Enjoying a standard cloud-native time trying to get CI to work, as per.

The one thing that consistently makes me sad about the family of CI systems derived from github actions, is how none of them make it easy to build a customized image in which to do a CI run. And so, every CI run on the planet ends up apt-get installing stuff every single time.

A CI system that gave a shit about efficiency would make it _trivial_ to build a builder image with all the tools you need, and then use it in actual CI runs. Instead, everyone's constantly reinstalling a million packages, downloading a million files, poking a million external things just to get to the point where it can compile stuff.

(This is not a request for help or 'splaining. Yes, I know it's _possible_ to set things up properly. Read again: I'm talking about what the tool makes _easy_, because it doesn't matter what your tool _can_ do, what matters is where the path of least resistance lies.)

Neighbouring rant: a good container-based CI system should make it hard to depend on random build images you don't control. It makes me break out in hives when I see blog posts going "I needed to left-pad a string, fortunately I found the action definitely-not-pwned/left-pad@latest !" Again a consequence of making it hard to build custom step images, it should be so trivial that it never occurs to you to do anything but build your own task images from a traceable source.