fix(python): scope Instructor compatibility shim

<!-- lance-gatekeeper-fix:v1 agent=5c80c44c083b3b8ad0da595419d468fc
generation=1 -->

## Root cause

The legacy synchronous Python table called `delete` on a shared, mutable
`lance.Dataset`. Concurrent table operations could hold a PyO3 borrow
while delete requested an exclusive borrow, producing `RuntimeError:
Already borrowed`. The current async-backed binding fixes this by
cloning its thread-safe Rust table handle before awaiting, but that
concurrency contract had no regression coverage.

## Fix

- Document why delete must clone the Rust table handle before entering
its async future.
- Add a barrier-synchronized regression test that deletes distinct rows
through one shared table from eight Python threads.
- Verify every delete commits exactly one row, every commit gets a
distinct version, and no rows remain.

## Validation

- `cargo check --quiet --features remote --tests --examples`
- `cargo fmt --all -- --check`
- `uv run --extra tests --extra dev ruff format --check
python/tests/test_table.py`
- `uv run --extra tests --extra dev ruff check
python/tests/test_table.py`
- `uv run --extra tests --extra dev pytest
python/tests/test_table.py::test_concurrent_deletes_are_thread_safe
python/tests/test_table.py::test_delete
python/tests/test_table.py::test_delete_expr
python/tests/test_table.py::test_delete_expr_async -q` (4 passed)
- Manual stress reproduction: 100 concurrent deletes on one table
completed at versions 2–101 with zero rows remaining.

Fixes #530

Co-authored-by: Gatefixer <313497061+lancedb-gatefixer[bot]@users.noreply.github.com>

Cargo.lock

@@ -601,7 +601,7 @@ dependencies = [
  "bytes",
  "fastrand",
  "hex",
- "http 1.4.2",
+ "http 1.5.0",
  "sha1 0.10.6",
  "time",
  "tokio",
@@ -664,7 +664,7 @@ dependencies = [
  "bytes-utils",
  "fastrand",
  "http 0.2.12",
- "http 1.4.2",
+ "http 1.5.0",
  "http-body 0.4.6",
  "http-body 1.1.0",
  "percent-encoding",
@@ -694,7 +694,7 @@ dependencies = [
  "bytes",
  "fastrand",
  "http 0.2.12",
- "http 1.4.2",
+ "http 1.5.0",
  "http-body-util",
  "regex-lite",
  "tracing",
@@ -719,7 +719,7 @@ dependencies = [
  "bytes",
  "fastrand",
  "http 0.2.12",
- "http 1.4.2",
+ "http 1.5.0",
  "regex-lite",
  "tracing",
 ]
@@ -743,7 +743,7 @@ dependencies = [
  "bytes",
  "fastrand",
  "http 0.2.12",
- "http 1.4.2",
+ "http 1.5.0",
  "regex-lite",
  "tracing",
 ]
@@ -773,7 +773,7 @@ dependencies = [
  "hex",
  "hmac 0.13.0",
  "http 0.2.12",
- "http 1.4.2",
+ "http 1.5.0",
  "http-body 1.1.0",
  "lru",
  "percent-encoding",
@@ -802,7 +802,7 @@ dependencies = [
  "bytes",
  "fastrand",
  "http 0.2.12",
- "http 1.4.2",
+ "http 1.5.0",
  "regex-lite",
  "tracing",
 ]
@@ -826,7 +826,7 @@ dependencies = [
  "bytes",
  "fastrand",
  "http 0.2.12",
- "http 1.4.2",
+ "http 1.5.0",
  "regex-lite",
  "tracing",
 ]
@@ -851,7 +851,7 @@ dependencies = [
  "aws-types",
  "fastrand",
  "http 0.2.12",
- "http 1.4.2",
+ "http 1.5.0",
  "regex-lite",
  "tracing",
 ]
@@ -873,7 +873,7 @@ dependencies = [
  "hex",
  "hmac 0.13.0",
  "http 0.2.12",
- "http 1.4.2",
+ "http 1.5.0",
  "p256",
  "percent-encoding",
  "ring",
@@ -906,7 +906,7 @@ dependencies = [
  "bytes",
  "crc-fast",
  "hex",
- "http 1.4.2",
+ "http 1.5.0",
  "http-body 1.1.0",
  "http-body-util",
  "md-5 0.11.0",
@@ -940,7 +940,7 @@ dependencies = [
  "bytes-utils",
  "futures-core",
  "futures-util",
- "http 1.4.2",
+ "http 1.5.0",
  "http-body 1.1.0",
  "http-body-util",
  "percent-encoding",
@@ -961,7 +961,7 @@ dependencies = [
  "h2 0.3.27",
  "h2 0.4.14",
  "http 0.2.12",
- "http 1.4.2",
+ "http 1.5.0",
  "http-body 0.4.6",
  "hyper 0.14.32",
  "hyper 1.9.0",
@@ -1023,7 +1023,7 @@ dependencies = [
  "bytes",
  "fastrand",
  "http 0.2.12",
- "http 1.4.2",
+ "http 1.5.0",
  "http-body 0.4.6",
  "http-body 1.1.0",
  "http-body-util",
@@ -1044,7 +1044,7 @@ dependencies = [
  "aws-smithy-types",
  "bytes",
  "http 0.2.12",
- "http 1.4.2",
+ "http 1.5.0",
  "pin-project-lite",
  "tokio",
  "tracing",
@@ -1070,7 +1070,7 @@ checksum = "7442cb268338f0eb8278140a107c046756aa01093d8ef5e99628d34ae09c94f5"
 dependencies = [
  "aws-smithy-runtime-api",
  "aws-smithy-types",
- "http 1.4.2",
+ "http 1.5.0",
 ]
 
 [[package]]
@@ -1084,7 +1084,7 @@ dependencies = [
  "bytes-utils",
  "futures-core",
  "http 0.2.12",
- "http 1.4.2",
+ "http 1.5.0",
  "http-body 0.4.6",
  "http-body 1.1.0",
  "http-body-util",
@@ -1132,7 +1132,7 @@ dependencies = [
  "axum-core",
  "bytes",
  "futures-util",
- "http 1.4.2",
+ "http 1.5.0",
  "http-body 1.1.0",
  "http-body-util",
  "hyper 1.9.0",
@@ -1165,7 +1165,7 @@ dependencies = [
  "async-trait",
  "bytes",
  "futures-util",
- "http 1.4.2",
+ "http 1.5.0",
  "http-body 1.1.0",
  "http-body-util",
  "mime",
@@ -3842,7 +3842,7 @@ dependencies = [
  "fnv",
  "futures-core",
  "futures-sink",
- "http 1.4.2",
+ "http 1.5.0",
  "indexmap 2.14.0",
  "slab",
  "tokio",
@@ -3953,7 +3953,7 @@ checksum = "629d8f3bbeda9d148036d6b0de0a3ab947abd08ce90626327fc3547a49d59d97"
 dependencies = [
  "dirs",
  "futures",
- "http 1.4.2",
+ "http 1.5.0",
  "indicatif",
  "libc",
  "log",
@@ -3976,7 +3976,7 @@ checksum = "430b33fa84f92796d4d263070b6c0d3ca219df7b9a0e1853ee431029b1612bcd"
 dependencies = [
  "async-trait",
  "bytes",
- "http 1.4.2",
+ "http 1.5.0",
  "more-asserts",
  "serde",
  "thiserror 2.0.18",
@@ -4041,9 +4041,9 @@ dependencies = [
 
 [[package]]
 name = "http"
-version = "1.4.2"
+version = "1.5.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "6970f50e31d6fc17d3fa27329444bfa74e196cf62e95052a3f6fee181dba6425"
+checksum = "918d3568bebf352712bc2ef3d46a8bcf1a75b373be6539de198e9105cbbf9ce0"
 dependencies = [
  "bytes",
  "itoa",
@@ -4067,7 +4067,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "ca2a8f2913ee65f60facd6a5905613afaa448497a0230cc41ce022d93290bc2c"
 dependencies = [
  "bytes",
- "http 1.4.2",
+ "http 1.5.0",
 ]
 
 [[package]]
@@ -4078,7 +4078,7 @@ checksum = "b021d93e26becf5dc7e1b75b1bed1fd93124b374ceb73f43d4d4eafec896a64a"
 dependencies = [
  "bytes",
  "futures-core",
- "http 1.4.2",
+ "http 1.5.0",
  "http-body 1.1.0",
  "pin-project-lite",
 ]
@@ -4145,7 +4145,7 @@ dependencies = [
  "futures-channel",
  "futures-core",
  "h2 0.4.14",
- "http 1.4.2",
+ "http 1.5.0",
  "http-body 1.1.0",
  "httparse",
  "httpdate",
@@ -4177,7 +4177,7 @@ version = "0.27.9"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "33ca68d021ef39cf6463ab54c1d0f5daf03377b70561305bb89a8f83aab66e0f"
 dependencies = [
- "http 1.4.2",
+ "http 1.5.0",
  "hyper 1.9.0",
  "hyper-util",
  "rustls 0.23.40",
@@ -4211,7 +4211,7 @@ dependencies = [
  "bytes",
  "futures-channel",
  "futures-util",
- "http 1.4.2",
+ "http 1.5.0",
  "http-body 1.1.0",
  "hyper 1.9.0",
  "ipnet",
@@ -5182,7 +5182,7 @@ dependencies = [
  "chrono",
  "futures",
  "goosefs-sdk",
- "http 1.4.2",
+ "http 1.5.0",
  "io-uring",
  "lance-arrow",
  "lance-core",
@@ -5414,7 +5414,7 @@ dependencies = [
  "goosefs-sdk",
  "half",
  "hf-hub",
- "http 1.4.2",
+ "http 1.5.0",
  "http-body 1.1.0",
  "lance",
  "lance-arrow",
@@ -6080,15 +6080,15 @@ dependencies = [
 
