Bump version: 0.38.0-beta.16 → 0.38.0

Updates the Rust workspace and Java lance-core dependency to Lance v11.0.0.

Includes compatibility adjustments for the Lance 11 object-store, table-listing, and shard-manifest APIs.

.bumpversion.toml

@@ -1,5 +1,5 @@
 [tool.bumpversion]
-current_version = "0.38.0-beta.11"
+current_version = "0.38.0"
 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 +3455,9 @@ checksum = "42703706b716c37f96a77aea830392ad231f44c9e9a67872fa5548707e11b11c"
 
 [[package]]
 name = "fsst"
-version = "12.0.0-beta.2"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.2#dafa4642658d996b3e31dde91e02f72db7860d7e"
+version = "11.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f727719438dfdb74f358a347c91ff81b6e7084a6421f34de3e473ce271f10caa"
 dependencies = [
  "arrow-array",
  "rand 0.9.5",
@@ -4815,8 +4816,9 @@ checksum = "e037a2e1d8d5fdbd49b16a4ea09d5d6401c1f29eca5ff29d03d3824dba16256a"
 
 [[package]]
 name = "lance"
-version = "12.0.0-beta.2"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.2#dafa4642658d996b3e31dde91e02f72db7860d7e"
+version = "11.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "be781f40c7a75f9eae2188a2f71174acb7a360dca97163db40b041d0828dea48"
 dependencies = [
  "arc-swap",
  "arrow",
@@ -4888,8 +4890,9 @@ dependencies = [
 
 [[package]]
 name = "lance-arrow"
-version = "12.0.0-beta.2"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.2#dafa4642658d996b3e31dde91e02f72db7860d7e"
+version = "11.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "fb97fd9875f3036d7c2561aa5b16eb87b80ccabaa4eeb5e6099b19cc662f1cd8"
 dependencies = [
  "arrow-array",
  "arrow-buffer",
@@ -4911,7 +4914,8 @@ dependencies = [
 [[package]]
 name = "lance-arrow-scalar"
 version = "58.0.0"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.2#dafa4642658d996b3e31dde91e02f72db7860d7e"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "771f68b04b47f3addf781116f65061808de94b05e1e9411c23c18f32d14ebe79"
 dependencies = [
  "arrow-array",
  "arrow-buffer",
@@ -4925,17 +4929,20 @@ dependencies = [
 [[package]]
 name = "lance-arrow-stats"
 version = "58.0.0"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.2#dafa4642658d996b3e31dde91e02f72db7860d7e"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "dd47ec33c90bf29f688fd02118e37d3a5ad5c339caa3163f89e417dc0867001f"
 dependencies = [
  "arrow-array",
  "arrow-schema",
+ "half",
  "lance-arrow-scalar",
 ]
 
 [[package]]
 name = "lance-bitpacking"
-version = "12.0.0-beta.2"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.2#dafa4642658d996b3e31dde91e02f72db7860d7e"
+version = "11.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6f45658c5b2dc9aada41b66ee44b83af3fa888b7385ae414bae951b12a9f1cd3"
 dependencies = [
  "arrayref",
  "crunchy",
@@ -4945,8 +4952,9 @@ dependencies = [
 
 [[package]]
 name = "lance-core"
-version = "12.0.0-beta.2"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.2#dafa4642658d996b3e31dde91e02f72db7860d7e"
+version = "11.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "27af3df3a7d08897efccd04461df31cedf0880c4b86a055ddce48e423d27f967"
 dependencies = [
  "arrow-array",
  "arrow-buffer",
@@ -4983,8 +4991,9 @@ dependencies = [
 
 [[package]]
 name = "lance-datafusion"
-version = "12.0.0-beta.2"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.2#dafa4642658d996b3e31dde91e02f72db7860d7e"
+version = "11.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "c377f837df5296e92f9fad724c83c1bef4e74d5af6e5a9312e9307e1dead8614"
 dependencies = [
  "arrow",
  "arrow-array",
@@ -5013,8 +5022,9 @@ dependencies = [
 
 [[package]]
 name = "lance-datagen"
-version = "12.0.0-beta.2"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.2#dafa4642658d996b3e31dde91e02f72db7860d7e"
+version = "11.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "778e1a5065fa4bc184e36e32681f10f8f4680ad8cedc9377b4c088dce8c5b8da"
 dependencies = [
  "arrow",
  "arrow-array",
@@ -5031,8 +5041,9 @@ dependencies = [
 
 [[package]]
 name = "lance-derive"
-version = "12.0.0-beta.2"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.2#dafa4642658d996b3e31dde91e02f72db7860d7e"
+version = "11.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "13e5e95e0fd3d74f7938f4bee623041421b323b5c61f242c8622a1f48a202527"
 dependencies = [
  "proc-macro2",
  "quote",
@@ -5041,8 +5052,9 @@ dependencies = [
 
 [[package]]
 name = "lance-encoding"
-version = "12.0.0-beta.2"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.2#dafa4642658d996b3e31dde91e02f72db7860d7e"
+version = "11.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "1625653c55c65f3426e281f6e29b54c603f38a40bd4cebd707bd4f3ea48be6c5"
 dependencies = [
  "arrow-arith",
  "arrow-array",
@@ -5075,8 +5087,9 @@ dependencies = [
 
 [[package]]
 name = "lance-file"
-version = "12.0.0-beta.2"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.2#dafa4642658d996b3e31dde91e02f72db7860d7e"
+version = "11.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b7e13c9266b478fc98f36ee19347c4658f7a6613fed77778b1a455fe1b88552e"
 dependencies = [
  "arrow-arith",
  "arrow-array",
@@ -5107,8 +5120,9 @@ dependencies = [
 
 [[package]]
 name = "lance-index"
-version = "12.0.0-beta.2"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.2#dafa4642658d996b3e31dde91e02f72db7860d7e"
+version = "11.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b0e0cb95f2c4f341c4dd04ac60f6a89ea26a6f75e09570225cbda4854c8b088e"
 dependencies = [
  "arc-swap",
  "arrow",
@@ -5172,8 +5186,9 @@ dependencies = [
 
 [[package]]
 name = "lance-index-core"
-version = "12.0.0-beta.2"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.2#dafa4642658d996b3e31dde91e02f72db7860d7e"
+version = "11.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "79ccd371977c1f7168da259d66ad37154f23146f093d46136bc7f79559f00f2c"
 dependencies = [
  "arrow-array",
  "arrow-schema",
@@ -5195,8 +5210,9 @@ dependencies = [
 
 [[package]]
 name = "lance-io"
-version = "12.0.0-beta.2"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.2#dafa4642658d996b3e31dde91e02f72db7860d7e"
+version = "11.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "414d50997391b1ac83dc183c1612fdff88f58b959806078dc4c5e465154566de"
 dependencies = [
  "arrow",
  "arrow-array",
@@ -5236,8 +5252,9 @@ dependencies = [
 
 [[package]]
 name = "lance-linalg"
-version = "12.0.0-beta.2"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.2#dafa4642658d996b3e31dde91e02f72db7860d7e"
+version = "11.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6ff55b152ef23a56d7ba7e4d1b2c9cf0cc79aef6ee607c115597557ea4059f41"
 dependencies = [
  "arrow-array",
  "arrow-schema",
@@ -5251,8 +5268,9 @@ dependencies = [
 
 [[package]]
 name = "lance-namespace"
-version = "12.0.0-beta.2"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.2#dafa4642658d996b3e31dde91e02f72db7860d7e"
+version = "11.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "09991c13ab282b731e323619613914e08da9cc82b312f904e58c128b23f2f0e3"
 dependencies = [
  "arrow",
  "async-trait",
@@ -5264,8 +5282,9 @@ dependencies = [
 
 [[package]]
 name = "lance-namespace-impls"
-version = "12.0.0-beta.2"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.2#dafa4642658d996b3e31dde91e02f72db7860d7e"
+version = "11.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0ec0bc005f6bb8f120774eb4a9ba02e10463d8338167a46cbb1391d46680a174"
 dependencies = [
  "arrow",
  "arrow-ipc",
@@ -5318,8 +5337,9 @@ dependencies = [
 
 [[package]]
 name = "lance-select"
-version = "12.0.0-beta.2"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.2#dafa4642658d996b3e31dde91e02f72db7860d7e"
+version = "11.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e8f676a2a1837cc85b77feb5326d3296827da964e40d67144f646563302a6ce9"
 dependencies = [
  "arrow-array",
  "arrow-buffer",
@@ -5333,8 +5353,9 @@ dependencies = [
 
 [[package]]
 name = "lance-table"
-version = "12.0.0-beta.2"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.2#dafa4642658d996b3e31dde91e02f72db7860d7e"
+version = "11.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "bd33054347395048b1d842dfb85a13f7801392c2da5f39425db62f00a481744b"
 dependencies = [
  "arrow",
  "arrow-array",
@@ -5374,8 +5395,9 @@ dependencies = [
 
 [[package]]
 name = "lance-testing"
-version = "12.0.0-beta.2"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.2#dafa4642658d996b3e31dde91e02f72db7860d7e"
+version = "11.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ecc9ad9ae24f045dfddd538a39e55a28fa7e1ca6ad9f23e20d4087eaf2bb66f7"
 dependencies = [
  "arrow-array",
  "arrow-schema",
@@ -5388,8 +5410,9 @@ dependencies = [
 
 [[package]]
 name = "lance-tokenizer"
-version = "12.0.0-beta.2"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.2#dafa4642658d996b3e31dde91e02f72db7860d7e"
+version = "11.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "3bfa6f0164c8b7056150f5682ce4d415a335b59b04c479873fda04b200117d27"
 dependencies = [
  "frostem",
  "icu_segmenter",
@@ -5402,7 +5425,7 @@ dependencies = [
 
 [[package]]
 name = "lancedb"
-version = "0.38.0-beta.11"
+version = "0.38.0-beta.15"
 dependencies = [
  "ahash",
  "anyhow",
@@ -5490,7 +5513,7 @@ dependencies = [
 
 [[package]]
 name = "lancedb-nodejs"
-version = "0.38.0-beta.11"
+version = "0.38.0-beta.15"
 dependencies = [
  "arrow-array",
  "arrow-buffer",
@@ -5515,7 +5538,7 @@ dependencies = [
 
 [[package]]
 name = "lancedb-python"
-version = "0.38.0-beta.11"
+version = "0.38.0-beta.15"
 dependencies = [
  "arrow",
  "async-trait",

Cargo.toml

@@ -13,20 +13,20 @@ categories = ["database-implementations"]
 rust-version = "1.91.0"
 
