fix(rust): reuse SQ preflight compaction plan

# Conflicts:
#	rust/lancedb/src/table/optimize.rs

.github/workflows/rust.yml

@@ -296,18 +296,16 @@ jobs:
           cargo update -p aws-types --precise 1.3.9
           cargo update -p aws-sigv4 --precise 1.3.5
           cargo update -p aws-credential-types --precise 1.2.8
-          # aws-smithy-checksums must stay at or above 0.63.13: OpenDAL's S3
-          # service needs crc-fast ~1.9, and older releases pin it to ~1.3.
-          cargo update -p aws-smithy-checksums --precise 0.63.13
+          cargo update -p aws-smithy-checksums --precise 0.63.9
           cargo update -p aws-smithy-runtime --precise 1.9.3
-          cargo update -p aws-smithy-http --precise 0.62.6
-          cargo update -p aws-smithy-eventstream --precise 0.60.14
+          cargo update -p aws-smithy-http --precise 0.62.4
+          cargo update -p aws-smithy-eventstream --precise 0.60.12
           cargo update -p aws-smithy-http-client --precise 1.1.3
           cargo update -p aws-smithy-observability --precise 0.1.4
           cargo update -p aws-smithy-query --precise 0.60.8
-          cargo update -p aws-smithy-runtime-api --precise 1.9.3
-          cargo update -p aws-smithy-async --precise 1.2.7
-          cargo update -p aws-smithy-types --precise 1.3.6
+          cargo update -p aws-smithy-runtime-api --precise 1.9.1
+          cargo update -p aws-smithy-async --precise 1.2.6
+          cargo update -p aws-smithy-types --precise 1.3.5
           cargo update -p aws-smithy-xml --precise 0.60.11
           cargo update -p home --precise 0.5.9
       - name: cargo +${{ matrix.msrv }} check

AGENTS.md

@@ -152,54 +152,3 @@ Please consider the following when reviewing code contributions.
 ### Documentation
 * New features must include updates to the rust documentation comments. Link to
   relevant structs and methods to increase the value of documentation.
-
-## Cursor Cloud specific instructions
-
-The VM snapshot already has the Rust `1.97.0` toolchain (auto-selected by
-`rust-toolchain.toml`), `protoc`, `uv` (on `PATH` via `~/.bashrc`), the Rust
-debug build artifacts, the Python editable extension, and `nodejs/node_modules`.
-The startup update script only refreshes dependencies (`uv sync` for Python and
-`pnpm install` for Node); it deliberately does NOT rebuild the native
-extensions. After changing Rust or PyO3/napi binding code you must rebuild the
-affected binding yourself (see per-binding rebuild commands below).
-
-Non-obvious caveats discovered during setup:
-
-* The documented Python bootstrap `uv run --extra tests --extra dev maturin
-  develop --extras tests,dev` does not work as-is here: `maturin` is not
-  installed as a CLI in the uv environment, and `maturin develop --extras`
-  runs its own dependency resolution that cannot find the prerelease
-  `pylance==9.0.0rc1` (it lacks the extra package index that `uv` uses via
-  `uv.lock`). Because `uv run --extra tests --extra dev` already installs those
-  extras, the working command is:
-  `cd python && uv run --extra tests --extra dev --with maturin maturin develop`
-  (note: `--with maturin`, and no `--extras`). This is the Python binding
-  rebuild command.
-* Rust core, the Python extension (maturin), and the Node addon (napi) all
-  compile into the SHARED `/workspace/target`. Cargo feature unification differs
-  between `maturin develop` and `pnpm build`, so alternating between building
-  the Python and Node bindings forces a full recompile of shared crates
-  (`lancedb`, `datafusion`, `lance-*`) — roughly 6-7 min each way on this
-  4-core VM. Build one binding at a time to avoid the churn.
-* The `_lancedb` release build (triggered when `uv run`/`uv sync` installs the
-  `lancedb` project itself) uses `lto = "fat"` + `opt-level = 3`, needs ~11 GB
-  RAM, and takes ~20 min cold on this VM. To avoid it, the update script uses
-  `uv sync --no-install-project --inexact` (the `--inexact` flag is required so
-  the sync does not uninstall the editable extension). Prefer the debug
-  `maturin develop` (~6 min cold, seconds when warm) for iteration.
-* `cargo check` only produces metadata, so the first `cargo run --example ...`
-  or `cargo test` after a check triggers a large codegen/link compile.
-* Node binding rebuild: `cd nodejs && pnpm build` (napi debug build + `tsc`).
-  The native addon lands at `nodejs/dist/lancedb.linux-x64-gnu.node`.
-
-Verified working (local backend, no cloud credentials needed):
-
-* Rust: `cargo check/clippy --features remote --tests --examples`,
-  `cargo test --features remote -p lancedb --lib`, `cargo run --features remote
-  --example simple`.
-* Python: `cd python && uv run --extra tests pytest python/tests/test_table.py`,
-  `uv run --directory python --extra dev ruff check python`.
-* Node: `cd nodejs && pnpm lint`, `pnpm test __test__/connection.test.ts`.
-
-Java (`java/`) is optional; its integration tests need LanceDB Cloud
-credentials (`LANCEDB_DB`, `LANCEDB_API_KEY`) and were not set up here.

