I could be building it wrong but my build works normally for 100’s of existing test cases without RR, then when I add RR, many of them fail either with no solution, or corrupt solution.
It is in an empty build directory. Also single-threaded Spooles. I previously used commit e4292fd but tried again with ff7f8a7 and it’s the same type of problem, just different numbers.
Perfect. Can you share the steps you use for the build?
It’s quite complicated and I would need to share some files that I can’t do on here. It’s using MSYS2 on Windows and I manually changed some files to get it to work. I’ve used the same process for the official CCX releases for many years and it’s been OK with them.
I see, I understand. I’ll try to fix the issue, but if it works for me I’ll call it done.
Fair enough. I should add that it did give the correct reaction force in at least one case of a single spring element. The problems seemed to happen more with hex and tet meshes.
I can confirm the bug. I compiled a single-threaded MINGW64 version and “RR” makes the other ouput garbage. Since it works on Linux, it is something MINGW64 specific. Seems like a fun bug…
That’s great you’ve been able to reproduce it. Yea sounds horrible. Could be a memory bug which might be hidden on Linux by luck.
Yep, my favourite kind. Btw thanks for the testing and the example.