The accuracy cost is trivial as well, it's a trade-off. I'm pretty sure the implementers were fully aware that regex can't properly handle every case and decided not to care.
..in source size. What I meant was while for a single source file, performance implications aren't noticeable. But usually when we run a tool like this, we happen to do it on an entire project, e.g. imagine running it on Kubernetes repo. If it were to parse entire syntax trees, the performance implications would be significant.