Add C# csm-tcp-router-client SDK

[Copilot]

• Create SDK/csharp/ directory mirroring SDK/python/
• Implement single-file SDK src/CsmTcpRouter/CsmTcpRouter.cs
• Add CsmTcpRouter.csproj library project (netstandard2.0 + net8.0, NuGet metadata)
• Add xUnit test project tests/CsmTcpRouter.Tests/
• Add examples/BasicUsage/ console example
• Add VS solution file CsmTcpRouter.sln
• Add README.md, README.zh-cn.md, CHANGELOG.md, LICENSE
• Add .github/workflows/CSharp_SDK.yml
• Update .gitignore for C# build artifacts
• Build and run all tests locally
• Address review feedback:
  • Replace flaky port-1 assumption with GetClosedPort() helper that binds a TcpListener to port 0, captures the OS-assigned port, then stops the listener
  • Replace Thread.Sleep in UnsubscribeStatus_RemovesCallback with ManualResetEventSlim.Wait(150) assertion
  • WaitForServerAsync now awaits the connect task in both success and timeout-loss paths so faults are observed; proactively closes the probe when the delay wins so the in-flight connect attempt is cancelled
  • WaitForRespAsync and WaitForCmdRespAsync now force _transport.Disconnect() on timeout so a late RESP / CMD_RESP from the timed-out command can no longer be misattributed to the next request (protocol v0 has no correlation id)

[Copilot]

Pull request overview

Adds a new C#/.NET client SDK under SDK/csharp/ that mirrors the existing Python client’s protocol and feature surface, including a single-file implementation, tests, example app, packaging metadata, and CI workflow.

Changes:

• Introduces the single-file .NET client implementation (TcpRouterClient, protocol codec, models, exceptions, transport).
• Adds xUnit protocol + integration tests with an in-process loopback MockServer.
• Adds docs, runnable console example, NuGet packaging metadata, solution file, CI workflow, and .gitignore updates.

Reviewed changes

Copilot reviewed 14 out of 15 changed files in this pull request and generated 5 comments.

Show a summary per file

File	Description
SDK/csharp/src/CsmTcpRouter/CsmTcpRouter.cs	Core single-file SDK: protocol codec, transport, client APIs, models, exceptions
SDK/csharp/src/CsmTcpRouter/CsmTcpRouter.csproj	Multi-targeted NuGet packaging metadata and deterministic CI build settings
SDK/csharp/tests/CsmTcpRouter.Tests/ProtocolTests.cs	Protocol codec + model/error parsing unit tests
SDK/csharp/tests/CsmTcpRouter.Tests/MockServer.cs	Loopback TCP mock server used by integration tests
SDK/csharp/tests/CsmTcpRouter.Tests/ClientIntegrationTests.cs	End-to-end tests for connect, commands, subscriptions, callbacks, timeouts, disconnect behavior
SDK/csharp/tests/CsmTcpRouter.Tests/CsmTcpRouter.Tests.csproj	Test project dependencies + reference to the SDK project
SDK/csharp/examples/BasicUsage/Program.cs	Runnable example showing typical client flows
SDK/csharp/examples/BasicUsage/BasicUsage.csproj	Example project referencing the SDK
SDK/csharp/README.md	English SDK documentation and API overview
SDK/csharp/README.zh-cn.md	Chinese SDK documentation and API overview
SDK/csharp/CHANGELOG.md	Initial changelog entry for v0.1.0
SDK/csharp/LICENSE	MIT license for the C# SDK distribution
SDK/csharp/CsmTcpRouter.sln	Solution grouping src/tests/examples
.github/workflows/CSharp_SDK.yml	Cross-OS build/test + pack + tag-gated NuGet publish workflow
.gitignore	Ignore .NET build artifacts and NuGet outputs

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

These tests assume TCP port 1 is closed to trigger connection/availability failures. On some machines/containers that port can be open, which would make the tests flaky. Consider allocating an unused port more deterministically (e.g., bind a TcpListener to port 0 to obtain a free port, stop it, then test against that port) or use an address/port combination guaranteed to fail in the test environment.

