Merge remote-tracking branch 'origin/main' into gatekeeper/fix-2325-1

# Conflicts: # python/python/lancedb/remote/table.py
feat: support blob computed column refresh (#4081 )
2026-09-13 00:32:31 +00:00 · 2026-08-28 09:45:02 +00:00 · 2026-08-28 17:23:49 +08:00 · 2026-08-27 19:06:07 -07:00 · 2026-08-28 00:52:01 +00:00 · 2026-08-27 17:47:27 -07:00
37 changed files with 2488 additions and 611 deletions
@@ -1,5 +1,5 @@
 [tool.bumpversion]
-current_version = "0.38.0-beta.11"
+current_version = "0.38.0-beta.12"
 parse = """(?x)
    (?P<major>0|[1-9]\\d*)\\.
    (?P<minor>0|[1-9]\\d*)\\.
@@ -40,40 +40,31 @@ jobs:
          - target: aarch64-apple-darwin
            host: macos-latest
            features: fp16kernels
+            # Fat LTO was ~111 of this job's ~113 minutes.
+            lto: thin
+            codegen_units: 16
            pre_build: |-
              brew install protobuf
-              # Fat LTO (the workspace default in .cargo/config.toml) is
-              # single-threaded and is the peak-memory step of the build. On
-              # this runner it accounted for ~111 of the job's ~113 minutes,
-              # making it the critical path of the entire publish pipeline.
-              # ThinLTO parallelizes it across the runner's cores, for a few
-              # percent of runtime performance.
-              export CARGO_PROFILE_RELEASE_LTO=thin
-              export CARGO_PROFILE_RELEASE_CODEGEN_UNITS=16
          - target: x86_64-pc-windows-msvc
            host: windows-2025
            features: ","
+            # The lower peak also keeps this on the standard 4-core runner.
+            lto: thin
+            codegen_units: 16
            pre_build: |-
              choco install --no-progress protoc ninja nasm
              tail -n 1000 /c/ProgramData/chocolatey/logs/chocolatey.log
              # There is an issue where choco doesn't add nasm to the path
              export PATH="$PATH:/c/Program Files/NASM"
              nasm -v
-              # See the ThinLTO note on aarch64-apple-darwin above. Keeping
-              # peak memory down is also what lets this run on the standard
-              # 4-core runner: the 8-core larger runner was only needed to
-              # stop fat LTO from OOMing rustc-LLVM.
-              export CARGO_PROFILE_RELEASE_LTO=thin
-              export CARGO_PROFILE_RELEASE_CODEGEN_UNITS=16
          - target: aarch64-pc-windows-msvc
            host: windows-2025
            features: ","
+            lto: thin
+            codegen_units: 16
            pre_build: |-
                choco install --no-progress protoc
                rustup target add aarch64-pc-windows-msvc
-                # See the ThinLTO note on aarch64-apple-darwin above.
-                export CARGO_PROFILE_RELEASE_LTO=thin
-                export CARGO_PROFILE_RELEASE_CODEGEN_UNITS=16
          - target: x86_64-unknown-linux-gnu
            host: ubuntu-latest
            features: fp16kernels
@@ -103,6 +94,14 @@ jobs:
            # https://github.com/napi-rs/napi-rs/blob/main/debian-aarch64.Dockerfile
            docker: ghcr.io/napi-rs/napi-rs/nodejs-rust:lts-debian-aarch64
            features: "fp16kernels"
+            # Fat LTO OOM-killed rustc every nightly; even with lld it peaked
+            # at 31391 MiB of the runner's 32 GiB.
+            lto: thin
+            codegen_units: 16
+            # arm64 Linux links through GNU `ld` where x86_64 defaults to
+            # `rust-lld`, which is why only arm64 OOM'd. lld cut the largest
+            # linker process 7.0 -> 4.0 GiB (lancedb/sophon#7313).
+            linker: /tmp/aarch64-lld-clang
            pre_build: |-
              set -e &&
              apt-get update &&
@@ -112,9 +111,30 @@ jobs:
              # AT_HWCAP2 (added in Linux 3.17). Define it for aws-lc-sys.
              export CFLAGS="$CFLAGS -DAT_HWCAP2=26" &&
              rustup target add aarch64-unknown-linux-gnu
+              # Not `&&`-chained: in dash, errexit does not fire for a
+              # non-final command in an `&&` list, so failures were ignored.
+              #
+              # A wrapper rather than `-C link-arg` because the per-target
+              # rustflags variable does not reach every unit that links, while
+              # the linker variable does. `clang` because GCC silently ignores
+              # `-fuse-ld=lld` unless built with lld support. Two echoes
+              # because printf's newline escape gets rewritten to `;` between
+              # here and the container.
+              echo '#!/bin/sh' > /tmp/aarch64-lld-clang
+              echo 'exec clang --target=aarch64-unknown-linux-gnu --sysroot=/usr/aarch64-unknown-linux-gnu/aarch64-unknown-linux-gnu/sysroot --gcc-toolchain=/usr/aarch64-unknown-linux-gnu -fuse-ld=lld "$@"' >> /tmp/aarch64-lld-clang
+              chmod 0755 /tmp/aarch64-lld-clang
+              # Fail now, not at the cdylib link ~30 minutes later. Linking at
+              # all also proves lld resolved; clang errors out when it cannot.
+              echo 'int main(void){return 0;}' > /tmp/probe.c
+              /tmp/aarch64-lld-clang /tmp/probe.c -o /tmp/probe
+              readelf -h /tmp/probe | grep AArch64
          - target: aarch64-unknown-linux-musl
            host: ubuntu-2404-8x-x64
            features: ","
+            # Fat LTO took the whole runner down. lld cannot help: it died
+            # inside rustc's LLVM, before any linker was spawned.
+            lto: thin
+            codegen_units: 16
            pre_build: |-
              set -e &&
              sudo apt-get update &&
@@ -123,6 +143,19 @@ jobs:
              export EXTRA_ARGS="-x"
    name: build - ${{ matrix.settings.target }}
    runs-on: ${{ matrix.settings.host }}
+    # On the job, not exported from `pre_build`: `Swatinem/rust-cache` hashes
+    # `CARGO_*` into its cache key before any step runs, so a step-local export
+    # leaves the key unchanged while cargo still rebuilds cold. The ThinLTO
+    # legs had been doing that every run.
+    #
+    # Not `RUSTFLAGS`: setting it, even to "", discards every config-file
+    # rustflag, silently dropping .cargo/config.toml's `target-cpu` and
+    # `target-feature` from the published binaries.
+    env:
+      CARGO_PROFILE_RELEASE_LTO: ${{ matrix.settings.lto || 'fat' }}
+      CARGO_PROFILE_RELEASE_CODEGEN_UNITS: ${{ matrix.settings.codegen_units || '1' }}
+      # Empty elsewhere: a per-target variable is only read for that triple.
+      CARGO_TARGET_AARCH64_UNKNOWN_LINUX_GNU_LINKER: ${{ matrix.settings.linker }}
    defaults:
      run:
        working-directory: nodejs
@@ -169,19 +202,15 @@ jobs:
          # creating ref). The nightly cadence also keeps entries inside
          # GitHub's 7-day eviction window, which a tag-only trigger would not.
          save-if: ${{ github.ref == 'refs/heads/main' }}
-      # Docker builds can use rust-cache too. `target/` already lives on the
-      # host because the whole workspace is bind-mounted into the container, and
-      # rust-cache's prune and save run host-side, so they can manage it -- which
-      # is what keeps the entry to dependency artifacts rather than a multi-GB
-      # copy of everything.
+      # Docker builds can use rust-cache too: the workspace is bind-mounted, so
+      # `target/` lives on the host and rust-cache's prune keeps the entry
+      # small.
      #
      # Two differences from the native builds. The container's CARGO_HOME is
-      # bind-mounted from `.cargo-cache` rather than the host's ~/.cargo, so that
-      # has to be cached explicitly. And the key is derived from the *host* rustc
-      # version, which is not the compiler that produced these artifacts; that is
-      # safe because cargo fingerprints the real compiler and rebuilds on a
-      # mismatch, it just means a base-image toolchain bump costs one cold build
-      # instead of invalidating the key.
+      # bind-mounted from `.cargo-cache` rather than ~/.cargo, so that is cached
+      # explicitly. And the key uses the *host* rustc version, not the compiler
+      # that built these artifacts -- safe, since cargo fingerprints the real
+      # one; a base-image bump just costs one cold build.
      - name: Cache cargo (docker builds)
        uses: Swatinem/rust-cache@v2
        if: ${{ matrix.settings.docker }}
@@ -210,9 +239,14 @@ jobs:
          # cache step above saves. Previously the registry mounts pointed at
          # `.cargo/...`, a path nothing cached, so the container re-downloaded
          # the whole crate registry on every run.
+          #
+          # `docker run` inherits nothing; `-e NAME` carries the job's `env:` in.
          options: "--user 0:0 -v ${{ github.workspace }}/.cargo-cache/git/db:/usr/local/cargo/git/db \
                    -v ${{ github.workspace }}/.cargo-cache/registry/cache:/usr/local/cargo/registry/cache \
                    -v ${{ github.workspace }}/.cargo-cache/registry/index:/usr/local/cargo/registry/index \
+                    -e CARGO_PROFILE_RELEASE_LTO \
+                    -e CARGO_PROFILE_RELEASE_CODEGEN_UNITS \
+                    -e CARGO_TARGET_AARCH64_UNKNOWN_LINUX_GNU_LINKER \
                    -v ${{ github.workspace }}:/build -w /build/nodejs"
          run: |
            set -e
@@ -256,6 +290,18 @@ jobs:
        if: always()
        run: df -h
        shell: bash
+      - name: Report peak memory
+        if: always() && runner.os == 'Linux'
+        shell: bash
+        run: |
+          peak=$(find /sys/fs/cgroup -name memory.peak -readable \
+                 -exec cat {} + 2>/dev/null | sort -n | tail -1)
+          if [ -n "$peak" ]; then
+            echo "peak memory: $((peak / 1024 / 1024)) MiB"
+          else
+            echo "peak memory: unavailable (no readable cgroup v2 memory.peak)"
+          fi
+          free -g || true
      - name: Upload artifact
        uses: actions/upload-artifact@v7
        with:
@@ -1597,9 +1597,9 @@ checksum = "613afe47fcd5fac7ccf1db93babcb082c5994d996f20b8b159f2ad1658eb5724"

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

 [[package]]
 name = "fsst"
-version = "12.0.0-beta.4"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.4#474ae88dd75c437d5d506859b627510978e502e9"
+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.4"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.4#474ae88dd75c437d5d506859b627510978e502e9"
+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.4"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.4#474ae88dd75c437d5d506859b627510978e502e9"
+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.4#474ae88dd75c437d5d506859b627510978e502e9"
+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.4#474ae88dd75c437d5d506859b627510978e502e9"
+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.4"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.4#474ae88dd75c437d5d506859b627510978e502e9"
+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.4"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.4#474ae88dd75c437d5d506859b627510978e502e9"
+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.4"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.4#474ae88dd75c437d5d506859b627510978e502e9"
+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.4"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.4#474ae88dd75c437d5d506859b627510978e502e9"
+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.4"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.4#474ae88dd75c437d5d506859b627510978e502e9"
+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.4"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.4#474ae88dd75c437d5d506859b627510978e502e9"
+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.4"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.4#474ae88dd75c437d5d506859b627510978e502e9"
+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.4"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.4#474ae88dd75c437d5d506859b627510978e502e9"
+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.4"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.4#474ae88dd75c437d5d506859b627510978e502e9"
+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.4"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.4#474ae88dd75c437d5d506859b627510978e502e9"
+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.4"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.4#474ae88dd75c437d5d506859b627510978e502e9"
+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.4"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.4#474ae88dd75c437d5d506859b627510978e502e9"
+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.4"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.4#474ae88dd75c437d5d506859b627510978e502e9"
+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.4"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.4#474ae88dd75c437d5d506859b627510978e502e9"
+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.4"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.4#474ae88dd75c437d5d506859b627510978e502e9"
+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.4"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.4#474ae88dd75c437d5d506859b627510978e502e9"
+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.4"
-source = "git+https://github.com/lance-format/lance.git?tag=v12.0.0-beta.4#474ae88dd75c437d5d506859b627510978e502e9"
+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.11"
+version = "0.38.0-beta.12"
 dependencies = [
 "ahash",
 "anyhow",
@@ -5490,7 +5490,7 @@ dependencies = [

 [[package]]
 name = "lancedb-nodejs"
-version = "0.38.0-beta.11"
+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.11"
+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.4", default-features = false, "tag" = "v12.0.0-beta.4", "git" = "https://github.com/lance-format/lance.git" }
-lance-core = { "version" = "=12.0.0-beta.4", "tag" = "v12.0.0-beta.4", "git" = "https://github.com/lance-format/lance.git" }
-lance-datagen = { "version" = "=12.0.0-beta.4", "tag" = "v12.0.0-beta.4", "git" = "https://github.com/lance-format/lance.git" }
-lance-file = { "version" = "=12.0.0-beta.4", "tag" = "v12.0.0-beta.4", "git" = "https://github.com/lance-format/lance.git" }
-lance-io = { "version" = "=12.0.0-beta.4", default-features = false, "tag" = "v12.0.0-beta.4", "git" = "https://github.com/lance-format/lance.git" }
-lance-index = { "version" = "=12.0.0-beta.4", "tag" = "v12.0.0-beta.4", "git" = "https://github.com/lance-format/lance.git" }
-lance-linalg = { "version" = "=12.0.0-beta.4", "tag" = "v12.0.0-beta.4", "git" = "https://github.com/lance-format/lance.git" }
-lance-namespace = { "version" = "=12.0.0-beta.4", "tag" = "v12.0.0-beta.4", "git" = "https://github.com/lance-format/lance.git" }
-lance-namespace-impls = { "version" = "=12.0.0-beta.4", default-features = false, "tag" = "v12.0.0-beta.4", "git" = "https://github.com/lance-format/lance.git" }
-lance-table = { "version" = "=12.0.0-beta.4", "tag" = "v12.0.0-beta.4", "git" = "https://github.com/lance-format/lance.git" }
-lance-testing = { "version" = "=12.0.0-beta.4", "tag" = "v12.0.0-beta.4", "git" = "https://github.com/lance-format/lance.git" }
-lance-datafusion = { "version" = "=12.0.0-beta.4", "tag" = "v12.0.0-beta.4", "git" = "https://github.com/lance-format/lance.git" }
-lance-encoding = { "version" = "=12.0.0-beta.4", "tag" = "v12.0.0-beta.4", "git" = "https://github.com/lance-format/lance.git" }
-lance-arrow = { "version" = "=12.0.0-beta.4", "tag" = "v12.0.0-beta.4", "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
@@ -131,18 +131,13 @@ allow = [
    "BSD-3-Clause",
    "ISC",
    "Unicode-3.0",
-    "Unicode-DFS-2016",
    "Zlib",
    "CC0-1.0",
    "MPL-2.0",
    "BSL-1.0",
-    "OpenSSL",
    # 0BSD ("BSD Zero Clause") is effectively public domain — no attribution
    # required. Pulled in by `mock_instant`.
    "0BSD",
-    # bzip2-1.0.6 is the permissive upstream bzip2 license (BSD-like). Pulled
-    # in by `libbz2-rs-sys`, the pure-Rust bzip2 implementation.
