Previously, the tokenizer had horrendous errors when lexing raw literal strings. These have been removed. The remaining issue still persists, that empty doc comments are serialized in c in a _weird_ way. They should be merged.