 [[package]]
 name = "napi-build"
-version = "2.3.2"
+version = "2.4.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "c9c366d2c8c60b86fa632df75f745509b52f9128f91a6bad4c796e44abb505e1"
+checksum = "5282704fbe8d49b0cf8b08e3f33233416a528658f205c7e5ace63b582de0b11c"
 
 [[package]]
 name = "napi-derive"
-version = "3.6.0"
+version = "3.6.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "a49c513341a61a16a10af6efcce46b30d0822ba2d4fb197d24d33dfc199c78d5"
+checksum = "4d5c9c02556ea6dc99dffd36c1ce60141411657438501a125b675776d011ce92"
 dependencies = [
  "convert_case",
  "ctor 1.0.5",
@@ -6100,9 +6100,9 @@ dependencies = [
 
 [[package]]
 name = "napi-derive-backend"
-version = "6.0.0"
+version = "6.1.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "4747005fa3e2c9989ac45a723a514c5db2411238b72981a3cda4c701a9dfea17"
+checksum = "d60b5d773ad46c698c8cc2cd9fde0b283d39cbb7f71c04bee633c7bdba4423bd"
 dependencies = [
  "convert_case",
  "proc-macro2",
@@ -6356,7 +6356,7 @@ dependencies = [
  "futures-channel",
  "futures-core",
  "futures-util",
- "http 1.4.2",
+ "http 1.5.0",
  "http-body-util",
  "httparse",
  "humantime",
@@ -6477,7 +6477,7 @@ dependencies = [
  "base64 0.22.1",
  "bytes",
  "futures",
- "http 1.4.2",
+ "http 1.5.0",
  "http-body 1.1.0",
  "jiff",
  "log",
@@ -6502,7 +6502,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "0d6f81ba6960e3fae1882f253b114b21d7e444e1534f209c7737a79f6243eb6f"
 dependencies = [
  "futures",
- "http 1.4.2",
+ "http 1.5.0",
  "mea",
  "opendal-core",
 ]
@@ -6546,7 +6546,7 @@ checksum = "0030644366ef5d8cbe3a4a5822bf99a4aafddc1666e9d24b44d158d9062fc76a"
 dependencies = [
  "base64 0.22.1",
  "bytes",
- "http 1.4.2",
+ "http 1.5.0",
  "log",
  "opendal-core",
  "opendal-service-azure-common",
@@ -6567,7 +6567,7 @@ checksum = "6dea4908d490143a9b0b7f7a790e139ff829b06a023f670455ed3d44f664b361"
 dependencies = [
  "base64 0.22.1",
  "bytes",
- "http 1.4.2",
+ "http 1.5.0",
  "log",
  "opendal-core",
  "opendal-service-azure-common",
@@ -6585,7 +6585,7 @@ version = "0.57.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "9b489f13c42e69d69bdd72952b634356ec43a7881a20259b38b540fcecdf4051"
 dependencies = [
- "http 1.4.2",
+ "http 1.5.0",
  "opendal-core",
 ]
 
@@ -6596,7 +6596,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "aa8cafe9729213375c7331019b0cb756ad3e1aff7f45cd32c45eae91ebde8901"
 dependencies = [
  "bytes",
- "http 1.4.2",
+ "http 1.5.0",
  "log",
  "opendal-core",
  "quick-xml 0.39.4",
@@ -6614,7 +6614,7 @@ checksum = "48de101aac565ed06af4b47903c24eafd249075553ec1fb18256751c45148d47"
 dependencies = [
  "async-trait",
  "bytes",
- "http 1.4.2",
+ "http 1.5.0",
  "log",
  "opendal-core",
  "percent-encoding",
@@ -6649,7 +6649,7 @@ checksum = "c4922661976a1d40794a2adfbdb888cc3c23097690f825a92f773af38908a848"
 dependencies = [
  "bytes",
  "hf-xet",
- "http 1.4.2",
+ "http 1.5.0",
  "log",
  "opendal-core",
  "percent-encoding",
@@ -6665,7 +6665,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "328fa55e8888cbdfe00826bfea2a79042422b720e8369e9e021e46121dea5ace"
 dependencies = [
  "bytes",
- "http 1.4.2",
+ "http 1.5.0",
  "log",
  "opendal-core",
  "quick-xml 0.39.4",
@@ -6684,7 +6684,7 @@ dependencies = [
  "base64 0.22.1",
  "bytes",
  "crc32c",
- "http 1.4.2",
+ "http 1.5.0",
  "log",
  "md-5 0.11.0",
  "opendal-core",
@@ -7583,7 +7583,7 @@ version = "0.14.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "343d3bd7056eda839b03204e68deff7d1b13aba7af2b2fd16890697274262ee7"
 dependencies = [
- "heck 0.5.0",
+ "heck 0.4.1",
  "itertools 0.14.0",
  "log",
  "multimap",
@@ -8226,7 +8226,7 @@ checksum = "57ac2757f3140aa2e213b554148ae0b52733e624fc6723f0cc6bb3d440176c95"
 dependencies = [
  "anyhow",
  "form_urlencoded",
- "http 1.4.2",
+ "http 1.5.0",
  "log",
  "percent-encoding",
  "reqsign-core",
@@ -8244,7 +8244,7 @@ dependencies = [
  "anyhow",
  "bytes",
  "form_urlencoded",
- "http 1.4.2",
+ "http 1.5.0",
  "log",
  "percent-encoding",
  "quick-xml 0.39.4",
@@ -8266,7 +8266,7 @@ dependencies = [
  "base64 0.22.1",
  "bytes",
  "form_urlencoded",
- "http 1.4.2",
+ "http 1.5.0",
  "jsonwebtoken",
  "log",
  "pem",
@@ -8291,7 +8291,7 @@ dependencies = [
  "futures",
  "hex",
  "hmac 0.12.1",
- "http 1.4.2",
+ "http 1.5.0",
  "jiff",
  "log",
  "percent-encoding",
@@ -8318,7 +8318,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "35cc609b49c69e76ecaceb775a03f792d1ed3e7755ab3548d4534fd801e3242e"
 dependencies = [
  "form_urlencoded",
- "http 1.4.2",
+ "http 1.5.0",
  "jsonwebtoken",
  "log",
  "percent-encoding",
@@ -8338,7 +8338,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "e128f19525861dbded59e1e7c17653a8ed63d573ca04aed708d552dbef5bb32a"
 dependencies = [
  "anyhow",
- "http 1.4.2",
+ "http 1.5.0",
  "log",
  "percent-encoding",
  "reqsign-core",
@@ -8358,7 +8358,7 @@ dependencies = [
  "futures-core",
  "futures-util",
  "h2 0.4.14",
- "http 1.4.2",
+ "http 1.5.0",
  "http-body 1.1.0",
  "http-body-util",
  "hyper 1.9.0",
@@ -8402,7 +8402,7 @@ dependencies = [
  "bytes",
  "futures-core",
  "futures-util",
- "http 1.4.2",
+ "http 1.5.0",
  "http-body 1.1.0",
  "http-body-util",
  "hyper 1.9.0",
@@ -8456,7 +8456,7 @@ checksum = "199dda04a536b532d0cc04d7979e39b1c763ea749bf91507017069c00b96056f"
 dependencies = [
  "anyhow",
  "async-trait",
- "http 1.4.2",
+ "http 1.5.0",
  "reqwest 0.13.3",
  "thiserror 2.0.18",
  "tower-service",
@@ -8498,9 +8498,9 @@ dependencies = [
 
 [[package]]
 name = "rkyv"
-version = "0.8.16"
+version = "0.8.17"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "73389e0c99e664f919275ab5b5b0471391fe9a8de61e1dff9b1eaf56a90f16e3"
+checksum = "815cc8a37159a463064825246cadb07961e25cd9885908606f6d08a98d8f8874"
 dependencies = [
  "bytecheck",
  "bytes",
@@ -8517,9 +8517,9 @@ dependencies = [
 
 [[package]]
 name = "rkyv_derive"
-version = "0.8.16"
+version = "0.8.17"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "5d2ed0b54125315fb36bd021e82d314d1c126548f871634b483f46b31d13cac6"
+checksum = "c0ed1a78a1b19d184b0daa629dd9a024573173ec7d485b287cb369fb3607cc1c"
 dependencies = [
  "proc-macro2",
  "quote",
@@ -9277,7 +9277,7 @@ version = "0.8.9"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "c1c97747dbf44bb1ca44a561ece23508e99cb592e862f22222dcf42f51d1e451"
 dependencies = [
- "heck 0.5.0",
+ "heck 0.4.1",
  "proc-macro2",
  "quote",
  "syn 2.0.117",
@@ -9289,7 +9289,7 @@ version = "0.9.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "54254b8531cafa275c5e096f62d48c81435d1015405a91198ddb11e967301d40"
 dependencies = [
- "heck 0.5.0",
+ "heck 0.4.1",
  "proc-macro2",
  "quote",
  "syn 2.0.117",
@@ -9730,7 +9730,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "32497e9a4c7b38532efcdebeef879707aa9f794296a4f0244f6f69e9bc8574bd"
 dependencies = [
  "fastrand",
- "getrandom 0.4.2",
+ "getrandom 0.3.4",
  "once_cell",
  "rustix",
  "windows-sys 0.61.2",
@@ -10058,7 +10058,7 @@ dependencies = [
  "base64 0.22.1",
  "bytes",
  "h2 0.4.14",
- "http 1.4.2",
+ "http 1.5.0",
  "http-body 1.1.0",
  "http-body-util",
  "hyper 1.9.0",
@@ -10114,7 +10114,7 @@ checksum = "1e9cd434a998747dd2c4276bc96ee2e0c7a2eadf3cae88e52be55a05fa9053f5"
 dependencies = [
  "bitflags 2.11.1",
  "bytes",
- "http 1.4.2",
+ "http 1.5.0",
  "http-body 1.1.0",
  "http-body-util",
  "pin-project-lite",
@@ -10134,7 +10134,7 @@ dependencies = [
  "bytes",
  "futures-core",
  "futures-util",
- "http 1.4.2",
+ "http 1.5.0",
  "http-body 1.1.0",
  "http-body-util",
  "pin-project-lite",
@@ -11155,7 +11155,7 @@ dependencies = [
  "clap",
  "crc32fast",
  "futures",
- "http 1.4.2",
+ "http 1.5.0",
  "hyper 1.9.0",
  "lazy_static",
  "more-asserts",
@@ -11229,7 +11229,7 @@ dependencies = [
  "chrono",
  "clap",
  "gearhash",
- "http 1.4.2",
+ "http 1.5.0",
  "itertools 0.14.0",
  "lazy_static",
  "more-asserts",