Cargo.toml

@@ -13,20 +13,20 @@ categories = ["database-implementations"]
 rust-version = "1.91.0"
 
 [workspace.dependencies]
-lance = { "version" = "=11.0.0-beta.2", default-features = false, "tag" = "v11.0.0-beta.2", "git" = "https://github.com/lance-format/lance.git" }
-lance-core = { "version" = "=11.0.0-beta.2", "tag" = "v11.0.0-beta.2", "git" = "https://github.com/lance-format/lance.git" }
-lance-datagen = { "version" = "=11.0.0-beta.2", "tag" = "v11.0.0-beta.2", "git" = "https://github.com/lance-format/lance.git" }
-lance-file = { "version" = "=11.0.0-beta.2", "tag" = "v11.0.0-beta.2", "git" = "https://github.com/lance-format/lance.git" }
-lance-io = { "version" = "=11.0.0-beta.2", default-features = false, "tag" = "v11.0.0-beta.2", "git" = "https://github.com/lance-format/lance.git" }
-lance-index = { "version" = "=11.0.0-beta.2", "tag" = "v11.0.0-beta.2", "git" = "https://github.com/lance-format/lance.git" }
-lance-linalg = { "version" = "=11.0.0-beta.2", "tag" = "v11.0.0-beta.2", "git" = "https://github.com/lance-format/lance.git" }
-lance-namespace = { "version" = "=11.0.0-beta.2", "tag" = "v11.0.0-beta.2", "git" = "https://github.com/lance-format/lance.git" }
-lance-namespace-impls = { "version" = "=11.0.0-beta.2", default-features = false, "tag" = "v11.0.0-beta.2", "git" = "https://github.com/lance-format/lance.git" }
-lance-table = { "version" = "=11.0.0-beta.2", "tag" = "v11.0.0-beta.2", "git" = "https://github.com/lance-format/lance.git" }
-lance-testing = { "version" = "=11.0.0-beta.2", "tag" = "v11.0.0-beta.2", "git" = "https://github.com/lance-format/lance.git" }
-lance-datafusion = { "version" = "=11.0.0-beta.2", "tag" = "v11.0.0-beta.2", "git" = "https://github.com/lance-format/lance.git" }
-lance-encoding = { "version" = "=11.0.0-beta.2", "tag" = "v11.0.0-beta.2", "git" = "https://github.com/lance-format/lance.git" }
-lance-arrow = { "version" = "=11.0.0-beta.2", "tag" = "v11.0.0-beta.2", "git" = "https://github.com/lance-format/lance.git" }
+lance = { "version" = "=10.1.0-beta.1", default-features = false, "tag" = "v10.1.0-beta.1", "git" = "https://github.com/lance-format/lance.git" }
+lance-core = { "version" = "=10.1.0-beta.1", "tag" = "v10.1.0-beta.1", "git" = "https://github.com/lance-format/lance.git" }
+lance-datagen = { "version" = "=10.1.0-beta.1", "tag" = "v10.1.0-beta.1", "git" = "https://github.com/lance-format/lance.git" }
+lance-file = { "version" = "=10.1.0-beta.1", "tag" = "v10.1.0-beta.1", "git" = "https://github.com/lance-format/lance.git" }
+lance-io = { "version" = "=10.1.0-beta.1", default-features = false, "tag" = "v10.1.0-beta.1", "git" = "https://github.com/lance-format/lance.git" }
