clang-cuda-l4, clang-cuda-p4, and clang-cuda-t4 are not reliable

[gkistanova]

Here is the example of false negatives (when failure does not seem related to the commit a builder builds) from these 3 builders:

• [Docs][Clang] Missing DR status for C++23-era papers in cxx_status.html llvm-project#68846 (comment),
• [Docs][Clang] Missing DR status for C++23-era papers in cxx_status.html llvm-project#68846 (comment),
• [Docs][Clang] Missing DR status for C++23-era papers in cxx_status.html llvm-project#68846 (comment).

For the mean time I marked them silent (no failure reporting to blame lists or github).

[fhahn]

Thanks!

[Artem-B]

Those would be libc-on-GPU tests. AFAICT the bots do report a real problem. The fact that the failure may be attributed to a wrong commit probably means that we didn't trigger the tests on all the changes relevant to libc, so the first commit to the LLVM or clang sources is the first to report the failure.

@jhuber6 : FYI. The libc on GPU tests are still broken, though in a different way.

[jhuber6]

Can you link me to where the bot currently lives?