Activity

1. MusicalNinjaDad

   added

   docsDocumentation in the Doc dirDocumentation in the Doc dir

   on Jun 3, 2024
2. Eclips4

   added

   topic-pathlib

   on Jun 3, 2024
3. henryiii commented on Jun 3, 2024

   

   henryiii

   on Jun 3, 2024

   Contributor

   More actions

   I can’t quickly replicate on macOS 14 Intel or macOS 14 ARM locally. There must be something specific about the setup when it happened in cibuildwheel that I’m not capturing.
4. serhiy-storchaka commented on Jun 4, 2024

   

   serhiy-storchaka

   on Jun 4, 2024

   Member

   More actions

   You can get a NotADirectoryError if an intermediate path refers to a file.

   >>> os.unlink('/etc/hosts/file')
   Traceback (most recent call last):
     File "<python-input-1>", line 1, in <module>
       os.unlink('/etc/hosts/file')
       ~~~~~~~~~^^^^^^^^^^^^^^^^^^^
   NotADirectoryError: [Errno 20] Not a directory: '/etc/hosts/file'

   👍React with 👍2Reacted by Henry Schreiner and MusicalNinjaDad
5. MusicalNinjaDad commented on Jun 4, 2024

   

   MusicalNinjaDad

   on Jun 4, 2024

   Author

   More actions

   Thanks ... I just found that through my debugging too...

   Would it be useful to add a comment to that effect in the docs?
6. MusicalNinjaDad

   changed the title [-]pathlib.Path.unlink raises a `NotADirectoryError` on macos if the directory does not exist, documentation expects `FileNotFoundError[/-] [+]pathlib.Path.unlink raises a `NotADirectoryError` if some element of the path is actually a file - this is undocumented[/+] on Jun 4, 2024
7. barneygale commented on Jun 12, 2024

   

   barneygale

   on Jun 12, 2024

   Last edited by barneygale

   Contributor

   More actions

   The error comes from the operating system, and it can occur from many pathlib.Path methods. Other exceptions like PermissionError are also possible. I don't want to start documenting this stuff inline in the pathlib docs because it's hard to know where to draw the line.

   This bit is a tiny bit misleading:

   If missing_ok is false (the default), FileNotFoundError is raised if the path does not exist.

   It could instead say:

   If missing_ok is false (the default), FileNotFoundError is raised if the path does not exist, as reported by the operating system.
8. MusicalNinjaDad commented on Jun 13, 2024

   

   MusicalNinjaDad

   on Jun 13, 2024

   Last edited by MusicalNinjaDad

   Author

   More actions

   That makes sense. I can see how it would quickly become unwieldly to inline all the potential exceptions.

   How about adjusting the wording to:

   This method propagates any :exc:`OSError` encountered during removal.
   
   If *missing_ok* is true, :exc:`FileNotFoundError` exceptions will be
   ignored (same behavior as the POSIX ``rm -f`` command).

   This way the reader does not expect to only see a FileNotFoundError but is attuned to the possibility of a range of errors being raised.

   I've pushed this proposal to the PR in case you think it's OK
9. MusicalNinjaDad commented on Jun 13, 2024

   

   MusicalNinjaDad

   on Jun 13, 2024

   Author

   More actions

   On a more general note, as a user of pathlib; the real value in the Concrete Paths is the abstraction they provide over os specifics. This abstraction breaks relatively quickly once errors are encountered, particularly in cases where different OS handle scenarios differently.

   Would you be interested in adding a short section to the documentation which focusses on this topic as additional help for users? I would be happy to produce a proposal.
10. barneygale commented on Jul 3, 2024

    

    barneygale

    on Jul 3, 2024

    Contributor

    More actions

    Thinking about it some more... maybe we should treat NotADirectoryError like FileNotFoundError in the code, i.e. suppress the exception when missing_ok=True. The Python test suite's os_helper.unlink() ignores specifically these two exceptions, and I think that NotADirectoryError can only mean that the path doesn't exist. Sorry for flip-flopping on this.
11. MusicalNinjaDad commented on Jul 3, 2024

    

    MusicalNinjaDad

    on Jul 3, 2024

    Last edited by MusicalNinjaDad

    Author

    More actions

    @barneygale, don't worry about "flip-flopping". I often find it's the simple things that are actually hard because of how they slowly highlight subtle things as you work through them.

    That's a pretty good idea to suppress NotADirectoryError as logically, this means that the file must be missing - because an intermediate path refers to a file.
    I can imagine that there are 1 or 2 additional such cases which semantically are the equivalent of "file missing" ... possibly something related to symlinks ...(Nope, can't see anything when looking through the subclasses of OSError)

    I'll let that thought rotate subconsciously for a bit and look at the possibility of suppressing NotADirectoryError - if you like the change then you can remove the docs label, if not, then I can revert the commit (?)
12. MusicalNinjaDad commented on Jul 5, 2024

    

    MusicalNinjaDad

    on Jul 5, 2024

    Author

    More actions

    Given what I just found in testing, I would suggest that treating NotADirectoryError and FileNotFoundError as equivalent is the right thing to do:

    Windows doesn't raise NotADirectoryError. That means that the same code will raise on Mac/Linux but not on Windows if someone uses missing_ok=True and hits the case where an intermediate path refers to a file.
13. tu-dc commented on Mar 23, 2026

    

    tu-dc

    on Mar 23, 2026

    More actions

    I think a documentation change is a good idea, regardless of what happens with unlink(). I noticed similar behavior with mkdir(), and as a not-super-experienced Python user, I would've never thought to also catch NotADirectoryError if I were a Windows user.