 [workspace.dependencies]
-lance = { "version" = "=12.0.0-beta.2", default-features = false, "tag" = "v12.0.0-beta.2", "git" = "https://github.com/lance-format/lance.git" }
-lance-core = { "version" = "=12.0.0-beta.2", "tag" = "v12.0.0-beta.2", "git" = "https://github.com/lance-format/lance.git" }
-lance-datagen = { "version" = "=12.0.0-beta.2", "tag" = "v12.0.0-beta.2", "git" = "https://github.com/lance-format/lance.git" }
-lance-file = { "version" = "=12.0.0-beta.2", "tag" = "v12.0.0-beta.2", "git" = "https://github.com/lance-format/lance.git" }
-lance-io = { "version" = "=12.0.0-beta.2", default-features = false, "tag" = "v12.0.0-beta.2", "git" = "https://github.com/lance-format/lance.git" }
-lance-index = { "version" = "=12.0.0-beta.2", "tag" = "v12.0.0-beta.2", "git" = "https://github.com/lance-format/lance.git" }
-lance-linalg = { "version" = "=12.0.0-beta.2", "tag" = "v12.0.0-beta.2", "git" = "https://github.com/lance-format/lance.git" }
-lance-namespace = { "version" = "=12.0.0-beta.2", "tag" = "v12.0.0-beta.2", "git" = "https://github.com/lance-format/lance.git" }
-lance-namespace-impls = { "version" = "=12.0.0-beta.2", default-features = false, "tag" = "v12.0.0-beta.2", "git" = "https://github.com/lance-format/lance.git" }
-lance-table = { "version" = "=12.0.0-beta.2", "tag" = "v12.0.0-beta.2", "git" = "https://github.com/lance-format/lance.git" }
-lance-testing = { "version" = "=12.0.0-beta.2", "tag" = "v12.0.0-beta.2", "git" = "https://github.com/lance-format/lance.git" }
-lance-datafusion = { "version" = "=12.0.0-beta.2", "tag" = "v12.0.0-beta.2", "git" = "https://github.com/lance-format/lance.git" }
-lance-encoding = { "version" = "=12.0.0-beta.2", "tag" = "v12.0.0-beta.2", "git" = "https://github.com/lance-format/lance.git" }
-lance-arrow = { "version" = "=12.0.0-beta.2", "tag" = "v12.0.0-beta.2", "git" = "https://github.com/lance-format/lance.git" }
+lance = { "version" = "=11.0.0", default-features = false }
+lance-core = "=11.0.0"
+lance-datagen = "=11.0.0"
+lance-file = "=11.0.0"
+lance-io = { "version" = "=11.0.0", default-features = false }
+lance-index = "=11.0.0"
+lance-linalg = "=11.0.0"
+lance-namespace = "=11.0.0"
+lance-namespace-impls = { "version" = "=11.0.0", default-features = false }
+lance-table = "=11.0.0"
+lance-testing = "=11.0.0"
+lance-datafusion = "=11.0.0"
+lance-encoding = "=11.0.0"
+lance-arrow = "=11.0.0"
 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.11</version>
+    <version>0.38.0</version>
 </dependency>
 ```
 

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.11</version>
+      <version>0.38.0-final.0</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.11</version>
+    <version>0.38.0-final.0</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>12.0.0-beta.2</lance-core.version>
+        <lance-core.version>11.0.0</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.11"
+version = "0.38.0"
 publish = false
 license.workspace = true
 description.workspace = true

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/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/npm/darwin-arm64/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-darwin-arm64",
-	"version": "0.38.0-beta.11",
+	"version": "0.38.0",
 	"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.11",
+	"version": "0.38.0",
 	"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.11",
+	"version": "0.38.0",
 	"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.11",
+	"version": "0.38.0",
 	"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.11",
+	"version": "0.38.0",
 	"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.11",
+  "version": "0.38.0",
   "os": [
     "win32"
   ],

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

@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-win32-x64-msvc",
-	"version": "0.38.0-beta.11",
+	"version": "0.38.0",
 	"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.11",
+  "version": "0.38.0",
   "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": {

python/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "lancedb-python"
-version = "0.38.0-beta.11"
+version = "0.38.0"
 publish = false
 edition.workspace = true
 description = "Python bindings for LanceDB"

python/python/lancedb/functions.py

@@ -49,11 +49,25 @@ from pydantic import (
     model_validator,
 )
 
+from .schema import is_blob_v2_field as _is_blob_v2_field
+
 _Int32 = conint(strict=True, ge=-(2**31), le=2**31 - 1)
 _UInt32 = conint(strict=True, ge=0, le=2**32 - 1)
 _UInt64 = conint(strict=True, ge=0, le=2**64 - 1)
 
 
+def _validate_gpu_wire_marker(value: Any) -> bool:
+    if value is not True:
+        raise ValueError("runtime.gpu must be true")
+    return True
+
+
+def _normalize_gpu_marker(value: bool) -> Optional[bool]:
+    if not isinstance(value, bool):
+        raise ValueError("gpu must be a boolean")
+    return True if value else None
+
+
 class _FrozenDict(dict):
     def _immutable(self, *args, **kwargs):
         raise TypeError("remote canonical values are immutable")
@@ -239,6 +253,23 @@ class PythonRuntimeSpec(_RemoteValue):
     python_version: Optional[str] = None
     environment: Optional[PythonEnvironmentSpec] = None
     env: Optional[Mapping[str, str]] = None
+    gpu: Optional[bool] = None
+
+    @model_validator(mode="before")
+    @classmethod
+    def _discard_unknown_runtime_payload(cls, value):
+        if isinstance(value, Mapping):
+            kind = value.get("kind")
+            if isinstance(kind, str) and kind not in {"python", "python_v2"}:
+                return {"kind": kind}
+        return value
+
+    @field_validator("gpu", mode="before")
+    @classmethod
+    def _validate_gpu_marker(cls, value):
+        if value is None:
+            return None
+        return _validate_gpu_wire_marker(value)
 
     @model_validator(mode="after")
     def _validate_runtime_kind(self):
@@ -247,18 +278,28 @@ class PythonRuntimeSpec(_RemoteValue):
                 raise ValueError("python runtime requires python_version")
             if self.environment is None:
                 raise ValueError("python runtime requires environment")
+            if self.gpu is not None:
+                raise ValueError("python runtime with gpu requires kind='python_v2'")
+        elif self.kind == "python_v2":
+            if self.python_version is None:
+                raise ValueError("python_v2 runtime requires python_version")
+            if self.environment is None:
+                raise ValueError("python_v2 runtime requires environment")
+            if self.gpu is None:
+                raise ValueError("python_v2 runtime requires gpu")
         else:
             object.__setattr__(self, "python_version", None)
             object.__setattr__(self, "environment", None)
             object.__setattr__(self, "env", None)
+            object.__setattr__(self, "gpu", None)
         return self
 
 
 class FunctionVersion(_RemoteValue):
     """An exact immutable Function version returned by Enterprise.
 
-    Scheduling resources, priority, concurrency, and retry policy belong to
-    the submitting Job and are not part of this identity.
+    The GPU execution requirement is part of this identity. CPU and memory sizing,
+    priority, concurrency, and retry policy belong to the execution platform.
     """
 
     name: str
@@ -479,6 +520,7 @@ class RefreshColumnResult(_RemoteValue):
 
 
 _FUNCTION_NAME = re.compile(r"^[A-Za-z_][A-Za-z0-9_.-]*$")
+_FUNCTION_BLOB_V2_TYPE = "blob_v2"
 
 
 _GRAMMAR_PRIMITIVES = (
@@ -495,6 +537,7 @@ _GRAMMAR_PRIMITIVES = (
     (pa.float32(), "float32"),
     (pa.float64(), "float64"),
     (pa.string(), "utf8"),
+    (pa.large_string(), "large_utf8"),
     (pa.binary(), "binary"),
     (pa.date32(), "date32"),
     (pa.date64(), "date64"),
@@ -502,31 +545,177 @@ _GRAMMAR_PRIMITIVES = (
 
 
 def _canonical_arrow_type(data_type: pa.DataType) -> str:
-    """The server's V1 Function type grammar. Anything outside it is rejected
-    here rather than at registration."""
+    """The compact Function grammar, or canonical exact JSON for nested types."""
+    grammar = _grammar_arrow_type(data_type)
+    if grammar is not None:
+        return grammar
+    exact = _exact_arrow_type(data_type)
+    return json.dumps(exact, ensure_ascii=False, sort_keys=True, separators=(",", ":"))
+
+
+def _grammar_arrow_type(data_type: pa.DataType) -> Optional[str]:
     for candidate, name in _GRAMMAR_PRIMITIVES:
         if data_type == candidate:
             return name
     if pa.types.is_list(data_type) or pa.types.is_large_list(data_type):
+        item = _grammar_list_item(data_type)
+        if item is None:
+            return None
         prefix = "list" if pa.types.is_list(data_type) else "large_list"
-        return f"{prefix}<{_canonical_list_item(data_type)}>"
+        return f"{prefix}<{item}>"
     if pa.types.is_fixed_size_list(data_type) and data_type.list_size > 0:
-        return (
-            f"fixed_size_list<{_canonical_list_item(data_type)}, {data_type.list_size}>"
-        )
-    raise TypeError(f"unsupported Arrow type for Function signature: {data_type}")
+        item = _grammar_list_item(data_type)
+        if item is not None:
+            return f"fixed_size_list<{item}, {data_type.list_size}>"
+    return None
 
 
-def _canonical_list_item(data_type: pa.DataType) -> str:
+def _grammar_list_item(data_type: pa.DataType) -> Optional[str]:
     """The grammar names only the item type; it always means a non-nullable
-    child called `item`, so any other child metadata cannot be represented."""
+    child called `item`, so other child properties require exact JSON."""
     child = data_type.value_field
     if child.name != "item" or child.nullable or child.metadata:
+        return None
+    return _grammar_arrow_type(child.type)
+
+
+def _validate_exact_arrow_field(field: pa.Field) -> None:
+    if not field.name:
         raise TypeError(
-            "unsupported Arrow type for Function signature: list items must be a "
-            f"non-nullable field named 'item', got {child}"
+            "unsupported Arrow type for Function signature: field names "
+            "must not be empty"
         )
-    return _canonical_arrow_type(child.type)
+    if _is_blob_v2_field(field):
+        if not _has_supported_blob_v2_layout(field):
+            raise TypeError(
+                "unsupported Arrow type for Function signature: lance.blob.v2 "
+                f"requires a supported Blob storage layout, got {field}"
+            )
+    elif field.metadata:
+        raise TypeError(
+            "unsupported Arrow type for Function signature: field metadata "
+            f"is not supported, got {field}"
+        )
+
+
+def _has_supported_blob_v2_layout(field: pa.Field) -> bool:
+    data_type = field.type
+    if isinstance(data_type, pa.ExtensionType):
+        data_type = data_type.storage_type
+    if not pa.types.is_struct(data_type):
+        return False
+
+    fields = tuple(data_type)
+
+    def matches(spec, compare_nullable) -> bool:
+        return len(fields) == len(spec) and all(
+            actual.name == name
+            and actual.type == expected_type
+            and (not check_nullable or actual.nullable == nullable)
+            for actual, (name, expected_type, nullable), check_nullable in zip(
+                fields, spec, compare_nullable
+            )
+        )
+
+    logical_minimal = (
+        ("data", pa.large_binary(), True),
+        ("uri", pa.utf8(), True),
+    )
+    logical_full = logical_minimal + (
+        ("position", pa.uint64(), True),
+        ("size", pa.uint64(), True),
+    )
+    prepared = (
+        ("kind", pa.uint8(), True),
+        ("data", pa.large_binary(), True),
+        ("uri", pa.utf8(), True),
+        ("blob_id", pa.uint32(), True),
+        ("blob_size", pa.uint64(), True),
+        ("position", pa.uint64(), True),
+    )
+    descriptor = (
+        ("kind", pa.uint8(), False),
+        ("position", pa.uint64(), False),
+        ("size", pa.uint64(), False),
+        ("blob_id", pa.uint32(), False),
+        ("blob_uri", pa.utf8(), False),
+    )
+    return (
+        matches(logical_minimal, (True, True))
+        or matches(logical_full, (True, True, False, False))
+        or matches(prepared, (True,) * len(prepared))
+        or matches(descriptor, (False,) * len(descriptor))
+    )
+
+
+def _canonical_arrow_field(field: pa.Field) -> str:
+    _validate_exact_arrow_field(field)
+    if _is_blob_v2_field(field):
+        return _FUNCTION_BLOB_V2_TYPE
+    return _canonical_arrow_type(field.type)
+
+
+def _exact_arrow_field(field: pa.Field) -> dict[str, Any]:
+    _validate_exact_arrow_field(field)
+    if _is_blob_v2_field(field):
+        raise TypeError(
+            "unsupported Arrow type for Function signature: nested Blob v2 "
+            "fields are not supported; declare Blob parameters or named result "
+            "fields directly"
+        )
+    value = {
+        "name": field.name,
+        "nullable": field.nullable,
+        "type": _exact_arrow_type(field.type),
+    }
+    return value
+
+
+def _exact_arrow_type(data_type: pa.DataType) -> dict[str, Any]:
+    for candidate, name in _GRAMMAR_PRIMITIVES:
+        if data_type == candidate:
+            return {"type": name}
+    if pa.types.is_struct(data_type):
+        fields = list(data_type)
+        names = [field.name for field in fields]
+        if not fields or len(set(names)) != len(names):
+            raise TypeError(
+                "unsupported Arrow type for Function signature: structs must have "
+                "non-empty, uniquely named fields"
+            )
+        return {
+            "type": "struct",
+            "fields": [_exact_arrow_field(field) for field in fields],
+        }
+    if (
+        pa.types.is_list(data_type)
+        or pa.types.is_large_list(data_type)
+        or pa.types.is_fixed_size_list(data_type)
+    ):
+        if pa.types.is_fixed_size_list(data_type):
+            if data_type.value_field.name != "item":
+                raise TypeError(
+                    "unsupported Arrow type for Function signature: fixed-size list "
+                    "items must be named 'item'"
+                )