-    "bzip2-1.0.6",
    # CDLA-Permissive-2.0 is a permissive data license used by `webpki-roots`
    # for the Mozilla CA root bundle. Data-only, distribution-compatible.
    "CDLA-Permissive-2.0",
@@ -150,12 +145,7 @@ allow = [
 confidence-threshold = 0.8
 # Per-crate license exceptions: allow a license for a specific crate only,
 # rather than globally via the `allow` list above.
-exceptions = [
-    # CDDL-1.0 (copyleft) is pulled in only as a dev/profiling dependency via
-    # `inferno` -> `pprof` -> `lance-testing`; it is a test dependency that we
-    # do not distribute, so scope the allowance to `inferno` alone.
-    { allow = ["CDDL-1.0"], crate = "inferno" },
-]
+exceptions = []
 # Crates whose license cannot be determined from Cargo metadata but whose
 # license we've manually confirmed from upstream. Keep this list minimal.
 [[licenses.clarify]]
@@ -14,7 +14,7 @@ Add the following dependency to your `pom.xml`:
 <dependency>
    <groupId>com.lancedb</groupId>
    <artifactId>lancedb-core</artifactId>
-    <version>0.38.0-beta.11</version>
+    <version>0.38.0-beta.12</version>
 </dependency>
 ```

@@ -42,6 +42,8 @@ listing a storage directory.

 ::: lancedb.table.Table

+::: lancedb.table.CompactionOptions
+
 ::: lancedb.table.FragmentStatistics

 ::: lancedb.table.FragmentSummaryStats
@@ -223,9 +225,13 @@ tokens = list(
 Blob columns store large binary values out of line so they can be read lazily
 instead of being materialized with the rest of the row.

-::: lancedb.blob
+`lancedb.BlobType` is `lance.blob.BlobType` when pylance is installed. Without
+pylance, LanceDB uses a matching `lance.blob.v2` extension type so blob columns
+still work. Queries return descriptors. Call
+[`fetch_blob_files`][lancedb.table.Table.fetch_blob_files] for lazy reads or
+[`fetch_blobs`][lancedb.table.Table.fetch_blobs] for eager bytes.

-::: lancedb.BlobType
+::: lancedb.blob

 ::: lancedb._blob.BlobFile
    options:
@@ -8,7 +8,7 @@
    <parent>
      <groupId>com.lancedb</groupId>
      <artifactId>lancedb-parent</artifactId>
-      <version>0.38.0-beta.11</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.11</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.4</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.11"
+version = "0.38.0-beta.12"
 publish = false
 license.workspace = true
 description.workspace = true
@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-darwin-arm64",
-	"version": "0.38.0-beta.11",
+	"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.11",
+	"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.11",
+	"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.11",
+	"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.11",
+	"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.11",
+  "version": "0.38.0-beta.12",
  "os": [
    "win32"
  ],
@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-win32-x64-msvc",
-	"version": "0.38.0-beta.11",
+	"version": "0.38.0-beta.12",
 	"os": ["win32"],
 	"cpu": ["x64"],
 	"main": "lancedb.win32-x64-msvc.node",
@@ -1,12 +1,12 @@
 {
  "name": "@lancedb/lancedb",
-  "version": "0.38.0-beta.11",
+  "version": "0.38.0-beta.12",
  "lockfileVersion": 3,
  "requires": true,
  "packages": {
    "": {
      "name": "@lancedb/lancedb",
-      "version": "0.38.0-beta.11",
+      "version": "0.38.0-beta.12",
      "cpu": [
        "x64",
        "arm64"
@@ -11,7 +11,7 @@
    "ann"
  ],
  "private": false,
-  "version": "0.38.0-beta.11",
+  "version": "0.38.0-beta.12",
  "main": "dist/index.js",
  "exports": {
    ".": "./dist/index.js",
@@ -1,6 +1,6 @@
 [package]
 name = "lancedb-python"
-version = "0.38.0-beta.11"
+version = "0.38.0-beta.12"
 publish = false
 edition.workspace = true
 description = "Python bindings for LanceDB"
@@ -6,7 +6,7 @@ import importlib.metadata
 import os
 from concurrent.futures import ThreadPoolExecutor
 from datetime import timedelta
-from typing import Dict, Optional, Union, Any, List, Iterable
+from typing import Dict, Optional, Union, Any, List, Iterable, TYPE_CHECKING

 __version__ = importlib.metadata.version("lancedb")

@@ -20,7 +20,7 @@ from .db import AsyncConnection, DBConnection, LanceDBConnection
 from .remote import ClientConfig
 from .remote.db import RemoteDBConnection
 from .expr import Expr, col, lit, func
-from .schema import blob, vector, BlobType
+from .schema import blob, vector
 from .job import AsyncJob, Job
 from .functions import (
    FunctionArtifactRequest as FunctionArtifactRequest,
@@ -38,7 +38,7 @@ from .materialized_view import (
    MaterializedView,
    MaterializedViewDefinition,
 )
-from .table import AsyncTable, Table
+from .table import AsyncTable, CompactionOptions, Table
 from .types import BaseTokenizerType
 from ._lancedb import Session
 from .namespace import (
@@ -49,6 +49,19 @@ from .namespace import (
 )


+if TYPE_CHECKING:
+    from lance.blob import BlobType as BlobType
+
+
+def __getattr__(name: str):
+    if name == "BlobType":
+        from .schema import BlobType
+
+        globals()["BlobType"] = BlobType
+        return BlobType
+    raise AttributeError(f"module {__name__!r} has no attribute {name!r}")
+
+
 def _check_s3_bucket_with_dots(
    uri: str, storage_options: Optional[Dict[str, str]]
 ) -> None:
@@ -545,6 +558,7 @@ __all__ = [
    "AsyncJob",
    "AsyncLanceNamespaceDBConnection",
    "AsyncTable",
+    "CompactionOptions",
    "FtsToken",
    "col",
    "Expr",
@@ -12,7 +12,7 @@ from typing import TYPE_CHECKING, Optional, Union
 import pyarrow as pa

 from .expr import Expr
-from .schema import blob_v2_column_paths
+from .schema import row_addressable_blob_v2_paths
 from .types import BlobMode, QueryProjection, QueryProjectionSpec

 if TYPE_CHECKING:
@@ -119,7 +119,7 @@ def blob_v2_projection_sources(
    schema: pa.Schema,
    projection: QueryProjection,
 ) -> dict[str, str]:
-    blob_columns = blob_v2_column_paths(schema)
+    blob_columns = row_addressable_blob_v2_paths(schema)
    if not blob_columns:
        return {}
    columns = set(blob_columns)
@@ -140,7 +140,9 @@ def v2_projection_needs_row_id(
 ) -> bool:
    if with_row_id:
        return False
-    return projection_includes_blob_column(projection, blob_v2_column_paths(schema))
+    return projection_includes_blob_column(
+        projection, row_addressable_blob_v2_paths(schema)
+    )


 def blob_auto_row_id_for_scan(
@@ -374,6 +374,7 @@ class Table:
        *,
        cleanup_since_ms: Optional[int] = None,
        delete_unverified: Optional[bool] = None,
+        compaction_options: Optional[Dict[str, Any]] = None,
    ) -> OptimizeStats: ...
    async def uri(self) -> str: ...
    async def initial_storage_options(self) -> Optional[Dict[str, str]]: ...
@@ -67,7 +67,16 @@ from ..query import (
    LanceTakeQueryBuilder,
    LanceVectorQueryBuilder,
 )
-from ..table import AsyncTable, BlobMode, Branches, IndexStatistics, Query, Table, Tags
+from ..table import (
+    AsyncTable,
+    BlobMode,
+    Branches,
+    CompactionOptions,
+    IndexStatistics,
+    Query,
+    Table,
+    Tags,
+)
 from ..types import BaseTokenizerType


@@ -953,6 +962,7 @@ class RemoteTable(Table):
        *,
        cleanup_older_than: Optional[timedelta] = None,
        delete_unverified: bool = False,
+        compaction_options: Optional[CompactionOptions] = None,
    ):
        """
        optimize() is a no-op on LanceDB Cloud.
@@ -4,30 +4,34 @@

 """Schema helpers for Lance blob columns."""

+import importlib
+from typing import TYPE_CHECKING
+
 import pyarrow as pa
+import pyarrow.ipc
+
+if TYPE_CHECKING:
+    from lance.blob import BlobType as BlobType

 _BLOB_EXTENSION_NAME = "lance.blob.v2"
 _BLOB_V1_KEY = "lance-encoding:blob"
 _ARROW_EXT_NAME_KEY = "ARROW:extension:name"
+_BLOB_V2_STORAGE_TYPE = pa.struct(
+    [
+        pa.field("data", pa.large_binary(), nullable=True),
+        pa.field("uri", pa.utf8(), nullable=True),
+        pa.field("position", pa.uint64(), nullable=True),
+        pa.field("size", pa.uint64(), nullable=True),
+    ]
+)
+_resolved_blob_type = None


-class BlobType(pa.ExtensionType):
-    """PyArrow extension type for a Lance blob v2 column.
-
-    Queries return descriptors; call :meth:`~lancedb.table.Table.fetch_blob_files`
-    for lazy reads or :meth:`~lancedb.table.Table.fetch_blobs` for eager bytes.
-    """
+class _FallbackBlobType(pa.ExtensionType):
+    """lance.blob.v2 extension type used when pylance is not installed."""

    def __init__(self) -> None:
-        storage_type = pa.struct(
-            [
-                pa.field("data", pa.large_binary(), nullable=True),
-                pa.field("uri", pa.utf8(), nullable=True),
-                pa.field("position", pa.uint64(), nullable=True),
-                pa.field("size", pa.uint64(), nullable=True),
-            ]
-        )
-        super().__init__(storage_type, _BLOB_EXTENSION_NAME)
+        pa.ExtensionType.__init__(self, _BLOB_V2_STORAGE_TYPE, _BLOB_EXTENSION_NAME)

    def __arrow_ext_serialize__(self) -> bytes:
        return b""
@@ -35,23 +39,16 @@ class BlobType(pa.ExtensionType):
    @classmethod
    def __arrow_ext_deserialize__(
        cls, storage_type: pa.DataType, serialized: bytes
-    ) -> "BlobType":
+    ) -> "_FallbackBlobType":
        return cls()

    def __reduce__(self):
-        # Ensure pickle round-trips on older pyarrow (apache/arrow#35599).
        return type(self).__arrow_ext_deserialize__, (
            self.storage_type,
            self.__arrow_ext_serialize__(),
        )


-try:
-    pa.register_extension_type(BlobType())  # type: ignore[arg-type]
-except pa.ArrowKeyError:
-    pass
-
-
 def _metadata_value(metadata: dict, key: str):
    return metadata.get(key.encode()) or metadata.get(key)

@@ -92,43 +89,105 @@ def is_blob_like_field(field: pa.Field) -> bool:
    return is_blob_v2_field(field) or _metadata_marks_legacy_blob(field.metadata or {})


-def _collect_blob_paths(schema: pa.Schema, is_blob) -> list[str]:
-    paths: list[str] = []
+def _collect_blob_paths(schema: pa.Schema, is_blob) -> list[tuple[str, bool]]:
+    """Walk the schema and return (path, has_list_ancestor) for each blob field."""
