Experienced during an integration test up to CatTrap, with GDB diagnosing the problem.
I haven't yet diagnosed under which specific conditions this happens.
Correcting the ABI I'm binding to appears to have fixed this issue, but I've still (according to assertions in FcFini) got a memory leak. FC_DEBUG=16 should tell me where it occurs. Or maybe try using ltrace again.
Presumably fixed in the reimplementation!