chore: merge main into probe validation repair

docs/src/js/classes/AutoQuery.md

@@ -417,6 +417,8 @@ Collect the results as an Arrow
 
 ArrowTable.
 
+Empty results retain the query's output schema.
+
 #### Inherited from
 
 `StandardQueryBase.toArrow`

docs/src/js/classes/Query.md

@@ -491,6 +491,8 @@ Collect the results as an Arrow
 
 ArrowTable.
 
+Empty results retain the query's output schema.
+
 #### Inherited from
 
 `StandardQueryBase.toArrow`

docs/src/js/classes/QueryBase.md

@@ -252,6 +252,8 @@ Collect the results as an Arrow
 
 ArrowTable.
 
+Empty results retain the query's output schema.
+
 ***
 
 ### withRowId()

docs/src/js/classes/TakeQuery.md

@@ -267,6 +267,8 @@ Collect the results as an Arrow
 
 ArrowTable.
 
+Empty results retain the query's output schema.
+
 #### Inherited from
 
 [`QueryBase`](QueryBase.md).[`toArrow`](QueryBase.md#toarrow)

docs/src/js/classes/VectorQuery.md

@@ -742,6 +742,8 @@ Collect the results as an Arrow
 
 ArrowTable.
 
+Empty results retain the query's output schema.
+
 #### Inherited from
 
 `StandardQueryBase.toArrow`

nodejs/__test__/query.test.ts

@@ -110,6 +110,129 @@ describe("Query outputSchema", () => {
   });
 });
 
+describe("Query empty results", () => {
+  let tmpDir: tmp.DirResult;
+  let table: Table;
+
+  beforeEach(async () => {
+    tmpDir = tmp.dirSync({ unsafeCleanup: true });
+    const db = await connect(tmpDir.name);
+    const schema = new Schema(
+      [
+        new Field("id", new Int64(), false),
+        new Field(
+          "text",
+          new Utf8(),
+          true,
+          new Map([["description", "searchable text"]]),
+        ),
+        new Field(
+          "vector",
+          new FixedSizeList(2, new Field("item", new Float32())),
+          true,
+        ),
+      ],
+      new Map([["source", "empty-query-regression"]]),
+    );
+    table = await db.createTable(
+      "test",
+      makeArrowTable([{ id: 1n, text: "hello", vector: [1, 2] }], { schema }),
+    );
+  });
+
+  afterEach(() => {
+    tmpDir.removeCallback();
+  });
+
+  it.each([
+    {
+      name: "filtered scan",
+      query: (table: Table) => table.query().where("id < 0"),
+      fields: ["id", "text", "vector"],
+    },
+    {
+      name: "zero limit",
+      query: (table: Table) => table.query().limit(0),
+      fields: ["id", "text", "vector"],
+    },
+    {
+      name: "column projection",
+      query: (table: Table) =>
+        table.query().where("id < 0").select(["text", "id"]),
+      fields: ["text", "id"],
+    },
+    {
+      name: "dynamic projection",
+      query: (table: Table) =>
+        table.query().where("id < 0").select({ doubled: "id * 2" }),
+      fields: ["doubled"],
+    },
+    {
+      name: "filtered vector search",
+      query: (table: Table) =>
+        table.search([0.5, 0.5]).where("id < 0").limit(3),
+      fields: ["id", "text", "vector", "_distance"],
+    },
+    {
+      name: "vector projection with row id",
+      query: (table: Table) =>
+        table
+          .vectorSearch([0.5, 0.5])
+          .where("id < 0")
+          .select(["id"])
+          .withRowId(),
+      fields: ["id", "_distance", "_rowid"],
+    },
+    {
+      name: "fast search without an index",
+      query: (table: Table) => table.search([0.5, 0.5]).fastSearch(),
+      fields: ["id", "text", "vector", "_rowid", "_distance"],
+    },
+    {
+      name: "take with no offsets",
+      query: (table: Table) => table.takeOffsets([]).select(["text"]),
+      fields: ["text"],
+    },
+  ])("preserves the output schema for $name", async ({ query, fields }) => {
+    const result = await query(table).toArrow({ maxBatchLength: 1 });
+
+    expect(result.numRows).toBe(0);
+    expect(result.schema.fields.map((field) => field.name)).toEqual(fields);
+    for (const name of fields) {
+      expect(result.getChild(name)).not.toBeNull();
+      expect(result.getChild(name)?.length).toBe(0);
+    }
+  });
+
+  it("preserves types, nullability, and metadata", async () => {
+    const result = await table.query().where("id < 0").toArrow();
+    const schema = await table.schema();
+
+    expect(result.schema).toEqual(schema);
+    expect(result.getChild("id")?.type.toString()).toBe("Int64");
+    expect(result.schema.fields[0].nullable).toBe(false);
+    expect(result.schema.fields[1].metadata.get("description")).toBe(
+      "searchable text",
+    );
+    expect(result.schema.metadata.get("source")).toBe("empty-query-regression");
+  });
+
+  it("preserves the score column for full-text search with no matches", async () => {
+    await table.createIndex("text", { config: Index.fts() });
+    const result = await table
+      .search("missing", "fts")
+      .select(["id"])
+      .toArrow();
+
+    expect(result.numRows).toBe(0);
+    expect(result.schema.fields.map((field) => field.name)).toEqual([
+      "id",
+      "_score",
+    ]);
+    expect(result.getChild("_score")?.type.toString()).toBe("Float32");
+  });
+});
+
 describe("Search pagination", () => {
   let tmpDir: tmp.DirResult;
   let table: Table;

