aobolensk wrote: > Looks good to me. > > If the event is not null we still generate invalid code right? Do we have a test for that? Yes, that's right, that is existing problem. I haven't found any tests for that, actually https://github.com/llvm/llvm-project/pull/212754