Issue 240795
Summary [lldb][windows][aarch64] test_run_python_os_step fails intermittently on the lldb-aarch64-windows builder
Labels new issue
Assignees
Reporter nodecoda
    ### Summary

`lldb-api :: functionalities/plugins/python_os_plugin/TestPythonOSPlugin.py::test_run_python_os_step`
fails intermittently on the `lldb-aarch64-windows` buildbot builder (builder 141,
worker `linaro-armv8-windows-msvc-05`), starting 2026-11-09 ~18:55Z. The failure is
deterministic in shape, not random noise:

    AssertionError: None != 'main.c' : Make sure we stepped to the next line in main.c

i.e. after `thread.StepOver()` on the OS-plugin virtual thread, the frame's
`SBLineEntry` is invalid (`GetFileSpec().GetFilename()` is `None`).

The sibling test in the same file, `test_python_os_plugin`, passes in the same run, so
plugin loading and the synthetic thread list are fine; only the single-step path is
affected.

### Environment

* Builder: `lldb-aarch64-windows` (https://lab.llvm.org/buildbot/#/builders/141)
* triple: `aarch64-pc-windows-msvc`, `--cmake-build-type Release`
* dotest is invoked with `--env LLDB_LAUNCH_FLAG_USE_PIPES=1`
* test launches the inferior with `target.LaunchSimple()` after `self.dbg.SetAsync(False)`

### Failure

Example build: https://lab.llvm.org/buildbot/#/builders/141/builds/25497

```
FAIL: LLDB (…/clang.exe-aarch64) :: test_run_python_os_step
      (TestPythonOSPlugin.PluginPythonOSPlugin.test_run_python_os_step)

Traceback (most recent call last):
  File "…/TestPythonOSPlugin.py", line 25, in test_run_python_os_step
    self.run_python_os_step()
  File "…/TestPythonOSPlugin.py", line 246, in run_python_os_step
    self.assertEqual(
AssertionError: None != 'main.c' : Make sure we stepped to the next line in main.c
```

The same assertion text is produced by every failing build I checked
(25454, 25470, 25491, 25493, 25497).

### Frequency and scope

Over the 198 completed builds 25301–25498 (2026-11-06 13:46Z – 2026-11-10 10:10Z):

| item | count |
|---|---|
| builds red overall | 54 |
| builds failing this test | **20** (first: 25454, 2026-11-09 18:55Z; last: 25497, 2026-11-10 09:44Z) |
| builds 25454–25498 (i.e. since the first occurrence) | 45, of which **20** hit this failure (~44 %) |
| builds failing this test before 2026-11-09 18:55Z | **0** |

So this is a new failure (~44 % of builds since its first appearance) that appeared late on
2026-11-09, not a long-standing background flake. Before 2026-11-09 the builder's red builds
were caused by unrelated tests (`TestStepOverBreakpoint`, `TestConsecutiveBreakpoints`,
`TestBreakpointIgnoreCount`, ...).

It does **not** reproduce on `lldb-x86_64-win` (builder 211): builds 12224–12237
(2026-11-10 01:36Z – 10:08Z, same revisions) are all green, with `test-check-lldb-api`
run to completion. So it is specific to Windows/ARM64 (or to that worker), not to the
Windows PDB configuration in general.

### Suspected regression window

* last green build of builder 141 before the current red streak: **25447** (2026-11-09 15:59Z), revision `62a26509c606`
* first red build: **25448** (2026-11-09 16:42Z), revision `b5e55c9ca78f` — 19 commits later
* first build hitting *this* test: **25454** (2026-11-09 18:55Z), revision `91d4667646ff`

Two non-test lldb changes land in that window:

* `caf7fd4e0690` — "[lldb] Interrupt the read thread when stopping it" (touches
  `ThreadedCommunication::StopReadThread`, i.e. the communication/read-thread path; this
  builder runs the tests with `LLDB_LAUNCH_FLAG_USE_PIPES=1`)
* `0497136c66a2` — "[lldb][test] Build PDB test inferiors with column info" (adds
  `-gcolumn-info` to `MAKE_PDB` builds in `Makefile.rules`, changing the line tables of
  every Windows test inferior)

I have no Windows/ARM64 machine, so I could not bisect the window myself.

Note that the same window also flipped other Windows tests from expected-failure to
passing: `XPASS: TestInlineStepping.py` occurs in 36 builds and — worth stressing — **all
of them are in builds 25448–25485**, i.e. they start with the very first red build and
none occurs earlier in the 198-build history. `XPASS: TestPlatformConnect.py` occurs in 20
builds (25454–25497). Several distinct Windows tests changing outcome at the same commit
boundary points at a systematic change in Windows line-table/stop behaviour rather than at
one broken test, which is consistent with `0497136c66a2` being involved. Both candidates
should be bisected before concluding.

### Why this matters beyond the builder being red

Buildbot attributes a failure to the last committing revision, so every one of these 20
failures lands as a comment on an unrelated PR. Example: PR #240280 (a LoongArch
codegen-test-only change, merged 2026-11-10 09:37Z as `e9286131f085`) received the
failure notice for build 25497 even though it cannot possibly affect lldb: this very test
had already failed 19 times before that PR merged, the first time at 2026-11-09 18:55Z,
~15 hours earlier.

### Suggested next steps

1. Bisect on a Windows/ARM64 machine (or by reverting `caf7fd4e0690` / `0497136c66a2`
   locally on top of `91d4667646ff`) to find the actual trigger.
2. Until that is done, temporarily restore the previous guard
   (`@skipIfWindows` on `test_run_python_os_step`) or mark it `XFAIL` for
   `oslist=["windows"], archs=["aarch64"]`, so the builder stops producing false
   attributions on unrelated PRs.
3. Separately track the `TestInlineStepping` / `TestPlatformConnect` `XPASS` regressions
   on the same builder, if they are not fixed by the same change.

---

*This issue was drafted with AI assistance and reviewed by a human before posting.*

_______________________________________________
llvm-bugs mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs

Reply via email to