I submit for your consideration a new working version (with documentation) of an extensible, subclassable PurePath/Path to close bpo-24132, a 6-year-old bug. I had previously submitted a PR which added classes to pathlib as a path to solving this, but these commits introduce no new public classes, breaking changes, or deviations from PEP 428.

The crux of the issue here is that Path (and PurePath) are factories which generate their subclasses upon instantiation and as such are actually dependent upon their flavoured subclasses. To get around this, we attach _flavour to PurePath/Path (and any non PurePosixPath/PureWindowsPath/PosixPath/WindowsPath class) at time of instantiation. Then any subclass will have the essential _flavour attribute available to it. Moreover, in the case that the class being instantiated is PurePath/Path, instead of instantiating another new instance, we just update the class attributes in place so that it becomes its flavoured subclass. My thanks goes out to Barney Gale for making suggestions that inspired this approach.

In addition, I should mention that @barneygale is working on implementing an AbstractPath interface which will allow for even further extensibility of pathlib. His idea thread on this is here. This was an active consideration as I was writing this, and I specifically chose an implementation which I think works with what he is doing. (Barney, I hope that you find that this to be true - I look forward to your comments.) To verify this, I created a stub version of AbstractPath derived from these commits which I hope further shows the viability of this approach. That code can be found here.

https://bugs.python.org/issue24132