nodejs/__test__/remote.test.ts

@@ -14,6 +14,14 @@ import {
   TlsConfig,
   connect,
 } from "../lancedb";
+import {
+  Table as ArrowTable,
+  Field,
+  Int64,
+  RecordBatch,
+  Schema,
+  tableToIPC,
+} from "../lancedb/arrow";
 import {
   HeaderProvider,
   OAuthHeaderProvider,
@@ -117,6 +125,66 @@ describe("remote connection", () => {
     expect(requests).toEqual(["/v1/table/test/describe/"]);
   });
 
+  it.each([false, true])(
+    "preserves an empty query's schema with an empty batch: %s",
+    async (withEmptyBatch) => {
+      const schema = new Schema(
+        [new Field("doubled", new Int64(), false)],
+        new Map([["source", "query-output"]]),
+      );
+      const result = new ArrowTable(
+        schema,
+        withEmptyBatch ? [new RecordBatch(schema, undefined)] : [],
+      );
+      const response = Buffer.from(tableToIPC(result, "stream"));
+      let queryRequests = 0;
+
+      await withMockDatabase(
+        (req, res) => {
+          if (req.url?.endsWith("/describe/")) {
+            res.writeHead(200, { "Content-Type": "application/json" }).end(
+              JSON.stringify({
+                name: "items",
+                version: 1,
+                schema: {
+                  fields: [
+                    { name: "id", type: { type: "int64" }, nullable: false },
+                  ],
+                },
+              }),
+            );
+          } else if (req.url?.endsWith("/query/")) {
+            queryRequests++;
+            req.resume();
+            req.on("end", () => {
+              res
+                .writeHead(200, {
+                  "Content-Type": "application/vnd.apache.arrow.stream",
+                })
+                .end(response);
+            });
+          } else {
+            res.writeHead(404).end();
+          }
+        },
+        async (db) => {
+          const table = await db.openTable("items");
+          const result = await table
+            .query()
+            .where("id < 0")
+            .select({ doubled: "id * 2" })
+            .toArrow();
+
+          expect(result.numRows).toBe(0);
+          expect(result.schema).toEqual(schema);
+          expect(result.getChild("doubled")?.length).toBe(0);
+        },
+      );
+
+      expect(queryRequests).toBe(1);
+    },
+  );
+
   it("rejects the external blob opt-in without blocking regular adds", async () => {
     let insertRequests = 0;
     const describeRequests: string[] = [];

