This patch as-is doesn't build. Building it requires another change that I know is wrong, so I'm posting it below to ask for your help:

I was just reading through https://llvm.org/docs/CodeReview.html#lgtm-how-a-patch-is-accepted and I realized that I might have given the wrong impression that I'm entitled to give "final approval". I'm not, so I'm sending this message to make it clear (I guess you all already know that). Nevertheless, the patch does look good to me (in the English meaning of the sentence, not in the formal patch acceptance process meaning).

Thanks, João, that's exactly what I needed. For the record, the use of -DLLVM_ENABLE_ASSERTIONS=ON makes it reproducible. I don't understand why, yet, but there you go.

First of all, let me say that I think that your patch is correct. Both the fix itself and the test case. I haven't checked that all options that are parsed in CompilerInvocation::ParseCodeGenArgs have some sort of initialization in CompilerInvocation::GenerateCodeGenArgs. Anyhow, it seems logical that they should, so your patch looks good to me.

Hi Joao,