+    paths: list[tuple[str, bool]] = []

-    def walk(fields, prefix: str) -> None:
+    def walk(fields, prefix: str, has_list_ancestor: bool) -> None:
        for field in fields:
            path = f"{prefix}.{field.name}" if prefix else field.name
            if is_blob(field):
-                paths.append(path)
+                paths.append((path, has_list_ancestor))
            elif pa.types.is_struct(field.type):
-                walk(field.type, path)
+                walk(field.type, path, has_list_ancestor)
            elif (
                pa.types.is_list(field.type)
                or pa.types.is_large_list(field.type)
                or pa.types.is_fixed_size_list(field.type)
            ):
-                walk([field.type.value_field], path)
+                walk([field.type.value_field], path, True)

-    walk(schema, "")
+    walk(schema, "", False)
    return paths


 def blob_column_paths(schema: pa.Schema) -> list[str]:
    """Dotted paths of blob-like columns (v2 extension or legacy metadata)."""
-    return _collect_blob_paths(schema, is_blob_like_field)
+    return [path for path, _ in _collect_blob_paths(schema, is_blob_like_field)]


 def blob_v2_column_paths(schema: pa.Schema) -> list[str]:
-    return _collect_blob_paths(schema, is_blob_v2_field)
+    return [path for path, _ in _collect_blob_paths(schema, is_blob_v2_field)]
+
+
+def row_addressable_blob_v2_paths(schema: pa.Schema) -> list[str]:
+    """Blob v2 paths with one blob addressable by table row id.
+
+    ``fetch_blobs`` and the descriptor row-id ride-along address one blob per
+    row, so a blob inside a list container has no row-id slot and no fetch
+    path. Those columns still store and query as raw descriptors.
+    """
+    return [
+        path
+        for path, has_list_ancestor in _collect_blob_paths(schema, is_blob_v2_field)
+        if not has_list_ancestor
+    ]


 def schema_has_blob_field(schema: pa.Schema) -> bool:
    return bool(blob_column_paths(schema))


+def _deserialize_registered_type(extension_type: pa.ExtensionType) -> pa.DataType:
+    """Return the type Arrow reconstructs for this extension name."""
+    schema = pa.schema([pa.field("value", extension_type)])
+    restored = pa.ipc.read_schema(schema.serialize())
+    return restored.field("value").type
+
+
+def _resolve_blob_type():
+    """Return the BlobType class this process should use.
+
+    pylance's class when it owns the lance.blob.v2 registry entry,
+    otherwise LanceDB's fallback. A different registered class is an error.
+    """
+    global _resolved_blob_type
+    if _resolved_blob_type is not None:
+        return _resolved_blob_type
+    try:
+        blob_module = importlib.import_module("lance.blob")
+    except ModuleNotFoundError as err:
+        if err.name not in ("lance", "lance.blob"):
+            raise
+    else:
+        blob_type = getattr(blob_module, "BlobType", None)
+        if blob_type is not None:
+            registered_type = _deserialize_registered_type(blob_type())
+            if type(registered_type) is not blob_type:
+                registered_cls = type(registered_type)
+                raise ValueError(
+                    "lance.blob.v2 is already registered by "
+                    f"{registered_cls.__module__}.{registered_cls.__qualname__}"
+                )
+            _resolved_blob_type = blob_type
+            return blob_type
+    try:
+        pa.register_extension_type(_FallbackBlobType())  # type: ignore[arg-type]
+    except pa.ArrowKeyError as err:
+        raise ValueError(
+            "lance.blob.v2 is already registered by another extension class"
+        ) from err
+    _resolved_blob_type = _FallbackBlobType
+    return _resolved_blob_type
+
+
 def blob(name: str, nullable: bool = True) -> pa.Field:
-    """Create a Lance blob v2 column field."""
-    return pa.field(name, BlobType(), nullable=nullable)
+    """Create a Lance blob v2 column field.
+
+    When pylance is installed this is ``lance.blob.BlobType``.
+    """
+    blob_type = _resolve_blob_type()
+    return pa.field(name, blob_type(), nullable=nullable)


 def vector(dimension: int, value_type: pa.DataType = pa.float32()) -> pa.DataType:
@@ -155,3 +214,11 @@ def vector(dimension: int, value_type: pa.DataType = pa.float32()) -> pa.DataTyp
    ... ])
    """
    return pa.list_(value_type, dimension)
+
+
+def __getattr__(name: str):
+    if name == "BlobType":
+        blob_type = _resolve_blob_type()
+        globals()["BlobType"] = blob_type
+        return blob_type
+    raise AttributeError(f"module {__name__!r} has no attribute {name!r}")
@@ -22,6 +22,7 @@ from typing import (
    Optional,
    Sequence,
    Tuple,
+    TypedDict,
    Union,
    overload,
 )
@@ -104,7 +105,12 @@ from .util import (
    value_to_sql,
 )
 from .index import lang_mapping
-from .schema import blob_v2_column_paths, schema_has_blob_field
+from .schema import (
+    blob_v2_column_paths,
+    is_blob_v2_field,
+    row_addressable_blob_v2_paths,
+    schema_has_blob_field,
+)


 def _should_push_down_query_table(
@@ -222,6 +228,88 @@ IndexConfigType = Union[
    FTS,
 ]

+
+class CompactionOptions(TypedDict, total=False):
+    """Options that control file compaction during table optimization.
+
+    Unspecified options use Lance's defaults.
+
+    Compaction planning is row based. Lowering ``target_rows_per_fragment``
+    based on the expected row size can bound later compaction passes once
+    oversized fragments have been rewritten. It does not split an existing
+    fragment, so the first pass over an oversized fragment is not subject to
+    that bound. ``max_bytes_per_file`` limits output file size, not compaction
+    memory. Source budgets keep whole planned tasks; if the first task exceeds
+    a budget, that run performs no compaction work.
+
+    Examples
+    --------
+    Derive a steady-state row target from the expected row size:
+
+    >>> desired_fragment_bytes = 750 * 1024 * 1024
+    >>> average_row_bytes = 1_500_000
+    >>> options: CompactionOptions = {
+    ...     "target_rows_per_fragment": max(
+    ...         1, desired_fragment_bytes // average_row_bytes
+    ...     ),
+    ... }
+    >>> await table.optimize(compaction_options=options)  # doctest: +SKIP
+    """
+
+    target_rows_per_fragment: int
+    """Target rows per fragment; existing oversized fragments are not split."""
+
+    max_rows_per_group: int
+    """Maximum number of rows per row group (default: 1,024)."""
+
+    max_bytes_per_file: Optional[int]
+    """Maximum output data-file size; this does not bound compaction memory."""
+
+    materialize_deletions: bool
+    """Whether to rewrite fragments containing deleted rows (default: True)."""
+
+    materialize_deletions_threshold: float
+    """Minimum deleted-row fraction that makes a fragment eligible (default: 0.1)."""
+
+    num_threads: Optional[int]
+    """Number of compaction tasks to run in parallel."""
+
+    batch_size: Optional[int]
+    """Number of rows per input scan batch."""
+
+    io_buffer_size: Optional[int]
+    """Maximum number of bytes queued in the input scan I/O buffer."""
+
+    defer_index_remap: bool
+    """Whether to defer index remapping during compaction (default: False)."""
+
+    index_remap_mode: Literal["direct", "compact"]
+    """How to construct the old-to-new row-address mapping."""
+
+    compaction_mode: Optional[
+        Literal["reencode", "try_binary_copy", "force_binary_copy"]
+    ]
+    """Whether compaction re-encodes data or uses binary copying."""
+
+    binary_copy_read_batch_bytes: Optional[int]
+    """Number of bytes read per batch during binary-copy compaction."""
+
+    max_source_fragments: Optional[int]
+    """Maximum number of source fragments compacted in one run."""
+
+    max_source_rows: Optional[int]
+    """Maximum live source rows per run, applied to whole planned tasks."""
+
+    max_source_bytes: Optional[int]
+    """Maximum source bytes per run, applied to whole planned tasks."""
+
+    excluded_fragment_ids: List[int]
+    """Fragment IDs to leave unchanged and use as planning boundaries."""
+
+    max_overlays_per_fragment: Optional[int]
+    """Maximum overlays before a fragment is fully compacted."""
+
+
 # Known distance metrics for legacy API detection
 KNOWN_METRICS = {"l2", "cosine", "dot", "hamming"}

@@ -426,6 +514,7 @@ def _cast_to_target_schema(

    def gen():
        for batch in reader:
+            batch = _coerce_blob_write_columns(batch, reordered_schema)
            # Table but not RecordBatch has cast.
            cast_batches = (
                pa.Table.from_batches([batch]).cast(reordered_schema).to_batches()
@@ -438,6 +527,166 @@ def _cast_to_target_schema(
    return pa.RecordBatchReader.from_batches(reordered_schema, gen())


+def _coerce_blob_write_columns(
+    batch: pa.RecordBatch, target_schema: pa.Schema
+) -> pa.RecordBatch:
+    """Materialize blob storage structs before the stream leaves Python.
+
+    merge_insert requires its source reader to already match the table's
+    physical schema. Unlike add and insert, it does not pass through
+    LanceDB's Rust blob coercion, so preserving binary input here would
+    reach Lance as binary and fail the schema check.
+    """
+    columns = []
+    fields = []
+    changed = False
+    for field, column in zip(batch.schema, batch.columns):
+        target_field = target_schema.field(field.name)
+        coerced = _coerce_blob_value(column, target_field)
+        if coerced is not column:
+            column = coerced
+            field = pa.field(
+                field.name,
+                coerced.type,
+                field.nullable,
+                target_field.metadata,
+            )
+            changed = True
+        columns.append(column)
+        fields.append(field)
+    if not changed:
+        return batch
+    return pa.RecordBatch.from_arrays(
+        columns, schema=pa.schema(fields, metadata=batch.schema.metadata)
+    )
+
+
+def _coerce_blob_value(column: pa.Array, target_field: pa.Field) -> pa.Array:
+    if is_blob_v2_field(target_field) and _can_coerce_to_blob(column.type):
+        return _coerce_value_to_blob(column, target_field)
+
+    target_type = target_field.type
+    if pa.types.is_struct(target_type) and pa.types.is_struct(column.type):
+        children = []
+        fields = []
+        changed = False
+        for source_field in column.type:
+            source_column = column.field(source_field.name)
+            nested_target = next(
+                (field for field in target_type if field.name == source_field.name),
+                None,
+            )
+            if nested_target is None:
+                children.append(source_column)
+                fields.append(source_field)
+                continue
+            coerced = _coerce_blob_value(source_column, nested_target)
+            if coerced is not source_column:
+                changed = True
+            child_array, child_type = _physical_array_and_type(coerced)
+            children.append(child_array)
+            fields.append(
+                pa.field(
+                    source_field.name,
+                    child_type,
+                    source_field.nullable,
+                    nested_target.metadata,
+                )
+            )
+        if not changed:
+            return column
+        return pa.StructArray.from_arrays(
+            children,
+            fields=fields,
+            mask=column.is_null() if column.null_count else None,
+        )
+
+    if _is_list_like(target_type) and _is_list_like(column.type):
+        return _coerce_blob_list_values(column, target_type.value_field)
+
+    return column
+
+
+def _coerce_blob_list_values(
+    column: pa.Array, target_value_field: pa.Field
+) -> pa.Array:
+    """Coerce blob values inside a list column, preserving offsets and nulls.
+
+    Works on the raw child values window instead of ``pc.list_flatten`` because
+    flatten drops values spanned by null slots, which would misalign offsets.
+    """
+    mask = column.is_null() if column.null_count else None
+    if pa.types.is_fixed_size_list(column.type):
+        list_size = column.type.list_size
+        values = column.values.slice(column.offset * list_size, len(column) * list_size)
+        coerced = _coerce_blob_value(values, target_value_field)
+        if coerced is values:
+            return column
+        physical_values, _ = _physical_array_and_type(coerced)
+        return pa.FixedSizeListArray.from_arrays(physical_values, list_size, mask=mask)
+    offsets = column.offsets
+    first_offset = offsets[0].as_py()
+    values = column.values.slice(
+        first_offset,
+        offsets[-1].as_py() - first_offset,
+    )
+    coerced = _coerce_blob_value(values, target_value_field)
+    if coerced is values:
+        return column
+    physical_values, _ = _physical_array_and_type(coerced)
+    if first_offset:
+        offsets = pc.subtract(offsets, pa.scalar(first_offset, offsets.type))
+    if pa.types.is_large_list(column.type):
+        return pa.LargeListArray.from_arrays(offsets, physical_values, mask=mask)
+    return pa.ListArray.from_arrays(offsets, physical_values, mask=mask)
+
+
+def _coerce_value_to_blob(values: pa.Array, target_field: pa.Field) -> pa.Array:
+    if pa.types.is_null(values.type):
+        data = pa.nulls(len(values), type=pa.large_binary())
+    elif pa.types.is_large_binary(values.type):
+        data = values
+    else:
+        data = values.cast(pa.large_binary())
+    length = len(values)
+    storage_type = target_field.type
+    if isinstance(storage_type, pa.ExtensionType):
+        storage_type = storage_type.storage_type
+    storage_fields = list(storage_type)
+    children = []
+    for storage_field in storage_fields:
+        if storage_field.name == "data":
+            children.append(data)
+        else:
+            children.append(pa.nulls(length, type=storage_field.type))
+    storage = pa.StructArray.from_arrays(
+        children,
+        fields=storage_fields,
+        mask=values.is_null() if values.null_count else None,
+    )
+    if isinstance(target_field.type, pa.ExtensionType):
+        return pa.ExtensionArray.from_storage(target_field.type, storage)
+    return storage
+
+
+def _physical_array_and_type(array: pa.Array) -> tuple[pa.Array, pa.DataType]:
+    if isinstance(array.type, pa.ExtensionType):
+        return array.storage, array.type.storage_type
+    return array, array.type
+
+
+def _can_coerce_to_blob(data_type: pa.DataType) -> bool:
+    return _is_binary_like(data_type) or pa.types.is_null(data_type)
+
+
+def _is_binary_like(data_type: pa.DataType) -> bool:
+    return (
+        pa.types.is_binary(data_type)
+        or pa.types.is_large_binary(data_type)
+        or pa.types.is_binary_view(data_type)
+    )
+
+
 def _field_extension_name(field: pa.Field) -> Optional[str]:
    extension_name = getattr(field.type, "extension_name", None)
    if extension_name is not None:
@@ -464,63 +713,71 @@ def _align_field_types(
        target_field = next((f for f in target_fields if f.name == field.name), None)
        if target_field is None:
            raise ValueError(f"Field '{field.name}' not found in target schema")
-        # Preserve arrow.json input until it reaches Lance. LanceDB exposes stored
-        # JSON columns as lance.json (JSONB-backed LargeBinary), but casting the
-        # input to that storage type here merely relabels the raw JSON bytes as
-        # JSONB. Lance must see arrow.json so it can perform the JSONB encoding.