nodejs/lancedb/query.ts

@@ -41,26 +41,6 @@ export async function* RecordBatchIterator(
   }
 }
 
-class RecordBatchIterable<
-  NativeQueryType extends NativeQuery | NativeVectorQuery | NativeTakeQuery,
-> implements AsyncIterable<RecordBatch>
-{
-  private inner: NativeQueryType;
-  private options?: QueryExecutionOptions;
-
-  constructor(inner: NativeQueryType, options?: QueryExecutionOptions) {
-    this.inner = inner;
-    this.options = options;
-  }
-
-  // biome-ignore lint/suspicious/noExplicitAny: skip
-  [Symbol.asyncIterator](): AsyncIterator<RecordBatch<any>, any, undefined> {
-    return RecordBatchIterator(
-      this.inner.execute(this.options?.maxBatchLength, this.options?.timeoutMs),
-    );
-  }
-}
-
 /**
  * Options that control the behavior of a particular query execution
  */
@@ -271,13 +251,20 @@ export class QueryBase<
     return RecordBatchIterator(this.nativeExecute());
   }
 
-  /** Collect the results as an Arrow @see {@link ArrowTable}. */
+  /**
+   * Collect the results as an Arrow @see {@link ArrowTable}.
+   *
+   * Empty results retain the query's output schema.
+   */
   async toArrow(options?: Partial<QueryExecutionOptions>): Promise<ArrowTable> {
     const batches = [];
-    const inner = await this.getInner();
-    for await (const batch of new RecordBatchIterable(inner, options)) {
+    const iterator = this.nativeExecute(options);
+    for await (const batch of RecordBatchIterator(iterator)) {
       batches.push(batch);
     }
+    if (batches.length === 0) {
+      return tableFromIPC((await iterator).schema());
+    }
     return new ArrowTable(batches);
   }
 

nodejs/src/iterator.rs

@@ -7,6 +7,8 @@ use lancedb::ipc::batches_to_ipc_file;
 use napi::bindgen_prelude::*;
 use napi_derive::napi;
 
+use crate::util::schema_to_buffer;
+
 /** Typescript-style Async Iterator over RecordBatches */
 #[napi]
 pub struct RecordBatchIterator {
@@ -19,6 +21,12 @@ impl RecordBatchIterator {
         Self { inner }
     }
 
+    /// Return the executed stream's output schema as an Arrow IPC file.
+    #[napi]
+    pub fn schema(&self) -> napi::Result<Buffer> {
+        schema_to_buffer(&self.inner.schema())
+    }
+
     #[napi(catch_unwind)]
     pub async unsafe fn next(&mut self) -> napi::Result<Option<Buffer>> {
         if let Some(rst) = self.inner.next().await {

python/python/lancedb/db.py

@@ -122,7 +122,48 @@ def _view_description(
 
 
 class DBConnection(EnforceOverrides):
-    """An active LanceDB connection interface."""
+    """An active LanceDB connection interface.
+
+    Use [close][lancedb.db.DBConnection.close] to release the connection's
+    underlying resources, or use the connection as a context manager to close it
+    automatically when leaving the block, including when an exception is raised.
+
+    Examples
+    --------
+    >>> import lancedb
+    >>> with lancedb.connect("memory://") as db:
+    ...     assert db.is_open()
+    >>> db.is_open()
+    False
+    """
+
+    def __enter__(self) -> DBConnection:
+        return self
+
+    def __exit__(self, *_) -> None:
+        self.close()
+
+    @abstractmethod
+    def is_open(self) -> bool:
+        """Return True if the connection is open."""
+        pass
+
+    @abstractmethod
+    def close(self) -> None:
+        """Close the connection, releasing any underlying resources.
+
+        It is safe to call this method multiple times. Database operations on a
+        closed connection raise ``RuntimeError: Connection is closed``.
+
+        Examples
+        --------
+        >>> import lancedb
+        >>> db = lancedb.connect("memory://")
+        >>> db.close()
+        >>> db.is_open()
+        False
+        """
+        pass
 
     def list_namespaces(
         self,
@@ -1275,6 +1316,14 @@ class LanceDBConnection(DBConnection):
     def __repr__(self) -> str:
         return f"{self.__class__.__name__}(uri={self._conn.uri!r})"
 
