So we don't loop around for no reason when a compilation is already done. This will be probabilistic since we won't be using atomic load/store.
<rdar://problem/86968671>