Songbook

https://zpevnik.skorepova.info/

Note

The canonical repository lives at forgejo.isbl.cz. If you're viewing this anywhere else, it's a mirror.

I accept pull requests from both code.nolog.cz and GitHub. You can also just email me a patch — that's fun.

Developing

Two setups are available. Fullstack also works for frontend-only work, but is more involved to set up — in particular, Nix doesn't work on Windows outside of WSL. You can start with the frontend-only setup and move to fullstack later. However, since I use fullstack setup for developing, it's generally more polished.

Frontend-only

Use this if you're only working on the frontend and don't want the extra setup of running the backend locally.

Prerequisites: pnpm

cd frontend
pnpm i
pnpm run gen:api

The dev server proxies /api to https://zpevnik.skorepova.info by default (override with the API_PROXY_TARGET env var if needed). Now you can run:

pnpm run dev

or, for Storybook:

pnpm run storybook

Fullstack

Use this if you're working on the backend, or want the full stack running locally.

Prerequisites: Nix with flakes enabled

Run nix run in the repo root to get everything running (backend, frontend, and database) in one go.

If you'd rather run things yourself, run nix develop in the repo root to enter the dev shell (Node, pnpm, etc.) (for convenience, you can instead use direnv and run direnv allow once).

AI Policy

I maintained this project for almost a decade with zero AI. Lately, I didn't really touch it except couple of weeks before summer camp, but I needed some features. To do that, I had to let go of my perfectionism (perfect; the enemy of good) and just let ThE aGeNTiC HarNEsS do some of the things I wanted to do for years. I just use it in particular way.

My approach. Figure out a small, commit-shaped change first, then ask the model to do that one little thing. Validate that the thing is what you wanted (don't trust anyone that reading code is dead; it's not). Repeat. Basically the same thing that software engineering always was: break it down, figure out how to validate it, do it. LLMs improve the last part - do not outsource the thinking.

A COMPUTER CAN NEVER BE HELD ACCOUNTABLE

THEREFORE A COMPUTER MUST NEVER MAKE A MANAGEMENT DECISION

• IBM Training Manual from 1979

Want to see something improved? Ask the robot, send PR. Drop a note in the PR description telling me:

• that you used AI;
• whether you ran the thing to confirm it works and fits your use case;
• but MOST IMPORTANTLY: what is the the problem you're trying to solve.

I don't care about code beauty or whatever. I care if people enjoy using this thing. And if it'll randomly break and I'll have to fix it two weeks before summer camp (it's kind of a tradition at this point though).

Comments. Delete the pointless comments though, please. There might be some that are warranted, but it's about 10% of what the robot produces.

Training. There's no license on this code, so by default everything here is copyrighted - all rights reserved. Friends and other creatures who ask nicely will probably get permission to use it for whatever, training included. The big labs won't ask anyway, so this paragraph is mostly for the people who would (the hosting is getting scraped even though this project always had robots.txt set to deny).

(yes, this section was AI-drafted; I didn't have the spoons until I hated the AI writing too much. I do have a favourite spoon though. I actually ended up rewriting the section a lot though; fuck rule of threes; emdashes are great tho)