@richardpl
Another issue I found is if previously call mysofa_load, and later mysofa_open with same sofa file and sample rate to one reported in mysofa_load, length of IRs: N changes.
The resulted IRs are bad, because I get artifacts when filtering in time domain, the one in frequency domain are still present but much less noticeable.