-        if (
-            _field_extension_name(field) == "arrow.json"
-            and _field_extension_name(target_field) == "lance.json"
-        ):
-            new_fields.append(field)
-            continue
-        if pa.types.is_struct(target_field.type):
-            if pa.types.is_struct(field.type):
-                new_type = pa.struct(
-                    _align_field_types(
-                        field.type.fields,
-                        target_field.type.fields,
-                    )
+        new_fields.append(_align_field(field, target_field))
+    return new_fields
+
+
+def _align_list_value_field(
+    value_field: pa.Field, target_value_field: pa.Field
+) -> pa.Field:
+    # A list has exactly one child, so the inferred child name ("item") aligns
+    # positionally and adopts the table's child name; pa.Table.cast renames it.
+    return _align_field(value_field, target_value_field).with_name(
+        target_value_field.name
+    )
+
+
+def _align_field(field: pa.Field, target_field: pa.Field) -> pa.Field:
+    # Preserve arrow.json input until it reaches Lance. LanceDB exposes stored
+    # JSON columns as lance.json (JSONB-backed LargeBinary), but casting the
+    # input to that storage type here merely relabels the raw JSON bytes as
+    # JSONB. Lance must see arrow.json so it can perform the JSONB encoding.
+    if (
+        _field_extension_name(field) == "arrow.json"
+        and _field_extension_name(target_field) == "lance.json"
+    ):
+        return field
+    if pa.types.is_struct(target_field.type):
+        if pa.types.is_struct(field.type):
+            new_type = pa.struct(
+                _align_field_types(
+                    field.type.fields,
+                    target_field.type.fields,
                )
-            else:
-                new_type = target_field.type
-        elif pa.types.is_list(target_field.type):
-            if _is_list_like(field.type):
-                new_type = pa.list_(
-                    _align_field_types(
-                        [field.type.value_field],
-                        [target_field.type.value_field],
-                    )[0]
-                )
-            else:
-                new_type = target_field.type
-        elif pa.types.is_large_list(target_field.type):
-            if _is_list_like(field.type):
-                new_type = pa.large_list(
-                    _align_field_types(
-                        [field.type.value_field],
-                        [target_field.type.value_field],
-                    )[0]
-                )
-            else:
-                new_type = target_field.type
-        elif pa.types.is_fixed_size_list(target_field.type):
-            if _is_list_like(field.type):
-                new_type = pa.list_(
-                    _align_field_types(
-                        [field.type.value_field],
-                        [target_field.type.value_field],
-                    )[0],
-                    target_field.type.list_size,
-                )
-            else:
-                new_type = target_field.type
+            )
        else:
            new_type = target_field.type
-        new_fields.append(
-            pa.field(field.name, new_type, field.nullable, target_field.metadata)
-        )
-    return new_fields
+    elif pa.types.is_list(target_field.type):
+        if _is_list_like(field.type):
+            new_type = pa.list_(
+                _align_list_value_field(
+                    field.type.value_field, target_field.type.value_field
+                )
+            )
+        else:
+            new_type = target_field.type
+    elif pa.types.is_large_list(target_field.type):
+        if _is_list_like(field.type):
+            new_type = pa.large_list(
+                _align_list_value_field(
+                    field.type.value_field, target_field.type.value_field
+                )
+            )
+        else:
+            new_type = target_field.type
+    elif pa.types.is_fixed_size_list(target_field.type):
+        if _is_list_like(field.type):
+            new_type = pa.list_(
+                _align_list_value_field(
+                    field.type.value_field, target_field.type.value_field
+                ),
+                target_field.type.list_size,
+            )
+        else:
+            new_type = target_field.type
+    else:
+        new_type = target_field.type
+    return pa.field(field.name, new_type, field.nullable, target_field.metadata)


 def _infer_subschema(
@@ -589,7 +846,7 @@ def sanitize_create_table(
        schema = data.schema
    else:
        if schema is not None:
-            data = pa.Table.from_pylist([], schema)
+            data = pa.Table.from_batches([], schema=schema)
    if schema is None:
        if data is None:
            raise ValueError("Either data or schema must be provided")
@@ -1875,6 +2132,7 @@ class Table(ABC):
        cleanup_older_than: Optional[timedelta] = None,
        delete_unverified: bool = False,
        retrain: bool = False,
+        compaction_options: Optional[CompactionOptions] = None,
    ):
        """
        Optimize the on-disk data and indices for better performance.
@@ -1907,6 +2165,11 @@ class Table(ABC):

        retrain: bool, default False
            This parameter is no longer used and is deprecated.
+        compaction_options: CompactionOptions, optional
+            Options that control file compaction. For large rows, derive a lower
+            ``target_rows_per_fragment`` from the expected row size to bound later
+            passes. This does not cap the first pass over an existing oversized
+            fragment; see [CompactionOptions][lancedb.table.CompactionOptions].

        Notes
        -----
@@ -1991,9 +2254,11 @@ class Table(ABC):
            Function columns are supported only on LanceDB Cloud and
            Enterprise.
        computed: Dict[str, str], optional
-            A map of column name to a SQL expression defining the column. The
-            column's type and inputs are derived from the expression, so no
-            data type is supplied.
+            A mapping from output column names to SQL expressions derives each
+            output field from its expression. A direct projection of a Blob v2
+            field inherits Blob v2 semantics; other expressions derive their
+            ordinary Arrow type. Mapping order is declaration and dependency
+            order.

            Unlike ``transforms``, the expression is stored rather than
            evaluated now: the column is committed with no values, and rows get
@@ -2698,7 +2963,7 @@ class LanceTable(Table):
            arrow_tbl = self.to_arrow()
            if blob_mode == "descriptions":
                arrow_tbl = strip_auto_row_ids(
-                    arrow_tbl, blob_v2_column_paths(self.schema)
+                    arrow_tbl, row_addressable_blob_v2_paths(self.schema)
                )
            return arrow_tbl.to_pandas(**kwargs)

@@ -4025,6 +4290,7 @@ class LanceTable(Table):
        cleanup_older_than: Optional[timedelta] = None,
        delete_unverified: bool = False,
        retrain: bool = False,
+        compaction_options: Optional[CompactionOptions] = None,
    ):
        """
        Optimize the on-disk data and indices for better performance.
@@ -4057,6 +4323,11 @@ class LanceTable(Table):

        retrain: bool, default False
            This parameter is no longer used and is deprecated.
+        compaction_options: CompactionOptions, optional
+            Options that control file compaction. For large rows, derive a lower
+            ``target_rows_per_fragment`` from the expected row size to bound later
+            passes. This does not cap the first pass over an existing oversized
+            fragment; see [CompactionOptions][lancedb.table.CompactionOptions].

        Notes
        -----
@@ -4071,6 +4342,7 @@ class LanceTable(Table):
                cleanup_older_than=cleanup_older_than,
                delete_unverified=delete_unverified,
                retrain=retrain,
+                compaction_options=compaction_options,
            )
        )

@@ -5102,7 +5374,9 @@ class AsyncTable:
        if blob_mode == "descriptions" or not schema_has_blob_field(schema):
            arrow_tbl = await self.to_arrow()
            if blob_mode == "descriptions":
-                arrow_tbl = strip_auto_row_ids(arrow_tbl, blob_v2_column_paths(schema))
+                arrow_tbl = strip_auto_row_ids(
+                    arrow_tbl, row_addressable_blob_v2_paths(schema)
+                )
            return arrow_tbl.to_pandas(**kwargs)

        if blob_mode == "lazy" and get_uri_scheme(await self.uri()) == "memory":
@@ -6092,8 +6366,11 @@ class AsyncTable:
            Function columns are supported only on LanceDB Cloud and
            Enterprise.
        computed: Dict[str, str], optional
-            A map of column name to a SQL expression defining the column. The
-            column's type and inputs are derived from the expression.
+            A mapping from output column names to SQL expressions derives each
+            output field from its expression. A direct projection of a Blob v2
+            field inherits Blob v2 semantics; other expressions derive their
+            ordinary Arrow type. Mapping order is declaration and dependency
+            order.

            Unlike ``transforms``, the expression is stored rather than
            evaluated now: the column is committed with no values, and rows get
@@ -6467,6 +6744,7 @@ class AsyncTable:
        cleanup_older_than: Optional[timedelta] = None,
        delete_unverified: bool = False,
        retrain=False,
+        compaction_options: Optional[CompactionOptions] = None,
    ) -> OptimizeStats:
        """
        Optimize the on-disk data and indices for better performance.
@@ -6499,6 +6777,11 @@ class AsyncTable:

        retrain: bool, default False
            This parameter is no longer used and is deprecated.
+        compaction_options: CompactionOptions, optional
+            Options that control file compaction. For large rows, derive a lower
+            ``target_rows_per_fragment`` from the expected row size to bound later
+            passes. This does not cap the first pass over an existing oversized
+            fragment; see [CompactionOptions][lancedb.table.CompactionOptions].

