Merge origin/main into gatekeeper/fix-3759-1

fix(python): support blob lists in dict appends
fix(python): preserve nested extension carriers
2026-09-08 22:39:05 +00:00 · 2026-08-28 01:19:19 +00:00 · 2026-08-26 23:47:47 +00:00 · 2026-08-26 23:21:40 +00:00 · 2026-08-26 22:55:54 +00:00
49 changed files with 11950 additions and 2021 deletions
@@ -1,5 +1,5 @@
 [tool.bumpversion]
-current_version = "0.38.0-beta.14"
+current_version = "0.38.0-beta.12"
 parse = """(?x)
    (?P<major>0|[1-9]\\d*)\\.
    (?P<minor>0|[1-9]\\d*)\\.
@@ -44,27 +44,3 @@ 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:
-          - "*"
@@ -29,14 +29,12 @@ jobs:
    steps:
      - uses: actions/setup-node@v6
        with:
-          node-version: "24"
-      - uses: pnpm/action-setup@v6
-        with:
-          version: 11.1.1
+          node-version: "18"
      # 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": {
@@ -45,11 +43,7 @@ jobs:
              "body-leading-blank": [0, "always"]
            }
          }' > .commitlintrc.js
-      - run: >
-          pnpm dlx
-          --package @commitlint/cli@21.2.2
-          --package @commitlint/config-conventional@21.2.2
-          commitlint --extends @commitlint/config-conventional --verbose <<< $COMMIT_MSG
+      - run: npx commitlint --extends @commitlint/config-conventional --verbose <<< $COMMIT_MSG
        env:
          COMMIT_MSG: >
            ${{ github.event.pull_request.title }}
@@ -60,7 +54,7 @@ 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.\
@@ -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 `pnpm run docs` in nodejs)
+          # API reference (the js/ tree comes from `npm 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
@@ -55,7 +55,9 @@ jobs:
      - name: Set up node
        uses: actions/setup-node@v6
        with:
-          node-version: 24
+          node-version: 20
+          cache: 'npm'
+          cache-dependency-path: docs/package-lock.json
      - name: Install node dependencies
        working-directory: nodejs
        run: |
@@ -47,8 +47,9 @@ jobs:
        version: 11.1.1
    - uses: actions/setup-node@v6
      with:
-        # Build on a supported LTS; the matrix job below covers every
-        # Node version the library claims to support.
+        # 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).
        node-version: 24
        cache: 'pnpm'
        cache-dependency-path: nodejs/pnpm-lock.yaml
@@ -83,7 +84,7 @@ jobs:
    timeout-minutes: 30
    strategy:
      matrix:
-        node-version: [ "22", "24", "26" ]
+        node-version: [ "18", "20" ]
    runs-on: "ubuntu-22.04"
    defaults:
      run:
@@ -100,9 +101,9 @@ jobs:
    - uses: actions/setup-node@v6
      name: Setup Node.js 24 for build
      with:
-        # 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.
+        # 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.
        node-version: 24
        cache: 'pnpm'
        cache-dependency-path: nodejs/pnpm-lock.yaml
@@ -151,9 +152,9 @@ jobs:
        S3_TEST: "1"
        # Newer @smithy/core uses dynamic ESM imports.
        NODE_OPTIONS: "--experimental-vm-modules"
-      # 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
+      # Invoke jest directly because pnpm 11 itself requires Node 22+
+      # while the matrix tests on older Node versions.
+      run: npx jest --verbose
    - name: Test examples
      working-directory: ./
      env:
@@ -163,7 +164,7 @@ jobs:
      run: |
        python ci/mock_openai.py &
        cd nodejs/examples
-        node_modules/.bin/jest --testEnvironment jest-environment-node-single-context --verbose
+        npx 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
@@ -184,7 +185,8 @@ jobs:
        version: 11.1.1
    - uses: actions/setup-node@v6
      with:
-        # pnpm 11 requires Node >= 22.13.
+        # pnpm 11 requires Node >= 22.13; use 24 since 22 hits EOL
+        # in October.
        node-version: 24
        cache: 'pnpm'
        cache-dependency-path: nodejs/pnpm-lock.yaml
@@ -168,7 +168,8 @@ jobs:
      - name: Setup node
        uses: actions/setup-node@v6
        with:
-          # pnpm 11 requires Node >= 22.13.
+          # pnpm 11 requires Node >= 22.13; use 24 since 22 hits EOL
+          # in October.
          node-version: 24
          cache: pnpm
          cache-dependency-path: nodejs/pnpm-lock.yaml
@@ -250,7 +251,7 @@ jobs:
          run: |
            set -e
            ${{ matrix.settings.pre_build }}
-            node_modules/.bin/napi build --platform --release \
+            npx napi build --platform --release \
              --features ${{ matrix.settings.features }} \
              --target ${{ matrix.settings.target }} \
              --dts ../lancedb/native.d.ts \
@@ -270,7 +271,7 @@ jobs:
      - name: Build
        run: |
          ${{ matrix.settings.pre_build }}
-          node_modules/.bin/napi build --platform --release \
+          npx napi build --platform --release \
              --features ${{ matrix.settings.features }} \
              --target ${{ matrix.settings.target }} \
              --dts ../lancedb/native.d.ts \
@@ -338,7 +339,7 @@ jobs:
          - target: aarch64-unknown-linux-gnu
            host: ubuntu-2404-8x-arm64
        node:
-          - '22'
+          - '20'
    runs-on: ${{ matrix.settings.host }}
    defaults:
      run:
@@ -384,9 +385,9 @@ jobs:
      - name: Move built files
        run: cp dist/native.d.ts dist/native.js dist/*.node lancedb/
      - name: Test bindings
-        # 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
+        # Invoke jest directly because pnpm 11 itself requires Node 22+
+        # while the matrix tests on older Node versions.
+        run: npx jest --verbose
  publish:
    name: Publish
    runs-on: ubuntu-latest
@@ -0,0 +1,22 @@
+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 }}
@@ -0,0 +1,22 @@
+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 }}
@@ -20,10 +20,7 @@ repos:
    hooks:
      - id: local-biome-check
        name: biome check
-        # 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/
+        entry: npx @biomejs/biome@1.8.3 check --config-path nodejs/biome.json nodejs/
        language: system
        types: [text]
        files: "nodejs/.*"
@@ -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 `pnpm` lint, format, build, and docs commands in `nodejs`.
+* TypeScript changes: run the relevant `npm`/`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 `pnpm build` to generate TypeScript definitions.
+2. Run `npm run 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 `pnpm run docs` to generate TypeScript documentation.
+6. Run `npm run docs` to generate TypeScript documentation.

 ## Python API reference

@@ -1597,9 +1597,9 @@ checksum = "613afe47fcd5fac7ccf1db93babcb082c5994d996f20b8b159f2ad1658eb5724"

 [[package]]
 name = "chacha20"
-version = "0.10.2"
+version = "0.10.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "65c35e4b699c7e15ccbe7ee35c005e4fc0a278d22238a2857e6ce2dadeda1b06"
+checksum = "6f8d983286843e49675a4b7a2d174efe136dc93a18d69130dd18198a6c167601"
 dependencies = [
 "cfg-if 1.0.4",
 "cpufeatures 0.3.0",
@@ -3455,8 +3455,8 @@ checksum = "42703706b716c37f96a77aea830392ad231f44c9e9a67872fa5548707e11b11c"

 [[package]]
 name = "fsst"
-version = "12.0.0-beta.5"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.5#556637791d0048c2b4f1342dd84b67c8bbd65259"
+version = "12.0.0-beta.2"
+source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.2#dafa4642658d996b3e31dde91e02f72db7860d7e"
 dependencies = [
 "arrow-array",
 "rand 0.9.5",
@@ -4815,8 +4815,8 @@ checksum = "e037a2e1d8d5fdbd49b16a4ea09d5d6401c1f29eca5ff29d03d3824dba16256a"

 [[package]]
 name = "lance"
-version = "12.0.0-beta.5"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.5#556637791d0048c2b4f1342dd84b67c8bbd65259"
+version = "12.0.0-beta.2"
+source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.2#dafa4642658d996b3e31dde91e02f72db7860d7e"
 dependencies = [
 "arc-swap",
 "arrow",
@@ -4888,8 +4888,8 @@ dependencies = [

 [[package]]
 name = "lance-arrow"
-version = "12.0.0-beta.5"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.5#556637791d0048c2b4f1342dd84b67c8bbd65259"
+version = "12.0.0-beta.2"
+source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.2#dafa4642658d996b3e31dde91e02f72db7860d7e"
 dependencies = [
 "arrow-array",
 "arrow-buffer",
@@ -4911,7 +4911,7 @@ dependencies = [
 [[package]]
 name = "lance-arrow-scalar"
 version = "58.0.0"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.5#556637791d0048c2b4f1342dd84b67c8bbd65259"
+source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.2#dafa4642658d996b3e31dde91e02f72db7860d7e"
 dependencies = [
 "arrow-array",
 "arrow-buffer",
@@ -4925,7 +4925,7 @@ dependencies = [
 [[package]]
 name = "lance-arrow-stats"
 version = "58.0.0"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.5#556637791d0048c2b4f1342dd84b67c8bbd65259"
+source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.2#dafa4642658d996b3e31dde91e02f72db7860d7e"
 dependencies = [
 "arrow-array",
 "arrow-schema",
@@ -4934,8 +4934,8 @@ dependencies = [

 [[package]]
 name = "lance-bitpacking"
-version = "12.0.0-beta.5"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.5#556637791d0048c2b4f1342dd84b67c8bbd65259"
+version = "12.0.0-beta.2"
+source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.2#dafa4642658d996b3e31dde91e02f72db7860d7e"
 dependencies = [
 "arrayref",
 "crunchy",
@@ -4945,8 +4945,8 @@ dependencies = [

 [[package]]
 name = "lance-core"
-version = "12.0.0-beta.5"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.5#556637791d0048c2b4f1342dd84b67c8bbd65259"
+version = "12.0.0-beta.2"
+source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.2#dafa4642658d996b3e31dde91e02f72db7860d7e"
 dependencies = [
 "arrow-array",
 "arrow-buffer",
@@ -4983,8 +4983,8 @@ dependencies = [

 [[package]]
 name = "lance-datafusion"
-version = "12.0.0-beta.5"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.5#556637791d0048c2b4f1342dd84b67c8bbd65259"
+version = "12.0.0-beta.2"
+source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.2#dafa4642658d996b3e31dde91e02f72db7860d7e"
 dependencies = [
 "arrow",
 "arrow-array",
@@ -5013,8 +5013,8 @@ dependencies = [

 [[package]]
 name = "lance-datagen"
-version = "12.0.0-beta.5"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.5#556637791d0048c2b4f1342dd84b67c8bbd65259"
+version = "12.0.0-beta.2"
+source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.2#dafa4642658d996b3e31dde91e02f72db7860d7e"
 dependencies = [
 "arrow",
 "arrow-array",
@@ -5031,8 +5031,8 @@ dependencies = [

 [[package]]
 name = "lance-derive"
-version = "12.0.0-beta.5"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.5#556637791d0048c2b4f1342dd84b67c8bbd65259"
+version = "12.0.0-beta.2"
+source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.2#dafa4642658d996b3e31dde91e02f72db7860d7e"
 dependencies = [
 "proc-macro2",
 "quote",
@@ -5041,8 +5041,8 @@ dependencies = [

 [[package]]
 name = "lance-encoding"
-version = "12.0.0-beta.5"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.5#556637791d0048c2b4f1342dd84b67c8bbd65259"
+version = "12.0.0-beta.2"
+source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.2#dafa4642658d996b3e31dde91e02f72db7860d7e"
 dependencies = [
 "arrow-arith",
 "arrow-array",
@@ -5075,8 +5075,8 @@ dependencies = [

