I have observed sporadic crashes when multi-threading is enabled on Apple M processors.
The weaker memory model of ARM architecture compared to X86 reveals subtle concurrency bugs.
I am suspecting the ConcurrentFlyweight lock-free datastructre ConcurrentFlyweight.h.
One immediate way to mitigate this would be to use stronger std::memory_order::seq_cst here and there at least on ARM architectures.
It would be nice to have an Apple ARM runner for the CI.