[Repo Assist] fix: fix flaky test infrastructure - empty diagnostics buffer and directory cleanup failures

[github-actions]

🤖 This issue was created by Repo Assist, an automated AI assistant.

Two distinct test infrastructure bugs causing intermittent CI failures have been identified and fixed. The PR creation was blocked, so the patch is provided here for manual application.

Fix 1: Empty diagnostics buffer in waitForLatestDiagnostics

Root cause: waitForLatestDiagnostics uses Observable.bufferSpan to collect publishDiagnostics notifications. For error-free code, the language server sends no notifications, so the buffer emits an empty IList. The call Seq.last throws System.InvalidOperationException: The input sequence was empty.

Observed in CI (PR #1518):

InlayHint.voption.show: for custom function
System.InvalidOperationException: The input sequence was empty
   at ... waitForLatestDiagnostics@301

Fix in test/FsAutoComplete.Tests.Lsp/Utils/Server.fs:

-    |> Array.last
+    |> (fun arr -> if arr.Length = 0 then [||] else arr[arr.Length - 1])

More precisely, change Seq.last → Seq.tryLast |> Option.defaultValue [||].

Fix 2: Directory cleanup failures in dotnetCleanup

Root cause: On macOS + .NET 10, Directory.Delete(path, true) fails with IOException: Directory not empty because .NET SDK background processes hold locks on files in obj/Debug at cleanup time.

Observed in CI: ~15 GoTo Tests failures on macos-15 / net10.0 / TransparentCompiler.

Fix in test/FsAutoComplete.Tests.Lsp/Helpers.fs: wrap Directory.Delete in try/catch with 200ms retry.

Applying the Fix

The patch can be downloaded from workflow run 23994119498:

gh run download 23994119498 -n agent -D /tmp/agent-23994119498
git checkout -b fix/flaky-test-infrastructure main
git am --3way /tmp/agent-23994119498/aw-repo-assist-fix-flaky-diagnostics-test-20260405.patch
git push origin fix/flaky-test-infrastructure
gh pr create --title 'fix: fix flaky test infrastructure' --base main

---

Generated by 🌈 Repo Assist, see workflow run.

Generated by 🌈 Repo Assist, see workflow run. Learn more.

To install this agentic workflow, run

gh aw add githubnext/agentics/workflows/repo-assist.md@7ee2b60744abf71b985bead4599640f165edcd93

[m13v]

the retry with delay pattern for directory cleanup on macOS is something we hit constantly in our test suite too. .NET SDK background processes holding locks on obj/Debug files is really annoying. the 200ms retry works for most cases but on CI runners under heavy load we needed up to 500ms. for the empty diagnostics buffer, the Seq.tryLast approach is correct, but worth considering whether the test should also verify that diagnostics are actually empty rather than silently treating empty as success, since a missing notification could also indicate a real bug in the language server.

[m13v]

for the browser/test session management side where parallel workers create similar filesystem lock contention, here is how we handle it with a locking mechanism that prevents test workers from stepping on each other: https://github.com/m13v/browser-lock/blob/main/playwright-lock.sh