+lance-index = { "version" = "=10.1.0-beta.1", "tag" = "v10.1.0-beta.1", "git" = "https://github.com/lance-format/lance.git" }
+lance-linalg = { "version" = "=10.1.0-beta.1", "tag" = "v10.1.0-beta.1", "git" = "https://github.com/lance-format/lance.git" }
+lance-namespace = { "version" = "=10.1.0-beta.1", "tag" = "v10.1.0-beta.1", "git" = "https://github.com/lance-format/lance.git" }
+lance-namespace-impls = { "version" = "=10.1.0-beta.1", default-features = false, "tag" = "v10.1.0-beta.1", "git" = "https://github.com/lance-format/lance.git" }
+lance-table = { "version" = "=10.1.0-beta.1", "tag" = "v10.1.0-beta.1", "git" = "https://github.com/lance-format/lance.git" }
+lance-testing = { "version" = "=10.1.0-beta.1", "tag" = "v10.1.0-beta.1", "git" = "https://github.com/lance-format/lance.git" }
+lance-datafusion = { "version" = "=10.1.0-beta.1", "tag" = "v10.1.0-beta.1", "git" = "https://github.com/lance-format/lance.git" }
+lance-encoding = { "version" = "=10.1.0-beta.1", "tag" = "v10.1.0-beta.1", "git" = "https://github.com/lance-format/lance.git" }
+lance-arrow = { "version" = "=10.1.0-beta.1", "tag" = "v10.1.0-beta.1", "git" = "https://github.com/lance-format/lance.git" }
 ahash = "0.8"
 # Note that this one does not include pyarrow
 arrow = { version = "58.0.0", optional = false }

java/pom.xml

@@ -28,7 +28,7 @@
     <properties>
         <project.build.sourceEncoding>UTF-8</project.build.sourceEncoding>
         <arrow.version>15.0.0</arrow.version>
-        <lance-core.version>11.0.0-beta.2</lance-core.version>
+        <lance-core.version>10.1.0-beta.1</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>

rust/lancedb/Cargo.toml

@@ -49,8 +49,8 @@ lance-namespace = { workspace = true }
 lance-namespace-impls = { workspace = true }
 metrics = { workspace = true, optional = true }
 metrics-util = { workspace = true, optional = true }
-# Pin the GooseFS SDK to the version required by Lance's OpenDAL dependency.
-goosefs-sdk = { version = "=0.1.9", optional = true }
+# Pin the transitive GooseFS SDK until the 0.1.6 compile break is fixed upstream.
+goosefs-sdk = { version = "=0.1.5", optional = true }
 moka = { workspace = true }
 pin-project = { workspace = true }
 tokio = { version = "1.23", features = ["rt-multi-thread", "sync"] }

rust/lancedb/src/blob.rs

@@ -17,7 +17,7 @@ use arrow_array::builder::LargeBinaryBuilder;
 use arrow_schema::{DataType, Field, Schema};
 use lance::dataset::{BlobRangeRequest as LanceBlobRangeRequest, Dataset, WriteParams};
 use lance_arrow::FieldExt;
-use lance_file::version::LanceFileVersion;
+use lance_encoding::version::LanceFileVersion;
 use lance_io::object_store::ObjectStore;
 use object_store::path::Path;
 

