Fix the last known case where pathlib can mangle the meaning of a path. This brings pathlib in line with IEEE Std 1003.1-2017, where trailing slashes are meaningful to path resolution and should not be discarded.

Changes

In several important respects, paths with trailing slashes behave differently to their slash-less counterparts:

• Paths with and without trailing slashes compare unequal and generate different hash codes
• __str__(), __fspath__() and related representations include any trailing slash
• glob() patterns ending with a slash will now generate results ending with a slash, matching glob.glob() behaviour
• match() now observes trailing slashes, and so its pattern language exactly matches that of glob().

To manipulate a trailing slash, we add these new methods/properties:

• has_trailing_sep - read-only boolean indicating whether a trailing slash is present
• with_trailing_sep() - returns a new path with a trailing slash present
• without_trailing_sep() - returns a new path with a trailing slash omitted

Backwards compatibility

Empty segments given to the PurePath initialiser do not generate new segments, so str(PurePath("foo", "")) results in "foo", not "foo/".

Methods concerned with dirnames and basenames ignore any trailing slash:

• name, stem, suffix and suffixes retrieve the last non-empty path segment, and so any trailing slash are ignored
• with_name(), with_stem() and with_suffix() replace the last non-empty path segment
• parent and parents ignore any trailing slash
• relative_to() and is_relative_to() ignore any trailing slashes in self and other.
• The .parts tuple works exactly as before, and doesn't distinguish paths with trailing separators
  • Not sure if this is right tbh

Dependencies

• GH-112361: Speed up pathlib by removing some temporary objects. #112362
• GH-106747: Improve Path.glob() expectations in pathlib tests #112365
• GH-112675: Move path joining tests into test_posixpath and test_ntpath #112676
• GH-65238: Add test cases for trailing slash handling in pathlib #113248

Future work

Once this lands, we can take advantage of the fact that pathlib does not mangle paths. Thus:

• We can make __fspath__() return an unnormalized path for a tasty speed boost in many applications
  • Optimize pathlib.PurePath.__fspath__() #102783
• We can call into pathlib from shutil, which means we can add methods like chown(), move() and rmtree() without code duplication
  • Add pathlib.chown method #64978
  • Support moving across filesystems in pathlib.Path, as shutil.move() does #73991

• Issue: pathlib strips trailing slash #65238