 [[package]]
 name = "lance-file"
-version = "12.0.0-beta.5"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.5#556637791d0048c2b4f1342dd84b67c8bbd65259"
+version = "12.0.0-beta.2"
+source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.2#dafa4642658d996b3e31dde91e02f72db7860d7e"
 dependencies = [
 "arrow-arith",
 "arrow-array",
@@ -5107,8 +5107,8 @@ dependencies = [

 [[package]]
 name = "lance-index"
-version = "12.0.0-beta.5"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.5#556637791d0048c2b4f1342dd84b67c8bbd65259"
+version = "12.0.0-beta.2"
+source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.2#dafa4642658d996b3e31dde91e02f72db7860d7e"
 dependencies = [
 "arc-swap",
 "arrow",
@@ -5172,8 +5172,8 @@ dependencies = [

 [[package]]
 name = "lance-index-core"
-version = "12.0.0-beta.5"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.5#556637791d0048c2b4f1342dd84b67c8bbd65259"
+version = "12.0.0-beta.2"
+source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.2#dafa4642658d996b3e31dde91e02f72db7860d7e"
 dependencies = [
 "arrow-array",
 "arrow-schema",
@@ -5195,8 +5195,8 @@ dependencies = [

 [[package]]
 name = "lance-io"
-version = "12.0.0-beta.5"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.5#556637791d0048c2b4f1342dd84b67c8bbd65259"
+version = "12.0.0-beta.2"
+source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.2#dafa4642658d996b3e31dde91e02f72db7860d7e"
 dependencies = [
 "arrow",
 "arrow-array",
@@ -5236,8 +5236,8 @@ dependencies = [

 [[package]]
 name = "lance-linalg"
-version = "12.0.0-beta.5"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.5#556637791d0048c2b4f1342dd84b67c8bbd65259"
+version = "12.0.0-beta.2"
+source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.2#dafa4642658d996b3e31dde91e02f72db7860d7e"
 dependencies = [
 "arrow-array",
 "arrow-schema",
@@ -5251,8 +5251,8 @@ dependencies = [

 [[package]]
 name = "lance-namespace"
-version = "12.0.0-beta.5"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.5#556637791d0048c2b4f1342dd84b67c8bbd65259"
+version = "12.0.0-beta.2"
+source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.2#dafa4642658d996b3e31dde91e02f72db7860d7e"
 dependencies = [
 "arrow",
 "async-trait",
@@ -5264,8 +5264,8 @@ dependencies = [

 [[package]]
 name = "lance-namespace-impls"
-version = "12.0.0-beta.5"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.5#556637791d0048c2b4f1342dd84b67c8bbd65259"
+version = "12.0.0-beta.2"
+source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.2#dafa4642658d996b3e31dde91e02f72db7860d7e"
 dependencies = [
 "arrow",
 "arrow-ipc",
@@ -5318,8 +5318,8 @@ dependencies = [

 [[package]]
 name = "lance-select"
-version = "12.0.0-beta.5"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.5#556637791d0048c2b4f1342dd84b67c8bbd65259"
+version = "12.0.0-beta.2"
+source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.2#dafa4642658d996b3e31dde91e02f72db7860d7e"
 dependencies = [
 "arrow-array",
 "arrow-buffer",
@@ -5333,8 +5333,8 @@ dependencies = [

 [[package]]
 name = "lance-table"
-version = "12.0.0-beta.5"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.5#556637791d0048c2b4f1342dd84b67c8bbd65259"
+version = "12.0.0-beta.2"
+source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.2#dafa4642658d996b3e31dde91e02f72db7860d7e"
 dependencies = [
 "arrow",
 "arrow-array",
@@ -5374,8 +5374,8 @@ dependencies = [

 [[package]]
 name = "lance-testing"
-version = "12.0.0-beta.5"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.5#556637791d0048c2b4f1342dd84b67c8bbd65259"
+version = "12.0.0-beta.2"
+source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.2#dafa4642658d996b3e31dde91e02f72db7860d7e"
 dependencies = [
 "arrow-array",
 "arrow-schema",
@@ -5388,8 +5388,8 @@ dependencies = [

 [[package]]
 name = "lance-tokenizer"
-version = "12.0.0-beta.5"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.5#556637791d0048c2b4f1342dd84b67c8bbd65259"
+version = "12.0.0-beta.2"
+source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.2#dafa4642658d996b3e31dde91e02f72db7860d7e"
 dependencies = [
 "frostem",
 "icu_segmenter",
@@ -5402,7 +5402,7 @@ dependencies = [

 [[package]]
 name = "lancedb"
-version = "0.38.0-beta.14"
+version = "0.38.0-beta.12"
 dependencies = [
 "ahash",
 "anyhow",
@@ -5490,7 +5490,7 @@ dependencies = [

 [[package]]
 name = "lancedb-nodejs"
-version = "0.38.0-beta.14"
+version = "0.38.0-beta.12"
 dependencies = [
 "arrow-array",
 "arrow-buffer",
@@ -5515,7 +5515,7 @@ dependencies = [

 [[package]]
 name = "lancedb-python"
-version = "0.38.0-beta.14"
+version = "0.38.0-beta.12"
 dependencies = [
 "arrow",
 "async-trait",
@@ -13,20 +13,20 @@ categories = ["database-implementations"]
 rust-version = "1.91.0"

 [workspace.dependencies]
-lance = { "version" = "=12.0.0-beta.5", default-features = false, "tag" = "v12.0.0-beta.5", "git" = "https://github.com/lance-format/lance.git" }
-lance-core = { "version" = "=12.0.0-beta.5", "tag" = "v12.0.0-beta.5", "git" = "https://github.com/lance-format/lance.git" }
-lance-datagen = { "version" = "=12.0.0-beta.5", "tag" = "v12.0.0-beta.5", "git" = "https://github.com/lance-format/lance.git" }
-lance-file = { "version" = "=12.0.0-beta.5", "tag" = "v12.0.0-beta.5", "git" = "https://github.com/lance-format/lance.git" }
-lance-io = { "version" = "=12.0.0-beta.5", default-features = false, "tag" = "v12.0.0-beta.5", "git" = "https://github.com/lance-format/lance.git" }
-lance-index = { "version" = "=12.0.0-beta.5", "tag" = "v12.0.0-beta.5", "git" = "https://github.com/lance-format/lance.git" }
-lance-linalg = { "version" = "=12.0.0-beta.5", "tag" = "v12.0.0-beta.5", "git" = "https://github.com/lance-format/lance.git" }
-lance-namespace = { "version" = "=12.0.0-beta.5", "tag" = "v12.0.0-beta.5", "git" = "https://github.com/lance-format/lance.git" }
-lance-namespace-impls = { "version" = "=12.0.0-beta.5", default-features = false, "tag" = "v12.0.0-beta.5", "git" = "https://github.com/lance-format/lance.git" }
-lance-table = { "version" = "=12.0.0-beta.5", "tag" = "v12.0.0-beta.5", "git" = "https://github.com/lance-format/lance.git" }
-lance-testing = { "version" = "=12.0.0-beta.5", "tag" = "v12.0.0-beta.5", "git" = "https://github.com/lance-format/lance.git" }
-lance-datafusion = { "version" = "=12.0.0-beta.5", "tag" = "v12.0.0-beta.5", "git" = "https://github.com/lance-format/lance.git" }
-lance-encoding = { "version" = "=12.0.0-beta.5", "tag" = "v12.0.0-beta.5", "git" = "https://github.com/lance-format/lance.git" }
-lance-arrow = { "version" = "=12.0.0-beta.5", "tag" = "v12.0.0-beta.5", "git" = "https://github.com/lance-format/lance.git" }
+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" }
 lancedb = { path = "rust/lancedb", default-features = false }
 ahash = "0.8"
 # Note that this one does not include pyarrow
@@ -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 && pnpm dlx license-checker@25 --markdown --out NODEJS_THIRD_PARTY_LICENSES.md
+	cd nodejs && npx license-checker --markdown --out NODEJS_THIRD_PARTY_LICENSES.md
 	cd java && ./mvnw license:aggregate-add-third-party -q
@@ -12,12 +12,16 @@ 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
+  git add Cargo.lock nodejs/package-lock.json
  git commit --amend --no-edit
 else
-  git add Cargo.lock
+  git add Cargo.lock nodejs/package-lock.json
  git commit -m "Update lockfiles"
 fi
@@ -131,13 +131,18 @@ 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",
@@ -145,7 +150,12 @@ 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 = []
+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" },
+]
 # Crates whose license cannot be determined from Cargo metadata but whose
 # license we've manually confirmed from upstream. Keep this list minimal.
 [[licenses.clarify]]
@@ -47,24 +47,22 @@ pytest -vv python/tests/docs

 ### Checking typescript examples

-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.
+The `@lancedb/lancedb` package must be built before running the tests:

 ```shell
 pushd nodejs
-pnpm install
-pnpm build
+npm ci
+npm run build
 popd
 ```

-Then you can run the examples by going to the `nodejs/examples` directory, which is a
-separate pnpm package with its own lockfile:
+Then you can run the examples by going to the `nodejs/examples` directory and
+running the tests like a normal npm package:

 ```shell
 pushd nodejs/examples
-pnpm install
-pnpm test
+npm ci
+npm test
 popd
 ```