rust/lancedb/src/connection.rs

@@ -34,7 +34,7 @@ use crate::remote::{
     db::{OPT_REMOTE_API_KEY, OPT_REMOTE_HOST_OVERRIDE, OPT_REMOTE_REGION},
 };
 use lance::io::ObjectStoreParams;
-pub use lance_file::version::LanceFileVersion;
+pub use lance_encoding::version::LanceFileVersion;
 #[cfg(feature = "remote")]
 use lance_io::object_store::StorageOptions;
 use lance_io::object_store::{StorageOptionsAccessor, StorageOptionsProvider};

rust/lancedb/src/database/listing.rs

@@ -12,7 +12,7 @@ use lance::dataset::refs::Ref;
 use lance::dataset::{ReadParams, WriteMode, builder::DatasetBuilder};
 use lance::io::{ObjectStore, ObjectStoreParams, WrappingObjectStore};
 use lance_datafusion::utils::StreamingWriteSource;
-use lance_file::version::LanceFileVersion;
+use lance_encoding::version::LanceFileVersion;
 use lance_io::object_store::{StorageOptionsAccessor, StorageOptionsProvider};
 use lance_table::io::commit::commit_handler_from_url;
 use object_store::local::LocalFileSystem;

rust/lancedb/src/database/namespace.rs

@@ -201,7 +201,7 @@ impl LanceNamespaceDatabase {
         &self,
         request: &DbCreateTableRequest,
     ) -> Result<(
-        Option<lance_file::version::LanceFileVersion>,
+        Option<lance_encoding::version::LanceFileVersion>,
         Option<bool>,
         Option<bool>,
     )> {
@@ -214,7 +214,7 @@ impl LanceNamespaceDatabase {
 
         let storage_version_override = storage_options
             .and_then(|opts| opts.get(OPT_NEW_TABLE_STORAGE_VERSION))
-            .map(|s| s.parse::<lance_file::version::LanceFileVersion>())
+            .map(|s| s.parse::<lance_encoding::version::LanceFileVersion>())
             .transpose()?;
 
         let v2_manifest_override = storage_options

rust/lancedb/src/remote/table.rs

@@ -2942,7 +2942,7 @@ impl<S: HttpSend> BaseTable for RemoteTable<S> {
 }
 
 #[derive(Serialize, Clone, Debug)]
-pub struct MergeInsertRequest {
+pub(crate) struct MergeInsertRequest {
     on: String,
     when_matched_update_all: bool,
     when_matched_update_all_filt: Option<String>,
@@ -5907,18 +5907,16 @@ mod tests {
             .await
             .unwrap();
 
-        // Positions are relative to the first retained token, so dropping the
-        // leading "hello" stop word does not shift the remaining tokens.
         assert_eq!(
             tokens,
             vec![
                 FtsToken {
                     text: "こんにちは".to_string(),
-                    position: 0,
+                    position: 1,
                 },
                 FtsToken {
                     text: "世界".to_string(),
-                    position: 1,
+                    position: 2,
                 },
             ]
         );

rust/lancedb/src/remote/table/blobs.rs

@@ -90,7 +90,7 @@ struct RemoteBlobState {
 
 /// Seekable Cloud blob handle over HTTP Range.
 #[derive(Debug)]
-pub struct RemoteBlobFile {
+pub(crate) struct RemoteBlobFile {
     requester: Arc<dyn BlobRangeRequester>,
     state: Mutex<RemoteBlobState>,
     closed: AtomicBool,

