Take for example takeWhile. Internally it uses span.
Does that mean it just references the input bytestring? Probably not, if so, is there a way to achieve this?
The motivating usecase is a large (>2gb) file that I want to map into memory and extract bytestrings pointing into the mapped memory.
Yes, all substring-like operations on ByteStrings are O(1), as you can see in the documentation, and make a shallow copy with a different offset/length.
If you don't need this, use copy to get a full copy of parsed results, so the original huge string could be garbage-collected.
Additionally, consider Lazy incarnations of mmap and Attoparsec, it may probably be more optimal in case of a consecutive parsing of the large chunk.