@@ -86,7 +84,6 @@ The new files should be checked into the repository.

 ```shell
 pushd nodejs
-# `pnpm docs` would invoke pnpm's built-in `docs` command, not the script.
-pnpm run docs
+npm run docs
 popd
 ```
@@ -0,0 +1,135 @@
+{
+  "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
+    }
+  }
+}
@@ -0,0 +1,20 @@
+{
+  "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"
+  }
+}
@@ -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.14</version>
+    <version>0.38.0-beta.12</version>
 </dependency>
 ```

@@ -0,0 +1,17 @@
+{
+  "include": [
+    "src/*.ts",
+  ],
+  "compilerOptions": {
+    "target": "es2022",
+    "module": "nodenext",
+    "declaration": true,
+    "outDir": "./dist",
+    "strict": true,
+    "allowJs": true,
+    "resolveJsonModule": true,
+  },
+  "exclude": [
+    "./dist/*",
+  ]
+}
@@ -8,7 +8,7 @@
    <parent>
      <groupId>com.lancedb</groupId>
      <artifactId>lancedb-parent</artifactId>
-      <version>0.38.0-beta.14</version>
+      <version>0.38.0-beta.12</version>
      <relativePath>../pom.xml</relativePath>
    </parent>

@@ -6,7 +6,7 @@

    <groupId>com.lancedb</groupId>
    <artifactId>lancedb-parent</artifactId>
-    <version>0.38.0-beta.14</version>
+    <version>0.38.0-beta.12</version>
    <packaging>pom</packaging>
    <name>${project.artifactId}</name>
    <description>LanceDB Java SDK Parent POM</description>
@@ -28,7 +28,7 @@
    <properties>
        <project.build.sourceEncoding>UTF-8</project.build.sourceEncoding>
        <arrow.version>15.0.0</arrow.version>
-        <lance-core.version>12.0.0-beta.5</lance-core.version>
+        <lance-core.version>12.0.0-beta.2</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>
@@ -1,7 +1,7 @@
 [package]
 name = "lancedb-nodejs"
 edition.workspace = true
-version = "0.38.0-beta.14"
+version = "0.38.0-beta.12"
 publish = false
 license.workspace = true
 description.workspace = true
@@ -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(">= 22");
-    expect(packageJson.peerDependencies["@types/node"]).toBe(">=22");
+    expect(packageJson.engines.node).toBe(">= 18");
+    expect(packageJson.peerDependencies["@types/node"]).toBe(">=18");
    expect(packageJson.peerDependenciesMeta["@types/node"]).toEqual({
      optional: true,
    });
@@ -3,7 +3,6 @@

 import * as http from "http";
 import { RequestListener } from "http";
-import packageJson = require("../package.json");
 import {
  ClientConfig,
  Connection,
@@ -71,13 +70,7 @@ async function withMockDatabase(
  try {
    await callback(db);
  } finally {
-    // `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());
-    });
+    server.close();
  }
 }

@@ -138,7 +131,7 @@ describe("remote connection", () => {
      (req, res) => {
        expect(req.headers["x-api-key"]).toEqual("fake");
        expect(req.headers["user-agent"]).toEqual(
-          `LanceDB-Node-Client/${packageJson.version}`,
+          `LanceDB-Node-Client/${process.env.npm_package_version}`,
        );

        const body = JSON.stringify({ tables: [] });
@@ -8,8 +8,7 @@
    "//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",
-    "//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",
+    "test": "node --experimental-vm-modules node_modules/.bin/jest --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",
@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-darwin-arm64",
-	"version": "0.38.0-beta.14",
+	"version": "0.38.0-beta.12",
 	"os": ["darwin"],
 	"cpu": ["arm64"],
 	"main": "lancedb.darwin-arm64.node",
@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-linux-arm64-gnu",
-	"version": "0.38.0-beta.14",
+	"version": "0.38.0-beta.12",
 	"os": ["linux"],
 	"cpu": ["arm64"],
 	"main": "lancedb.linux-arm64-gnu.node",
@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-linux-arm64-musl",
-	"version": "0.38.0-beta.14",
+	"version": "0.38.0-beta.12",
 	"os": ["linux"],
 	"cpu": ["arm64"],
 	"main": "lancedb.linux-arm64-musl.node",
@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-linux-x64-gnu",
-	"version": "0.38.0-beta.14",
+	"version": "0.38.0-beta.12",
 	"os": ["linux"],
 	"cpu": ["x64"],
 	"main": "lancedb.linux-x64-gnu.node",
@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-linux-x64-musl",
-	"version": "0.38.0-beta.14",
+	"version": "0.38.0-beta.12",
 	"os": ["linux"],
 	"cpu": ["x64"],
 	"main": "lancedb.linux-x64-musl.node",
@@ -1,6 +1,6 @@
 {
  "name": "@lancedb/lancedb-win32-arm64-msvc",
-  "version": "0.38.0-beta.14",
+  "version": "0.38.0-beta.12",
  "os": [
    "win32"
  ],
@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-win32-x64-msvc",
-	"version": "0.38.0-beta.14",
+	"version": "0.38.0-beta.12",
 	"os": ["win32"],
 	"cpu": ["x64"],
 	"main": "lancedb.win32-x64-msvc.node",
@@ -11,7 +11,7 @@
    "ann"
  ],
  "private": false,
-  "version": "0.38.0-beta.14",
+  "version": "0.38.0-beta.12",
  "main": "dist/index.js",
  "exports": {
    ".": "./dist/index.js",
@@ -67,7 +67,7 @@
    "timeout": "3m"
  },
  "engines": {
-    "node": ">= 22"
+    "node": ">= 18"
  },
  "packageManager": "pnpm@11.1.1",
  "cpu": ["x64", "arm64"],
@@ -101,7 +101,7 @@
    "openai": "4.29.2"
  },
  "peerDependencies": {
-    "@types/node": ">=22",
+    "@types/node": ">=18",
    "apache-arrow": ">=15.0.0 <=18.1.0"
  },
  "peerDependenciesMeta": {
@@ -1,6 +1,6 @@
 [package]
 name = "lancedb-python"
-version = "0.38.0-beta.14"
+version = "0.38.0-beta.12"
 publish = false
 edition.workspace = true
 description = "Python bindings for LanceDB"
@@ -54,18 +54,6 @@ _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")
@@ -251,23 +239,6 @@ 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):
@@ -276,28 +247,18 @@ 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.

-    The GPU execution requirement is part of this identity. CPU and memory sizing,
-    priority, concurrency, and retry policy belong to the execution platform.
+    Scheduling resources, priority, concurrency, and retry policy belong to
+    the submitting Job and are not part of this identity.
    """

    name: str
@@ -541,107 +502,31 @@ _GRAMMAR_PRIMITIVES = (


 def _canonical_arrow_type(data_type: pa.DataType) -> str:
-    """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]:
+    """The server's V1 Function type grammar. Anything outside it is rejected
+    here rather than at registration."""
    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}<{item}>"
+        return f"{prefix}<{_canonical_list_item(data_type)}>"
    if pa.types.is_fixed_size_list(data_type) and data_type.list_size > 0:
-        item = _grammar_list_item(data_type)
-        if item is not None:
-            return f"fixed_size_list<{item}, {data_type.list_size}>"
-    return None
+        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}")


-def _grammar_list_item(data_type: pa.DataType) -> Optional[str]:
+def _canonical_list_item(data_type: pa.DataType) -> str:
    """The grammar names only the item type; it always means a non-nullable
-    child called `item`, so other child properties require exact JSON."""
+    child called `item`, so any other child metadata cannot be represented."""
    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: field names "
-            "must not be empty"
+            "unsupported Arrow type for Function signature: list items must be a "
+            f"non-nullable field named 'item', got {child}"
        )
-    if field.metadata:
-        raise TypeError(
-            "unsupported Arrow type for Function signature: field metadata "
-            f"is not supported, got {field}"
-        )
-
-
-def _exact_arrow_field(field: pa.Field) -> dict[str, Any]:
-    _validate_exact_arrow_field(field)
-    return {
-        "name": field.name,
-        "nullable": field.nullable,
-        "type": _exact_arrow_type(field.type),
-    }
-
-
-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}")
+    return _canonical_arrow_type(child.type)


 def _list_of(item: pa.DataType) -> pa.DataType:
@@ -715,11 +600,8 @@ 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):
-        _validate_exact_arrow_field(output)
        if output.nullable:
            raise ValueError("Function output must be non-nullable")
        fields = tuple(output.type)
@@ -735,7 +617,6 @@ 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(
@@ -748,8 +629,6 @@ 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")
@@ -778,10 +657,6 @@ 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:
@@ -1035,7 +910,6 @@ class UdfDefinition:
        pip: tuple[str, ...],
        env: Mapping[str, str],
        python_version: Optional[str],
-        gpu: bool = False,
        conda: tuple[str, ...] = (),
        conda_channels: tuple[str, ...] = (),
    ):
@@ -1064,14 +938,12 @@ 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_v2" if gpu_marker is not None else "python",
+            kind="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(
@@ -1117,7 +989,6 @@ 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]: ...
@@ -1132,7 +1003,6 @@ 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] = (),
 ):
@@ -1165,10 +1035,6 @@ 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
@@ -1193,11 +1059,6 @@ 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:
@@ -1209,7 +1070,6 @@ 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),
        )
@@ -67,15 +67,7 @@ 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


@@ -559,12 +559,18 @@ def _coerce_blob_list_values(


 def _coerce_value_to_blob(values: pa.Array, target_field: pa.Field) -> pa.Array:
-    if pa.types.is_null(values.type):
-        data = pa.nulls(len(values), type=pa.large_binary())
+    if _is_string_like(values.type):
+        carrier_name = "uri"
+        carrier = values
+    elif pa.types.is_null(values.type):
+        carrier_name = None
+        carrier = None
    elif pa.types.is_large_binary(values.type):
-        data = values
+        carrier_name = "data"
+        carrier = values
    else:
-        data = values.cast(pa.large_binary())
+        carrier_name = "data"
+        carrier = values.cast(pa.large_binary())
    length = len(values)
    storage_type = target_field.type
    if isinstance(storage_type, pa.ExtensionType):
@@ -572,8 +578,8 @@ def _coerce_value_to_blob(values: pa.Array, target_field: pa.Field) -> pa.Array:
    storage_fields = list(storage_type)
    children = []
    for storage_field in storage_fields:
-        if storage_field.name == "data":
-            children.append(data)
+        if storage_field.name == carrier_name:
+            children.append(carrier.cast(storage_field.type))
        else:
            children.append(pa.nulls(length, type=storage_field.type))
    storage = pa.StructArray.from_arrays(
@@ -593,7 +599,11 @@ def _physical_array_and_type(array: pa.Array) -> tuple[pa.Array, pa.DataType]:


 def _can_coerce_to_blob(data_type: pa.DataType) -> bool:
-    return _is_binary_like(data_type) or pa.types.is_null(data_type)
+    return (
+        _is_binary_like(data_type)
+        or _is_string_like(data_type)
+        or pa.types.is_null(data_type)
+    )


 def _is_binary_like(data_type: pa.DataType) -> bool:
@@ -604,6 +614,15 @@ def _is_binary_like(data_type: pa.DataType) -> bool:
    )


+def _is_string_like(data_type: pa.DataType) -> bool:
+    predicates = ("is_string", "is_large_string", "is_string_view")
+    return any(
+        predicate(data_type)
+        for name in predicates
+        if (predicate := getattr(pa.types, name, None)) is not None
+    )
+
+
 def _field_extension_name(field: pa.Field) -> Optional[str]:
    extension_name = getattr(field.type, "extension_name", None)
    if extension_name is not None:
@@ -618,6 +637,187 @@ def _field_extension_name(field: pa.Field) -> Optional[str]:
    return extension_name


+_JSON_EXTENSION_NAMES = {"arrow.json", "lance.json"}
+_BLOB_EXTENSION_NAME = "lance.blob.v2"
+
+
+def _field_contains_write_extension(field: pa.Field) -> bool:
+    extension_name = _field_extension_name(field)
+    if (
+        extension_name in _JSON_EXTENSION_NAMES
+        or extension_name == _BLOB_EXTENSION_NAME
+    ):
+        return True
+    if pa.types.is_struct(field.type):
+        return any(_field_contains_write_extension(child) for child in field.type)
+    if (
+        pa.types.is_list(field.type)
+        or pa.types.is_large_list(field.type)
+        or pa.types.is_fixed_size_list(field.type)
+    ):
+        return _field_contains_write_extension(field.type.value_field)
+    return False
+
+
+def _with_field_type(
+    field: pa.Field,
+    data_type: pa.DataType,
+    *,
+    name: Optional[str] = None,
+    metadata: Optional[dict] = None,
+) -> pa.Field:
+    return pa.field(
