         0  1  [1.0, 2.0]
         1  2  [3.0, 4.0]
         2  3  [5.0, 6.0]
-        >>> table.update(where="x = 2", values={"vector": [10.0, 10]})
+        >>> table.update(where=col("x") == 2, values={"vector": [10.0, 10]})
         UpdateResult(rows_updated=1, version=2)
         >>> table.to_pandas()
            x        vector
@@ -5061,7 +5278,9 @@ class AsyncTable:
         if blob_mode == "descriptions" or not schema_has_blob_field(schema):
             arrow_tbl = await self.to_arrow()
             if blob_mode == "descriptions":
-                arrow_tbl = strip_auto_row_ids(arrow_tbl, blob_v2_column_paths(schema))
+                arrow_tbl = strip_auto_row_ids(
+                    arrow_tbl, row_addressable_blob_v2_paths(schema)
+                )
             return arrow_tbl.to_pandas(**kwargs)
 
         if blob_mode == "lazy" and get_uri_scheme(await self.uri()) == "memory":
@@ -5365,6 +5584,7 @@ class AsyncTable:
         fill_value: Optional[float] = None,
         progress: Optional[Union[bool, Callable, Any]] = None,
         write_parallelism: Optional[int] = None,
+        allow_external_blob_outside_bases: bool = False,
     ) -> AddResult:
         """Add more data to the [AsyncTable][lancedb.table.AsyncTable].
 
@@ -5395,6 +5615,9 @@ class AsyncTable:
             data in flight. Defaults to an estimate based on the data size,
             capped at the number of CPU cores. Lower this if bulk ingestion is
             using too much memory.
+        allow_external_blob_outside_bases: bool, default False
+            Allow blob URIs outside registered bases. See :meth:`Table.add`.
+            Local tables only.
 
         """
         schema = await self.schema()
@@ -5431,6 +5654,7 @@ class AsyncTable:
                 mode or "append",
                 progress=progress,
                 write_parallelism=write_parallelism,
+                allow_external_blob_outside_bases=allow_external_blob_outside_bases,
             )
         except RuntimeError as e:
             if "Cast error" in str(e):
@@ -5955,7 +6179,7 @@ class AsyncTable:
         self,
         updates: Optional[Dict[str, Any]] = None,
         *,
-        where: Optional[str] = None,
+        where: Optional[Union[str, Expr]] = None,
         updates_sql: Optional[Dict[str, str]] = None,
     ) -> UpdateResult:
         """
@@ -5970,9 +6194,11 @@ class AsyncTable:
             The updates to apply.  The keys should be the name of the column to
             update.  The values should be the new values to assign.  This is
             required unless updates_sql is supplied.
-        where: str, optional
-            An SQL filter that controls which rows are updated. For example, 'x = 2'
-            or 'x IN (1, 2, 3)'.  Only rows that satisfy this filter will be udpated.
+        where: str or [Expr][lancedb.expr.Expr], optional
+            The filter condition. Can be a SQL string or a type-safe
+            [Expr][lancedb.expr.Expr] built with [col][lancedb.expr.col] and
+            [lit][lancedb.expr.lit]. Only rows that satisfy this filter will
+            be updated.
         updates_sql: dict, optional
             The updates to apply, expressed as SQL expression strings.  The keys should
             be column names. The values should be SQL expressions.  These can be SQL
@@ -5990,13 +6216,14 @@ class AsyncTable:
         --------
         >>> import asyncio
         >>> import lancedb
+        >>> from lancedb.expr import col
         >>> import pandas as pd
         >>> async def demo_update():
         ...     data = pd.DataFrame({"x": [1, 2], "vector": [[1, 2], [3, 4]]})
         ...     db = await lancedb.connect_async("./.lancedb")
         ...     table = await db.create_table("my_table", data)
         ...     # x is [1, 2], vector is [[1, 2], [3, 4]]
-        ...     await table.update({"vector": [10, 10]}, where="x = 2")
+        ...     await table.update({"vector": [10, 10]}, where=col("x") == 2)
         ...     # x is [1, 2], vector is [[1, 2], [10, 10]]
         ...     await table.update(updates_sql={"x": "x + 1"})
         ...     # x is [2, 3], vector is [[1, 2], [10, 10]]
@@ -6010,7 +6237,8 @@ class AsyncTable:
         if updates is not None:
             updates_sql = {k: value_to_sql(v) for k, v in updates.items()}
 
-        return await self._inner.update(updates_sql, where)
+        predicate = where.to_sql() if isinstance(where, Expr) else where
+        return await self._inner.update(updates_sql, predicate)
 
     async def add_columns(
         self,
@@ -6042,8 +6270,11 @@ class AsyncTable:
             Function columns are supported only on LanceDB Cloud and
             Enterprise.
         computed: Dict[str, str], optional
-            A map of column name to a SQL expression defining the column. The
-            column's type and inputs are derived from the expression.
+            A mapping from output column names to SQL expressions derives each
+            output field from its expression. A direct projection of a Blob v2
+            field inherits Blob v2 semantics; other expressions derive their
+            ordinary Arrow type. Mapping order is declaration and dependency
+            order.
 
             Unlike ``transforms``, the expression is stored rather than
             evaluated now: the column is committed with no values, and rows get

python/python/tests/docs/test_basic.py

@@ -105,7 +105,7 @@ def test_quickstart(tmp_path):
     tbl.create_index(num_sub_vectors=1)
     # --8<-- [end:create_index]
     # --8<-- [start:delete_rows]
-    tbl.delete('item = "fizz"')
+    tbl.delete("item = 'fizz'")
     # --8<-- [end:delete_rows]
     # --8<-- [start:drop_table]
     db.drop_table("my_table")
@@ -201,7 +201,7 @@ async def test_quickstart_async(tmp_path):
     await tbl.create_index("vector")
     # --8<-- [end:create_index_async]
     # --8<-- [start:delete_rows_async]
-    await tbl.delete('item = "fizz"')
+    await tbl.delete("item = 'fizz'")
     # --8<-- [end:delete_rows_async]
     # --8<-- [start:drop_table_async]
     await db.drop_table("my_table_async")

python/python/tests/docs/test_guide_tables.py

@@ -266,7 +266,7 @@ def test_table():
     tbl.add(pydantic_model_items)
     # --8<-- [end:add_table_from_pydantic]
     # --8<-- [start:delete_row]
-    tbl.delete('item = "fizz"')
+    tbl.delete("item = 'fizz'")
     # --8<-- [end:delete_row]
     # --8<-- [start:delete_specific_row]
     data = [
@@ -538,7 +538,7 @@ async def test_table_async():
     await async_tbl.add(pydantic_model_items)
     # --8<-- [end:add_table_async_from_pydantic]
     # --8<-- [start:delete_row_async]
-    await async_tbl.delete('item = "fizz"')
+    await async_tbl.delete("item = 'fizz'")
     # --8<-- [end:delete_row_async]
     # --8<-- [start:delete_specific_row_async]
     data = [

python/python/tests/test_blob.py

@@ -2,17 +2,41 @@
 # SPDX-FileCopyrightText: Copyright The LanceDB Authors
 
 import io
+import subprocess
+import sys
+import textwrap
 
+import lance
 import pyarrow as pa
 import pyarrow.compute as pc
 import pytest
+from lance.blob import BlobType as LanceBlobType
 
 import lancedb
-from lancedb._blob import read_row_ids_from_hits, stash_auto_row_ids
+from lancedb._blob import (
+    blob_v2_projection_sources,
+    read_row_ids_from_hits,
+    stash_auto_row_ids,
+)
+from lancedb.expr import col
 from lancedb.index import FTS
 from lancedb.schema import blob_column_paths, blob_v2_column_paths
 
 
+_HIDE_LANCE_BLOB = """\
+import importlib.abc
+import sys
+
+class _MissingLanceBlob(importlib.abc.MetaPathFinder):
+    def find_spec(self, fullname, path, target=None):
+        if fullname == "lance.blob" or fullname.startswith("lance.blob."):
+            raise ModuleNotFoundError(fullname, name="lance.blob")
+
+sys.modules.pop("lance.blob", None)
+sys.meta_path.insert(0, _MissingLanceBlob())
+"""
+
+
 def _blob_table(name, rows):
     db = lancedb.connect("memory:///")
     schema = pa.schema([pa.field("id", pa.int64()), lancedb.blob("image")])
@@ -46,6 +70,181 @@ def test_blob_factory_declares_v2_field():
     field = lancedb.blob("image")
     assert isinstance(field.type, pa.ExtensionType)
     assert field.type.extension_name == "lance.blob.v2"
+    assert lancedb.BlobType is LanceBlobType
+    assert type(field.type) is LanceBlobType
+
+
+def test_blob_type_works_without_pylance():
+    script = _HIDE_LANCE_BLOB + textwrap.dedent(
+        """\
+        import lancedb
+        import pyarrow as pa
+
+        field = lancedb.blob("image")
+        if not isinstance(field.type, pa.ExtensionType):
+            raise SystemExit("expected an extension type")
+        if field.type.extension_name != "lance.blob.v2":
+            raise SystemExit(field.type.extension_name)
+        if lancedb.BlobType is not type(field.type):
+            raise SystemExit("BlobType is not the field type class")
+        if lancedb.BlobType.__module__ != "lancedb.schema":
+            raise SystemExit(lancedb.BlobType.__module__)
+
+        db = lancedb.connect("memory:///")
+        table = db.create_table(
+            "images",
+            schema=pa.schema([pa.field("id", pa.int64()), field]),
+        )
+        table.add([{"id": 1, "image": b"hello"}])
+        result = (
+            table.merge_insert("id")
+            .when_matched_update_all()
+            .when_not_matched_insert_all()
+            .execute([{"id": 1, "image": b"updated"}, {"id": 2, "image": b"inserted"}])
+        )
+        if result.num_updated_rows != 1 or result.num_inserted_rows != 1:
+            raise SystemExit(
+                f"merge_insert rows updated={result.num_updated_rows} "
+                f"inserted={result.num_inserted_rows}"
+            )
+        """
+    )
+    result = subprocess.run(
+        [sys.executable, "-c", script],
+        capture_output=True,
+        text=True,
+        check=False,
+    )
+    assert result.returncode == 0, result.stderr
+
+
+def test_blob_resolves_pylance_type_without_eager_import():
+    script = textwrap.dedent(
+        """\
+        import sys
+        import lancedb
+
+        if "lance.blob" in sys.modules:
+            raise SystemExit("import lancedb imported lance.blob")
+        field = lancedb.blob("image")
+        from lance.blob import BlobType
+
+        if type(field.type) is not BlobType:
+            raise SystemExit(f"{type(field.type)} is not {BlobType}")
+        import lance
+
+        image = lance.blob_array([b"x"])
+        if type(image.type) is not BlobType:
+            raise SystemExit("blob_array used a different class")
+        if type(image.type) is not type(field.type):
+            raise SystemExit("field and array classes differ")
+        """
+    )
+    result = subprocess.run(
+        [sys.executable, "-c", script],
+        capture_output=True,
+        text=True,
+        check=False,
+    )
+    assert result.returncode == 0, result.stderr
+
+
+def test_blob_fallback_fails_if_name_already_registered():
+    script = _HIDE_LANCE_BLOB + textwrap.dedent(
+        """\
+        import pyarrow as pa
+
+        class OtherBlobType(pa.ExtensionType):
+            def __init__(self):
+                super().__init__(
+                    pa.struct([pa.field("data", pa.large_binary())]),
+                    "lance.blob.v2",
+                )
+
+            def __arrow_ext_serialize__(self):
+                return b""
+
+            @classmethod
+            def __arrow_ext_deserialize__(cls, storage_type, serialized):
+                return cls()
+
+        pa.register_extension_type(OtherBlobType())
+        import lancedb
+
+        try:
+            lancedb.blob("image")
+        except ValueError as err:
+            if "already registered" not in str(err):
+                raise SystemExit(err)
+        else:
+            raise SystemExit("expected ValueError")
+        """
+    )
+    result = subprocess.run(
+        [sys.executable, "-c", script],
+        capture_output=True,
+        text=True,
+        check=False,
+    )
+    assert result.returncode == 0, result.stderr
+
+
+def test_blob_type_rejects_competing_registration_with_pylance():
+    script = textwrap.dedent(
+        """\
+        import pyarrow as pa
+        import pyarrow.ipc
+
+        class OtherBlobType(pa.ExtensionType):
+            def __init__(self):
+                super().__init__(
+                    pa.struct(
+                        [
+                            pa.field("data", pa.large_binary()),
+                            pa.field("uri", pa.utf8()),
+                            pa.field("position", pa.uint64()),
+                            pa.field("size", pa.uint64()),
+                        ]
+                    ),
+                    "lance.blob.v2",
+                )
+
+            def __arrow_ext_serialize__(self):
+                return b""
+
+            @classmethod
+            def __arrow_ext_deserialize__(cls, storage_type, serialized):
+                return cls()
+
+        pa.register_extension_type(OtherBlobType())
+
+        from lance.blob import BlobType
+
+        if BlobType is OtherBlobType:
+            raise SystemExit("pylance BlobType was replaced")
+        schema = pa.schema([pa.field("value", BlobType())])
+        restored = pa.ipc.read_schema(schema.serialize())
+        if type(restored.field("value").type) is not OtherBlobType:
+            raise SystemExit(type(restored.field("value").type))
+
+        import lancedb
+
+        try:
+            lancedb.blob("image")
+        except ValueError as err:
+            if "__main__.OtherBlobType" not in str(err):
+                raise SystemExit(err)
+        else:
+            raise SystemExit("expected ValueError")
+        """
+    )
+    result = subprocess.run(
+        [sys.executable, "-c", script],
+        capture_output=True,
+        text=True,
+        check=False,
+    )
+    assert result.returncode == 0, result.stderr
 
 
 def test_blob_v2_column_paths_include_list_children():
@@ -70,6 +269,14 @@ def test_blob_v2_column_paths_include_list_children():
     ]
 
 
+def test_blob_v2_projection_sources_use_typed_column_name():
+    schema = pa.schema([lancedb.blob("blob")])
+
+    assert blob_v2_projection_sources(schema, {"blob_alias": col("blob")}) == {
+        "blob_alias": "blob"
+    }
+
+
 def _legacy_v1_table(name):
     db = lancedb.connect("memory:///")
     schema = pa.schema(
@@ -166,6 +373,20 @@ async def test_async_table_to_pandas_descriptions_mode_omits_row_id():
     assert set(descriptor.keys()) == {"kind", "position", "size", "blob_id", "blob_uri"}
 
 
+@pytest.mark.asyncio
+async def test_async_typed_blob_projection_preserves_source_column():
+    db = await lancedb.connect_async("memory:///typed_blob_projection")
+    schema = pa.schema([pa.field("id", pa.int64()), lancedb.blob("blob")])
+    table = await db.create_table("typed_blob_projection", schema=schema)
+    await table.add([{"id": 1, "blob": b"alpha"}])
+
+    hits = await table.query().select({"blob_alias": col("blob")}).to_arrow()
+
+    assert "_lance_row_id" in hits.schema.field("blob_alias").type.names
+    blobs = await table.fetch_blobs("blob", hits)
+    assert blobs.to_pylist() == [b"alpha"]
+
+
 def test_fetch_blobs_round_trip():
     table = _blob_table(
         "round_trip",
@@ -176,6 +397,292 @@ def test_fetch_blobs_round_trip():
     assert [blobs[0].as_py(), blobs[1].as_py()] == [b"alpha", b"beta"]
 