+    @override
+    def is_open(self) -> bool:
+        return self._conn.is_open()
+
+    @override
+    def close(self) -> None:
+        self._conn.close()
+
     @override
     def serialize(self) -> str:
         import json

python/python/lancedb/namespace.py

@@ -489,6 +489,14 @@ class LanceNamespaceDBConnection(DBConnection):
             )
         self._uri = self._inner.uri
 
+    @override
+    def is_open(self) -> bool:
+        return self._inner.is_open()
+
+    @override
+    def close(self) -> None:
+        self._inner.close()
+
     @override
     def serialize(self) -> str:
         import json

python/python/lancedb/remote/db.py

@@ -1054,6 +1054,10 @@ class RemoteDBConnection(DBConnection):
         """
         return LOOP.run(self._conn.namespace_client())
 
-    async def close(self):
-        """Close the connection to the database."""
+    @override
+    def is_open(self) -> bool:
+        return self._conn.is_open()
+
+    @override
+    def close(self) -> None:
         self._conn.close()

python/python/lancedb/table.py

@@ -4332,9 +4332,9 @@ class LanceTable(Table):
             )
             if storage_options is None:
                 storage_options = {}
-            storage_options["new_table_enable_v2_manifest_paths"] = (
+            storage_options["new_table_enable_v2_manifest_paths"] = str(
                 enable_v2_manifest_paths
-            )
+            ).lower()
 
         self._table = LOOP.run(
             self._conn._conn.create_table(
@@ -4766,7 +4766,7 @@ class LanceTable(Table):
         [LanceTable.uses_v2_manifest_paths][lancedb.table.LanceTable.uses_v2_manifest_paths]
         to check if the table is already using the new path style.
         """
-        LOOP.run(self._table.migrate_v2_manifest_paths())
+        LOOP.run(self._table.migrate_manifest_paths_v2())
 
     @deprecation.deprecated(
         deprecated_in="0.33.1",
@@ -6107,7 +6107,17 @@ class AsyncTable:
     @overload
     async def search(
         self,
-        query: Optional[str] = None,
+        query: None = None,
+        vector_column_name: Optional[str] = None,
+        query_type: QueryType = "auto",
+        ordering_field_name: Optional[str] = None,
+        fts_columns: Optional[Union[str, List[str]]] = None,
+    ) -> AsyncQuery: ...
+
+    @overload
+    async def search(
+        self,
+        query: str,
         vector_column_name: Optional[str] = None,
         query_type: Literal["auto"] = ...,
         ordering_field_name: Optional[str] = None,
@@ -6117,7 +6127,7 @@ class AsyncTable:
     @overload
     async def search(
         self,
-        query: Optional[str] = None,
+        query: str,
         vector_column_name: Optional[str] = None,
         query_type: Literal["hybrid"] = ...,
         ordering_field_name: Optional[str] = None,
@@ -6127,7 +6137,7 @@ class AsyncTable:
     @overload
     async def search(
         self,
-        query: Optional[Union[VEC, "PIL.Image.Image", Tuple]] = None,
+        query: Union[VEC, "PIL.Image.Image", Tuple],
         vector_column_name: Optional[str] = None,
         query_type: Literal["auto"] = ...,
         ordering_field_name: Optional[str] = None,
@@ -6137,7 +6147,7 @@ class AsyncTable:
     @overload
     async def search(
         self,
-        query: Optional[str] = None,
+        query: str,
         vector_column_name: Optional[str] = None,
         query_type: Literal["fts"] = ...,
         ordering_field_name: Optional[str] = None,
@@ -6147,9 +6157,7 @@ class AsyncTable:
     @overload
     async def search(
         self,
-        query: Optional[
-            Union[VEC, str, "PIL.Image.Image", Tuple, FullTextQuery]
-        ] = None,
+        query: Union[VEC, str, "PIL.Image.Image", Tuple, FullTextQuery],
         vector_column_name: Optional[str] = None,
         query_type: Literal["vector"] = ...,
         ordering_field_name: Optional[str] = None,
@@ -6165,7 +6173,7 @@ class AsyncTable:
         query_type: QueryType = "auto",
         ordering_field_name: Optional[str] = None,
         fts_columns: Optional[Union[str, List[str]]] = None,
-    ) -> Union[AsyncHybridQuery, AsyncFTSQuery, AsyncVectorQuery]:
+    ) -> Union[AsyncQuery, AsyncHybridQuery, AsyncFTSQuery, AsyncVectorQuery]:
         """Create a search query to find the nearest neighbors
         of the given query vector. We currently support [vector search](https://lancedb.com/docs/search/vector-search/)
         and [full-text search](https://lancedb.com/docs/search/full-text-search/).
@@ -6180,8 +6188,9 @@ class AsyncTable:
             - *default None*.
             Acceptable types are: list, np.ndarray, PIL.Image.Image
 