+            if data_type.list_size <= 0:
+                raise TypeError(
+                    f"unsupported Arrow type for Function signature: {data_type}"
+                )
+        value: dict[str, Any] = {
+            "type": (
+                "list"
+                if pa.types.is_list(data_type)
+                else "large_list"
+                if pa.types.is_large_list(data_type)
+                else "fixed_size_list"
+            ),
+            "fields": [_exact_arrow_field(data_type.value_field)],
+        }
+        if pa.types.is_fixed_size_list(data_type):
+            value["length"] = data_type.list_size
+        return value
+    raise TypeError(f"unsupported Arrow type for Function signature: {data_type}")
 
 
 def _list_of(item: pa.DataType) -> pa.DataType:
@@ -600,8 +789,15 @@ def _callable_parameters(function: Callable[..., Any]) -> tuple[inspect.Paramete
 
 def _function_output(output: pa.DataType | pa.Field | pa.Schema) -> FunctionOutput:
     if isinstance(output, pa.Schema):
+        if output.metadata:
+            raise TypeError("Function output schema metadata is not supported")
         fields = tuple(output)
-    elif isinstance(output, pa.Field) and pa.types.is_struct(output.type):
+    elif (
+        isinstance(output, pa.Field)
+        and not _is_blob_v2_field(output)
+        and pa.types.is_struct(output.type)
+    ):
+        _validate_exact_arrow_field(output)
         if output.nullable:
             raise ValueError("Function output must be non-nullable")
         fields = tuple(output.type)
@@ -617,11 +813,12 @@ def _function_output(output: pa.DataType | pa.Field | pa.Schema) -> FunctionOutp
             raise TypeError(
                 "output_schema must be a PyArrow DataType, Field, or Schema"
             )
+        _validate_exact_arrow_field(field)
         if field.nullable:
             raise ValueError("Function output must be non-nullable")
         return FunctionOutput(
             kind="scalar",
-            arrow_type=_canonical_arrow_type(field.type),
+            arrow_type=_canonical_arrow_field(field),
             nullable=False,
         )
 
@@ -629,6 +826,8 @@ def _function_output(output: pa.DataType | pa.Field | pa.Schema) -> FunctionOutp
         raise ValueError("named-struct Function output must contain at least one field")
     if any(field.nullable for field in fields):
         raise ValueError("Function output fields must be non-nullable")
+    for field in fields:
+        _validate_exact_arrow_field(field)
     names = [field.name for field in fields]
     if len(set(names)) != len(names):
         raise ValueError("Function output field names must be unique")
@@ -637,7 +836,7 @@ def _function_output(output: pa.DataType | pa.Field | pa.Schema) -> FunctionOutp
         fields=tuple(
             FunctionResultField(
                 name=field.name,
-                arrow_type=_canonical_arrow_type(field.type),
+                arrow_type=_canonical_arrow_field(field),
                 nullable=False,
             )
             for field in fields
@@ -657,6 +856,10 @@ def _infer_signature(
     if input_schema is not None:
         if not isinstance(input_schema, pa.Schema):
             raise TypeError("input_schema must be a PyArrow Schema")
+        if input_schema.metadata:
+            raise TypeError("Function input schema metadata is not supported")
+        for field in input_schema:
+            _validate_exact_arrow_field(field)
         expected = tuple(parameter.name for parameter in parameters)
         actual = tuple(input_schema.names)
         if actual != expected:
@@ -667,7 +870,7 @@ def _infer_signature(
         inputs = tuple(
             FunctionParameter(
                 name=field.name,
-                arrow_type=_canonical_arrow_type(field.type),
+                arrow_type=_canonical_arrow_field(field),
                 nullable=field.nullable,
             )
             for field in input_schema
@@ -690,7 +893,9 @@ def _infer_signature(
         inputs.append(
             FunctionParameter(
                 name=parameter.name,
-                arrow_type=_canonical_arrow_type(data_type),
+                arrow_type=_canonical_arrow_field(
+                    pa.field(parameter.name, data_type, nullable=nullable)
+                ),
                 nullable=nullable,
             )
         )
@@ -910,6 +1115,7 @@ class UdfDefinition:
         pip: tuple[str, ...],
         env: Mapping[str, str],
         python_version: Optional[str],
+        gpu: bool = False,
         conda: tuple[str, ...] = (),
         conda_channels: tuple[str, ...] = (),
     ):
@@ -938,12 +1144,14 @@ class UdfDefinition:
         signature = _infer_signature(function, input_schema, output_schema)
         source = _package_source(function)
         digest = f"sha256:{hashlib.sha256(source).hexdigest()}"
+        gpu_marker = _normalize_gpu_marker(gpu)
         runtime = PythonRuntimeSpec(
-            kind="python",
+            kind="python_v2" if gpu_marker is not None else "python",
             python_version=python_version
             or f"{sys.version_info.major}.{sys.version_info.minor}",
             environment=environment_spec,
             env=environment,
+            gpu=gpu_marker,
         )
         self._function = function
         self._request = FunctionRegistrationRequest(
@@ -989,6 +1197,7 @@ def udf(
     pip: tuple[str, ...] | list[str] = (),
     env: Optional[Mapping[str, str]] = None,
     python_version: Optional[str] = None,
+    gpu: bool = False,
     conda: tuple[str, ...] | list[str] = (),
     conda_channels: tuple[str, ...] | list[str] = (),
 ) -> Callable[[Callable[..., Any]], UdfDefinition]: ...
@@ -1003,6 +1212,7 @@ def udf(
     pip: tuple[str, ...] | list[str] = (),
     env: Optional[Mapping[str, str]] = None,
     python_version: Optional[str] = None,
+    gpu: bool = False,
     conda: tuple[str, ...] | list[str] = (),
     conda_channels: tuple[str, ...] | list[str] = (),
 ):
@@ -1035,6 +1245,10 @@ def udf(
         Environment variables included in the Function definition.
     python_version : str, optional
         Remote Python major/minor version. Defaults to the client version.
+    gpu : bool, default False
+        Whether every remote execution requires a GPU. The execution platform
+        selects one compatible GPU for each worker. The requirement is part of
+        the immutable Function version.
 
     The packaged artifact is a snapshot: the function source plus exactly
     the module-level names it references (modules as imports, importable
@@ -1059,6 +1273,11 @@ def udf(
     ...     return value * 2
     >>> score(1.5)
     3.0
+    >>> @udf(pip=["cupy-cuda12x"], gpu=True)
+    ... def gpu_score(value: int) -> int:
+    ...     return value * 2
+    >>> gpu_score.registration_request.runtime.gpu
+    True
     """
 
     def decorate(target: Callable[..., Any]) -> UdfDefinition:
@@ -1070,6 +1289,7 @@ def udf(
             pip=tuple(pip),
             env={} if env is None else env,
             python_version=python_version,
+            gpu=gpu,
             conda=tuple(conda),
             conda_channels=tuple(conda_channels),
         )

python/python/lancedb/remote/table.py

@@ -67,7 +67,15 @@ from ..query import (
     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
 
 

python/python/lancedb/table.py

@@ -2165,9 +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
@@ -6268,8 +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/test_first_class_function_slice2.py

@@ -19,7 +19,13 @@ import pyarrow as pa
 import pytest
 
 import lancedb
-from lancedb.functions import UdfDefinition, udf
+from lancedb.functions import (
+    PythonRuntimeSpec,
+    UdfDefinition,
+    _canonical_arrow_type,
+    _GRAMMAR_PRIMITIVES,
+    udf,
+)
 
 THRESHOLD = 20
 _CACHE = None
@@ -89,6 +95,58 @@ def test_udf_conda_environment():
         udf(name="channels", conda_channels=["conda-forge"])(lambda value: value)
 
 
+def test_udf_gpu_marker_uses_gpu_runtime():
+    @udf(pip=["cupy-cuda12x"], gpu=True)
+    def double_on_gpu(value: int) -> int:
+        return value * 2
+
+    request = json.loads(double_on_gpu.registration_request.to_canonical_json())
+    assert request["runtime"]["kind"] == "python_v2"
+    assert request["runtime"]["gpu"] is True
+
+    @udf(pip=["pyarrow"])
+    def cpu_function(value: int) -> int:
+        return value
+
+    cpu_runtime = json.loads(cpu_function.registration_request.to_canonical_json())[
+        "runtime"
+    ]
+    assert cpu_runtime["kind"] == "python"
+    assert "gpu" not in cpu_runtime
+
+    def identity(value: int) -> int:
+        return value
+
+    for invalid in [None, 0, 1, -1, 1.5, "", "true", "1", "H100"]:
+        with pytest.raises(ValueError, match="gpu must be a boolean"):
+            udf(name="invalid_gpu", gpu=invalid)(identity)
+
+    base_runtime = {
+        "kind": "python_v2",
+        "python_version": "3.12",
+        "environment": {"kind": "pip"},
+    }
+    runtime = PythonRuntimeSpec.model_validate({**base_runtime, "gpu": True})
+    assert runtime.gpu is True
+    for invalid in [False, 1, 0, "", "true", "1", "H100"]:
+        with pytest.raises(ValueError, match="runtime.gpu must be true"):
+            PythonRuntimeSpec.model_validate({**base_runtime, "gpu": invalid})
+
+
+def test_unknown_runtime_discards_payload_before_known_field_validation():
+    for payload in [
+        {"kind": "python_v3", "gpu": {"model": "H100"}},
+        {"kind": "python_v3", "resources": []},
+        {
+            "kind": "python_v3",
+            "environment": {"kind": []},
+            "python_version": 3.15,
+        },
+    ]:
+        runtime = PythonRuntimeSpec.model_validate(payload)
+        assert runtime.to_canonical_json() == '{"kind":"python_v3"}'
+
+
 def test_udf_packages_attribute_access_and_body_imports():
     @udf
     def word_norm(body: str) -> float:
@@ -168,9 +226,7 @@ def test_udf_resolves_module_globals_before_builtins(tmp_path):
         udf(module.uses_callable_shadow)
 
 
-def test_canonical_arrow_type_is_exactly_the_grammar():
-    from lancedb.functions import _GRAMMAR_PRIMITIVES, _canonical_arrow_type
-
+def test_canonical_arrow_type_prefers_the_compact_grammar():
     golden = json.loads(
         (
             Path(__file__).parents[3]
@@ -181,14 +237,19 @@ def test_canonical_arrow_type_is_exactly_the_grammar():
         case["arrow_type"] for case in golden["valid"] if "<" not in case["arrow_type"]
     ]
     assert [name for _, name in _GRAMMAR_PRIMITIVES] == primitives
+    assert _canonical_arrow_type(pa.list_(pa.field("item", pa.float32(), False))) == (
+        "list<float32>"
+    )
+    assert (
+        _canonical_arrow_type(pa.large_list(pa.field("item", pa.float32(), False)))
+        == "large_list<float32>"
+    )
     for outside in [
         pa.timestamp("us"),
         pa.decimal128(10, 2),
-        pa.large_string(),
         pa.large_binary(),
         pa.binary(4),
         pa.duration("s"),
-        pa.struct([pa.field("a", pa.int32())]),
         pa.list_(pa.float32(), 0),
         pa.list_(pa.timestamp("us")),
     ]:
@@ -378,14 +439,27 @@ def test_udf_recursion_versus_a_rebound_module_name(tmp_path):
         udf(raw_fact)
 
 
-def test_canonical_arrow_type_rejects_unrepresentable_list_children():
-    from lancedb.functions import _canonical_arrow_type
-
+def test_canonical_arrow_type_uses_exact_json_for_list_child_properties():
+    nullable = pa.list_(pa.float32())
+    assert json.loads(_canonical_arrow_type(nullable)) == {
+        "type": "list",
+        "fields": [
+            {
+                "name": "item",
+                "nullable": True,
+                "type": {"type": "float32"},
+            }
+        ],
+    }
+    named = pa.list_(pa.field("custom", pa.float32(), nullable=False))
+    assert json.loads(_canonical_arrow_type(named))["fields"][0]["name"] == "custom"
     for outside in [
-        pa.list_(pa.float32()),  # pyarrow default: nullable child
-        pa.list_(pa.field("custom", pa.float32(), nullable=False)),
         pa.list_(pa.field("item", pa.float32(), nullable=False, metadata={"k": "v"})),
         pa.list_(pa.field("item", pa.float32(), nullable=False), 0),
+        pa.list_(
+            pa.field("item", pa.float32(), nullable=False, metadata={"k": "v"}), 3
+        ),
+        pa.list_(pa.field("custom", pa.float32(), nullable=False), 3),
     ]:
         with pytest.raises(TypeError, match="unsupported Arrow type"):
             _canonical_arrow_type(outside)
@@ -395,6 +469,29 @@ def test_canonical_arrow_type_rejects_unrepresentable_list_children():
         )
         == "fixed_size_list<float32, 3>"
     )
+    fixed = json.loads(_canonical_arrow_type(pa.list_(pa.float32(), 3)))
+    assert fixed == {
+        "type": "fixed_size_list",
+        "fields": [
+            {
+                "name": "item",
+                "nullable": True,
+                "type": {"type": "float32"},
+            }
+        ],
+        "length": 3,
+    }
+    large = json.loads(_canonical_arrow_type(pa.large_list(pa.float32())))
+    assert large["type"] == "large_list"
+    assert large["fields"][0]["nullable"] is True
+
+    for invalid_struct in [
+        pa.struct([]),
+        pa.struct([pa.field("a", pa.int32()), pa.field("a", pa.int64())]),
+        pa.struct([pa.field("", pa.int32())]),
+    ]:
+        with pytest.raises(TypeError, match="unsupported Arrow type"):
+            _canonical_arrow_type(invalid_struct)
 
 
 def _calls_missing(value: int) -> int:
@@ -432,6 +529,7 @@ def _arrow_type_from_golden(spec: dict) -> pa.DataType:
         "null": pa.null(),
         "bool": pa.bool_(),
         "utf8": pa.string(),
+        "large_utf8": pa.large_string(),
         "binary": pa.binary(),
         "float16": pa.float16(),
         "float32": pa.float32(),
@@ -448,8 +546,6 @@ def test_arrow_type_grammar_matches_the_shared_golden():
             / "rust/lancedb/tests/fixtures/first_class_functions/v1/arrow_types.json"
         ).read_text()
     )
-    from lancedb.functions import _canonical_arrow_type
-
     emitted = {
         case["arrow_type"]: _canonical_arrow_type(_arrow_type_from_golden(case["json"]))