+        name or field.name,
+        data_type,
+        nullable=field.nullable,
+        metadata=field.metadata if metadata is None else metadata,
+    )
+
+
+def _with_list_value_field(
+    data_type: pa.DataType, value_field: pa.Field
+) -> pa.DataType:
+    if pa.types.is_list(data_type):
+        return pa.list_(value_field)
+    if pa.types.is_large_list(data_type):
+        return pa.large_list(value_field)
+    return pa.list_(value_field, data_type.list_size)
+
+
+def _extension_storage_field(field: pa.Field) -> pa.Field:
+    """Return a from-pylist-compatible field for nested write extensions."""
+    extension_name = _field_extension_name(field)
+    if extension_name in _JSON_EXTENSION_NAMES:
+        metadata = dict(field.metadata or {})
+        metadata[b"ARROW:extension:name"] = b"arrow.json"
+        return _with_field_type(field, pa.string(), metadata=metadata)
+    if extension_name == _BLOB_EXTENSION_NAME:
+        metadata = dict(field.metadata or {})
+        metadata[b"ARROW:extension:name"] = _BLOB_EXTENSION_NAME.encode()
+        metadata[b"ARROW:extension:metadata"] = b""
+        storage_type = getattr(field.type, "storage_type", field.type)
+        return _with_field_type(field, storage_type, metadata=metadata)
+    if pa.types.is_struct(field.type):
+        children = [_extension_storage_field(child) for child in field.type]
+        return _with_field_type(field, pa.struct(children))
+    if _is_list_like(field.type):
+        value_field = _extension_storage_field(field.type.value_field)
+        return _with_field_type(field, _with_list_value_field(field.type, value_field))
+    return field
+
+
+def _prepare_extension_field(
+    field: pa.Field, target_field: pa.Field
+) -> Tuple[pa.Field, bool]:
+    extension_name = _field_extension_name(target_field)
+    if extension_name in _JSON_EXTENSION_NAMES:
+        metadata = dict(field.metadata or {})
+        metadata[b"ARROW:extension:name"] = b"arrow.json"
+        return _with_field_type(field, pa.string(), metadata=metadata), True
+    if extension_name == _BLOB_EXTENSION_NAME and pa.types.is_null(field.type):
+        return _with_field_type(field, pa.large_binary()), True
+
+    if pa.types.is_struct(field.type) and pa.types.is_struct(target_field.type):
+        target_children = {child.name: child for child in target_field.type}
+        children = []
+        changed = False
+        for child in field.type:
+            target_child = target_children.get(child.name)
+            if target_child is None:
+                children.append(child)
+                continue
+            prepared, child_changed = _prepare_extension_field(child, target_child)
+            children.append(prepared)
+            changed = changed or child_changed
+        if changed:
+            return _with_field_type(field, pa.struct(children)), True
+
+    if _is_list_like(field.type) and _is_list_like(target_field.type):
+        target_value_field = target_field.type.value_field
+        if _field_contains_write_extension(target_value_field):
+            prepared = _extension_storage_field(target_value_field)
+            data_type = _with_list_value_field(target_field.type, prepared)
+            return _with_field_type(field, data_type), True
+
+    return field, False
+
+
+def _prepare_extension_value(
+    value: Any, target_field: pa.Field, *, within_list: bool = False
+) -> Any:
+    """Shape raw nested blob values for PyArrow's struct construction."""
+    if value is None:
+        return None
+
+    extension_name = _field_extension_name(target_field)
+    if extension_name == _BLOB_EXTENSION_NAME and within_list:
+        if isinstance(value, (bytes, bytearray, memoryview)):
+            return {"data": value}
+        if isinstance(value, str):
+            return {"uri": value}
+        return value
+
+    if pa.types.is_struct(target_field.type) and isinstance(value, dict):
+        target_children = {child.name: child for child in target_field.type}
+        return {
+            name: _prepare_extension_value(
+                child_value, target_children[name], within_list=within_list
+            )
+            if name in target_children
+            else child_value
+            for name, child_value in value.items()
+        }
+
+    if _is_list_like(target_field.type) and isinstance(value, (list, tuple)):
+        return [
+            _prepare_extension_value(
+                item, target_field.type.value_field, within_list=True
+            )
+            for item in value
+        ]
+
+    return value
+
+
+def _prepare_extension_list(data: DATA, target_schema: pa.Schema) -> DATA:
+    """Give inferred list columns the logical type required by extensions."""
+    if not isinstance(data, list) or not data or not isinstance(data[0], dict):
+        return data
+
+    target_fields = {field.name: field for field in target_schema}
+    if not any(
+        _field_contains_write_extension(field) for field in target_fields.values()
+    ):
+        return data
+
+    inferred = pa.Table.from_pylist(data)
+    fields = []
+    changed = False
+    for field in inferred.schema:
+        target_field = target_fields.get(field.name)
+        if target_field is None:
+            fields.append(field)
+            continue
+        prepared, field_changed = _prepare_extension_field(field, target_field)
+        fields.append(prepared)
+        changed = changed or field_changed
+
+    if not changed:
+        return inferred
+
+    insert_schema = pa.schema(fields, metadata=inferred.schema.metadata)
+    prepared_data = [
+        {
+            name: _prepare_extension_value(value, target_fields[name])
+            if name in target_fields
+            else value
+            for name, value in row.items()
+        }
+        for row in data
+    ]
+    return pa.Table.from_pylist(prepared_data, schema=insert_schema)
+
+
 def _align_field_types(
    fields: List[pa.Field],
    target_fields: List[pa.Field],
@@ -2165,11 +2365,9 @@ class Table(ABC):
            Function columns are supported only on LanceDB Cloud and
            Enterprise.
        computed: Dict[str, str], optional
-            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.
+            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.

            Unlike ``transforms``, the expression is stored rather than
            evaluated now: the column is committed with no values, and rows get
@@ -5626,6 +5824,9 @@ class AsyncTable:
        if fill_value is None:
            fill_value = 0.0

+        if mode != "overwrite":
+            data = _prepare_extension_list(data, schema)
+
        # _santitize_data is an old code path, but we will use it until the
        # new code path is ready.
        if mode == "overwrite":
@@ -6270,11 +6471,8 @@ class AsyncTable:
            Function columns are supported only on LanceDB Cloud and
            Enterprise.
        computed: Dict[str, str], optional
-            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.
+            A map of column name to a SQL expression defining the column. The
+            column's type and inputs are derived from the expression.

            Unlike ``transforms``, the expression is stored rather than
            evaluated now: the column is committed with no values, and rows get
@@ -710,6 +710,80 @@ def test_fetch_blobs_preserves_null_and_empty_values():
    assert blobs[3].as_py() == b"present"


+def test_add_all_null_list_to_blob_column():
+    table = _blob_table("all_null_add", [{"id": 1, "image": None}])
+
+    hits = table.search().to_arrow()
+    blobs = table.fetch_blobs("image", hits)
+    assert len(blobs) == 1
+    assert blobs[0].as_py() is None
+
+
+def test_add_all_null_list_to_blob_column_with_sanitizer():
+    db = lancedb.connect("memory:///")
+    schema = pa.schema([pa.field("id", pa.int64()), lancedb.blob("image")])
+    table = db.create_table("all_null_sanitized_add", schema=schema)
+
+    table.add([{"id": 1, "image": None}], on_bad_vectors="fill")
+
+    hits = table.search().to_arrow()
+    blobs = table.fetch_blobs("image", hits)
+    assert len(blobs) == 1
+    assert blobs[0].as_py() is None
+
+
+def test_add_all_null_list_to_nested_blob_column():
+    db = lancedb.connect("memory:///")
+    blob_field = lancedb.blob("image")
+    info_field = pa.field("info", pa.struct([blob_field]))
+    info = pa.StructArray.from_arrays(
+        [_blob_array("image", [b"seed"])], fields=[blob_field]
+    )
+    seed = pa.Table.from_arrays(
+        [pa.array([0], type=pa.int64()), info],
+        schema=pa.schema([pa.field("id", pa.int64()), info_field]),
+    )
+    table = db.create_table("nested_null_add", data=seed)
+
+    table.add([{"id": 1, "info": {"image": None}}])
+    table.add([{"id": 2, "info": {"image": None}}], on_bad_vectors="fill")
+
+    hits = table.search().where("id > 0").to_arrow()
+    blobs = table.fetch_blobs("info.image", hits)
+    assert len(blobs) == 2
+    assert all(blob.as_py() is None for blob in blobs)
+
+
+@pytest.mark.parametrize("large_list", [False, True], ids=["list", "large_list"])
+def test_add_list_of_dicts_to_blob_list_column(large_list):
+    db = lancedb.connect("memory:///")
+    blob_field = lancedb.blob("image")
+    blob_values = _blob_array("image", [b"seed"])
+    if large_list:
+        items_field = pa.field("items", pa.large_list(blob_field))
+        items = pa.LargeListArray.from_arrays(
+            pa.array([0, 1], type=pa.int64()), blob_values
+        )
+    else:
+        items_field = pa.field("items", pa.list_(blob_field))
+        items = pa.ListArray.from_arrays(pa.array([0, 1], type=pa.int32()), blob_values)
+    seed = pa.Table.from_arrays(
+        [pa.array([0], type=pa.int64()), items],
+        schema=pa.schema([pa.field("id", pa.int64()), items_field]),
+    )
+    table = db.create_table(f"blob_{large_list}_list_add", data=seed)
+
+    table.add([{"id": 1, "items": [None]}])
+    table.add(
+        [{"id": 2, "items": [b"a", None]}],
+        on_bad_vectors="fill",
+    )
+
+    ids = table.search().select(["id"]).to_arrow()["id"].to_pylist()
+    assert sorted(ids) == [0, 1, 2]
+    assert pa.types.is_large_list(table.schema.field("items").type) is large_list
+
+
 def test_fetch_blob_ranges_aligns_repeated_ranges_and_nulls():
    table = _blob_table(
        "range_alignment",
@@ -1230,6 +1304,7 @@ def test_add_external_uri_string_round_trips_with_flag(tmp_path):
    table = db.create_table("external_string", schema=schema)
    table.add(
        [{"id": 1, "image": blob_path.as_uri()}],
+        on_bad_vectors="fill",
        allow_external_blob_outside_bases=True,
    )

@@ -19,7 +19,7 @@ import pyarrow as pa
 import pytest

 import lancedb
-from lancedb.functions import PythonRuntimeSpec, UdfDefinition, udf
+from lancedb.functions import UdfDefinition, udf

 THRESHOLD = 20
 _CACHE = None
@@ -89,58 +89,6 @@ 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:
@@ -220,7 +168,7 @@ def test_udf_resolves_module_globals_before_builtins(tmp_path):
        udf(module.uses_callable_shadow)


-def test_canonical_arrow_type_prefers_the_compact_grammar():
+def test_canonical_arrow_type_is_exactly_the_grammar():
    from lancedb.functions import _GRAMMAR_PRIMITIVES, _canonical_arrow_type

    golden = json.loads(
@@ -233,13 +181,6 @@ def test_canonical_arrow_type_prefers_the_compact_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),
@@ -247,6 +188,7 @@ def test_canonical_arrow_type_prefers_the_compact_grammar():
        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")),
    ]:
@@ -436,29 +378,14 @@ def test_udf_recursion_versus_a_rebound_module_name(tmp_path):
        udf(raw_fact)


-def test_canonical_arrow_type_uses_exact_json_for_list_child_properties():
+def test_canonical_arrow_type_rejects_unrepresentable_list_children():
    from lancedb.functions import _canonical_arrow_type

-    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)
@@ -468,29 +395,6 @@ def test_canonical_arrow_type_uses_exact_json_for_list_child_properties():
        )
        == "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:
@@ -578,105 +482,6 @@ def test_explicit_arrow_schema_is_deterministic():
    assert signature.output.nullable is False


-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"):

@@ -720,72 +525,6 @@ 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)
@@ -786,6 +786,55 @@ async def test_add_async(mem_db_async: AsyncConnection):
    assert await table.count_rows() == 3


+@pytest.mark.skipif(not hasattr(pa, "json_"), reason="requires PyArrow JSON type")
+@pytest.mark.asyncio
+@pytest.mark.parametrize(