-            - If None then the select/where/limit clauses are applied to filter
-            the table
+            - If None then a plain [AsyncQuery][lancedb.query.AsyncQuery] is
+            returned, equivalent to calling [query][lancedb.table.AsyncTable.query].
+            The select/where/limit clauses are applied to filter the table.
         vector_column_name: str, optional
             The name of the vector column to search.
 
@@ -6209,10 +6218,13 @@ class AsyncTable:
 
         Returns
         -------
-        LanceQueryBuilder
+        AsyncQuery, AsyncHybridQuery, AsyncFTSQuery, or AsyncVectorQuery
             A query builder object representing the query.
         """
 
+        if query is None:
+            return self.query()
+
         def is_embedding(query):
             return isinstance(query, (list, np.ndarray, pa.Array, pa.ChunkedArray))
 

python/python/tests/test_db.py

@@ -2,6 +2,7 @@
 # SPDX-FileCopyrightText: Copyright The LanceDB Authors
 
 
+import contextlib
 import inspect
 import re
 import sys
@@ -405,6 +406,49 @@ async def test_connect(tmp_path):
     assert str(db) == f"ListingDatabase(uri={tmp_path}, read_consistency_interval=5s)"
 
 
+@pytest.fixture(params=["memory", "local", "namespace", "from_inner"])
+def sync_db(request, tmp_path):
+    if request.param == "memory":
+        return lancedb.connect("memory://")
+    if request.param == "namespace":
+        return lancedb.connect_namespace("dir", {"root": str(tmp_path)})
+    db = lancedb.connect(tmp_path)
+    if request.param == "from_inner":
+        return lancedb.db.LanceDBConnection.from_inner(db._inner, None)
+    return db
+
+
+def test_sync_close(sync_db):
+    assert sync_db.is_open()
+    assert sync_db.close() is None
+    assert not sync_db.is_open()
+    assert sync_db.close() is None
+
+    with pytest.warns(DeprecationWarning, match="table_names"):
+        with pytest.raises(RuntimeError, match="Connection is closed"):
+            sync_db.table_names()
+    with pytest.raises(RuntimeError, match="Connection is closed"):
+        sync_db.list_tables()
+    with pytest.raises(RuntimeError, match="Connection is closed"):
+        sync_db.open_table("test")
+
+
+@pytest.mark.parametrize("raise_error", [False, True])
+def test_sync_context_manager(sync_db, raise_error):
+    with contextlib.ExitStack() as stack:
+        if raise_error:
+            stack.enter_context(pytest.raises(ValueError, match="test error"))
+        with sync_db as db:
+            assert db is sync_db
+            assert db.is_open()
+            if raise_error:
+                raise ValueError("test error")
+
+    assert not sync_db.is_open()
+    with pytest.raises(RuntimeError, match="Connection is closed"):
+        sync_db.list_tables()
+
+
 @pytest.mark.asyncio
 async def test_close(mem_db_async: lancedb.AsyncConnection):
     assert mem_db_async.is_open()
@@ -495,6 +539,27 @@ async def test_create_exist_ok_async(tmp_db_async: lancedb.AsyncConnection):
     #     await db.create_table("test", schema=bad_schema, exist_ok=True)
 