nodejs/src/table.rs

@@ -339,7 +339,9 @@ impl Table {
         let transforms = NewColumnTransform::SqlExpressions(transforms);
         let res = self
             .inner_ref()?
-            .add_columns(transforms, None)
+            .add_columns()
+            .transform(transforms)
+            .execute()
             .await
             .default_error()?;
         Ok(res.into())
@@ -356,7 +358,9 @@ impl Table {
         let transforms = NewColumnTransform::AllNulls(schema);
         let res = self
             .inner_ref()?
-            .add_columns(transforms, None)
+            .add_columns()
+            .transform(transforms)
+            .execute()
             .await
             .default_error()?;
         Ok(res.into())

python/python/lancedb/db.py

@@ -707,6 +707,9 @@ class LanceDBConnection(DBConnection):
         self._namespace_client_properties = namespace_client_properties
         if _inner is not None:
             self._conn = _inner
+            # Native-derived wrappers resolve this in their async reconstruction
+            # path so construction never synchronously re-enters LOOP.
+            self._read_consistency_interval = read_consistency_interval
             self._cached_namespace_client = None
             return
 
@@ -756,11 +759,14 @@ class LanceDBConnection(DBConnection):
         # storage_options.  Also, this class really shouldn't be holding any state
         # beyond _conn.
         self._conn = AsyncConnection(LOOP.run(do_connect()))
+        # Keep property access synchronous so debugger introspection cannot wait on
+        # the background loop while that thread is suspended at a breakpoint.
+        self._read_consistency_interval = read_consistency_interval
         self._cached_namespace_client: Optional[LanceNamespace] = None
 
     @property
     def read_consistency_interval(self) -> Optional[timedelta]:
-        return LOOP.run(self._conn.get_read_consistency_interval())
+        return self._read_consistency_interval
 
     @property
     def session(self) -> Optional[Session]:
@@ -771,8 +777,16 @@ class LanceDBConnection(DBConnection):
         return self._conn.uri
 
     @classmethod
-    def from_inner(cls, inner: LanceDbConnection):
-        return cls(None, _inner=inner)
+    def from_inner(
+        cls,
+        inner: LanceDbConnection,
+        read_consistency_interval: Optional[timedelta],
+    ):
+        return cls(
+            None,
+            read_consistency_interval=read_consistency_interval,
+            _inner=inner,
+        )
 
     def __repr__(self) -> str:
         return f"{self.__class__.__name__}(uri={self._conn.uri!r})"

python/python/lancedb/embeddings/instructor.py

@@ -3,6 +3,7 @@
 
 
 from typing import List
+from urllib.parse import unquote, urlparse
 
 import numpy as np
 
@@ -125,9 +126,20 @@ class InstructorEmbeddingFunction(TextEmbeddingFunction):
 
     @weak_lru(maxsize=1)
     def get_model(self):
-        instructor_embedding = attempt_import_or_raise(
-            "InstructorEmbedding", "InstructorEmbedding"
-        )
+        huggingface_hub = attempt_import_or_raise("huggingface_hub", "huggingface-hub")
+        missing = object()
+        original_cached_download = getattr(huggingface_hub, "cached_download", missing)
+        if original_cached_download is missing:
+            huggingface_hub.cached_download = _cached_download(huggingface_hub)
+
+        try:
+            instructor_embedding = attempt_import_or_raise(
+                "InstructorEmbedding", "InstructorEmbedding"
+            )
+        finally:
+            if original_cached_download is missing:
+                del huggingface_hub.cached_download
+
         torch = attempt_import_or_raise("torch", "torch")
 
         model = instructor_embedding.INSTRUCTOR(self.name)
@@ -140,3 +152,44 @@ class InstructorEmbeddingFunction(TextEmbeddingFunction):
                 model, {torch.nn.Linear}, dtype=torch.qint8
             )
         return model
+
+
+def _cached_download(huggingface_hub):
+    """Provide the legacy download API used by sentence-transformers 2.2.x."""
+
+    def cached_download(
+        *,
+        url,
+        cache_dir=None,
+        force_filename=None,
+        library_name=None,
+        library_version=None,
+        user_agent=None,
+        use_auth_token=None,
+        **_,
+    ):
+        path = urlparse(url).path.lstrip("/")
+        try:
+            repo_id, resolved_path = path.split("/resolve/", maxsplit=1)
+            revision, filename = resolved_path.split("/", maxsplit=1)
+        except ValueError as err:
+            raise ValueError(f"Unsupported Hugging Face Hub URL: {url}") from err
+
+        repo_id = unquote(repo_id)
+        revision = unquote(revision)
+        filename = unquote(filename)
+        # sentence-transformers derives force_filename from this Hub path with
+        # os.path.join. Using the URL path beneath local_dir produces the same
+        # local destination without sending Windows separators to the Hub.
+        return huggingface_hub.hf_hub_download(
+            repo_id=repo_id,
+            filename=filename,
+            revision=revision,
+            local_dir=cache_dir,
+            library_name=library_name,
+            library_version=library_version,
+            user_agent=user_agent,
+            token=use_auth_token,
+        )
+
+    return cached_download

python/python/lancedb/merge.py

@@ -92,8 +92,10 @@ class LanceMergeInsertBuilder(object):
         self._when_not_matched_by_source_delete = True
         if isinstance(condition, Expr):
             self._when_not_matched_by_source_condition_expr = condition._inner
-        elif condition is not None:
+            self._when_not_matched_by_source_condition = None
+        else:
             self._when_not_matched_by_source_condition = condition
+            self._when_not_matched_by_source_condition_expr = None
         return self
 
     def use_index(self, use_index: bool) -> LanceMergeInsertBuilder:

python/python/lancedb/permutation.py

@@ -226,7 +226,7 @@ class PermutationBuilder:
 
         async def do_execute():
             inner_tbl = await self._async.execute()
-            return LanceTable.from_inner(inner_tbl)
+            return await LanceTable.from_inner(inner_tbl)
 
         return LOOP.run(do_execute())
 

python/python/lancedb/query.py

@@ -2697,7 +2697,7 @@ class LanceHybridQueryBuilder(LanceQueryBuilder):
             self._fts_query.phrase_query(True)
         if self._distance_type:
             self._vector_query.metric(self._distance_type)
-        if self._minimum_nprobes:
+        if self._minimum_nprobes is not None:
             self._vector_query.minimum_nprobes(self._minimum_nprobes)
         if self._maximum_nprobes is not None:
             self._vector_query.maximum_nprobes(self._maximum_nprobes)

python/python/lancedb/table.py

@@ -1595,8 +1595,8 @@ class Table(ABC):
         Prefer this over :meth:`fetch_blobs` for large payloads. ``row_ids`` is
         a ``list[int]`` or a query ``pyarrow.Table`` carrying row identity via
         ``_rowid`` or a ``_lance_row_id`` field on the blob descriptor. Null
-        rows are ``None``. Unsupported on LanceDB Cloud, where
-        :meth:`fetch_blobs` returns full bytes instead.
+        rows are ``None``. Remote tables require LanceDB Cloud server 0.5.0 or
+        newer.
         """
 
     @abstractmethod
@@ -2182,11 +2182,15 @@ class LanceTable(Table):
         return self.name
 
     @classmethod
-    def from_inner(cls, tbl: LanceDBTable):
-        from .db import LanceDBConnection
+    async def from_inner(cls, tbl: LanceDBTable):
+        from .db import AsyncConnection, LanceDBConnection
 
         async_tbl = AsyncTable(tbl)
-        conn = LanceDBConnection.from_inner(tbl.database())
+        inner_conn = tbl.database()
+        read_consistency_interval = await AsyncConnection(
+            inner_conn
+        ).get_read_consistency_interval()
+        conn = LanceDBConnection.from_inner(inner_conn, read_consistency_interval)
         return cls(
             conn,
             async_tbl.name,

python/python/tests/test_db.py

@@ -77,6 +77,23 @@ def test_sync_repr_does_not_use_background_loop(tmp_path, monkeypatch):
     assert repr(table) == f"LanceTable(name='test', _conn={db!r})"
 
