I'm surprised at how expensive this is. The type of development I do doesn't have string processing in the hot path, so I've never benchmarked these sort of uni

I'm surprised at how expensive this is. The type of development I do doesn't have string processing in the hot path, so I've never benchmarked these sort of unicode operations, and I don't doubt that the bstr Rust library is reasonably well optimized. But 1us/byte checking for word boundaries - that's on the order of 10k instructions per byte! I knew that unicode was more complex than ASCII but yikes!