feat(node): accept ArrayBuffer, Blob/File, and URL as blob column inputs (#4313)

Closes #4270. ## Summary Blob v2 columns now accept more of the containers TypeScript code commonly has, both as the column value and inside `{ data }` / `{ uri }`: | Input | Handling | |---|---| | `ArrayBuffer` | `new Uint8Array(value)`, no copy | | `URL` | stored as `value.href` | | `Blob` / `File` | `await value.arrayBuffer()` in the async write paths: `createTable`, `add`, `mergeInsert` | `makeArrowTable` is synchronous, so it can't read a `Blob`. Given one, it throws an error that names the field and row and points to the async APIs or `new Uint8Array(await value.arrayBuffer())`. It does not write a null blob. ### Types `BlobInput` is now the accepted input type. It is widened to the types above and exported from the package entry point with `BlobData` and `BlobUri`, so callers can annotate rows and `tsc` catches bad inputs. The struct forms are a union that requires exactly one of `data` / `uri`. The old normalized `{ data, uri }` shape is renamed to `BlobValue` (internal). ### Implementation - `blob.ts`: `normalizeBlobInput` (sync: `ArrayBuffer`/`URL`), `resolveBlobInput` (async: `Blob`/`File`). `coerceBlobValue` normalizes first. - `arrow.ts`: `mapBlobInputs` walks only the blob positions in the schema, including blobs nested in structs and lists. It copies records and arrays only where a value changed, so caller data is never mutated. `resolveBlobInputs` runs before `makeArrowTable` in `convertToTable` (used by `add` / `mergeInsert`) and in `createTable`. Each distinct `Blob` is read once. - `makeArrowTable` now coerces every blob value to the full `{ data, uri }` struct **before** schema inference. This also fixes two bugs that existed before this change: - Rows mixing plain bytes with `{ data }` failed inference. - A `{ uri }`-only struct produced a blob struct missing its `data` child. Blob validation errors now include the field path and row. - Also fixed along the way: - `add` / `mergeInsert` on a table with an embedding function rejected most blob values, because `ensureNestedFieldsExist` filled blob columns in like ordinary structs. It now skips them. - With a schema, a struct field that is `null` in one row and a record in a later row failed inference. Primitives and arrays given for a struct field are now rejected instead of being written as an all-null struct. - Blob structs with keys other than `data` / `uri` are rejected, which catches typos. Class instances with `data` / `uri` are handled like plain objects. ## Testing - Unit tests in `__test__/blob.test.ts`: coercion of the new types, rejection and error messages for `Blob` in the sync path, nested lists and structs, `resolveBlobInputs` (only blob columns touched, caller data untouched, shared `Blob` read once). - Table tests in `__test__/table.test.ts`: - `createTable`, `add`, and `mergeInsert` round-trip `ArrayBuffer` / `Blob` / `File` bytes through `fetchBlobs`. - `URL` and `{ uri: URL }` are added to the external-blob round-trip test from #4317 (`allowExternalBlobOutsideBases`). - Full `jest` suite passes locally after the rebase (938 passed). `pnpm lint` is clean and `pnpm run docs` has been regenerated. Streaming inputs (`ReadableStream`, Node `Readable`, async iterables) remain out of scope, as the issue says. 🤖 Generated with [Claude Code](https://claude.com/claude-code) https://claude.ai/code/session_012U4KdHofT5edv3LcXHXbLv --------- Co-authored-by: Claude <noreply@anthropic.com>
2026-10-22 03:32:28 +00:00 · 2026-10-05 12:35:00 -07:00
parent 11b679a2db
commit a68c7be202
12 changed files with 879 additions and 38 deletions
@@ -790,6 +790,43 @@ describe.each(arrowVersions)("Arrow %#", (arrow: ApacheArrow) => {
      }
    });