 
+def test_read_consistency_interval_does_not_use_background_loop(tmp_path, monkeypatch):
+    from lancedb.background_loop import LOOP
+    from lancedb.db import LanceDBConnection
+
+    consistency_interval = timedelta(seconds=5)
+    db = lancedb.connect(tmp_path, read_consistency_interval=consistency_interval)
+    db_from_inner = LanceDBConnection.from_inner(db._inner, consistency_interval)
+
+    def fail_run(*args, **kwargs):
+        raise AssertionError("properties should not use the Python background loop")
+
+    monkeypatch.setattr(LOOP, "run", fail_run)
+
+    assert db.read_consistency_interval == consistency_interval
+    assert db_from_inner.read_consistency_interval == consistency_interval
+
+
 def test_ingest_pd(tmp_path):
     db = lancedb.connect(tmp_path)
 

python/python/tests/test_embeddings.py

@@ -1,8 +1,11 @@
 # SPDX-License-Identifier: Apache-2.0
 # SPDX-FileCopyrightText: Copyright The LanceDB Authors
 
+import ntpath
 import os
 import pickle
+import sys
+from types import ModuleType
 from typing import List, Optional, Union
 from unittest.mock import MagicMock, patch
 
@@ -522,6 +525,59 @@ def test_embedding_function_safe_model_dump(embedding_type):
             )
 
 
+def test_instructor_embedding_supports_huggingface_hub_without_cached_download(
+    tmp_path, monkeypatch
+):
+    from lancedb.embeddings.instructor import InstructorEmbeddingFunction
+
+    hub_download = MagicMock(return_value="/cache/1_Pooling/config.json")
+    huggingface_hub = ModuleType("huggingface_hub")
+    huggingface_hub.hf_hub_download = hub_download
+    torch = ModuleType("torch")
+    monkeypatch.setitem(sys.modules, "huggingface_hub", huggingface_hub)
+    monkeypatch.setitem(sys.modules, "torch", torch)
+    monkeypatch.delitem(sys.modules, "InstructorEmbedding", raising=False)
+    monkeypatch.syspath_prepend(str(tmp_path))
+
+    (tmp_path / "InstructorEmbedding.py").write_text(
+        "from huggingface_hub import cached_download\n\n"
+        "class INSTRUCTOR:\n"
+        "    def __init__(self, name):\n"
+        "        self.name = name\n"
+    )
+
+    embedding = InstructorEmbeddingFunction.create(show_progress_bar=False)
+    instructor_model = embedding.get_model()
+
+    assert instructor_model.name == "hkunlp/instructor-base"
+    assert not hasattr(huggingface_hub, "cached_download")
+
+    instructor_embedding = sys.modules["InstructorEmbedding"]
+    path = instructor_embedding.cached_download(
+        url=(
+            "https://huggingface.co/hkunlp/instructor-base/resolve/abc123/"
+            "1_Pooling/config.json"
+        ),
+        cache_dir="/cache",
+        force_filename=ntpath.join("1_Pooling", "config.json"),
+        library_name="sentence-transformers",
+        library_version="2.2.2",
+        use_auth_token="token",
+    )
+
+    assert path == "/cache/1_Pooling/config.json"
+    hub_download.assert_called_once_with(
+        repo_id="hkunlp/instructor-base",
+        filename="1_Pooling/config.json",
+        revision="abc123",
+        local_dir="/cache",
+        library_name="sentence-transformers",
+        library_version="2.2.2",
+        user_agent=None,
+        token="token",
+    )
+
+
 @patch("time.sleep")
 def test_retry(mock_sleep):
     test_function = MagicMock(side_effect=[Exception] * 9 + ["result"])

python/python/tests/test_hybrid_query.py

@@ -123,6 +123,19 @@ async def test_async_hybrid_query_default_limit(table: AsyncTable):
     assert texts.count("a") == 1
 
 
+def test_hybrid_query_minimum_nprobes_zero_raises(sync_table: Table):
+    # minimum_nprobes(0) must raise the same validation error a plain vector
+    # query raises, not silently no-op because 0 is falsy.
+    with pytest.raises(ValueError, match="minimum_nprobes must be greater than 0"):
+        (
+            sync_table.search(query_type="hybrid")
+            .vector([0.0, 0.4])
+            .text("dog")
+            .minimum_nprobes(0)
+            .to_arrow()
+        )
+
+
 def test_hybrid_query_distance_range(sync_table: Table):
     reranker = RRFReranker(return_score="all")
     result = (

python/python/tests/test_permutation.py

@@ -6,6 +6,7 @@ import math
 import pytest
 
 from lancedb import DBConnection, Table, connect
+from lancedb.background_loop import LOOP
 from lancedb.permutation import Permutation, Permutations, permutation_builder
 
 
@@ -31,6 +32,25 @@ def test_split_random_ratios(mem_db):
     assert 65 <= split_1_count <= 75  # ~70% ± tolerance
 
 
+def test_execute_does_not_reenter_background_loop(tmp_path, monkeypatch):
+    import threading
+
+    db = connect(tmp_path)
+    tbl = db.create_table("test_table", pa.table({"x": range(10)}))
+    original_run = LOOP.run
+
+    def fail_on_reentry(future):
+        assert threading.current_thread() is not LOOP.thread
+        return original_run(future)
+
+    monkeypatch.setattr(LOOP, "run", fail_on_reentry)
+
+    permutation_tbl = permutation_builder(tbl).execute()
+
+    assert permutation_tbl.count_rows() == 10
+    assert permutation_tbl._conn.read_consistency_interval is None
+
+
 def test_split_random_counts(mem_db):
     """Test random splitting with absolute counts."""
     tbl = mem_db.create_table(

python/python/tests/test_remote_db.py

@@ -2055,6 +2055,24 @@ def blob_remote_table(*, server_version=Version("0.5.0")):
             request.send_header("phalanx-version", str(server_version))
             request.end_headers()
             request.wfile.write(json.dumps(BLOB_DESCRIBE_RESPONSE).encode())
+        elif request.path.startswith("/v1/table/test/blob/image/"):
+            path = request.path.partition("?")[0]
+            row_id = int(path.split("/")[-2])
+            payload = {10: b"alpha", 20: None, 30: b"gamma"}[row_id]
+            if payload is None:
+                request.send_response(204)
+                request.end_headers()
+                return
+            byte_range = request.headers["Range"].removeprefix("bytes=")
+            start_text, end_text = byte_range.split("-", maxsplit=1)
+            start = int(start_text)
+            end = int(end_text) if end_text else len(payload) - 1
+            chunk = payload[start : end + 1]
+            request.send_response(206)
+            request.send_header("Content-Range", f"bytes {start}-{end}/{len(payload)}")
+            request.send_header("Content-Length", str(len(chunk)))
+            request.end_headers()
+            request.wfile.write(chunk)
         elif request.path == "/v1/table/test/query/":
             content_len = int(request.headers.get("Content-Length", 0))
             body = json.loads(request.rfile.read(content_len))
@@ -2092,8 +2110,21 @@ def test_remote_blob_columns_and_fetch():
         assert table.blob_columns() == ["image"]
         blobs = table.fetch_blobs("image", [10, 20, 30])
         assert blobs.to_pylist() == [b"alpha", None, b"gamma"]
-        with pytest.raises(NotImplementedError, match="Use fetch_blobs for full bytes"):
-            table.fetch_blob_files("image", [10, 20, 30])
+
+
+def test_remote_blob_files_are_lazy_seekable_handles():
+    with blob_remote_table() as table:
+        files = table.fetch_blob_files("image", [10, 20, 30])
+
+        assert len(files) == 3
+        alpha, null_row, gamma = files
+        assert null_row is None
+        assert alpha is not None
+        assert gamma is not None
+        assert alpha.size() == 5
+        assert alpha.read_range(1, 3) == b"lph"
+        gamma.seek(2)
+        assert gamma.read() == b"mma"
 
 
 def test_remote_blob_fetch_accepts_query_table():

python/python/tests/test_table.py

@@ -6,6 +6,7 @@ import os
 import sys
 import threading
 import warnings
+from concurrent.futures import ThreadPoolExecutor
 from datetime import date, datetime, timedelta
 from time import sleep
 from typing import List
@@ -2124,6 +2125,27 @@ def test_delete(mem_db: DBConnection):
     assert table.to_arrow()["id"].to_pylist() == [1]
 
 
+def test_concurrent_deletes_are_thread_safe(mem_db: DBConnection):
+    num_workers = 8
+    table = mem_db.create_table(
+        "my_table", data=[{"id": row_id} for row_id in range(num_workers)]
+    )
+    barrier = threading.Barrier(num_workers)
+
+    def delete(row_id: int):
+        barrier.wait()
+        return table.delete(f"id = {row_id}")
+
+    with ThreadPoolExecutor(max_workers=num_workers) as pool:
+        results = list(pool.map(delete, range(num_workers)))
+
+    assert all(result.num_deleted_rows == 1 for result in results)
+    assert sorted(result.version for result in results) == list(
+        range(2, num_workers + 2)
+    )
+    assert table.count_rows() == 0
+
+
 def test_delete_expr(mem_db: DBConnection):
     table = mem_db.create_table(
         "my_table",
@@ -2364,6 +2386,29 @@ def test_merge_insert_by_source_delete_expr(mem_db: DBConnection):
     assert table.to_arrow().sort_by("a") == expected
 