 
+def test_merge_insert_writes_python_bytes():
+    table = _blob_table("merge_bytes", [{"id": 1, "image": b"before"}])
+    result = (
+        table.merge_insert("id")
+        .when_matched_update_all()
+        .when_not_matched_insert_all()
+        .execute([{"id": 1, "image": b"updated"}, {"id": 2, "image": b"inserted"}])
+    )
+    assert result.num_updated_rows == 1
+    assert result.num_inserted_rows == 1
+    by_id = _row_ids_by_id(table)
+    blobs = table.fetch_blobs("image", [by_id[1], by_id[2]])
+    assert blobs.to_pylist() == [b"updated", b"inserted"]
+
+
+def test_merge_insert_bytes_after_reopen_without_touching_blob_type(tmp_path):
+    db = lancedb.connect(tmp_path)
+    schema = pa.schema([pa.field("id", pa.int64()), lancedb.blob("image")])
+    table = db.create_table("images", schema=schema)
+    table.add([{"id": 1, "image": b"hello"}])
+
+    script = textwrap.dedent(
+        f"""\
+        import lancedb
+
+        db = lancedb.connect({str(tmp_path)!r})
+        table = db.open_table("images")
+        image_type = table.schema.field("image").type
+        if type(image_type).__name__ != "StructType":
+            raise SystemExit(f"expected StructType, got {{type(image_type)}}")
+        result = (
+            table.merge_insert("id")
+            .when_matched_update_all()
+            .when_not_matched_insert_all()
+            .execute(
+                [{{"id": 1, "image": b"updated"}}, {{"id": 2, "image": b"inserted"}}]
+            )
+        )
+        if result.num_updated_rows != 1 or result.num_inserted_rows != 1:
+            raise SystemExit(
+                f"rows updated={{result.num_updated_rows}} "
+                f"inserted={{result.num_inserted_rows}}"
+            )
+        hits = table.search().with_row_id(True).limit(10).to_arrow()
+        by_id = dict(zip(hits["id"].to_pylist(), hits["_rowid"].to_pylist()))
+        blobs = table.fetch_blobs("image", [by_id[1], by_id[2]])
+        if blobs.to_pylist() != [b"updated", b"inserted"]:
+            raise SystemExit(blobs.to_pylist())
+        """
+    )
+    result = subprocess.run(
+        [sys.executable, "-c", script],
+        capture_output=True,
+        text=True,
+        check=False,
+    )
+    assert result.returncode == 0, result.stderr
+
+
+def test_merge_insert_bytes_after_reopen_without_pylance(tmp_path):
+    db = lancedb.connect(tmp_path)
+    schema = pa.schema([pa.field("id", pa.int64()), lancedb.blob("image")])
+    table = db.create_table("images", schema=schema)
+    table.add([{"id": 1, "image": b"hello"}])
+
+    script = _HIDE_LANCE_BLOB + textwrap.dedent(
+        f"""\
+        import lancedb
+
+        db = lancedb.connect({str(tmp_path)!r})
+        table = db.open_table("images")
+        image_type = table.schema.field("image").type
+        if type(image_type).__name__ != "StructType":
+            raise SystemExit(f"expected StructType, got {{type(image_type)}}")
+        result = (
+            table.merge_insert("id")
+            .when_matched_update_all()
+            .when_not_matched_insert_all()
+            .execute(
+                [{{"id": 1, "image": b"updated"}}, {{"id": 2, "image": b"inserted"}}]
+            )
+        )
+        if result.num_updated_rows != 1 or result.num_inserted_rows != 1:
+            raise SystemExit(
+                f"rows updated={{result.num_updated_rows}} "
+                f"inserted={{result.num_inserted_rows}}"
+            )
+        hits = table.search().with_row_id(True).limit(10).to_arrow()
+        by_id = dict(zip(hits["id"].to_pylist(), hits["_rowid"].to_pylist()))
+        blobs = table.fetch_blobs("image", [by_id[1], by_id[2]])
+        if blobs.to_pylist() != [b"updated", b"inserted"]:
+            raise SystemExit(blobs.to_pylist())
+        """
+    )
+    result = subprocess.run(
+        [sys.executable, "-c", script],
+        capture_output=True,
+        text=True,
+        check=False,
+    )
+    assert result.returncode == 0, result.stderr
+
+
+def test_merge_insert_blob_array_into_reopened_unregistered_table(tmp_path):
+    db = lancedb.connect(tmp_path)
+    schema = pa.schema([pa.field("id", pa.int64()), lancedb.blob("image")])
+    table = db.create_table("images", schema=schema)
+    table.add([{"id": 1, "image": b"before"}])
+
+    script = textwrap.dedent(
+        f"""\
+        import pyarrow as pa
+        import lancedb
+
+        db = lancedb.connect({str(tmp_path)!r})
+        table = db.open_table("images")
+        image_type = table.schema.field("image").type
+        if type(image_type).__name__ != "StructType":
+            raise SystemExit(
+                f"expected StructType before lance import, got {{type(image_type)}}"
+            )
+
+        import lance
+
+        updates = pa.Table.from_arrays(
+            [
+                pa.array([1, 2], type=pa.int64()),
+                lance.blob_array([b"updated", b"inserted"]),
+            ],
+            names=["id", "image"],
+        )
+        result = (
+            table.merge_insert("id")
+            .when_matched_update_all()
+            .when_not_matched_insert_all()
+            .execute(updates)
+        )
+        if result.num_updated_rows != 1 or result.num_inserted_rows != 1:
+            raise SystemExit(
+                f"rows updated={{result.num_updated_rows}} "
+                f"inserted={{result.num_inserted_rows}}"
+            )
+        hits = table.search().with_row_id(True).limit(10).to_arrow()
+        by_id = dict(zip(hits["id"].to_pylist(), hits["_rowid"].to_pylist()))
+        blobs = table.fetch_blobs("image", [by_id[1], by_id[2]])
+        if blobs.to_pylist() != [b"updated", b"inserted"]:
+            raise SystemExit(blobs.to_pylist())
+        """
+    )
+    result = subprocess.run(
+        [sys.executable, "-c", script],
+        capture_output=True,
+        text=True,
+        check=False,
+    )
+    assert result.returncode == 0, result.stderr
+
+
+def test_add_all_null_blob_column():
+    db = lancedb.connect("memory:///")
+    schema = pa.schema([pa.field("id", pa.int64()), lancedb.blob("image")])
+    table = db.create_table("all_null", schema=schema)
+    table.add([{"id": 1, "image": None}, {"id": 2, "image": None}])
+    by_id = _row_ids_by_id(table)
+    blobs = table.fetch_blobs("image", [by_id[1], by_id[2]])
+    assert blobs.to_pylist() == [None, None]
+
+
+def test_create_table_nested_blob_schema_without_rows():
+    db = lancedb.connect("memory:///")
+    schema = pa.schema(
+        [
+            pa.field("id", pa.int64()),
+            pa.field("info", pa.struct([lancedb.blob("blob")])),
+            pa.field("images", pa.list_(lancedb.blob("image"))),
+        ]
+    )
+    table = db.create_table("nested_empty", schema=schema)
+    assert table.count_rows() == 0
+
+
+def test_merge_insert_nested_blob_dicts():
+    db = lancedb.connect("memory:///")
+    info = pa.StructArray.from_arrays(
+        [
+            pa.array(["first"], type=pa.string()),
+            _blob_array("blob", [b"before"]),
+        ],
+        names=["name", "blob"],
+    )
+    data = pa.Table.from_arrays(
+        [pa.array([1], type=pa.int64()), info],
+        names=["id", "info"],
+    )
+    table = db.create_table("nested_merge", data=data)
+    result = (
+        table.merge_insert("id")
+        .when_matched_update_all()
+        .execute([{"id": 1, "info": {"name": "first", "blob": b"after"}}])
+    )
+    assert result.num_updated_rows == 1
+    by_id = _row_ids_by_id(table)
+    blobs = table.fetch_blobs("info.blob", [by_id[1]])
+    assert blobs.to_pylist() == [b"after"]
+
+
+def _list_blob_table(name):
+    db = lancedb.connect("memory:///")
+    blob_field = lancedb.blob("image")
+    images = pa.ListArray.from_arrays(
+        pa.array([0, 1], type=pa.int32()), _blob_array("image", [b"before"])
+    )
+    data = pa.Table.from_arrays(
+        [pa.array([1], type=pa.int64()), images],
+        schema=pa.schema(
+            [pa.field("id", pa.int64()), pa.field("images", pa.list_(blob_field))]
+        ),
+    )
+    return db.create_table(name, data=data)
+
+
+def test_merge_insert_list_blob_dicts():
+    table = _list_blob_table("list_merge")
+    result = (
+        table.merge_insert("id")
+        .when_matched_update_all()
+        .when_not_matched_insert_all()
+        .execute([{"id": 1, "images": [b"one", b"two"]}, {"id": 2, "images": None}])
+    )
+    assert result.num_updated_rows == 1
+    assert result.num_inserted_rows == 1
+    hits = table.search().limit(10).to_arrow()
+    sizes = {
+        row["id"]: None if row["images"] is None else [d["size"] for d in row["images"]]
+        for row in hits.to_pylist()
+    }
+    assert sizes == {1: [3, 3], 2: None}
+
+
+def test_list_blob_column_queries_as_raw_descriptors():
+    table = _list_blob_table("list_query")
+    hits = table.search().limit(10).to_arrow()
+    element = hits.schema.field("images").type.value_type
+    assert pa.types.is_struct(element)
+    assert "_lance_row_id" not in element.names
+    with pytest.raises(ValueError, match="expected struct before segment"):
+        table.fetch_blobs("images.image", [0])
+
+
+def test_row_addressable_paths_exclude_list_children():
+    from lancedb.schema import row_addressable_blob_v2_paths
+
+    schema = pa.schema(
+        [
+            pa.field("id", pa.int64()),
+            pa.field("info", pa.struct([lancedb.blob("blob")])),
+            pa.field("images", pa.list_(lancedb.blob("image"))),
+        ]
+    )
+    assert blob_v2_column_paths(schema) == ["info.blob", "images.image"]
+    assert row_addressable_blob_v2_paths(schema) == ["info.blob"]
+
+
+def test_merge_insert_writes_pylance_blob_array():
+    table = _blob_table("merge_pylance", [{"id": 1, "image": b"before"}])
+    image = lance.blob_array([b"updated", b"inserted"])
+    assert type(image.type) is LanceBlobType
+    assert type(image.type) is type(lancedb.BlobType())
+    updates = pa.Table.from_arrays(
+        [pa.array([1, 2], type=pa.int64()), image], names=["id", "image"]
+    )
+
+    result = (
+        table.merge_insert("id")
+        .when_matched_update_all()
+        .when_not_matched_insert_all()
+        .execute(updates)
+    )
+
+    assert result.num_updated_rows == 1
+    assert result.num_inserted_rows == 1
+    by_id = _row_ids_by_id(table)
+    blobs = table.fetch_blobs("image", [by_id[1], by_id[2]])
+    assert blobs.to_pylist() == [b"updated", b"inserted"]
+
+
 def test_fetch_blobs_accepts_query_result():
     table = _blob_table("from_result", [{"id": 1, "image": b"gamma"}])
     hits = table.search().limit(10).to_arrow()
@@ -403,6 +910,50 @@ async def test_blob_v2_hybrid_fetch_blobs_async():
     assert {blobs[i].as_py() for i in range(len(blobs))} == {b"alpha", b"beta"}
 
 
+@pytest.mark.asyncio
+async def test_async_hybrid_typed_blob_projection_preserves_source_column():
+    db = await lancedb.connect_async("memory:///hybrid_typed_blob")
+    schema = pa.schema(
+        [
+            pa.field("id", pa.int64()),
+            pa.field("text", pa.utf8()),
+            pa.field("vector", pa.list_(pa.float32(), list_size=2)),
+            lancedb.blob("blob"),
+        ]
+    )
+    table = await db.create_table("hybrid_typed_blob", schema=schema)
+    await table.add(
+        [
+            {
+                "id": 1,
+                "text": "hello alpha",
+                "vector": [1.0, 0.0],
+                "blob": b"alpha",
+            },
+            {
+                "id": 2,
+                "text": "hello beta",
+                "vector": [0.9, 0.1],
+                "blob": b"beta",
+            },
+        ]
+    )
+    await table.create_index("text", config=FTS(with_position=False))
+
+    hits = await (
+        table.query()
+        .nearest_to([1.0, 0.0])
+        .nearest_to_text("hello")
+        .select({"blob_alias": col("blob")})
+        .limit(2)
+        .to_arrow()
+    )
+
+    assert "_lance_row_id" in hits.schema.field("blob_alias").type.names
+    blobs = await table.fetch_blobs("blob", hits)
+    assert {blobs[i].as_py() for i in range(len(blobs))} == {b"alpha", b"beta"}
+
+
 def test_blob_file_seek_read_and_read_range():
     payload = _identifiable_payload(1024)
     table = _blob_table("seek_read", [{"id": 1, "image": payload}])
@@ -617,3 +1168,71 @@ def test_fetch_blobs_nested_path_survives_sort_after_query():
 def _identifiable_payload(size: int) -> bytes:
     block = 256
     return b"".join(bytes([i % 256]) * block for i in range(size // block))
+
+
+def _external_uri_blob_array(uris):
+    blob_type = lancedb.blob("image").type
+    storage_type = blob_type.storage_type
+    child_names = [field.name for field in storage_type]
+    assert "uri" in child_names, "blob layout no longer has a uri child"
+    children = [
+        pa.array(uris if field.name == "uri" else [None] * len(uris), type=field.type)
+        for field in storage_type
+    ]
+    storage = pa.StructArray.from_arrays(children, fields=list(storage_type))
+    return pa.ExtensionArray.from_storage(blob_type, storage)
+
+
+def _external_uri_table_and_rows(name, uris):
+    db = lancedb.connect("memory:///")
+    schema = pa.schema([pa.field("id", pa.int64()), lancedb.blob("image")])
+    table = db.create_table(name, schema=schema)
+    rows = pa.Table.from_arrays(
+        [
+            pa.array(range(len(uris)), type=pa.int64()),
+            _external_uri_blob_array(uris),
+        ],
+        schema=schema,
+    )
+    return table, rows
+
+
+def test_add_external_uri_struct_round_trips_with_flag(tmp_path):
+    payload = b"external-uri-bytes"
+    blob_path = tmp_path / "payload.bin"
+    blob_path.write_bytes(payload)
+
+    table, rows = _external_uri_table_and_rows("external_struct", [blob_path.as_uri()])
+    table.add(rows, allow_external_blob_outside_bases=True)
+
+    hits = table.search().to_arrow()
+    blobs = table.fetch_blobs("image", hits)
+    assert blobs[0].as_py() == payload
+
+
+def test_add_external_uri_without_flag_raises(tmp_path):
+    blob_path = tmp_path / "payload.bin"
+    blob_path.write_bytes(b"unreachable")
+
+    table, rows = _external_uri_table_and_rows("external_no_flag", [blob_path.as_uri()])
+    with pytest.raises(ValueError, match="allow_external_blob_outside_bases"):
+        table.add(rows)
+    assert table.count_rows() == 0
+
+
+def test_add_external_uri_string_round_trips_with_flag(tmp_path):
+    payload = b"external-uri-bytes"
+    blob_path = tmp_path / "payload.bin"
+    blob_path.write_bytes(payload)
+
+    db = lancedb.connect("memory:///")
+    schema = pa.schema([pa.field("id", pa.int64()), lancedb.blob("image")])
+    table = db.create_table("external_string", schema=schema)
+    table.add(
+        [{"id": 1, "image": blob_path.as_uri()}],
+        allow_external_blob_outside_bases=True,
+    )
+
+    hits = table.search().to_arrow()
+    blobs = table.fetch_blobs("image", hits)
+    assert blobs[0].as_py() == payload

python/python/tests/test_expr.py

@@ -52,7 +52,7 @@ class TestExprConstruction:
     def test_func(self):
         e = func("lower", col("name"))
         assert isinstance(e, Expr)
-        assert e.to_sql() == "lower(name)"
+        assert e.to_sql() == "lower(`name`)"
 
     def test_func_unknown_raises(self):
         with pytest.raises(Exception):
@@ -115,7 +115,7 @@ class TestExprOperators:
     def test_and_operator(self):
         e = (col("age") > lit(18)) & (col("status") == lit("active"))
         assert isinstance(e, Expr)
-        assert e.to_sql() == "((age > 18) AND (status = 'active'))"
+        assert e.to_sql() == "((age > 18) AND (`status` = 'active'))"
 
     def test_or_operator(self):
         e = (col("a") == lit(1)) | (col("b") == lit(2))
@@ -166,7 +166,7 @@ class TestExprOperators:
     def test_coerce_plain_str(self):
         e = col("name") == "alice"
         assert isinstance(e, Expr)
-        assert e.to_sql() == "(name = 'alice')"
+        assert e.to_sql() == "(`name` = 'alice')"
 
     def test_reflexive_comparisons(self):
         # 10 < col("age") swaps to col("age") > 10
@@ -198,85 +198,85 @@ class TestExprBytesLiteral:
 
     def test_bytes_equality_expr_sql(self):
         e = col("data") == lit(b"\xca\xfe")
-        assert e.to_sql() == "(data = X'CAFE')"
+        assert e.to_sql() == "(`data` = X'CAFE')"
 
     def test_bytes_ne_expr_sql(self):
         e = col("data") != lit(b"\xff")
-        assert e.to_sql() == "(data <> X'FF')"
+        assert e.to_sql() == "(`data` <> X'FF')"
 
     def test_bytes_compound_expr_sql(self):
         e = (col("data") == lit(b"\x01")) & (col("id") > lit(5))
-        assert e.to_sql() == "((data = X'01') AND (id > 5))"
+        assert e.to_sql() == "((`data` = X'01') AND (id > 5))"
 
     def test_bytes_in_function_call(self):
         # Regression test: binary literals inside scalar function calls
         # used to fail because DataFusion's unparser does not support Binary
         # scalars.  Now handled via a placeholder-substitution rewrite.
         e = func("contains", col("data"), lit(b"\xff"))
-        assert e.to_sql() == "contains(data, X'FF')"
+        assert e.to_sql() == "contains(`data`, X'FF')"
 
     def test_bytes_in_not(self):
         e = ~(col("data") == lit(b"\xff"))
-        assert e.to_sql() == "NOT (data = X'FF')"
+        assert e.to_sql() == "NOT (`data` = X'FF')"
 
 
 class TestExprStringMethods:
     def test_lower(self):
         e = col("name").lower()
         assert isinstance(e, Expr)
-        assert e.to_sql() == "lower(name)"
+        assert e.to_sql() == "lower(`name`)"
 
     def test_upper(self):
         e = col("name").upper()
         assert isinstance(e, Expr)
-        assert e.to_sql() == "upper(name)"
+        assert e.to_sql() == "upper(`name`)"
 
     def test_contains(self):
         e = col("text").contains(lit("hello"))
         assert isinstance(e, Expr)
-        assert e.to_sql() == "contains(text, 'hello')"
+        assert e.to_sql() == "contains(`text`, 'hello')"
 
     def test_contains_with_str_coerce(self):
         e = col("text").contains("hello")
         assert isinstance(e, Expr)
-        assert e.to_sql() == "contains(text, 'hello')"
+        assert e.to_sql() == "contains(`text`, 'hello')"
 