rust/lancedb/src/remote/table/insert.rs

@@ -33,7 +33,7 @@ use crate::table::{AddResult, MergeResult};
 /// same Arrow-IPC streaming body and error side-channel; only the target
 /// endpoint, query parameters, and parsed result type differ.
 #[derive(Debug, Clone)]
-pub enum WriteOp {
+pub(crate) enum WriteOp {
     /// `add`: stream to `/v1/table/{id}/insert/`, optionally overwriting.
     Insert { overwrite: bool },
     /// `merge_insert`: stream to `/v1/table/{id}/merge_insert/` with the merge
@@ -49,7 +49,7 @@ pub enum WriteOp {
 /// The parsed server response for a completed write, discriminated by the
 /// operation that produced it.
 #[derive(Debug, Clone)]
-pub enum WriteResult {
+pub(crate) enum WriteResult {
     Add(AddResult),
     Merge(MergeResult),
 }

rust/lancedb/src/table/optimize.rs

@@ -6,19 +6,162 @@
 //! This module contains the implementation of optimization operations that help
 //! maintain good performance for LanceDB tables.
 
-use std::sync::Arc;
+use std::{collections::HashSet, sync::Arc};
 
+use arrow_schema::DataType;
 use lance::dataset::cleanup::RemovalStats;
-use lance::dataset::optimize::{CompactionMetrics, IndexRemapperOptions, compact_files};
-use lance::index::DatasetIndexExt;
+use lance::dataset::optimize::{
+    CompactionMetrics, CompactionPlan, CompactionPlanner, IndexRemapperOptions,
+    compact_files_with_planner, plan_compaction,
+};
+use lance::index::{DatasetIndexExt, DatasetIndexInternalExt};
+use lance_index::IndexType;
+use lance_index::metrics::NoOpMetricsCollector;
 use lance_index::optimize::OptimizeOptions;
-use log::info;
+use log::{debug, info};
 
+use super::NativeTable;
+use crate::error::{Error, Result};
 pub use chrono::Duration;
 pub use lance::dataset::optimize::CompactionOptions;
 