[Copilot]

WaitForServerAsync starts TcpClient.ConnectAsync and uses Task.WhenAny with a 1s delay, but it never awaits/observes the connect task when it faults or times out. This can leave unobserved exceptions and potentially keep an in-flight connect attempt running until the socket is disposed. Consider awaiting the connect task in both success and failure paths (to observe exceptions) and proactively closing the probe socket when the delay wins (or using a CancellationToken/WaitAsync timeout) so each retry is cleanly cancelled.

Suggested change

	varct=probe.ConnectAsync(host,port);
	varwinner=awaitTask.WhenAny(ct,Task.Delay(TimeSpan.FromSeconds(1))).ConfigureAwait(false);
	if(winner==ct&&!ct.IsFaulted)
	{
	try{probe.Close();}catch{/* ignore */}
	returntrue;
	}
	varconnectTask=probe.ConnectAsync(host,port);
	varwinner=awaitTask.WhenAny(connectTask,Task.Delay(TimeSpan.FromSeconds(1))).ConfigureAwait(false);
	if(winner==connectTask)
	{
	awaitconnectTask.ConfigureAwait(false);
	try{probe.Close();}catch{/* ignore */}
	returntrue;
	}
	try{probe.Close();}catch{/* ignore */}
	try
	{
	awaitconnectTask.ConfigureAwait(false);
	}
	catch(SocketException){/* not ready yet */}
	catch(IOException){/* not ready yet */}
	catch(ObjectDisposedException){/* connect aborted by closing probe */}

[Copilot]

If WaitForRespAsync times out, the connection remains open and subsequent SendAndWait calls can consume a late RESP from the previous request (there is no correlation ID in the protocol, and OnPacket always completes the current _respTcs). This can cause callers to receive the wrong response for a later command. A safer approach is to treat a RESP timeout as a protocol desync and force a Disconnect() (or otherwise prevent further requests until reconnect) so late frames cannot be misattributed.

[Copilot]

WaitForCmdRespAsync has the same desync risk as WaitForRespAsync: after a cmd-resp timeout the client stays connected, and a late CMD_RESP from the prior request can complete the next in-flight _cmdRespTcs, making a later Post/Subscribe appear to succeed incorrectly. Consider disconnecting (or marking the connection unusable) on CMD_RESP timeouts to avoid misattributing late frames.

[Copilot]

This test uses Thread.Sleep to assert that a callback is not invoked after UnsubscribeStatus. Sleep-based timing assertions are inherently flaky across CI environments. Prefer using a synchronization primitive (e.g., ManualResetEventSlim) and asserting it is not signaled within a timeout window, or push multiple status messages and assert the counter remains unchanged with a bounded wait loop.

Suggested change

	client.SubscribeStatus("Status","AI", _ =>Interlocked.Increment(refhits),DefaultTimeout);
	client.UnsubscribeStatus("Status","AI",DefaultTimeout);
	server.PushStatus("Status >> v1 <- AI");
	Thread.Sleep(150);
	usingvarev=newManualResetEventSlim();
	client.SubscribeStatus("Status","AI", _ =>{Interlocked.Increment(refhits);ev.Set();},DefaultTimeout);
	client.UnsubscribeStatus("Status","AI",DefaultTimeout);
	server.PushStatus("Status >> v1 <- AI");
	Assert.False(ev.Wait(150),"callback was invoked after unsubscribe");

[nevstop]

@copilot apply changes based on the comments in this thread

[Copilot]

@copilot apply changes based on the comments in this thread

Addressed all five review comments in 050485e:

• Replaced flaky port-1 usage in Connect_BadPort_Throws and WaitForServer_ReturnsFalseOnTimeout with a GetClosedPort() helper that binds a TcpListener to port 0, captures the OS-assigned port, then stops the listener.
• Reworked WaitForServerAsync to await the connect task in both success and delay-wins paths so faults are observed; the probe socket is closed proactively when the delay wins to abort the in-flight connect attempt.
• WaitForRespAsync and WaitForCmdRespAsync now force _transport.Disconnect() on timeout so a late RESP/CMD_RESP from the timed-out command cannot be misattributed to the next request (protocol v0 has no correlation id).
• UnsubscribeStatus_RemovesCallback now uses ManualResetEventSlim.Wait(150) instead of Thread.Sleep.

All 51 tests still pass.