 
+@pytest.mark.parametrize("enable_v2_manifest_paths", [False, True])
+@pytest.mark.parametrize("empty", [False, True])
+def test_create_table_deprecated_v2_manifest_paths(
+    tmp_path, enable_v2_manifest_paths, empty
+):
+    db = lancedb.connect(tmp_path)
+    data = None if empty else [{"id": 1}]
+
+    with pytest.warns(DeprecationWarning, match="enable_v2_manifest_paths"):
+        table = db.create_table(
+            "test",
+            data=data,
+            schema=pa.schema([("id", pa.int64())]),
+            enable_v2_manifest_paths=enable_v2_manifest_paths,
+        )
+
+    assert table.uses_v2_manifest_paths() == enable_v2_manifest_paths
+    assert table.to_arrow().to_pylist() == ([] if empty else data)
+    assert db.open_table("test").uses_v2_manifest_paths() == enable_v2_manifest_paths
+
+
 @pytest.mark.asyncio
 async def test_create_table_v2_manifest_paths_async(tmp_path):
     db_with_v2_paths = await lancedb.connect_async(

python/python/tests/test_query.py

@@ -24,6 +24,7 @@ from lancedb.pydantic import LanceModel, Vector
 from lancedb.query import (
     AsyncFTSQuery,
     AsyncHybridQuery,
+    AsyncQuery,
     AsyncQueryBase,
     AsyncVectorQuery,
     ColumnOrdering,
@@ -1304,6 +1305,27 @@ async def test_query_to_polars_async(table_async: AsyncTable):
     assert df.shape == (0, num_columns)
 
 
+@pytest.mark.asyncio
+@pytest.mark.parametrize("search_kwargs", [{}, {"query": None}])
+@pytest.mark.parametrize("with_vector", [False, True])
+async def test_async_search_without_query(mem_db_async, search_kwargs, with_vector):
+    data = [{"id": i} for i in range(10)]
+    if with_vector:
+        for row in data:
+            row["vector"] = [float(row["id"]), 1.0]
+    table = await mem_db_async.create_table("test", data)
+
+    query = await table.search(**search_kwargs)
+    assert isinstance(query, AsyncQuery)
+    assert await query.limit(3).to_arrow() == await table.query().limit(3).to_arrow()
+
+    query = await table.search(**search_kwargs)
+    assert await query.where("id >= 8").select(["id"]).limit(3).to_list() == [
+        {"id": 8},
+        {"id": 9},
+    ]
+
+
 @pytest.mark.asyncio
 async def test_none_query(table_async: AsyncTable):
     with pytest.raises(ValueError):

python/python/tests/test_remote_db.py

@@ -17,7 +17,7 @@ from packaging.version import Version
 
 import lancedb
 from lancedb.conftest import MockTextEmbeddingFunction
-from lancedb.query import ColumnOrdering
+from lancedb.query import AsyncQuery, ColumnOrdering
 from lancedb.remote import ClientConfig
 from lancedb.remote.errors import HttpError, RetryError
 import pytest
@@ -1521,6 +1521,46 @@ def test_query_sync_empty_query():
         assert data == expected
 
 
+@pytest.mark.asyncio
+@pytest.mark.parametrize("search_kwargs", [{}, {"query": None}])
+async def test_async_search_without_query(search_kwargs):
+    expected = pa.table({"id": [7, 8, 9]})
+
+    def handler(request):
+        if request.path == "/v1/table/test/describe/":
+            request.send_response(200)
+            request.send_header("Content-Type", "application/json")
+            request.end_headers()
+            request.wfile.write(b'{"version": 1, "schema": {"fields": []}}')
+        elif request.path == "/v1/table/test/query/":
+            body = json.loads(
+                request.rfile.read(int(request.headers["Content-Length"]))
+            )
+            assert body == {
+                "k": 3,
+                "filter": "id >= 7",
+                "vector": [],
+                "columns": ["id"],
+                "prefilter": True,
+                "version": None,
+            }
+            request.send_response(200)
+            request.send_header("Content-Type", "application/vnd.apache.arrow.file")
+            request.end_headers()
+            with pa.ipc.new_file(request.wfile, schema=expected.schema) as writer:
+                writer.write_table(expected)
+        else:
+            request.send_response(404)
+            request.end_headers()
+
+    async with mock_lancedb_connection_async(handler) as db:
+        table = await db.open_table("test")
+        query = await table.search(**search_kwargs)
+        assert isinstance(query, AsyncQuery)
+        result = await query.where("id >= 7").select(["id"]).limit(3).to_arrow()
+        assert result == expected
+
+
 def test_query_sync_maximal():
     def handler(body):
         assert body == {
@@ -2192,15 +2232,46 @@ async def test_header_provider_overrides_static_headers():
 
