feat(xmtp_api_d14n): XmtpMlsBidiStreams trait + v3 Subscribe transport

[tylerhawkes]

Stack 3/4 of the XIP-83 bidi client lane: #3769 → #3770 → #3771 → #3772.

Native-only XmtpMlsBidiStreams trait in xmtp_proto::api_client — subscribe_bidi(BoxStream<SubscribeRequest>) -> Stream<Result<SubscribeResponse>> — and its V3Client implementation: prost-encode outbound frames over bidi_stream("/xmtp.mls.api.v1.MlsApi/Subscribe"), decode inbound via the existing XmtpStream. Browsers stay on XmtpMlsStreams + watchdog (gRPC-Web cannot full-duplex).

🤖 Generated with Claude Code

Note

Add XmtpMlsBidiStreams trait and bidirectional streaming support to the v3 Subscribe transport

• Introduces the XmtpMlsBidiStreams trait in xmtp_proto (native-only) defining a subscribe_bidi method for full-duplex MLS subscriptions per XIP-83.
• Implements subscribe_bidi for V3Client in bidi.rs, encoding outbound SubscribeRequest frames and decoding inbound SubscribeResponse frames over the underlying Client::bidi_stream.
• Adds bidi_stream to the Client trait with a default implementation returning "not supported"; GrpcClient overrides this on native targets using tonic bidirectional streaming.
• Propagates bidi_stream through all middleware layers: AuthMiddleware, MultiNodeClient, ReadWriteClient, ReadonlyClient, and BoxedClient.
• Behavioral Change: ReadonlyClient now enforces write-deny checks on bidirectional streaming paths; wasm targets always return an unsupported error.

^{Macroscope summarized c793213.}

[claude]

Claude finished @tylerhawkes's task —— View job

---

Code Review Completed

Review Date: June 25, 2026

Summary

This PR successfully introduces the XIP-83 bidirectional streaming capability for native-only targets. The author has already addressed multiple rounds of feedback, including the critical AuthMiddleware forwarding issue mentioned in the most recent comment. The implementation is thorough and follows existing patterns well.

✅ What's Good

1. Consistent middleware implementation — All middleware wrappers (AuthMiddleware, ReadonlyClient, ReadWriteClient, MultiNodeClient) now properly forward bidi_stream with appropriate semantics (auth/readonly/routing, etc.), mirroring their stream implementations (crates/xmtp_api_d14n/src/middleware/auth.rs:185-197)

2. Proper blanket impls — Box<T>, Arc<T>, and &T forwarders for both Client::bidi_stream and XmtpMlsBidiStreams ensure type erasure works seamlessly (crates/xmtp_proto/src/traits.rs:199-206, crates/xmtp_proto/src/api_client/impls.rs:407-443)

3. Native-only gating — Appropriate use of if_native! prevents compilation on platforms without full-duplex support, with clear documentation explaining the limitation (crates/xmtp_api_grpc/src/grpc_client/client.rs:196-224)

4. Comprehensive test coverage — The bidi transport implementation has solid unit tests verifying encoding/decoding, error handling, and middleware forwarding (crates/xmtp_api_d14n/src/queries/v3/bidi.rs:52-175, crates/xmtp_api_d14n/src/middleware/auth.rs:291-312)

5. RPC instrumentation — The #[xmtp_common::rpc_span] annotation on subscribe_bidi provides observability at the appropriate boundary layer (crates/xmtp_api_d14n/src/queries/v3/bidi.rs:29)

📊 Test Coverage Note

Codecov reports 72.56% patch coverage with 31 lines missing. Most uncovered lines are:

• Default trait implementations (expected — only exercised when a concrete type lacks full-duplex support)
• Error path in traits.rs:157-159 (the "unsupported transport" default)
• Some blanket impl forwarding paths (also expected — these are fallback code paths)

This is acceptable given these are defensive/fallback code paths that are difficult to meaningfully test.

🟢 Verdict

Ready to merge. All previous review comments have been addressed, the implementation is sound, and test coverage is appropriate for the changes. No blocking issues remain.

---

[codecov]

Codecov Report

❌ Patch coverage is 69.56522% with 42 lines in your changes missing coverage. Please review.
✅ Project coverage is 84.44%. Comparing base (1e1a195) to head (c793213).
⚠️ Report is 4 commits behind head on main.

Files with missing lines	Patch %	Lines
crates/xmtp_api_grpc/src/grpc_client/client.rs	22.22%	14 Missing ⚠️
crates/xmtp_proto/src/traits.rs	0.00%	8 Missing ⚠️
crates/xmtp_api_d14n/src/middleware/auth.rs	77.77%	4 Missing ⚠️
crates/xmtp_proto/src/api_client/impls.rs	0.00%	4 Missing ⚠️
...es/xmtp_api_d14n/src/middleware/readonly_client.rs	0.00%	3 Missing ⚠️
crates/xmtp_api_d14n/src/queries/v3/bidi.rs	96.29%	3 Missing ⚠️
...pi_d14n/src/middleware/multi_node_client/client.rs	0.00%	2 Missing ⚠️
...pi_d14n/src/middleware/read_write_client/client.rs	0.00%	2 Missing ⚠️
crates/xmtp_proto/src/traits/boxed_client.rs	0.00%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #3771      +/-   ##
==========================================
- Coverage   84.44%   84.44%   -0.01%     
==========================================
  Files         408      409       +1     
  Lines       59880    60013     +133     
==========================================
+ Hits        50567    50675     +108     
- Misses       9313     9338      +25     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[macroscopeapp]

Approvability

Verdict: Needs human review

This PR introduces new bidirectional streaming infrastructure (XIP-83 trait and v3 Subscribe transport) which represents a new feature capability. While the implementation is well-structured and the author owns all changed files, new features introducing new communication patterns warrant human review.

^{You can customize Macroscope's approvability policy. Learn more.}

[cameronvoell]

bidi finally! i'll leave the claude comment about test coverage up to your discretion 🙏

Dismissing prior approval to re-evaluate b65f00e