        Notes
        -----
@@ -6525,6 +6808,7 @@ class AsyncTable:
        return await self._inner.optimize(
            cleanup_since_ms=cleanup_since_ms,
            delete_unverified=delete_unverified,
+            compaction_options=compaction_options,
        )

    async def list_indices(self) -> Iterable[IndexConfig]:
@@ -2,10 +2,15 @@
 # SPDX-FileCopyrightText: Copyright The LanceDB Authors

 import io
+import subprocess
+import sys
+import textwrap

+import lance
 import pyarrow as pa
 import pyarrow.compute as pc
 import pytest
+from lance.blob import BlobType as LanceBlobType

 import lancedb
 from lancedb._blob import (
@@ -18,6 +23,20 @@ from lancedb.index import FTS
 from lancedb.schema import blob_column_paths, blob_v2_column_paths


+_HIDE_LANCE_BLOB = """\
+import importlib.abc
+import sys
+
+class _MissingLanceBlob(importlib.abc.MetaPathFinder):
+    def find_spec(self, fullname, path, target=None):
+        if fullname == "lance.blob" or fullname.startswith("lance.blob."):
+            raise ModuleNotFoundError(fullname, name="lance.blob")
+
+sys.modules.pop("lance.blob", None)
+sys.meta_path.insert(0, _MissingLanceBlob())
+"""
+
+
 def _blob_table(name, rows):
    db = lancedb.connect("memory:///")
    schema = pa.schema([pa.field("id", pa.int64()), lancedb.blob("image")])
@@ -51,6 +70,181 @@ def test_blob_factory_declares_v2_field():
    field = lancedb.blob("image")
    assert isinstance(field.type, pa.ExtensionType)
    assert field.type.extension_name == "lance.blob.v2"
+    assert lancedb.BlobType is LanceBlobType
+    assert type(field.type) is LanceBlobType
+
+
+def test_blob_type_works_without_pylance():
+    script = _HIDE_LANCE_BLOB + textwrap.dedent(
+        """\
+        import lancedb
+        import pyarrow as pa
+
+        field = lancedb.blob("image")
+        if not isinstance(field.type, pa.ExtensionType):
+            raise SystemExit("expected an extension type")
+        if field.type.extension_name != "lance.blob.v2":
+            raise SystemExit(field.type.extension_name)
+        if lancedb.BlobType is not type(field.type):
+            raise SystemExit("BlobType is not the field type class")
+        if lancedb.BlobType.__module__ != "lancedb.schema":
+            raise SystemExit(lancedb.BlobType.__module__)
+
+        db = lancedb.connect("memory:///")
+        table = db.create_table(
+            "images",
+            schema=pa.schema([pa.field("id", pa.int64()), field]),
+        )
+        table.add([{"id": 1, "image": b"hello"}])
+        result = (
+            table.merge_insert("id")
+            .when_matched_update_all()
+            .when_not_matched_insert_all()
+            .execute([{"id": 1, "image": b"updated"}, {"id": 2, "image": b"inserted"}])
+        )
+        if result.num_updated_rows != 1 or result.num_inserted_rows != 1:
+            raise SystemExit(
+                f"merge_insert rows updated={result.num_updated_rows} "
+                f"inserted={result.num_inserted_rows}"
+            )
+        """
+    )
+    result = subprocess.run(
+        [sys.executable, "-c", script],
+        capture_output=True,
+        text=True,
+        check=False,
+    )
+    assert result.returncode == 0, result.stderr
+
+
+def test_blob_resolves_pylance_type_without_eager_import():
+    script = textwrap.dedent(
+        """\
+        import sys
+        import lancedb
+
+        if "lance.blob" in sys.modules:
+            raise SystemExit("import lancedb imported lance.blob")
+        field = lancedb.blob("image")
+        from lance.blob import BlobType
+
+        if type(field.type) is not BlobType:
+            raise SystemExit(f"{type(field.type)} is not {BlobType}")
+        import lance
+
+        image = lance.blob_array([b"x"])
+        if type(image.type) is not BlobType:
+            raise SystemExit("blob_array used a different class")
+        if type(image.type) is not type(field.type):
+            raise SystemExit("field and array classes differ")
+        """
+    )
+    result = subprocess.run(
+        [sys.executable, "-c", script],
+        capture_output=True,
+        text=True,
+        check=False,
+    )
+    assert result.returncode == 0, result.stderr
+
+
+def test_blob_fallback_fails_if_name_already_registered():
+    script = _HIDE_LANCE_BLOB + textwrap.dedent(
+        """\
+        import pyarrow as pa
+
+        class OtherBlobType(pa.ExtensionType):
+            def __init__(self):
+                super().__init__(
+                    pa.struct([pa.field("data", pa.large_binary())]),
+                    "lance.blob.v2",
+                )
+
+            def __arrow_ext_serialize__(self):
+                return b""
+
+            @classmethod
+            def __arrow_ext_deserialize__(cls, storage_type, serialized):
+                return cls()
+
+        pa.register_extension_type(OtherBlobType())
+        import lancedb
+
+        try:
+            lancedb.blob("image")
+        except ValueError as err:
+            if "already registered" not in str(err):
+                raise SystemExit(err)
+        else:
+            raise SystemExit("expected ValueError")
+        """
+    )
+    result = subprocess.run(
+        [sys.executable, "-c", script],
+        capture_output=True,
+        text=True,
+        check=False,
+    )
+    assert result.returncode == 0, result.stderr
+
+
+def test_blob_type_rejects_competing_registration_with_pylance():
+    script = textwrap.dedent(
+        """\
+        import pyarrow as pa
+        import pyarrow.ipc
+
+        class OtherBlobType(pa.ExtensionType):
+            def __init__(self):
+                super().__init__(
+                    pa.struct(
+                        [
+                            pa.field("data", pa.large_binary()),
+                            pa.field("uri", pa.utf8()),
+                            pa.field("position", pa.uint64()),
+                            pa.field("size", pa.uint64()),
+                        ]
+                    ),
+                    "lance.blob.v2",
+                )
+
+            def __arrow_ext_serialize__(self):
+                return b""
+
+            @classmethod
+            def __arrow_ext_deserialize__(cls, storage_type, serialized):
+                return cls()
+
+        pa.register_extension_type(OtherBlobType())
+
+        from lance.blob import BlobType
+
+        if BlobType is OtherBlobType:
+            raise SystemExit("pylance BlobType was replaced")
+        schema = pa.schema([pa.field("value", BlobType())])
+        restored = pa.ipc.read_schema(schema.serialize())
+        if type(restored.field("value").type) is not OtherBlobType:
+            raise SystemExit(type(restored.field("value").type))
+
+        import lancedb
+
+        try:
+            lancedb.blob("image")
+        except ValueError as err:
+            if "__main__.OtherBlobType" not in str(err):
+                raise SystemExit(err)
+        else:
+            raise SystemExit("expected ValueError")
+        """
+    )
+    result = subprocess.run(
+        [sys.executable, "-c", script],
+        capture_output=True,
+        text=True,
+        check=False,
+    )
+    assert result.returncode == 0, result.stderr


 def test_blob_v2_column_paths_include_list_children():
@@ -203,6 +397,292 @@ def test_fetch_blobs_round_trip():
    assert [blobs[0].as_py(), blobs[1].as_py()] == [b"alpha", b"beta"]


+def test_merge_insert_writes_python_bytes():
+    table = _blob_table("merge_bytes", [{"id": 1, "image": b"before"}])
+    result = (
+        table.merge_insert("id")
+        .when_matched_update_all()
+        .when_not_matched_insert_all()
+        .execute([{"id": 1, "image": b"updated"}, {"id": 2, "image": b"inserted"}])
+    )
+    assert result.num_updated_rows == 1
+    assert result.num_inserted_rows == 1
+    by_id = _row_ids_by_id(table)
+    blobs = table.fetch_blobs("image", [by_id[1], by_id[2]])
+    assert blobs.to_pylist() == [b"updated", b"inserted"]
+
+
+def test_merge_insert_bytes_after_reopen_without_touching_blob_type(tmp_path):
+    db = lancedb.connect(tmp_path)
+    schema = pa.schema([pa.field("id", pa.int64()), lancedb.blob("image")])
+    table = db.create_table("images", schema=schema)
+    table.add([{"id": 1, "image": b"hello"}])
+
+    script = textwrap.dedent(
+        f"""\
+        import lancedb
+
+        db = lancedb.connect({str(tmp_path)!r})
+        table = db.open_table("images")
+        image_type = table.schema.field("image").type
+        if type(image_type).__name__ != "StructType":
+            raise SystemExit(f"expected StructType, got {{type(image_type)}}")
+        result = (
+            table.merge_insert("id")
+            .when_matched_update_all()
+            .when_not_matched_insert_all()
+            .execute(
+                [{{"id": 1, "image": b"updated"}}, {{"id": 2, "image": b"inserted"}}]
+            )
+        )
+        if result.num_updated_rows != 1 or result.num_inserted_rows != 1:
+            raise SystemExit(
+                f"rows updated={{result.num_updated_rows}} "
+                f"inserted={{result.num_inserted_rows}}"
+            )
+        hits = table.search().with_row_id(True).limit(10).to_arrow()
+        by_id = dict(zip(hits["id"].to_pylist(), hits["_rowid"].to_pylist()))
+        blobs = table.fetch_blobs("image", [by_id[1], by_id[2]])
+        if blobs.to_pylist() != [b"updated", b"inserted"]:
+            raise SystemExit(blobs.to_pylist())
+        """
+    )
+    result = subprocess.run(
+        [sys.executable, "-c", script],
+        capture_output=True,
+        text=True,
+        check=False,
+    )
+    assert result.returncode == 0, result.stderr
+
+
+def test_merge_insert_bytes_after_reopen_without_pylance(tmp_path):
+    db = lancedb.connect(tmp_path)
+    schema = pa.schema([pa.field("id", pa.int64()), lancedb.blob("image")])
+    table = db.create_table("images", schema=schema)
+    table.add([{"id": 1, "image": b"hello"}])
+
+    script = _HIDE_LANCE_BLOB + textwrap.dedent(
+        f"""\
+        import lancedb
+
+        db = lancedb.connect({str(tmp_path)!r})
+        table = db.open_table("images")
+        image_type = table.schema.field("image").type
+        if type(image_type).__name__ != "StructType":
+            raise SystemExit(f"expected StructType, got {{type(image_type)}}")
+        result = (
+            table.merge_insert("id")
+            .when_matched_update_all()
+            .when_not_matched_insert_all()
+            .execute(
+                [{{"id": 1, "image": b"updated"}}, {{"id": 2, "image": b"inserted"}}]
+            )
+        )
+        if result.num_updated_rows != 1 or result.num_inserted_rows != 1:
+            raise SystemExit(
+                f"rows updated={{result.num_updated_rows}} "
+                f"inserted={{result.num_inserted_rows}}"
+            )
+        hits = table.search().with_row_id(True).limit(10).to_arrow()
+        by_id = dict(zip(hits["id"].to_pylist(), hits["_rowid"].to_pylist()))
+        blobs = table.fetch_blobs("image", [by_id[1], by_id[2]])
+        if blobs.to_pylist() != [b"updated", b"inserted"]:
+            raise SystemExit(blobs.to_pylist())
+        """
+    )
+    result = subprocess.run(
+        [sys.executable, "-c", script],
+        capture_output=True,
+        text=True,
+        check=False,
+    )
+    assert result.returncode == 0, result.stderr
+
+
+def test_merge_insert_blob_array_into_reopened_unregistered_table(tmp_path):
+    db = lancedb.connect(tmp_path)
+    schema = pa.schema([pa.field("id", pa.int64()), lancedb.blob("image")])
+    table = db.create_table("images", schema=schema)
+    table.add([{"id": 1, "image": b"before"}])
+
+    script = textwrap.dedent(
+        f"""\
+        import pyarrow as pa
+        import lancedb
+
+        db = lancedb.connect({str(tmp_path)!r})
+        table = db.open_table("images")
+        image_type = table.schema.field("image").type
+        if type(image_type).__name__ != "StructType":
+            raise SystemExit(
+                f"expected StructType before lance import, got {{type(image_type)}}"
+            )
+
+        import lance
+
+        updates = pa.Table.from_arrays(
+            [
+                pa.array([1, 2], type=pa.int64()),
+                lance.blob_array([b"updated", b"inserted"]),
+            ],
+            names=["id", "image"],
+        )
+        result = (
+            table.merge_insert("id")
+            .when_matched_update_all()
+            .when_not_matched_insert_all()
+            .execute(updates)
+        )
+        if result.num_updated_rows != 1 or result.num_inserted_rows != 1:
+            raise SystemExit(
+                f"rows updated={{result.num_updated_rows}} "
+                f"inserted={{result.num_inserted_rows}}"
+            )
+        hits = table.search().with_row_id(True).limit(10).to_arrow()
+        by_id = dict(zip(hits["id"].to_pylist(), hits["_rowid"].to_pylist()))
+        blobs = table.fetch_blobs("image", [by_id[1], by_id[2]])
+        if blobs.to_pylist() != [b"updated", b"inserted"]:
+            raise SystemExit(blobs.to_pylist())
+        """
+    )
+    result = subprocess.run(
+        [sys.executable, "-c", script],
+        capture_output=True,
+        text=True,
+        check=False,
+    )
+    assert result.returncode == 0, result.stderr
+
+
+def test_add_all_null_blob_column():
+    db = lancedb.connect("memory:///")
+    schema = pa.schema([pa.field("id", pa.int64()), lancedb.blob("image")])
+    table = db.create_table("all_null", schema=schema)
+    table.add([{"id": 1, "image": None}, {"id": 2, "image": None}])
+    by_id = _row_ids_by_id(table)
+    blobs = table.fetch_blobs("image", [by_id[1], by_id[2]])
+    assert blobs.to_pylist() == [None, None]
+
+
+def test_create_table_nested_blob_schema_without_rows():
+    db = lancedb.connect("memory:///")
+    schema = pa.schema(
+        [
+            pa.field("id", pa.int64()),
+            pa.field("info", pa.struct([lancedb.blob("blob")])),
+            pa.field("images", pa.list_(lancedb.blob("image"))),
+        ]
+    )
+    table = db.create_table("nested_empty", schema=schema)
+    assert table.count_rows() == 0
+
+
+def test_merge_insert_nested_blob_dicts():
+    db = lancedb.connect("memory:///")
+    info = pa.StructArray.from_arrays(
+        [
+            pa.array(["first"], type=pa.string()),
+            _blob_array("blob", [b"before"]),
+        ],
+        names=["name", "blob"],
+    )
+    data = pa.Table.from_arrays(
+        [pa.array([1], type=pa.int64()), info],
+        names=["id", "info"],
+    )
+    table = db.create_table("nested_merge", data=data)
+    result = (
+        table.merge_insert("id")
+        .when_matched_update_all()
+        .execute([{"id": 1, "info": {"name": "first", "blob": b"after"}}])
+    )
+    assert result.num_updated_rows == 1
+    by_id = _row_ids_by_id(table)
+    blobs = table.fetch_blobs("info.blob", [by_id[1]])
+    assert blobs.to_pylist() == [b"after"]
+
+
+def _list_blob_table(name):
+    db = lancedb.connect("memory:///")
+    blob_field = lancedb.blob("image")
+    images = pa.ListArray.from_arrays(
+        pa.array([0, 1], type=pa.int32()), _blob_array("image", [b"before"])
+    )
+    data = pa.Table.from_arrays(
+        [pa.array([1], type=pa.int64()), images],
+        schema=pa.schema(