+    it("will allow null values around struct values with a schema", function () {
+      const schema = new Schema([
+        new Field("value", new Struct([new Field("nested", new Int32())])),
+      ]);
+      for (const { records, nullIndex } of [
+        {
+          records: [{ value: null }, { value: { nested: 2 } }],
+          nullIndex: 0,
+        },
+        {
+          records: [{ value: { nested: 1 } }, { value: null }],
+          nullIndex: 1,
+        },
+      ]) {
+        const table = makeArrowTable(records, { schema });
+        const values = table.getChild("value");
+
+        expect(values?.nullCount).toBe(1);
+        expect(values?.get(nullIndex)).toBeNull();
+      }
+    });
+
+    it("will reject non-object values for struct fields in a schema", function () {
+      const schema = new Schema([
+        new Field("value", new Struct([new Field("nested", new Int32())])),
+      ]);
+      for (const records of [
+        [{ value: 5 }],
+        [{ value: 5 }, { value: { nested: 1 } }],
+        [{ value: { nested: 1 } }, { value: "oops" }],
+      ]) {
+        expect(() => makeArrowTable(records, { schema })).toThrow(
+          /Expected a struct value for field value at row \d, got/,
+        );
+      }
+    });
+
    it("will allow a schema to be provided", async function () {
      await checkTableCreation(
        async (records, _, schema) =>
@@ -1,9 +1,25 @@
 // SPDX-License-Identifier: Apache-2.0
 // SPDX-FileCopyrightText: Copyright The LanceDB Authors

-import { Field, Int64, List, Schema, Struct, Utf8 } from "apache-arrow";
-import { makeArrowTable } from "../lancedb/arrow";
+import {
+  Field,
+  Float,
+  Float32,
+  Int32,
+  Int64,
+  List,
+  Schema,
+  Struct,
+  Utf8,
+} from "apache-arrow";
+import {
+  convertToTable,
+  makeArrowTable,
+  resolveBlobInputs,
+} from "../lancedb/arrow";
 import { BlobFile, blob, coerceBlobValue, isBlobField } from "../lancedb/blob";
+import { EmbeddingFunction, LanceSchema } from "../lancedb/embedding";
+import { getRegistry } from "../lancedb/embedding/registry";

 describe("blob()", () => {
  it("marks the field as lance.blob.v2", () => {
@@ -46,6 +62,13 @@ describe("blob()", () => {
  });
 });

+class BlobStruct {
+  constructor(
+    public data?: unknown,
+    public uri?: unknown,
+  ) {}
+}
+
 describe("coerceBlobValue", () => {
  it.each([
    ["Buffer", Buffer.from("x"), { data: Buffer.from("x"), uri: null }],
@@ -65,6 +88,41 @@ describe("coerceBlobValue", () => {
      { uri: "s3://bucket/key" },