+    ("values", "expected"),
+    [
+        ([None], [None]),
+        ([None, '{"k": 1}'], [None, '{"k":1}']),
+        (['{"k": 2}'], ['{"k":2}']),
+    ],
+)
+async def test_add_list_of_dicts_to_json_column(
+    mem_db_async: AsyncConnection, values, expected
+):
+    schema = pa.schema([pa.field("id", pa.int64()), pa.field("value", pa.json_())])
+    table = await mem_db_async.create_table("json_list_add", schema=schema)
+
+    await table.add([{"id": idx, "value": value} for idx, value in enumerate(values)])
+
+    rows = (await table.to_arrow()).sort_by("id").to_pylist()
+    assert [row["value"] for row in rows] == expected
+
+
+@pytest.mark.skipif(not hasattr(pa, "json_"), reason="requires PyArrow JSON type")
+@pytest.mark.asyncio
+async def test_add_list_of_dicts_to_nested_json_column(
+    mem_db_async: AsyncConnection,
+):
+    json_field = pa.field("value", pa.json_())
+    info_field = pa.field("info", pa.struct([json_field]))
+    info = pa.StructArray.from_arrays(
+        [pa.array(['{"seed": 0}'], type=pa.json_())], fields=[json_field]
+    )
+    seed = pa.Table.from_arrays(
+        [pa.array([0], type=pa.int64()), info],
+        schema=pa.schema([pa.field("id", pa.int64()), info_field]),
+    )
+    table = await mem_db_async.create_table("nested_json_list_add", data=seed)
+
+    await table.add([{"id": 1, "info": {"value": '{"k": 1}'}}])
+    await table.add([{"id": 2, "info": {"value": '{"k": 2}'}}], on_bad_vectors="fill")
+
+    rows = (await table.to_arrow()).sort_by("id").to_pylist()
+    assert rows == [
+        {"id": 0, "info": {"value": '{"seed":0}'}},
+        {"id": 1, "info": {"value": '{"k":1}'}},
+        {"id": 2, "info": {"value": '{"k":2}'}},
+    ]
+
+
 def test_add_overwrite_infers_vector_schema(mem_db: DBConnection):
    """Overwrite should infer vector columns the same way create_table does.

@@ -4087,29 +4136,6 @@ 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)
@@ -1,6 +1,6 @@
 [package]
 name = "lancedb"
-version = "0.38.0-beta.14"
+version = "0.38.0-beta.12"
 edition.workspace = true
 description = "LanceDB: A serverless, low-latency vector database for AI applications"
 license.workspace = true
@@ -207,33 +207,6 @@ 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
@@ -246,27 +219,22 @@ impl PythonRuntimeSpec {
    pub fn kind(&self) -> &str {
        match self {
            Self::Python { .. } => "python",
-            Self::PythonV2 { .. } => "python_v2",
            Self::Unrecognized { kind } => kind,
        }
    }

-    /// The Python version for a known Python runtime, or `None` for an unknown kind.
+    /// The Python version for the V1 runtime, or `None` for an unknown kind.
    pub fn python_version(&self) -> Option<&str> {
        match self {
-            Self::Python { python_version, .. } | Self::PythonV2 { python_version, .. } => {
-                Some(python_version)
-            }
+            Self::Python { python_version, .. } => Some(python_version),
            Self::Unrecognized { .. } => None,
        }
    }

-    /// The Python environment for a known Python runtime, or `None` for an unknown kind.
+    /// The Python environment for the V1 runtime, or `None` for an unknown kind.
    pub fn environment(&self) -> Option<&PythonEnvironmentSpec> {
        match self {
-            Self::Python { environment, .. } | Self::PythonV2 { environment, .. } => {
-                Some(environment)
-            }
+            Self::Python { environment, .. } => Some(environment),
            Self::Unrecognized { .. } => None,
        }
    }
@@ -274,73 +242,38 @@ impl PythonRuntimeSpec {
    /// Environment variables, or `None` for an unknown kind.
    pub fn env(&self) -> Option<&BTreeMap<String, String>> {
        match self {
-            Self::Python { env, .. } | Self::PythonV2 { env, .. } => Some(env),
+            Self::Python { 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 PythonRuntimeV1Wire {
-    python_version: String,
-    environment: PythonEnvironmentSpec,
+struct PythonRuntimeWire {
+    kind: String,
+    #[serde(default)]
+    python_version: Option<String>,
+    #[serde(default)]
+    environment: Option<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 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 }),
+        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 })
        }
    }
 }
@@ -354,8 +287,6 @@ 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)]
@@ -373,19 +304,6 @@ 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),
@@ -395,8 +313,8 @@ impl Serialize for PythonRuntimeSpec {

 /// Immutable Function version returned by the Enterprise catalog.
 ///
-/// The GPU execution requirement is part of this identity. CPU and memory sizing,
-/// priority, concurrency, and retry policy belong to the execution platform.
+/// Scheduling resources, priority, concurrency, and retry policy belong to
+/// the submitting Job and are not part of this identity.
 #[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
 pub struct FunctionVersion {
    name: String,
@@ -671,7 +589,7 @@ impl_json!(RefreshColumnResult);

 #[cfg(test)]
 mod conda_environment_tests {
-    use super::{PythonEnvironmentSpec, PythonRuntimeSpec};
+    use super::PythonEnvironmentSpec;

    #[test]
    fn conda_channels_round_trip_and_pip_stays_bare() {
@@ -690,45 +608,4 @@ 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"}"#
-            );
-        }
-    }
 }
@@ -3180,8 +3180,8 @@ impl<S: HttpSend> BaseTable for RemoteTable<S> {
            self.schema().await?.as_ref(),
            "schema evolution",
        )?;
-        // The server plans the declaration against its table schema, including
-        // Blob v2 semantics inherited by a direct field projection.
+        // The server plans the declaration: expression validation, type
+        // inference and the persisted binding all happen there.
        let entries = columns
            .iter()
            .map(
@@ -5734,8 +5734,9 @@ mod tests {
            ))
            .execute()
            .await;
-        let Err(err) = result else {
-            panic!("legacy remote query unexpectedly succeeded")
+        let err = match result {
+            Ok(_) => panic!("legacy remote query unexpectedly succeeded"),
+            Err(err) => err,
        };

        assert!(
@@ -7387,8 +7388,8 @@ mod tests {
        assert_eq!(result.version, if old_server { 0 } else { 43 });
    }

-    /// A declaration is sent as `{name, computed}` for the server to plan; the
-    /// client never types the expression itself.
+    /// A declaration is sent as `{name, computed}` entries 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() {
@@ -7464,93 +7465,6 @@ 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| {
@@ -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, while a remote one sends the
-    /// expression for the server to plan.
+    /// validates and types the expression itself, a remote one sends the text
+    /// for the server to plan.
    async fn add_computed_columns(
        &self,
        _columns: &[(String, String)],
@@ -5763,13 +5763,6 @@ 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`
@@ -9,35 +9,29 @@
 //! refresh fills the rows.
 //!
 //! The rule is tagged by kind ([`ComputedColumnKind`]) because kinds differ in
-//! where the column's type and inputs come from. A SQL expression determines
-//! its inputs and physical result type. A direct projection of a Blob v2 field
-//! also inherits that field's semantic type while execution continues to use
-//! `LargeBinary`. A kind resolved through a registry cannot be typed without
-//! consulting it.
-//! Registered Functions use an exact remote version plus a schema-level
-//! Function binding; unknown newer kinds remain readable and fail closed
-//! before mutation.
+//! where the column's type and inputs come from. A SQL expression is
+//! self-describing -- both are derived from the expression, so a caller writes
+//! neither -- while a kind resolved through a registry cannot be typed without
+//! consulting it. Registered Functions use an exact remote version plus a
+//! schema-level Function binding; unknown newer kinds remain readable and fail
+//! closed before mutation.
 //!
 //! [`computed_columns`] and [`computed_column_from_field`] read declarations
 //! back off a schema.

-use std::collections::{BTreeSet, HashMap, HashSet};
+use std::collections::{BTreeSet, HashMap};
 use std::sync::Arc;

 use arrow_schema::{DataType, Field as ArrowField, Fields, Schema as ArrowSchema, SchemaRef};
-use datafusion_common::{ScalarValue, tree_node::TreeNode};
-use datafusion_expr::Expr;
+use datafusion_common::tree_node::TreeNode;
 use datafusion_physical_plan::PhysicalExpr;
 use lance::dataset::NewColumnTransform;
-use lance_arrow::FieldExt;
-use lance_core::datatypes::{BLOB_V2_DESC_FIELD, format_field_path_minimal, parse_field_path};
 use lance_datafusion::planner::Planner;
 use lance_namespace::models::{JsonArrowDataType, JsonArrowField, JsonArrowSchema};
 use serde::{Deserialize, Serialize};
 use serde_json::Value;

 use crate::function::{FunctionApplication, FunctionBinding};
-use crate::utils::resolve_arrow_field_path;
 use crate::{Error, Result};

 /// Field metadata key marking a column as computed. The value is `"true"`.
@@ -547,12 +541,7 @@ fn ensure_known_binding_shape(value: &Value) -> Result<()> {
    Ok(())
 }

-struct ResolvedFieldPath<'a> {
-    root: &'a ArrowField,
-    leaf: &'a ArrowField,
-}
-
-fn resolve_field_path<'a>(schema: &'a ArrowSchema, path: &str) -> Result<ResolvedFieldPath<'a>> {
+fn resolve_field_path<'a>(schema: &'a ArrowSchema, path: &str) -> Result<&'a ArrowField> {
    let parts = lance_core::datatypes::parse_field_path(path).map_err(|e| {
        invalid_function(format!("invalid Function input field path '{path}': {e}"))
    })?;
@@ -561,33 +550,25 @@ fn resolve_field_path<'a>(schema: &'a ArrowSchema, path: &str) -> Result<Resolve
            "Function input field path cannot be empty",
        ));
    };
-    let root = schema
+    let mut field = schema
        .field_with_name(root)
        .map_err(|_| invalid_function(format!("unknown Function input column '{path}'")))?;
-    let mut leaf = root;
    for child in children {
-        let DataType::Struct(fields) = leaf.data_type() else {
+        let DataType::Struct(fields) = field.data_type() else {
            return Err(invalid_function(format!(
                "Function input field path '{path}' traverses a non-struct field"
            )));
        };
-        leaf = fields
+        field = fields
            .iter()
            .find(|field| field.name() == child)
            .map(AsRef::as_ref)
            .ok_or_else(|| invalid_function(format!("unknown Function input column '{path}'")))?;
    }
-    Ok(ResolvedFieldPath { root, leaf })
+    Ok(field)
 }

-fn canonical_input_arrow_type(field: &ArrowField) -> Result<String> {
-    if field.is_blob_v2() {
-        return Ok("blob_v2".to_string());
-    }
-    let schema =
-        lance_namespace::schema::arrow_schema_to_json(&ArrowSchema::new(vec![field.clone()]))
-            .map_err(|e| invalid_function(format!("invalid Function input schema: {e}")))?;
-    let field = schema.fields.into_iter().next().unwrap();
+fn canonical_input_arrow_type(field: &JsonArrowField) -> Result<String> {
    if field.r#type.fields.is_none() && field.r#type.length.is_none() {
        Ok(field.r#type.r#type.clone())
    } else {
@@ -676,95 +657,10 @@ fn parse_output_arrow_type(raw: &str) -> Result<JsonArrowDataType> {
    Ok(data_type)
 }

-fn parse_output_arrow_field(name: &str, nullable: bool, raw: &str) -> Result<JsonArrowField> {
-    if raw.trim() == "blob_v2" {
-        let schema =
-            lance_namespace::schema::arrow_schema_to_json(&ArrowSchema::new(vec![crate::blob(
-                name, nullable,
-            )]))
-            .map_err(|e| invalid_function(format!("invalid Function Blob type: {e}")))?;
-        return Ok(schema.fields.into_iter().next().unwrap());
-    }
-
-    fn parse_nested(name: &str, nullable: bool, raw: &str) -> Result<JsonArrowField> {
-        let raw = raw.trim();
-        if raw == "blob_v2" {
-            return parse_output_arrow_field(name, nullable, raw);
-        }
-        if let Some(inner) = raw
-            .strip_prefix("list<")
-            .and_then(|value| value.strip_suffix('>'))
-        {
-            let mut data_type = JsonArrowDataType::new("list".to_string());
-            data_type.fields = Some(vec![parse_nested("item", false, inner)?]);
-            return Ok(JsonArrowField::new(name.to_string(), nullable, data_type));
-        }
-        if let Some(inner) = raw