     def test_chained_lower_eq(self):
         e = col("name").lower() == lit("alice")
         assert isinstance(e, Expr)
-        assert e.to_sql() == "(lower(name) = 'alice')"
+        assert e.to_sql() == "(lower(`name`) = 'alice')"
 
 
 class TestExprCast:
     def test_cast_string(self):
         e = col("id").cast("string")
         assert isinstance(e, Expr)
-        assert e.to_sql() == "CAST(id AS VARCHAR)"
+        assert e.to_sql() == "arrow_cast(id, 'Utf8')"
 
     def test_cast_int32(self):
         e = col("score").cast("int32")
         assert isinstance(e, Expr)
-        assert e.to_sql() == "CAST(score AS INTEGER)"
+        assert e.to_sql() == "arrow_cast(score, 'Int32')"
 
     def test_cast_float64(self):
         e = col("val").cast("float64")
         assert isinstance(e, Expr)
-        assert e.to_sql() == "CAST(val AS DOUBLE)"
+        assert e.to_sql() == "arrow_cast(val, 'Float64')"
 
     def test_cast_pyarrow_type(self):
         e = col("score").cast(pa.int32())
         assert isinstance(e, Expr)
-        assert e.to_sql() == "CAST(score AS INTEGER)"
+        assert e.to_sql() == "arrow_cast(score, 'Int32')"
 
     def test_cast_pyarrow_float64(self):
         e = col("val").cast(pa.float64())
         assert isinstance(e, Expr)
-        assert e.to_sql() == "CAST(val AS DOUBLE)"
+        assert e.to_sql() == "arrow_cast(val, 'Float64')"
 
     def test_cast_pyarrow_string(self):
         e = col("id").cast(pa.string())
         assert isinstance(e, Expr)
-        assert e.to_sql() == "CAST(id AS VARCHAR)"
+        assert e.to_sql() == "arrow_cast(id, 'Utf8')"
 
     def test_cast_pyarrow_and_string_equivalent(self):
         # pa.int32() and "int32" should produce equivalent SQL
@@ -597,14 +597,14 @@ class TestExprIsin:
     def test_isin_strs(self):
         assert (
             col("status").isin(["active", "pending"]).to_sql()
-            == "status IN ('active', 'pending')"
+            == "`status` IN ('active', 'pending')"
         )
 
     def test_isin_coerces_and_mixes(self):
         assert col("id").isin([lit(1), 2]).to_sql() == "id IN (1, 2)"
 
     def test_isin_empty(self):
-        assert col("id").isin([]).to_sql() == "id IN ()"
+        assert col("id").isin([]).to_sql() == "false"
 
     def test_isin_filter(self, simple_table):
         result = simple_table.search().where(col("id").isin([1, 3, 5])).to_arrow()

python/python/tests/test_first_class_function_slice2.py

@@ -69,6 +69,26 @@ def _run_packaged(definition, *args):
     return namespace[definition.registration_request.artifact.entrypoint](*args)
 
 
+def test_udf_conda_environment():
+    @udf(conda=["scipy", "numpy"], conda_channels=["conda-forge", "defaults"])
+    def halve(value: float) -> float:
+        return value / 2
+
+    request = json.loads(halve.registration_request.to_canonical_json())
+    assert request["runtime"]["environment"] == {
+        "kind": "conda",
+        "packages": ["numpy", "scipy"],
+        "channels": ["conda-forge", "defaults"],
+    }
+    pip_request = json.loads(normalize_score.registration_request.to_canonical_json())
+    assert "channels" not in pip_request["runtime"]["environment"]
+
+    with pytest.raises(ValueError, match="not both"):
+        udf(name="both", pip=["numpy"], conda=["numpy"])(lambda value: value)
+    with pytest.raises(ValueError, match="requires conda"):
+        udf(name="channels", conda_channels=["conda-forge"])(lambda value: value)
+
+
 def test_udf_packages_attribute_access_and_body_imports():
     @udf
     def word_norm(body: str) -> float:

python/python/tests/test_fts.py

@@ -25,6 +25,7 @@ from lancedb.db import DBConnection
 from lancedb.index import FTS
 from lancedb.query import (
     BoostQuery,
+    DocumentGranularity,
     MatchQuery,
     MultiMatchQuery,
     PhraseQuery,
@@ -245,6 +246,55 @@ def test_create_inverted_index_rejects_invalid_block_size(table):
         table.create_index("text", config=FTS(block_size=129))
 
 
+def test_list_element_document_granularity(tmp_path):
+    docs_type = pa.list_(pa.struct([pa.field("content", pa.string())]))
+    docs = pa.array(
+        [
+            [
+                {"content": "alpha beta"},
+                None,
+                {"content": ""},
+                {"content": "the and"},
+                {"content": "alpha beta"},
+            ]
+        ],
+        type=docs_type,
+    )
+    table = ldb.connect(tmp_path).create_table(
+        "list_element_docs", pa.table({"id": [0], "docs": docs})
+    )
+    row_table = ldb.connect(tmp_path).create_table(
+        "row_docs", pa.table({"id": [0], "docs": docs})
+    )
+    row_table.create_index("docs.content", config=FTS())
+    row_result = row_table.search(MatchQuery("alpha", "docs.content")).to_arrow()
+    assert row_result.num_rows == 1
+    assert "_doc_index" not in row_result.column_names
+
+    granularity = DocumentGranularity.LIST_ELEMENT
+    table.create_index(
+        "docs.content",
+        config=FTS(with_position=True, document_granularity=granularity),
+    )
+    assert table.list_indices()[0].columns == ["docs.content"]
+
+    def coordinates(query):
+        result = table.search(query).limit(10).to_arrow()
+        doc_index_type = result.schema.field("_doc_index").type
+        assert pa.types.is_list(doc_index_type)
+        assert doc_index_type.value_type == pa.uint32()
+        return sorted(result["_doc_index"].to_pylist())
+
+    assert coordinates(
+        MatchQuery("alpha", "docs.content", document_granularity=granularity)
+    ) == [[0], [4]]
+    assert coordinates(
+        PhraseQuery("alpha beta", "docs.content", document_granularity=granularity)
+    ) == [[0], [4]]
+    assert coordinates(MatchQuery("alpha", "docs.content")) == [[0], [4]]
+    assert FTS().document_granularity is DocumentGranularity.ROW
+
+
 def test_create_inverted_index_respects_build_memory_limit(table):
     with pytest.raises(ValueError, match="exceeds worker memory limit"):
         table.create_index(
@@ -1089,6 +1139,20 @@ def test_fts_query_to_json():
     )
     assert json_str == expected
 
+    # Test MatchQuery with list-element document granularity
+    match_query = MatchQuery(
+        "hello world",
+        "text",
+        document_granularity=DocumentGranularity.LIST_ELEMENT,
+    )
+    json_str = match_query.to_json()
+    expected = (
+        '{"match":{"column":"text","terms":"hello world","boost":1.0,'
+        '"fuzziness":0,"max_expansions":50,"operator":"Or","prefix_length":0,'
+        '"document_granularity":"list_element"}}'
+    )
+    assert json_str == expected
+
     # Test MatchQuery with options
     match_query = MatchQuery("puppy", "text", fuzziness=2, boost=1.5, prefix_length=3)
     json_str = match_query.to_json()
@@ -1098,6 +1162,19 @@ def test_fts_query_to_json():
     )
     assert json_str == expected
 
+    # Test PhraseQuery with list-element document granularity
+    phrase_query = PhraseQuery(
+        "quick brown fox",
+        "title",
+        document_granularity=DocumentGranularity.LIST_ELEMENT,
+    )
+    json_str = phrase_query.to_json()
+    expected = (
+        '{"phrase":{"column":"title","terms":"quick brown fox","slop":0,'
+        '"document_granularity":"list_element"}}'
+    )
+    assert json_str == expected
+
     # Test PhraseQuery
     phrase_query = PhraseQuery("quick brown fox", "title")
     json_str = phrase_query.to_json()

python/python/tests/test_query.py

@@ -675,6 +675,21 @@ def test_distance_range(table: lancedb.table.Table):
     assert res["_distance"].to_pylist() == [min_dist, max_dist]
 
 
+@pytest.mark.parametrize("expression", ["1 - _distance", "1.0 - _distance"])
+def test_select_arithmetic_with_distance(table, expression):
+    result = (
+        table.search([10, 10])
+        .select({"similarity": expression, "_distance": "_distance"})
+        .distance_type("cosine")
+        .to_arrow()
+    )
+
+    assert result.schema.field("similarity").type == pa.float32()
+    assert result["similarity"].to_pylist() == pytest.approx(
+        [1 - distance for distance in result["_distance"].to_pylist()]
+    )
+
+
 @pytest.mark.asyncio
 async def test_distance_range_async(table_async: AsyncTable):
     q = [0, 0]
@@ -897,6 +912,23 @@ def test_query_builder_batches(table):
     assert rs_list["id"][1] == 2
 
 
+def test_batch_vector_query_shares_filtered_flat_scan(table):
+    query = (
+        table.search([[1.0, 2.0], [3.0, 4.0]])
+        .where("id > 0", prefilter=True)
+        .limit(1)
+        .select(["id"])
+    )
+
+    plan = query.explain_plan(verbose=True)
+    assert "KNNVectorDistance: queries=2" in plan
+    assert "UnionExec" not in plan
+
+    results = query.to_arrow()
+    assert len(results) == 2
+    assert results["query_index"].to_pylist() == [0, 1]
+
+
 def test_dynamic_projection(table):
     rs = (
         LanceVectorQueryBuilder(table, [0, 0], "vector")

python/python/tests/test_remote_db.py

@@ -1618,6 +1618,49 @@ def test_query_sync_fts():
         )
 
 
+def test_query_sync_fts_document_granularity():
+    from lancedb.query import DocumentGranularity, MatchQuery
+
+    def handler(body):
+        assert body == {
+            "full_text_query": {
+                "query": {
+                    "match": {
+                        "column": "docs.content",
+                        "terms": "alpha",
+                        "boost": 1.0,
+                        "fuzziness": 0,
+                        "max_expansions": 50,
+                        "operator": "Or",
+                        "prefix_length": 0,
+                        "document_granularity": "list_element",
+                    }
+                }
+            },
+            "k": 10,
+            "prefilter": True,
+            "vector": [],
+            "version": None,
+        }
+        return pa.table(
+            {
+                "id": [1, 1],
+                "_doc_index": pa.array([[0], [4]], type=pa.list_(pa.uint32())),
+            }
+        )
+
+    with query_test_table(handler, server_version=Version("0.6.0")) as table:
+        result = table.search(
+            MatchQuery(
+                "alpha",
+                "docs.content",
+                document_granularity=DocumentGranularity.LIST_ELEMENT,
+            )
+        ).to_arrow()
+
+    assert result["_doc_index"].to_pylist() == [[0], [4]]
+
+
 def test_query_sync_hybrid():
     def handler(body):
         if "full_text_query" in body:

python/python/tests/test_s3.py

@@ -4,6 +4,7 @@
 
 import asyncio
 import copy
+from concurrent.futures import ThreadPoolExecutor
 from datetime import timedelta
 import threading
 
@@ -86,6 +87,25 @@ def test_s3_lifecycle(s3_bucket: str):
     asyncio.run(test())
 
 
+@pytest.mark.s3_test
+def test_concurrent_open_table(s3_bucket: str):
+    uri = f"s3://{s3_bucket}/test_concurrent_open_table"
+    db = lancedb.connect(uri, storage_options=copy.copy(CONFIG))
+    db.create_table("test", pa.table({"x": [1, 2, 3]}))
+
+    num_workers = 32
+    barrier = threading.Barrier(num_workers)
+
+    def open_and_count(_):
+        barrier.wait()
+        return db.open_table("test").count_rows()
+
+    with ThreadPoolExecutor(max_workers=num_workers) as pool:
+        row_counts = list(pool.map(open_and_count, range(num_workers)))
+
+    assert row_counts == [3] * num_workers
+
+
 @pytest.fixture()
 def kms_key():
     kms = get_boto3_client("kms", endpoint_url=CONFIG["aws_endpoint"])

python/python/tests/test_table.py

@@ -11,6 +11,7 @@ import warnings
 import weakref
 from concurrent.futures import ThreadPoolExecutor
 from datetime import date, datetime, timedelta
+from decimal import Decimal
 from time import sleep
 from typing import List
 from unittest.mock import patch
@@ -336,6 +337,21 @@ async def test_update_async(mem_db_async: AsyncConnection):
     assert await table.count_rows("id == 10") == 1
 
 
+@pytest.mark.asyncio
+async def test_update_expr_filter_literals_async(mem_db_async: AsyncConnection):
+    values = ["5", "4.66e-84", "it's"]
+    table = await mem_db_async.create_table(
+        "update_expr_literals",
+        data=[{"field": value, "result": "original"} for value in values],
+    )
+
+    for value in values:
+        update_res = await table.update({"result": value}, where=col("field") == value)
+        assert update_res.rows_updated == 1
+
+    assert (await table.to_arrow())["result"].to_pylist() == values
+
+
 def test_create_table(mem_db: DBConnection):
     schema = pa.schema(
         {
@@ -2343,6 +2359,148 @@ def test_update(mem_db: DBConnection):
     assert np.allclose(v, np.array([[1.2, 1.9], [1.1, 1.1]]))
 
 
+def test_update_expr_filter_literals(mem_db: DBConnection):
+    values = ["5", "4.66e-84", "it's"]
+    table = mem_db.create_table(
+        "update_expr_literals",
+        data=[{"field": value, "result": "original"} for value in values],
+    )
+
+    for value in values:
+        update_res = table.update(where=col("field") == value, values={"result": value})
+        assert update_res.rows_updated == 1
+
+    assert table.to_arrow()["result"].to_pylist() == values
+
+
+def test_update_expr_filter_preserves_typed_semantics(mem_db: DBConnection):
+    low = Decimal("1.234567890123456789")
+    high = Decimal("1.234567890123456790")
+    decimal_schema = pa.schema(
+        [("val", pa.decimal128(19, 18)), ("result", pa.string())]
+    )
+    decimal_table = mem_db.create_table(
+        "update_expr_decimal",
+        pa.table(
+            {"val": [low, high], "result": ["old", "old"]},
+            schema=decimal_schema,
+        ),
+    )
+    predicate = col("val") < lit(high)
+    assert decimal_table.search().where(predicate).to_arrow().num_rows == 1
+    result = decimal_table.update(where=predicate, values={"result": "new"})
+    assert result.rows_updated == 1
+
+    keyword_table = mem_db.create_table(
+        "update_expr_keyword", [{"null": 1, "result": "old"}]
+    )
+    predicate = col("null") == 1
+    assert keyword_table.search().where(predicate).to_arrow().num_rows == 1
+    result = keyword_table.update(where=predicate, values={"result": "new"})
+    assert result.rows_updated == 1
+
+    empty_in_table = mem_db.create_table(
+        "update_expr_empty_in", [{"id": 1, "result": "old"}]
+    )
+    predicate = col("id").isin([])
+    assert empty_in_table.search().where(predicate).to_arrow().num_rows == 0
+    result = empty_in_table.update(where=predicate, values={"result": "new"})
+    assert result.rows_updated == 0
+
+    marker = "__lancedb_binary_placeholder_0__"
+    binary_schema = pa.schema(
+        [("payload", pa.binary()), ("text", pa.string()), ("result", pa.string())]
+    )
+    binary_table = mem_db.create_table(
+        "update_expr_binary",
+        pa.table(
+            {
+                "payload": [b"\x01", b"\x02"],
+                "text": ["other", marker],
+                "result": ["old", "old"],
+            },
+            schema=binary_schema,
+        ),
+    )
+    predicate = (col("payload") == lit(b"\x01")) | (col("text") == marker)
+    assert binary_table.search().where(predicate).to_arrow().num_rows == 2
+    result = binary_table.update(where=predicate, values={"result": "new"})
+    assert result.rows_updated == 2
+
+    nonfinite_table = mem_db.create_table(
+        "update_expr_nonfinite",
+        [{"x": 1.0, "result": "old"}, {"x": 2.0, "result": "old"}],
+    )
+    predicate = col("x") < float("inf")
+    assert nonfinite_table.search().where(predicate).to_arrow().num_rows == 2
+    result = nonfinite_table.update(where=predicate, values={"result": "new"})
+    assert result.rows_updated == 2
+
+    float16_table = mem_db.create_table(
+        "update_expr_float16",
+        [{"x": 1.0, "result": "old"}, {"x": 3.0, "result": "old"}],
+    )
+    predicate = col("x").cast(pa.float16()) < 2.0
+    assert float16_table.search().where(predicate).to_arrow().num_rows == 1
+    result = float16_table.update(where=predicate, values={"result": "new"})
+    assert result.rows_updated == 1
+
+    string_cast_table = mem_db.create_table(
+        "update_expr_string_cast",
+        [{"x": 1, "result": "old"}, {"x": 2, "result": "old"}],
+    )
+    predicate = col("x").cast("string") == "1"
+    assert string_cast_table.search().where(predicate).to_arrow().num_rows == 1
+    result = string_cast_table.update(where=predicate, values={"result": "new"})
+    assert result.rows_updated == 1
+
+    quoted_identifier_schema = pa.schema(
+        [("payload", pa.binary()), ("odd'name", pa.int64()), ("result", pa.string())]
+    )
+    quoted_identifier_table = mem_db.create_table(
+        "update_expr_quoted_identifier",
+        pa.table(
+            {"payload": [b"\x01"], "odd'name": [1], "result": ["old"]},
+            schema=quoted_identifier_schema,
+        ),
+    )
+    predicate = (col("payload") == lit(b"\x01")) & (col("odd'name") == 1)
+    assert quoted_identifier_table.search().where(predicate).to_arrow().num_rows == 1
+    result = quoted_identifier_table.update(where=predicate, values={"result": "new"})
+    assert result.rows_updated == 1
+
+    decimal256_schema = pa.schema(
+        [("val", pa.decimal256(40, 2)), ("result", pa.string())]
+    )
+    decimal256_table = mem_db.create_table(
+        "update_expr_decimal256",
+        pa.table(
+            {
+                "val": [Decimal("1.00"), Decimal("3.00")],
+                "result": ["old", "old"],
+            },
+            schema=decimal256_schema,
+        ),
+    )
+    predicate = col("val") < lit(Decimal("2.00")).cast(pa.decimal256(40, 2))
+    assert decimal256_table.search().where(predicate).to_arrow().num_rows == 1
+    result = decimal256_table.update(where=predicate, values={"result": "new"})
+    assert result.rows_updated == 1
+
+    binary_empty_table = mem_db.create_table(
+        "update_expr_binary_empty",
+        pa.table(
+            {"payload": [b"\x01", b"\x02"], "result": ["old", "old"]},
+            schema=pa.schema([("payload", pa.binary()), ("result", pa.string())]),
+        ),
+    )
+    predicate = (col("payload") == lit(b"\x01")).isin([])
+    assert binary_empty_table.search().where(predicate).to_arrow().num_rows == 0
+    assert predicate.to_sql() == "false"
+    result = binary_empty_table.update(where=predicate, values={"result": "new"})
+    assert result.rows_updated == 0
+
+
 def test_update_with_arrow_scalar(mem_db: DBConnection):
     schema = pa.schema({"id": pa.int64(), "vector": pa.list_(pa.float32(), 4)})
     table = mem_db.create_table("my_table", schema=schema)
@@ -3929,6 +4087,29 @@ def test_computed_column_rejects_transforms_and_computed_together(tmp_path):
         table.add_columns({"a": "x + 1"}, computed={"b": "x * 2"})
 