      { data: null, uri: "s3://bucket/key" },
    ],
+    [
+      "ArrayBuffer",
+      new Uint8Array([120]).buffer,
+      { data: new Uint8Array([120]), uri: null },
+    ],
+    [
+      "URL",
+      new URL("https://example.com/blob"),
+      { data: null, uri: "https://example.com/blob" },
+    ],
+    [
+      "data struct with ArrayBuffer",
+      { data: new Uint8Array([121]).buffer },
+      { data: new Uint8Array([121]), uri: null },
+    ],
+    [
+      "uri struct with URL",
+      { uri: new URL("file:///tmp/a.png") },
+      { data: null, uri: "file:///tmp/a.png" },
+    ],
+    [
+      "class instance with Uint8Array data",
+      new BlobStruct(new Uint8Array([122])),
+      { data: new Uint8Array([122]), uri: null },
+    ],
+    [
+      "class instance with ArrayBuffer data",
+      new BlobStruct(new Uint8Array([122]).buffer),
+      { data: new Uint8Array([122]), uri: null },
+    ],
+    [
+      "class instance with URL uri",
+      new BlobStruct(undefined, new URL("s3://bucket/key")),
+      { data: null, uri: "s3://bucket/key" },
+    ],
    ["null", null, null],
  ])("accepts %s", (_name, input, expected) => {
    expect(coerceBlobValue(input)).toEqual(expected);
@@ -76,7 +134,7 @@ describe("coerceBlobValue", () => {
    [
      "Int16Array",
      new Int16Array([1]),
-      /Blob data must be Buffer or Uint8Array/,
+      /Blob data must be Buffer, Uint8Array, ArrayBuffer, or Blob/,
    ],
    [
      "both data and uri",
@@ -88,11 +146,160 @@ describe("coerceBlobValue", () => {
      { data: null, uri: null },
      /exactly one of 'data' or 'uri'/,
    ],
+    ["Blob", new Blob(["x"]), /must be read asynchronously/],
+    ["data struct with Blob", { data: new Blob(["x"]) }, /asynchronously/],
+    [
+      "class instance with Blob data",
+      new BlobStruct(new Blob(["x"])),
+      /asynchronously/,
+    ],
+    [
+      "unknown struct field",
+      { data: Buffer.from("y"), mime: "image/png" },
+      /Blob struct values only support 'data' and 'uri', got 'mime'/,
+    ],
+    [
+      "misspelled struct field",
+      { uri: "s3://bucket/key", dat: Buffer.from("y") },
+      /got 'dat'/,
+    ],
+    [
+      "SharedArrayBuffer",
+      new SharedArrayBuffer(2),
+      /Blob column values must be Buffer, Uint8Array, ArrayBuffer, Blob/,
+    ],
+    ["Date", new Date(0), /Blob column values must be/],
  ])("rejects %s", (_name, input, message) => {
    expect(() => coerceBlobValue(input)).toThrow(message);
  });
 });

