FSSpecParser implementation for URI handling

[brl0]

Update: Python3.13 naming for "flavour" is now "parser", see ParserBase https://github.com/barneygale/pathlib-abc/blob/e429fd2ae079d3c623e687715a8133f4c4769ed2/pathlib_abc/__init__.py#L38

Issue

Starting new issue as requested based on comment in issue #26.

after digging deeper into the code, and exploring the pathlib implementation, I am wondering if UPath would benefit from having something like a _URI_Flavour derived from the pathlib._Flavour class. This might make it easier to have consistent path handling across the fsspec based implementations.

The base class: pathlib._Flavour

The posix implementation: pathlib._PosixFlavour

Here is a list of the members of the class for an idea of what may need to be implemented:

pathlib._PosixFlavour

• altsep
• casefold
• casefold_parts
• compile_pattern
• gethomedir
• has_drv
• is_reserved
• is_supported
• join_parsed_parts
• make_uri
• parse_parts
• pathmod
• resolve
• sep
• splitroot

Of course it probably makes sense to base as much as possible on fsspec functionality, especially considering its ability for url chaining.
Here are some possibly related functions from fsspec.core:

• _un_chain
• url_to_fs
• split_protocol
• strip_protocol

[ap--]

A few notes regarding flavours:

With #114 coming up, future versions of universal_pathlib will probably derive from the pathlib.PathBase base class (probably available in python>=3.13). Support for older pythons (<3.13) will likely be provided through a backport of PathBase in pathlib2 (see: jazzband/pathlib2#70 (comment)) pathlib-abc (http://github.com/barneygale/pathlib-abc)

As far as I understand it, custom "flavour" implementations would then be equivalent to overriding PathBase.pathmod (see: python/cpython#106533) which is basically using a subset of the public interface of posixpath and ntpath.

This is relevant for providing support for urllib.parse.urljoin-like joining behavior.

[merwok]

See this message about the timeline and back/future-port of the ABC: drivendataorg/cloudpathlib#347 (comment)

[ap--]

It looks like we should implemenent our custom pathlib_abc.PathModuleBase for all Pythons when we start to rely on pathlib_abc

See: barneygale/pathlib-abc#19

[ap--]

Note to self:

This should be relatively simple to achieve in the Python-3.12 support PR I have already drafted:

https://github.com/ap--/universal_pathlib/blob/2878bff22e32dbc5240766f51b3e7d31c00b4b9a/upath/core312plus.py#L38-L58

[ap--]

see: python/cpython#114575

[ap--]

This is basically implemented now via upath._flavour