All commit authors signed the Contributor License Agreement.

@eryksun have a mo to review? thanks

This also fixes my issue #85871 I think!

As was discussed in previous comments, this PR changes the behavior of ntpath.splitdrive() to restrict support for drives in extended paths to just drive-letter and GUID volume names and the "UNC" device. OTOH, "\\.\" paths will continue to support any device name as a drive.

For example, splitdrive() will no longer support any of the following builtin device names (alternate names for "C:" in this case) as a drive in an extended path:

>>> os.path.realpath('//?/BootPartition/')
'\\\\?\\C:\\'
>>> os.path.realpath('//?/Harddisk0Partition2/')
'\\\\?\\C:\\'
>>> os.path.realpath('//?/HarddiskVolume2/')
'\\\\?\\C:\\'

The existing implementation handles these as absolute paths, i.e. paths that have both a drive and a root (per the definition used by both pathlib and the C++ filesystem library):

>>> os.path.splitdrive('//?/BootPartition/')
('//?/BootPartition', '/')
>>> os.path.splitdrive('//?/Harddisk0Partition2/')
('//?/Harddisk0Partition2', '/')
>>> os.path.splitdrive('//?/HarddiskVolume2/')
('//?/HarddiskVolume2', '/')

The new restricted behavior will affect the results of ismount(), join(), split(), relpath(), and commonpath().

Thanks Eryk, will fix.

I've adjusted the patch so that it only affects paths beginning \\?\UNC\. Thanks again for the help Eryk.

@eryksun do you approve of the PR now?

Any other feedback @eryksun? Thanks

As far as I can see, this PR is ready to merge.

Other problems of interest:

• UNC and device paths that have no drive (e.g. "//server" or "//?/UNC/server") are mishandled by join(). They're invalid, but they're still absolute and should not be joined to a drive.
• Device-path drives that end in a colon are mishandled as DOS drives by join(). For example, "//./C:" joined with "spam" should be "//./C:\spam", not a device named "C:spam".
• Device paths that have no root are wrongly classified as relative by isabs(). UNC and device paths are always absolute, even if they lack a path on the device (e.g. "//./Volume{GUID}") or are invalid (e.g. more than two initial slashes; or no drive).

@eryksun Could you file those other problems as new issues? Seems like we always have at least one "UNC isn't quite right" issue open, so may as well continue the tradition 😆