fix(nodejs): add better error handling when missing embedding functions

[universalmind303]

note:
running the default lint command npm run lint -- --fix seems to have made a lot of unrelated changes.

[universalmind303]

this is the relevant code change. Every thing else is just from running the linter

[universalmind303]

this is the relevant code change. Every thing else is just from running the linter

[universalmind303]

@wjones127 @westonpace, I'm unable to reproduce the CI failures locally. Could one of you take a look and see if you are able to reproduce.

[westonpace]

@wjones127 @westonpace, I'm unable to reproduce the CI failures locally. Could one of you take a look and see if you are able to reproduce.

@universalmind303 It does reproduce locally for me. It appears that it's this line:

    if (data[0][field.name] === undefined) {

I suspect data[0] is undefined.

[westonpace]

I might be reading this wrong but does this prevent a user from providing a fixed-size-list column directly without using the embeddings function? In other words, if a user is calculating their own embeddings (not using lancedb to do it) will it still work?

Also, any change added to node/src/arrow.ts is probably appropriate for nodejs/src/arrow.ts.

[universalmind303]

I might be reading this wrong but does this prevent a user from providing a fixed-size-list column directly without using the embeddings function? In other words, if a user is calculating their own embeddings (not using lancedb to do it) will it still work?

Users can still provide their own embeddings manually without using the embed functionality. It will only error out if there is a vector field in the schema with no matching implementation (manual or embed function)

Example:

     const schema = new Schema([
        new Field("id", new Int32()),
        new Field("text", new Utf8()),
        new Field(
          "vector",
          new FixedSizeList(2, new Field("item", new Float32(), true))
        ),
      ]);

user provides schema and manual embeddings (OK)

      const data = [
        { id: 1, text: "foo", vector: [0.1, 0.2] },
        { id: 2, text: "bar", vector: [0.3, 0.4] },
      ];
      let table = await con.createTable({
        name: "test",
        data,
        schema,
      });

user provides schema and embedding function (OK)

      const embeddingFunction = {...}
      const data = [
        { id: 1, text: "foo"},
        { id: 2, text: "bar"},
      ];
      let table = await con.createTable({
        name: "test",
        data,
        schema,
        embeddingFunction
      });

User provides schema, but no embeddings or function (Not OK)

      const data = [
        { id: 1, text: "foo"},
        { id: 2, text: "bar"},
      ];
      let table = await con.createTable({
        name: "test",
        data,
        schema
      });

[westonpace]

Got it, looks good.