+describe("resolveBlobInputs", () => {
+  it("reads Blob and File values in blob columns only", async () => {
+    const schema = new Schema([
+      new Field("name", new Utf8(), true),
+      blob("image"),
+      new Field("images", new List(blob("image")), true),
+    ]);
+    const shared = new Blob(["same"]);
+    const notABlobColumn = new Blob(["other"]);
+    const data = [
+      {
+        name: notABlobColumn,
+        image: new File(["file"], "a.txt"),
+        images: [shared, { data: shared }, null],
+      },
+      { name: "x", image: Buffer.from("buf"), images: null },
+    ];
+    const resolved = await resolveBlobInputs(data, schema);
+
+    expect(resolved[0].name).toBe(notABlobColumn);
+    expect(Buffer.from(resolved[0].image as Uint8Array).toString()).toBe(
+      "file",
+    );
+    const images = resolved[0].images as [Uint8Array, { data: Uint8Array }];
+    expect(Buffer.from(images[0]).toString()).toBe("same");
+    expect(Buffer.from(images[1].data).toString()).toBe("same");
+    // Rows without Blob values and the caller's records are left alone.
+    expect(resolved[1]).toBe(data[1]);
+    expect(data[0].images?.[0]).toBe(shared);
+  });
+
+  it("reads each distinct Blob once, bare or inside data structs", async () => {
+    class CountingBlob extends Blob {
+      calls = 0;
+      async arrayBuffer(): Promise<ArrayBuffer> {
+        this.calls++;
+        return super.arrayBuffer();
+      }
+    }
+    const shared = new CountingBlob(["same"]);
+    const schema = new Schema([blob("image")]);
+    const resolved = await resolveBlobInputs(
+      [
+        { image: { data: shared } },
+        { image: { data: shared } },
+        { image: shared },
+      ],
+      schema,
+    );
+    expect(shared.calls).toBe(1);
+    const first = (resolved[0].image as { data: Uint8Array }).data;
+    expect(Buffer.from(first).toString()).toBe("same");
+    expect((resolved[1].image as { data: Uint8Array }).data).toBe(first);
+    expect(resolved[2].image).toBe(first);
+  });
+
+  it("reads a Blob held by a class instance", async () => {
+    const schema = new Schema([blob("image")]);
+    const [row] = await resolveBlobInputs(
+      [{ image: new BlobStruct(new Blob(["hi"])) }],
+      schema,
+    );
+    expect(
+      Buffer.from((row.image as { data: Uint8Array }).data).toString(),
+    ).toBe("hi");
+  });
+
+  it("returns the input when there is nothing to read", async () => {
+    const data = [{ image: Buffer.from("x") }];
+    const schema = new Schema([blob("image")]);
+    await expect(resolveBlobInputs(data, schema)).resolves.toBe(data);
+    await expect(resolveBlobInputs(data)).resolves.toBe(data);
+  });
+});
+
+describe("convertToTable blob columns with embedding functions", () => {
+  // Embedding metadata routes rows through ensureNestedFieldsExist, which must
+  // leave blob values for the blob coercion instead of treating them as structs.
+  class LengthEmbedding extends EmbeddingFunction<string> {
+    ndims() {
+      return 2;
+    }
+    embeddingDataType(): Float {
+      return new Float32();
+    }
+    async computeSourceEmbeddings(data: string[]) {
+      return data.map((text) => [text.length, 1]);
+    }
+  }
+  getRegistry().register("blob-test-length")(LengthEmbedding);
+  const func = getRegistry().get<LengthEmbedding>("blob-test-length")!.create();
+  const embeddingSchema = LanceSchema({
+    text: func.sourceField(new Utf8()),
+    vector: func.vectorField(),
+  });
+  const schema = new Schema(
+    [...embeddingSchema.fields, blob("image")],
+    embeddingSchema.metadata,
+  );
+
+  it.each([
+    ["Uint8Array", new Uint8Array([104, 105]), "hi", null],
+    ["ArrayBuffer", new Uint8Array([104, 105]).buffer, "hi", null],
+    ["Blob", new Blob(["hi"]), "hi", null],
+    ["URI string", "s3://b/k", null, "s3://b/k"],
+    ["URL", new URL("s3://b/k"), null, "s3://b/k"],
+    ["data struct", { data: new Uint8Array([104, 105]) }, "hi", null],
+  ])("accepts %s input", async (_name, image, data, uri) => {
+    const table = await convertToTable([{ text: "a", image }], undefined, {
+      schema,
+    });
+    const column = table.getChild("image")!;
+    expect(column.nullCount).toBe(0);
+    const bytes = column.getChild("data")!.get(0);
+    expect(bytes == null ? null : Buffer.from(bytes).toString()).toBe(data);
+    expect(column.getChild("uri")!.get(0)).toBe(uri);
+  });
+
+  it("writes a missing blob as null", async () => {
+    const table = await convertToTable([{ text: "a" }], undefined, { schema });
+    const column = table.getChild("image")!;
+    expect(column.nullCount).toBe(1);
+    expect(column.get(0)).toBeNull();
+  });
+});
+
 describe("BlobFile", () => {
  it("rejects constructing BlobFile without a native handle", () => {
    expect(() => new (BlobFile as unknown as { new (): BlobFile })()).toThrow(
@@ -103,26 +310,122 @@ describe("BlobFile", () => {