 
+def test_computed_column_blob_projection_inherits_semantics(tmp_path):
+    schema = pa.schema([pa.field("id", pa.int64()), lancedb.blob("image")])
+    db = lancedb.connect(tmp_path)
+    table = db.create_table("computed_column_blob", schema=schema)
+    table.add(
+        [
+            {"id": 1, "image": b"hello"},
+            {"id": 2, "image": b""},
+            {"id": 3, "image": None},
+        ]
+    )
+
+    table.add_columns(computed={"image_copy": "image", "second_copy": "image_copy"})
+    assert table.refresh_column("image_copy").rows_filled == 2
+    assert table.refresh_column("second_copy").rows_filled == 2
+    assert table.blob_columns() == ["image", "image_copy", "second_copy"]
+
+    hits = table.search().with_row_id(True).limit(10).to_arrow()
+    rows = sorted(zip(hits["id"].to_pylist(), hits["_rowid"].to_pylist()))
+    copied = table.fetch_blobs("second_copy", [row_id for _, row_id in rows])
+    assert copied.to_pylist() == [b"hello", b"", None]
+
+
 @pytest.mark.asyncio
 async def test_computed_column_async(tmp_path):
     db = await lancedb.connect_async(tmp_path)

python/python/tests/test_util.py

@@ -7,6 +7,7 @@ import pathlib
 from typing import Optional
 
 import lance
+from lance.blob import BlobType as LanceBlobType
 from lancedb.conftest import MockTextEmbeddingFunction
 from lancedb.embeddings.base import EmbeddingFunctionConfig
 from lancedb.embeddings.registry import EmbeddingFunctionRegistry
@@ -907,6 +908,165 @@ def test_cast_to_target_schema():
     assert output == expected
 
 
+def test_cast_to_target_schema_coerces_binary_to_blob_v2():
+    data = pa.table({"image": pa.array([b"hello", None], type=pa.binary())})
+    target = pa.schema([lancedb.blob("image")])
+
+    output = _cast_to_target_schema(data.to_reader(), target).read_all()
+
+    image = output["image"].chunk(0)
+    assert type(image.type) is lancedb.BlobType
+    assert image.storage.to_pylist() == [
+        {"data": b"hello", "uri": None, "position": None, "size": None},
+        None,
+    ]
+
+
+def test_cast_to_target_schema_coerces_binary_to_metadata_blob_struct():
+    storage = lancedb.blob("image").type.storage_type
+    target = pa.schema(
+        [
+            pa.field(
+                "image",
+                storage,
+                metadata={
+                    b"ARROW:extension:name": b"lance.blob.v2",
+                    b"ARROW:extension:metadata": b"",
+                },
+            )
+        ]
+    )
+    data = pa.table({"image": pa.array([b"hello", None], type=pa.binary())})
+
+    output = _cast_to_target_schema(data.to_reader(), target).read_all()
+
+    image = output["image"].chunk(0)
+    assert not isinstance(image.type, pa.ExtensionType)
+    assert image.to_pylist() == [
+        {"data": b"hello", "uri": None, "position": None, "size": None},
+        None,
+    ]
+
+
+def test_cast_to_target_schema_coerces_nested_binary_blob():
+    data = pa.table(
+        {
+            "info": pa.array(
+                [{"blob": b"hello"}, {"blob": None}],
+                type=pa.struct([pa.field("blob", pa.binary())]),
+            )
+        }
+    )
+    target = pa.schema([pa.field("info", pa.struct([lancedb.blob("blob")]))])
+
+    output = _cast_to_target_schema(data.to_reader(), target).read_all()
+
+    blob = output["info"].chunk(0).field("blob")
+    assert type(blob.type) is lancedb.BlobType
+    assert blob.storage.to_pylist() == [
+        {"data": b"hello", "uri": None, "position": None, "size": None},
+        None,
+    ]
+
+
+def test_cast_to_target_schema_coerces_list_binary_blob_with_inferred_child_name():
+    data = pa.table(
+        {"images": pa.array([[b"a", b"b"], None], type=pa.list_(pa.binary()))}
+    )
+    target = pa.schema([pa.field("images", pa.list_(lancedb.blob("image")))])
+
+    output = _cast_to_target_schema(data.to_reader(), target).read_all()
+
+    images = output["images"].chunk(0)
+    assert images.type.value_field.name == "image"
+    assert type(images.type.value_type) is lancedb.BlobType
+    assert images.to_pylist()[1] is None
+    assert images.values.storage.to_pylist() == [
+        {"data": b"a", "uri": None, "position": None, "size": None},
+        {"data": b"b", "uri": None, "position": None, "size": None},
+    ]
+
+
+def test_list_blob_coercion_preserves_null_slots_with_nonzero_extent():
+    child = pa.field("image", pa.binary())
+    source = pa.ListArray.from_arrays(
+        pa.array([0, 2, 4], type=pa.int32()),
+        pa.array([b"a", b"b", b"dead", b"beef"], type=pa.binary()),
+        mask=pa.array([False, True]),
+    ).cast(pa.list_(child))
+    target = pa.schema([pa.field("images", pa.list_(lancedb.blob("image")))])
+
+    output = _cast_to_target_schema(
+        pa.table({"images": source}).to_reader(), target
+    ).read_all()
+
+    images = output["images"].chunk(0)
+    assert images.to_pylist()[1] is None
+    assert [b["data"] for b in images.to_pylist()[0]] == [b"a", b"b"]
+
+
+def test_fixed_size_list_blob_coercion_keeps_null_rows():
+    child = pa.field("frame", pa.binary())
+    source = (
+        pa.FixedSizeListArray.from_arrays(
+            pa.array([b"a", b"b", b"c", b"d"], type=pa.binary()), 2
+        )
+        .take(pa.array([0, None], type=pa.int32()))
+        .cast(pa.list_(child, 2))
+    )
+    target = pa.schema([pa.field("frames", pa.list_(lancedb.blob("frame"), 2))])
+
+    output = _cast_to_target_schema(
+        pa.table({"frames": source}).to_reader(), target
+    ).read_all()
+
+    frames = output["frames"].chunk(0)
+    assert frames.to_pylist()[1] is None
+    assert [b["data"] for b in frames.to_pylist()[0]] == [b"a", b"b"]
+
+
+def test_cast_to_target_schema_accepts_pylance_blob_v2():
+    target_type = lancedb.BlobType()
+    source = lance.blob_array([b"hello", None])
+    assert type(source.type) is LanceBlobType
+    assert type(source.type) is type(target_type)
+    data = pa.table({"image": source})
+    target = pa.schema([pa.field("image", target_type)])
+
+    output = _cast_to_target_schema(data.to_reader(), target).read_all()
+
+    image = output["image"].chunk(0)
+    assert type(image.type) is LanceBlobType
+    assert image.type == target_type
+    assert image.storage.to_pylist() == [
+        {"data": b"hello", "uri": None, "position": None, "size": None},
+        None,
+    ]
+
+
+def test_cast_to_target_schema_rejects_different_blob_v2_class():
+    class OtherBlobType(pa.ExtensionType):
+        def __init__(self):
+            super().__init__(lancedb.BlobType().storage_type, "lance.blob.v2")
+
+        def __arrow_ext_serialize__(self) -> bytes:
+            return b""
+
+        @classmethod
+        def __arrow_ext_deserialize__(
+            cls, storage_type: pa.DataType, serialized: bytes
+        ) -> "OtherBlobType":
+            return cls()
+
+    storage = lance.blob_array([b"hello"]).storage
+    source = pa.ExtensionArray.from_storage(OtherBlobType(), storage)
+    data = pa.table({"image": source})
+    target = pa.schema([lancedb.blob("image")])
+
+    with pytest.raises(pa.ArrowTypeError, match="different extension type"):
+        _cast_to_target_schema(data.to_reader(), target).read_all()
+
+
 def test_sanitize_data_stream():
     # Make sure we don't collect the whole stream when running sanitize_data
     schema = pa.schema({"a": pa.int32()})

python/src/expr.rs

@@ -130,6 +130,14 @@ impl PyExpr {
 
     // ── utilities ────────────────────────────────────────────────────────────
 
+    /// Return the referenced column name for a bare column expression.
+    fn column_name(&self) -> Option<String> {
+        match &self.0 {
+            DfExpr::Column(column) if column.relation.is_none() => Some(column.name.clone()),
+            _ => None,
+        }
+    }
+
     /// Render the expression as a SQL string (useful for debugging).
     fn to_sql(&self) -> PyResult<String> {
         lancedb::expr::expr_to_sql_string(&self.0).map_err(|e| PyValueError::new_err(e.to_string()))

python/src/index.rs

@@ -8,7 +8,7 @@ use lancedb::index::vector::{
 };
 use lancedb::index::{
     Index as LanceDbIndex,
-    scalar::{BTreeIndexBuilder, FmIndexBuilder, FtsIndexBuilder},
+    scalar::{BTreeIndexBuilder, DocumentGranularity, FmIndexBuilder, FtsIndexBuilder},
 };
 use pyo3::IntoPyObject;
 use pyo3::types::PyStringMethods;
@@ -60,7 +60,11 @@ pub fn extract_index_params(source: &Option<Bound<'_, PyAny>>) -> PyResult<Lance
                     .ngram_min_length(params.ngram_min_length)
                     .ngram_max_length(params.ngram_max_length)
                     .ngram_prefix_only(params.prefix_only)
-                    .custom_stop_words(params.custom_stop_words);
+                    .custom_stop_words(params.custom_stop_words)
+                    .document_granularity(
+                        DocumentGranularity::try_from(params.document_granularity.as_str())
+                            .map_err(|err| PyValueError::new_err(err.to_string()))?,
+                    );
                 if let Some(memory_limit) = params.memory_limit {
                     inner_opts = inner_opts.memory_limit_mb(memory_limit);
                 }
@@ -221,6 +225,7 @@ struct FtsParams {
     block_size: usize,
     memory_limit: Option<u64>,
     num_workers: Option<usize>,
+    document_granularity: String,
 }
 
 #[derive(FromPyObject)]
@@ -481,6 +486,7 @@ mod tests {
     block_size = 128
     memory_limit = 2048
     num_workers = 7
+    document_granularity = 'row'
 
 config = FTS()",
                 None,

python/src/query.rs

@@ -1,6 +1,7 @@
 // SPDX-License-Identifier: Apache-2.0
 // SPDX-FileCopyrightText: Copyright The LanceDB Authors
 
+use std::collections::HashMap;
 use std::sync::Arc;
 use std::time::Duration;
 
@@ -16,8 +17,8 @@ use arrow::pyarrow::FromPyArrow;
 use arrow::pyarrow::IntoPyArrow;
 use arrow::pyarrow::ToPyArrow;
 use lancedb::index::scalar::{
-    BooleanQuery, BoostQuery, FtsQuery, FullTextSearchQuery, MatchQuery, MultiMatchQuery, Occur,
-    Operator, PhraseQuery,
+    BooleanQuery, BoostQuery, DocumentGranularity, FtsQuery, FullTextSearchQuery, MatchQuery,
+    MultiMatchQuery, Occur, Operator, PhraseQuery,
 };
 use lancedb::query::AnalyzePlanDistributedMetrics;
 use lancedb::query::QueryBase;
@@ -76,8 +77,16 @@ impl<'a, 'py> FromPyObject<'a, 'py> for PyLanceDB<FtsQuery> {
                 let max_expansions = ob.getattr("max_expansions")?.extract()?;
                 let operator = ob.getattr("operator")?.extract::<String>()?;
                 let prefix_length = ob.getattr("prefix_length")?.extract()?;
+                let document_granularity = ob
+                    .getattr("document_granularity")?
+                    .extract::<Option<String>>()?
+                    .map(|value| {
+                        DocumentGranularity::try_from(value.as_str())
+                            .map_err(|err| PyValueError::new_err(err.to_string()))
+                    })
+                    .transpose()?;
 
-                Ok(Self(
+                let mut query =
                     MatchQuery::new(query)
                         .with_column(Some(column))
                         .with_boost(boost)
@@ -86,21 +95,32 @@ impl<'a, 'py> FromPyObject<'a, 'py> for PyLanceDB<FtsQuery> {
                         .with_operator(Operator::try_from(operator.as_str()).map_err(|e| {
                             PyValueError::new_err(format!("Invalid operator: {}", e))
                         })?)
-                        .with_prefix_length(prefix_length)
-                        .into(),
-                ))
+                        .with_prefix_length(prefix_length);
+                if let Some(document_granularity) = document_granularity {
+                    query = query.with_document_granularity(document_granularity);
+                }
+                Ok(Self(query.into()))
             }
             "PhraseQuery" => {
                 let query = ob.getattr("query")?.extract()?;
                 let column = ob.getattr("column")?.extract()?;
                 let slop = ob.getattr("slop")?.extract()?;
+                let document_granularity = ob
+                    .getattr("document_granularity")?
+                    .extract::<Option<String>>()?
+                    .map(|value| {
+                        DocumentGranularity::try_from(value.as_str())
+                            .map_err(|err| PyValueError::new_err(err.to_string()))
+                    })
+                    .transpose()?;
 
-                Ok(Self(
-                    PhraseQuery::new(query)
-                        .with_column(Some(column))
-                        .with_slop(slop)
-                        .into(),
-                ))
+                let mut query = PhraseQuery::new(query)
+                    .with_column(Some(column))
+                    .with_slop(slop);
+                if let Some(document_granularity) = document_granularity {
+                    query = query.with_document_granularity(document_granularity);
+                }
+                Ok(Self(query.into()))
             }
             "BoostQuery" => {
                 let positive: Self = ob.getattr("positive")?.extract()?;
@@ -167,6 +187,13 @@ impl<'py> IntoPyObject<'py> for PyLanceDB<FtsQuery> {
                 kwargs.set_item("max_expansions", query.max_expansions)?;
                 kwargs.set_item::<_, &str>("operator", query.operator.into())?;
                 kwargs.set_item("prefix_length", query.prefix_length)?;
+                if let Some(document_granularity) = query.document_granularity {
+                    let value = match document_granularity {
+                        DocumentGranularity::Row => "row",
+                        DocumentGranularity::ListElement => "list_element",
+                    };
+                    kwargs.set_item("document_granularity", value)?;
+                }
                 namespace
                     .getattr(intern!(py, "MatchQuery"))?
                     .call((query.terms, query.column.unwrap()), Some(&kwargs))
@@ -174,6 +201,13 @@ impl<'py> IntoPyObject<'py> for PyLanceDB<FtsQuery> {
             FtsQuery::Phrase(query) => {
                 let kwargs = PyDict::new(py);
                 kwargs.set_item("slop", query.slop)?;
+                if let Some(document_granularity) = query.document_granularity {
+                    let value = match document_granularity {
+                        DocumentGranularity::Row => "row",
+                        DocumentGranularity::ListElement => "list_element",
+                    };
+                    kwargs.set_item("document_granularity", value)?;
+                }
                 namespace
                     .getattr(intern!(py, "PhraseQuery"))?
                     .call((query.terms, query.column.unwrap()), Some(&kwargs))
@@ -292,6 +326,7 @@ pub struct PyQueryRequest {
     pub filter: Option<PyQueryFilter>,
     pub full_text_search: Option<PyLanceDB<FtsQuery>>,
     pub select: PySelect,
+    pub select_source_columns: Option<HashMap<String, String>>,
     pub fast_search: Option<bool>,
     pub with_row_id: Option<bool>,
     pub use_lsm: Option<bool>,
@@ -322,6 +357,7 @@ impl From<AnyQuery> for PyQueryRequest {
                 full_text_search: query_request
                     .full_text_search
                     .map(|fts| PyLanceDB(fts.query)),
+                select_source_columns: PySelect::source_columns(&query_request.select),
                 select: PySelect(query_request.select),
                 fast_search: Some(query_request.fast_search),
                 with_row_id: Some(query_request.with_row_id),
@@ -347,6 +383,7 @@ impl From<AnyQuery> for PyQueryRequest {
                 offset: vector_query.base.offset,
                 filter: vector_query.base.filter.map(PyQueryFilter),
                 full_text_search: None,
+                select_source_columns: PySelect::source_columns(&vector_query.base.select),
                 select: PySelect(vector_query.base.select),
                 fast_search: Some(vector_query.base.fast_search),
                 with_row_id: Some(vector_query.base.with_row_id),
@@ -379,6 +416,25 @@ impl From<AnyQuery> for PyQueryRequest {
 #[derive(Clone)]
 pub struct PySelect(Select);
 