 
+def test_merge_insert_by_source_delete_reconfigure(mem_db: DBConnection):
+    # Calling when_not_matched_by_source_delete() again with no condition must
+    # widen the delete to unconditional, not keep the earlier condition around.
+    table = mem_db.create_table(
+        "my_table",
+        data=pa.table({"a": [1, 2, 3], "b": ["a", "b", "c"]}),
+    )
+    new_data = pa.table({"a": [2, 4], "b": ["x", "z"]})
+
+    merge_insert_res = (
+        table.merge_insert("a")
+        .when_matched_update_all()
+        .when_not_matched_insert_all()
+        .when_not_matched_by_source_delete("a > 2")
+        .when_not_matched_by_source_delete()
+        .execute(new_data)
+    )
+    assert merge_insert_res.num_deleted_rows == 2
+
+    expected = pa.table({"a": [2, 4], "b": ["x", "z"]})
+    assert table.to_arrow().sort_by("a") == expected
+
+
 @pytest.mark.asyncio
 async def test_merge_insert_by_source_delete_expr_async(
     mem_db_async: AsyncConnection,

python/src/table.rs

@@ -426,9 +426,11 @@ pub struct PyBlobFile {
 impl PyBlobFile {
     fn read_bytes(self_: PyRef<'_, Self>) -> PyResult<Py<PyBytes>> {
         let inner = self_.inner.clone();
-        let bytes = block_on(async move { inner.read().await })
+        let py = self_.py();
+        let bytes = py
+            .detach(move || block_on(async move { inner.read().await }))
             .map_err(|e| PyRuntimeError::new_err(format!("blob read failed: {e}")))?;
-        Ok(PyBytes::new(self_.py(), bytes.as_ref()).unbind())
+        Ok(PyBytes::new(py, bytes.as_ref()).unbind())
     }
 
     pub fn read(self_: PyRef<'_, Self>) -> PyResult<Bound<'_, PyAny>> {
@@ -444,24 +446,32 @@ impl PyBlobFile {
 
     fn close(self_: PyRef<'_, Self>) -> PyResult<()> {
         let inner = self_.inner.clone();
-        block_on(async move { inner.close().await })
+        self_
+            .py()
+            .detach(move || block_on(async move { inner.close().await }))
             .map_err(|e| PyRuntimeError::new_err(format!("blob close failed: {e}")))
     }
 
     fn is_closed(self_: PyRef<'_, Self>) -> bool {
         let inner = self_.inner.clone();
-        block_on(async move { inner.is_closed().await })
+        self_
+            .py()
+            .detach(move || block_on(async move { inner.is_closed().await }))
     }
 
     fn seek(self_: PyRef<'_, Self>, position: u64) -> PyResult<()> {
         let inner = self_.inner.clone();
-        block_on(async move { inner.seek(position).await })
+        self_
+            .py()
+            .detach(move || block_on(async move { inner.seek(position).await }))
             .map_err(|e| PyRuntimeError::new_err(format!("blob seek failed: {e}")))
     }
 
     fn tell(self_: PyRef<'_, Self>) -> PyResult<u64> {
         let inner = self_.inner.clone();
-        block_on(async move { inner.tell().await })
+        self_
+            .py()
+            .detach(move || block_on(async move { inner.tell().await }))
             .map_err(|e| PyRuntimeError::new_err(format!("blob tell failed: {e}")))
     }
 
@@ -475,16 +485,20 @@ impl PyBlobFile {
             .checked_add(length as u64)
             .ok_or_else(|| PyValueError::new_err("offset + length overflowed"))?;
         let inner = self_.inner.clone();
-        let bytes = block_on(async move { inner.read_range(offset..end).await })
+        let py = self_.py();
+        let bytes = py
+            .detach(move || block_on(async move { inner.read_range(offset..end).await }))
             .map_err(|e| PyRuntimeError::new_err(format!("blob read_range failed: {e}")))?;
-        Ok(PyBytes::new(self_.py(), bytes.as_ref()).unbind())
+        Ok(PyBytes::new(py, bytes.as_ref()).unbind())
     }
 
     fn read_up_to(self_: PyRef<'_, Self>, length: usize) -> PyResult<Py<PyBytes>> {
         let inner = self_.inner.clone();
-        let bytes = block_on(async move { inner.read_up_to(length).await })
-            .map_err(|e| PyRuntimeError::new_err(format!("blob read failed: {e}")))?;
-        Ok(PyBytes::new(self_.py(), bytes.as_ref()).unbind())
+        let py = self_.py();
+        let bytes = py
+            .detach(move || block_on(async move { inner.read_up_to(length).await }))
+            .map_err(|e| PyRuntimeError::new_err(format!("blob read_up_to failed: {e}")))?;
+        Ok(PyBytes::new(py, bytes.as_ref()).unbind())
     }
 }
 
@@ -731,6 +745,9 @@ impl Table {
 
     #[allow(private_interfaces)]
     pub fn delete(self_: PyRef<'_, Self>, condition: PredicateArg) -> PyResult<Bound<'_, PyAny>> {
+        // Do not hold the Python borrow across the await.  The cloned Rust table
+        // handle is thread-safe and allows deletes on the same Python table to
+        // run concurrently without PyO3 reporting "Already borrowed".
         let inner = self_.inner_ref()?.clone();
         future_into_py(self_.py(), async move {
             let result = match &condition {
@@ -1361,7 +1378,12 @@ impl Table {
 
         let inner = self_.inner_ref()?.clone();
         future_into_py(self_.py(), async move {
-            let result = inner.add_columns(definitions, None).await.infer_error()?;
+            let result = inner
+                .add_columns()
+                .transform(definitions)
+                .execute()
+                .await
+                .infer_error()?;
             Ok(AddColumnsResult::from(result))
         })
     }
@@ -1375,7 +1397,12 @@ impl Table {
 
         let inner = self_.inner_ref()?.clone();
         future_into_py(self_.py(), async move {
-            let result = inner.add_columns(transform, None).await.infer_error()?;
+            let result = inner
+                .add_columns()
+                .transform(transform)
+                .execute()
+                .await
+                .infer_error()?;
             Ok(AddColumnsResult::from(result))
         })
     }

rust/lancedb/src/blob.rs

@@ -9,6 +9,7 @@
 //!
 //! Blob tables require Lance file format >= 2.2 and stable row ids at create.
 
+use std::ops::Range;
 use std::sync::Arc;
 
 use arrow_array::LargeBinaryArray;
@@ -17,10 +18,202 @@ use arrow_schema::{DataType, Field, Schema};
 use lance::dataset::{BlobRangeRequest as LanceBlobRangeRequest, Dataset, WriteParams};
 use lance_arrow::FieldExt;
 use lance_encoding::version::LanceFileVersion;
+use lance_io::object_store::ObjectStore;
+use object_store::path::Path;
 
 use crate::error::{Error, Result};
 
