Does anyone know what the “llvmclang.diff” patch does to “llvm/utils/lit/lit/TestRunner.py” on the windows buildbots? It looks like it’s a local patch to the buildbots and not part of the LLVM codebase. I tried emailing the maintainer of the buildbot but didn’t get any reply. Thanks.
Ping again :).
I am concerned about this local patch which seems to pass a test that is supposed to fail on windows (I tried it on 3 different machines). clang-tools-extra: r187428 was reverted because this test was marked XFAIL on windows but was passing on the buildbots. There are at least two bugs behind this which I am trying to fix currently but this change of behavior between dev builds and buildbots is a bit concerning. We should just incorporate the changes into trunk or at least document this on llvm.org. Thanks.
Are you sure this wasn’t just a mingw and msys vs cmd and MSVC issue? It seems unlikely that this patch is the source of the discrepancy.
Looking at the name of the builder, it shouldn’t be using mingw/msys. There seems to be other builders in the buildbot that uses mingw/msys. That being said, I am not entirely sure that this patch is the problem, but the content of this patch seems to be unknown which is concerning to me. It is possible that the test is passing on the build bot for some other reason, I just want to rule out the unknown patch from this problem J.
In a general workflow, I think a user should expect to see same results between the dev machines and the buildbots for certain documented configurations. This is the broken build I was talking about: