Comment by danpalmer
5 years ago
He strikes me as someone who has been burnt by the packaging issues that have been around in the ecosystem for years.
You're right that the existence of PyPI doesn't mean everyone is forced to link back to it, but damn it would make a lot of stuff a hell of a lot easier if it did. I've been bitten by Debian packages being years out of date, I've seen packages modified by maintainers in ways that are poorly documented and useful for Debian users rather than Python ecosystem participants.
I've managed to work my way out of that now and would never go back to that world if I can help it as the trade of reduced engineering quality (along various axes) for increased "freedom" is not one that I value (personally, or at work). So I can really empathise with this persons position.
I think they handled it poorly, communicated poorly, came to the wrong conclusions, and left the discussion in a bad place, but all of that just screams frustration to me, and I can understand where it comes from.
> I've been bitten by Debian packages being years out of date
I've been bitten far harder by pypi packages trying to download random opaque binaries, miscompiling their own binaries, vendoring and failing to update large packages, having impossible-to-satisfy dependency requirements (relying on pip only ever making a half-assed attempt at satisfying them) and generally making strange assumptions about the target system.
That is indeed also a problem! Essentially it comes down to wrapping sets of dependency management. Thankfully the "engineering" side of Python is fairly good at this in my experience, but I have seen the "data science" side of Python do some of this.
The whole point of nix is to decrease the chances of being burnt by dependency management. This is why packages don't rely on _just installing packages via pip_.