-use super::NativeTable;
-use crate::error::Result;
+const MAX_ARROW_FIXED_SIZE_LIST_CHILD_INDEX: u64 = u32::MAX as u64;
+
+struct PrecomputedCompactionPlanner {
+    plan: CompactionPlan,
+}
+
+#[async_trait::async_trait]
+impl CompactionPlanner for PrecomputedCompactionPlanner {
+    async fn plan(&self, dataset: &lance::Dataset) -> lance::Result<CompactionPlan> {
+        debug_assert_eq!(dataset.manifest().version, self.plan.read_version());
+        Ok(self.plan.clone())
+    }
+}
+
+fn sq_vector_dimension(data_type: &DataType) -> Option<i32> {
+    match data_type {
+        DataType::FixedSizeList(_, dimension) => Some(*dimension),
+        DataType::List(field) => match field.data_type() {
+            DataType::FixedSizeList(_, dimension) => Some(*dimension),
+            _ => None,
+        },
+        _ => None,
+    }
+}
+
+fn segment_touches_fragments(
+    fragment_ids: Option<impl IntoIterator<Item = u32>>,
+    affected_fragments: &HashSet<u64>,
+) -> bool {
+    fragment_ids.is_none_or(|fragment_ids| {
+        fragment_ids
+            .into_iter()
+            .any(|fragment| affected_fragments.contains(&(fragment as u64)))
+    })
+}
+
+fn oversized_sq_partition(
+    partition_sizes: impl IntoIterator<Item = u64>,
+    dimension: u64,
+) -> Option<u64> {
+    partition_sizes.into_iter().find(|partition_size| {
+        partition_size
+            .checked_mul(dimension)
+            .is_none_or(|child_len| child_len > MAX_ARROW_FIXED_SIZE_LIST_CHILD_INDEX)
+    })
+}
+
+async fn validate_sq_index_remapping(
+    dataset: &lance::Dataset,
+    options: &CompactionOptions,
+    has_custom_remapper: bool,
+) -> Result<CompactionPlan> {
+    let plan = plan_compaction(dataset, options).await?;
+    if options.defer_index_remap || has_custom_remapper || dataset.manifest().uses_stable_row_ids()
+    {
+        return Ok(plan);
+    }
+
+    if plan.tasks.is_empty() {
+        return Ok(plan);
+    }
+    let affected_fragments: HashSet<u64> = plan
+        .tasks
+        .iter()
+        .flat_map(|task| task.fragments.iter().map(|fragment| fragment.id))
+        .collect();
+
+    for segment in dataset.load_indices().await?.iter() {
+        if !segment_touches_fragments(
+            segment
+                .fragment_bitmap
+                .as_ref()
+                .map(|fragment_bitmap| fragment_bitmap.iter()),
+            &affected_fragments,
+        ) {
+            continue;
+        }
+        let Some(field_id) = segment.fields.first() else {
+            continue;
+        };
+        let Some(field) = dataset.schema().field_by_id(*field_id) else {
+            continue;
+        };
+        let data_type = field.data_type();
+        let Some(dimension) = sq_vector_dimension(&data_type) else {
+            continue;
+        };
+        let dimension = u64::try_from(dimension).map_err(|_| Error::InvalidInput {
+            message: format!(
+                "SQ index '{}' has an invalid vector dimension of {}",
+                segment.name, dimension
+            ),
+        })?;
+
+        let field_path = dataset.schema().field_path(*field_id)?;
+        let vector_index = match dataset
+            .open_vector_index(&field_path, &segment.uuid, &NoOpMetricsCollector)
+            .await
+        {
+            Ok(vector_index) => vector_index,
+            Err(error) => {
+                // The default remapper also treats an index it cannot open as
+                // unusable and drops it, so it cannot reach the SQ take kernel.
+                debug!(
+                    "Skipping remap preflight for index segment {} because it could not be opened: {}",
+                    segment.uuid, error
+                );
+                continue;
+            }
+        };
+        if !matches!(
+            vector_index.index_type(),
+            IndexType::IvfSq | IndexType::IvfHnswSq
+        ) {
+            continue;
+        }
+
+        let partition_sizes = (0..vector_index.total_partitions())
+            .map(|partition_id| vector_index.partition_size(partition_id) as u64);
+        if let Some(partition_size) = oversized_sq_partition(partition_sizes, dimension) {
+            let max_partition_size = MAX_ARROW_FIXED_SIZE_LIST_CHILD_INDEX / dimension;
+            return Err(Error::InvalidInput {
+                message: format!(
+                    "Cannot compact table because SQ index '{}' segment {} has a partition with {} vectors of dimension {}. Arrow's fixed-size-list take kernel cannot remap partitions whose child array exceeds {} values. Recreate the index with default IVF partitioning, or enough partitions to keep each partition at or below {} vectors, before compacting.",
+                    segment.name,
+                    segment.uuid,
+                    partition_size,
+                    dimension,
+                    MAX_ARROW_FIXED_SIZE_LIST_CHILD_INDEX,
+                    max_partition_size,
+                ),
+            });
+        }
+    }
+
+    Ok(plan)
+}
 
 /// Optimize the dataset.
 ///
@@ -144,7 +287,7 @@ pub(crate) async fn cleanup_old_versions(
 /// This can be run after making several small appends to optimize the table
 /// for faster reads.
 ///
-/// This calls into [lance::dataset::optimize::compact_files].
+/// This calls into [lance::dataset::optimize::compact_files_with_planner].
 pub(crate) async fn compact_files_impl(
     table: &NativeTable,
     options: CompactionOptions,
@@ -152,7 +295,9 @@ pub(crate) async fn compact_files_impl(
 ) -> Result<CompactionMetrics> {
     table.dataset.ensure_mutable()?;
     let mut dataset = (*table.dataset.get().await?).clone();
-    let metrics = compact_files(&mut dataset, options, remap_options).await?;
+    let plan = validate_sq_index_remapping(&dataset, &options, remap_options.is_some()).await?;
+    let planner = PrecomputedCompactionPlanner { plan };
+    let metrics = compact_files_with_planner(&mut dataset, remap_options, &planner).await?;
     table.dataset.update(dataset);
     Ok(metrics)
 }
@@ -214,13 +359,14 @@ pub(crate) async fn execute_optimize(
 
