Dropping Python 2 support

[Eric-Arellano]

Hello. Thank you for this awesome library!

I'm wondering if you have thought about removing Python 2 support?

The main benefit would be being able to move the type hints from the type stub library directly to the source implementation. This would simplify things for type-hint users because type hints would work out of the box, rather than needing to install a separate library and needing to ensure that MyPy points to the stubs.

--

Thanks to setup.py's python_requires, Python 2 users would safely resolve ordered-set==3.1.1 when calling pip2 install ordered-set. Meanwhile, Python 3 users would resolve the newest version, e.g. ordered-set==4.0.0.

That is, this won't negatively impact Python 2 users at all outside of missing possible future changes (although, this library seems very stable already).

--

Given the go-ahead, I'd be happy to put up a PR that inlines the type hints, tweaks setup.py, and runs MyPy in CI.

(Type stubs give some issues to the wrapper around MyPy that I'm using, so I would love to see this change and am eager to help however possible.)

--

Finally, see https://python3statement.org for precedent from most of the Python ecosystem for no longer supporting Python 2 once it goes EOL in 5 days. Includes Pytest, TensorFlow, Apache Spark, and Requests. https://python3statement.org

[Eric-Arellano]

I realized that the typing change does not in any way depend on dropping Python 2, thanks to PEP 561. See #60.

Due to the greater ecosystem initiative to drop support for Python 2, I still personally think it would be great to drop Python 2 support for new releases, although I understand possibly not wanting to do this because the burden of Python 2 support is so minimal for this library.

The benefits (I see) to dropping are:

• Support the greater ecosystem initiative to drop Python 2.
• Can use modern versions of tools like Pytest 5, whereas now you'd need to use Pytest 4.
• Can inline the type hints, which improves code readability and reduces duplication.
• Remove the collections vs. collections.abc compatibility code.

The downside is Python 2 users missing out on any future changes to this library (although they can keep using 3.1.1.). ordered-set seems to be a very foundational library and the maintenance burden is light, so I think both choices are compelling and justified.

Either way, thanks again for this awesome library!

[rspeer]

I'm definitely interested in dropping Python 2 support and have a branch where I've done that, bringing the type hints inline. Would you like to take a look at #61 ?