https://gcc.gnu.org/bugzilla/show_bug.cgi?id=67518
--- Comment #2 from Joost VandeVondele <Joost.VandeVondele at mat dot ethz.ch> --- yes, on current trunk this doesn't fail anymore, not sure if it has gone latent or was fixed with some of the recent activity. Unfortunately, I seemingly still have a wrong code case, which I have not been able to reduce. If I can, I'll file a different PR.