[✨ Triage] dotnet/runtime#118064 by amanasifkhalid - `System.Net.WebSockets.Client.Tests.CancelTest_SharedHandler_Lo ...

[MihuBot]

Triage for dotnet/runtime#118064.
Repo filter: All networking issues.
MihuBot version: 7cce3b.
Ping MihaZupan for any issues.

This is a test triage report generated by AI, aimed at helping the triage team quickly identify past issues/PRs that may be related.
Take any conclusions with a large grain of salt.

dotnet/runtime#118064: `System.Net.WebSockets.Client.Tests.CancelTest_SharedHandler_Loopback.ConnectAsync_Cancel_ThrowsCancellationException` failing with NRE on CoreCLR android-arm64 by amanasifkhalid
[Tool] Searching for SafeDeleteSslContext.WriteToConnection NullReferenceException, System.Net.WebSockets.Client.Tests.CancelTest_SharedHandler_Loopback.ConnectAsync_Cancel_ThrowsCancellationException android-arm64, WebSockets NullReferenceException android, SafeDeleteSslContext NullReferenceException (IncludeOpen=True, IncludeClosed=True, IncludeIssues=True, IncludePullRequests=True, Repository=dotnet/runtime)
[Tool] Found 49 issues, 63 comments, 49 returned results (7217 ms)

Here are the most relevant related issues and discussions for issue #118064 (System.Net.WebSockets.Client.Tests.CancelTest_SharedHandler_Loopback.ConnectAsync_Cancel_ThrowsCancellationException failing with NRE on CoreCLR android-arm64, specifically with System.NullReferenceException at System.Net.SafeDeleteSslContext.WriteToConnection):

---

1. Issue #25734 (April 2018) - "SafeDeleteSslContext.WriteToConnection throws System.ArgumentNullException"

• Summary: This issue describes sporadic CI failures with exceptions in SafeDeleteSslContext.WriteToConnection, including ArgumentNullException and process crashes. The failures occur when a client opens an SSL connection asynchronously and then closes it, possibly while the handshake is still in progress.
• Key Discussion Points:
  • The root cause was identified as locking on an object that gets nulled out in Dispose.
  • Suggestions included guarding null-outs behind SafeHandle.IsClosed or never nulling out those fields and letting GC handle them.
  • The consensus was to avoid nulling out fields in Dispose to prevent such exceptions.
• Relevance: The stack trace and context are very similar to the current issue, indicating a long-standing race/disposal problem in this area of the code.

---

2. PR #117982 (July 2025) - "[Apple] Lock over the same object in SafeDeleteSslContext to serialize access to the buffers"

• Summary: This PR addresses synchronization issues in SafeDeleteSslContext by introducing a dedicated lock object to ensure all access to shared buffers is serialized.
• Key Discussion Points:
  • The PR was motivated by issues found while working on another PR and aims to prevent race conditions in buffer access.
  • While the PR is focused on OSX, the underlying synchronization problem is similar to what could cause NREs in other platforms.
• Relevance: Shows recent, active work in this area, and the kind of race that could lead to NREs in WriteToConnection.

---

3. Issue #49600 (March 2021) - "Use after dispose bug in SafeDeleteSslContext.WriteToConnection on OSX"

• Summary: Reports a use-after-dispose bug in SafeDeleteSslContext.WriteToConnection, exposed by changes that started nulling out a buffer in Dispose.
• Key Discussion Points:
  • The issue is not a regression but was exposed by recent changes.
  • The root cause is similar: fields being nulled out while native code may still reference them.
• Relevance: Directly related to the type of NRE seen in the current issue.

---

4. PR #49945 (March 2021) - "catch exceptions in callbacks from native code on OSX in SafeDeleteSslContext"

• Summary: This PR changes the handling of exceptions thrown from managed callbacks invoked by native code, returning error codes instead of letting exceptions propagate.
• Key Discussion Points:
  • The PR is OSX-specific but addresses the general problem of managed exceptions (including NRE) in native callbacks.
  • The author notes that a longer-term fix is needed to avoid using SafeHandle after dispose.
• Relevance: Shows a mitigation for the symptoms, but not a root-cause fix.

---

5. Issue #55736 (July 2021) - "Mono: AOT compiler can't compile System.Net.SafeDeleteSslContext:ReadFromConnection (Mac Catalyst)"

• Summary: Reports an AOT compilation failure for ReadFromConnection, a sibling to WriteToConnection, due to missing attributes.
• Key Discussion Points:
  • Fixed by adding UnmanagedCallersOnly attributes (see PR #55947).
• Relevance: Not directly about NRE, but shows platform-specific issues in this code.

---

6. Issue #52645 (May 2021) - "WebSockets test failures on Android"

• Summary: Tracks various WebSockets test failures on Android, including resource handling and configuration issues.
• Key Discussion Points:
  • Some failures were due to resource files and trimming, not NREs.
• Relevance: Shows Android-specific test instability in WebSockets, but not directly about NRE in WriteToConnection.

---

7. Issue #47582 (January 2021) - "Undocumented NullReferenceException is Raised by ClientWebSocket.ReceiveAsync"

• Summary: Reports that ReceiveAsync and SendAsync can throw NREs if called after a failed OpenAsync.
• Key Discussion Points:
  • Suggests that a more appropriate exception should be thrown.
• Relevance: Shows that NREs can surface in WebSocket code paths after failed or canceled operations.

---

8. PR #69527 (May 2022) - "avoid allocation of SafeFreeSslCredentials and SafeDeleteSslContext on Linux"

• Summary: Refactors how SafeDeleteSslContext is used on Linux, merging it with SafeSslHandle.
• Key Discussion Points:
  • Some discussion about null checks and disposal.
• Relevance: Shows ongoing refactoring in this area, but not directly about NREs.

---

9. Issue #104599 (July 2024) - "WebSocket exception with LLVM and full AOT Enabled in .NET Android App"

• Summary: Reports WebSocket failures on Android with LLVM and full AOT, but the stack trace is about connection failures, not NREs.
• Key Discussion Points:
  • The issue is about connection failures, not NREs in WriteToConnection.
• Relevance: Shows Android-specific WebSocket issues, but not the same failure.

---

Conclusions

• The NRE in SafeDeleteSslContext.WriteToConnection is a long-standing, cross-platform race/disposal issue, especially when native code can call back into managed code after disposal.
• There have been several attempts to mitigate or fix these issues, especially on OSX, but the underlying pattern is present on other platforms (including Android).
• Recent PRs (like #117982) show ongoing work to improve synchronization and prevent these races, but the problem is not fully solved for all platforms.
• Android-specific WebSocket test failures are common, but this particular NRE is less frequently reported on Android, suggesting a possible gap in platform-specific handling or test coverage.

---

Actionable Summary

• This is a known class of issue (race/disposal in SafeDeleteSslContext) with multiple historical and recent reports, especially on OSX but also relevant to Android.
• Recent work (PR #117982) may help, but platform-specific investigation for Android is warranted.
• No direct duplicate for this exact NRE on Android, but the root cause and mitigation strategies are well-documented in the referenced issues and PRs.

---

If you need more details on any specific issue or PR, let me know!