lutzroeder/netron · error · Error

Expected 'backend_name'.

Error message

Expected 'backend_name'.

What it means

Generated binary decoder for caffe2.BackendOptions rejects data missing the required 'backend_name' field. The verifier runs after the field loop; if the wire bytes never carried field 1 (backend_name), it throws this error.

Source

Thrown at source/caffe2-proto.js:1132

    static decode(reader, length) {
        const message = new caffe2.BackendOptions();
        const end = length === undefined ? reader.length : reader.position + length;
        while (reader.position < end) {
            const tag = reader.uint32();
            switch (tag >>> 3) {
                case 1:
                    message.backend_name = reader.string();
                    break;
                case 2:
                    message.option.push(caffe2.MapFieldEntry.decode(reader, reader.uint32()));
                    break;
                default:
                    reader.skipType(tag & 7);
                    break;
            }
        }
        if (!Object.prototype.hasOwnProperty.call(message, 'backend_name')) {
            throw new Error("Expected 'backend_name'.");
        }
        return message;
    }

    static decodeText(reader) {
        const message = new caffe2.BackendOptions();
        reader.start();
        while (!reader.end()) {
            const tag = reader.tag();
            switch (tag) {
                case "backend_name":
                    message.backend_name = reader.string();
                    break;
                case "option":
                    message.option.push(caffe2.MapFieldEntry.decodeText(reader));
                    break;
                default:
                    reader.field(tag, message);

View on GitHub (pinned to d8a543f5f8)

Solutions

1. Slice the sub-buffer using the correct field length from the parent message rather than guessing offsets.
2. Regenerate caffe2-proto.js from the .proto matching the data producer.
3. Prefer decoding the parent PartitionInfo and accessing .backend_options instead of manual slicing.
4. Sanity-check buffer length > 0 before decoding.

Example fix

// before
const opts = caffe2.BackendOptions.decode(buf); // throws Expected 'backend_name'

// after
const info = caffe2.PartitionInfo.decode(fullBuf);
const opts = info.backend_options; // decoded in context, offsets handled internally

if (sub.length===0 || (sub[0]>>3)!==1) throw new RangeError('buffer does not start at BackendOptions.backend_name');

const hasBackendName = (o) => o != null && typeof o.backend_name === 'string';

try { opts = caffe2.BackendOptions.decode(sub); } catch (e) { if (/Expected 'backend_name'/.test(e.message)) { decode parent instead; } else throw e; }

When it happens

Trigger: BackendOptions.decode() on bytes missing the backend_name field — wrong buffer offset when slicing from a parent PartitionInfo message, empty buffers, or schema drift between proto versions.

Common situations: Parsing PARTITIONING backend options embedded in caffe2 plan/model files; extracting sub-messages with incorrect subarray bounds; decoding data from a different caffe2 build.

Related errors

• Expected 'begin'.
• Expected 'end'.
• Expected 'precision'.
• Expected 'scale'.
• Expected 'bias'.

AI-assisted analysis of lutzroeder/netron@d8a543f5f8 (2026-08-27). Data as JSON: /api/errors/231e2ee84996e72b. Report an issue: GitHub.