Motivations for Custom Packaging/DSL vs. Extending brew?

[KMJ-007]

I'm curious about the decision to build a full custom system (YAML DSL) instead of iterating on Homebrew's infra (e.g., taps, bottles, Ruby formulas). Homebrew isn't perfect (non-atomic updates, dep conflicts, etc.), but it has a huge ecosystem.

What specific bottlenecks did you hit with Homebrew that made a ground-up rewrite worthwhile? Any key achievements in SPS v1/SPS2 that validated this path (e.g., perf gains, easier maintenance)?
Would love your thoughts

thanks for the great work!

[KMJ-007]

I am aware that you mention about this in the readme of sps1

Another reason is that I made some choices on internal architecture in the beginning that I have come to see as suboptimal but can not easily refactor so this will allow me to go for the better option right from the get go (go for rx/tx for status propagation from the start, build the core fully async etc.)

but i am more curious about from brew side, cause I also think there is room for improvement. Sometimes, LLVM upgrades take 3 hours, which is not acceptable

[alexykn]

You are kidding? LLVM upgrades with brew take you three hours? They should not :P

Well again, one of the main things is to see whether I am able too xD. Managing automatic upstream tracking and building + publishing is a whole other thing than just doing the client side. Honestly I still am not sure I can but I'm going to try, it should be possible seeing other package managers exist 👍

The other thing is that brew bottles use a weird "repo" system where they host and download binary packages directly from ghcr.io / Github. I think I would prefer more classical repositories like dnf / zypper etc. have them.

Also I don't really like using ruby for a DSL, I am torn between using yaml or a approach similar to rpm specs. But honestly it's whatever, ruby works so no real reason there to call my approach better either.

Atomicity I could have done with brew bottles too, I just did not really have that in mind when starting sps1. All in all I don't think brew is that bad, it's just I'd do some things differently and I wanna see if I can pull this off. Also this is not that important a feature given that on mac the pm is third party and separated from the regular system dirs anyway so a update should never break the base system.

I am kinda taking a break from coding this right now though. I've seen that I relied too much on AI and at the moment am mostly doing smaller projects / stuff at work entirely without chadgpt just too get more fluent coding. Set my goal for finishing this by Q1 2027.

Ahhh what am I saying, to be honest... I am procrastinating because I reached the part where most shit on the client works and I have to get to the package building and automation part which is way harder (especially given my lack of hardware) lol.

Any way I will motivate myself eventually. The point is, my motivation is mostly me wanting to see if I can and having fun building it the way I want it.

Also by building this from the ground up I learned to understand many decisions the people behind brew made. Still don't agree with all of them but I think there isn't a single one where I can't see the reason and I even came to the same or a very similar conclusion with many things.