[llvm] [MachineVerifier] Report errors from one thread at a time (PR #111605)

Ellis Hoag via llvm-commits llvm-commits at lists.llvm.org
Fri Oct 11 10:43:16 PDT 2024


ellishg wrote:

> Probably fine, but I don't think we really need the verifier to be thread safe. Step 1 of debugging any issue would be to get a standalone reproducer which fails

I did find a reproducer, but because it included multiple failing modules, the error was scrambled, so I couldn't minimize the reproducer easily. I couldn't even tell which modules were failing until I introduced a lock in the reporting code.

> We also should probably stop dumping the entire function on failure.

Makes sense to me, but that should probably be a separate PR.

https://github.com/llvm/llvm-project/pull/111605


More information about the llvm-commits mailing list