Kākāpō – A web bundler for Nix strings with context

by todsacerdoti • 2 years ago

96 41 2 years ago

colordrops2• 2 years ago

I would be completely happy with Nix as a universal build and deploy language.

It is amazing what the Nix team has built. It's horribly complex, especially once you start peering behind the nixpkgs curtain, and yet it works incredibly well and is almost universally adaptable.

I've got my 2 personal machines as well as my homelab and a few other appliances all running off of it now.

pxc• 2 years ago

Me, too, more or less. I think there's room for improvement, but Nix is a nice language for that niche.

imiric• 2 years ago

The ideas of Nix are very powerful, but the experience of using it can be frustrating. It desperately needs a UX overhaul, or at the very least a distro that wraps it into a user-friendly package.

I would happily adopt it as a build and deploy system if the UX was improved, but I currently hesitate to recommend it to my teammates because of its rough edges and steep learning curve.

whateveracct2• 2 years ago

There's been several UX improvements merged lately. For instance, better error messages upon runtime type errors iirc.

As is, Nix isn't actually hard if people learn how to learn. I see coworkers have trouble with it but don't even know how to use the repl to debug. I always debug their issues in an instant.

So docs and education would be my focus.

hamandcheese2• 2 years ago

This is pretty neat. I wonder if the idea could be extended more generally to "un-nixify" packages - i.e. replace references to the nix store with relative paths.

I can envision some applications for distributing packages for non-nix systems while still relying on nix to build them.

tadfisher• 2 years ago

There's an older attempt at a universal Nix bundler here: https://github.com/matthewbauer/nix-bundle

The flake schema has a "bundler" output that can point to an executable like this to bundle a flake's outputs for any supported target. Not sure about the state of the art, though.

hamandcheese2• 2 years ago

My main gripe with this (and guix mentioned in a sibling comment) is lack of macOS support. I believe macOS support is a large part of why Nix is able to become so popular.

mik1998• 2 years ago

Guix has this feature - guix pack https://guix.gnu.org/manual/en/html_node/Invoking-guix-pack....

xyzzy_plugh2• 2 years ago

Not really. Guix only really targets Linux and it achieves relocatable packs by virtualizing the filesystem in a variety of extremely non-portable ways. You can do this with Nix as well, of course, Guix wins no points here.

Neither supports non-absolute paths which is the crux of the issue.

xyzzy_plugh2• 2 years ago

This can (and has) been done but sooner or later you step on a landmine package for which this is not possible (reliably patching the output), so you end up rebuilding (reliably patching the input) and now you're inevitably rebuilding the entire world. It would've been easier to just do that in the first place.

That being said I've long suspected there is a happy middle ground where you hack at the problem from both sides. On one side you provide F(x) -> x' where x is an absolute path closure and x' is a relative path closure, for any supported subgraphs in the build graph. Then, for the rest, you rebuild from source using any transformed inputs you can. In other words, when you're evaluating x', if you have x in a cache already, and you can rewrite it, do so. If not, just build it from scratch.

The problem is at some point you have to prove that both methods of producing x' result in the same exact output. This is not easy! However you can possibly prove it for any arbitrary closure by simply round tripping. For example, assuming the transform F(x) -> x' is cheap, and at some point you build x, which is almost guaranteed, then allow x'' == x and when you build x just try and retarget it anywhere, and then back again.

This doesn't prove it actually works, but might prove that it's correct, which is acceptable to me. I haven't got far enough to try this out yet, of course. Be my guest!

blueflow• 2 years ago

Has someone german-speaking looked over the name before it got selected?

elpocko• 2 years ago

It means "poopy butt" in german. I giggle every time I hear about the Kakapo bird.

elliottkember• 2 years ago

Given the bird's character – I think it would appreciate this!

timClicks2• 2 years ago

The vowels are different. So, although they look the same, the words wouldn't sound the same.

mrlonglong3• 2 years ago

Kakapo is also Maori for a very cute little flightless parrot native to the New Zealand islands.

baconhigh3• 2 years ago

If you mean Kākāpō - yes it is!

I saw this is already a Github issue for this project - but use of a Māori bird name is considered tapu (holy or sacred) to Māori.

Hopefully that doesn't distract from the project, the developer should probably consider the implications of choosing such a name (it may look like cultural appropriation).

Kia ora!

hatefulmoron5• 2 years ago

The author/developer lives in New Zealand. I don't see how it's 'cultural appropriation', to any extent that naming a programming project after a bird can even be cultural appropriation.

lambtontrojan• 2 years ago

Have they tried asking the bird?

mikeywazowski• 2 years ago

We use Maori bird names for all sorts of things in NZ, such as my child's kindergarten room. Is there a difference between that and this? Or do you think neither is okay?

timClicks2• 2 years ago

Known is the emoji world as party parrot. The bird's name is Sirocco. https://www.doc.govt.nz/sirocco

zdragnar• 2 years ago

Oh wow, I'm not meme-hip and had no idea that came from a parrot mating with the back of a guy's head.

Work will never be same (developers on my team love to use it in slack reactions)

silverquiet• 2 years ago

Douglas Adams is mostly known for his Sci-Fi books, but he also wrote a book about endangered species in the late 80's/early 90's that he said was his favorite work. One of those he visited was the Kakapo, and his infatuation with the bird was clear from his writing. Certainly he got me to like them as well.

robotguy• 2 years ago

I’m a big Adam’s fan and Last Chance to See is my favorite book of his. It’s the first thing I thought of when I saw the title.

If you’re interested, try the audiobook. It’s narrated by Adam’s himself.

ptek• 2 years ago

seems to be in good taste, now if it was to be a name for malware perhaps the playful Kea would be better.

Upvoted for the name, from Kiwiland myself :)

kish10• 2 years ago

You maybe interested in a project I've been working on (nix-server-tools): https://github.com/kish10/nix-server-tools.

The goal is to create tools using Nix to make it easy to deploy applications on a server.

----

So far the project can:

- Create docker-compose files that: deploy a `Caddy` server & server applications in a reverse proxy through the Docker bridge network.

---- This way Caddy & Let's Encrypt takes care of TSL certificates, while thanks to Nix & Docker we simply need to create docker-compose files. (for an easy deployment)

- Deploy a `paperless-ngx` application that backups files using `borgbackup` & sends an error email when the backup fails.

----

Next week I plan to add:

- A Nix generated `cloud-init.yaml` file to make it easy to initialize & setup new servers (for easy & quick deployment of applications that need to run on additional servers).

----

Examples:

- Deploying a "hello-test-app": https://github.com/kish10/nix-server-tools/blob/main/reverse...

- Deploying a "hello-test-app" & "paperless-ngx -- with borgbackup & error emails": https://github.com/kish10/nix-server-tools/blob/main/reverse...

----

I just started open sourcing the code this week, so there isn't a lot of documentation & examples.

I could definitely use feedback & contributions, so I'd appreciate it if you would take a look.

Kākāpō – A web bundler for Nix strings with context

Related Stories