 
 def test_close():
-    """Test that close() works without AttributeError."""
-    import asyncio
-
     def handler(req):
         req.send_response(200)
         req.end_headers()
 
     with mock_lancedb_connection(handler) as db:
-        asyncio.run(db.close())
+        assert db.close() is None
+        assert not db.is_open()
+        assert db.close() is None
+
+        with pytest.warns(DeprecationWarning, match="table_names"):
+            with pytest.raises(RuntimeError, match="Connection is closed"):
+                db.table_names()
+        with pytest.raises(RuntimeError, match="Connection is closed"):
+            db.list_tables()
+        with pytest.raises(RuntimeError, match="Connection is closed"):
+            db.open_table("test")
+
+
+@pytest.mark.parametrize("raise_error", [False, True])
+def test_sync_context_manager(raise_error):
+    def handler(req):
+        req.send_response(200)
+        req.send_header("Content-Type", "application/json")
+        req.end_headers()
+        req.wfile.write(b'{"tables": []}')
+
+    with mock_lancedb_connection(handler) as db:
+        with contextlib.ExitStack() as stack:
+            if raise_error:
+                stack.enter_context(pytest.raises(ValueError, match="test error"))
+            with db as conn:
+                assert conn is db
+                assert conn.is_open()
+                assert conn.list_tables().tables == []
+                if raise_error:
+                    raise ValueError("test error")
+
+        assert not db.is_open()
+        with pytest.raises(RuntimeError, match="Connection is closed"):
+            db.list_tables()
 
 
 @pytest.mark.parametrize("exception", [KeyboardInterrupt, SystemExit, GeneratorExit])

python/python/tests/test_table.py

@@ -113,6 +113,32 @@ def test_basic(mem_db: DBConnection):
     assert table.to_arrow() == expected_data
 
 
+@pytest.mark.parametrize("enable_v2", [False, True])
+def test_migrate_v2_manifest_paths(tmp_path, enable_v2):
+    db = lancedb.connect(
+        tmp_path,
+        storage_options={"new_table_enable_v2_manifest_paths": str(enable_v2).lower()},
+    )
+    table = db.create_table("calls", [{"id": 1, "vector": [1.0, 1.0]}])
+    table.add([{"id": 2, "vector": [2.0, 2.0]}])
+    expected_data = table.to_arrow()
+    expected_versions = table.list_versions()
+    assert table.uses_v2_manifest_paths() == enable_v2
+
+    # Migration is also safe to repeat on a table already using v2 paths.
+    for _ in range(2):
+        table.migrate_v2_manifest_paths()
+        assert table.uses_v2_manifest_paths()
+        reopened = db.open_table("calls")
+        assert reopened.uses_v2_manifest_paths()
+        assert reopened.to_arrow() == expected_data
+        assert reopened.list_versions() == expected_versions
+
+    manifests = list((tmp_path / "calls.lance" / "_versions").glob("*.manifest"))
+    assert len(manifests) == len(expected_versions)
+    assert all(len(path.stem) == 20 and path.stem.isdigit() for path in manifests)
+
+
 def test_search_preserves_nulls_from_sliced_arrow_table(mem_db: DBConnection):
     data = pa.table(
         {

rust/lancedb/src/remote/table.rs

@@ -3101,7 +3101,13 @@ impl<S: HttpSend> BaseTable for RemoteTable<S> {
         // loop can re-execute the plan (and re-stream the body) on each retry.
         // This mirrors the old `send_streaming(with_retry=true)` path, which
         // likewise buffered the reader to support retries.
-        let batches = new_data.collect::<std::result::Result<Vec<_>, _>>()?;
+        let schema = RecordBatchReader::schema(new_data.as_ref());
+        let mut batches = new_data.collect::<std::result::Result<Vec<_>, _>>()?;
+        // An empty reader still carries a schema. Keep it in an empty batch so
+        // the buffered source remains scannable and can be replayed on retries.
+        if batches.is_empty() {
+            batches.push(RecordBatch::new_empty(schema));
+        }
         let source: Box<dyn Scannable> = Box::new(batches);
         let rescannable = source.rescannable();
         let input: Arc<dyn ExecutionPlan> =
@@ -4801,6 +4807,104 @@ mod tests {
         }
     }
 