-            .strip_prefix("large_list<")
-            .and_then(|value| value.strip_suffix('>'))
-        {
-            let mut data_type = JsonArrowDataType::new("large_list".to_string());
-            data_type.fields = Some(vec![parse_nested("item", false, inner)?]);
-            return Ok(JsonArrowField::new(name.to_string(), nullable, data_type));
-        }
-        if let Some((inner, size)) = split_fixed_size_list(raw) {
-            let mut data_type = JsonArrowDataType::new("fixed_size_list".to_string());
-            data_type.fields = Some(vec![parse_nested("item", false, inner)?]);
-            data_type.length = Some(i64::from(size));
-            return Ok(JsonArrowField::new(name.to_string(), nullable, data_type));
-        }
-        Ok(JsonArrowField::new(
-            name.to_string(),
-            nullable,
-            parse_output_arrow_type(raw)?,
-        ))
-    }
-
-    let field = parse_nested(name, nullable, raw)?;
-    lance_namespace::schema::convert_json_arrow_field(&field)
-        .map_err(|e| invalid_function(format!("unsupported Function Arrow type '{raw}': {e}")))?;
-    Ok(field)
-}
-
-fn function_fields_equivalent(actual: &ArrowField, expected: &ArrowField) -> bool {
-    if actual.is_blob_v2() || expected.is_blob_v2() {
-        return actual.is_blob_v2()
-            && expected.is_blob_v2()
-            && actual.data_type() == expected.data_type();
-    }
-    match (actual.data_type(), expected.data_type()) {
-        (DataType::Struct(actual), DataType::Struct(expected)) => {
-            actual.len() == expected.len()
-                && actual.iter().zip(expected).all(|(actual, expected)| {
-                    actual.name() == expected.name()
-                        && actual.is_nullable() == expected.is_nullable()
-                        && function_fields_equivalent(actual, expected)
-                })
-        }
-        (DataType::List(actual), DataType::List(expected))
-        | (DataType::LargeList(actual), DataType::LargeList(expected)) => {
-            actual.name() == expected.name()
-                && actual.is_nullable() == expected.is_nullable()
-                && function_fields_equivalent(actual, expected)
-        }
-        (
-            DataType::FixedSizeList(actual, actual_size),
-            DataType::FixedSizeList(expected, expected_size),
-        ) => {
-            actual_size == expected_size
-                && actual.name() == expected.name()
-                && actual.is_nullable() == expected.is_nullable()
-                && function_fields_equivalent(actual, expected)
-        }
-        _ => actual.data_type() == expected.data_type(),
-    }
-}
-
 fn ensure_binding_matches_schema(schema: &ArrowSchema, binding: &FunctionBinding) -> Result<()> {
    let mut input_fields = Vec::with_capacity(binding.inputs().len());
    for input in binding.inputs() {
-        let resolved = resolve_field_path(schema, &input.field_path)?;
-        let field = resolved.leaf;
+        let field = resolve_field_path(schema, &input.field_path)?;
        if field
            .metadata()
            .get(COMPUTED_COLUMN_META_KEY)
@@ -792,7 +688,12 @@ fn ensure_binding_matches_schema(schema: &ArrowSchema, binding: &FunctionBinding
            input.nullable,
        )
        .with_metadata(field.metadata().clone());
-        if canonical_input_arrow_type(&parameter_field)? != input.arrow_type {
+        let json = lance_namespace::schema::arrow_schema_to_json(&ArrowSchema::new(vec![
+            parameter_field.clone(),
+        ]))
+        .map_err(|e| invalid_function(format!("invalid Function input schema: {e}")))?;
+        let json_field = json.fields.into_iter().next().unwrap();
+        if canonical_input_arrow_type(&json_field)? != input.arrow_type {
            return Err(invalid_function(format!(
                "Function input '{}' type no longer matches binding '{}'",
                input.field_path,
@@ -814,11 +715,6 @@ fn ensure_binding_matches_schema(schema: &ArrowSchema, binding: &FunctionBinding
        )));
    }

-    let expected_inputs = binding
-        .inputs()
-        .iter()
-        .map(|input| input.field_path.clone())
-        .collect::<Vec<_>>();
    let mut output_fields = Vec::with_capacity(binding.outputs().len());
    for output in binding.outputs() {
        let field = schema.field_with_name(&output.output_name).map_err(|_| {
@@ -835,39 +731,21 @@ fn ensure_binding_matches_schema(schema: &ArrowSchema, binding: &FunctionBinding
                binding.binding_id()
            )));
        }
-        let expected_field = parse_output_arrow_field(field.name(), true, &output.arrow_type)?;
-        let expected_field = lance_namespace::schema::convert_json_arrow_field(&expected_field)
+        let expected_type = parse_output_arrow_type(&output.arrow_type)?;
+        let expected_type = lance_namespace::schema::convert_json_arrow_type(&expected_type)
            .map_err(|e| invalid_function(format!("invalid Function output type: {e}")))?;
-        if !function_fields_equivalent(field, &expected_field) {
+        if field.data_type() != &expected_type {
            return Err(invalid_function(format!(
                "Function output '{}' type no longer matches binding '{}'",
                output.output_name,
                binding.binding_id()
            )));
        }
-        let metadata = field.metadata();
-        let declared_inputs = metadata
-            .get(INPUTS_META_KEY)
-            .and_then(|raw| serde_json::from_str::<Vec<String>>(raw).ok());
-        if metadata.get(COMPUTED_COLUMN_META_KEY).map(String::as_str) != Some("true")
-            || metadata.get(KIND_META_KEY).map(String::as_str) != Some(FUNCTION_KIND)
-            || metadata
-                .get(FUNCTION_BINDING_ID_META_KEY)
-                .map(String::as_str)
-                != Some(binding.binding_id())
-            || metadata
-                .get(FUNCTION_OUTPUT_ORDINAL_META_KEY)
-                .and_then(|value| value.parse::<u32>().ok())
-                != Some(output.output_ordinal)
-            || declared_inputs.as_deref() != Some(expected_inputs.as_slice())
-        {
-            return Err(invalid_function(format!(
-                "Function output '{}' declaration metadata does not match binding '{}'",
-                output.output_name,
-                binding.binding_id()
-            )));
-        }
-        output_fields.push(expected_field);
+        output_fields.push(ArrowField::new(
+            field.name().clone(),
+            field.data_type().clone(),
+            true,
+        ));
    }
    let output_schema =
        lance_namespace::schema::arrow_schema_to_json(&ArrowSchema::new(output_fields))
@@ -894,7 +772,7 @@ pub(crate) fn plan_function_application(
    application: &FunctionApplication,
    output_name: Option<&str>,
 ) -> Result<FunctionDeclarationPlan> {
-    ensure_supported_function_metadata(schema)?;
+    ensure_no_function_bindings_for_mutation(schema, "Function binding declaration")?;
    if application.has_unknown_fields() {
        return Err(Error::NotSupported {
            message: "Function application contains fields from a newer contract".into(),
@@ -944,9 +822,8 @@ pub(crate) fn plan_function_application(
                input.parameter
            ))
        })?;
-        let resolved = resolve_field_path(schema, path)?;
-        if resolved
-            .root
+        let field = resolve_field_path(schema, path)?;
+        if field
            .metadata()
            .get(COMPUTED_COLUMN_META_KEY)
            .map(String::as_str)
@@ -956,17 +833,21 @@ pub(crate) fn plan_function_application(
                "Function input '{path}' is computed; computed-on-computed bindings are not supported"
            )));
        }
-        let field = resolved.leaf;
        let parameter_field = ArrowField::new(
            input.parameter.clone(),
            field.data_type().clone(),
            field.is_nullable(),
        )
        .with_metadata(field.metadata().clone());
+        let input_schema = lance_namespace::schema::arrow_schema_to_json(&ArrowSchema::new(vec![
+            parameter_field.clone(),
+        ]))
+        .map_err(|e| invalid_function(format!("invalid Function input schema: {e}")))?;
+        let json_field = input_schema.fields.into_iter().next().unwrap();
        input_bindings.push(FunctionInputTarget {
            parameter: input.parameter.clone(),
            field_path: path.to_string(),
-            arrow_type: canonical_input_arrow_type(&parameter_field)?,
+            arrow_type: canonical_input_arrow_type(&json_field)?,
            nullable: field.is_nullable(),
        });
        input_fields.push(parameter_field);
@@ -995,19 +876,16 @@ pub(crate) fn plan_function_application(
                    "Function logical outputs must be non-nullable during NULL assignment",
                ));
            }
-            let output_field = parse_output_arrow_field(
-                name,
-                true,
-                output.arrow_type.as_deref().ok_or_else(|| {
+            let data_type =
+                parse_output_arrow_type(output.arrow_type.as_deref().ok_or_else(|| {
                    invalid_function("scalar Function output is missing its Arrow type")
-                })?,
-            )?;
+                })?)?;
            outputs.push(FunctionOutputTarget {
                result_field: WHOLE_RESULT_FIELD.to_string(),
                output_name: name.to_string(),
                output_ordinal: 0,
            });
-            output_fields.push(output_field);
+            output_fields.push(JsonArrowField::new(name.to_string(), true, data_type));
        }
        "named_struct" => {
            if output.fields.is_empty() {
@@ -1051,7 +929,13 @@ pub(crate) fn plan_function_application(
                let fields = output
                    .fields
                    .iter()
-                    .map(|field| parse_output_arrow_field(&field.name, false, &field.arrow_type))
+                    .map(|field| {
+                        Ok(JsonArrowField::new(
+                            field.name.clone(),
+                            false,
+                            parse_output_arrow_type(&field.arrow_type)?,
+                        ))
+                    })
                    .collect::<Result<Vec<_>>>()?;
                let mut data_type = JsonArrowDataType::new("struct".to_string());
                data_type.fields = Some(fields);
@@ -1078,7 +962,11 @@ pub(crate) fn plan_function_application(
                        output_name: name.clone(),
                        output_ordinal: ordinal as u32,
                    });
-                    output_fields.push(parse_output_arrow_field(name, true, &field.arrow_type)?);
+                    output_fields.push(JsonArrowField::new(
+                        name.clone(),
+                        true,
+                        parse_output_arrow_type(&field.arrow_type)?,
+                    ));
                }
            }
        }
@@ -1218,20 +1106,15 @@ pub(crate) fn ensure_no_foreign_declarations<'a>(
    fields: impl IntoIterator<Item = &'a Arc<ArrowField>>,
 ) -> Result<()> {
    for field in fields {
-        ensure_no_foreign_declaration(field)?;
-    }
-    Ok(())
-}
-
-fn ensure_no_foreign_declaration(field: &ArrowField) -> Result<()> {
-    if field.metadata().keys().any(|k| is_declaration_key(k)) {
-        return Err(Error::InvalidInput {
-            message: format!(
-                "field '{}' carries computed-column metadata; declare computed columns \
-                 with add_columns().computed()",
-                field.name()
-            ),
-        });
+        if field.metadata().keys().any(|k| is_declaration_key(k)) {
+            return Err(Error::InvalidInput {
+                message: format!(
+                    "field '{}' carries computed-column metadata; declare computed columns \
+                     with add_columns().computed()",
+                    field.name()
+                ),
+            });
+        }
    }
    Ok(())
 }
@@ -1279,154 +1162,15 @@ pub(crate) struct BoundExpression {
    /// The columns the expression names, as written; nested inputs keep
    /// their dotted path.
    pub inputs: Vec<String>,
-    /// The top-level columns evaluation reads, in physical-expression order.
-    /// A nested input appears through its root.
+    /// The top-level columns evaluation reads, in [`Self::read_schema`]
+    /// order. A nested input appears through its root.
    pub roots: Vec<String>,
+    /// The projected schema evaluation runs against.
+    pub read_schema: SchemaRef,