 describe("makeArrowTable blob columns", () => {
  it.each([
-    ["ArrayBuffer", new Uint8Array([104]).buffer],
-    ["Blob", new Blob(["hello"])],
-    ["File", new File(["hello"], "hello.txt")],
-    ["URL", new URL("https://example.com/blob")],
    ["ReadableStream", new ReadableStream()],
-    ["data struct with ArrayBuffer", { data: new Uint8Array([104]).buffer }],
-    ["uri struct with URL", { uri: new URL("https://example.com/blob") }],
+    ["Int16Array", new Int16Array([1])],
  ])("rejects %s input rather than writing a null blob", (_name, image) => {
    const schema = new Schema([blob("image")]);
    expect(() => makeArrowTable([{ image }], { schema })).toThrow(/image/);
  });

-  it("rejects Blob instances with enumerable fields as one invalid value", () => {
+  it.each([
+    ["Blob", new Blob(["hello"])],
+    ["File", new File(["hello"], "hello.txt")],
+    ["data struct with Blob", { data: new Blob(["hello"]) }],
+    [
+      "Blob with enumerable fields",
+      Object.assign(new Blob(["hello"]), { position: 0 }),
+    ],
+  ])("points %s input at the async APIs", (_name, image) => {
    const schema = new Schema([blob("image")]);
-    const image = Object.assign(new Blob(["hello"]), { position: 0 });
-    expect(() => makeArrowTable([{ image }], { schema })).toThrow(
-      "Unsupported object value for field image at row 0.",
+    expect(() => makeArrowTable([{ id: 1, image }], { schema })).toThrow(
+      /blob field image at row 0: Blob and File values must be read asynchronously/,
    );
  });

+  it.each([
+    ["ArrayBuffer", new Uint8Array([104, 105]).buffer, "hi", null],
+    [
+      "data struct with ArrayBuffer",
+      { data: new Uint8Array([104, 105]).buffer },
+      "hi",
+      null,
+    ],
+    ["URL", new URL("https://example.com/a"), null, "https://example.com/a"],
+    [
+      "uri struct with URL",
+      { uri: new URL("file:///tmp/a.png") },
+      null,
+      "file:///tmp/a.png",
+    ],
+  ])("accepts %s input", (_name, image, data, uri) => {
+    const schema = new Schema([blob("image")]);
+    const input = { image };
+    const table = makeArrowTable([input], { schema });
+    const column = table.getChild("image")!;
+    expect(column.nullCount).toBe(0);
+    const bytes = column.getChild("data")!.get(0);
+    expect(bytes == null ? null : Buffer.from(bytes).toString()).toBe(data);
+    expect(column.getChild("uri")!.get(0)).toBe(uri);
+    // The caller's record is left untouched.
+    expect(input.image).toBe(image);
+  });
+
+  it.each([
+    ["Uint8Array", new Uint8Array([104, 105]), "hi", null],
+    ["URI string", "s3://b/k", null, "s3://b/k"],
+    ["data struct", { data: new Uint8Array([104, 105]) }, "hi", null],
+  ])("accepts %s input after a null row", (_name, image, data, uri) => {
+    const schema = new Schema([blob("image")]);
+    const table = makeArrowTable([{ image: null }, { image }], { schema });
+    const column = table.getChild("image")!;
+    expect(column.nullCount).toBe(1);
+    expect(column.get(0)).toBeNull();
+    const bytes = column.getChild("data")!.get(1);
+    expect(bytes == null ? null : Buffer.from(bytes).toString()).toBe(data);
+    expect(column.getChild("uri")!.get(1)).toBe(uri);
+  });
+
+  it("accepts a null struct before one holding a blob", () => {
+    const schema = new Schema([
+      new Field("info", new Struct([blob("image")]), true),
+    ]);
+    const table = makeArrowTable(
+      [{ info: null }, { info: { image: new Uint8Array([97]) } }],
+      { schema },
+    );
+    const info = table.getChild("info")!;
+    expect(info.get(0)).toBeNull();
+    const image = (info.get(1) as { image: { data: Uint8Array } }).image;
+    expect(Buffer.from(image.data).toString()).toBe("a");
+  });
+
+  it("skips a missing blob column named like an Object.prototype member", () => {
+    const schema = new Schema([
+      new Field("id", new Int32(), true),
+      blob("constructor"),
+    ]);
+    const table = makeArrowTable([{ id: 1 }], { schema });
+    expect(table.schema.fields.map((f) => f.name)).toEqual(["id"]);
+  });
+
+  it("accepts ArrayBuffer and URL inside lists and structs", () => {
+    const schema = new Schema([
+      new Field("images", new List(blob("image")), true),
+      new Field("info", new Struct([blob("image")]), true),
+    ]);
+    const table = makeArrowTable(
+      [
+        {
+          images: [new Uint8Array([97]).buffer, new URL("s3://b/k")],
+          info: { image: new Uint8Array([98]).buffer },
+        },
+      ],
+      { schema },
+    );
+    const images = Array.from(
+      table.getChild("images")!.get(0) as Iterable<{
+        data: Uint8Array | null;
+        uri: string | null;
+      }>,
+    );
+    expect(Buffer.from(images[0].data!).toString()).toBe("a");
+    expect(images[1].uri).toBe("s3://b/k");
+    const info = table.getChild("info")!.get(0) as {
+      image: { data: Uint8Array };
+    };
+    expect(Buffer.from(info.image.data).toString()).toBe("b");
+  });
+
  it("coerces Buffer input onto a blob field", () => {
    const schema = new Schema([
      new Field("id", new Int64(), true),
@@ -9,6 +9,7 @@ import * as tmp from "tmp";