-pub use lance::dataset::BlobFile;
+/// Seekable handle for one blob value, backed by local storage or a remote
+/// HTTP byte-range endpoint.
+#[derive(Debug)]
+pub struct BlobFile {
+    inner: BlobFileInner,
+}
+
+#[derive(Debug)]
+enum BlobFileInner {
+    Native(lance::dataset::BlobFile),
+    #[cfg(feature = "remote")]
+    Remote(Box<crate::remote::table::blobs::RemoteBlobFile>),
+}
+
+impl From<lance::dataset::BlobFile> for BlobFile {
+    fn from(value: lance::dataset::BlobFile) -> Self {
+        Self {
+            inner: BlobFileInner::Native(value),
+        }
+    }
+}
+
+#[cfg(feature = "remote")]
+impl From<crate::remote::table::blobs::RemoteBlobFile> for BlobFile {
+    fn from(value: crate::remote::table::blobs::RemoteBlobFile) -> Self {
+        Self {
+            inner: BlobFileInner::Remote(Box::new(value)),
+        }
+    }
+}
+
+impl BlobFile {
+    /// Inline reader over a data-file slice.
+    pub fn new_inline(
+        object_store: Arc<ObjectStore>,
+        path: Path,
+        position: u64,
+        size: u64,
+    ) -> Self {
+        lance::dataset::BlobFile::new_inline(object_store, path, position, size).into()
+    }
+
+    /// Dedicated sidecar-file reader.
+    pub fn new_dedicated(object_store: Arc<ObjectStore>, path: Path, size: u64) -> Self {
+        lance::dataset::BlobFile::new_dedicated(object_store, path, size).into()
+    }
+
+    /// Packed reader for a slice in a shared sidecar.
+    pub fn new_packed(
+        object_store: Arc<ObjectStore>,
+        path: Path,
+        position: u64,
+        size: u64,
+    ) -> Self {
+        lance::dataset::BlobFile::new_packed(object_store, path, position, size).into()
+    }
+
+    /// External reader at a resolved object location.
+    pub fn new_external(
+        object_store: Arc<ObjectStore>,
+        path: Path,
+        uri: String,
+        position: u64,
+        size: u64,
+    ) -> Self {
+        lance::dataset::BlobFile::new_external(object_store, path, uri, position, size).into()
+    }
+
+    /// Close the handle.
+    pub async fn close(&self) -> lance_core::Result<()> {
+        match &self.inner {
+            BlobFileInner::Native(file) => file.close().await,
+            #[cfg(feature = "remote")]
+            BlobFileInner::Remote(file) => file.close().await,
+        }
+    }
+
+    /// Whether the handle is closed.
+    pub async fn is_closed(&self) -> bool {
+        match &self.inner {
+            BlobFileInner::Native(file) => file.is_closed().await,
+            #[cfg(feature = "remote")]
+            BlobFileInner::Remote(file) => file.is_closed(),
+        }
+    }
+
+    /// Read a range without moving the cursor.
+    pub async fn read_range(&self, range: Range<u64>) -> lance_core::Result<bytes::Bytes> {
+        match &self.inner {
+            BlobFileInner::Native(file) => file.read_range(range).await,
+            #[cfg(feature = "remote")]
+            BlobFileInner::Remote(file) => file.read_range(range).await,
+        }
+    }
+
+    /// Read ranges without moving the cursor.
+    pub async fn read_ranges(
+        &self,
+        ranges: &[Range<u64>],
+    ) -> lance_core::Result<Vec<bytes::Bytes>> {
+        match &self.inner {
+            BlobFileInner::Native(file) => file.read_ranges(ranges).await,
+            #[cfg(feature = "remote")]
+            BlobFileInner::Remote(file) => file.read_ranges(ranges).await,
+        }
+    }
+
+    /// Read from the cursor to the end.
+    pub async fn read(&self) -> lance_core::Result<bytes::Bytes> {
+        match &self.inner {
+            BlobFileInner::Native(file) => file.read().await,
+            #[cfg(feature = "remote")]
+            BlobFileInner::Remote(file) => file.read().await,
+        }
+    }
+
+    /// Read up to `len` bytes and advance the cursor.
+    pub async fn read_up_to(&self, len: usize) -> lance_core::Result<bytes::Bytes> {
+        match &self.inner {
+            BlobFileInner::Native(file) => file.read_up_to(len).await,
+            #[cfg(feature = "remote")]
+            BlobFileInner::Remote(file) => file.read_up_to(len).await,
+        }
+    }
+
+    /// Move the cursor to `new_cursor`.
+    pub async fn seek(&self, new_cursor: u64) -> lance_core::Result<()> {
+        match &self.inner {
+            BlobFileInner::Native(file) => file.seek(new_cursor).await,
+            #[cfg(feature = "remote")]
+            BlobFileInner::Remote(file) => file.seek(new_cursor).await,
+        }
+    }
+
+    /// Current cursor position.
+    pub async fn tell(&self) -> lance_core::Result<u64> {
+        match &self.inner {
+            BlobFileInner::Native(file) => file.tell().await,
+            #[cfg(feature = "remote")]
+            BlobFileInner::Remote(file) => file.tell().await,
+        }
+    }
+
+    /// Blob length in bytes.
+    pub fn size(&self) -> u64 {
+        match &self.inner {
+            BlobFileInner::Native(file) => file.size(),
+            #[cfg(feature = "remote")]
+            BlobFileInner::Remote(file) => file.size(),
+        }
+    }
+
+    /// Physical byte offset in the data file. `None` on remote handles. The
+    /// Cloud byte-range route does not expose storage layout.
+    pub fn position(&self) -> Option<u64> {
+        match &self.inner {
+            BlobFileInner::Native(file) => Some(file.position()),
+            #[cfg(feature = "remote")]
+            BlobFileInner::Remote(_) => None,
+        }
+    }
+
+    /// Path of the data file holding the blob. `None` on remote handles. The
+    /// Cloud byte-range route does not expose storage layout.
+    pub fn data_path(&self) -> Option<&Path> {
+        match &self.inner {
+            BlobFileInner::Native(file) => Some(file.data_path()),
+            #[cfg(feature = "remote")]
+            BlobFileInner::Remote(_) => None,
+        }
+    }
+
+    /// Native storage layout. `None` on remote handles. The Cloud byte-range
+    /// route does not expose layout.
+    pub fn kind(&self) -> Option<lance_core::datatypes::BlobKind> {
+        match &self.inner {
+            BlobFileInner::Native(file) => Some(file.kind()),
+            #[cfg(feature = "remote")]
+            BlobFileInner::Remote(_) => None,
+        }
+    }
+
+    /// External URI for native handles. Remote handles do not expose storage URIs.
+    pub fn uri(&self) -> Option<&str> {
+        match &self.inner {
+            BlobFileInner::Native(file) => file.uri(),
+            #[cfg(feature = "remote")]
+            BlobFileInner::Remote(_) => None,
+        }
+    }
+}
 
 /// One row-specific blob range read request.
 ///
@@ -264,7 +457,10 @@ pub(crate) async fn take_blob_files_aligned(
 
     let handles = dataset.take_blobs(row_ids, column).await?;
     ensure_all_row_ids_resolved(column, row_ids.len(), handles.len())?;
-    Ok(handles)
+    Ok(handles
+        .into_iter()
+        .map(|handle| handle.map(Into::into))
+        .collect())
 }
 
 #[cfg(test)]

rust/lancedb/src/remote/table.rs

@@ -1,7 +1,7 @@
 // SPDX-License-Identifier: Apache-2.0
 // SPDX-FileCopyrightText: Copyright The LanceDB Authors
 
-mod blobs;
+pub mod blobs;
 pub mod insert;
 
 use self::insert::{RemoteWriteExec, WriteOp};
@@ -3089,10 +3089,12 @@ mod tests {
             Box::pin(table.delete("false").map_ok(|_| ())),
             Box::pin(
                 table
-                    .add_columns(
-                        NewColumnTransform::SqlExpressions(vec![("x".into(), "y".into())]),
-                        None,
-                    )
+                    .add_columns()
+                    .transform(NewColumnTransform::SqlExpressions(vec![(
+                        "x".into(),
+                        "y".into(),
+                    )]))
+                    .execute()
                     .map_ok(|_| ()),
             ),
             Box::pin(async {
@@ -4300,32 +4302,9 @@ mod tests {
             "fetch_blobs",
         );
 
-        let message = table
-            .fetch_blob_files("image", &[1])
-            .await
-            .unwrap_err()
-            .to_string();
-        assert!(
-            message.contains("fetch_blob_files is not supported on LanceDB Cloud"),
-            "got: {message}"
-        );
-        assert!(
-            !message.contains("Use fetch_blobs"),
-            "old server must not be told to use fetch_blobs, got: {message}"
-        );
-    }
-
-    #[tokio::test]
-    async fn test_blob_files_point_at_fetch_blobs_on_a_blob_capable_server() {
-        let table = Table::new_with_handler_version(
-            "my_table",
-            semver::Version::new(0, 5, 0),
-            |_| -> http::Response<String> { panic!("fetch_blob_files must not reach the server") },
-        );
-
         assert_not_supported_error(
             table.fetch_blob_files("image", &[1]).await.unwrap_err(),
-            "Use fetch_blobs for full bytes",
+            "requires LanceDB Cloud server 0.5.0 or newer",
         );
     }
 
@@ -6411,13 +6390,12 @@ mod tests {
         });
 
         let result = table
-            .add_columns(
-                NewColumnTransform::SqlExpressions(vec![
-                    ("b".into(), "a + 1".into()),
-                    ("x".into(), "cast(NULL as int32)".into()),
-                ]),
-                None,
-            )
+            .add_columns()
+            .transform(NewColumnTransform::SqlExpressions(vec![
+                ("b".into(), "a + 1".into()),
+                ("x".into(), "cast(NULL as int32)".into()),
+            ]))
+            .execute()
             .await
             .unwrap();
 
@@ -7142,10 +7120,12 @@ mod tests {
             }
             "add_columns" => {
                 let _ = table
-                    .add_columns(
-                        NewColumnTransform::SqlExpressions(vec![("c".into(), "a + 1".into())]),
-                        None,
-                    )
+                    .add_columns()
+                    .transform(NewColumnTransform::SqlExpressions(vec![(
+                        "c".into(),
+                        "a + 1".into(),
+                    )]))
+                    .execute()
                     .await;
             }
             "drop_columns" => {
@@ -9903,10 +9883,12 @@ mod tests {
             .await
             .unwrap();
         branch
-            .add_columns(
-                NewColumnTransform::SqlExpressions(vec![("b".into(), "a + 1".into())]),
-                None,
-            )
+            .add_columns()
+            .transform(NewColumnTransform::SqlExpressions(vec![(
+                "b".into(),
+                "a + 1".into(),
+            )]))
+            .execute()
             .await
             .unwrap();
         branch

rust/lancedb/src/table.rs

@@ -3,6 +3,7 @@
 
 //! LanceDB Table APIs
 
+use crate::blob::BlobFile;
 use arrow_array::{LargeBinaryArray, RecordBatch, RecordBatchReader};
 use arrow_schema::{Schema, SchemaRef};
 use async_trait::async_trait;
@@ -12,7 +13,6 @@ use datafusion_physical_plan::ExecutionPlan;
 use datafusion_physical_plan::display::DisplayableExecutionPlan;
 use futures::StreamExt;
 use futures::stream::FuturesUnordered;
-use lance::dataset::BlobFile;
 pub use lance::dataset::ColumnAlteration;
 pub use lance::dataset::NewColumnTransform;
 pub use lance::dataset::ReadParams;
@@ -65,6 +65,7 @@ use crate::utils::{PatchReadParam, PatchWriteParam, resolve_arrow_field_path};
 use self::dataset::DatasetConsistencyWrapper;
 use self::merge::MergeInsertBuilder;
 
+pub mod add_columns;
 mod add_data;
 pub mod branch_merge;
 mod create_index;
@@ -79,6 +80,7 @@ pub mod schema_evolution;
 pub mod update;
 pub mod write_progress;
 use crate::index::waiter::wait_for_index;
+pub use add_columns::AddColumnsBuilder;
 #[cfg(feature = "remote")]
 pub(crate) use add_data::PreprocessingOutput;
 pub use add_data::{AddDataBuilder, AddDataMode, AddResult, NaNVectorBehavior};
@@ -1620,12 +1622,8 @@ impl Table {
     }
 
     /// Add new columns to the table, providing values to fill in.
-    pub async fn add_columns(
-        &self,
-        transforms: NewColumnTransform,
-        read_columns: Option<Vec<String>>,
-    ) -> Result<AddColumnsResult> {
-        self.inner.add_columns(transforms, read_columns).await
+    pub fn add_columns(&self) -> AddColumnsBuilder {
+        AddColumnsBuilder::new(self.inner.clone())
     }
 
     /// Change a column's name or nullability.

