Dependencies in python has evolved a lot in the last 2 years. You can use a simple setup.cfg to make them, and you can use a wheel like a jar if you want.
That's not so much the issue - as Python package/dependency problems are not so much a problem of 'how it works' - it's an issue because nobody is on the same page. It's byzantine.
If python dependencies boiled down to 'point to a directory' well then fine, everything would work that way and we'd build tools around it, docs would be clear etc. etc..
But as someone who uses Python seldom and loves the syntax, I generally stay away from it mostly for that reason.
And lack of typing which is obviously another discussion.
Actually, "point to a directory" works. Any directory can be added usong sys.path.append() and its python content becomes importable. No needs to do anything more.
You are right, "almost always the case" was a baseless claim and binary wheels are better than nothing.
However, how many projects still have "psycopg2" in their setup.py (or is it requirements.txt? ;) instead of "psycopg2-binary"? I just found out that the latter exists when I checked PyPI.
I'm not pretending all is fine in the kingdom. But a lot of big Python issues (2 vs 3, encoding, deps) are a in a fantastic state compared to years ago and I think it's important that people don't stay stuck on a bad impression from past mistakes.
It's also for me a way to acknowledge the huge work the community has been putting into fixing what's been asked.
