LLVM 14 does not use opaque pointers yet, slowness should not be introduced by them at the moment. But when that happens, it is natural to expect some kind of performance loss that has to be regained by removing all pointer bitcasts.
That is good to know. Is there a slowdown in general with crystal? This seems like it would be hard to measure. The more features that are added the more specs are added.
I think you would have to add a benchmark and then compile it over many crystal versions.