         for case in golden["valid"]
@@ -482,6 +578,250 @@ def test_explicit_arrow_schema_is_deterministic():
     assert signature.output.nullable is False
 
 
+def test_blob_fields_use_the_scalar_function_semantic_type():
+    @udf(
+        input_schema=pa.schema([lancedb.blob("image", nullable=False)]),
+        output_schema=lancedb.blob("result", nullable=False),
+    )
+    def copy_blob(image):
+        return image
+
+    signature = copy_blob.registration_request.signature
+    assert signature.inputs[0].arrow_type == "blob_v2"
+    assert signature.output.kind == "scalar"
+    assert signature.output.arrow_type == "blob_v2"
+
+
+def test_named_struct_function_can_include_a_blob_result_field():
+    @udf(
+        input_schema=pa.schema([lancedb.blob("image", nullable=False)]),
+        output_schema=pa.schema(
+            [
+                lancedb.blob("thumbnail", nullable=False),
+                pa.field("width", pa.int32(), nullable=False),
+            ]
+        ),
+    )
+    def inspect_blob(image):
+        return {"thumbnail": image, "width": 1}
+
+    output = inspect_blob.registration_request.signature.output
+    assert output.kind == "named_struct"
+    assert [(field.name, field.arrow_type) for field in output.fields] == [
+        ("thumbnail", "blob_v2"),
+        ("width", "int32"),
+    ]
+
+
+def test_metadata_marked_blob_field_uses_the_semantic_type():
+    extension = lancedb.blob("image", nullable=False).type
+    storage = (
+        extension.storage_type if isinstance(extension, pa.ExtensionType) else extension
+    )
+    metadata_blob = pa.field(
+        "image",
+        storage,
+        nullable=False,
+        metadata={"ARROW:extension:name": "lance.blob.v2"},
+    )
+
+    @udf(
+        input_schema=pa.schema([metadata_blob]),
+        output_schema=pa.field("size", pa.int64(), nullable=False),
+    )
+    def blob_size(image):
+        return len(image)
+
+    assert blob_size.registration_request.signature.inputs[0].arrow_type == "blob_v2"
+
+
+def test_blob_marker_rejects_invalid_storage_layout():
+    malformed = pa.field(
+        "image",
+        pa.int64(),
+        nullable=False,
+        metadata={"ARROW:extension:name": "lance.blob.v2"},
+    )
+
+    with pytest.raises(TypeError, match="requires a supported Blob storage layout"):
+
+        @udf(
+            input_schema=pa.schema([malformed]),
+            output_schema=pa.field("size", pa.int64(), nullable=False),
+        )
+        def blob_size(image):
+            return len(image)
+
+
+def test_nested_blob_signature_field_has_a_clear_error():
+    nested = pa.field(
+        "value",
+        pa.struct([lancedb.blob("image", nullable=False)]),
+        nullable=False,
+    )
+    with pytest.raises(TypeError, match="nested Blob v2 fields are not supported"):
+
+        @udf(
+            input_schema=pa.schema([nested]),
+            output_schema=pa.field("size", pa.int64(), nullable=False),
+        )
+        def blob_size(value):
+            return len(value["image"])
+
+
+def test_nested_non_blob_extension_is_not_silently_unwrapped():
+    class TestExtension(pa.ExtensionType):
+        def __init__(self):
+            super().__init__(pa.int64(), "test.function.extension")
+
+        def __arrow_ext_serialize__(self):
+            return b""
+
+        @classmethod
+        def __arrow_ext_deserialize__(cls, storage_type, serialized):
+            return cls()
+
+    nested = pa.field(
+        "value",
+        pa.struct([pa.field("extended", TestExtension(), nullable=False)]),
+        nullable=False,
+    )
+    with pytest.raises(TypeError, match="unsupported Arrow type"):
+
+        @udf(
+            input_schema=pa.schema([nested]),
+            output_schema=pa.field("result", pa.int64(), nullable=False),
+        )
+        def extension_value(value):
+            return value["extended"]
+
+
+def test_explicit_large_utf8_schemas_use_the_canonical_function_name():
+    input_schema = pa.schema([pa.field("text", pa.large_string(), nullable=True)])
+    output_schema = pa.field("result", pa.large_string(), nullable=False)
+
+    @udf(input_schema=input_schema, output_schema=output_schema)
+    def preserve(text):
+        return text
+
+    signature = preserve.registration_request.signature
+    assert signature.inputs[0].arrow_type == "large_utf8"
+    assert signature.inputs[0].nullable is True
+    assert signature.output.arrow_type == "large_utf8"
+    assert signature.output.nullable is False
+
+    nested = pa.struct([pa.field("text", pa.large_string(), nullable=True)])
+    assert json.loads(_canonical_arrow_type(nested)) == {
+        "type": "struct",
+        "fields": [
+            {
+                "name": "text",
+                "nullable": True,
+                "type": {"type": "large_utf8"},
+            }
+        ],
+    }
+
+
+def test_nested_struct_output_uses_canonical_exact_json():
+    token = pa.struct(
+        [
+            pa.field("position", pa.int32(), nullable=False),
+            pa.field("value", pa.string(), nullable=False),
+            pa.field("length", pa.int32(), nullable=False),
+        ]
+    )
+    analysis = pa.struct(
+        [
+            pa.field("normalized_text", pa.string(), nullable=False),
+            pa.field("has_content", pa.bool_(), nullable=False),
+            pa.field(
+                "metrics",
+                pa.struct(
+                    [
+                        pa.field("character_count", pa.int64(), nullable=False),
+                        pa.field("word_count", pa.int32(), nullable=False),
+                        pa.field("average_word_length", pa.float64(), nullable=False),
+                    ]
+                ),
+                nullable=False,
+            ),
+            pa.field(
+                "diagnostics",
+                pa.struct(
+                    [
+                        pa.field("status", pa.string(), nullable=False),
+                        pa.field(
+                            "normalization",
+                            pa.struct(
+                                [
+                                    pa.field("changed", pa.bool_(), nullable=False),
+                                    pa.field(
+                                        "original_length", pa.int64(), nullable=False
+                                    ),
+                                ]
+                            ),
+                            nullable=False,
+                        ),
+                    ]
+                ),
+                nullable=False,
+            ),
+            pa.field(
+                "token_preview",
+                pa.list_(pa.field("item", token, nullable=False)),
+                nullable=False,
+            ),
+        ]
+    )
+
+    @udf(
+        input_schema=pa.schema([pa.field("text", pa.string(), nullable=False)]),
+        output_schema=pa.field("analysis", analysis, nullable=False),
+    )
+    def analyze(text):
+        return {"normalized_text": text}
+
+    output = analyze.registration_request.signature.output
+    assert output.kind == "named_struct"
+    assert [field.name for field in output.fields] == [
+        "normalized_text",
+        "has_content",
+        "metrics",
+        "diagnostics",
+        "token_preview",
+    ]
+    metrics = json.loads(output.fields[2].arrow_type)
+    assert metrics == {
+        "type": "struct",
+        "fields": [
+            {
+                "name": "character_count",
+                "nullable": False,
+                "type": {"type": "int64"},
+            },
+            {
+                "name": "word_count",
+                "nullable": False,
+                "type": {"type": "int32"},
+            },
+            {
+                "name": "average_word_length",
+                "nullable": False,
+                "type": {"type": "float64"},
+            },
+        ],
+    }
+    preview = json.loads(output.fields[4].arrow_type)
+    assert preview["type"] == "list"
+    assert preview["fields"][0]["type"]["type"] == "struct"
+    assert [field["name"] for field in preview["fields"][0]["type"]["fields"]] == [
+        "position",
+        "value",
+        "length",
+    ]
+
+
 def test_annotation_and_explicit_schema_validation_fail_closed():
     with pytest.raises(TypeError, match="missing Function annotations"):
 
@@ -525,6 +865,72 @@ def test_annotation_and_explicit_schema_validation_fail_closed():
         def nullable_explicit(value):
             return value
 
+    for invalid_field in [
+        pa.field("", pa.int32(), nullable=False),
+        pa.field("result", pa.int32(), nullable=False, metadata={"k": "v"}),
+    ]:
+        with pytest.raises(TypeError, match="unsupported Arrow type"):
+
+            @udf(
+                input_schema=pa.schema([pa.field("value", pa.int64())]),
+                output_schema=pa.schema([invalid_field]),
+            )
+            def invalid_explicit_field(value):
+                return value
+
+    with pytest.raises(TypeError, match="unsupported Arrow type"):
+
+        @udf(
+            input_schema=pa.schema(
+                [pa.field("value", pa.int64(), metadata={"k": "v"})]
+            ),
+            output_schema=pa.int64(),
+        )
+        def input_field_metadata(value):
+            return value
+
+    with pytest.raises(TypeError, match="unsupported Arrow type"):
+
+        @udf(
+            input_schema=pa.schema([pa.field("value", pa.int64())]),
+            output_schema=pa.field(
+                "result", pa.int64(), nullable=False, metadata={"k": "v"}
+            ),
+        )
+        def scalar_output_field_metadata(value):
+            return value
+
+    struct_type = pa.struct([pa.field("value", pa.int64(), nullable=False)])
+    with pytest.raises(TypeError, match="unsupported Arrow type"):
+
+        @udf(
+            input_schema=pa.schema([pa.field("value", pa.int64())]),
+            output_schema=pa.field(
+                "result", struct_type, nullable=False, metadata={"k": "v"}
+            ),
+        )
+        def struct_output_field_metadata(value):
+            return {"value": value}
+
+    for input_schema, output_schema in [
+        (
+            pa.schema([pa.field("value", pa.int64())], metadata={"k": "v"}),
+            pa.int64(),
+        ),
+        (
+            pa.schema([pa.field("value", pa.int64())]),
+            pa.schema(
+                [pa.field("result", pa.int64(), nullable=False)],
+                metadata={"k": "v"},
+            ),
+        ),
+    ]:
+        with pytest.raises(TypeError, match="schema metadata"):
+
+            @udf(input_schema=input_schema, output_schema=output_schema)
+            def schema_metadata(value):
+                return value
+
 
 def test_local_function_catalog_operations_are_not_supported(tmp_path):
     db = lancedb.connect(tmp_path)

python/python/tests/test_table.py

@@ -4087,6 +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)

rust/lancedb/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "lancedb"
-version = "0.38.0-beta.11"
+version = "0.38.0"
 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::{ReadDirOptions, StorageOptionsAccessor, StorageOptionsProvider};
+use lance_io::object_store::{StorageOptionsAccessor, StorageOptionsProvider};
 use lance_table::io::commit::commit_handler_from_url;
 use object_store::local::LocalFileSystem;
 use snafu::ResultExt;
@@ -282,14 +282,11 @@ 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.
+/// The table a listed child directory holds, or `None` if it is not a table at all.
 ///
 /// 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()?
@@ -297,6 +294,75 @@ fn table_name(location: &object_store::path::Path, dir_suffix: &str) -> Option<S
         .map(String::from)
         .filter(|name| !name.is_empty())
 }
+
+/// One page of the table directories under the database directory, in key order.
+struct DirPage {
+    /// The table directories the page holds, as the store lists them.
+    common_prefixes: Vec<object_store::path::Path>,
+    /// Resumes after this page, or `None` when the page reached the end of the level.
+    page_token: Option<String>,
+}
+
+/// Where a listed location sits inside the database directory — the space page tokens live
+/// in — or `None` if it is not a child of that directory at all. Matching both halves of the
+/// prefix drops a location that merely starts with the directory's name (`dbx/y` against
+/// `db/`) as well as the marker object some stores keep for the directory itself.
+fn relative_key<'a>(prefix: Option<&str>, location: &'a str) -> Option<&'a str> {
+    let relative = match prefix {
+        Some(prefix) => location.strip_prefix(prefix)?,
+        None => location,
+    };
+    (!relative.is_empty()).then_some(relative)
+}
+
+/// One page of the table directories under `base_path`, one directory level deep.
+///
+/// Lance 11 exposes no paginated directory listing, so the level is listed in full and paged
+/// locally: table directories go into key order (a directory's key keeps its trailing `/`,
+/// so a token is never a table name), the page is the smallest `limit` of them past
+/// `page_token`, and the token handed back is the key of the last directory the page took —
+/// so a page that took nothing ends the listing rather than resuming from a position no page
+/// ever reached. Only `<name>.lance/` directories enter the page: loose objects, other
+/// directories, and a bare `.lance/` never take a page slot or name a token, which keeps a
+/// page to exactly one listing of the level. Correct on every store, at the cost of that one
+/// full-level listing per page.
+async fn read_dir_page(
+    object_store: &ObjectStore,
+    base_path: &object_store::path::Path,
+    page_token: Option<String>,
+    limit: Option<usize>,
+) -> Result<DirPage> {
+    let listed = object_store.list_with_delimiter(Some(base_path)).await?;
+    let prefix = {
+        let base = base_path.as_ref();
+        (!base.is_empty()).then(|| format!("{base}/"))
+    };
+    let table_dir_suffix = format!(".{LANCE_EXTENSION}/");
+    let mut children: Vec<(String, object_store::path::Path)> = listed
+        .common_prefixes
+        .into_iter()
+        .filter_map(|location| {