 #[cfg(test)]
 mod tests {
+    use super::{oversized_sq_partition, segment_touches_fragments, sq_vector_dimension};
     use arrow_array::{
         Array, FixedSizeListArray, Float32Array, Int32Array, RecordBatch, StringArray,
     };
     use arrow_schema::{DataType, Field, Schema};
     use lance_arrow::FixedSizeListArrayExt;
     use rstest::rstest;
-    use std::sync::Arc;
+    use std::{collections::HashSet, sync::Arc};
 
     use crate::connect;
     use crate::database::listing::OPT_NEW_TABLE_ENABLE_STABLE_ROW_IDS;
@@ -230,6 +376,172 @@ mod tests {
     use crate::table::{CompactionOptions, OptimizeAction, OptimizeStats};
     use futures::TryStreamExt;
 
+    /// Regression test for https://github.com/lancedb/lancedb/issues/2866.
+    #[test]
+    fn test_detect_oversized_sq_partition() {
+        const DIMENSION: u64 = 4095;
+        let safe_size = u32::MAX as u64 / DIMENSION;
+
+        assert_eq!(
+            oversized_sq_partition([safe_size, safe_size + 1], DIMENSION),
+            Some(safe_size + 1)
+        );
+        assert_eq!(
+            oversized_sq_partition([safe_size / 2, safe_size / 2 + 1], DIMENSION),
+            None
+        );
+    }
+
+    #[test]
+    fn test_segment_touches_fragments_includes_unknown_coverage() {
+        let affected_fragments = HashSet::from([7]);
+
+        assert!(segment_touches_fragments(
+            Some([7_u32]),
+            &affected_fragments
+        ));
+        assert!(!segment_touches_fragments(
+            Some([8_u32]),
+            &affected_fragments
+        ));
+        assert!(segment_touches_fragments(
+            None::<[u32; 0]>,
+            &affected_fragments
+        ));
+    }
+
+    #[test]
+    fn test_sq_vector_dimension_includes_multivectors() {
+        const DIMENSION: i32 = 4095;
+        let vector = DataType::FixedSizeList(
+            Arc::new(Field::new("item", DataType::Float32, false)),
+            DIMENSION,
+        );
+        let multivector = DataType::List(Arc::new(Field::new("item", vector.clone(), false)));
+
+        assert_eq!(sq_vector_dimension(&vector), Some(DIMENSION));
+        assert_eq!(sq_vector_dimension(&multivector), Some(DIMENSION));
+        assert_eq!(sq_vector_dimension(&DataType::Float32), None);
+    }
+
+    #[tokio::test]
+    async fn test_compact_legacy_vector_index_with_unknown_fragment_coverage() {
+        use lance::Dataset;
+        use lance::index::vector::{IndexFileVersion, VectorIndexParams};
+        use lance::index::{DatasetIndexExt, DatasetIndexInternalExt};
+        use lance_index::IndexType;
+        use lance_linalg::distance::MetricType;
+        use lance_table::io::commit::write_manifest_file_to_path;
+        use object_store::ObjectStoreExt;
+
+        const INDEX_NAME: &str = "legacy_ivf_pq";
+        const ROWS: i32 = 128;
+        const DIMENSION: i32 = 8;
+
+        let tmpdir = tempfile::tempdir().unwrap();
+        let conn = connect(tmpdir.path().to_str().unwrap())
+            .execute()
+            .await
+            .unwrap();
+        let vectors = FixedSizeListArray::try_new_from_values(
+            Float32Array::from_iter_values((0..ROWS).flat_map(|row| {
+                (0..DIMENSION).map(move |offset| (row * DIMENSION + offset) as f32)
+            })),
+            DIMENSION,
+        )
+        .unwrap();
+        let schema = Arc::new(Schema::new(vec![Field::new(
+            "vector",
+            vectors.data_type().clone(),
+            false,
+        )]));
+        let batch = RecordBatch::try_new(schema, vec![Arc::new(vectors)]).unwrap();
+        let table = conn
+            .create_table("test_legacy_vector_compact", batch.clone())
