@tirkarthi / @mariocj89 - can you see any problems with this?

My username in GitHub is @tirkarthi :) Regarding the __version__ value I think it's left as it is when mock was merged to stdlib in Python 3. There is an open issue around removing __version__ for standard library at https://bugs.python.org/issue31826 and there are several notes like library being maintained externally, decimal module having it for standard etc. There are many modules with similar value like argparse that were merged to stdlib. I don't think anyone is using __version__ for this change to be a breakage. I am +0 on this.

rg "__version__ =\s['\"]\d+." Lib | grep -v '/test/'
Lib/imaplib.py:__version__ = "2.58"
Lib/optparse.py:__version__ = "1.5.3"
Lib/cgi.py:__version__ = "2.6"
Lib/argparse.py:__version__ = '1.1'
Lib/unittest/mock.py:__version__ = '1.0'
Lib/tabnanny.py:__version__ = "6"
Lib/ipaddress.py:__version__ = '1.0'
Lib/socketserver.py:__version__ = "0.4"
Lib/re.py:__version__ = "2.2.1"
Lib/_pydecimal.py:__version__ = '1.70'    # Highest version of the spec this complies with
Lib/platform.py:__version__ = '1.0.8'
Lib/logging/__init__.py:__version__ = "0.5.1.2"
Lib/json/__init__.py:__version__ = '2.0.9'
Lib/http/server.py:__version__ = "0.6"
Lib/tkinter/font.py:__version__ = "0.9"
Lib/tkinter/ttk.py:__version__ = "0.3.1"
Lib/ctypes/__init__.py:__version__ = "1.1.0"
Lib/wsgiref/simple_server.py:__version__ = "0.2"
Lib/ctypes/macholib/__init__.py:__version__ = '1.0'

Cool, I'm comfortable removing this as you're +0. It's now confusing because it clashes with the __version__ that the backport needs to have for sensible release to pypi.

@cjw296: Please replace # with GH- in the commit message next time. Thanks!

Juts a heads up that this actually breaks things: https://bugzilla.redhat.com/show_bug.cgi?id=1797690

Why are they relying on this version number? What are they using it for?

I have no idea. IMHO they should stop doing that, yet this change affects something at least. I suggest documenting it at https://docs.python.org/3.9/whatsnew/3.9.html

Ah, fair, I should have done a blurb for this. No idea if I can separately put one of those in?

Upstream issue for stem: torproject/stem#56

Ah, fair, I should have done a blurb for this. No idea if I can separately put one of those in?

No idea either. However I think that blurbs only do changelogs, https://docs.python.org/3.9/whatsnew/3.9.html is maintained separately.

Hello, Stem's author here.

Why are they relying on this version number? What are they using it for?

Only display purposes when running the tests. Stem 1.8 supports Python 2.6-3.x, and with 2.x Mock often came from PyPI. As such the version sometimes mattered in test failures.

Is there a reason to break backward compatibility now rather than Python 4.x?

I'm sorry, but I don't considering this a break in backwards compatibility.

This __version__ was incorrect and unmaintained. If you are using unittest.mock then the version is that of the Python distribution in use. If you are using the backport from PyPI, then it is the version of the mock package you have installed.

I'm sorry, but I don't considering this a break in backwards compatibility.

Gotcha. We might need to agree to disagree on this. Python 3.6 has a version attribute whereas Python 3.9 does not, causing anyone that references it to raise an AttributeError.

Is it a useless attribute? Absolutely, and I firmly agree with removing it in Python 4.x. But as a fellow user of symantic versioning if I was in your shoes I'd flag this attribute as obsolete and remove it in Python 4.x but that said, your show. :)

Python regularly breaks backward compatibility in subtle ways between X.Y and X.Y+1. It does not follow semver. See also https://www.python.org/dev/peps/pep-0387/

Python regularly breaks backward compatibility in subtle ways

Interesting, I didn't realize that. PEP 387 seems to advise otherwise [1] but that said, not a big whoop. Clearly this is an obscure issue and we can hotfix on our end if necessary.

[1] "Unless it is going through the deprecation process below, the behavior of an API must not change between any two consecutive releases."

Unless it is going through the deprecation process below, the behavior of an API must not change between any two consecutive releases.

Yes. That doesn't mean "break in 4.x", but rather "deprecate in 3.9, remove in 3.10". Neither happened here.

The confusion here IMHO is based on the fact that __version__ was not considered API. Whether or not that is reasonable assumption, I don't know. It could be deprecated via module __getattr__, if needed.

I took the liberty to attach this to bpo-31826.

I should have done a blurb for this. No idea if I can separately put one of those in?

You can; use bpo-31826 as the issue to link them together.