Python packaging must be getting better - a datapoint

Thought this may be of interest


Presumably, what’s primarily improved is the rate of wheel adoption.


Which is an indication that publishers have improved their success rate at being able to build wheels, which is something a lot of us have been prioritising for years.

You didn’t suggest it, but I wouldn’t want anyone reading this to think the improvement happened by accident :slight_smile:


Of course.

Adoption of the wheel format has a cumulative effect; if your dependencies are also available as wheels, your own project becomes easier to install.

And yes, I do think that building wheels has become easier, in no small part due to adoption of pyproject.toml - in particular, the PEP 517/518 build-system table. In the old days, it was common enough for’s logic to have a circular dependency - calling setup requires importing a third-party library to compute the arguments, but setup is supposed to be responsible for installing that library.

1 Like