+    #[rstest]
+    #[case::no_batches_insert(false, false)]
+    #[case::empty_batch_insert(true, false)]
+    #[case::no_batches_delete(false, true)]
+    #[case::empty_batch_delete(true, true)]
+    #[tokio::test]
+    async fn test_merge_insert_empty_source(
+        #[case] has_batch: bool,
+        #[case] delete_unmatched: bool,
+    ) {
+        let mut fields = vec![Field::new("id", DataType::Int64, false)];
+        if !delete_unmatched {
+            fields.extend([
+                Field::new("k", DataType::Int64, false),
+                Field::new(
+                    "vector",
+                    DataType::FixedSizeList(
+                        Arc::new(Field::new("item", DataType::Float32, true)),
+                        2,
+                    ),
+                    true,
+                ),
+                Field::new("s", DataType::Utf8, true),
+            ]);
+        }
+        let schema = Arc::new(Schema::new_with_metadata(
+            fields,
+            HashMap::from([("source".to_string(), "empty".to_string())]),
+        ));
+        let batches = if has_batch {
+            vec![Ok(RecordBatch::new_empty(schema.clone()))]
+        } else {
+            vec![]
+        };
+        let data: Box<dyn RecordBatchReader + Send> =
+            Box::new(RecordBatchIterator::new(batches, schema.clone()));
+        let attempts = Arc::new(AtomicUsize::new(0));
+        let attempts_ref = attempts.clone();
+        let num_deleted_rows = if delete_unmatched { 3 } else { 0 };
+
+        let table = Table::new_with_handler("my_table", move |request| {
+            assert_eq!(request.method(), "POST");
+            assert_eq!(request.url().path(), "/v1/table/my_table/merge_insert/");
+            assert_eq!(request.headers()[CONTENT_TYPE], ARROW_STREAM_CONTENT_TYPE);
+            let params = request.url().query_pairs().collect::<HashMap<_, _>>();
+            assert_eq!(params["on"], "id");
+            assert_eq!(
+                params["when_not_matched_insert_all"],
+                (!delete_unmatched).to_string()
+            );
+            assert_eq!(
+                params["when_not_matched_by_source_delete"],
+                delete_unmatched.to_string()
+            );
+
+            let body = request.body().unwrap().as_bytes().unwrap();
+            let reader = StreamReader::try_new(Cursor::new(body), None).unwrap();
+            assert_eq!(reader.schema(), schema);
+            for batch in reader {
+                assert_eq!(batch.unwrap().num_rows(), 0);
+            }
+
+            // The empty source must retain its schema when replayed after a conflict.
+            if attempts_ref.fetch_add(1, Ordering::SeqCst) == 0 {
+                http::Response::builder()
+                    .status(409)
+                    .body(String::new())
+                    .unwrap()
+            } else {
+                http::Response::builder()
+                    .status(200)
+                    .body(
+                        json!({
+                            "version": 43,
+                            "num_deleted_rows": num_deleted_rows,
+                            "num_inserted_rows": 0,
+                            "num_updated_rows": 0,
+                        })
+                        .to_string(),
+                    )
+                    .unwrap()
+            }
+        });
+
+        let mut merge = table.merge_insert(&["id"]);
+        if delete_unmatched {
+            merge.when_not_matched_by_source_delete(None);
+        } else {
+            merge.when_not_matched_insert_all();
+        }
+        let result = merge.execute(data).await.unwrap();
+        assert_eq!(attempts.load(Ordering::SeqCst), 2);
+        assert_eq!(result.version, 43);
+        assert_eq!(result.num_deleted_rows, num_deleted_rows);
+        assert_eq!(result.num_inserted_rows, 0);
+        assert_eq!(result.num_updated_rows, 0);
+    }
+
     #[tokio::test]
     async fn test_merge_insert_composite_key() {
         let batch = RecordBatch::try_new(