+            let key = format!("{}/", relative_key(prefix.as_deref(), location.as_ref())?);
+            (key.len() > table_dir_suffix.len() && key.ends_with(&table_dir_suffix))
+                .then_some((key, location))
+        })
+        .collect();
+    children.sort_unstable_by(|(left, _), (right, _)| left.cmp(right));
+    if let Some(resume) = &page_token {
+        children.retain(|(key, _)| key > resume);
+    }
+    let total = children.len();
+    children.truncate(limit.unwrap_or(total).min(total));
+    let page_token = match children.last() {
+        Some((last, _)) if children.len() < total => Some(last.clone()),
+        _ => None,
+    };
+    Ok(DirPage {
+        common_prefixes: children.into_iter().map(|(_, location)| location).collect(),
+        page_token,
+    })
+}
+
 const ENGINE: &str = "engine";
 const MIRRORED_STORE: &str = "mirroredStore";
 
@@ -982,8 +1048,7 @@ impl Database for ListingDatabase {
         let mut tables = Vec::new();
         let mut page_token = request.page_token.filter(|token| !token.is_empty());
 
-        // A page of nothing: the store rejects a limit of zero, and no table was handed over
-        // for a token to resume after.
+        // A page of nothing: no table was handed over for a token to resume after.
         if limit == Some(0) {
             return Ok(ListTablesResponse {
                 context: None,
@@ -992,35 +1057,21 @@ impl Database for ListingDatabase {
             });
         }
 
-        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;
-            }
-        }
+        // The page holds only table directories, so one call — and the one full-level
+        // listing behind it — fills it.
+        let page = read_dir_page(
+            &self.object_store,
+            &self.base_path,
+            page_token.take(),
+            limit,
+        )
+        .await?;
+        page_token = page.page_token;
+        tables.extend(
+            page.common_prefixes
+                .iter()
+                .filter_map(|location| table_name(location, &dir_suffix)),
+        );
 
         Ok(ListTablesResponse {
             context: None,
@@ -1666,8 +1717,8 @@ mod tests {
     }
 
     /// 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.
+    /// under the database prefix are not. They never take a page slot, so even a `limit`
+    /// smaller than the clutter ahead of the first table returns that table.
     #[tokio::test]
     async fn test_listing_ignores_non_table_children() {
         let (tempdir, db) = setup_database().await;
@@ -1686,6 +1737,37 @@ mod tests {
         assert_eq!(page.tables, vec!["real"]);
     }
 
+    /// The Lance 11 fallback pages locally over one full-level listing, so a bounded page
+    /// costs exactly one listing call — clutter ahead of the first table must not buy extra
+    /// round trips.
+    #[tokio::test]
+    async fn test_one_full_listing_per_public_page() {
+        use crate::io::object_store::io_tracking::IoStatsHolder;
+        use lance_io::object_store::WrappingObjectStore;
+
+        let (tempdir, mut 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 io_stats = IoStatsHolder::default();
+        let mut tracked_store = (*db.object_store).clone();
+        tracked_store.inner =
+            io_stats.wrap(&tracked_store.store_prefix, tracked_store.inner.clone());
+        db.object_store = Arc::new(tracked_store);
+
+        let page = db
+            .list_tables(ListTablesRequest {
+                limit: Some(1),
+                ..Default::default()
+            })
+            .await
+            .unwrap();
+
+        assert_eq!(page.tables, vec!["real"]);
+        assert_eq!(io_stats.incremental_stats().read_iops, 1);
+    }
+
     #[tokio::test]
     async fn listing_ignores_empty_table_name() {
         let (tempdir, db) = setup_database().await;

rust/lancedb/src/function.rs

@@ -15,6 +15,9 @@ use serde_json::Value;
 
 use crate::{Error, Result};
 
+/// Semantic Function type for a Blob v2 value.
+pub const FUNCTION_BLOB_V2_TYPE: &str = "blob_v2";
+
 fn invalid_json(error: impl std::fmt::Display) -> Error {
     Error::InvalidInput {
         message: format!("invalid remote Function JSON: {error}"),
@@ -207,6 +210,33 @@ pub enum PythonRuntimeSpec {
         environment: PythonEnvironmentSpec,
         env: BTreeMap<String, String>,
     },
+    /// The GPU-enabled Sophon-managed Python runtime.
+    ///
+    /// # Examples
+    ///
+    /// ```
+    /// use std::collections::BTreeMap;
+    /// use lancedb::function::{PythonEnvironmentSpec, PythonRuntimeSpec};
+    ///
+    /// let runtime = PythonRuntimeSpec::PythonV2 {
+    ///     python_version: "3.12".to_string(),
+    ///     environment: PythonEnvironmentSpec {
+    ///         kind: "pip".to_string(),
+    ///         packages: vec!["cupy-cuda12x".to_string()],
+    ///         channels: Vec::new(),
+    ///         path: None,
+    ///         modules: Vec::new(),
+    ///         image: None,
+    ///     },
+    ///     env: BTreeMap::new(),
+    /// };
+    /// assert!(runtime.requires_gpu());
+    /// ```
+    PythonV2 {
+        python_version: String,
+        environment: PythonEnvironmentSpec,
+        env: BTreeMap<String, String>,
+    },
     /// A runtime kind introduced by a newer server.
     ///
     /// Unknown payload fields are intentionally not retained because the
@@ -219,22 +249,27 @@ impl PythonRuntimeSpec {
     pub fn kind(&self) -> &str {
         match self {
             Self::Python { .. } => "python",
+            Self::PythonV2 { .. } => "python_v2",
             Self::Unrecognized { kind } => kind,
         }
     }
 
-    /// The Python version for the V1 runtime, or `None` for an unknown kind.
+    /// The Python version for a known Python runtime, or `None` for an unknown kind.
     pub fn python_version(&self) -> Option<&str> {
         match self {
-            Self::Python { python_version, .. } => Some(python_version),
+            Self::Python { python_version, .. } | Self::PythonV2 { python_version, .. } => {
+                Some(python_version)
+            }
             Self::Unrecognized { .. } => None,
         }
     }
 
-    /// The Python environment for the V1 runtime, or `None` for an unknown kind.
+    /// The Python environment for a known Python runtime, or `None` for an unknown kind.
     pub fn environment(&self) -> Option<&PythonEnvironmentSpec> {
         match self {
-            Self::Python { environment, .. } => Some(environment),
+            Self::Python { environment, .. } | Self::PythonV2 { environment, .. } => {
+                Some(environment)
+            }
             Self::Unrecognized { .. } => None,
         }
     }
@@ -242,38 +277,73 @@ impl PythonRuntimeSpec {
     /// Environment variables, or `None` for an unknown kind.
     pub fn env(&self) -> Option<&BTreeMap<String, String>> {
         match self {
-            Self::Python { env, .. } => Some(env),
+            Self::Python { env, .. } | Self::PythonV2 { env, .. } => Some(env),
             Self::Unrecognized { .. } => None,
         }
     }
+
+    /// Whether the runtime requires a GPU selected by the execution platform.
+    pub fn requires_gpu(&self) -> bool {
+        matches!(self, Self::PythonV2 { .. })
+    }
 }
 
 #[derive(Deserialize)]
-struct PythonRuntimeWire {
-    kind: String,
-    #[serde(default)]
-    python_version: Option<String>,
-    #[serde(default)]
-    environment: Option<PythonEnvironmentSpec>,
+struct PythonRuntimeV1Wire {
+    python_version: String,
+    environment: PythonEnvironmentSpec,
     #[serde(default)]
     env: BTreeMap<String, String>,
+    #[serde(default)]
+    gpu: Option<Value>,
+}
+
+#[derive(Deserialize)]
+struct PythonRuntimeV2Wire {
+    python_version: String,
+    environment: PythonEnvironmentSpec,
+    #[serde(default)]
+    env: BTreeMap<String, String>,
+    gpu: bool,
 }
 
 impl<'de> Deserialize<'de> for PythonRuntimeSpec {
     fn deserialize<D: Deserializer<'de>>(deserializer: D) -> std::result::Result<Self, D::Error> {
-        let wire = PythonRuntimeWire::deserialize(deserializer)?;
-        if wire.kind == "python" {
-            Ok(Self::Python {
-                python_version: wire
-                    .python_version
-                    .ok_or_else(|| de::Error::missing_field("python_version"))?,
-                environment: wire
-                    .environment
-                    .ok_or_else(|| de::Error::missing_field("environment"))?,
-                env: wire.env,
-            })
-        } else {
-            Ok(Self::Unrecognized { kind: wire.kind })
+        let value = Value::deserialize(deserializer)?;
+        let kind = value
+            .get("kind")
+            .ok_or_else(|| de::Error::missing_field("kind"))?
+            .as_str()
+            .ok_or_else(|| de::Error::custom("runtime.kind must be a string"))?
+            .to_string();
+        match kind.as_str() {
+            "python" => {
+                let wire: PythonRuntimeV1Wire =
+                    serde_json::from_value(value).map_err(de::Error::custom)?;
+                if wire.gpu.is_some() {
+                    return Err(de::Error::custom(
+                        "python runtime with gpu requires kind='python_v2'",
+                    ));
+                }
+                Ok(Self::Python {
+                    python_version: wire.python_version,
+                    environment: wire.environment,
+                    env: wire.env,
+                })
+            }
+            "python_v2" => {
+                let wire: PythonRuntimeV2Wire =
+                    serde_json::from_value(value).map_err(de::Error::custom)?;
+                if !wire.gpu {
+                    return Err(de::Error::custom("runtime.gpu must be true"));
+                }
+                Ok(Self::PythonV2 {
+                    python_version: wire.python_version,
+                    environment: wire.environment,
+                    env: wire.env,
+                })
+            }
+            _ => Ok(Self::Unrecognized { kind }),
         }
     }
 }
@@ -287,6 +357,8 @@ impl Serialize for PythonRuntimeSpec {
             environment: &'a PythonEnvironmentSpec,
             #[serde(skip_serializing_if = "BTreeMap::is_empty")]
             env: &'a BTreeMap<String, String>,
+            #[serde(skip_serializing_if = "Option::is_none")]
+            gpu: Option<bool>,
         }
 
         #[derive(Serialize)]
@@ -304,6 +376,19 @@ impl Serialize for PythonRuntimeSpec {
                 python_version,
                 environment,
                 env,
+                gpu: None,
+            }
+            .serialize(serializer),
+            Self::PythonV2 {
+                python_version,
+                environment,
+                env,
+            } => PythonRuntimeRef {
+                kind: "python_v2",
+                python_version,
+                environment,
+                env,
+                gpu: Some(true),
             }
             .serialize(serializer),
             Self::Unrecognized { kind } => UnrecognizedRuntimeRef { kind }.serialize(serializer),
@@ -313,8 +398,8 @@ impl Serialize for PythonRuntimeSpec {
 
 /// Immutable Function version returned by the Enterprise catalog.
 ///
-/// Scheduling resources, priority, concurrency, and retry policy belong to
-/// the submitting Job and are not part of this identity.
+/// The GPU execution requirement is part of this identity. CPU and memory sizing,
+/// priority, concurrency, and retry policy belong to the execution platform.
 #[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
 pub struct FunctionVersion {
     name: String,
@@ -589,7 +674,7 @@ impl_json!(RefreshColumnResult);
 
 #[cfg(test)]
 mod conda_environment_tests {
-    use super::PythonEnvironmentSpec;
+    use super::{PythonEnvironmentSpec, PythonRuntimeSpec};
 
     #[test]
     fn conda_channels_round_trip_and_pip_stays_bare() {
@@ -608,4 +693,45 @@ mod conda_environment_tests {
             serde_json::from_str(r#"{"kind":"pip","packages":["numpy"]}"#).unwrap();
         assert!(!serde_json::to_string(&pip).unwrap().contains("channels"));
     }
+
+    #[test]
+    fn gpu_python_runtime_marker_round_trips_and_validates() {
+        let runtime: PythonRuntimeSpec = serde_json::from_str(
+            r#"{"kind":"python_v2","python_version":"3.12","environment":{"kind":"pip"},"gpu":true}"#,
+        )
+        .unwrap();
+        assert_eq!(runtime.kind(), "python_v2");
+        assert!(runtime.requires_gpu());
+        assert_eq!(
+            super::canonical_json(&runtime).unwrap(),
+            r#"{"environment":{"kind":"pip"},"gpu":true,"kind":"python_v2","python_version":"3.12"}"#
+        );
+
+        for invalid in [
+            r#"{"kind":"python","python_version":"3.12","environment":{"kind":"pip"},"gpu":true}"#,
+            r#"{"kind":"python_v2","python_version":"3.12","environment":{"kind":"pip"}}"#,
+            r#"{"kind":"python_v2","python_version":"3.12","environment":{"kind":"pip"},"gpu":1}"#,
+            r#"{"kind":"python_v2","python_version":"3.12","environment":{"kind":"pip"},"gpu":false}"#,
+            r#"{"kind":"python_v2","python_version":"3.12","environment":{"kind":"pip"},"gpu":"true"}"#,
+            r#"{"kind":"python_v2","python_version":"3.12","environment":{"kind":"pip"},"gpu":"H100"}"#,
+        ] {
+            assert!(serde_json::from_str::<PythonRuntimeSpec>(invalid).is_err());
+        }
+    }
+
+    #[test]
+    fn unknown_runtime_discards_payload_before_known_field_validation() {
+        for encoded in [
+            r#"{"kind":"python_v3","gpu":{"model":"H100"}}"#,
+            r#"{"kind":"python_v3","resources":[]}"#,
+            r#"{"kind":"python_v3","python_version":3.15,"environment":{"kind":[]}}"#,
+        ] {
+            let runtime: PythonRuntimeSpec = serde_json::from_str(encoded).unwrap();
+            assert_eq!(runtime.kind(), "python_v3");
+            assert_eq!(
+                super::canonical_json(&runtime).unwrap(),
+                r#"{"kind":"python_v3"}"#
+            );
+        }
+    }
 }

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, list::PaginatedListStore, path::Path,
+    UploadPart, path::Path,