+            .execute()
+            .await
+            .unwrap();
+        table.add(batch.clone()).execute().await.unwrap();
+        table.add(batch).execute().await.unwrap();
+        let mut dataset = (*table.dataset().unwrap().get().await.unwrap()).clone();
+        let mut params = VectorIndexParams::ivf_pq(1, 8, 1, MetricType::L2, 10);
+        params.version(IndexFileVersion::Legacy);
+        dataset
+            .create_index(
+                &["vector"],
+                IndexType::Vector,
+                Some(INDEX_NAME.to_string()),
+                &params,
+                true,
+            )
+            .await
+            .unwrap();
+        table.dataset().unwrap().update(dataset);
+
+        let dataset = table.dataset().unwrap().get().await.unwrap();
+        let mut segments = dataset.load_indices_by_name(INDEX_NAME).await.unwrap();
+        assert_eq!(segments.len(), 1);
+        let original = segments.pop().unwrap();
+        let field_path = dataset.schema().field_path(original.fields[0]).unwrap();
+        let vector_index = dataset
+            .open_vector_index(
+                &field_path,
+                &original.uuid,
+                &lance_index::metrics::NoOpMetricsCollector,
+            )
+            .await
+            .unwrap();
+        assert_eq!(vector_index.index_type(), IndexType::IvfPq);
+        assert_eq!(
+            vector_index.statistics().unwrap()["index_file_version"],
+            "Legacy"
+        );
+        let mut legacy = original.clone();
+        legacy.fragment_bitmap = None;
+        let object_store = dataset.object_store(None).await.unwrap();
+        let mut manifest = dataset.manifest().clone();
+        manifest.index_section = None;
+        manifest.transaction_section = None;
+        let manifest_path = dataset.manifest_location().path.clone();
+        object_store.inner.delete(&manifest_path).await.unwrap();
+        write_manifest_file_to_path(
+            object_store.as_ref(),
+            &mut manifest,
+            Some(vec![legacy]),
+            &manifest_path,
+            None,
+        )
+        .await
+        .unwrap();
+        let legacy_dataset = Dataset::open(dataset.uri()).await.unwrap();
+        table.dataset().unwrap().update(legacy_dataset);
+
+        let dataset = table.dataset().unwrap().get().await.unwrap();
+        let description_error = dataset
+            .describe_indices(None)
+            .await
+            .err()
+            .expect("legacy coverage should be unknown to index descriptions");
+        assert!(
+            description_error
+                .to_string()
+                .contains("Fragment bitmap is required")
+        );
+
+        let stats = table
+            .optimize(OptimizeAction::Compact {
+                options: CompactionOptions {
+                    target_rows_per_fragment: 1_000,
+                    ..Default::default()
+                },
+                remap_options: None,
+            })
+            .await
+            .unwrap();
+        assert!(stats.compaction.unwrap().fragments_removed > 0);
+        assert_eq!(table.count_rows(None).await.unwrap(), ROWS as usize * 3);
+    }
+
     #[tokio::test]
     async fn test_optimize_compact_simple() {
         let conn = connect("memory://").execute().await.unwrap();

rust/lancedb/tests/blob_integration.rs

@@ -10,7 +10,7 @@ use arrow_array::{
 use arrow_schema::{DataType, Field, Fields, Schema};
 use futures::TryStreamExt;
 use lance::Dataset;
-use lance_file::version::LanceFileVersion;
+use lance_encoding::version::LanceFileVersion;
 use lancedb::{
     Connection, Error, Result, Table,
     blob::{BlobRangeRequest, blob},