Yes of course. If I run pyflakes or mypy on our code, it's a complete shitshow. But that's not the point.

The point is, for better or worse, however imperfect our code is, it run cleanly and predictably in older Python interpreters. When I have to correct legacy stuff that is known to work well, I compromise hundreds of hours of formal and informal testing.

Imperfect code that has been running flawlessly for a long time and has proven its reliability is better than more perfect code that hasn't been tested as much.

In fact, in certain industries like the aero industry, it doesn't matter if you find slightly bad code after the system has been certified: it's frozen and you leave it the hell alone unless it's critical. Fortunately we're not exactly in that situation, but we do have customers who require - and pay for - configuration control, and those Python issues kind of make everything more difficult needlessly for us. Lucky for us, our Python packages are mostly support code, so it's not too critical. But we do have to be careful and thorough.