It'd be nice to be able to track down all the CI system inefficiencies out there, too, and the biggest culprits, along with just setting up local caches. Rather than repeatedly grabbing the same pinned dependencies for testing and building, it would be far better to just bake them into the necessary container image. (Or even just do things locally, in a way that takes advantage of local caches.)
I've previously strenuously objected to the notion that an Internet connection can be relied upon for build processes, or that it should be necessary for building without a carefully thought-out reason. I like to test by deliberately disabling the network connection, in fact. The most secure and reproducible builds will use tools and artifacts you already have, after all.
Found via https://discuss.python.org/t/_/104294 . This thread also offers the previously submitted (https://news.ycombinator.com/item?id=41629698) proposal https://kristoff.it/blog/python-training-wheels/ , which cites my calculation re the economic value of all this bandwidth. (PyPI bandwidth use has roughly doubled again in the last two years, to my understanding.)
It'd be nice to be able to track down all the CI system inefficiencies out there, too, and the biggest culprits, along with just setting up local caches. Rather than repeatedly grabbing the same pinned dependencies for testing and building, it would be far better to just bake them into the necessary container image. (Or even just do things locally, in a way that takes advantage of local caches.)
I've previously strenuously objected to the notion that an Internet connection can be relied upon for build processes, or that it should be necessary for building without a carefully thought-out reason. I like to test by deliberately disabling the network connection, in fact. The most secure and reproducible builds will use tools and artifacts you already have, after all.
[dead]