today I fell down a terrible rabbithole of internet software development community arguments

until today I was aware of a package manager and package description language called Nix, and a competitor called Guix that had a similar design philosophy but diverged in its free software policy and in avoiding the Nix language

my reaction when I learned of Guix was "man, Nix is already pretty small, splitting it further feels like it hurts the survival / success chances of both subdivisions"

anyway as of today I'm aware of:
- Lix, a fork of the Nix package manager seemingly motivated by concerns with the community structure around Nix (e.g. lack of moderation policies, conflicts of interests with for-profits)
- tvix, a complete reimplementation of the Nix package manager, motivated seemingly by technical rather than social issues
- aux, "an alternative to the Nix ecosystem", currently at the planning stage, seemingly founded for similar reasons to Lix but with a much broader scope

it seems like the Nix community is having kind of a bad time, which sucks for knowing where I should focus my efforts. I place my hope in the idea that no-one can really pass up allowing people to use the nixpkgs repository, so contributions there are "stickier". But only time will tell, probably.

recently I've been thinking about the following two strategies for meeting social needs:

1. goal-oriented: you think explicitly "what things do I want more of, socially", and then "what people do I already know who can provide this thing, or else what kind of people might I meet who can do so" and then "how can I go about making that happen". (Of course you can include e.g. the happiness of other people in "things I want more of".)
2. heuristic, vibes-based: you think generally about what kind of person to be, what kind of social opportunities to take, and act based on what feels good and enriching without having an explicit story for how it will help you achieve anything in particular

I think people are often quite goal-oriented about finding a romantic relationship, but quite vibes-based about developing a relationship, although sometimes goal-oriented again about things like marriage and kids. On the other hand, I get the sense that people are usually less goal-oriented about finding and developing friendships.

I have a lot more thoughts than this but I don't have time to elaborate on them right now and refuse to put this post in Draft Hell. But I think probably the trick is to get better at figuring out when each strategy is the right one, or what assumptions they rest on.

In Ideological Abuse, Busyness, and the Importance of Rest, Ozy Brennan writes about (among other things) Quaker meetings:

In the Anglosphere, most Quaker Meetings for Worship (i.e. Quaker church services) are unprogrammed. To the uninitiated, unprogrammed Meetings seem more like meditation sessions. Everyone enters a room at a particular time and sits in silence. Occasionally, someone will stand up and speak extemperaneously for a few minutes about some spiritual or ethical topic (a “vocal ministry”). After an hour, everyone stands up and goes to get coffee.
[...] very often, thirty minutes in, I notice something about my life that’s completely obvious, and yet that I’d missed for weeks or months: [...]

(a lot of the stuff I'm cutting here is good but I don't want to just reproduce the thing wholesale, go read it)

I'm thinking about doing something like this myself in my own home, probably with someone else I can talk into doing it with me for accountability purposes. One thing I notice trying to come up with something like this is just how many degrees of freedom there are in designing the details:

• An hour seems kind of long, so maybe I'll start out with half an hour.
• Every week? Every two weeks? Every day?
• The thing where people spontaneously get up and talk sounds interesting but also quite disruptive? Do I actually just want to talk about stuff at the end? If I keep the extemporaneous stuff, I can imagine wanting to reply to what someone says, but that seems like it would devolve into a discussion and effectively end the session. It's not even obvious that that's a bad outcome, but I think I don't want it. Should I have no replies at all? I considered replies but no counter-replies, but I think that's probably not sustainable.
• Do we summarise at the end at all? Maybe that puts pressure on people to come up with Content? Maybe it's good to let people have a time in which they don't come up with anything? Part of what I'm interested in exploring here is the idea of slowing down my usually relentless internal narrative, in which case the biggest success would be thinking nothing at all. But maybe even that's not in tension with having something to say at the end.
• Is there some kind of way that I should be nudging people to spend their time? Is it actually best to be as unprescriptive as possible? Does that risk people just feeling aimless and restless and doing nothing? Can I have prompts that aim to fill space without replacing anything that already fills it, or is it inevitable that any prompts will tend to crowd out what comes naturally?