+            [pa.field("id", pa.int64()), pa.field("images", pa.list_(blob_field))]
+        ),
+    )
+    return db.create_table(name, data=data)
+
+
+def test_merge_insert_list_blob_dicts():
+    table = _list_blob_table("list_merge")
+    result = (
+        table.merge_insert("id")
+        .when_matched_update_all()
+        .when_not_matched_insert_all()
+        .execute([{"id": 1, "images": [b"one", b"two"]}, {"id": 2, "images": None}])
+    )
+    assert result.num_updated_rows == 1
+    assert result.num_inserted_rows == 1
+    hits = table.search().limit(10).to_arrow()
+    sizes = {
+        row["id"]: None if row["images"] is None else [d["size"] for d in row["images"]]
+        for row in hits.to_pylist()
+    }
+    assert sizes == {1: [3, 3], 2: None}
+
+
+def test_list_blob_column_queries_as_raw_descriptors():
+    table = _list_blob_table("list_query")
+    hits = table.search().limit(10).to_arrow()
+    element = hits.schema.field("images").type.value_type
+    assert pa.types.is_struct(element)
+    assert "_lance_row_id" not in element.names
+    with pytest.raises(ValueError, match="expected struct before segment"):
+        table.fetch_blobs("images.image", [0])
+
+
+def test_row_addressable_paths_exclude_list_children():
+    from lancedb.schema import row_addressable_blob_v2_paths
+
+    schema = pa.schema(
+        [
+            pa.field("id", pa.int64()),
+            pa.field("info", pa.struct([lancedb.blob("blob")])),
+            pa.field("images", pa.list_(lancedb.blob("image"))),
+        ]
+    )
+    assert blob_v2_column_paths(schema) == ["info.blob", "images.image"]
+    assert row_addressable_blob_v2_paths(schema) == ["info.blob"]
+
+
+def test_merge_insert_writes_pylance_blob_array():
+    table = _blob_table("merge_pylance", [{"id": 1, "image": b"before"}])
+    image = lance.blob_array([b"updated", b"inserted"])
+    assert type(image.type) is LanceBlobType
+    assert type(image.type) is type(lancedb.BlobType())
+    updates = pa.Table.from_arrays(
+        [pa.array([1, 2], type=pa.int64()), image], names=["id", "image"]
+    )
+
+    result = (
+        table.merge_insert("id")
+        .when_matched_update_all()
+        .when_not_matched_insert_all()
+        .execute(updates)
+    )
+
+    assert result.num_updated_rows == 1
+    assert result.num_inserted_rows == 1
+    by_id = _row_ids_by_id(table)
+    blobs = table.fetch_blobs("image", [by_id[1], by_id[2]])
+    assert blobs.to_pylist() == [b"updated", b"inserted"]
+
+
 def test_fetch_blobs_accepts_query_result():
    table = _blob_table("from_result", [{"id": 1, "image": b"gamma"}])
    hits = table.search().limit(10).to_arrow()
@@ -13,7 +13,7 @@ from concurrent.futures import ThreadPoolExecutor
 from datetime import date, datetime, timedelta
 from decimal import Decimal
 from time import sleep
-from typing import List
+from typing import Any, List
 from unittest.mock import patch

 import lancedb
@@ -3876,6 +3876,100 @@ async def test_optimize(mem_db_async: AsyncConnection):
    assert await table.query().to_arrow() == pa.table({"x": [[1], [2]]})


+@pytest.mark.asyncio
+async def test_optimize_compaction_options(mem_db_async: AsyncConnection):
+    table = await mem_db_async.create_table("test", data=[{"x": 1}])
+    await table.add([{"x": 2}])
+
+    stats = await table.optimize(
+        compaction_options={
+            "target_rows_per_fragment": 1,
+            "batch_size": 1,
+            "num_threads": 1,
+        }
+    )
+    assert stats.compaction.fragments_removed == 0
+    assert stats.compaction.fragments_added == 0
+
+    stats = await table.optimize(compaction_options={"target_rows_per_fragment": 3})
+    assert stats.compaction.fragments_removed == 2
+    assert stats.compaction.fragments_added == 1
+
+    with pytest.raises(ValueError, match="Invalid compaction option: unknown"):
+        await table.optimize(compaction_options={"unknown": 1})
+
+
+@pytest.mark.parametrize("option", ["max_source_rows", "max_source_bytes"])
+@pytest.mark.asyncio
+async def test_optimize_compaction_source_limits(
+    mem_db_async: AsyncConnection, option: str
+):
+    table = await mem_db_async.create_table("test", data=[{"x": 1}])
+    await table.add([{"x": 2}])
+
+    stats = await table.optimize(
+        compaction_options={
+            "target_rows_per_fragment": 3,
+            option: 1,
+        }
+    )
+    assert stats.compaction.fragments_removed == 0
+    assert stats.compaction.fragments_added == 0
+
+
+@pytest.mark.asyncio
+async def test_optimize_compaction_excluded_fragments(mem_db_async: AsyncConnection):
+    table = await mem_db_async.create_table("test", data=[{"x": 1}])
+    await table.add([{"x": 2}])
+
+    stats = await table.optimize(
+        compaction_options={
+            "target_rows_per_fragment": 3,
+            "excluded_fragment_ids": [0],
+        }
+    )
+    assert stats.compaction.fragments_removed == 0
+    assert stats.compaction.fragments_added == 0
+
+    stats = await table.optimize(compaction_options={"target_rows_per_fragment": 3})
+    assert stats.compaction.fragments_removed == 2
+    assert stats.compaction.fragments_added == 1
+
+
+@pytest.mark.parametrize(
+    ("option", "value", "message"),
+    [
+        ("target_rows_per_fragment", 0, "must be between 1 and 4294967295"),
+        ("max_rows_per_group", 0, "must be between 1 and 4294967295"),
+        ("batch_size", 0, "must be between 1 and 4294967295"),
+        ("num_threads", 0, "must be greater than 0"),
+        ("target_rows_per_fragment", 2**32, "must be between 1 and 4294967295"),
+        ("max_rows_per_group", 2**32, "must be between 1 and 4294967295"),
+        ("batch_size", 2**32, "must be between 1 and 4294967295"),
+        ("io_buffer_size", 2**63, "must be at most 9223372036854775807"),
+        ("max_source_rows", 0, "must be greater than 0"),
+        ("max_source_bytes", 0, "must be greater than 0"),
+        (
+            "excluded_fragment_ids",
+            [-1],
+            "must contain values between 0 and 4294967295",
+        ),
+        (
+            "excluded_fragment_ids",
+            [2**32],
+            "must contain values between 0 and 4294967295",
+        ),
+    ],
+)
+@pytest.mark.asyncio
+async def test_optimize_compaction_options_validation(
+    mem_db_async: AsyncConnection, option: str, value: Any, message: str
+):
+    table = await mem_db_async.create_table("test", data=[{"x": 1}])
+    with pytest.raises(ValueError, match=message):
+        await table.optimize(compaction_options={option: value})
+
+
@pytest.mark.asyncio
 async def test_optimize_delete_unverified(tmp_db_async: AsyncConnection, tmp_path):
    table = await tmp_db_async.create_table(
@@ -4087,6 +4181,29 @@ def test_computed_column_rejects_transforms_and_computed_together(tmp_path):
        table.add_columns({"a": "x + 1"}, computed={"b": "x * 2"})


+def test_computed_column_blob_projection_inherits_semantics(tmp_path):
+    schema = pa.schema([pa.field("id", pa.int64()), lancedb.blob("image")])
+    db = lancedb.connect(tmp_path)
+    table = db.create_table("computed_column_blob", schema=schema)
+    table.add(
+        [
+            {"id": 1, "image": b"hello"},
+            {"id": 2, "image": b""},
+            {"id": 3, "image": None},
+        ]
+    )
+
+    table.add_columns(computed={"image_copy": "image", "second_copy": "image_copy"})
+    assert table.refresh_column("image_copy").rows_filled == 2
+    assert table.refresh_column("second_copy").rows_filled == 2
+    assert table.blob_columns() == ["image", "image_copy", "second_copy"]
+
+    hits = table.search().with_row_id(True).limit(10).to_arrow()
+    rows = sorted(zip(hits["id"].to_pylist(), hits["_rowid"].to_pylist()))
+    copied = table.fetch_blobs("second_copy", [row_id for _, row_id in rows])
+    assert copied.to_pylist() == [b"hello", b"", None]
+
+
@pytest.mark.asyncio
 async def test_computed_column_async(tmp_path):
    db = await lancedb.connect_async(tmp_path)
@@ -7,6 +7,7 @@ import pathlib
 from typing import Optional

 import lance
+from lance.blob import BlobType as LanceBlobType
 from lancedb.conftest import MockTextEmbeddingFunction
 from lancedb.embeddings.base import EmbeddingFunctionConfig
 from lancedb.embeddings.registry import EmbeddingFunctionRegistry
@@ -907,6 +908,165 @@ def test_cast_to_target_schema():
    assert output == expected


+def test_cast_to_target_schema_coerces_binary_to_blob_v2():
+    data = pa.table({"image": pa.array([b"hello", None], type=pa.binary())})
+    target = pa.schema([lancedb.blob("image")])
+
+    output = _cast_to_target_schema(data.to_reader(), target).read_all()
+
+    image = output["image"].chunk(0)
+    assert type(image.type) is lancedb.BlobType
+    assert image.storage.to_pylist() == [
+        {"data": b"hello", "uri": None, "position": None, "size": None},
+        None,
+    ]
+
+
+def test_cast_to_target_schema_coerces_binary_to_metadata_blob_struct():
+    storage = lancedb.blob("image").type.storage_type
+    target = pa.schema(
+        [
+            pa.field(
+                "image",
+                storage,
+                metadata={
+                    b"ARROW:extension:name": b"lance.blob.v2",
+                    b"ARROW:extension:metadata": b"",
+                },
+            )
+        ]
+    )
+    data = pa.table({"image": pa.array([b"hello", None], type=pa.binary())})
+
+    output = _cast_to_target_schema(data.to_reader(), target).read_all()
+
+    image = output["image"].chunk(0)
+    assert not isinstance(image.type, pa.ExtensionType)
+    assert image.to_pylist() == [
+        {"data": b"hello", "uri": None, "position": None, "size": None},
+        None,
+    ]
+
+
+def test_cast_to_target_schema_coerces_nested_binary_blob():
+    data = pa.table(
+        {
+            "info": pa.array(
+                [{"blob": b"hello"}, {"blob": None}],
+                type=pa.struct([pa.field("blob", pa.binary())]),
+            )
+        }
+    )
+    target = pa.schema([pa.field("info", pa.struct([lancedb.blob("blob")]))])
+
+    output = _cast_to_target_schema(data.to_reader(), target).read_all()
+
+    blob = output["info"].chunk(0).field("blob")
+    assert type(blob.type) is lancedb.BlobType
+    assert blob.storage.to_pylist() == [
+        {"data": b"hello", "uri": None, "position": None, "size": None},
+        None,
+    ]
+
+
+def test_cast_to_target_schema_coerces_list_binary_blob_with_inferred_child_name():
+    data = pa.table(
+        {"images": pa.array([[b"a", b"b"], None], type=pa.list_(pa.binary()))}
+    )
+    target = pa.schema([pa.field("images", pa.list_(lancedb.blob("image")))])
+
+    output = _cast_to_target_schema(data.to_reader(), target).read_all()
+
+    images = output["images"].chunk(0)
+    assert images.type.value_field.name == "image"
+    assert type(images.type.value_type) is lancedb.BlobType
+    assert images.to_pylist()[1] is None
+    assert images.values.storage.to_pylist() == [
+        {"data": b"a", "uri": None, "position": None, "size": None},
+        {"data": b"b", "uri": None, "position": None, "size": None},
+    ]
+
+
+def test_list_blob_coercion_preserves_null_slots_with_nonzero_extent():
+    child = pa.field("image", pa.binary())
+    source = pa.ListArray.from_arrays(
+        pa.array([0, 2, 4], type=pa.int32()),
+        pa.array([b"a", b"b", b"dead", b"beef"], type=pa.binary()),
+        mask=pa.array([False, True]),
+    ).cast(pa.list_(child))
+    target = pa.schema([pa.field("images", pa.list_(lancedb.blob("image")))])
+
+    output = _cast_to_target_schema(
+        pa.table({"images": source}).to_reader(), target
+    ).read_all()
+
+    images = output["images"].chunk(0)
+    assert images.to_pylist()[1] is None
+    assert [b["data"] for b in images.to_pylist()[0]] == [b"a", b"b"]
+
+
+def test_fixed_size_list_blob_coercion_keeps_null_rows():
+    child = pa.field("frame", pa.binary())
+    source = (
+        pa.FixedSizeListArray.from_arrays(
+            pa.array([b"a", b"b", b"c", b"d"], type=pa.binary()), 2
+        )
+        .take(pa.array([0, None], type=pa.int32()))
+        .cast(pa.list_(child, 2))
+    )
+    target = pa.schema([pa.field("frames", pa.list_(lancedb.blob("frame"), 2))])
+
+    output = _cast_to_target_schema(
+        pa.table({"frames": source}).to_reader(), target
+    ).read_all()
+
+    frames = output["frames"].chunk(0)
+    assert frames.to_pylist()[1] is None
+    assert [b["data"] for b in frames.to_pylist()[0]] == [b"a", b"b"]
+
+
+def test_cast_to_target_schema_accepts_pylance_blob_v2():
+    target_type = lancedb.BlobType()
+    source = lance.blob_array([b"hello", None])
+    assert type(source.type) is LanceBlobType
+    assert type(source.type) is type(target_type)
+    data = pa.table({"image": source})
+    target = pa.schema([pa.field("image", target_type)])
+
+    output = _cast_to_target_schema(data.to_reader(), target).read_all()
+
+    image = output["image"].chunk(0)
+    assert type(image.type) is LanceBlobType
+    assert image.type == target_type
+    assert image.storage.to_pylist() == [
+        {"data": b"hello", "uri": None, "position": None, "size": None},
+        None,
+    ]
+
+
+def test_cast_to_target_schema_rejects_different_blob_v2_class():
+    class OtherBlobType(pa.ExtensionType):
+        def __init__(self):
+            super().__init__(lancedb.BlobType().storage_type, "lance.blob.v2")
+
+        def __arrow_ext_serialize__(self) -> bytes:
+            return b""
+
+        @classmethod
+        def __arrow_ext_deserialize__(
+            cls, storage_type: pa.DataType, serialized: bytes
+        ) -> "OtherBlobType":
+            return cls()
+
+    storage = lance.blob_array([b"hello"]).storage
+    source = pa.ExtensionArray.from_storage(OtherBlobType(), storage)
+    data = pa.table({"image": source})
+    target = pa.schema([lancedb.blob("image")])
+
+    with pytest.raises(pa.ArrowTypeError, match="different extension type"):
+        _cast_to_target_schema(data.to_reader(), target).read_all()
+
+
 def test_sanitize_data_stream():
    # Make sure we don't collect the whole stream when running sanitize_data
    schema = pa.schema({"a": pa.int32()})
@@ -20,8 +20,9 @@ use arrow::{
 use lancedb::blob::{BlobFile, BlobRangeRequest};
 use lancedb::index::scalar::FtsIndexBuilder;
 use lancedb::table::{
-    AddDataMode, ColumnAlteration, Duration, FieldMetadataUpdate, FtsToken as LanceDbFtsToken,
-    NewColumnTransform, OptimizeAction, OptimizeOptions, Ref, Table as LanceDbTable,
+    AddDataMode, ColumnAlteration, CompactionMode, CompactionOptions, Duration,
+    FieldMetadataUpdate, FtsToken as LanceDbFtsToken, IndexRemapMode, NewColumnTransform,
+    OptimizeAction, OptimizeOptions, Ref, Table as LanceDbTable,
 };
 use lancedb::tokenize as lancedb_tokenize;
 use pyo3::{
@@ -100,6 +101,128 @@ enum PredicateArg {
    Sql(String),
 }

+fn validate_positive_u32(value: u64, name: &str) -> PyResult<usize> {
+    if !(1..=u32::MAX as u64).contains(&value) {
+        return Err(PyValueError::new_err(format!(
+            "{name} must be between 1 and {}",
+            u32::MAX
+        )));
+    }
+    Ok(value as usize)
+}
+
+fn positive_u32(value: &Bound<'_, PyAny>, name: &str) -> PyResult<usize> {
+    validate_positive_u32(value.extract()?, name)
+}
+
+fn optional_positive_u32(value: &Bound<'_, PyAny>, name: &str) -> PyResult<Option<usize>> {
+    value
+        .extract::<Option<u64>>()?