rust/lancedb/src/table/add_columns.rs

@@ -0,0 +1,161 @@
+// SPDX-License-Identifier: Apache-2.0
+// SPDX-FileCopyrightText: Copyright The LanceDB Authors
+
+//! Builder for adding columns to a table.
+
+use std::sync::Arc;
+
+use lance::dataset::NewColumnTransform;
+
+use super::BaseTable;
+use super::schema_evolution::AddColumnsResult;
+use crate::{Error, Result};
+
+/// Adds columns to a table. See [`Table::add_columns`](super::Table::add_columns).
+pub struct AddColumnsBuilder {
+    parent: Arc<dyn BaseTable>,
+    transform: Option<NewColumnTransform>,
+    read_columns: Option<Vec<String>>,
+}
+
+impl std::fmt::Debug for AddColumnsBuilder {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        f.debug_struct("AddColumnsBuilder")
+            .field("parent", &self.parent)
+            .field("has_transform", &self.transform.is_some())
+            .field("read_columns", &self.read_columns)
+            .finish()
+    }
+}
+
+impl AddColumnsBuilder {
+    pub(crate) fn new(parent: Arc<dyn BaseTable>) -> Self {
+        Self {
+            parent,
+            transform: None,
+            read_columns: None,
+        }
+    }
+
+    /// Set how the new columns' values are produced. Required.
+    pub fn transform(mut self, transform: NewColumnTransform) -> Self {
+        self.transform = Some(transform);
+        self
+    }
+
+    /// Limit which existing columns a [`NewColumnTransform::BatchUDF`] mapper
+    /// receives. Every other transform determines what it reads, so setting
+    /// this alongside one is an error rather than a silent no-op.
+    pub fn read_columns(mut self, columns: impl IntoIterator<Item = impl Into<String>>) -> Self {
+        self.read_columns = Some(columns.into_iter().map(Into::into).collect());
+        self
+    }
+
+    /// Add the columns.
+    pub async fn execute(self) -> Result<AddColumnsResult> {
+        let Self {
+            parent,
+            transform,
+            read_columns,
+        } = self;
+
+        let Some(transform) = transform else {
+            return Err(Error::InvalidInput {
+                message: "add_columns requires a transform".into(),
+            });
+        };
+
+        if read_columns.is_some() && !matches!(transform, NewColumnTransform::BatchUDF(_)) {
+            return Err(Error::InvalidInput {
+                message: "read_columns applies only to a BatchUDF transform; \
+                          every other transform determines what it reads"
+                    .into(),
+            });
+        }
+
+        parent.add_columns(transform, read_columns).await
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use std::sync::Arc;
+
+    use arrow_array::{Int32Array, RecordBatch, record_batch};
+    use arrow_schema::{DataType, Field, Schema};
+    use lance::dataset::{BatchUDF, NewColumnTransform};
+
+    use crate::Table;
+    use crate::connect;
+
+    async fn table_with_two_columns(name: &str) -> Table {
+        let conn = connect("memory://").execute().await.unwrap();
+        let batch = record_batch!(("x", Int32, [1, 2, 3]), ("y", Int32, [10, 20, 30])).unwrap();
+        conn.create_table(name, batch).execute().await.unwrap()
+    }
+
+    #[tokio::test]
+    async fn test_requires_a_transform() {
+        let table = table_with_two_columns("no_transform").await;
+        let err = table.add_columns().execute().await.unwrap_err();
+        assert!(
+            err.to_string().contains("requires a transform"),
+            "got: {err}"
+        );
+    }
+
+    #[tokio::test]
+    async fn test_read_columns_with_sql_expressions_is_rejected() {
+        let table = table_with_two_columns("read_cols_sql").await;
+        let err = table
+            .add_columns()
+            .transform(NewColumnTransform::SqlExpressions(vec![(
+                "doubled".into(),
+                "x * 2".into(),
+            )]))
+            .read_columns(["x"])
+            .execute()
+            .await
+            .unwrap_err();
+        assert!(err.to_string().contains("BatchUDF"), "got: {err}");
+
+        let schema = table.schema().await.unwrap();
+        assert!(
+            schema.field_with_name("doubled").is_err(),
+            "a rejected call must not commit"
+        );
+    }
+
+    #[tokio::test]
+    async fn test_read_columns_limits_what_a_batch_udf_sees() {
+        let table = table_with_two_columns("read_cols_udf").await;
+
+        let output_schema = Arc::new(Schema::new(vec![Field::new("sum", DataType::Int32, true)]));
+        let mapper_schema = output_schema.clone();
+        let udf = BatchUDF {
+            mapper: Box::new(move |batch: &RecordBatch| {
+                assert!(batch.column_by_name("x").is_some());
+                assert!(batch.column_by_name("y").is_none(), "y was not requested");
+                let x = batch["x"].as_any().downcast_ref::<Int32Array>().unwrap();
+                let doubled: Int32Array = x.iter().map(|v| v.map(|v| v * 2)).collect();
+                Ok(RecordBatch::try_new(
+                    mapper_schema.clone(),
+                    vec![Arc::new(doubled)],
+                )?)
+            }),
+            output_schema,
+            result_checkpoint: None,
+        };
+
+        table
+            .add_columns()
+            .transform(NewColumnTransform::BatchUDF(udf))
+            .read_columns(["x"])
+            .execute()
+            .await
+            .unwrap();
+
+        let schema = table.schema().await.unwrap();
+        assert!(schema.field_with_name("sum").is_ok());
+    }
+}

rust/lancedb/src/table/add_data.rs

@@ -576,10 +576,12 @@ mod tests {
 
         // Add a new physical column AFTER the embedding column.
         table
-            .add_columns(
-                NewColumnTransform::SqlExpressions(vec![("score".into(), "42.0".into())]),
-                None,
-            )
+            .add_columns()
+            .transform(NewColumnTransform::SqlExpressions(vec![(
+                "score".into(),
+                "42.0".into(),
+            )]))
+            .execute()
             .await
             .unwrap();
 
@@ -683,7 +685,9 @@ mod tests {
             true,
         )]));
         table
-            .add_columns(NewColumnTransform::AllNulls(nested_schema), None)
+            .add_columns()
+            .transform(NewColumnTransform::AllNulls(nested_schema))
+            .execute()
             .await
             .unwrap();
 

rust/lancedb/src/table/schema_evolution.rs

@@ -193,10 +193,12 @@ mod tests {
 
         // Add a computed column
         let result = table
-            .add_columns(
-                NewColumnTransform::SqlExpressions(vec![("doubled".into(), "id * 2".into())]),
-                None,
-            )
+            .add_columns()
+            .transform(NewColumnTransform::SqlExpressions(vec![(
+                "doubled".into(),
+                "id * 2".into(),
+            )]))
+            .execute()
             .await
             .unwrap();
 
@@ -251,13 +253,12 @@ mod tests {
 
         // Add multiple columns at once
         table
-            .add_columns(
-                NewColumnTransform::SqlExpressions(vec![
-                    ("y".into(), "x + 1".into()),
-                    ("z".into(), "x * x".into()),
-                ]),
-                None,
-            )
+            .add_columns()
+            .transform(NewColumnTransform::SqlExpressions(vec![
+                ("y".into(), "x + 1".into()),
+                ("z".into(), "x * x".into()),
+            ]))
+            .execute()
             .await
             .unwrap();
 
@@ -283,10 +284,12 @@ mod tests {
 
         // Add a column with a constant value
         table
-            .add_columns(
-                NewColumnTransform::SqlExpressions(vec![("constant".into(), "42".into())]),
-                None,
-            )
+            .add_columns()
+            .transform(NewColumnTransform::SqlExpressions(vec![(
+                "constant".into(),
+                "42".into(),
+            )]))
+            .execute()
             .await
             .unwrap();
 
@@ -659,10 +662,12 @@ mod tests {
 
         // Add column increments version
         let add_result = table
-            .add_columns(
-                NewColumnTransform::SqlExpressions(vec![("c".into(), "a + b".into())]),
-                None,
-            )
+            .add_columns()
+            .transform(NewColumnTransform::SqlExpressions(vec![(
+                "c".into(),
+                "a + b".into(),
+            )]))
+            .execute()
             .await
             .unwrap();
         assert!(add_result.version > v1);

rust/lancedb/tests/blob_integration.rs

@@ -9,14 +9,17 @@ use arrow_array::{
 };
 use arrow_schema::{DataType, Field, Fields, Schema};
 use futures::TryStreamExt;
+use lance::Dataset;
 use lance_encoding::version::LanceFileVersion;
 use lancedb::{
     Connection, Error, Result, Table,
     blob::{BlobRangeRequest, blob},
     connect, connect_namespace,
-    database::listing::OPT_NEW_TABLE_ENABLE_STABLE_ROW_IDS,
+    database::listing::{
+        ListingDatabaseOptions, NewTableConfig, OPT_NEW_TABLE_ENABLE_STABLE_ROW_IDS,
+    },
     query::{ExecutableQuery, QueryBase},
-    table::{AddDataMode, CompactionOptions, OptimizeAction},
+    table::{AddDataMode, CompactionOptions, OptimizeAction, OptimizeStats},
 };
 use tempfile::tempdir;
 
