Hi Ludo, On Mon, 16 Sep 2019 14:24:48 +0200 Ludovic Courtès wrote: > Does that ring a bell? Yes, I've brought that up upstream and upstream is more than willing to work on this and debug this problem if there is a system to debug it on. However, as far as I understand we decided not to give thepowersgang (authors of mrustc) a login to a Guix machine--therefore, the situation will not improve. The problem is NOT reproducible in Debian with the same gcc version. Maybe I'll get my home internet set up next month and put a Guix machine on it, but right now I only have mobile internet (with very slow upload and behind NAT). As it is now, I cannot reasonably give someone a Guix machine already setup to debug this problem. The problem is 100% reproducible and I estimate would be easy to fix for the authors--and maybe would fix the similar armhf problems as well. So if someone could put a Guix machine on the internet and give thepowersgang access, that would be great. I can't right now. If that happens, I can instruct thepowersgang how to enter an environment where this problem can be reproduced and fixed. Even at the last FOSDEM, Chris Marusich and I saw this problem and fixed part of it--by now we got upstream attention. (After all this inertia maybe we lost upstream attention again--we'll see) > Any ideas of a fix or workaround we could apply? No, but thepowersgang might find it very quickly. They guess it might be some C undefined behavior being used by the mrustc->C translator, or a problem with the struct layout (although I've checked the latter and it should be fine).