rnk wrote: > FWIW, I'm seeing the test failure on Windows x64 with a debug build of > Clang's unit tests as of a fresh fetch of main this morning.
Sorry about that! I leaned on the Windows premerge tests to find Windows issues (they did, that was useful), but I'm pretty sure it uses release, not debug, builds. I won't be able to repro any windows build failures until Monday, and it's likely I'll be buried in other stuff. If it's a singular test case, I would disable the test (rename it to add the `DISABLED_` prefix to the test name as in 743c32eb81) to unblock work. Reverting and relanding will create new stale gtest executables in everyone's build directories. If you want to go ahead and debug it, the next steps I would try are adding all the various `LLVMInitializeAll*` API calls to the gtest test case. I think what's going on here is that the TargetInfo classes are registered, but the MCTargetDesc or register info classes are not registered. This happens because some other gtest case running earlier registers the targets, and now the CodeGenTest links in the LLVM Target backends, when previously they were not present in the old gtest binary. https://github.com/llvm/llvm-project/pull/134196 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
