[ITA] llvm

[Jeremy Drake]

On Sun, 15 Jun 2025, Jeremy Drake via Cygwin-apps wrote:

> I pushed update to 20.1.7 and synced patches with upstream, current
> test results:
> llvm:
> Total Discovered Tests: 65227
>   Skipped          :    52 (0.08%)
>   Unsupported      :  2197 (3.37%)
>   Passed           : 62757 (96.21%)
>   Expectedly Failed:   193 (0.30%)
>   Failed           :    28 (0.04%)
>
> clang:
> Total Discovered Tests: 46308
>   Skipped          :     8 (0.02%)
>   Unsupported      :   181 (0.39%)
>   Passed           : 46069 (99.48%)
>   Expectedly Failed:    31 (0.07%)
>   Failed           :    19 (0.04%)
>
> lld:
> Total Discovered Tests: 1066
>   Unsupported:   10 (0.94%)
>   Passed     : 1056 (99.06%)


With 21.1.0-rc1, with a patch applied to avoid the pseudo-relocation in
debug info issue, I get

llvm:
Total Discovered Tests: 68765
  Skipped          :    52 (0.08%)
  Unsupported      :  2461 (3.58%)
  Passed           : 66031 (96.02%)
  Expectedly Failed:   192 (0.28%)
  Failed           :    29 (0.04%)

clang:
Total Discovered Tests: 47721
  Skipped          :     8 (0.02%)
  Unsupported      :   545 (1.14%)
  Passed           : 47116 (98.73%)
  Expectedly Failed:    29 (0.06%)
  Failed           :    23 (0.05%)


lld:
Total Discovered Tests: 1097
  Unsupported:   10 (0.91%)
  Passed     : 1087 (99.09%)


There is now a test in clang that is hanging for me,
Index/crash-recovery-modules.m.  Attaching a debugger sees it waiting for
a mutex in a global destructor.  It is a recursive mutex, but is marked as
being owned by a thread that no longer exists.  Given this is a "crash
recovery" test, I'm guessing it happens to crash while the mutex is held.
It doesn't happen for me when I run the test manually, it crashes as it is
supposed to, catches the SIGILL, and shuts down properly.