    /// The compiled expression.
    pub physical: Arc<dyn PhysicalExpr>,
    /// The type the expression yields.
    pub data_type: DataType,
-    /// Blob v2 leaves the scan must materialize as `LargeBinary`.
-    pub blob_paths: Vec<String>,
-    /// A directly projected Blob v2 field whose semantics the output inherits.
-    projected_blob_field: Option<ArrowField>,
-}
-
-fn is_direct_field_projection(expr: &Expr) -> bool {
-    match expr {
-        Expr::Column(_) => true,
-        Expr::ScalarFunction(function)
-            if function.name() == "get_field" && function.args.len() == 2 =>
-        {
-            is_direct_field_projection(&function.args[0])
-                && matches!(
-                    &function.args[1],
-                    Expr::Literal(ScalarValue::Utf8(Some(_)), _)
-                )
-        }
-        _ => false,
-    }
-}
-
-fn projected_blob_field(schema: &ArrowSchema, expr: &Expr) -> Result<Option<ArrowField>> {
-    if !is_direct_field_projection(expr) {
-        return Ok(None);
-    }
-    let paths = Planner::column_names_in_expr(expr);
-    let [path] = paths.as_slice() else {
-        return Ok(None);
-    };
-    let (_, field) = resolve_arrow_field_path(schema, path)?;
-    Ok(field.is_blob_v2().then_some(field))
-}
-
-fn collect_blob_paths(field: &ArrowField, parent: &[String], paths: &mut Vec<Vec<String>>) {
-    let mut path = parent.to_vec();
-    path.push(field.name().clone());
-    if field.is_blob_v2() {
-        paths.push(path);
-        return;
-    }
-    match field.data_type() {
-        DataType::Struct(children) => {
-            for child in children {
-                collect_blob_paths(child, &path, paths);
-            }
-        }
-        DataType::List(child)
-        | DataType::LargeList(child)
-        | DataType::FixedSizeList(child, _)
-        | DataType::Map(child, _) => collect_blob_paths(child, &path, paths),
-        _ => {}
-    }
-}
-
-fn schema_blob_paths(schema: &ArrowSchema) -> Vec<Vec<String>> {
-    let mut paths = Vec::new();
-    for field in schema.fields() {
-        collect_blob_paths(field, &[], &mut paths);
-    }
-    paths
-}
-
-fn transform_blob_field(
-    field: &ArrowField,
-    parent: &[String],
-    materialized: &HashSet<Vec<String>>,
-) -> ArrowField {
-    let mut path = parent.to_vec();
-    path.push(field.name().clone());
-    if field.is_blob_v2() {
-        if materialized.contains(&path) {
-            return ArrowField::new(field.name(), DataType::LargeBinary, field.is_nullable());
-        }
-        return ArrowField::new(
-            field.name(),
-            BLOB_V2_DESC_FIELD.data_type().clone(),
-            field.is_nullable(),
-        )
-        .with_metadata(BLOB_V2_DESC_FIELD.metadata().clone());
-    }
-
-    let data_type = match field.data_type() {
-        DataType::Struct(children) => DataType::Struct(
-            children
-                .iter()
-                .map(|child| Arc::new(transform_blob_field(child, &path, materialized)))
-                .collect(),
-        ),
-        DataType::List(child) => {
-            DataType::List(Arc::new(transform_blob_field(child, &path, materialized)))
-        }
-        DataType::LargeList(child) => {
-            DataType::LargeList(Arc::new(transform_blob_field(child, &path, materialized)))
-        }
-        DataType::FixedSizeList(child, size) => DataType::FixedSizeList(
-            Arc::new(transform_blob_field(child, &path, materialized)),
-            *size,
-        ),
-        DataType::Map(child, sorted) => DataType::Map(
-            Arc::new(transform_blob_field(child, &path, materialized)),
-            *sorted,
-        ),
-        _ => return field.clone(),
-    };
-    ArrowField::new(field.name(), data_type, field.is_nullable())
-        .with_metadata(field.metadata().clone())
-}
-
-fn blob_runtime_schema(schema: &ArrowSchema, materialized: &HashSet<Vec<String>>) -> SchemaRef {
-    Arc::new(ArrowSchema::new_with_metadata(
-        schema
-            .fields()
-            .iter()
-            .map(|field| Arc::new(transform_blob_field(field, &[], materialized)))
-            .collect::<Fields>(),
-        schema.metadata().clone(),
-    ))
-}
-
-fn referenced_blob_paths(schema: &ArrowSchema, inputs: &[String]) -> Result<Vec<Vec<String>>> {
-    let input_paths = inputs
-        .iter()
-        .map(|input| {
-            parse_field_path(input).map_err(|error| Error::InvalidInput {
-                message: format!("invalid computed-column input path '{input}': {error}"),
-            })
-        })
-        .collect::<Result<Vec<_>>>()?;
-    Ok(schema_blob_paths(schema)
-        .into_iter()
-        .filter(|blob_path| {
-            input_paths.iter().any(|input_path| {
-                input_path.len() <= blob_path.len()
-                    && input_path
-                        .iter()
-                        .zip(blob_path)
-                        .all(|(input, blob)| input == blob)
-            })
-        })
-        .collect())
 }

 /// Parse, resolve and compile `expression` against `schema`.
@@ -1441,18 +1185,10 @@ pub(crate) fn bind(schema: SchemaRef, column: &str, expression: &str) -> Result<
        message,
    };

-    // Blob v2 is a semantic type whose runtime expression ABI is
-    // `LargeBinary`. Parse against that ABI first so a direct Blob reference
-    // is not mistaken for its storage descriptor struct.
-    let all_blob_paths = schema_blob_paths(schema.as_ref())
-        .into_iter()
-        .collect::<HashSet<_>>();
-    let parsing_schema = blob_runtime_schema(schema.as_ref(), &all_blob_paths);
-    let planner = Planner::new(parsing_schema);
+    let planner = Planner::new(schema.clone());
    let parsed = planner
        .parse_expr(expression)
        .map_err(|e| invalid(e.to_string()))?;
-    let projected_blob_field = projected_blob_field(schema.as_ref(), &parsed)?;

    // A declaration is evaluated more than once -- staging and writing are
    // separate passes, and a refresh years later replays the same text -- so
@@ -1482,19 +1218,13 @@ pub(crate) fn bind(schema: SchemaRef, column: &str, expression: &str) -> Result<
    inputs.sort();
    inputs.dedup();

-    let blob_paths = referenced_blob_paths(schema.as_ref(), &inputs)?;
-    let runtime_schema = blob_runtime_schema(
-        schema.as_ref(),
-        &blob_paths.iter().cloned().collect::<HashSet<_>>(),
-    );
-
    // A nested input is recorded by its path but read through its root
    // column; Schema::index_of resolves top-level names only. Resolved here
    // rather than left to the planner so an unknown column names itself in
    // the error instead of surfacing as a plan failure.
    let mut indices = Vec::with_capacity(inputs.len());
    for input in &inputs {
-        let index = runtime_schema
+        let index = schema
            .index_of(root(input))
            .map_err(|_| invalid(format!("unknown column '{input}'")))?;
        if !indices.contains(&index) {
@@ -1507,7 +1237,7 @@ pub(crate) fn bind(schema: SchemaRef, column: &str, expression: &str) -> Result<
    // compiles the expression has to be built on the projected schema
    // evaluation will actually read.
    let read_schema = Arc::new(
-        runtime_schema
+        schema
            .project(&indices)
            .map_err(|e| invalid(e.to_string()))?,
    );
@@ -1517,8 +1247,7 @@ pub(crate) fn bind(schema: SchemaRef, column: &str, expression: &str) -> Result<
        .map(|field| field.name().clone())
        .collect();

-    let runtime_planner = Planner::new(runtime_schema);
-    let optimized = runtime_planner
+    let optimized = planner
        .optimize_expr(parsed)
        .map_err(|e| invalid(e.to_string()))?;
    let physical = Planner::new(read_schema.clone())
@@ -1531,16 +1260,9 @@ pub(crate) fn bind(schema: SchemaRef, column: &str, expression: &str) -> Result<
    Ok(BoundExpression {
        inputs,
        roots,
+        read_schema,
        physical,
        data_type,
-        blob_paths: blob_paths
-            .iter()
-            .map(|path| {
-                let segments = path.iter().map(String::as_str).collect::<Vec<_>>();
-                format_field_path_minimal(&segments)
-            })
-            .collect(),
-        projected_blob_field,
    })
 }

@@ -1556,7 +1278,7 @@ pub(crate) fn bind(schema: SchemaRef, column: &str, expression: &str) -> Result<
 /// batch may declare `a` and then `b = a + 1` in one commit. Refresh order
 /// then matters, and refresh enforces it: `b` is refused while `a` still has
 /// unfilled rows.
-fn plan_declarations(schema: SchemaRef, columns: &[(String, String)]) -> Result<Vec<ArrowField>> {
+pub(crate) fn plan(schema: SchemaRef, columns: &[(String, String)]) -> Result<Vec<ArrowField>> {
    if columns.is_empty() {
        return Err(Error::InvalidInput {
            message: "at least one computed column is required".into(),
@@ -1575,19 +1297,8 @@ fn plan_declarations(schema: SchemaRef, columns: &[(String, String)]) -> Result<

        // Declared columns start entirely null, so nullability is a property
        // of the declaration rather than of what the expression yields.
-        let computed_metadata = computed_column_metadata(expression, &bound.inputs);
-        let field = match bound.projected_blob_field {
-            Some(source) => {
-                let mut metadata = source.metadata().clone();
-                metadata.retain(|key, _| !is_declaration_key(key));
-                metadata.extend(computed_metadata);
-                source
-                    .with_name(name)
-                    .with_nullable(true)
-                    .with_metadata(metadata)
-            }
-            None => ArrowField::new(name, bound.data_type, true).with_metadata(computed_metadata),
-        };
+        let field = ArrowField::new(name, bound.data_type, true)
+            .with_metadata(computed_column_metadata(expression, &bound.inputs));
        schema = Arc::new(ArrowSchema::new_with_metadata(
            schema
                .fields()
@@ -1603,10 +1314,6 @@ fn plan_declarations(schema: SchemaRef, columns: &[(String, String)]) -> Result<
    Ok(fields)
 }

-pub(crate) fn plan(schema: SchemaRef, columns: &[(String, String)]) -> Result<Vec<ArrowField>> {
-    plan_declarations(schema, columns)
-}
-
 /// Run the schema-level checks of
 /// [`AddColumnsBuilder::computed`](super::AddColumnsBuilder::computed) against
 /// `schema` without committing: the Function-binding guard and the planning of
@@ -1645,7 +1352,7 @@ pub(crate) fn declare(
    schema: SchemaRef,
    columns: &[(String, String)],
 ) -> Result<NewColumnTransform> {
-    let fields = plan_declarations(schema, columns)?;
+    let fields = plan(schema, columns)?;
    Ok(NewColumnTransform::AllNulls(Arc::new(ArrowSchema::new(
        fields,
    ))))
@@ -1771,44 +1478,6 @@ mod tests {
        );
    }

-    #[test]
-    fn test_direct_blob_projection_inherits_semantics() {
-        let schema = Arc::new(ArrowSchema::new(vec![crate::blob("image", false)]));
-        let fields = plan(
-            schema,
-            &[
-                ("first".to_string(), "image".to_string()),
-                ("second".to_string(), "first".to_string()),
-            ],
-        )
-        .unwrap();
-
-        for field in &fields {
-            assert!(field.is_blob_v2());
-            assert!(field.is_nullable());
-        }
-        assert_eq!(
-            fields[1]
-                .metadata()
-                .get(EXPRESSION_META_KEY)
-                .map(String::as_str),
-            Some("first")
-        );
-    }
-
-    #[test]
-    fn test_blob_expression_transformation_does_not_inherit_semantics() {
-        let schema = Arc::new(ArrowSchema::new(vec![crate::blob("image", true)]));
-        let fields = plan(
-            schema,
-            &[("payload".to_string(), "coalesce(image, image)".to_string())],
-        )
-        .unwrap();
-
-        assert!(!fields[0].is_blob_v2());
-        assert_eq!(fields[0].data_type(), &DataType::LargeBinary);
-    }
-
    /// The binding reaches the schema only if `AllNulls` carries per-field
    /// metadata through the commit. The whole representation rests on it.
    #[tokio::test]
@@ -2676,44 +2345,6 @@ mod tests {
        .unwrap()
    }

-    fn blob_function_application() -> FunctionApplication {
-        FunctionApplication::from_json(
-            r#"{
-                "function":{"name":"copy_blob","version":"fv_exact"},
-                "inputs":[
-                    {"parameter":"image","kind":"column","value":{"path":"image"}}
-                ],
