Seems impossible to read and verify. Wouldn’t it be simpler and more consistent to have a first parse the IP into an internal format then perform all logic on that?
So not speaking about this library specifically but yes, separating the address parsing part (https://www.netmeister.org/blog/inet_aton.html is a fun read) from the "is this address considered private?" part should make things much simpler.
Vast majority of related code in this space that I've seen works like that. I'm actually struggling to come up with counterexamples.
Regexes can be fine in order to parse the string representations initially. But for everything else I would stick with byte arrays etc.
I'm skeptical of that concern: executing multiple regular expression matches on a string is never going to be faster than parsing the entire string once, into a 4-octet structure, and then performing integer comparisons on the octets.