 };
 
 use async_trait::async_trait;
@@ -187,14 +187,6 @@ 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, list::PaginatedListStore, path::Path,
+    UploadPart, path::Path,
 };
 
 #[derive(Debug, Default)]
@@ -57,14 +57,6 @@ 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/remote/table.rs

@@ -3180,8 +3180,8 @@ impl<S: HttpSend> BaseTable for RemoteTable<S> {
             self.schema().await?.as_ref(),
             "schema evolution",
         )?;
-        // The server plans the declaration: expression validation, type
-        // inference and the persisted binding all happen there.
+        // The server plans the declaration against its table schema, including
+        // Blob v2 semantics inherited by a direct field projection.
         let entries = columns
             .iter()
             .map(
@@ -5734,9 +5734,8 @@ mod tests {
             ))
             .execute()
             .await;
-        let err = match result {
-            Ok(_) => panic!("legacy remote query unexpectedly succeeded"),
-            Err(err) => err,
+        let Err(err) = result else {
+            panic!("legacy remote query unexpectedly succeeded")
         };
 
         assert!(
@@ -7388,8 +7387,8 @@ mod tests {
         assert_eq!(result.version, if old_server { 0 } else { 43 });
     }
 
-    /// A declaration is sent as `{name, computed}` entries for the server to
-    /// plan; the client never types the expression itself.
+    /// A declaration is sent as `{name, computed}` for the server to plan; the
+    /// client never types the expression itself.
     #[tokio::test]
     async fn test_add_computed_columns_sends_the_expression() {
         let table = Table::new_with_handler("my_table", |request| match request.url().path() {
@@ -7465,6 +7464,93 @@ mod tests {
         assert_eq!(result.version, 8);
     }
 
+    #[tokio::test]
+    async fn test_add_function_column_allows_an_existing_binding() {
+        let binding = crate::function::FunctionBinding::from_json(include_str!(
+            "../../tests/fixtures/first_class_functions/v1/remote_function_binding.json"
+        ))
+        .unwrap();
+        let binding_metadata = crate::table::computed_columns::function_bindings_metadata(
+            std::slice::from_ref(&binding),
+        )
+        .unwrap();
+        let mut fields = vec![
+            Field::new("title", DataType::Utf8, true),
+            Field::new("body", DataType::Utf8, true),
+        ];
+        fields.extend(binding.outputs().iter().map(|output| {
+            let data_type = match output.arrow_type.as_str() {
+                "utf8" => DataType::Utf8,
+                "int64" => DataType::Int64,
+                other => panic!("unexpected fixture output type {other}"),
+            };
+            Field::new(&output.output_name, data_type, true).with_metadata(
+                crate::table::computed_columns::function_computed_column_metadata(
+                    binding.binding_id(),
+                    output.output_ordinal,
+                    &["title".into(), "body".into()],
+                ),
+            )
+        }));
+        let schema = Schema::new_with_metadata(
+            fields,
+            HashMap::from([(
+                crate::table::computed_columns::FUNCTION_BINDINGS_META_KEY.to_string(),
+                binding_metadata,
+            )]),
+        );
+        let table =
+            Table::new_with_handler("my_table", move |request| match request.url().path() {
+                "/v1/table/my_table/describe/" => http::Response::builder()
+                    .status(200)
+                    .body(describe_response(&schema))
+                    .unwrap(),
+                "/v1/table/my_table/add_columns/" => {
+                    let actual: serde_json::Value =
+                        serde_json::from_slice(request.body().unwrap().as_bytes().unwrap())
+                            .unwrap();
+                    assert_eq!(
+                        actual["new_columns"],
+                        serde_json::json!([
+                            {"name":"secondary_text","all_null":true},
+                            {"name":"secondary_token_count","all_null":true}
+                        ])
+                    );
+                    http::Response::builder()
+                        .status(200)
+                        .body(r#"{"version":10}"#.to_string())
+                        .unwrap()
+                }
+                path => panic!("Unexpected path: {path}"),
+            });
+        let application = crate::function::FunctionApplication::from_json(
+            r#"{
+                "function":{"name":"text_features","version":"fv_01K3TEXT"},
+                "inputs":[
+                    {"parameter":"title","kind":"column","value":{"path":"title"}},
+                    {"parameter":"body","kind":"column","value":{"path":"body"}}
+                ],
+                "output":{"kind":"named_struct","fields":[
+                    {"name":"normalized_text","arrow_type":"utf8","nullable":false},
+                    {"name":"token_count","arrow_type":"int64","nullable":false}
+                ]},
+                "columns":{
+                    "normalized_text":"secondary_text",
+                    "token_count":"secondary_token_count"
+                }
+            }"#,
+        )
+        .unwrap();
+
+        let result = table
+            .add_columns()
+            .function(application)
+            .execute()
+            .await
+            .unwrap();
+        assert_eq!(result.version, 10);
+    }
+
     #[tokio::test]
     async fn test_add_fixed_size_list_function_column_declares_the_vector_type() {
         let table = Table::new_with_handler("my_table", |request| {

rust/lancedb/src/table.rs

@@ -750,8 +750,8 @@ pub trait BaseTable: std::fmt::Display + std::fmt::Debug + Send + Sync {
     /// Declare computed columns, each defined by a SQL expression.
     ///
     /// Where the declaration is planned depends on the backend: a local table
-    /// validates and types the expression itself, a remote one sends the text
-    /// for the server to plan.
+    /// validates and types the expression itself, while a remote one sends the
+    /// expression for the server to plan.
     async fn add_computed_columns(
         &self,
         _columns: &[(String, String)],
@@ -4183,14 +4183,6 @@ mod tests {
                 parent_list_calls: self.parent_list_calls.clone(),
             })
         }
-
-        fn wrap_paginated(
-            &self,
-            _store_prefix: &str,
-            _original: Arc<dyn object_store::list::PaginatedListStore>,
-        ) -> Option<Arc<dyn object_store::list::PaginatedListStore>> {
-            None
-        }
     }
 
     #[tokio::test]
@@ -4294,14 +4286,6 @@ mod tests {
             self.called.store(true, Ordering::Relaxed);
             original
         }
-
-        fn wrap_paginated(
-            &self,
-            _store_prefix: &str,
-            original: Arc<dyn object_store::list::PaginatedListStore>,
-        ) -> Option<Arc<dyn object_store::list::PaginatedListStore>> {
-            Some(original)
-        }
     }
 
     #[tokio::test]
@@ -5763,6 +5747,13 @@ mod tests {
         assert!(index_bytes > 0);
         assert_eq!(with_index, data_only + index_bytes);
 
+        // Release builds reject unstable overlay datasets unless explicitly opted in.
+        if !lance_table::feature_flags::can_read_dataset(
+            lance_table::feature_flags::FLAG_UNSTABLE_DATA_OVERLAY_FILES,
+        ) {
+            return;
+        }
+
         // Commit an overlay file supplying new `foo` values for the first three
         // rows of fragment 0. There is no high-level API that writes overlays
         // yet, so write the overlay's data file and commit the `DataOverlay`

rust/lancedb/src/table/query/lsm.rs

@@ -300,7 +300,7 @@ async fn build_read_context(
     for shard_id in shard_ids {
         let manifest_store =
             ShardManifestStore::new(store.clone(), &base_path, shard_id, scan_batch_size);
-        if let Some(manifest) = manifest_store.latest().await? {
+        if let Some(manifest) = manifest_store.read_latest().await? {
             snapshots.push(snapshot_from_manifest(shard_id, &manifest, &exclude));
         }
     }

rust/lancedb/src/table/refresh.rs

@@ -7,6 +7,16 @@
 //! therefore idempotent and does not observe input mutation -- once a row is
 //! filled, changing what the expression reads leaves the stored result alone.
 //!
+//! A column's computed inputs are filled first -- the dependency graph is
+//! walked once, each reachable column filled once in dependency order, each
+//! fill its own commit. Every fill in the pass, the requested column's
+//! included, covers only the fragments of the snapshot the pass started
+//! from: a commit may rebase over a concurrent append, and the fragment that
+//! admits carries placeholder nulls no earlier fill covered, so it waits for
+//! a later refresh rather than being read as values. Two concurrent fills of
+//! one input collide on its field in lance's conflict check, so a dependent
+//! fill can only commit over inputs that were durable when it read them.
+//!
 //! Two passes per fragment. The first scans only the unfilled live rows and
 //! evaluates the expression over them, which yields the exact fill count and
 //! decides whether the fragment is staged at all -- a fragment where nothing
@@ -19,10 +29,14 @@
 //! inputs masked to null first, so a poison value in a row nobody is filling
 //! cannot fail the refresh.
 
+use std::collections::HashSet;
 use std::sync::Arc;
 
-use arrow_array::{ArrayRef, BooleanArray, RecordBatch, RecordBatchOptions};
-use arrow_schema::Schema as ArrowSchema;
+use arrow_array::{
+    Array, ArrayRef, BooleanArray, LargeBinaryArray, RecordBatch, RecordBatchOptions, StructArray,
+    new_null_array,
+};
+use arrow_schema::{DataType, Field as ArrowField, Schema as ArrowSchema};
 use datafusion_expr::ColumnarValue;
 use futures::{Stream, StreamExt, TryStreamExt};
 use lance::Dataset;
@@ -30,7 +44,7 @@ use lance::dataset::WriteDestination;
 use lance::dataset::fragment::FileFragment;
 use lance::dataset::transaction::Operation;
 use lance_core::ROW_ID;
-use lance_core::datatypes::Schema as LanceSchema;
+use lance_core::datatypes::{BlobHandling, Schema as LanceSchema};
 use serde::{Deserialize, Serialize};
 
 use super::computed_columns::{BoundExpression, ComputedColumnKind, computed_column_from_field};
@@ -41,7 +55,8 @@ use crate::{Error, Result};
 /// The result of refreshing a computed column.
 #[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize, Default)]
 pub struct RefreshColumnResult {
-    /// Rows that had a value computed.
+    /// Rows that had a value computed, in the requested column only; inputs
+    /// filled on its behalf are not counted.
     #[serde(default)]
     pub rows_filled: u64,
     /// The commit version associated with the operation.
@@ -52,6 +67,7 @@ pub struct RefreshColumnResult {
 struct RefreshExecution {
     result: RefreshColumnResult,
     source_version: u64,
+    published_version: Option<u64>,
 }
 
 /// Internal implementation of the refresh logic.
@@ -74,7 +90,12 @@ async fn execute_refresh_column_with_source(
 
     let expression = declared_expression(&dataset, column)?;
     let schema = Arc::new(ArrowSchema::from(dataset.schema()));
-    let bound = Arc::new(super::computed_columns::bind(schema, column, &expression)?);
+    let bound = Arc::new(super::computed_columns::bind(
+        schema.clone(),
+        column,
+        &expression,
+    )?);
+    ensure_inputs_filled(&dataset, &schema, column, &bound).await?;
     let field = dataset
         .schema()
         .field(column)
@@ -87,6 +108,7 @@ async fn execute_refresh_column_with_source(
         fields: vec![field.clone()],
         metadata: Default::default(),
     };
+    let output_is_blob = field.is_blob_v2();
 
     let mut rows_filled = 0u64;
     let mut replacements = Vec::new();
@@ -96,29 +118,30 @@ async fn execute_refresh_column_with_source(
             continue;
         }
         rows_filled += gained;
-        let values = fill_stream(&dataset, &fragment, bound.clone(), column).await?;
+        let values =
+            fill_stream(&dataset, &fragment, bound.clone(), column, output_is_blob).await?;
         replacements.push(fragment.write_columns(values, &column_schema).await?);
     }
 
+    let source_version = dataset.version().version;
     if replacements.is_empty() {
-        let source_version = dataset.version().version;
         return Ok(RefreshExecution {
             result: RefreshColumnResult {
                 rows_filled: 0,
                 version: source_version,
             },
             source_version,
+            published_version: None,
         });
     }
 
-    let read_version = dataset.version().version;
     // The dataset's own session, so registrations and caches survive the
     // commit being installed on the handle.
     let session = dataset.session();
     let new_dataset = Dataset::commit(
         WriteDestination::Dataset(dataset.clone()),
         Operation::DataReplacement { replacements },
-        Some(read_version),
+        Some(source_version),
         None,
         None,
         session,
@@ -133,10 +156,52 @@ async fn execute_refresh_column_with_source(
             rows_filled,
             version,
         },
-        source_version: read_version,
+        source_version,
+        published_version: Some(version),
     })
 }
 
+/// Refuse while a computed input still has rows a refresh of it would fill:
+/// read now, its placeholder null would be evaluated as a value and kept.
+async fn ensure_inputs_filled(
+    dataset: &Dataset,
+    schema: &Arc<ArrowSchema>,
+    column: &str,
+    bound: &BoundExpression,
+) -> Result<()> {
+    for input in &bound.roots {
+        let Some(declaration) = schema
+            .field_with_name(input)
+            .ok()
+            .and_then(computed_column_from_field)
+        else {
+            continue;
+        };
+        let ComputedColumnKind::Sql { expression } = &declaration.kind else {
+            return Err(Error::NotSupported {