+impl PySelect {
+    fn source_columns(select: &Select) -> Option<HashMap<String, String>> {
+        match select {
+            Select::Expr(pairs) => Some(
+                pairs
+                    .iter()
+                    .filter_map(|(output, expr)| match expr {
+                        lancedb::expr::DfExpr::Column(column) if column.relation.is_none() => {
+                            Some((output.clone(), column.name.clone()))
+                        }
+                        _ => None,
+                    })
+                    .collect(),
+            ),
+            _ => None,
+        }
+    }
+}
+
 impl<'py> IntoPyObject<'py> for PySelect {
     type Target = PyAny;
     type Output = Bound<'py, Self::Target>;

python/src/table.rs

@@ -780,15 +780,19 @@ impl Table {
         })
     }
 
-    #[pyo3(signature = (data, mode, progress=None, write_parallelism=None))]
+    #[pyo3(signature = (data, mode, progress=None, write_parallelism=None, allow_external_blob_outside_bases=false))]
     pub fn add<'a>(
         self_: PyRef<'a, Self>,
         data: PyScannable,
         mode: String,
         progress: Option<Py<PyAny>>,
         write_parallelism: Option<usize>,
+        allow_external_blob_outside_bases: bool,
     ) -> PyResult<Bound<'a, PyAny>> {
-        let mut op = self_.inner_ref()?.add(data);
+        let mut op = self_
+            .inner_ref()?
+            .add(data)
+            .allow_external_blob_outside_bases(allow_external_blob_outside_bases);
         if mode == "append" {
             op = op.mode(AddDataMode::Append);
         } else if mode == "overwrite" {

rust/lancedb/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "lancedb"
-version = "0.38.0-beta.10"
+version = "0.38.0-beta.12"
 edition.workspace = true
 description = "LanceDB: A serverless, low-latency vector database for AI applications"
 license.workspace = true

rust/lancedb/src/database/listing.rs

@@ -13,7 +13,7 @@ use lance::dataset::{ReadParams, WriteMode, builder::DatasetBuilder};
 use lance::io::{ObjectStore, ObjectStoreParams, WrappingObjectStore};
 use lance_datafusion::utils::StreamingWriteSource;
 use lance_file::version::LanceFileVersion;
-use lance_io::object_store::{StorageOptionsAccessor, StorageOptionsProvider};
+use lance_io::object_store::{ReadDirOptions, StorageOptionsAccessor, StorageOptionsProvider};
 use lance_table::io::commit::commit_handler_from_url;
 use object_store::local::LocalFileSystem;
 use snafu::ResultExt;
@@ -281,6 +281,22 @@ impl std::fmt::Display for ListingDatabase {
 }
 
 const LANCE_EXTENSION: &str = "lance";
+
+/// The table a listed child of the database names, or `None` if the child is not a table.
+///
+/// A table is the directory `<name>.lance`; a loose file or any other directory under the
+/// database prefix belongs to something else. `dir_suffix` is `.lance`, built once by the
+/// caller rather than per child.
+/// The table a listed child directory holds, or `None` if it is not a table at all.
+///
+/// Only directories are considered, so a loose object named like a table is not one.
+fn table_name(location: &object_store::path::Path, dir_suffix: &str) -> Option<String> {
+    location
+        .filename()?
+        .strip_suffix(dir_suffix)
+        .map(String::from)
+        .filter(|name| !name.is_empty())
+}
 const ENGINE: &str = "engine";
 const MIRRORED_STORE: &str = "mirroredStore";
 
@@ -944,51 +960,72 @@ impl Database for ListingDatabase {
         Ok(f)
     }
 
+    /// List the tables in the database, a page at a time.
+    ///
+    /// The page_token is opaque, unlike the `start_after` parameter of [`Self::table_names()`].
+    ///
+    /// When there are no more results, the returned page_token will be None.
+    ///
+    /// `limit` is the maximum number of tables to return in the response. But it is possible
+    /// for the response to contain fewer than `limit` tables, even when there are more tables
+    /// to return. Clients should check the returned page_token to determine if there are
+    /// more results, rather than relying on the number of tables returned.
+    ///
+    /// The order that results are returned in not guaranteed to be stable across calls,
+    /// so clients should not rely on it.
     async fn list_tables(&self, request: ListTablesRequest) -> Result<ListTablesResponse> {
         if request.id.as_ref().map(|v| !v.is_empty()).unwrap_or(false) {
             return self.namespace_database().list_tables(request).await;
         }
-        let mut f = self
-            .object_store
-            .read_dir(self.base_path.clone())
-            .await?
-            .iter()
-            .map(Path::new)
-            .filter(|path| {
-                let is_lance = path
-                    .extension()
-                    .and_then(|e| e.to_str())
-                    .map(|e| e == LANCE_EXTENSION);
-                is_lance.unwrap_or(false)
-            })
-            .filter_map(|p| p.file_stem().and_then(|s| s.to_str().map(String::from)))
-            .collect::<Vec<String>>();
-        f.sort();
+        let limit = request.limit.map(|limit| limit.max(0) as usize);
+        let dir_suffix = format!(".{LANCE_EXTENSION}");
+        let mut tables = Vec::new();
+        let mut page_token = request.page_token.filter(|token| !token.is_empty());
 
-        // Handle pagination with page_token
-        if let Some(ref page_token) = request.page_token {
-            let index = f
-                .iter()
-                .position(|name| name.as_str() > page_token.as_str())
-                .unwrap_or(f.len());
-            f.drain(0..index);
+        // A page of nothing: the store rejects a limit of zero, and no table was handed over
+        // for a token to resume after.
+        if limit == Some(0) {
+            return Ok(ListTablesResponse {
+                context: None,
+                tables,
+                page_token: None,
+            });
         }
 
-        // Determine if there's a next page. The token is the last name of this page,
-        // not the first of the next one: the next page resumes strictly after the
-        // token, so naming the next page's first entry would skip it.
-        let next_page_token = match request.limit {
-            Some(limit) if f.len() > limit as usize => {
-                f.truncate(limit as usize);
-                f.last().cloned()
+        loop {
+            // Ask only for what the page still has room for, so a database holding more
+            // than one page costs one request per page rather than one per table.
+            let listing = self
+                .object_store
+                .read_dir_page(
+                    self.base_path.clone(),
+                    ReadDirOptions {
+                        page_token: page_token.take(),
+                        limit: limit.map(|limit| limit - tables.len()),
+                    },
+                )
+                .await?;
+            page_token = listing.page_token;
+            // Only child directories can be tables, and the store already separates them
+            // out, so the objects in the page are not looked at.
+            tables.extend(
+                listing
+                    .result
+                    .common_prefixes
+                    .iter()
+                    .filter_map(|location| table_name(location, &dir_suffix)),
+            );
+            // Children that are not tables leave the page short of the limit, so keep
+            // going until the page is full or the database runs out.
+            if page_token.is_none() || limit.is_none_or(|limit| tables.len() >= limit) {
+                break;
             }
-            _ => None,
-        };
+        }
 
         Ok(ListTablesResponse {
             context: None,
-            tables: f,
-            page_token: next_page_token,
+            tables,
+            page_token,
         })
     }
 
@@ -1476,7 +1513,7 @@ mod tests {
     use crate::table::{AnyQuery, WriteOptions};
     use arrow_array::{Int32Array, RecordBatch, StringArray};
     use arrow_schema::{DataType, Field, Schema, SchemaRef};
-    use futures::{TryStreamExt, stream::once};
+    use futures::{TryStreamExt, future::try_join_all, stream::once};
     use std::path::PathBuf;
     use std::sync::Arc;
     use std::time::Duration;
@@ -1484,6 +1521,182 @@ mod tests {
     use tokio::sync::Barrier;
     use tokio::time::timeout;
 
+    async fn create_tables(db: &ListingDatabase, names: &[&str]) {
+        let schema = Arc::new(Schema::new(vec![Field::new("id", DataType::Int32, false)]));
+        for name in names {
+            db.create_table(CreateTableRequest {
+                name: name.to_string(),
+                namespace_path: vec![],
+                data: Box::new(RecordBatch::new_empty(schema.clone())) as Box<dyn Scannable>,
+                mode: CreateTableMode::Create,
+                write_options: Default::default(),
+                location: None,
+                namespace_client: None,
+            })
+            .await
+            .unwrap();
+        }
+    }
+
+    /// Every table in the database, taken `limit` at a time, which is how a caller walks a
+    /// listing: the token ends the walk, never a short page.
+    async fn walk(db: &ListingDatabase, limit: Option<i32>) -> Vec<String> {
+        let mut seen = Vec::new();
+        let mut page_token = None;
+        loop {
+            let page = db
+                .list_tables(ListTablesRequest {
+                    limit,
+                    page_token,
+                    ..Default::default()
+                })
+                .await
+                .unwrap();
+            seen.extend(page.tables);
+            page_token = page.page_token;
+            if page_token.is_none() {
+                return seen;
+            }
+            assert!(
+                seen.len() < 100,
+                "the walk is serving tables more than once"
+            );
+        }
+    }
+
+    /// Paging with the returned token has to visit every table exactly once, whatever the
+    /// page size, with nothing lost or repeated at a boundary.
+    #[rstest::rstest]
+    #[tokio::test]
+    async fn test_list_tables_pages_over_every_table_once(#[values(1, 2, 3, 5, 10)] limit: i32) {
+        let (_tempdir, db) = setup_database().await;
+        create_tables(&db, &["a", "b", "c", "d", "e"]).await;
+
+        assert_eq!(walk(&db, Some(limit)).await, vec!["a", "b", "c", "d", "e"]);
+    }
+
+    /// The token is opaque: it is whatever resumes the store the database sits on, not a
+    /// table name. Callers hand it back and nothing else.
+    ///
+    /// Nothing validates a token, so one invented by a caller is read as a position rather
+    /// than refused — which is why the token has to come back from a previous page.
+    #[tokio::test]
+    async fn test_the_page_token_is_not_a_table_name() {
+        let (_tempdir, db) = setup_database().await;
+        create_tables(&db, &["a", "b", "c"]).await;
+
+        let page = db
+            .list_tables(ListTablesRequest {
+                limit: Some(1),
+                ..Default::default()
+            })
+            .await
+            .unwrap();
+
+        assert_eq!(page.tables, vec!["a"]);
+        let token = page.page_token.expect("two tables are still to come");
+        assert_ne!(token, "a");
+
+        // Handing it back is the only thing a caller does with it, and it resumes.
+        let rest = db
+            .list_tables(ListTablesRequest {
+                page_token: Some(token),
+                ..Default::default()
+            })
+            .await
+            .unwrap();
+        assert_eq!(rest.tables, vec!["b", "c"]);
+    }
+
+    /// A limit the listing does not fill leaves no token behind, so a caller paging by token
+    /// stops without asking for an empty page.
+    #[tokio::test]
+    async fn test_a_listing_that_runs_out_has_no_token() {
+        let (_tempdir, db) = setup_database().await;
+        create_tables(&db, &["a", "b"]).await;
+
+        let page = db
+            .list_tables(ListTablesRequest {
+                limit: Some(10),
+                ..Default::default()
+            })
+            .await
+            .unwrap();
+
+        assert_eq!(page.tables, vec!["a", "b"]);
+        assert_eq!(page.page_token, None);
+    }
+
+    /// An empty page token means "from the start", which is how a client looping on a token
+    /// spells its first request.
+    #[tokio::test]
+    async fn test_an_empty_page_token_lists_from_the_start() {
+        let (_tempdir, db) = setup_database().await;
+        create_tables(&db, &["a", "b"]).await;
+
+        let page = db
+            .list_tables(ListTablesRequest {
+                page_token: Some(String::new()),
+                ..Default::default()
+            })
+            .await
+            .unwrap();
+
+        assert_eq!(page.tables, vec!["a", "b"]);
+    }
+
+    /// Listing follows the order the object store lists directories in, so a name that
+    /// extends another comes first: the `-` of `users-archive.lance` sorts below the `.` of
+    /// `users.lance`. Pagination pushes its cursor into the list request, so it cannot report
+    /// an order other than the one it resumes in.
+    #[tokio::test]
+    async fn test_listing_order_follows_the_store_not_the_table_name() {
+        let (_tempdir, db) = setup_database().await;
+        create_tables(&db, &["users", "users-archive", "users.old"]).await;
+
+        assert_eq!(
+            walk(&db, None).await,
+            vec!["users-archive", "users", "users.old"]
+        );
+        // And paging reports the same order, so a walk sees each table once.
+        assert_eq!(
+            walk(&db, Some(1)).await,
+            vec!["users-archive", "users", "users.old"]
+        );
+    }
+
+    /// Only directories named `<name>.lance` are tables; loose files and other directories
+    /// under the database prefix are not. A page spent on them is filled from the next one,
+    /// so a page holding only non-tables does not read as an empty database.
+    #[tokio::test]
+    async fn test_listing_ignores_non_table_children() {
+        let (tempdir, db) = setup_database().await;
+        create_tables(&db, &["real"]).await;
+        std::fs::write(tempdir.path().join("aaa-loose.lance"), b"not a table").unwrap();
+        create_dir_all(tempdir.path().join("aaa-scratch")).unwrap();
+
+        let page = db
+            .list_tables(ListTablesRequest {
+                limit: Some(1),
+                ..Default::default()
+            })
+            .await
+            .unwrap();
+
+        assert_eq!(page.tables, vec!["real"]);
+    }
+
+    #[tokio::test]
+    async fn listing_ignores_empty_table_name() {
+        let (tempdir, db) = setup_database().await;
+        create_dir_all(tempdir.path().join(".lance")).unwrap();
+        let page = db.list_tables(ListTablesRequest::default()).await.unwrap();
+        assert!(
+            page.tables.is_empty(),
+            "invalid empty table name was listed"
+        );
+    }
+
     async fn setup_database() -> (tempfile::TempDir, ListingDatabase) {
         let tempdir = tempdir().unwrap();
         let uri = tempdir.path().to_str().unwrap();
@@ -1614,6 +1827,59 @@ mod tests {
         );
     }
 
+    #[tokio::test]
+    async fn test_concurrent_open_table_reuses_connection_object_store() {
+        let tempdir = tempdir().unwrap();
+        let uri = tempdir.path().to_str().unwrap();
+        let session = Arc::new(lance::session::Session::default());
+        let request = ConnectRequest {
+            uri: uri.to_string(),
+            #[cfg(feature = "remote")]
+            client_config: Default::default(),
+            options: Default::default(),
+            namespace_client_properties: Default::default(),
+            manifest_enabled: false,
+            read_consistency_interval: None,
+            session: Some(session.clone()),
+        };
+        let db = ListingDatabase::connect_with_options(&request)
+            .await
+            .unwrap();
+        let schema = Arc::new(Schema::new(vec![Field::new("id", DataType::Int32, false)]));
+        db.create_table(CreateTableRequest {
+            name: "test".to_string(),
+            namespace_path: vec![],
+            data: Box::new(RecordBatch::new_empty(schema)) as Box<dyn Scannable>,
+            mode: CreateTableMode::Create,
+            write_options: Default::default(),
+            location: None,
+            namespace_client: None,
+        })
+        .await
+        .unwrap();
+
+        let before = session.store_registry().stats();
+        let opened_tables = try_join_all((0..32).map(|_| {
+            db.open_table(OpenTableRequest {
+                name: "test".to_string(),
+                namespace_path: vec![],
+                index_cache_size: None,
+                lance_read_params: None,
+                location: None,
+                namespace_client: None,
+                managed_versioning: None,
+            })
+        }))
+        .await
+        .unwrap();
+        let after = session.store_registry().stats();
+
+        assert_eq!(opened_tables.len(), 32);
+        assert_eq!(after.misses, before.misses);
+        assert_eq!(after.active_stores, before.active_stores);
+        assert!(after.hits >= before.hits + 32);
+    }
+
     #[tokio::test]
     async fn test_listing_database_root_ops_do_not_create_manifest() {
         let tempdir = tempdir().unwrap();

rust/lancedb/src/expr.rs

@@ -19,6 +19,7 @@
 
 mod sql;
 
