https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85569
--- Comment #7 from Alexandre Oliva <aoliva at gcc dot gnu.org> --- I realized the preprocessed testcase was the work-around rather than the base failure mode. What confused me was that even the work-around was failing, and that's what my patch is for. https://gcc.gnu.org/ml/gcc-patches/2018-11/msg01969.html I spent some time looking into the initial testcase, but to no avail so far.