At https://docs.python.org/3/library/pathlib.html#pure-paths one can read

> Each element of pathsegments can be either a string or bytes object representing a path segment; it can also be another path object:

which is a lie:

>>> pathlib.PurePath(b"/foo")
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/bru/code/cpython/Lib/pathlib.py", line 609, in __new__
    return cls._from_parts(args)
  File "/home/bru/code/cpython/Lib/pathlib.py", line 638, in _from_parts
    drv, root, parts = self._parse_args(args)
  File "/home/bru/code/cpython/Lib/pathlib.py", line 630, in _parse_args
    % type(a))
TypeError: argument should be a path or str object, not <class 'bytes'>

So either
(1) the doc is wrong
(2) PathLib path management fails: it should decode bytes parts with os.fsdecode()
I doubt I tagged both components. I'll be happy to provide a fix once you decide what is the right solution.

I take this opportunity to share an itch: filesystem encoding on Unix cannot be reliably determined. sys.getfilesystemencoding() is only a preference and there is no guarantee that an arbitrary file will respect it. This is extensively discussed in the following thread: https://mail.python.org/pipermail/python-dev/2014-August/135873.html
What is the right way to deal with those? If I use "surrogateescape" (see PEP383) how can I display the fake-unicode path to the user? `print()` does seems to use strict encoding. Should I encode it with "surrogateescape" or "ignore" myself beforehand?

Interesting. The doc is wrong here: pathlib was designed so that it only accepts text strings.

> If I use "surrogateescape" (see PEP383) how can I display the
> fake-unicode path to the user? `print()` does seems to use strict
> encoding. Should I encode it with "surrogateescape" or "ignore" myself 
> beforehand?

Yes, you should probably encode it yourself. If you are sure your terminal can eat the original bytestring, then use "surrogateescape". Otherwise, "replace" sounds better so that the user knows there are some undecodable characters out there.

New changeset 7463c06f6e87 by Antoine Pitrou in branch '3.4':
Close #23904: fix pathlib documentation misleadingly mentioning that bytes objects are accepted in the PurePath constructor
https://hg.python.org/cpython/rev/7463c06f6e87

New changeset 386732087dfb by Antoine Pitrou in branch 'default':
Close #23904: fix pathlib documentation misleadingly mentioning that bytes objects are accepted in the PurePath constructor
https://hg.python.org/cpython/rev/386732087dfb

Thanks for the report!