+pub(crate) use sql::canonicalize_sql_predicate;
 pub use sql::expr_to_sql_string;
 
 use std::sync::Arc;
@@ -156,7 +157,7 @@ mod tests {
         use datafusion_common::ScalarValue;
         let expr = col("data").eq(lit(ScalarValue::Binary(Some(vec![0xca, 0xfe]))));
         let sql = expr_to_sql_string(&expr).unwrap();
-        assert_eq!(sql, "(data = X'CAFE')");
+        assert_eq!(sql, "(`data` = X'CAFE')");
     }
 
     #[test]
@@ -166,7 +167,7 @@ mod tests {
         let int_expr = col("id").gt(lit(5i64));
         let combined = bin_expr.and(int_expr);
         let sql = expr_to_sql_string(&combined).unwrap();
-        assert_eq!(sql, "((data = X'01') AND (id > 5))");
+        assert_eq!(sql, "((`data` = X'01') AND (id > 5))");
     }
 
     #[test]
@@ -184,7 +185,7 @@ mod tests {
         // serialized correctly (regression test for placeholder rewrite path).
         let expr = contains(col("data"), lit(ScalarValue::Binary(Some(vec![0xff]))));
         let sql = expr_to_sql_string(&expr).unwrap();
-        assert_eq!(sql, "contains(data, X'FF')");
+        assert_eq!(sql, "contains(`data`, X'FF')");
     }
 
     #[test]
@@ -195,7 +196,7 @@ mod tests {
             .eq(lit(ScalarValue::Binary(Some(vec![0xab, 0xcd]))))
             .not();
         let sql = expr_to_sql_string(&expr).unwrap();
-        assert_eq!(sql, "NOT (data = X'ABCD')");
+        assert_eq!(sql, "NOT (`data` = X'ABCD')");
     }
 
     #[test]
@@ -205,6 +206,122 @@ mod tests {
         assert!(sql.contains("IN"), "expected IN in: {}", sql);
     }
 
+    #[test]
+    fn test_empty_is_in() {
+        let expr = is_in(col("id"), vec![]);
+        assert_eq!(expr_to_sql_string(&expr).unwrap(), "false");
+    }
+
+    #[test]
+    fn test_empty_is_in_discards_binary_children() {
+        use datafusion_common::ScalarValue;
+
+        let expr = is_in(
+            col("payload").eq(lit(ScalarValue::Binary(Some(vec![0x01])))),
+            vec![],
+        );
+        assert_eq!(expr_to_sql_string(&expr).unwrap(), "false");
+    }
+
+    #[test]
+    fn test_keyword_identifier() {
+        let expr = col("null").eq(lit(1i64));
+        assert_eq!(expr_to_sql_string(&expr).unwrap(), "(`null` = 1)");
+    }
+
+    #[test]
+    fn test_decimal_literal_preserves_type() {
+        use datafusion_common::ScalarValue;
+
+        let expr = col("val").lt(lit(ScalarValue::Decimal128(
+            Some(1_234_567_890_123_456_790),
+            19,
+            18,
+        )));
+        let sql = expr_to_sql_string(&expr).unwrap();
+        assert_eq!(
+            sql,
+            "(val < arrow_cast('1.234567890123456790', 'Decimal128(19, 18)'))"
+        );
+    }
+
+    #[test]
+    fn test_non_finite_float_literal_preserves_type() {
+        let expr = col("x").lt(lit(f64::INFINITY));
+        assert_eq!(
+            expr_to_sql_string(&expr).unwrap(),
+            "(x < arrow_cast('inf', 'Float64'))"
+        );
+    }
+
+    #[test]
+    fn test_cast_uses_arrow_type_name() {
+        let string = expr_cast(col("x"), DataType::Utf8);
+        assert_eq!(
+            expr_to_sql_string(&string).unwrap(),
+            "arrow_cast(x, 'Utf8')"
+        );
+
+        let int32 = expr_cast(col("x"), DataType::Int32);
+        assert_eq!(
+            expr_to_sql_string(&int32).unwrap(),
+            "arrow_cast(x, 'Int32')"
+        );
+
+        let expr = expr_cast(col("x"), DataType::Float16).lt(lit(2.0));
+        assert_eq!(
+            expr_to_sql_string(&expr).unwrap(),
+            "(arrow_cast(x, 'Float16') < 2.0)"
+        );
+
+        let decimal = expr_cast(lit("2.00"), DataType::Decimal256(40, 2));
+        assert_eq!(
+            expr_to_sql_string(&decimal).unwrap(),
+            "arrow_cast('2.00', 'Decimal256(40, 2)')"
+        );
+    }
+
+    #[test]
+    fn test_binary_placeholder_does_not_rewrite_user_string() {
+        use datafusion_common::ScalarValue;
+
+        let marker = "__lancedb_binary_placeholder_0__";
+        let expr = col("payload")
+            .eq(lit(ScalarValue::Binary(Some(vec![0x01]))))
+            .or(col("text").eq(lit(marker)));
+        assert_eq!(
+            expr_to_sql_string(&expr).unwrap(),
+            "((payload = X'01') OR (`text` = '__lancedb_binary_placeholder_0__'))"
+        );
+    }
+
+    #[test]
+    fn test_binary_binding_skips_quoted_identifiers() {
+        use datafusion_common::ScalarValue;
+
+        let expr = col("payload")
+            .eq(lit(ScalarValue::Binary(Some(vec![0x01]))))
+            .and(col("odd'name").eq(lit(1i64)))
+            .and(col("odd`'name").eq(lit(2i64)));
+        assert_eq!(
+            expr_to_sql_string(&expr).unwrap(),
+            "(((payload = X'01') AND (`odd'name` = 1)) AND (`odd``'name` = 2))"
+        );
+    }
+
+    #[test]
+    fn test_binary_placeholder_collision_search_is_linear() {
+        use datafusion_common::ScalarValue;
+
+        let collision_shaped = format!("__lancedb_binary_placeholder_0__{}", "_".repeat(64_000));
+        let expr = col("payload")
+            .eq(lit(ScalarValue::Binary(Some(vec![0x01]))))
+            .and(col("text").eq(lit(collision_shaped.clone())));
+        let sql = expr_to_sql_string(&expr).unwrap();
+        assert!(sql.contains("X'01'"));
+        assert!(sql.contains(&format!("'{collision_shaped}'")));
+    }
+
     #[test]
     fn test_multiple_binary_literals() {
         use datafusion_common::ScalarValue;

rust/lancedb/src/expr/sql.rs

@@ -1,10 +1,27 @@
 // SPDX-License-Identifier: Apache-2.0
 // SPDX-FileCopyrightText: Copyright The LanceDB Authors
 
+use std::{
+    any::TypeId,
+    collections::{HashMap, HashSet},
+};
+
+use arrow_array::types::{
+    Decimal32Type, Decimal64Type, Decimal128Type, Decimal256Type, DecimalType,
+};
+use arrow_schema::DataType;
 use datafusion_common::ScalarValue;
 use datafusion_common::tree_node::{Transformed, TreeNode, TreeNodeRecursion};
 use datafusion_expr::Expr;
-use datafusion_sql::unparser::{self, dialect::Dialect};
+use datafusion_functions::core::expr_fn::{
+    arrow_cast as datafusion_arrow_cast, arrow_try_cast as datafusion_arrow_try_cast,
+};
+use datafusion_sql::sqlparser::{
+    dialect::{Dialect as SqlParserDialect, GenericDialect},
+    keywords::ALL_KEYWORDS,
+    tokenizer::{Token, Tokenizer},
+};
+use datafusion_sql::unparser::{self, dialect::Dialect as UnparserDialect};
 
 /// Unparser dialect that matches the quoting style expected by the Lance SQL
 /// parser.  Lance uses backtick (`` ` ``) as the only delimited-identifier
@@ -19,17 +36,74 @@ use datafusion_sql::unparser::{self, dialect::Dialect};
 ///   lower-case by the SQL parser, which would break case-sensitive schemas).
 struct LanceSqlDialect;
 
-impl Dialect for LanceSqlDialect {
+impl UnparserDialect for LanceSqlDialect {
     fn identifier_quote_style(&self, identifier: &str) -> Option<char> {
-        let needs_quote = identifier.chars().any(|c| c.is_ascii_uppercase())
-            || !identifier
-                .chars()
-                .enumerate()
-                .all(|(i, c)| c == '_' || c.is_ascii_alphabetic() || (i > 0 && c.is_ascii_digit()));
+        let identifier_upper = identifier.to_ascii_uppercase();
+        let needs_quote =
+            (identifier_upper != "ID" && ALL_KEYWORDS.contains(&identifier_upper.as_str()))
+                || identifier.chars().any(|c| c.is_ascii_uppercase())
+                || !identifier.chars().enumerate().all(|(i, c)| {
+                    c == '_' || c.is_ascii_alphabetic() || (i > 0 && c.is_ascii_digit())
+                });
         if needs_quote { Some('`') } else { None }
     }
 }
 
+/// Lance's tokenizer dialect with SQL-standard double-quoted identifiers added.
+///
+/// Keep this deliberately small: Lance's parser wraps `GenericDialect` and
+/// delegates only identifier recognition, leaving every other dialect option at
+/// its default. In particular, `/*! ... */` remains an ordinary block comment.
+#[derive(Debug, Default)]
+struct PredicateDialect(GenericDialect);
+
+impl SqlParserDialect for PredicateDialect {
+    fn dialect(&self) -> TypeId {
+        self.0.dialect()
+    }
+
+    fn is_identifier_start(&self, ch: char) -> bool {
+        self.0.is_identifier_start(ch)
+    }
+
+    fn is_identifier_part(&self, ch: char) -> bool {
+        self.0.is_identifier_part(ch)
+    }
+
+    fn is_delimited_identifier_start(&self, ch: char) -> bool {
+        ch == '"' || ch == '`'
+    }
+}
+
+/// Canonicalize a raw SQL predicate for Lance's parser.
+///
+/// Lance wraps [`GenericDialect`] for identifier recognition while retaining the
+/// default dialect behavior for every other lexical option. [`PredicateDialect`]
+/// mirrors that contract and additionally recognizes `"` as an identifier
+/// delimiter, allowing this function to rewrite only those identifier tokens.
+pub fn canonicalize_sql_predicate(predicate: &str) -> crate::Result<String> {
+    let dialect = PredicateDialect::default();
+    let tokens = Tokenizer::new(&dialect, predicate)
+        .with_unescape(false)
+        .tokenize()
+        .map_err(|err| crate::Error::InvalidInput {
+            message: format!("invalid SQL predicate: {err}"),
+        })?;
+
+    Ok(tokens
+        .into_iter()
+        .map(|token| match token {
+            Token::Word(word) if word.quote_style == Some('"') => {
+                // with_unescape(false) retains doubled double quotes. Decode
+                // those before escaping any backticks for Lance's delimiter.
+                let identifier = word.value.replace("\"\"", "\"").replace('`', "``");
+                format!("`{identifier}`")
+            }
+            other => other.to_string(),
+        })
+        .collect())
+}
+
 /// Prefix for placeholder strings inserted in place of binary literals.  Chosen
 /// to be extremely unlikely to occur in user data.
 const BINARY_PLACEHOLDER_PREFIX: &str = "__lancedb_binary_placeholder_";
@@ -39,24 +113,128 @@ fn bytes_to_hex_sql(bytes: &[u8]) -> String {
     format!("X'{hex}'")
 }
 
-/// Returns true if *expr* contains a `Binary` or `LargeBinary` scalar literal
-/// anywhere in its subtree.  DataFusion's SQL unparser cannot serialize those
-/// variants, so we route such expressions through a placeholder-substitution
-/// path that emits SQL `X'...'` byte-string literals.
-fn has_binary_literal(expr: &Expr) -> bool {
-    let mut found = false;
+fn string_literals(expr: &Expr) -> HashSet<String> {
+    let mut literals = HashSet::new();
     let _ = expr.apply(&mut |e: &Expr| {
-        if matches!(
-            e,
-            Expr::Literal(ScalarValue::Binary(_) | ScalarValue::LargeBinary(_), _)
-        ) {
-            found = true;
-            Ok(TreeNodeRecursion::Stop)
-        } else {
-            Ok(TreeNodeRecursion::Continue)
+        if let Expr::Literal(
+            ScalarValue::Utf8(Some(value))
+            | ScalarValue::LargeUtf8(Some(value))
+            | ScalarValue::Utf8View(Some(value)),
+            _,
+        ) = e
+        {
+            literals.insert(value.clone());
         }
+        Ok(TreeNodeRecursion::Continue)
     });
-    found
+    literals
+}
+
+fn typed_string_literal(value: String, data_type: DataType) -> Expr {
+    datafusion_arrow_cast(
+        Expr::Literal(ScalarValue::Utf8(Some(value)), None),
+        Expr::Literal(ScalarValue::Utf8(Some(data_type.to_string())), None),
+    )
+}
+
+fn next_binary_placeholder(user_strings: &HashSet<String>, next_id: &mut usize) -> String {
+    loop {
+        let placeholder = format!("{BINARY_PLACEHOLDER_PREFIX}{}__", *next_id);
+        *next_id += 1;
+        if !user_strings.contains(&placeholder) {
+            return placeholder;
+        }
+    }
+}
+
+fn bind_binary_literals(
+    sql: &str,
+    mut bindings: HashMap<String, Vec<u8>>,
+) -> crate::Result<String> {
+    let bytes = sql.as_bytes();
+    let mut output = Vec::with_capacity(bytes.len());
+    let mut index = 0;
+
+    // Walk SQL string tokens once. Placeholders are plain, unescaped string
+    // literals, so this remains linear even when user strings are large or
+    // deliberately resemble the placeholder prefix.
+    while index < bytes.len() {
+        if bytes[index] == b'`' {
+            let identifier_start = index;
+            index += 1;
+            let mut identifier_end = None;
+            while index < bytes.len() {
+                if bytes[index] == b'`' {
+                    if index + 1 < bytes.len() && bytes[index + 1] == b'`' {
+                        index += 2;
+                    } else {
+                        index += 1;
+                        identifier_end = Some(index);
+                        break;
+                    }
+                } else {
+                    index += 1;
+                }
+            }
+
+            let Some(identifier_end) = identifier_end else {
+                return Err(crate::Error::InvalidInput {
+                    message: "unterminated identifier while binding binary literal".to_string(),
+                });
+            };
+            output.extend_from_slice(&bytes[identifier_start..identifier_end]);
+            continue;
+        }
+
+        if bytes[index] != b'\'' {
+            output.push(bytes[index]);
+            index += 1;
+            continue;
+        }
+
+        let literal_start = index;
+        index += 1;
+        let content_start = index;
+        let mut escaped = false;
+        let mut content_end = None;
+        while index < bytes.len() {
+            if bytes[index] == b'\'' {
+                if index + 1 < bytes.len() && bytes[index + 1] == b'\'' {
+                    escaped = true;
+                    index += 2;
+                } else {
+                    content_end = Some(index);
+                    index += 1;
+                    break;
+                }
+            } else {
+                index += 1;
+            }
+        }
+
+        let Some(content_end) = content_end else {
+            return Err(crate::Error::InvalidInput {
+                message: "unterminated string while binding binary literal".to_string(),
+            });
+        };
+
+        let placeholder = &sql[content_start..content_end];
+        if !escaped && let Some(value) = bindings.remove(placeholder) {
+            output.extend_from_slice(bytes_to_hex_sql(&value).as_bytes());
+        } else {
+            output.extend_from_slice(&bytes[literal_start..index]);
+        }
+    }
+
+    if !bindings.is_empty() {
+        return Err(crate::Error::InvalidInput {
+            message: "failed to bind binary literal while serializing expression".to_string(),
+        });
+    }
+
+    String::from_utf8(output).map_err(|e| crate::Error::InvalidInput {
+        message: format!("failed to bind binary literal: {e}"),
+    })
 }
 
 fn run_unparser(expr: &Expr) -> crate::Result<String> {
@@ -69,25 +247,37 @@ fn run_unparser(expr: &Expr) -> crate::Result<String> {
 }
 