@@ -1075,3 +1078,252 @@ async fn fetch_blob_files_aligns_across_fragments_with_nulls_and_dups() -> Resul
     }
     Ok(())
 }
+
+/// Rows exercising the null/empty interleavings from
+/// <https://github.com/lancedb/lancedb/issues/3744>: a payload, a null, a valid
+/// empty value, then payloads whose descriptors a fragment rewrite used to zero.
+fn null_empty_input_batch() -> RecordBatch {
+    let owned = [
+        Some(dedicated_blob_bytes(1)),
+        None,
+        Some(Vec::new()),
+        Some(dedicated_blob_bytes(4)),
+        Some(dedicated_blob_bytes(5)),
+        Some(dedicated_blob_bytes(6)),
+    ];
+    let payloads: Vec<Option<&[u8]>> = owned.iter().map(|payload| payload.as_deref()).collect();
+    binary_input_batch(&[1, 2, 3, 4, 5, 6], &payloads)
+}
+
+/// One `(id, Some((payload length, first byte)))` per live row, or `(id, None)`
+/// for a null blob.  Comparing lengths and first bytes keeps failure output
+/// readable where comparing whole payloads would not.
+type BlobSummary = Vec<(i64, Option<(usize, Option<u8>)>)>;
+
+/// The rows [`null_empty_input_batch`] leaves behind after `id IN (1, 4)` is
+/// deleted: a null, a valid empty value, and the two payloads that follow them.
+fn expected_null_empty_survivors() -> BlobSummary {
+    vec![
+        (2, None),
+        (3, Some((0, None))),
+        (5, Some((DEDICATED_BLOB_LEN, Some(5)))),
+        (6, Some((DEDICATED_BLOB_LEN, Some(6)))),
+    ]
+}
+
+/// `optimize()` only rewrites a fragment when lance's compaction planner selects
+/// it — here because the delete pushes the fragment past
+/// `materialize_deletions_threshold` (0.1 by default; these tests delete 2 of 6
+/// rows). Without this check, a planner or threshold change upstream would leave
+/// both regression tests green while no rewrite happened at all.
+fn assert_compacted(stats: &OptimizeStats) {
+    let metrics = stats
+        .compaction
+        .as_ref()
+        .expect("OptimizeAction::All runs compaction");
+    assert!(
+        metrics.fragments_removed >= 1,
+        "optimize() rewrote no fragment, so this test proves nothing: {metrics:?}"
+    );
+}
+
+fn summarize(rows: &[(i64, Option<Vec<u8>>)]) -> BlobSummary {
+    rows.iter()
+        .map(|(id, payload)| {
+            (
+                *id,
+                payload
+                    .as_ref()
+                    .map(|bytes| (bytes.len(), bytes.first().copied())),
+            )
+        })
+        .collect()
+}
+
+async fn sorted_id_rowid(table: &Table) -> Result<Vec<(i64, u64)>> {
+    let mut pairs = collect_id_rowid(table).await?;
+    pairs.sort_by_key(|(id, _)| *id);
+    Ok(pairs)
+}
+
+/// `{position, size}` descriptors of a legacy v1 blob column, keyed by `id`.
+async fn v1_blob_descriptors(table: &Table) -> Result<Vec<(i64, Option<(u64, u64)>)>> {
+    let batches = table
+        .query()
+        .execute()
+        .await?
+        .try_collect::<Vec<_>>()
+        .await?;
+    let batch = arrow_select::concat::concat_batches(&batches[0].schema(), &batches).unwrap();
+    let ids = batch
+        .column_by_name("id")
+        .unwrap()
+        .as_any()
+        .downcast_ref::<Int64Array>()
+        .unwrap();
+    let descriptors = batch
+        .column_by_name("image")
+        .unwrap()
+        .as_any()
+        .downcast_ref::<StructArray>()
+        .expect("v1 blob column reads back as a descriptor struct");
+    let position = descriptors
+        .column_by_name("position")
+        .unwrap()
+        .as_any()
+        .downcast_ref::<UInt64Array>()
+        .unwrap();
+    let size = descriptors
+        .column_by_name("size")
+        .unwrap()
+        .as_any()
+        .downcast_ref::<UInt64Array>()
+        .unwrap();
+    let mut rows: Vec<(i64, Option<(u64, u64)>)> = (0..batch.num_rows())
+        .map(|row| {
+            let descriptor =
+                (!descriptors.is_null(row)).then(|| (position.value(row), size.value(row)));
+            (ids.value(row), descriptor)
+        })
+        .collect();
+    rows.sort_by_key(|(id, _)| *id);
+    Ok(rows)
+}
+
+/// Payload bytes of every live row of a legacy v1 blob column, keyed by `id`.
+/// [`Table::fetch_blobs`] rejects v1 columns, so read them through lance.
+async fn v1_blob_payloads(dataset_uri: &str, table: &Table) -> Result<Vec<(i64, Option<Vec<u8>>)>> {
+    let pairs = sorted_id_rowid(table).await?;
+    let row_ids: Vec<u64> = pairs.iter().map(|(_, row_id)| *row_id).collect();
+    let dataset = Arc::new(Dataset::open(dataset_uri).await?);
+    let files = dataset.take_blobs(&row_ids, "image").await?;
+    assert_eq!(
+        files.len(),
+        pairs.len(),
+        "take_blobs returned {} handles for {} live rows",
+        files.len(),
+        pairs.len()
+    );
+    let mut rows = Vec::with_capacity(pairs.len());
+    for ((id, _), file) in pairs.iter().zip(files) {
+        let payload = match file {
+            Some(file) => Some(file.read().await?.to_vec()),
+            None => None,
+        };
+        rows.push((*id, payload));
+    }
+    Ok(rows)
+}
+
+/// Length and first byte of every live blob v2 value, keyed by `id`.
+async fn blob_v2_values(table: &Table) -> Result<BlobSummary> {
+    let pairs = sorted_id_rowid(table).await?;
+    let row_ids: Vec<u64> = pairs.iter().map(|(_, row_id)| *row_id).collect();
+    let bytes = table.fetch_blobs("image", &row_ids).await?;
+    Ok(pairs
+        .iter()
+        .enumerate()
+        .map(|(slot, (id, _))| {
+            let value = (!bytes.is_null(slot))
+                .then(|| (bytes.value(slot).len(), bytes.value(slot).first().copied()));
+            (*id, value)
+        })
+        .collect())
+}
+
+/// Regression test for [#3744]: on storage 2.0 (legacy v1 descriptors),
+/// compaction rewrote every payload following a null or empty value in the same
+/// fragment as `{position: 0, size: 0}`, so the payload bytes read back as `b""`
+/// and the new fragment no longer referenced them at all.
+///
+/// [#3744]: https://github.com/lancedb/lancedb/issues/3744
+#[tokio::test]
+async fn optimize_preserves_v1_blob_payloads_with_null_and_empty() -> Result<()> {
+    let tmp = tempdir().unwrap();
+    let db_uri = tmp.path().to_str().unwrap().to_string();
+    let db = connect(&db_uri)
+        .database_options(&ListingDatabaseOptions {
+            new_table_config: NewTableConfig {
+                data_storage_version: Some(LanceFileVersion::V2_0),
+                ..Default::default()
+            },
+            ..Default::default()
+        })
+        .execute()
+        .await?;
+    let legacy = Field::new("image", DataType::LargeBinary, true).with_metadata(
+        std::collections::HashMap::from([("lance-encoding:blob".to_string(), "true".to_string())]),
+    );
+    let schema = Arc::new(Schema::new(vec![
+        Field::new("id", DataType::Int64, false),
+        legacy,
+    ]));
+    let table = db.create_empty_table("t", schema).execute().await?;
+    table.add(null_empty_input_batch()).execute().await?;
+    assert_eq!(
+        storage_format_version(&table).await,
+        LanceFileVersion::V2_0.resolve(),
+        "v1 blob descriptors only exist below storage 2.2"
+    );
+    let dataset_uri = table.uri().await?;
+
+    // Any rewrite triggers it; deleting rows is the shape from the issue.
+    table.delete("id IN (1, 4)").await?;
+    let descriptors_before = v1_blob_descriptors(&table).await?;
+    let before = v1_blob_payloads(&dataset_uri, &table).await?;
+    assert_eq!(
+        summarize(&before),
+        expected_null_empty_survivors(),
+        "test setup no longer produces the null/empty/payload mix"
+    );
+
+    let stats = table.optimize(OptimizeAction::All).await?;
+    assert_compacted(&stats);
+
+    let descriptors_after = v1_blob_descriptors(&table).await?;
+    let after = v1_blob_payloads(&dataset_uri, &table).await?;
+    assert_eq!(
+        summarize(&after),
+        summarize(&before),
+        "optimize() lost blob payloads; descriptors before={descriptors_before:?} after={descriptors_after:?}"
+    );
+    assert!(after == before, "optimize() changed blob payload bytes");
+    Ok(())
+}
+
+/// Regression test for the blob v2 half of [#3744]: compaction rewrote a valid
+/// empty value as null, destroying the null-vs-empty distinction.
+///
+/// [#3744]: https://github.com/lancedb/lancedb/issues/3744
+#[tokio::test]
+async fn optimize_preserves_blob_v2_null_and_empty_distinction() -> Result<()> {
+    let tmp = tempdir().unwrap();
+    let db = connect(tmp.path().to_str().unwrap()).execute().await?;
+    let table = db
+        .create_empty_table("t", blob_table_schema())
+        .execute()
+        .await?;
+    table.add(null_empty_input_batch()).execute().await?;
+    assert!(
+        storage_format_version(&table).await >= LanceFileVersion::V2_2,
+        "blob v2 columns require storage >= 2.2"
+    );
+
+    table.delete("id IN (1, 4)").await?;
+    let before = blob_v2_values(&table).await?;
+    assert_eq!(
+        before,
+        expected_null_empty_survivors(),
+        "test setup no longer produces the null/empty/payload mix"
+    );
+
+    let stats = table.optimize(OptimizeAction::All).await?;
+    assert_compacted(&stats);
+
+    assert_eq!(
+        blob_v2_values(&table).await?,
+        before,
+        "optimize() changed blob v2 values"
+    );
+    Ok(())
+}