+                message: format!(
+                    "computed column '{column}' reads '{input}', whose fill state this \
+                     refresh cannot check; refresh '{input}' first"
+                ),
+            });
+        };
+        let input_bound = super::computed_columns::bind(schema.clone(), input, expression)?;
+        let mut unfilled = 0u64;
+        for fragment in dataset.get_fragments() {
+            unfilled += count_fragment_gains(dataset, &fragment, &input_bound, input).await?;
+        }
+        if unfilled > 0 {
+            return Err(Error::InvalidInput {
+                message: format!(
+                    "computed column '{column}' reads '{input}', which has {unfilled} unfilled \
+                     rows; refresh '{input}' first"
+                ),
+            });
+        }
+    }
+    Ok(())
+}
+
 /// Run the refresh as a [`Job`] in this process.
 pub(crate) async fn execute_refresh_column_async(
     table: &NativeTable,
@@ -160,8 +225,7 @@ pub(crate) async fn execute_refresh_column_async(
             rows_failed: 0,
             rows_remaining: 0,
             source_version: execution.source_version,
-            published_version: (execution.result.rows_filled > 0)
-                .then_some(execution.result.version),
+            published_version: execution.published_version,
         })
     })))
 }
@@ -236,12 +300,15 @@ fn evaluation_batch(
     mask_out: Option<&BooleanArray>,
 ) -> lance_core::Result<RecordBatch> {
     let mut columns = Vec::with_capacity(bound.roots.len());
+    let mut fields = Vec::with_capacity(bound.roots.len());
     for name in &bound.roots {
-        let column = batch.column_by_name(name).ok_or_else(|| {
+        let index = batch.schema_ref().index_of(name).map_err(|_| {
             lance_core::Error::invalid_input(format!(
                 "refreshing a computed column read no {name} column"
             ))
         })?;
+        let column = batch.column(index);
+        fields.push(batch.schema_ref().field(index).clone());
         // Rows outside the mask must not reach the expression: a value in a
         // deleted or already-filled row can be one it would choke on.
         columns.push(match mask_out {
@@ -250,7 +317,7 @@ fn evaluation_batch(
         });
     }
     Ok(RecordBatch::try_new_with_options(
-        bound.read_schema.clone(),
+        Arc::new(ArrowSchema::new(fields)),
         columns,
         &RecordBatchOptions::new().with_row_count(Some(batch.num_rows())),
     )?)
@@ -271,6 +338,99 @@ fn evaluate(bound: &BoundExpression, batch: &RecordBatch) -> lance_core::Result<
     }
 }
 
+fn materialized_blob_ids(schema: &LanceSchema, paths: &[String]) -> Result<HashSet<u32>> {
+    paths
+        .iter()
+        .map(|path| {
+            let field = schema
+                .resolve(path)
+                .and_then(|fields| fields.last().copied())
+                .ok_or_else(|| Error::InvalidInput {
+                    message: format!("computed Blob input '{path}' no longer exists"),
+                })?;
+            if !field.is_blob_v2() {
+                return Err(Error::InvalidInput {
+                    message: format!("computed Blob input '{path}' is no longer Blob v2"),
+                });
+            }
+            u32::try_from(field.id).map_err(|_| Error::InvalidInput {
+                message: format!(
+                    "computed Blob input '{path}' has invalid field id {}",
+                    field.id
+                ),
+            })
+        })
+        .collect()
+}
+
+fn configure_blob_inputs(
+    scanner: &mut lance::dataset::scanner::Scanner,
+    schema: &LanceSchema,
+    bound: &BoundExpression,
+    extra_blob_id: Option<u32>,
+) -> Result<()> {
+    let mut ids = materialized_blob_ids(schema, &bound.blob_paths)?;
+    ids.extend(extra_blob_id);
+    scanner.blob_handling(BlobHandling::SomeBlobsBinary(ids));
+    Ok(())
+}
+
+fn blob_array_from_binary(
+    array: &ArrayRef,
+    target_field: &ArrowField,
+) -> lance_core::Result<ArrayRef> {
+    let values = array
+        .as_any()
+        .downcast_ref::<LargeBinaryArray>()
+        .ok_or_else(|| {
+            lance_core::Error::invalid_input(format!(
+                "a Blob v2 computed output produced {}, expected LargeBinary",
+                array.data_type()
+            ))
+        })?;
+    let mut builder = lance::blob::BlobArrayBuilder::new(values.len());
+    for index in 0..values.len() {
+        if values.is_null(index) {
+            builder.push_null()?;
+        } else {
+            builder.push_bytes(values.value(index))?;
+        }
+    }
+    let minimal = builder.finish()?;
+    let minimal = minimal
+        .as_any()
+        .downcast_ref::<StructArray>()
+        .ok_or_else(|| lance_core::Error::internal("Blob builder returned a non-struct array"))?;
+    let DataType::Struct(target_fields) = target_field.data_type() else {
+        return Err(lance_core::Error::invalid_input(format!(
+            "Blob v2 output field '{}' has non-struct type {}",
+            target_field.name(),
+            target_field.data_type()
+        )));
+    };
+    let columns = target_fields
+        .iter()
+        .map(|field| match field.name().as_str() {
+            "data" | "uri" => minimal
+                .column_by_name(field.name())
+                .cloned()
+                .ok_or_else(|| {
+                    lance_core::Error::internal(format!("Blob builder omitted '{}'", field.name()))
+                }),
+            "position" | "size" => Ok(new_null_array(field.data_type(), minimal.len())),
+            name => Err(lance_core::Error::invalid_input(format!(
+                "Blob v2 output field '{}' has unsupported logical child '{name}'",
+                target_field.name()
+            ))),
+        })
+        .collect::<lance_core::Result<Vec<_>>>()?;
+    Ok(Arc::new(StructArray::try_new(
+        target_fields.clone(),
+        columns,
+        minimal.nulls().cloned(),
+    )?))
+}
+
 /// How many rows of one fragment would gain a value.
 ///
 /// Scans only the unfilled live rows -- deleted rows never reach the
@@ -289,6 +449,7 @@ async fn count_fragment_gains(
         .with_row_id()
         .filter(&format!("{} IS NULL", quote_identifier(column)))?
         .project(&bound.roots)?;
+    configure_blob_inputs(&mut scanner, dataset.schema(), bound, None)?;
 
     let mut gained = 0u64;
     let mut batches = scanner.try_into_stream().await?;
@@ -310,6 +471,7 @@ async fn fill_stream(
     fragment: &FileFragment,
     bound: Arc<BoundExpression>,
     column: &str,
+    output_is_blob: bool,
 ) -> Result<impl Stream<Item = lance_core::Result<RecordBatch>> + Send + use<>> {
     let mut projection: Vec<String> = bound.roots.clone();
     projection.push(column.to_string());
@@ -319,6 +481,20 @@ async fn fill_stream(
         .with_row_id()
         .include_deleted_rows()
         .project(&projection)?;
+    let output_blob_id = output_is_blob
+        .then(|| {
+            dataset
+                .schema()
+                .field(column)
+                .and_then(|field| u32::try_from(field.id).ok())
+        })
+        .flatten();
+    configure_blob_inputs(
+        &mut scanner,
+        dataset.schema(),
+        bound.as_ref(),
+        output_blob_id,
+    )?;
 
     let projected = Arc::new(ArrowSchema::new(vec![
         ArrowSchema::from(dataset.schema())
@@ -354,6 +530,11 @@ async fn fill_stream(
 
         let computed = evaluate(&bound, &evaluation_batch(&batch, &bound, Some(&keep))?)?;
         let merged = arrow_select::zip::zip(&fill, &computed, existing)?;
+        let merged = if output_is_blob {
+            blob_array_from_binary(&merged, projected.field(0))?
+        } else {
+            merged
+        };
         Ok(RecordBatch::try_new(projected.clone(), vec![merged])?)
     }))
 }
@@ -362,8 +543,12 @@ async fn fill_stream(
 mod tests {
     use std::sync::Arc;
 
-    use arrow_array::{Int32Array, record_batch};
+    use arrow_array::{
+        Array, ArrayRef, Int32Array, LargeBinaryArray, RecordBatch, StructArray, record_batch,
+    };
+    use arrow_schema::Field as ArrowField;
     use futures::TryStreamExt;
+    use lance_core::ROW_ID;
 
     use crate::connect;
     use crate::query::{ExecutableQuery, QueryBase, Select};
@@ -384,7 +569,8 @@ mod tests {
             .version)
     }
 
-    async fn read(table: &Table, column: &str) -> Vec<Option<i32>> {
+    async fn read(table: &Table, column: &str) -> Vec<Option<i64>> {
+        use arrow_array::{Array, Int64Array};
         let batches = table
             .query()
             .select(Select::columns(&[column]))
@@ -394,15 +580,19 @@ mod tests {
             .try_collect::<Vec<_>>()
             .await
             .unwrap();
-        let mut values: Vec<Option<i32>> = batches
+        let mut values: Vec<Option<i64>> = batches
             .iter()
             .flat_map(|batch| {
-                batch[column]
-                    .as_any()
-                    .downcast_ref::<Int32Array>()
-                    .unwrap()
-                    .iter()
-                    .collect::<Vec<_>>()
+                let array = &batch[column];
+                match array.as_any().downcast_ref::<Int32Array>() {
+                    Some(ints) => ints.iter().map(|v| v.map(i64::from)).collect::<Vec<_>>(),
+                    None => array
+                        .as_any()
+                        .downcast_ref::<Int64Array>()
+                        .unwrap()
+                        .iter()
+                        .collect::<Vec<_>>(),
+                }
             })
             .collect();
         values.sort();
@@ -414,6 +604,117 @@ mod tests {
         table.add(batch).execute().await.unwrap();
     }
 
+    #[test]
+    fn test_blob_output_matches_complete_logical_field() {
+        let values: ArrayRef = Arc::new(LargeBinaryArray::from(vec![
+            Some(b"hello".as_slice()),
+            None,
+        ]));
+        let field = ArrowField::new(
+            "image",
+            lance_core::datatypes::BLOB_V2_LOGICAL_TYPE.clone(),
+            true,
+        );
+
+        let output = super::blob_array_from_binary(&values, &field).unwrap();
+        assert_eq!(output.data_type(), field.data_type());
+        let output = output.as_any().downcast_ref::<StructArray>().unwrap();
+        assert_eq!(output.column_by_name("position").unwrap().null_count(), 2);
+        assert_eq!(output.column_by_name("size").unwrap().null_count(), 2);
+    }
+
+    /// The gate's reproducer: `b = coalesce(a, 0)` refreshed before `a`
+    /// must not bake zeros from `a`'s placeholder null. It is refused, and
+    /// names the input, until `a` is filled -- after every append too.
+    #[tokio::test]
+    async fn test_dependent_refresh_refuses_an_unfilled_input() {
+        let table = table_with("dependent_refresh_order", vec![1, 2, 3]).await;
+        table
+            .add_columns()
+            .computed("a", "x + 1")
+            .computed("b", "coalesce(a, 0)")
+            .execute()
+            .await
+            .unwrap();
+
+        let err = table.refresh_column("b").await.unwrap_err();
+        assert!(
+            matches!(&err, Error::InvalidInput { message } if message.contains("refresh 'a' first")),
+            "{err}"
+        );
+        assert_eq!(read(&table, "b").await, vec![None, None, None]);
+
+        assert_eq!(table.refresh_column("a").await.unwrap().rows_filled, 3);
+        assert_eq!(table.refresh_column("b").await.unwrap().rows_filled, 3);
+        assert_eq!(read(&table, "b").await, vec![Some(2), Some(3), Some(4)]);
+
+        append(&table, vec![10]).await;
+        assert!(table.refresh_column("b").await.is_err());
+        table.refresh_column("a").await.unwrap();
+        assert_eq!(table.refresh_column("b").await.unwrap().rows_filled, 1);
+        assert_eq!(
+            table.count_rows(Some("b = 0".to_string())).await.unwrap(),
+            0
+        );
+    }
+
+    /// Names that need quoting, and a nested input, survive the trip through
+    /// declaration metadata and the dependency check: the recorded inputs
+    /// are matched by name, never re-parsed as SQL.
+    #[tokio::test]
+    async fn test_dependent_refresh_handles_awkward_column_names() {
+        use arrow_array::{Int32Array, StructArray};
+        use arrow_schema::{DataType, Field, Fields};
+
+        let conn = connect("memory://").execute().await.unwrap();
+        let age_fields = Fields::from(vec![Field::new("age", DataType::Int32, true)]);
+        let meta = StructArray::new(
+            age_fields.clone(),
+            vec![Arc::new(Int32Array::from(vec![10, 20])) as _],
+            None,
+        );
+        let schema = Arc::new(arrow_schema::Schema::new(vec![
+            Field::new("camelCase", DataType::Int32, true),
+            Field::new("with-hyphen", DataType::Int32, true),
+            Field::new("meta", DataType::Struct(age_fields), true),
+        ]));
+        let batch = arrow_array::RecordBatch::try_new(
+            schema,
+            vec![
+                Arc::new(Int32Array::from(vec![1, 2])) as _,
+                Arc::new(Int32Array::from(vec![100, 200])) as _,
+                Arc::new(meta) as _,
+            ],
+        )
+        .unwrap();
+        let table = conn
+            .create_table("awkward_names", batch)
+            .execute()
+            .await
+            .unwrap();
+
+        table
+            .add_columns()
+            .computed("y", "`camelCase` * 2")
+            .computed("z", "coalesce(y, 0) + `with-hyphen` + meta.age")
+            .execute()
+            .await
+            .unwrap();
+        let z = crate::table::computed_columns::computed_columns(
+            table.schema().await.unwrap().as_ref(),
+        )
+        .into_iter()
+        .find(|c| c.name == "z")
+        .unwrap();
+        assert_eq!(z.inputs, vec!["meta.age", "with-hyphen", "y"]);
+
+        let err = table.refresh_column("z").await.unwrap_err();
+        assert!(err.to_string().contains("refresh 'y' first"), "{err}");
+        assert_eq!(table.refresh_column("y").await.unwrap().rows_filled, 2);
+        assert_eq!(table.refresh_column("z").await.unwrap().rows_filled, 2);
