`shutil.move` into child directory on a case insensitive OS

[danny0838]

Bug report

On a case insensitive filesystem, shutil.move(r'D:\Folder', r'd:\folder\subfolder') will run and cause recureive file/directory copying until hitting OS path length limitation.

Your environment

• CPython versions tested on: 11., 12.
• Operating system and architecture: Windows

Linked PRs

• gh-103165: prevent moving of folders into child folders in case insen… #103409

[arhadthedev]

@giampaolo (as an shutil expert)

[eryksun]

The function shutil._destinsrc() could use normcase(). For example:

def _destinsrc(src, dst):
    src = os.path.normcase(os.path.abspath(src))
    dst = os.path.normcase(os.path.abspath(dst))
    if not src.endswith(os.path.sep):
        src += os.path.sep
    if not dst.endswith(os.path.sep):
        dst += os.path.sep
    return dst.startswith(src)

On Windows, this could yield a false positive for a case-sensitive NTFS directory. On POSIX, it could yield a false negative for a case-insensitive directory or filesystem (e.g. exFAT or FAT32).

[danny0838]

@eryksun For a potential false positive/negative of a mounted filesystem with different case sensitivity, maybe the revised algorithm below could help?

1. Normalize path and force case normalization for both src and dst, check if dst is under src
   e.g. for this case, check r'd:\folder' and r'd:\folder\subfolder
2. Truncate the original paths at the same length, check for the sameness (i.e. inode or something) of the common paths.
   e.g. for this case, check if r'D:\Folder' and r'd:\folder' are same

[serhiy-storchaka]

The current code of _destinsrc() is full of issues, taking into account that abspath() is unreliable if the path contains symlinks and the .. components. Even without the .. components, symlinks, mount points and directory junctions allow to have lexically unrelated path referring to the same directory.

GNU mv just calls rename() and interprets errno=EINVAL as only caused by moving into a subdirectory of itself.

On Windows, rename() raises a PermissionError with WinError ERROR_ACCESS_DENIED or ERROR_SHARING_VIOLATION. It cannot be used to unambiguously detect such case.