Datasets:
CROSS3: Protocol Bridge — JSON to Message Translation
Goal
Fix 4 translation bugs and 2 error mapping bugs in the bridge service that translates Service A's JSON REST API responses into structured messages for Service B's consumer.
Requirements
- int64 fields must be translated as Python int (no truncation to 32-bit)
- bytes fields must be base64-decoded from JSON strings
- oneof fields must have exactly one variant set (not multiple)
- enum fields must be mapped from string names to integer values
- HTTP 404 from Service A must map to NOT_FOUND error code (5) for Service B
- HTTP 429 from Service A must map to RESOURCE_EXHAUSTED error code (8) for Service B
- All tests must pass:
pytest tests/
Supporting Documents
service_a/models.py— JSON data models from Service Aservice_b/schema.py— Message schema for Service B (proto3-style)bridge/translator.py— JSON→Message translation (4 bugs)bridge/error_mapper.py— HTTP status → error code (2 bugs)
Background
The bridge service sits between Service A (a REST API returning JSON) and Service B (a queue consumer expecting structured proto3-style messages). Because JSON and proto3 have different type semantics, every field crossing this boundary needs a careful type conversion.
Type Semantic Differences
| JSON Type | Proto3 Type | Issue |
|---|---|---|
| number | int64 | JSON numbers lose precision for values > 2^53; must not be masked to 32-bit |
| string | bytes | Binary data is base64-encoded in JSON; must be decoded to bytes |
| object | oneof | JSON may include multiple keys; proto3 oneof allows exactly one |
| string | enum | Enum names in JSON must be converted to integer codes |
Error Code Mapping
Service A returns HTTP status codes. Service B uses gRPC-style integer error codes:
| HTTP Status | Expected Error Code | Code Number |
|---|---|---|
| 200 | OK | 0 |
| 400 | INVALID_ARGUMENT | 3 |
| 401/403 | INVALID_ARGUMENT | 3 |
| 404 | NOT_FOUND | 5 |
| 429 | RESOURCE_EXHAUSTED | 8 |
| 5xx | INTERNAL | 13 |
Real-World Context
JSON-to-gRPC transcoding bridges are a common integration layer in microservice architectures, and the 6 bugs in this task reflect bugs documented in real transcoding implementations:
- int64 truncation (Bug 1): JavaScript's
JSON.parse()silently truncates 64-bit integers beyondNumber.MAX_SAFE_INTEGER(2^53-1). This caused data loss in the Twitter Snowflake ID migration (2010) and is documented in the gRPC-Gateway project FAQ. Fix: serialize int64 as strings in JSON. - bytes base64 handling (Bug 2): Proto3
bytesfields must be base64-encoded in JSON (RFC 4648). The gRPC-Gateway library had exactly this bug in v1.x — binary fields were passed through raw, corrupting payloads. CVE-class issue in several open-source gRPC-JSON bridges. - oneof variant mapping (Bug 3): Proto3
oneoffields in JSON use the field name as the key, not the oneof name. Incorrect mapping silently drops the value. - enum integer vs string (Bug 4): Proto3 JSON canonical form uses enum names (strings), not integers. Several gRPC web clients (grpc-web, grpc-gateway) have shipped this mismatch causing 0-value enums for all inputs.
- Error code mapping (Bugs 5–6): gRPC status codes (google.rpc.Code) are a strict mapping from HTTP status; 404→NOT_FOUND(5) and 429→RESOURCE_EXHAUSTED(8) are specified in google.rpc.Code and commonly misimplemented.