Segmentation fault when disabling DDI during simulation

[FinnHoller]

I have downloaded the new Spirit-2.2.0 release (revision b739e4e06f31c) and noticed a bug in the cpp-ui (qt). Previously it was possible to tweak the Heisenberg Hamiltonian parameters, while running a simulation (tested for LLG and MC). In this new version this still seems possible, although less responsive. Only when enabling or disabling the dipole-dipole interaction, the program crashes with a segmentation fault.

I have compiled the code from source, with following options: cmake -DSPIRIT_USE_OPENMP=OFF ..

System information:

• OS: Linux Mint 21.1 Vera
• CMAKE: cmake version 3.22.1
• CPP: cpp (Ubuntu 11.3.0-1ubuntu1~22.04) 11.3.0
• CPU: Intel Core i5-6400T

[FinnHoller]

I just tested and this does not happen in LLG when openMP is enabled. It still happens with MC.

[GPMueller]

I have the suspicion that the GUI might be passing a nullptr to the C API somewhere. Not sure when, but a few misuses of the C API in the GUIs seem to have crept in. On the develop branch I added a bunch of throw_if_nullptr checks around the arguments of the C API functions, which will generate a decent log message if the API is used incorrectly.

Could you try the develop branch to see whether the issue persists for you there?

[FinnHoller]

I have cloned and built the develop branch, but there are no logs when the segmentation fault happens, should these appear in the spirit log or on the standard output?

[GPMueller]

So the segfault still happens?

I'll try to reproduce this as soon as I can. In case it might be helpful please attach your input-config.

[FinnHoller]

Yes it does still happen.

This is the configuration I am using: test_cfg.txt

[FinnHoller]

I also got this error while changing the exchange interation during an LLG (solver: LBFGS_Atlas) simulation:

malloc(): unsorted double linked list corrupted
Segmentation fault