[Request] Add a CBOR codec to smithy-python, for wire parity with smithy-java

[humanzz]

Ask

Add a CBOR codec to smithy-python, mirroring the existing JSON codec (JSONCodec in smithy-json) and wire-compatible with smithy-java's Rpcv2CborCodec. The JSON half already exists in both languages; this is the missing CBOR half on the Python side, so shapes generated from one Smithy model can be serde'd in both Java and Python and stay wire-compatible.

Use case: one Smithy model, codegen'd into Java and Python (SageMaker endpoint invocation)

SageMaker endpoint invocation crosses a language boundary: the model server is Python, and the clients calling it are Java and Python. The goal is to model the request/response shapes once in Smithy, codegen typed classes into both languages, and serde them with matching codecs — CBOR on the wire (compact, binary, schema-driven), JSON when readability matters. The requirement is wire compatibility: bytes written by the Java codec have to deserialize with the Python codec, and vice versa.

Java today: smithy-java

A generated shape round-trips through CBOR or JSON by swapping the codec — serde is schema-driven, nothing hand-written:

import software.amazon.smithy.java.cbor.Rpcv2CborCodec;
import software.amazon.smithy.java.json.JsonCodec;
import software.amazon.smithy.java.core.serde.Codec;
import java.nio.ByteBuffer;

// `request` is a code-generated smithy-java shape (a SerializableShape).

// --- CBOR (RPC v2 CBOR) ---
Codec cbor = Rpcv2CborCodec.builder().build();
ByteBuffer encoded = cbor.serialize(request);
MyShape decoded = cbor.deserializeShape(encoded, MyShape.builder());

// --- JSON (same shapes, same API) ---
Codec json = JsonCodec.builder().build();
ByteBuffer encodedJson = json.serialize(request);
MyShape decodedJson = json.deserializeShape(encodedJson, MyShape.builder());

Python today (smithy-python): JSON only

JSONCodec implements the format-agnostic Codec protocol from smithy_core.codecs:

from smithy_json import JSONCodec

codec = JSONCodec()
encoded: bytes = codec.serialize(request)      # `request` is a code-generated shape
decoded = codec.deserialize(encoded, MyShape)

Python already has the generic Codec protocol (smithy-core) and concrete JSON (smithy-json) and XML (smithy-xml) codecs — but no CBOR codec (no smithy-cbor package).

[jonathan343]

Hey @humanzz!

Thanks for your interest in smithy python and for opening this issue.

To be fully transparent, our current approach has been to implement features that will help us unblock and expand our AWS service coverage. Below is the protocol distribution for all modeled AWS services (might be missing a few):

rest-json: 254
json: 143
query: 17
smithy-rpc-v2-cbor: 9
rest-xml: 4
ec2: 1

Total services: 428

JSON-based (rest-json & json) protocols apply to the most services, XML-based (query, rest-xml, and ec2) protocols are next, and lastly CBOR-based protocol (smithy-rpc-v2-cbor) is what remains. There are still non-codec level changes to get protocols like json and rest-xml unblocked which is what will get prioritized. However, once that's done CBOR will become the main focus.

I'll leave this issue open to track interest and future updates. Thanks again!