Note that this will break if we deal with non-utf8 paths. Fixing this
way because converting everythig to byte strings instead is a very
invasive task so it should be done as a specific feature to provide
support for non-utf8 paths at some point in the future (if needed).