+        assert_eq!(read(&table, "z").await, vec![Some(112), Some(224)]);
+    }
+
     #[tokio::test]
     async fn test_refresh_fills_a_declared_column() {
         let table = table_with("refresh_fills", vec![1, 2, 3]).await;
@@ -651,7 +952,8 @@ mod tests {
 
         let read_back = read(&table, "doubled").await;
         assert_eq!(read_back.len(), 20_000);
-        let mut expected: Vec<Option<i32>> = values.iter().map(|v| Some(v * 2)).collect();
+        let mut expected: Vec<Option<i64>> =
+            values.iter().map(|v| Some(i64::from(v * 2))).collect();
         expected.sort();
         assert_eq!(read_back, expected);
     }
@@ -1008,4 +1310,366 @@ mod tests {
         let err = table.refresh_column("embedding").await.unwrap_err();
         assert!(matches!(err, Error::NotSupported { message } if message.contains("udf")));
     }
+
+    fn blob_batch(ids: Vec<i32>, payloads: Vec<Option<&[u8]>>) -> RecordBatch {
+        use arrow_array::Int32Array;
+        use arrow_schema::{Field, Schema};
+
+        let mut builder = lance::blob::BlobArrayBuilder::new(payloads.len());
+        for payload in payloads {
+            match payload {
+                Some(payload) => builder.push_bytes(payload).unwrap(),
+                None => builder.push_null().unwrap(),
+            }
+        }
+        RecordBatch::try_new(
+            Arc::new(Schema::new(vec![
+                Field::new("id", arrow_schema::DataType::Int32, false),
+                crate::blob("image", true),
+            ])),
+            vec![Arc::new(Int32Array::from(ids)), builder.finish().unwrap()],
+        )
+        .unwrap()
+    }
+
+    async fn create_blob_table(path: &std::path::Path, batch: RecordBatch) -> Table {
+        let conn = connect(path.to_str().unwrap()).execute().await.unwrap();
+        conn.create_table("blobs", batch).execute().await.unwrap()
+    }
+
+    #[tokio::test]
+    async fn test_refresh_inherits_and_publishes_blob_output() {
+        use arrow_array::UInt64Array;
+        use lance_arrow::{
+            BLOB_DEDICATED_SIZE_THRESHOLD_META_KEY, BLOB_INLINE_SIZE_THRESHOLD_META_KEY,
+        };
+        use lance_core::datatypes::BlobKind;
+
+        use crate::table::schema_evolution::FieldMetadataUpdate;
+
+        let tmp = tempfile::tempdir().unwrap();
+        let table = create_blob_table(
+            tmp.path(),
+            blob_batch(
+                vec![1, 2, 3, 4],
+                vec![Some(b"hello"), Some(b"ab"), Some(b""), None],
+            ),
+        )
+        .await;
+        table
+            .add_columns()
+            .computed("image_copy", "image")
+            .execute()
+            .await
+            .unwrap();
+        table
+            .update_field_metadata(&[FieldMetadataUpdate::new("image_copy")
+                .set(BLOB_INLINE_SIZE_THRESHOLD_META_KEY, "1")
+                .set(BLOB_DEDICATED_SIZE_THRESHOLD_META_KEY, "4")])
+            .await
+            .unwrap();
+
+        let first_refresh = table.refresh_column("image_copy").await.unwrap();
+        assert_eq!(first_refresh.rows_filled, 3);
+        assert_eq!(
+            table.blob_columns().await.unwrap(),
+            vec!["image".to_string(), "image_copy".to_string()]
+        );
+
+        let batches = table
+            .query()
+            .with_row_id()
+            .execute()
+            .await
+            .unwrap()
+            .try_collect::<Vec<_>>()
+            .await
+            .unwrap();
+        let batch = arrow_select::concat::concat_batches(&batches[0].schema(), &batches).unwrap();
+        assert!(
+            batch
+                .column_by_name("image_copy")
+                .unwrap()
+                .as_any()
+                .is::<arrow_array::StructArray>()
+        );
+        let row_ids = batch
+            .column_by_name(ROW_ID)
+            .unwrap()
+            .as_any()
+            .downcast_ref::<UInt64Array>()
+            .unwrap()
+            .values()
+            .to_vec();
+        let original = table.fetch_blobs("image", &row_ids).await.unwrap();
+        let copied = table.fetch_blobs("image_copy", &row_ids).await.unwrap();
+        assert_eq!(original, copied);
+        let ids = batch
+            .column_by_name("id")
+            .unwrap()
+            .as_any()
+            .downcast_ref::<Int32Array>()
+            .unwrap();
+        let files = table
+            .fetch_blob_files("image_copy", &row_ids)
+            .await
+            .unwrap();
+        let mut layouts = ids
+            .values()
+            .iter()
+            .copied()
+            .zip(files)
+            .map(|(id, file)| (id, file.and_then(|file| file.kind())))
+            .collect::<Vec<_>>();
+        layouts.sort_by_key(|(id, _)| *id);
+        assert_eq!(
+            layouts,
+            vec![
+                (1, Some(BlobKind::Dedicated)),
+                (2, Some(BlobKind::Packed)),
+                (3, Some(BlobKind::Inline)),
+                (4, None),
+            ]
+        );
+
+        table
+            .add(blob_batch(vec![5], vec![Some(b"appended")]))
+            .execute()
+            .await
+            .unwrap();
+        table
+            .optimize(crate::table::OptimizeAction::Compact {
+                options: crate::table::CompactionOptions::default(),
+                remap_options: None,
+            })
+            .await
+            .unwrap();
+        assert_eq!(
+            table
+                .refresh_column("image_copy")
+                .await
+                .unwrap()
+                .rows_filled,
+            1
+        );
+        assert_eq!(
+            table
+                .refresh_column("image_copy")
+                .await
+                .unwrap()
+                .rows_filled,
+            0
+        );
+
+        table.checkout(first_refresh.version).await.unwrap();
+        assert_eq!(table.count_rows(None).await.unwrap(), 4);
+        assert_eq!(
+            table.blob_columns().await.unwrap(),
+            vec!["image".to_string(), "image_copy".to_string()]
+        );
+        table.checkout_latest().await.unwrap();
+    }
+
+    #[tokio::test]
+    async fn test_refresh_inherits_nested_struct_blob_input() {
+        use arrow_array::{Int32Array, StructArray, UInt64Array};
+        use arrow_schema::{DataType, Field, Fields, Schema};
+
+        let tmp = tempfile::tempdir().unwrap();
+        let mut blob_builder = lance::blob::BlobArrayBuilder::new(2);
+        blob_builder.push_bytes(b"nested").unwrap();
+        blob_builder.push_null().unwrap();
+        let blob_field = crate::blob("image", true);
+        let metadata_fields = Fields::from(vec![blob_field.clone()]);
+        let metadata = StructArray::new(
+            metadata_fields.clone(),
+            vec![blob_builder.finish().unwrap()],
+            None,
+        );
+        let batch = RecordBatch::try_new(
+            Arc::new(Schema::new(vec![
+                Field::new("id", DataType::Int32, false),
+                Field::new("metadata", DataType::Struct(metadata_fields), true),
+            ])),
+            vec![Arc::new(Int32Array::from(vec![1, 2])), Arc::new(metadata)],
+        )
+        .unwrap();
+        let table = create_blob_table(tmp.path(), batch).await;
+        table
+            .add_columns()
+            .computed("payload_copy", "metadata.image")
+            .execute()
+            .await
+            .unwrap();
+
+        assert_eq!(
+            table
+                .refresh_column("payload_copy")
+                .await
+                .unwrap()
+                .rows_filled,
+            1
+        );
+        assert_eq!(
+            table.blob_columns().await.unwrap(),
+            vec!["metadata.image".to_string(), "payload_copy".to_string()]
+        );
+        let batches = table
+            .query()
+            .with_row_id()
+            .execute()
+            .await
+            .unwrap()
+            .try_collect::<Vec<_>>()
+            .await
+            .unwrap();
+        let row_ids = batches[0]
+            .column_by_name(ROW_ID)
+            .unwrap()
+            .as_any()
+            .downcast_ref::<UInt64Array>()
+            .unwrap()
+            .values();
+        let payloads = table.fetch_blobs("payload_copy", row_ids).await.unwrap();
+        assert_eq!(payloads.value(0), b"nested");
+        assert!(payloads.is_null(1));
+    }
+
+    #[tokio::test]
+    async fn test_refresh_preserves_list_shape_when_materializing_blob_input() {
+        use arrow_array::{Int32Array, ListArray};
+        use arrow_buffer::{OffsetBuffer, ScalarBuffer};
+        use arrow_schema::{DataType, Field, Schema};
+
+        let tmp = tempfile::tempdir().unwrap();
+        let mut blob_builder = lance::blob::BlobArrayBuilder::new(3);
+        blob_builder.push_bytes(b"a").unwrap();
+        blob_builder.push_bytes(b"bb").unwrap();
+        blob_builder.push_null().unwrap();
+        let item = Arc::new(crate::blob("item", true));
+        let images = ListArray::new(
+            item.clone(),
+            OffsetBuffer::new(ScalarBuffer::from(vec![0, 2, 3])),
+            blob_builder.finish().unwrap(),
+            None,
+        );
+        let batch = RecordBatch::try_new(
+            Arc::new(Schema::new(vec![
+                Field::new("id", DataType::Int32, false),
+                Field::new("images", DataType::List(item), true),
+            ])),
+            vec![Arc::new(Int32Array::from(vec![1, 2])), Arc::new(images)],
+        )
+        .unwrap();
+        let table = create_blob_table(tmp.path(), batch).await;
+        table
+            .add_columns()
+            .computed("image_payloads", "images")
+            .execute()
+            .await
+            .unwrap();
+
+        assert_eq!(
+            table
+                .refresh_column("image_payloads")
+                .await
+                .unwrap()
+                .rows_filled,
+            2
+        );
+        let batches = table
+            .query()
+            .select(Select::columns(&["image_payloads"]))
+            .execute()
+            .await
+            .unwrap()
+            .try_collect::<Vec<_>>()
+            .await
+            .unwrap();
+        let output = batches[0]
+            .column_by_name("image_payloads")
+            .unwrap()
+            .as_any()
+            .downcast_ref::<ListArray>()
+            .unwrap();
+        assert_eq!(output.value_offsets(), &[0, 2, 3]);
+        assert!(output.values().as_any().is::<LargeBinaryArray>());
+    }
+
+    #[tokio::test]
+    async fn test_refresh_inherits_external_blob_input() {
+        use arrow_array::{Int32Array, StringArray, UInt64Array};
+        use arrow_schema::{DataType, Field, Schema};
+
+        let tmp = tempfile::tempdir().unwrap();
+        let payload = b"external-payload";
+        let path = tmp.path().join("payload.bin");
+        std::fs::write(&path, payload).unwrap();
+        let uri = url::Url::from_file_path(path).unwrap().to_string();
+        let conn = connect(tmp.path().join("db").to_str().unwrap())
+            .execute()
+            .await
+            .unwrap();
+        let table = conn
+            .create_empty_table(
+                "external",
+                Arc::new(Schema::new(vec![
+                    Field::new("id", DataType::Int32, false),
+                    crate::blob("image", true),
+                ])),
+            )
+            .execute()
+            .await
+            .unwrap();
+        let batch = RecordBatch::try_new(
+            Arc::new(Schema::new(vec![
+                Field::new("id", DataType::Int32, false),
+                Field::new("image", DataType::Utf8, true),
+            ])),
+            vec![
+                Arc::new(Int32Array::from(vec![1])),
+                Arc::new(StringArray::from(vec![Some(uri)])),
+            ],
+        )
+        .unwrap();
+        table
+            .add(batch)
+            .allow_external_blob_outside_bases(true)
+            .execute()
+            .await
+            .unwrap();
+        table
+            .add_columns()
+            .computed("payload_copy", "image")
+            .execute()
+            .await
+            .unwrap();
+
+        assert_eq!(
+            table
+                .refresh_column("payload_copy")
+                .await
+                .unwrap()
+                .rows_filled,
+            1
+        );
+        let batches = table
+            .query()
+            .with_row_id()
+            .execute()
+            .await
+            .unwrap()
+            .try_collect::<Vec<_>>()
+            .await
+            .unwrap();
+        let row_ids = batches[0]
+            .column_by_name(ROW_ID)
+            .unwrap()
+            .as_any()
+            .downcast_ref::<UInt64Array>()
+            .unwrap()
+            .values();
+        let payloads = table.fetch_blobs("payload_copy", row_ids).await.unwrap();
+        assert_eq!(payloads.value(0), payload);
+    }
 }

rust/lancedb/tests/fixtures/first_class_functions/v1/arrow_types.json

@@ -78,6 +78,12 @@
         "type": "utf8"
       }
     },
+    {
+      "arrow_type": "large_utf8",
+      "json": {
+        "type": "large_utf8"
+      }
+    },
     {
       "arrow_type": "binary",
       "json": {
@@ -171,6 +177,21 @@
         ]
       }
     },
+    {
+      "arrow_type": "list<large_utf8>",
+      "json": {
+        "type": "list",
+        "fields": [
+          {
+            "name": "item",
+            "nullable": false,
+            "type": {
+              "type": "large_utf8"
+            }
+          }
+        ]
+      }
+    },
     {
       "arrow_type": "large_list<utf8>",
       "json": {
@@ -186,6 +207,21 @@
         ]
       }
     },
+    {
+      "arrow_type": "large_list<large_utf8>",
+      "json": {
+        "type": "large_list",
+        "fields": [
+          {
+            "name": "item",
+            "nullable": false,
+            "type": {
+              "type": "large_utf8"
+            }
+          }
+        ]
+      }
+    },
     {
       "arrow_type": "fixed_size_list<float32, 384>",
       "json": {
@@ -330,4 +366,4 @@
     "timestamp[us]",
     "struct<a: int32>"
   ]
-}
+}