rdflib.namespace + rdflib.term Type Refactoring

[JaimieMurdock]

The class heirarchy for rdflib.namespace and rdflib.term is introducing errors in implementation.

- object
  - rdflib.namespace.ClosedNamespace
    - rdflib.namespace._RDFNamespace
  - rdflib.term.Node
    - rdflib.term.Identifier
      - rdflib.term.BNode
      - rdflib.term.Literal
      - rdflib.term.URIRef
        - rdflib.term.Genid
        - rdflib.term.RDFLibGenid
- str
  - rdflib.namespace.Namespace
  - rdflib.namespace.URIPattern

For example, the lack of relationship between Namespace and ClosedNamespace causes multiple definitions of term, __getitem__, __getattr__, and (with PR#1237) __contains__.

Additionally, some of the convenience of Namespace inheriting from str is now gone, so the ClosedNamespace loses some portability. For example, to show ref = URIRef(...) membership in a ns = Namespace(...), the code would be ref.startswith(ns). However, with a ClosedNamespace, this is an error and the code is ref.startswith(cns.uri) (ignoring for a moment validity of the membership relationship - just checking prefixes).

First Step

Refactoring the Namespace and ClosedNamespace heirarchy should be easy without destroying existing functionality.

Proposed class heirarchy:

- str
  - rdflib.namespace.Namespace
    - rdflib.namespace.ClosedNamespace
      - rdflib.namespace._RDFNamespace

This should add some new functions to the ClosedNamespace from the str type.

This may already have an open PR in #1213. However, the implementation there doesn't address duplicate implementation of __getitem__, __getattr__, and term.

Next Steps

Investigate issues with the URIRef representation. Identifier currently has multiple inheritence from Node and text_type. Double check representation.

There is a lot of added complexity coming from Python 2 support and six. Removing six and the now-EOL Python 2 compatibility could simplify long-term maintenance. (#1014 polled on this for 6.0.0, but there doesn't seem to be an issue open to remove it and coordinate?) We'd also be inheriting from stdlib str and not text_type!

Scope Creep

I'm unsure of what the proper granularity for a ticket in rdflib is, but if we want to add some outer bounds for scope: adding type annotations to the code base may help tremendously with finding errors like this.

Also, I'd like to contribute, not just make work for other people :) I've got dev cycles I can put on rdflib and am happy to discuss the best direction for my energy with the maintainers. Already meeting to discuss some work on PySHACL soon. If there's a public roadmap beyond the 6.0.0 tag or Wiki, I'd love to learn more!