Some of this will naturally self-correct: I think I'll have some perception of whether a given length of time is obviously too short or too long (although even there part of me suspects there's value in giving myself an uncomfortably long time, in the hope of not being able to sustain avoiding my problems for that long). But really testing and evaluating all the freedom here, especially with the tendencies for interactions between them (e.g. spontaneous discussion might be less disruptive if you have an hour than half an hour, because you have more time to "get back into it"), feels impractical. I think I have a harder time than some people accepting that I'm just going to make a bunch of guesses, and for many of them, never find out if I was right or not.

I expect the interested audience for this post to be pretty small. I encourage you to stop reading as soon as that seems right to you.

I just finished writing NixOS modules + packages for flexiprocity, e.g. config.nix. I haven't quite switched the "live" version over to the new host, but I think the only thing left is the data migration, which I'm expecting to not be a big deal.

My aim has been to have all the configuration in code, so that I don't have to remember what I did to set up the box, and I can easily recreate the box (e.g. when I move host, as I am doing). A few sore points:

• I'm using pulumi's hcloud provider for deploying the server itself. You can boot servers directly into Ubuntu, but they don't prepackage images for NixOS. However you can mount a NixOS ISO and boot from there. (Why do they natively support the ISO image but not a normal OS image? I don't know). The process of installing NixOS from ISO is pretty manual: I have to configure the server to mount the ISO, go into the console and run a bunch of commands to get it all set up correctly, (spend a bit of time confused about how to install GRUB an so on, have a few false reboots...), shut down, reconfigure the server to unmount the ISO, and boot it again. I could automate much of this with the API, but automating running the install process from the console means typing stuff into a VNC session, which sounds possible but not necessarily fun. (There's also nixos-infect which tries to turn a running Ubuntu (or etc.) host into a NixOS one, but that sounds a bit scary, and also maybe I do want to be able to set up my filesystems, which really sounds like I'd need to boot from ISO to do it).
• The smallest Hetzner server instance (4G RAM) can run nixos-rebuild, but it OOM'd at AWS (0.5G). If you want to use NixOS, you're not going to get away with as tiny a memory or disk footprint than if you didn't, I think. (Unless you do some fancy thing where you use Nix to build a Docker image or OS image, but I can't be bothered with that).
• It seems like packaging JavaScript stuff on NixOS is in a not-great state right now, compounded by the fact that I don't fully understand the interactions between node, npm, and e.g. yarn myself. For now I'm using npm exec for postgraphile instead of making a "real" package for it, which seems to work fine.
• I don't have a great solution for "state initialisation" (e.g. applying the initial database schema, creating the jwt key). My postgres server systemd unit just has a postStart hook that checks if the database exists and runs the schema if not. I'm not certain that postgraphile waits for this to happen before trying to connect.
• I don't have a great solution for secrets (e.g. my Google secret key for implementing login with Google). Currently I just do it manually over ssh.
• Getting the elm compiler to work in a network- and filesystem-isolated environment is surprisingly annoying, because it wants to download package metadata from the website. Luckily, the people who wrote elm2nix have already worked around this to some extent, but there's more I want from them.
• Writing a package that puts git revision information into the program was pretty painful. nix asks you what the source directory for your build is, and makes a copy before building in it, so you have to specify that to be the whole repo so you can pick up the .git dir and get git information from it. Worse, if you're submoduled in a larger repo, you don't even know what the whole repo is! I worked around this by letting the owner of the "supermodule" tell the package what's going on, but the "real fix" is probably going to be just not using git version information anymore.

All that said, I'm pretty happy that I can write config that manages a pretty disparate collection of languages, build processes, and system configurations, and I can put it all in one repo. I learned a bunch about how NixOS modules and packages worked, though I still have more to figure out.

I guess technically three repos: I'm still unsure how / if to publish my NixOS configuration.nix and Pulumi infra configuration, but it's not where most of the meat is, I think, so I'm not losing sleep over it.