Taking this to extreme degree — could wire up a GitHub/`claude` workflow that automatically generates pull request implementations of top-rated features (e.g. with branch previews hooked up so you can just try it and approve)
In the most recent game I automated builds for the initial build took 1-3 hours, but subsequent builds with the Library/ folder cached took 5-10 minutes.
Are you switching platforms between builds locally? (like building one for web / server / pc / mac etc). Platform switching essentially blows away the cache, making it take the full build time each time. There was an asset years back that would create folders for each platform's Library/ cache and switch between them when you switched build platforms.
In CI (eg using https://game.ci) you can prefix build cache keys with the platform to manage that automatically.
Unity Cloud Build is in general slower/more flakey/sometimes has cache issues compared to DIY CI, but it should speed up after a successful first build as long as it caches the Library/ folder, and they do separate that out by build platform properly. (Worth checking each build config setting is set to cache that)
Curious, are the slow build times locally, your own CI or Unity Cloud Build? (If local/own builders may be some ways to dramatically lower the build times eg by managing the Library/ cache per-platform.)
Happy to help anyone with optimizing these approaches it's been a pain to figure out
It is really what you want out of a hacker-stuck-in-corporate story.
There was a tale maybe 10+ years ago about someone who automated their job with a script or Excel sheet or macro and didn't tell anyone about it. Having a hard time tracking it down again, anyone remember what that was?
Folks who are here and interested in parallelization for CI/CD may also be interested in Dagger.io — I had heard about it on HN over the years but not played w it. It's basically a more fine-grained Docker-like executor with better caching and utilities for spinning up services and running tests.
Curious if anyone else has experiences with it, honestly been surprised at how little I've heard about it
I've been wondering this last year about the use case of compressing highly repetitive logs in a streaming fashion, and whether fine-tuning a model or combination of LLM / datastore might make a sort of adaptive online compression perform well (e.g. allowing central versioned coordination of compression steps/layers as they evolve)
Love the concept of being able to find and read a bunch of these. Real quick put together a quick redirect URL that jumps to a random text-and-code-on-a-page thing https://textandcode.page (can add more if there's other recs!)