-                "output":{"kind":"scalar","arrow_type":"blob_v2","nullable":false}
-            }"#,
-        )
-        .unwrap()
-    }
-
-    #[test]
-    fn test_blob_function_input_uses_the_semantic_binding_type() {
-        let schema = ArrowSchema::new(vec![crate::blob("image", true)]);
-        let plan =
-            plan_function_application(&schema, &blob_function_application(), Some("copy")).unwrap();
-
-        assert_eq!(plan.input_bindings[0].arrow_type, "blob_v2");
-        let input = lance_namespace::schema::convert_json_arrow_field(&plan.input_schema.fields[0])
-            .unwrap();
-        assert!(input.is_blob_v2());
-    }
-
-    #[test]
-    fn test_blob_function_output_is_plannable() {
-        let schema = ArrowSchema::new(vec![crate::blob("image", true)]);
-        let plan =
-            plan_function_application(&schema, &blob_function_application(), Some("copy")).unwrap();
-
-        let output =
-            lance_namespace::schema::convert_json_arrow_field(&plan.output_schema.fields[0])
-                .unwrap();
-        assert_eq!(output.name(), "copy");
-        assert!(output.is_blob_v2());
-    }
-
    fn function_binding_schema(title_nullable: bool, body_nullable: bool) -> ArrowSchema {
        ArrowSchema::new(vec![
            ArrowField::new("title", DataType::Utf8, title_nullable),
@@ -2723,37 +2354,6 @@ mod tests {
        ])
    }

-    fn valid_function_binding_schema(
-        title_nullable: bool,
-        body_nullable: bool,
-        binding: &FunctionBinding,
-    ) -> ArrowSchema {
-        let mut fields = function_binding_schema(title_nullable, body_nullable)
-            .fields()
-            .iter()
-            .map(|field| field.as_ref().clone())
-            .collect::<Vec<_>>();
-        let inputs = binding
-            .inputs()
-            .iter()
-            .map(|input| input.field_path.clone())
-            .collect::<Vec<_>>();
-        for output in binding.outputs() {
-            let index = fields
-                .iter()
-                .position(|field| field.name() == &output.output_name)
-                .unwrap();
-            fields[index] = fields[index]
-                .clone()
-                .with_metadata(function_computed_column_metadata(
-                    binding.binding_id(),
-                    output.output_ordinal,
-                    &inputs,
-                ));
-        }
-        ArrowSchema::new(fields)
-    }
-
    #[test]
    fn test_non_nullable_function_inputs_can_bind_to_nullable_parameters() {
        let binding = FunctionBinding::from_json(include_str!(
@@ -2761,11 +2361,7 @@ mod tests {
        ))
        .unwrap();

-        ensure_binding_matches_schema(
-            &valid_function_binding_schema(false, false, &binding),
-            &binding,
-        )
-        .unwrap();
+        ensure_binding_matches_schema(&function_binding_schema(false, false), &binding).unwrap();
    }

    #[test]
@@ -2778,11 +2374,8 @@ mod tests {
        raw_binding["input_schema"]["fields"][0]["nullable"] = Value::Bool(false);
        let binding: FunctionBinding = serde_json::from_value(raw_binding).unwrap();

-        let err = ensure_binding_matches_schema(
-            &valid_function_binding_schema(true, false, &binding),
-            &binding,
-        )
-        .unwrap_err();
+        let err = ensure_binding_matches_schema(&function_binding_schema(true, false), &binding)
+            .unwrap_err();
        assert!(
            matches!(&err, Error::InvalidInput { message }
                if message.contains("input column 'title' is nullable")
@@ -2793,73 +2386,6 @@ mod tests {
        );
    }

-    #[test]
-    fn test_second_binding_rejects_outputs_without_reciprocal_metadata() {
-        let binding = FunctionBinding::from_json(include_str!(
-            "../../tests/fixtures/first_class_functions/v1/remote_function_binding.json"
-        ))
-        .unwrap();
-        let schema = ArrowSchema::new_with_metadata(
-            function_binding_schema(true, true).fields().to_vec(),
-            HashMap::from([(
-                FUNCTION_BINDINGS_META_KEY.to_string(),
-                function_bindings_metadata(std::slice::from_ref(&binding)).unwrap(),
-            )]),
-        );
-
-        let err = plan_function_application(
-            &schema,
-            &named_struct_application(
-                r#"{"normalized_text":"secondary_text","token_count":"secondary_token_count"}"#,
-            ),
-            None,
-        )
-        .unwrap_err();
-        assert!(
-            matches!(&err, Error::InvalidInput { message }
-                if message.contains("declaration metadata")
-                    && message.contains("fb_01K3TEXT")),
-            "{err:?}"
-        );
-    }
-
-    #[test]
-    fn test_persisted_nested_input_keeps_leaf_level_validation() {
-        let mut raw_binding: Value = serde_json::from_str(include_str!(
-            "../../tests/fixtures/first_class_functions/v1/remote_function_binding.json"
-        ))
-        .unwrap();
-        raw_binding["inputs"][0]["field_path"] = Value::String("title.value".to_string());
-        let binding: FunctionBinding = serde_json::from_value(raw_binding).unwrap();
-        let title = ArrowField::new(
-            "title",
-            DataType::Struct(vec![ArrowField::new("value", DataType::Utf8, true)].into()),
-            true,
-        )
-        .with_metadata(HashMap::from([
-            (COMPUTED_COLUMN_META_KEY.to_string(), "true".to_string()),
-            (KIND_META_KEY.to_string(), SQL_KIND.to_string()),
-            (EXPRESSION_META_KEY.to_string(), "title".to_string()),
-        ]));
-        let mut fields = vec![title, ArrowField::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}"),
-            };
-            ArrowField::new(&output.output_name, data_type, true).with_metadata(
-                function_computed_column_metadata(
-                    binding.binding_id(),
-                    output.output_ordinal,
-                    &["title.value".into(), "body".into()],
-                ),
-            )
-        }));
-
-        ensure_binding_matches_schema(&ArrowSchema::new(fields), &binding).unwrap();
-    }
-
    #[test]
    fn test_function_binding_metadata_survives_schema_round_trip() {
        let binding = FunctionBinding::from_json(include_str!(
@@ -2908,36 +2434,9 @@ mod tests {
                output_ordinal: 1,
            } if binding_id == "fb_01K3TEXT"
        ));
-        let dependent_application = FunctionApplication::from_json(
-            r#"{
-                "function":{"name":"dependent","version":"fv_dependent"},
-                "inputs":[
-                    {"parameter":"text","kind":"column","value":{"path":"search_text"}}
-                ],
-                "output":{"kind":"scalar","arrow_type":"int64","nullable":false}
-            }"#,
-        )
-        .unwrap();
-        let err = plan_function_application(&reopened, &dependent_application, Some("dependent"))
+        let err = plan_function_application(&reopened, &named_struct_application("{}"), None)
            .unwrap_err();
-        assert!(
-            matches!(&err, Error::InvalidInput { message } if message.contains("computed-on-computed"))
-        );
-        let plan = plan_function_application(
-            &reopened,
-            &named_struct_application(
-                r#"{"normalized_text":"secondary_text","token_count":"secondary_token_count"}"#,
-            ),
-            None,
-        )
-        .unwrap();
-        assert_eq!(
-            plan.outputs
-                .iter()
-                .map(|output| output.output_name.as_str())
-                .collect::<Vec<_>>(),
-            ["secondary_text", "secondary_token_count"]
-        );
+        assert!(matches!(err, Error::NotSupported { .. }));
    }

    #[test]
@@ -3093,38 +2592,5 @@ mod tests {
        assert!(
            matches!(&err, Error::InvalidInput { message } if message.contains("computed-on-computed"))
        );
-
-        let nested_title = ArrowField::new(
-            "title",
-            DataType::Struct(vec![ArrowField::new("value", DataType::Utf8, true)].into()),
-            true,
-        )
-        .with_metadata(HashMap::from([
-            (COMPUTED_COLUMN_META_KEY.to_string(), "true".to_string()),
-            (KIND_META_KEY.to_string(), SQL_KIND.to_string()),
-            (
-                EXPRESSION_META_KEY.to_string(),
-                "struct('value')".to_string(),
-            ),
-        ]));
-        let nested_schema = ArrowSchema::new(vec![nested_title, schema.field(1).as_ref().clone()]);
-        let nested_application = FunctionApplication::from_json(
-            r#"{
-                "function":{"name":"text_features","version":"fv_exact"},
-                "inputs":[
-                    {"parameter":"title","kind":"column","value":{"path":"title.value"}},
-                    {"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}
-                ]}
-            }"#,
-        )
-        .unwrap();
-        let err = plan_function_application(&nested_schema, &nested_application, None).unwrap_err();
-        assert!(
-            matches!(&err, Error::InvalidInput { message } if message.contains("computed-on-computed"))
-        );
    }
 }
@@ -29,14 +29,10 @@
 //! 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::{
-    Array, ArrayRef, BooleanArray, LargeBinaryArray, RecordBatch, RecordBatchOptions, StructArray,
-    new_null_array,
-};
-use arrow_schema::{DataType, Field as ArrowField, Schema as ArrowSchema};
+use arrow_array::{ArrayRef, BooleanArray, RecordBatch, RecordBatchOptions};
+use arrow_schema::Schema as ArrowSchema;
 use datafusion_expr::ColumnarValue;
 use futures::{Stream, StreamExt, TryStreamExt};
 use lance::Dataset;
@@ -44,7 +40,7 @@ use lance::dataset::WriteDestination;
 use lance::dataset::fragment::FileFragment;
 use lance::dataset::transaction::Operation;
 use lance_core::ROW_ID;
-use lance_core::datatypes::{BlobHandling, Schema as LanceSchema};
+use lance_core::datatypes::Schema as LanceSchema;
 use serde::{Deserialize, Serialize};

 use super::computed_columns::{BoundExpression, ComputedColumnKind, computed_column_from_field};
@@ -108,7 +104,6 @@ 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();
@@ -118,8 +113,7 @@ async fn execute_refresh_column_with_source(
            continue;
        }
        rows_filled += gained;
-        let values =
-            fill_stream(&dataset, &fragment, bound.clone(), column, output_is_blob).await?;
+        let values = fill_stream(&dataset, &fragment, bound.clone(), column).await?;
        replacements.push(fragment.write_columns(values, &column_schema).await?);
    }

@@ -300,15 +294,12 @@ 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 index = batch.schema_ref().index_of(name).map_err(|_| {
+        let column = batch.column_by_name(name).ok_or_else(|| {
            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 {
@@ -317,7 +308,7 @@ fn evaluation_batch(
        });
    }
    Ok(RecordBatch::try_new_with_options(
-        Arc::new(ArrowSchema::new(fields)),
+        bound.read_schema.clone(),
        columns,
        &RecordBatchOptions::new().with_row_count(Some(batch.num_rows())),
    )?)
@@ -338,99 +329,6 @@ 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
@@ -449,7 +347,6 @@ 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?;
@@ -471,7 +368,6 @@ 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());
@@ -481,20 +377,6 @@ 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())
@@ -530,11 +412,6 @@ 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])?)
    }))
 }
@@ -543,12 +420,8 @@ async fn fill_stream(
 mod tests {
    use std::sync::Arc;

-    use arrow_array::{
-        Array, ArrayRef, Int32Array, LargeBinaryArray, RecordBatch, StructArray, record_batch,
-    };
-    use arrow_schema::Field as ArrowField;
+    use arrow_array::{Int32Array, record_batch};
    use futures::TryStreamExt;
-    use lance_core::ROW_ID;

    use crate::connect;
    use crate::query::{ExecutableQuery, QueryBase, Select};
@@ -604,25 +477,6 @@ 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.
@@ -1310,366 +1164,4 @@ 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);
-    }
 }
Author	SHA1	Message	Date
Gatefixer	05fce57841	Merge origin/main into gatekeeper/fix-3759-1	2026-08-28 01:19:19 +00:00
Gatefixer	43465f06d5	fix(python): support blob lists in dict appends	2026-08-26 23:47:47 +00:00
Gatefixer	bc762926ab	fix(python): preserve nested extension carriers	2026-08-26 23:21:40 +00:00
Gatefixer	2426f275cd	fix(python): accept null extension column batches	2026-08-26 22:55:54 +00:00