 pub fn expr_to_sql_string(expr: &Expr) -> crate::Result<String> {
-    // Fast path: no binary literals — DataFusion's unparser handles everything.
-    if !has_binary_literal(expr) {
-        return run_unparser(expr);
-    }
-
-    // Slow path: DataFusion's unparser cannot serialize `Binary`/`LargeBinary`
-    // scalars, so we rewrite each one to a unique string-literal placeholder,
-    // let the unparser do the rest of the work, then substitute the SQL
-    // `X'...'` byte-string literal back in.  This keeps the operator/function
-    // serialization logic centralized in DataFusion and works for every
-    // expression node type the unparser supports.
-    let mut bindings: Vec<Vec<u8>> = Vec::new();
+    // DataFusion's unparser needs a few adaptations before its SQL can be
+    // reparsed by Lance without changing the typed expression's semantics:
+    //
+    // * decimal literals need an explicit cast to preserve precision and scale;
+    // * casts need exact Arrow type names rather than SQL type aliases;
+    // * an empty IN list is valid in DataFusion but invalid SQL;
+    // * binary literals are unsupported by the unparser and need placeholders.
+    // Eliminate empty membership expressions before visiting their children.
+    // Otherwise a discarded binary child could leave behind a stale binding.
     let rewritten = expr
         .clone()
+        .transform(|e: Expr| match e {
+            Expr::InList(in_list) if in_list.list.is_empty() => Ok(Transformed::yes(
+                Expr::Literal(ScalarValue::Boolean(Some(in_list.negated)), None),
+            )),
+            other => Ok(Transformed::no(other)),
+        })
+        .map_err(|e| crate::Error::InvalidInput {
+            message: format!("failed to rewrite expression: {e}"),
+        })?
+        .data;
+
+    let user_strings = string_literals(&rewritten);
+    let mut next_placeholder_id = 0;
+    let mut binary_bindings = HashMap::new();
+    let rewritten = rewritten
         .transform(|e: Expr| match e {
             Expr::Literal(ScalarValue::Binary(Some(bytes)), m)
             | Expr::Literal(ScalarValue::LargeBinary(Some(bytes)), m) => {
-                let placeholder = format!("{}{}__", BINARY_PLACEHOLDER_PREFIX, bindings.len());
-                bindings.push(bytes);
+                let placeholder = next_binary_placeholder(&user_strings, &mut next_placeholder_id);
+                binary_bindings.insert(placeholder.clone(), bytes);
                 Ok(Transformed::yes(Expr::Literal(
                     ScalarValue::Utf8(Some(placeholder)),
                     m,
@@ -97,6 +287,57 @@ pub fn expr_to_sql_string(expr: &Expr) -> crate::Result<String> {
             | Expr::Literal(ScalarValue::LargeBinary(None), m) => {
                 Ok(Transformed::yes(Expr::Literal(ScalarValue::Null, m)))
             }
+            Expr::Literal(ScalarValue::Decimal32(Some(value), precision, scale), _m) => {
+                let value = Decimal32Type::format_decimal(value, precision, scale);
+                Ok(Transformed::yes(typed_string_literal(
+                    value,
+                    DataType::Decimal32(precision, scale),
+                )))
+            }
+            Expr::Literal(ScalarValue::Decimal64(Some(value), precision, scale), _m) => {
+                let value = Decimal64Type::format_decimal(value, precision, scale);
+                Ok(Transformed::yes(typed_string_literal(
+                    value,
+                    DataType::Decimal64(precision, scale),
+                )))
+            }
+            Expr::Literal(ScalarValue::Decimal128(Some(value), precision, scale), _m) => {
+                let value = Decimal128Type::format_decimal(value, precision, scale);
+                Ok(Transformed::yes(typed_string_literal(
+                    value,
+                    DataType::Decimal128(precision, scale),
+                )))
+            }
+            Expr::Literal(ScalarValue::Decimal256(Some(value), precision, scale), _m) => {
+                let value = Decimal256Type::format_decimal(value, precision, scale);
+                Ok(Transformed::yes(typed_string_literal(
+                    value,
+                    DataType::Decimal256(precision, scale),
+                )))
+            }
+            Expr::Literal(ScalarValue::Float16(Some(value)), _m) if !value.is_finite() => Ok(
+                Transformed::yes(typed_string_literal(value.to_string(), DataType::Float16)),
+            ),
+            Expr::Literal(ScalarValue::Float32(Some(value)), _m) if !value.is_finite() => Ok(
+                Transformed::yes(typed_string_literal(value.to_string(), DataType::Float32)),
+            ),
+            Expr::Literal(ScalarValue::Float64(Some(value)), _m) if !value.is_finite() => Ok(
+                Transformed::yes(typed_string_literal(value.to_string(), DataType::Float64)),
+            ),
+            Expr::Cast(cast) => Ok(Transformed::yes(datafusion_arrow_cast(
+                *cast.expr,
+                Expr::Literal(
+                    ScalarValue::Utf8(Some(cast.field.data_type().to_string())),
+                    None,
+                ),
+            ))),
+            Expr::TryCast(cast) => Ok(Transformed::yes(datafusion_arrow_try_cast(
+                *cast.expr,
+                Expr::Literal(
+                    ScalarValue::Utf8(Some(cast.field.data_type().to_string())),
+                    None,
+                ),
+            ))),
             other => Ok(Transformed::no(other)),
         })
         .map_err(|e| crate::Error::InvalidInput {
@@ -104,12 +345,58 @@ pub fn expr_to_sql_string(expr: &Expr) -> crate::Result<String> {
         })?
         .data;
 
-    let mut sql = run_unparser(&rewritten)?;
-    for (i, bytes) in bindings.iter().enumerate() {
-        // The unparser quotes string literals with single quotes, so the
-        // placeholder appears as `'__lancedb_binary_placeholder_<i>__'`.
-        let quoted = format!("'{}{}__'", BINARY_PLACEHOLDER_PREFIX, i);
-        sql = sql.replace(&quoted, &bytes_to_hex_sql(bytes));
+    let sql = run_unparser(&rewritten)?;
+    if binary_bindings.is_empty() {
+        Ok(sql)
+    } else {
+        bind_binary_literals(&sql, binary_bindings)
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::canonicalize_sql_predicate;
+
+    #[test]
+    fn normalizes_double_quoted_identifiers() {
+        assert_eq!(
+            canonicalize_sql_predicate(r#""PartyAbbrev" = 'D'"#).unwrap(),
+            "`PartyAbbrev` = 'D'"
+        );
+        assert_eq!(
+            canonicalize_sql_predicate(r#""MetaData"."userId" = 5"#).unwrap(),
+            "`MetaData`.`userId` = 5"
+        );
+        assert_eq!(
+            canonicalize_sql_predicate(r#""a""b" = 1"#).unwrap(),
+            "`a\"b` = 1"
+        );
+    }
+
+    #[test]
+    fn preserves_quotes_inside_literals_and_backticks() {
+        let filter = r#"name = 'Alice "Ace"' AND `quoted"field` = 1"#;
+        assert_eq!(canonicalize_sql_predicate(filter).unwrap(), filter);
+    }
+
+    #[test]
+    fn preserves_literals_and_comments_using_lance_dialect_rules() {
+        let predicate = r#"path = '\' AND "PartyAbbrev" = 'D' -- unmatched " in comment"#;
+        assert_eq!(
+            canonicalize_sql_predicate(predicate).unwrap(),
+            r#"path = '\' AND `PartyAbbrev` = 'D' -- unmatched " in comment"#
+        );
+
+        let predicate = r#"id = 1 /* unmatched " in block comment */"#;
+        assert_eq!(canonicalize_sql_predicate(predicate).unwrap(), predicate);
+
+        let predicate = r#"id = 1 /*! OR "PartyAbbrev" = 'D' */"#;
+        assert_eq!(canonicalize_sql_predicate(predicate).unwrap(), predicate);
+    }
+
+    #[test]
+    fn rejects_unterminated_double_quoted_identifier() {
+        let error = canonicalize_sql_predicate(r#""PartyAbbrev = 'D'"#).unwrap_err();
+        assert!(matches!(error, crate::Error::InvalidInput { .. }));
     }
-    Ok(sql)
 }

rust/lancedb/src/function.rs

@@ -186,6 +186,9 @@ pub struct PythonEnvironmentSpec {
     pub kind: String,
     #[serde(default, skip_serializing_if = "Vec::is_empty")]
     pub packages: Vec<String>,
+    /// Conda channels in priority order; conda environments only.
+    #[serde(default, skip_serializing_if = "Vec::is_empty")]
+    pub channels: Vec<String>,
     #[serde(default, skip_serializing_if = "Option::is_none")]
     pub path: Option<String>,
     #[serde(default, skip_serializing_if = "Vec::is_empty")]
@@ -583,3 +586,26 @@ impl RefreshColumnResult {
 }
 
 impl_json!(RefreshColumnResult);
+
+#[cfg(test)]
+mod conda_environment_tests {
+    use super::PythonEnvironmentSpec;
+
+    #[test]
+    fn conda_channels_round_trip_and_pip_stays_bare() {
+        let conda: PythonEnvironmentSpec = serde_json::from_str(
+            r#"{"kind":"conda","packages":["numpy"],"channels":["conda-forge"]}"#,
+        )
+        .unwrap();
+        assert_eq!(conda.channels, ["conda-forge"]);
+        assert!(
+            serde_json::to_string(&conda)
+                .unwrap()
+                .contains(r#""channels":["conda-forge"]"#)
+        );
+
+        let pip: PythonEnvironmentSpec =
+            serde_json::from_str(r#"{"kind":"pip","packages":["numpy"]}"#).unwrap();
+        assert!(!serde_json::to_string(&pip).unwrap().contains("channels"));
+    }
+}

rust/lancedb/src/index/scalar.rs

@@ -63,4 +63,5 @@ pub struct FmIndexBuilder {}
 pub use lance_index::scalar::FullTextSearchQuery;
 pub use lance_index::scalar::InvertedIndexParams as FtsIndexBuilder;
 pub use lance_index::scalar::InvertedIndexParams;
+pub use lance_index::scalar::inverted::DocumentGranularity;
 pub use lance_index::scalar::inverted::query::*;

rust/lancedb/src/io/object_store.rs

@@ -10,7 +10,7 @@ use lance::io::WrappingObjectStore;
 use object_store::{
     CopyOptions, Error, GetOptions, GetResult, ListResult, MultipartUpload, ObjectMeta,
     ObjectStore, ObjectStoreExt, PutMultipartOptions, PutOptions, PutPayload, PutResult, Result,
-    UploadPart, path::Path,
+    UploadPart, list::PaginatedListStore, path::Path,
 };
 
 use async_trait::async_trait;
@@ -187,6 +187,14 @@ impl WrappingObjectStore for MirroringObjectStoreWrapper {
             secondary: self.secondary.clone(),
         })
     }
+
+    fn wrap_paginated(
+        &self,
+        _store_prefix: &str,
+        original: Arc<dyn PaginatedListStore>,
+    ) -> Option<Arc<dyn PaginatedListStore>> {
+        Some(original)
+    }
 }
 
 // windows pathing can't be simply concatenated

rust/lancedb/src/io/object_store/io_tracking.rs

@@ -12,7 +12,7 @@ use lance::io::WrappingObjectStore;
 use object_store::{
     CopyOptions, GetOptions, GetResult, ListResult, MultipartUpload, ObjectMeta, ObjectStore,
     PutMultipartOptions, PutOptions, PutPayload, PutResult, RenameOptions, Result as OSResult,
-    UploadPart, path::Path,
+    UploadPart, list::PaginatedListStore, path::Path,
 };
 
 #[derive(Debug, Default)]
@@ -57,6 +57,14 @@ impl WrappingObjectStore for IoStatsHolder {
             stats: self.0.clone(),
         })
     }
+
+    fn wrap_paginated(
+        &self,
+        _store_prefix: &str,
+        original: Arc<dyn PaginatedListStore>,
+    ) -> Option<Arc<dyn PaginatedListStore>> {
+        Some(original)
+    }
 }
 
 impl IoTrackingStore {

rust/lancedb/src/job.rs

@@ -47,6 +47,10 @@ impl TerminalResult {
         }
     }
 
+    pub(crate) fn value(&self) -> Option<&Value> {
+        self.value.as_ref()
+    }
+
     fn decode<T: DeserializeOwned>(self) -> Result<T> {
         let value = self.value.ok_or_else(|| match &self.request_id {
             Some(request_id) => Error::Http {

rust/lancedb/src/materialized_view.rs

@@ -170,6 +170,15 @@ pub(crate) fn plan(
     filter: Option<&str>,
     limit: Option<u64>,
 ) -> Result<(MaterializedViewDefinition, Vec<ArrowField>, Lineage)> {
+    let filter = filter
+        .map(crate::expr::canonicalize_sql_predicate)
+        .transpose()
+        .map_err(|err| match err {
+            Error::InvalidInput { message } => Error::InvalidInput {
+                message: format!("invalid view filter: {message}"),
+            },
+            err => err,
+        })?;
     let projections: Vec<(String, String)> = if projections.is_empty() {
         source_schema
             .fields()
@@ -274,7 +283,7 @@ pub(crate) fn plan(
         declared.push(output);
     }
 
-    if let Some(filter) = filter {
+    if let Some(filter) = filter.as_deref() {
         let expr = planner
             .parse_filter(filter)
             .map_err(|e| Error::InvalidInput {
@@ -314,7 +323,7 @@ pub(crate) fn plan(
             .into_iter()
             .map(|(output, expression)| ViewProjection { output, expression })
             .collect(),
-        filter: filter.map(String::from),
+        filter,
         limit,
         inputs,
     };

rust/lancedb/src/materialized_view/refresh.rs

@@ -46,8 +46,9 @@ use lance_table::format::Fragment;
 use serde::{Deserialize, Serialize};
 
 use super::{
-    INCARNATION_META_KEY, MaterializedViewDefinition, REFRESHED_AT_MS_META_KEY,
-    SOURCE_ROW_ID_COLUMN, SOURCE_VERSION_META_KEY,
+    DEFINITION_META_KEY, INCARNATION_META_KEY, MaterializedViewDefinition,
+    REFRESHED_AT_MS_META_KEY, SOURCE_ROW_ID_COLUMN, SOURCE_VERSION_META_KEY,
+    definition_to_metadata,
 };
 use crate::database::OpenTableRequest;
 use crate::table::{NativeTable, NativeTableExt, Table};
@@ -197,8 +198,28 @@ pub(crate) async fn execute_refresh(
             ),
         });
     }
+    let definition_changed =
+        definition.filter != replanned.filter || definition.inputs != replanned.inputs;
     let definition = &replanned;
 
+    // A watermark written for a legacy raw filter certifies the rows that
+    // filter produced, not the canonical predicate above. Rebuild instead of
+    // accepting or advancing it, and persist the migrated definition in the
+    // same metadata commit that certifies the replacement rows.
+    if definition_changed {
+        return rebuild(
+            view_native,
+            &view_ds,
+            &source_ds,
+            source_version,
+            source_ts,
+            definition,
+            true,
+            expected_incarnation,
+        )
+        .await;
+    }
+
     let metadata = &view_ds.schema().metadata;
     let watermark: Option<u64> = metadata
         .get(SOURCE_VERSION_META_KEY)
@@ -257,6 +278,7 @@ pub(crate) async fn execute_refresh(
                         source_version,
                         source_ts,
                         definition,
+                        false,
                         expected_incarnation,
                     )
                     .await
@@ -271,6 +293,7 @@ pub(crate) async fn execute_refresh(
                 source_version,
                 source_ts,
                 definition,
+                false,
                 expected_incarnation,
             )
             .await
@@ -683,6 +706,7 @@ async fn incremental(
             view_ds.clone(),
             source_version,
             source_ts,
+            None,
             expected_incarnation,
         )
         .await?;
@@ -704,6 +728,7 @@ async fn incremental(
             published,
             source_version,
             source_ts,
+            None,
             expected_incarnation,
         )
         .await?;
@@ -775,6 +800,7 @@ async fn incremental(
             published,
             source_version,
             source_ts,
+            None,
             expected_incarnation,
         )
         .await?;
@@ -824,12 +850,14 @@ async fn incremental(
         appended,
         source_version,
         source_ts,
+        None,
         expected_incarnation,
     )
     .await?;
     Ok(Some(result))
 }
 
+#[allow(clippy::too_many_arguments)]
 async fn rebuild(
     view_native: &NativeTable,
     view_ds: &Dataset,
@@ -837,6 +865,7 @@ async fn rebuild(
     source_version: u64,
     source_ts: u128,
     definition: &MaterializedViewDefinition,
+    persist_definition: bool,
     expected_incarnation: Option<&str>,
 ) -> Result<RefreshMaterializedViewResult> {
     let rows_written = Arc::new(AtomicU64::new(0));
@@ -867,6 +896,7 @@ async fn rebuild(
         replaced,
         source_version,
         source_ts,
+        persist_definition.then_some(definition),
         expected_incarnation,
     )
     .await?;
@@ -981,6 +1011,7 @@ async fn stamp_watermark(
     mut dataset: Dataset,
     source_version: u64,
     source_ts: u128,
+    definition: Option<&MaterializedViewDefinition>,
     expected_incarnation: Option<&str>,
 ) -> Result<u64> {
     ensure_incarnation(&dataset, expected_incarnation, dataset.uri()).await?;
@@ -993,27 +1024,32 @@ async fn stamp_watermark(
         .get(INCARNATION_META_KEY)
         .cloned()
         .unwrap_or_else(|| uuid::Uuid::new_v4().to_string());
-    dataset
-        .update_schema_metadata([
-            (INCARNATION_META_KEY.to_string(), Some(incarnation)),
-            (
-                SOURCE_VERSION_META_KEY.to_string(),
-                Some(source_version.to_string()),
-            ),
-            (
-                SOURCE_VERSION_TS_META_KEY.to_string(),
-                Some(source_ts.to_string()),
-            ),
-            (
-                REFRESHED_AT_MS_META_KEY.to_string(),
-                Some(now_ms().to_string()),
-            ),
-            (
-                VIEW_VERSION_META_KEY.to_string(),
-                Some(predicted.to_string()),
-            ),
-        ])
-        .await?;
+    let mut metadata = vec![(INCARNATION_META_KEY.to_string(), Some(incarnation))];
+    if let Some(definition) = definition {
+        metadata.push((
+            DEFINITION_META_KEY.to_string(),
+            Some(definition_to_metadata(definition)?),
+        ));
+    }
+    metadata.extend([
+        (
+            SOURCE_VERSION_META_KEY.to_string(),
+            Some(source_version.to_string()),
+        ),
+        (
+            SOURCE_VERSION_TS_META_KEY.to_string(),
+            Some(source_ts.to_string()),
+        ),
+        (
+            REFRESHED_AT_MS_META_KEY.to_string(),
+            Some(now_ms().to_string()),
+        ),
+        (
+            VIEW_VERSION_META_KEY.to_string(),
+            Some(predicted.to_string()),
+        ),
+    ]);
+    dataset.update_schema_metadata(metadata).await?;
     let actual = dataset.version().version;
     if actual != predicted {
         return Err(Error::Runtime {
@@ -1585,6 +1621,106 @@ mod tests {
         assert_eq!(read(view.table(), "x").await, vec![20, 40]);
     }
 