+        .map(|value| validate_positive_u32(value, name))
+        .transpose()
+}
+
+fn optional_positive_usize(value: &Bound<'_, PyAny>, name: &str) -> PyResult<Option<usize>> {
+    let value: Option<usize> = value.extract()?;
+    if value == Some(0) {
+        return Err(PyValueError::new_err(format!(
+            "{name} must be greater than 0"
+        )));
+    }
+    Ok(value)
+}
+
+fn optional_positive_u64(value: &Bound<'_, PyAny>, name: &str) -> PyResult<Option<u64>> {
+    let value: Option<u64> = value.extract()?;
+    if value == Some(0) {
+        return Err(PyValueError::new_err(format!(
+            "{name} must be greater than 0"
+        )));
+    }
+    Ok(value)
+}
+
+fn u32_list(value: &Bound<'_, PyAny>, name: &str) -> PyResult<Vec<u32>> {
+    value
+        .extract::<Vec<i64>>()?
+        .into_iter()
+        .map(|value| {
+            u32::try_from(value).map_err(|_| {
+                PyValueError::new_err(format!(
+                    "{name} must contain values between 0 and {}",
+                    u32::MAX
+                ))
+            })
+        })
+        .collect()
+}
+
+fn optional_i64_bounded_u64(value: &Bound<'_, PyAny>, name: &str) -> PyResult<Option<u64>> {
+    let value: Option<u64> = value.extract()?;
+    if value.is_some_and(|value| value > i64::MAX as u64) {
+        return Err(PyValueError::new_err(format!(
+            "{name} must be at most {}",
+            i64::MAX
+        )));
+    }
+    Ok(value)
+}
+
+fn parse_compaction_options(options: Option<&Bound<'_, PyDict>>) -> PyResult<CompactionOptions> {
+    let mut parsed = CompactionOptions::default();
+    let Some(options) = options else {
+        return Ok(parsed);
+    };
+
+    for (key, value) in options.iter() {
+        let key: String = key.extract()?;
+        match key.as_str() {
+            "target_rows_per_fragment" => {
+                parsed.target_rows_per_fragment = positive_u32(&value, &key)?
+            }
+            "max_rows_per_group" => parsed.max_rows_per_group = positive_u32(&value, &key)?,
+            "max_bytes_per_file" => parsed.max_bytes_per_file = value.extract()?,
+            "materialize_deletions" => parsed.materialize_deletions = value.extract()?,
+            "materialize_deletions_threshold" => {
+                parsed.materialize_deletions_threshold = value.extract()?
+            }
+            "num_threads" => parsed.num_threads = optional_positive_usize(&value, &key)?,
+            "batch_size" => parsed.batch_size = optional_positive_u32(&value, &key)?,
+            "io_buffer_size" => parsed.io_buffer_size = optional_i64_bounded_u64(&value, &key)?,
+            "defer_index_remap" => parsed.defer_index_remap = value.extract()?,
+            "index_remap_mode" => {
+                let mode: String = value.extract()?;
+                parsed.index_remap_mode = IndexRemapMode::try_from(mode.as_str())
+                    .map_err(|err| PyValueError::new_err(err.to_string()))?;
+            }
+            "compaction_mode" => {
+                let mode: Option<String> = value.extract()?;
+                parsed.compaction_mode = mode
+                    .map(|mode| {
+                        CompactionMode::try_from(mode.as_str())
+                            .map_err(|err| PyValueError::new_err(err.to_string()))
+                    })
+                    .transpose()?;
+            }
+            "binary_copy_read_batch_bytes" => {
+                parsed.binary_copy_read_batch_bytes = value.extract()?
+            }
+            "max_source_fragments" => parsed.max_source_fragments = value.extract()?,
+            "max_source_rows" => parsed.max_source_rows = optional_positive_usize(&value, &key)?,
+            "max_source_bytes" => parsed.max_source_bytes = optional_positive_u64(&value, &key)?,
+            "excluded_fragment_ids" => parsed.excluded_fragment_ids = u32_list(&value, &key)?,
+            "max_overlays_per_fragment" => parsed.max_overlays_per_fragment = value.extract()?,
+            _ => {
+                return Err(PyValueError::new_err(format!(
+                    "Invalid compaction option: {key}"
+                )));
+            }
+        }
+    }
+
+    Ok(parsed)
+}
+
 /// Statistics about a compaction operation.
 #[pyclass(get_all, from_py_object)]
 #[derive(Clone, Debug)]
@@ -1340,13 +1463,15 @@ impl Table {
    }

    /// Optimize the on-disk data by compacting and pruning old data, for better performance.
-    #[pyo3(signature = (cleanup_since_ms=None, delete_unverified=None))]
-    pub fn optimize(
-        self_: PyRef<'_, Self>,
+    #[pyo3(signature = (cleanup_since_ms=None, delete_unverified=None, compaction_options=None))]
+    pub fn optimize<'py>(
+        self_: PyRef<'py, Self>,
        cleanup_since_ms: Option<u64>,
        delete_unverified: Option<bool>,
-    ) -> PyResult<Bound<'_, PyAny>> {
+        compaction_options: Option<&Bound<'py, PyDict>>,
+    ) -> PyResult<Bound<'py, PyAny>> {
        let inner = self_.inner_ref()?.clone();
+        let compaction_options = parse_compaction_options(compaction_options)?;
        let older_than = if let Some(ms) = cleanup_since_ms {
            if ms > i64::MAX as u64 {
                return Err(PyValueError::new_err(format!(
@@ -1362,7 +1487,7 @@ impl Table {
        future_into_py(self_.py(), async move {
            let compaction_stats = inner
                .optimize(OptimizeAction::Compact {
-                    options: lancedb::table::CompactionOptions::default(),
+                    options: compaction_options,
                    remap_options: None,
                })
                .await
@@ -1,6 +1,6 @@
 [package]
 name = "lancedb"
-version = "0.38.0-beta.11"
+version = "0.38.0-beta.12"
 edition.workspace = true
 description = "LanceDB: A serverless, low-latency vector database for AI applications"
 license.workspace = true
@@ -3180,8 +3180,8 @@ impl<S: HttpSend> BaseTable for RemoteTable<S> {
            self.schema().await?.as_ref(),
            "schema evolution",
        )?;
-        // The server plans the declaration: expression validation, type
-        // inference and the persisted binding all happen there.
+        // The server plans the declaration against its table schema, including
+        // Blob v2 semantics inherited by a direct field projection.
        let entries = columns
            .iter()
            .map(
@@ -5734,8 +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}` entries for the server to
-    /// plan; the client never types the expression itself.
+    /// A declaration is sent as `{name, computed}` for the server to plan; the
+    /// client never types the expression itself.
    #[tokio::test]
    async fn test_add_computed_columns_sends_the_expression() {
        let table = Table::new_with_handler("my_table", |request| match request.url().path() {
@@ -103,7 +103,9 @@ pub use lance::dataset::refs::{BranchContents, Ref, TagContents, Tags as LanceTa
 pub use lance::dataset::scanner::DatasetRecordBatchStream;
 pub use lance_index::optimize::OptimizeOptions;
 pub use lsm_stats::{BucketStats, GenerationStats, LsmStats, MemtableStats};
-pub use optimize::{CompactionOptions, OptimizeAction, OptimizeStats};
+pub use optimize::{
+    CompactionMode, CompactionOptions, IndexRemapMode, OptimizeAction, OptimizeStats,
+};
 pub use refresh::RefreshColumnResult;
 pub use schema_evolution::{
    AddColumnsResult, AlterColumnsResult, DropColumnsResult, FieldMetadataUpdate,
@@ -750,8 +752,8 @@ pub trait BaseTable: std::fmt::Display + std::fmt::Debug + Send + Sync {
    /// Declare computed columns, each defined by a SQL expression.
    ///
    /// Where the declaration is planned depends on the backend: a local table
-    /// validates and types the expression itself, a remote one sends the text
-    /// for the server to plan.
+    /// validates and types the expression itself, while a remote one sends the
+    /// expression for the server to plan.
    async fn add_computed_columns(
        &self,
        _columns: &[(String, String)],
@@ -9,29 +9,35 @@
 //! 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 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.
+//! 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.
 //!
 //! [`computed_columns`] and [`computed_column_from_field`] read declarations
 //! back off a schema.

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

 use arrow_schema::{DataType, Field as ArrowField, Fields, Schema as ArrowSchema, SchemaRef};
-use datafusion_common::tree_node::TreeNode;
+use datafusion_common::{ScalarValue, tree_node::TreeNode};
+use datafusion_expr::Expr;
 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"`.
@@ -1106,15 +1112,20 @@ pub(crate) fn ensure_no_foreign_declarations<'a>(
    fields: impl IntoIterator<Item = &'a Arc<ArrowField>>,
 ) -> Result<()> {
    for field in fields {
-        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()
-                ),
-            });
-        }
+        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()
+            ),
+        });
    }
    Ok(())
 }
@@ -1162,15 +1173,154 @@ 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 [`Self::read_schema`]
-    /// order. A nested input appears through its root.
+    /// The top-level columns evaluation reads, in physical-expression 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`.
@@ -1185,10 +1335,18 @@ pub(crate) fn bind(schema: SchemaRef, column: &str, expression: &str) -> Result<
        message,
    };

-    let planner = Planner::new(schema.clone());
+    // 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 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
@@ -1218,13 +1376,19 @@ 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 = schema
+        let index = runtime_schema
            .index_of(root(input))
            .map_err(|_| invalid(format!("unknown column '{input}'")))?;
        if !indices.contains(&index) {
@@ -1237,7 +1401,7 @@ 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(
-        schema
+        runtime_schema
            .project(&indices)
            .map_err(|e| invalid(e.to_string()))?,
    );
@@ -1247,7 +1411,8 @@ pub(crate) fn bind(schema: SchemaRef, column: &str, expression: &str) -> Result<
        .map(|field| field.name().clone())
        .collect();

-    let optimized = planner
+    let runtime_planner = Planner::new(runtime_schema);
+    let optimized = runtime_planner
        .optimize_expr(parsed)
        .map_err(|e| invalid(e.to_string()))?;
    let physical = Planner::new(read_schema.clone())
@@ -1260,9 +1425,16 @@ 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,
    })
 }

@@ -1275,10 +1447,10 @@ pub(crate) fn bind(schema: SchemaRef, column: &str, expression: &str) -> Result<
 /// is one a refresh can always act on.
 ///
 /// Each accepted column joins the schema the next one resolves against, so a
-/// batch may declare `a` and then `b = a + 1` in one commit. Refresh fills a
-/// column's computed inputs before the column, so the order of refresh calls
-/// does not matter.
-pub(crate) fn plan(schema: SchemaRef, columns: &[(String, String)]) -> Result<Vec<ArrowField>> {
+/// 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>> {
    if columns.is_empty() {
        return Err(Error::InvalidInput {
            message: "at least one computed column is required".into(),
@@ -1290,15 +1462,28 @@ pub(crate) fn plan(schema: SchemaRef, columns: &[(String, String)]) -> Result<Ve

    for (name, expression) in columns {
        if schema.field_with_name(name).is_ok() {
-            return Err(Error::ColumnAlreadyExists { name: name.clone() });
+            return Err(Error::ColumnAlreadyExists {
+                name: name.to_string(),
+            });
        }

        let bound = bind(schema.clone(), name, expression)?;

        // Declared columns start entirely null, so nullability is a property
        // of the declaration rather than of what the expression yields.
