[ROBO-5747] fix: don't log failed-accept Error on server teardown

[eduard-dumitru]

Disposing an IpcServer cancels the accept loop; on Linux a parked slot's pipe can surface as a non-OCE IOException ("Pipe is broken") rather than an OperationCanceledException, which was logged as "Failed to accept new connection". Suppress it when cancellation is already requested (genuine errors still log). Includes a real-pipe, Linux red->green regression test.

Test Results (on CI-RobotIntegration-Test)

PR #126 teardown-noise verification — coreipc IpcServer accept loop

Artifact: RobotLogs (per build). Search string in Robot.log: Failed to accept new connection
Full line (before fix): Failed to accept new connection. Ex: System.IO.IOException: Pipe is broken.

Build	Fix?	Leg	File in RobotLogs/	Matches	First 3 line #s
12439546	✅ with fix	Linux: Build	Robot.log	0	— (no matches)
12439546	✅ with fix	Docker: Linux	Robot-*.log	0	— (no matches)
12439546	✅ with fix	Mac: Build	stage still in progress — no logs published	n/a	—
12439546	✅ with fix	Windows ×3 / Serverless	Robot.log	0	—
12436764	❌ no fix	Linux: Build	Robot.log	2675	2466, 2472, 2478
12436764	❌ no fix	Docker: Linux	Robot-7f2da42b-c12b-4e07-8da5-247649cc0c57.log	2035	231, 237, 243
12436764	❌ no fix	Mac: Build	Robot.log	3125	2513, 2519, 2525
12436764	❌ no fix	Windows ×3 / Serverless	Robot.log	0	—

How to verify: open the file in the leg's RobotLogs zip, Find Failed to accept new connection →
before-fix legs hit it thousands of times (jump to the line #s above); fixed Linux/Docker legs return 0 matches.
Noise is Unix-only (Windows = 0 in both), matching the root cause PR #126 guards.

Test Results (on dev machine) — 20 runs per cell (80 total)

	Linux (WSL Debian, net6.0)	Windows (net6.0)
with fix	✅ 20 / 20 pass	✅ 20 / 20 pass
without fix	❌ 0 / 20 fail (expected)	✅ 20 / 20 pass

Findings

• Deterministic on Linux. Without the fix the test fails on every run (20/20), reproducing
  the exact production stack trace (IOException: Pipe is broken at
  NamedPipeServerTransport.AwaitConnection → IpcServer.Accepter.TryAccept). With the fix it
  passes on every run. No flakiness in either direction.
• Windows never manifests it (20/20 pass with and without the fix). There the accept
  cancellation is a clean OperationCanceledException (already suppressed); the non-OCE
  IOException only leaks on the Linux dispose-based cancellation path. This matches the
  original report, which was Linux.

How the Linux runs were executed

wsl -d Debian                       # default distro (docker-desktop) has no shell
cd /mnt/d/Code/coreipc
DOTNET_ROLL_FORWARD=Major \         # Debian has .NET 8/10 runtimes, not net6.0
  dotnet test src/UiPath.CoreIpc.Tests/UiPath.CoreIpc.Tests.csproj -f net6.0 \
  --filter "FullyQualifiedName~DisposingServerDoesNotLogFailedAcceptOnTeardown"

Follow-up — addressed Copilot review (commit 8c52ee4)

Three test-only nits from the Copilot review (the fix itself was unchanged):

• Serialized the whole test suite — [assembly: CollectionBehavior(DisableTestParallelization = true)]. The teardown test attaches a process-wide System.Diagnostics.Trace listener, which is unsafe under xUnit's default cross-collection parallelism; serializing removes the hazard entirely.
• using on the raw NamedPipeClientStream — exception-safe (no leak if ConnectAsync throws); the scope-end dispose is still an abrupt OS handle close.
• Dropped the fixed Task.Delay sleeps — Timeouts.Short for the slot re-park, and removed the post-dispose settle delay (DisposeAsync already awaits the accept-loop teardown, so any OnError has fired by then).

Re-verified with the new timings (Linux WSL/Debian, net6.0): 10/10 fail without the fix, 10/10 pass with it; Windows still passes.

[Copilot]

Pull request overview

This PR reduces teardown noise by preventing IpcServer’s accept loop from reporting “Failed to accept new connection” when the server is already shutting down (cancellation requested), and adds a regression test to validate the behavior using real named pipes.

Changes:

• Suppress _newConnection.OnError(ex) during accept-loop teardown when cancellation is already requested.
• Add a named-pipe regression test that captures System.Diagnostics.Trace error output and asserts teardown does not emit “Failed to accept new connection”.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 3 comments.

File	Description
src/UiPath.CoreIpc/Config/IpcServer.cs	Avoid reporting accept-loop exceptions to observers when cancellation is already requested (teardown race).
src/UiPath.CoreIpc.Tests/NamedPipeSmokeTests.Teardown.cs	Adds a real named-pipe regression test to ensure server disposal doesn’t log failed-accept errors.

---

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

[Copilot]

Pull request overview

Copilot reviewed 3 out of 3 changed files in this pull request and generated 1 comment.

[eduard-dumitru]

/azp run CI

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[danutboanta]

Please fix the node.js on Windows/Ubuntu jobs pipeline