+    #[tokio::test]
+    async fn test_mixed_case_filter_is_canonicalized_for_lineage_and_refresh() {
+        let conn = connect("memory://").execute().await.unwrap();
+        let batch = record_batch!(
+            ("id", Int32, [1, 2, 3]),
+            ("PartyAbbrev", Utf8, ["D", "R", "D"])
+        )
+        .unwrap();
+        conn.create_table("src", batch)
+            .write_options(crate::materialized_view::tests::stable_row_ids())
+            .execute()
+            .await
+            .unwrap();
+        conn.create_materialized_view("democrats", "src")
+            .select([("id", "id")])
+            .only_if(r#""PartyAbbrev" = 'D'"#)
+            .execute()
+            .await
+            .unwrap();
+
+        // Reopen from schema metadata so these assertions cover the stored
+        // predicate and lineage, not only the declaration-time handle.
+        let view = conn.open_materialized_view("democrats").await.unwrap();
+        assert_eq!(
+            view.definition().filter.as_deref(),
+            Some("`PartyAbbrev` = 'D'")
+        );
+        assert_eq!(view.definition().inputs, ["PartyAbbrev", "id"]);
+
+        let result = view.refresh().execute().await.unwrap();
+        assert_eq!(result.rows_written, 2);
+        assert_eq!(read(view.table(), "id").await, vec![1, 3]);
+    }
+
+    #[tokio::test]
+    async fn test_legacy_raw_filter_rebuilds_and_persists_canonical_definition() {
+        let conn = connect("memory://").execute().await.unwrap();
+        let batch = record_batch!(
+            ("id", Int32, [1, 2, 3]),
+            ("PartyAbbrev", Utf8, ["D", "R", "D"])
+        )
+        .unwrap();
+        conn.create_table("legacy_src", batch)
+            .write_options(crate::materialized_view::tests::stable_row_ids())
+            .execute()
+            .await
+            .unwrap();
+        let view = conn
+            .create_materialized_view("legacy_view", "legacy_src")
+            .select([("id", "id")])
+            .only_if(r#""PartyAbbrev" = 'X'"#)
+            .execute()
+            .await
+            .unwrap();
+        assert_eq!(view.refresh().execute().await.unwrap().rows_written, 0);
+
+        // Model a definition and up-to-date watermark written before filter
+        // canonicalization was applied to materialized views.
+        let mut legacy = view.definition().clone();
+        legacy.filter = Some(r#""PartyAbbrev" = 'D'"#.into());
+        legacy.inputs = vec!["id".into()];
+        let native = view.table().as_native().unwrap();
+        let mut dataset = native.dataset.get().await.unwrap().as_ref().clone();
+        let predicted = dataset.version().version + 1;
+        dataset
+            .update_schema_metadata([
+                (
+                    DEFINITION_META_KEY.to_string(),
+                    Some(definition_to_metadata(&legacy).unwrap()),
+                ),
+                (
+                    VIEW_VERSION_META_KEY.to_string(),
+                    Some(predicted.to_string()),
+                ),
+            ])
+            .await
+            .unwrap();
+        native.dataset.update(dataset);
+
+        let reopened = conn.open_materialized_view("legacy_view").await.unwrap();
+        let result = reopened.refresh().execute().await.unwrap();
+        assert_eq!(result.mode, RefreshMode::Rebuild);
+        assert_eq!(result.rows_written, 2);
+        assert_eq!(read(reopened.table(), "id").await, vec![1, 3]);
+
+        // A fresh handle proves the migration was stored alongside the new
+        // watermark and therefore happens only once.
+        let migrated = conn.open_materialized_view("legacy_view").await.unwrap();
+        assert_eq!(
+            migrated.definition().filter.as_deref(),
+            Some("`PartyAbbrev` = 'D'")
+        );
+        assert_eq!(migrated.definition().inputs, ["PartyAbbrev", "id"]);
+        assert_eq!(
+            migrated.refresh().execute().await.unwrap().mode,
+            RefreshMode::NoOp
+        );
+        assert_eq!(read(migrated.table(), "id").await, vec![1, 3]);
+    }
+
     #[tokio::test]
     async fn test_append_refreshes_incrementally() {
         let (_conn, source, view) = refreshed_doubled(vec![1, 2]).await;
@@ -2767,7 +2903,7 @@ mod tests {
         let stale = view_native.dataset.get().await.unwrap().as_ref().clone();
         view.table().delete("x = 1").await.unwrap();
 
-        let err = stamp_watermark(view_native, stale, 99, 99, None).await;
+        let err = stamp_watermark(view_native, stale, 99, 99, None, None).await;
         assert!(err.is_err());
 
         let result = view.refresh().execute().await.unwrap();

rust/lancedb/src/query.rs

@@ -399,6 +399,9 @@ pub trait QueryBase {
     /// x > 5 OR y = 'test'
     /// ```
     ///
+    /// Identifiers may be delimited with SQL-standard double quotes or
+    /// backticks. String literals must use single quotes.
+    ///
     /// Filtering performance can often be improved by creating a scalar index
     /// on the filter column(s).
     ///
@@ -913,6 +916,17 @@ impl QueryRequest {
     /// use different representations) the error is recorded and surfaced later
     /// by [`Self::check_filter`].
     pub(crate) fn add_filter(&mut self, new: QueryFilter) {
+        let new = match new {
+            QueryFilter::Sql(filter) => match crate::expr::canonicalize_sql_predicate(&filter) {
+                Ok(filter) => QueryFilter::Sql(filter),
+                Err(err) => {
+                    self.filter_error = Some(err.to_string());
+                    return;
+                }
+            },
+            other => other,
+        };
+
         self.filter = Some(match self.filter.take() {
             None => new,
             Some(existing) => match and_filters(existing, new) {
@@ -1174,12 +1188,12 @@ impl VectorQuery {
 
     /// Add another query vector to the search.
     ///
-    /// Multiple searches will be dispatched as part of the query.
-    /// This is a convenience method for adding multiple query vectors
-    /// to the search. It is not expected to be faster than issuing
-    /// multiple queries concurrently.
+    /// Multiple searches will be dispatched as a batch. Flat searches share
+    /// one table scan across the query vectors, avoiding the scan and memory
+    /// amplification of issuing the searches concurrently. Indexed searches
+    /// may still perform per-vector index work.
     ///
-    /// The output data will contain an additional columns `query_index` which
+    /// The output data will contain an additional column `query_index` which
     /// will contain the index of the query vector that was used to generate the
     /// result.
     pub fn add_query_vector(mut self, vector: impl IntoQueryVector) -> Result<Self> {
@@ -1646,10 +1660,14 @@ mod tests {
     use std::{collections::HashSet, sync::Arc};
 
     use super::*;
-    use arrow::{array::downcast_array, compute::concat_batches, datatypes::Int32Type};
+    use arrow::{
+        array::downcast_array,
+        compute::concat_batches,
+        datatypes::{Int32Type, UInt8Type},
+    };
     use arrow_array::{
-        FixedSizeListArray, Float32Array, Int32Array, RecordBatch, StringArray, cast::AsArray,
-        types::Float32Type,
+        FixedSizeListArray, Float32Array, Int32Array, RecordBatch, RecordBatchIterator,
+        StringArray, cast::AsArray, types::Float32Type,
     };
     use arrow_schema::{DataType, Field as ArrowField, Schema as ArrowSchema};
     use futures::{StreamExt, TryStreamExt};
@@ -1878,6 +1896,157 @@ mod tests {
         query.execute().await.unwrap();
     }
 
+    #[tokio::test]
+    async fn test_double_quoted_predicates_across_table_operations() {
+        let tmp_dir = tempdir().unwrap();
+        let dataset_path = tmp_dir.path().join("test.lance");
+        let uri = dataset_path.to_str().unwrap();
+        let schema = Arc::new(ArrowSchema::new(vec![
+            ArrowField::new("id", DataType::Int32, false),
+            ArrowField::new("PartyAbbrev", DataType::Utf8, false),
+            ArrowField::new("path", DataType::Utf8, false),
+        ]));
+        let batch = RecordBatch::try_new(
+            schema.clone(),
+            vec![
+                Arc::new(Int32Array::from(vec![1, 2, 3, 4])),
+                Arc::new(StringArray::from(vec!["D", "R", "R", "D"])),
+                Arc::new(StringArray::from(vec!["\\", "\\", "x", "x"])),
+            ],
+        )
+        .unwrap();
+
+        let conn = connect(uri).execute().await.unwrap();
+        let table = conn.create_table("parties", batch).execute().await.unwrap();
+        let batches = table
+            .query()
+            .only_if(r#""PartyAbbrev" = 'D'"#)
+            .execute()
+            .await
+            .unwrap()
+            .try_collect::<Vec<_>>()
+            .await
+            .unwrap();
+
+        assert_eq!(batches.iter().map(RecordBatch::num_rows).sum::<usize>(), 2);
+        assert_eq!(
+            table
+                .count_rows(Some(r#""PartyAbbrev" = 'D'"#.to_string()))
+                .await
+                .unwrap(),
+            2
+        );
+
+        // Public BaseTable dispatch cannot bypass canonicalization.
+        let query = AnyQuery::Query(QueryRequest {
+            filter: Some(QueryFilter::Sql(r#""PartyAbbrev" = 'D'"#.to_string())),
+            ..Default::default()
+        });
+        let batches = table
+            .base_table()
+            .query(&query, Default::default())
+            .await
+            .unwrap()
+            .try_collect::<Vec<_>>()
+            .await
+            .unwrap();
+        assert_eq!(batches.iter().map(RecordBatch::num_rows).sum::<usize>(), 2);
+        assert_eq!(
+            table
+                .base_table()
+                .count_rows(Some(crate::table::Filter::Sql(
+                    r#""PartyAbbrev" = 'D'"#.to_string(),
+                )))
+                .await
+                .unwrap(),
+            2
+        );
+
+        for predicate in [
+            r#"id = 1 -- unmatched " in a valid SQL comment"#,
+            r#"id = 1 /* unmatched " in a valid SQL comment */"#,
+            r#"id = 1 /*! OR "PartyAbbrev" = 'D' */"#,
+            r#"path = '\' AND "PartyAbbrev" = 'D'"#,
+        ] {
+            let batches = table
+                .query()
+                .only_if(predicate)
+                .execute()
+                .await
+                .unwrap()
+                .try_collect::<Vec<_>>()
+                .await
+                .unwrap();
+            assert_eq!(batches.iter().map(RecordBatch::num_rows).sum::<usize>(), 1);
+        }
+
+        // The same canonical predicate contract applies to both merge filters.
+        let source = RecordBatch::try_new(
+            schema.clone(),
+            vec![
+                Arc::new(Int32Array::from(vec![1, 2, 3])),
+                Arc::new(StringArray::from(vec!["D", "R", "R"])),
+                Arc::new(StringArray::from(vec!["\\", "\\", "x"])),
+            ],
+        )
+        .unwrap();
+        let mut merge = table.merge_insert(&["id"]);
+        merge.when_not_matched_by_source_delete(Some(r#""PartyAbbrev" = 'D'"#.to_string()));
+        let result = table
+            .base_table()
+            .merge_insert(
+                merge,
+                Box::new(RecordBatchIterator::new(vec![Ok(source)], schema.clone())),
+            )
+            .await
+            .unwrap();
+        assert_eq!(result.num_deleted_rows, 1);
+
+        let source = RecordBatch::try_new(
+            schema.clone(),
+            vec![
+                Arc::new(Int32Array::from(vec![1, 2, 3])),
+                Arc::new(StringArray::from(vec!["U", "U", "U"])),
+                Arc::new(StringArray::from(vec!["\\", "\\", "x"])),
+            ],
+        )
+        .unwrap();
+        let mut merge = table.merge_insert(&["id"]);
+        merge.when_matched_update_all(Some(r#"target."PartyAbbrev" = 'D'"#.to_string()));
+        merge
+            .execute(Box::new(RecordBatchIterator::new(vec![Ok(source)], schema)))
+            .await
+            .unwrap();
+        assert_eq!(
+            table
+                .count_rows(Some(r#""PartyAbbrev" = 'U'"#.to_string()))
+                .await
+                .unwrap(),
+            1
+        );
+
+        let update = table
+            .update()
+            .only_if(r#""PartyAbbrev" = 'R'"#)
+            .column("PartyAbbrev", "'X'");
+        table.base_table().update(update).await.unwrap();
+        assert_eq!(
+            table
+                .count_rows(Some(r#""PartyAbbrev" = 'X'"#.to_string()))
+                .await
+                .unwrap(),
+            2
+        );
+
+        let result = table
+            .base_table()
+            .delete(crate::table::Predicate::String(r#""PartyAbbrev" = 'X'"#))
+            .await
+            .unwrap();
+        assert_eq!(result.num_deleted_rows, 2);
+        assert_eq!(table.count_rows(None).await.unwrap(), 1);
+    }
+
     #[tokio::test]
     async fn test_select_with_transform() {
         let batches = make_non_empty_batches();
@@ -2334,7 +2503,8 @@ mod tests {
             .limit(1);
 
         let plan = query.explain_plan(true).await.unwrap();
-        assert!(plan.contains("UnionExec"));
+        assert!(plan.contains("KNNVectorDistance: queries=2"));
+        assert!(!plan.contains("UnionExec"));
 
         let results = query
             .execute()
@@ -2349,6 +2519,100 @@ mod tests {
         // We don't guarantee order.
         assert!(query_index.values().contains(&0));
         assert!(query_index.values().contains(&1));
+
+        // Batch KNN does not support a per-query offset, so offset queries keep
+        // the legacy per-vector plan to preserve their result semantics.
+        let offset_query = table
+            .query()
+            .nearest_to(&[0.1, 0.2, 0.3, 0.4])
+            .unwrap()
+            .add_query_vector(&[0.5, 0.6, 0.7, 0.8])
+            .unwrap()
+            .limit(1)
+            .offset(1);
+        assert!(
+            offset_query
+                .explain_plan(true)
+                .await
+                .unwrap()
+                .contains("UnionExec")
+        );
+        let offset_results = offset_query
+            .execute()
+            .await
+            .unwrap()
+            .try_collect::<Vec<_>>()
+            .await
+            .unwrap();
+        assert_eq!(
+            offset_results
+                .iter()
+                .map(RecordBatch::num_rows)
+                .sum::<usize>(),
+            2
+        );
+    }
+
+    #[tokio::test]
+    async fn test_multiple_binary_query_vectors() {
+        let vectors = FixedSizeListArray::from_iter_primitive::<UInt8Type, _, _>(
+            vec![
+                Some(vec![Some(0), Some(0)]),
+                Some(vec![Some(255), Some(255)]),
+            ],
+            2,
+        );
+        let schema = Arc::new(ArrowSchema::new(vec![
+            ArrowField::new("id", DataType::Int32, false),
+            ArrowField::new("vector", vectors.data_type().clone(), false),
+        ]));
+        let batch = RecordBatch::try_new(
+            schema,
+            vec![Arc::new(Int32Array::from(vec![0, 1])), Arc::new(vectors)],
+        )
+        .unwrap();
+
+        let conn = connect("memory://").execute().await.unwrap();
+        let table = conn
+            .create_table("binary_batch", batch)
+            .execute()
+            .await
+            .unwrap();
+        let query = table
+            .query()
+            .nearest_to(&[0.0, 0.0])
+            .unwrap()
+            .add_query_vector(&[255.0, 255.0])
+            .unwrap()
+            .distance_type(DistanceType::Hamming)
+            .limit(1);
+
+        // Binary queries retain the per-vector plan because Lance's binary
+        // nearest path requires primitive UInt8 query arrays.
+        assert!(
+            query
+                .explain_plan(true)
+                .await
+                .unwrap()
+                .contains("UnionExec")
+        );
+
+        let results = query
+            .execute()
+            .await
+            .unwrap()
+            .try_collect::<Vec<_>>()
+            .await
+            .unwrap();
+        let results = concat_batches(&results[0].schema(), &results).unwrap();
+        assert_eq!(results.num_rows(), 2);
+
+        let ids = results["id"].as_primitive::<Int32Type>();
+        assert!(ids.values().contains(&0));
+        assert!(ids.values().contains(&1));
+        let query_index = results["query_index"].as_primitive::<Int32Type>();
+        assert!(query_index.values().contains(&0));
+        assert!(query_index.values().contains(&1));
     }
 
     #[tokio::test]

rust/lancedb/src/remote/db.rs

@@ -87,6 +87,10 @@ impl ServerVersion {
     pub fn support_blobs(&self) -> bool {
         self.0 >= semver::Version::new(0, 5, 0)
     }
+
+    pub fn support_fts_document_granularity(&self) -> bool {
+        self.0 >= semver::Version::new(0, 6, 0)
+    }
 }
 
 pub const OPT_REMOTE_PREFIX: &str = "remote_database_";