 import {
  AutoQuery,
+  type BlobInput,
  Connection,
  MatchQuery,
  PhraseQuery,
@@ -2583,7 +2584,7 @@ describe("when dealing with blob columns", () => {
      new Field("id", new Int64(), true),
      blob("image"),
    ]);
-    const row = { id: 1n, image: new Uint8Array([104]).buffer };
+    const row = { id: 1n, image: new ReadableStream() };

    await expect(db.createTable("invalid", [row], { schema })).rejects.toThrow(
      /field image at row 0/,
@@ -2613,6 +2614,8 @@ describe("when dealing with blob columns", () => {
  it.each([
    ["URI string", (uri: string) => uri],
    ["URI struct", (uri: string) => ({ uri })],
+    ["URL", (uri: string) => new URL(uri)],
+    ["URL struct", (uri: string) => ({ uri: new URL(uri) })],
  ])(
    "round-trips an external blob from a %s after reopening",
    async (_label, blobValue) => {
@@ -2706,6 +2709,63 @@ describe("when dealing with blob columns", () => {
    expect(bytes[0]).toEqual(payload);
  });

+  it("accepts ArrayBuffer, Blob, and File in createTable and add", async () => {
+    const db = await connect(tmpDir.name);
+    const schema = new Schema([
+      new Field("id", new Int64(), true),
+      blob("image"),
+    ]);
+    const rows: { id: bigint; image: BlobInput }[] = [
+      { id: 1n, image: new TextEncoder().encode("array-buffer").buffer },
+      { id: 2n, image: new Blob(["blob"]) },
+      { id: 3n, image: { data: new File(["file"], "f.txt") } },
+    ];
+    const table = await db.createTable("widened", rows.slice(0, 2), {
+      schema,
+    });
+    await table.add(rows.slice(2));
+
+    const results = await table.query().select(["id"]).withRowId().toArray();
+    results.sort((a, b) => Number(a.id - b.id));
+    const bytes = await table.fetchBlobs(
+      "image",
+      results.map((row) => row._rowid as bigint),
+    );
+    expect(bytes.map((b) => b?.toString())).toEqual([
+      "array-buffer",
+      "blob",
+      "file",
+    ]);
+  });
+
+  it("accepts Blob values in mergeInsert", async () => {
+    const db = await connect(tmpDir.name);
+    const schema = new Schema([
+      new Field("id", new Int64(), true),
+      blob("image"),
+    ]);
+    const table = await db.createTable(
+      "merge_blobs",
+      [{ id: 1n, image: Buffer.from("old") }],
+      { schema },
+    );
+    await table
+      .mergeInsert("id")
+      .whenMatchedUpdateAll()
+      .whenNotMatchedInsertAll()
+      .execute([
+        { id: 1n, image: new Blob(["new"]) },
+        { id: 2n, image: new Uint8Array([104, 105]).buffer },
+      ]);
+    const results = await table.query().select(["id"]).withRowId().toArray();
+    results.sort((a, b) => Number(a.id - b.id));
+    const bytes = await table.fetchBlobs(
+      "image",
+      results.map((row) => row._rowid as bigint),
+    );
+    expect(bytes.map((b) => b?.toString())).toEqual(["new", "hi"]);
+  });
+
  it("discovers blob columns", async () => {
    const { table } = await openBlobTable();
    expect(await table.blobColumns()).toEqual(["image"]);