-        let field = ArrowField::new(name, bound.data_type, true)
-            .with_metadata(computed_column_metadata(expression, &bound.inputs));
+        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),
+        };
        schema = Arc::new(ArrowSchema::new_with_metadata(
            schema
                .fields()
@@ -1314,6 +1499,10 @@ pub(crate) fn plan(schema: SchemaRef, columns: &[(String, String)]) -> Result<Ve
    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
@@ -1352,7 +1541,7 @@ pub(crate) fn declare(
    schema: SchemaRef,
    columns: &[(String, String)],
 ) -> Result<NewColumnTransform> {
-    let fields = plan(schema, columns)?;
+    let fields = plan_declarations(schema, columns)?;
    Ok(NewColumnTransform::AllNulls(Arc::new(ArrowSchema::new(
        fields,
    ))))
@@ -1478,6 +1667,44 @@ 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]
@@ -36,6 +36,14 @@ pub(super) fn coerce_blob_expr(
    };

    let input_shape = match input_field.data_type() {
+        DataType::Null => {
+            let expr: Arc<dyn PhysicalExpr> = Arc::new(CastExpr::new(
+                input_expr,
+                table_field.data_type().clone(),
+                None,
+            ));
+            return Ok((expr, table_field.clone()));
+        }
        DataType::Binary | DataType::LargeBinary | DataType::BinaryView => BlobInputShape::Bytes,
        DataType::Utf8 | DataType::LargeUtf8 | DataType::Utf8View => BlobInputShape::String,
        DataType::Struct(children) => {
@@ -155,7 +163,7 @@ mod tests {
    use crate::blob::blob;
    use arrow_array::{
        Array, ArrayRef, BinaryArray, BinaryViewArray, Int32Array, Int64Array, LargeBinaryArray,
-        RecordBatch, StringArray, StringViewArray, StructArray, UInt8Array, UInt64Array,
+        NullArray, RecordBatch, StringArray, StringViewArray, StructArray, UInt8Array, UInt64Array,
    };
    use arrow_schema::Schema;
    use datafusion::prelude::SessionContext;
@@ -279,6 +287,18 @@ mod tests {
        assert_eq!(data.value(0), b"view");
    }

+    #[tokio::test]
+    async fn null_column_coerces_to_all_null_blob_struct() {
+        let batch = batch_with_image(
+            Field::new("image", DataType::Null, true),
+            Arc::new(NullArray::new(2)),
+        );
+        let coerced = coerce(batch, &blob_table_schema()).await;
+        let image = image_struct(&coerced);
+        assert!(image.is_null(0));
+        assert!(image.is_null(1));
+    }
+
    #[tokio::test]
    async fn binary_nulls_stay_null_after_coercion() {
        let batch = batch_with_image(
@@ -15,7 +15,7 @@ use lance_index::optimize::OptimizeOptions;
 use log::info;

 pub use chrono::Duration;
-pub use lance::dataset::optimize::CompactionOptions;
+pub use lance::dataset::optimize::{CompactionMode, CompactionOptions, IndexRemapMode};

 use super::NativeTable;
 use crate::error::Result;
Author	SHA1	Message	Date
Gatefixer	9bfd475702	Merge remote-tracking branch 'origin/main' into gatekeeper/fix-2325-1 # Conflicts: # python/python/lancedb/remote/table.py	2026-08-28 09:45:02 +00:00
Xuanwo	0559108fa9	feat: support blob computed column refresh (#4081 ) Computed-column planning currently sees Blob v2 storage descriptors, so expressions cannot consume payload bytes or preserve Blob semantics in their outputs. A computed declaration now derives its output field from its expression. A direct projection of a Blob v2 field inherits the source field's Blob metadata; other expressions retain their ordinary Arrow-inferred type. Declarations remain ordered, so the same rule applies across chained projections. Refresh materializes referenced Blob inputs as `LargeBinary` payload bytes and publishes inherited Blob outputs through Lance's Blob conversion path. Remote requests remain within the shared namespace contract as `{name, computed}`; the server planner is being updated in tandem to implement the same Blob-aware planning semantics, and remote enablement must be aligned with that server rollout. The existing null-as-unfilled contract remains unchanged. Row-level freshness and cell flags remain follow-up work.	2026-08-28 17:23:49 +08:00
Will Jones	6ab3b9eb30	ci: upgrade chacha20 to 0.10.2 (#4078 ) The pinned version was yanked due to UB in some SIMD kernels. Upgrading.	2026-08-27 19:06:07 -07:00
Lance Release	c94d9a2a16	Bump version: 0.38.0-beta.11 → 0.38.0-beta.12	2026-08-28 00:52:01 +00:00
Wyatt Alt	6c8aa22704	feat: let a computed-column batch read its own earlier declarations (#4072 ) `add_columns().computed()` accepted several columns in one call but bound each against the table's schema as it stood before the call, so `a` and `b = a + 1` had to be two commits. A server staging declarations behind other schema work has no atomic way to do that, and a caller reading the builder's plural signature reasonably expects the batch to be one. Each accepted column now joins the schema the next one resolves against, so the batch is planned and committed as one. Order is the dependency order; reading ahead is still an unknown column. `validate_declarations` exposes the schema-level checks -- the Function-binding guard and the planning -- without a commit, for callers that must reject before earlier work in the same request lands; LSM state is table state and stays a commit-time check. Refresh order matters for a dependent column: `b = coalesce(a, 0)` refreshed before `a` would bake zeros from `a`'s placeholder null, and the fill-once contract keeps them. Refresh now refuses, naming the input, while a computed input still has rows a refresh of it would fill -- the same probe refresh already uses to detect a no-op. Otherwise it is one snapshot and one commit, as before; a concurrent append is not in the commit and waits for the next refresh. Refreshing dependencies on the caller's behalf was considered and rejected: it is not how materialized views or our own backfill scheduler behave, and it needs multi-commit fencing that an explicit per-row fill marker would make unnecessary.	2026-08-27 17:47:27 -07:00
Will JonesandClaude Opus 5	84f46df876	ci(nodejs): fix nightly OOM on the aarch64 publish legs (#4077 ) The nightly `NPM Publish` run has failed every night since at least Aug 23, always on the same two legs: `aarch64-unknown-linux-gnu` and `aarch64-unknown-linux-musl`. The other five targets pass. rustc is OOM-killed during the fat-LTO codegen of the cdylib — `signal: 9` with no diagnostic, about 27 minutes in — and on the musl leg that takes the whole runner down with `The runner has received a shutdown signal`. Both legs now pass: \| leg \| before \| peak memory \| wall time \| \| --- \| --- \| --- \| --- \| \| `aarch64-unknown-linux-gnu` \| OOM-killed at ~27 min \| 31391 → 22851 MiB \| 38m43s → 22m04s \| \| `aarch64-unknown-linux-musl` \| runner killed at ~28 min \| >32 GiB → 16516 MiB \| ~40 min → 20m50s \| ThinLTO is most of that. Fat LTO is single-threaded, and its peak is consumed inside rustc's LLVM before any linker process is spawned — which is why it is the whole fix on musl, and why lld alone left the gnu leg still peaking at 31391 MiB against the runner's 32 GiB. Both legs now use the `lto: thin` / `codegen_units: 16` settings that darwin and both Windows legs already use, at a cost of a few percent runtime performance. lld covers the rest, on the gnu leg. arm64 Linux otherwise links through GNU `ld` where x86_64 already defaults to `rust-lld`, which is why only the arm64 legs hit this at all; on a comparable arm64 build (`lancedb/sophon#7313`) it cut the largest single linker process from 7.0 to 4.0 GiB and wall time by 35%. The flags live in a small wrapper script used as the linker rather than in `-C link-arg`, because the per-target rustflags variable does not reach every unit that links: dependency crates linking a dylib (`crc-fast`, `lance-arrow`) were invoked as bare `clang`, which targets the x86_64 host and fails with `Relocations in generic ELF (EM: 183)`. Separately, and affecting five legs rather than two: the three ThinLTO targets exported `CARGO_PROFILE_RELEASE_LTO` and `CARGO_PROFILE_RELEASE_CODEGEN_UNITS` from `pre_build`, which runs inside the build step — after the cache step. `Swatinem/rust-cache` computes its key when the action runs, before any step, so step-local values are invisible to it. The result is a loop that never converges: the key never changes, so restores are exact hits, an exact hit makes the post-run save a no-op, and cargo invalidates the restored artifacts anyway because the flags differ. Those legs have been rebuilding cold on every run. Both values move to job-level `env:` ahead of the cache step, driven by new `lto:`/`codegen_units:` matrix fields, and are forwarded into the containers with `-e` since `docker run` inherits nothing. Every leg's cache key shifts once as a result, so expect one cold rebuild. A `Report peak memory` step is added so whether these legs fit is a number rather than an inference from whether the runner survived. It produced the figures above. ## Not included Moving these legs to native arm64 runners. It would retire the zig cross path, the `AT_HWCAP2` workaround and the `TARGET_CC` override, and arm64 runners are billed roughly 37% below x64 at equal core count — but the `lts-debian-aarch64` image exists to link against the manylinux2014 sysroot's glibc 2.17, and building natively on ubuntu-24.04 would raise the minimum glibc for every published aarch64 binary. That is a user-facing decision, not a CI cleanup. Dropping these legs to smaller runners, which is where the real cost saving is — larger runners are billed even on public repos. On these numbers it is not available yet: musl at 16516 MiB is about 130 MiB over what a 16 GB standard runner has. Worth revisiting as a follow-up. ## Testing Cargo's rustflags precedence was checked locally rather than taken from the docs, since getting it wrong would silently change the published binaries. With a throwaway crate carrying both a `target.'cfg(all())'` and a per-target rustflags table: setting `RUSTFLAGS` discards both, and setting it to the empty string discards them too. That rules out routing the linker flag through a job-level `RUSTFLAGS`, because `env:` keys cannot be conditionally omitted and every other leg would then silently lose the `target-cpu`/`target-feature` settings in `.cargo/config.toml` — `+avx2` on x86_64 and `-crt-static` on aarch64-musl. --------- Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>	2026-08-27 17:35:06 -07:00
Drew	83cff3ab93	fix(python): use one blobv2 type and coerce blob writes by metadata (#4065 )	2026-08-27 17:07:29 -07:00
Will JonesandClaude Opus 5	b85776c22a	fix(listing)!: page table listings from the store's own cursor (#3979 ) BREAKING CHANGE: list_tables now provides tables in arbitrary order and the page token is now completely opaque. `table_names` retains the old behavior of lexical ordering and `start-after` semantics. Listing the tables in a directory database cost what the database held rather than what the page held. `ListingDatabase::list_tables` enumerated every child directory of the base path, sorted the names, then discarded all but the requested page — on every request, for every page. On object storage that is one full listing per page. This PR pages the store instead. `list_tables` asks for one page at a time through `ObjectStore::read_dir_page`, carrying the store's own continuation token, so a page is one request. Non-table children can leave a page short of its limit, so the walk continues until the page is full or the store runs out. --------- Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>	2026-08-27 15:52:41 -07:00
Xuanwo	670bda8725	Merge branch 'main' into gatekeeper/fix-2325-1	2026-08-27 18:13:07 +08:00
Gatefixer	10151b2dc8	Merge remote-tracking branch 'origin/main' into gatekeeper/fix-2325-1	2026-08-22 07:07:55 +00:00
Gatefixer	53b4c3b715	Merge origin/main into gatekeeper/fix-2325-1 # Conflicts: # python/python/lancedb/__init__.py	2026-08-22 07:07:52 +00:00
Gatefixer	638430fdb4	Merge origin/main into gatekeeper/fix-2325-1	2026-08-21 09:23:38 +00:00
Gatefixer	2221b8df6a	Merge origin/main into gatekeeper/fix-2325-1	2026-08-21 08:56:03 +00:00
Gatefixer	14eeae4bd4	Merge origin/main into gatekeeper/fix-2325-1	2026-08-21 08:47:29 +00:00
Gatefixer	320755ed55	Merge origin/main into gatekeeper/fix-2325-1 # Conflicts: # python/python/lancedb/__init__.py	2026-08-21 08:13:23 +00:00
Gatefixer	e55c2da7b1	fix(python): clarify compaction limits	2026-08-20 11:45:51 +00:00
Gatefixer	d33b05328c	Merge remote-tracking branch 'origin/main' into gatekeeper/fix-2325-1	2026-08-20 11:09:16 +00:00
Gatefixer	82f5355b71	fix(python): expose compaction source limits	2026-08-14 22:41:06 +00:00
Gatefixer	40cff9b644	Merge origin/main into gatekeeper/fix-2325-1	2026-08-14 21:59:28 +00:00
Gatefixer	edf95e53fc	Merge origin/main into gatekeeper/fix-2325-1	2026-08-07 19:46:40 +00:00
Gatefixer	0b5eba085d	Merge origin/main into gatekeeper/fix-2325-1	2026-08-07 19:32:06 +00:00
Gatefixer	21bf859c0b	fix(python): validate compaction option bounds	2026-08-06 06:30:38 +00:00
Gatefixer	e0499de959	fix(python): expose optimize compaction options	2026-08-06 05:01:42 +00:00