Stream: cull-os

Topic: Fixing what Nix did wrong

RGBCube (Jun 21 2024 at 18:38):

(copied from the matrix; this is about fixing flakes being a new concept and destroying the "everything is a derivation" concept of Nix; also see this post https://jade.fyi/blog/flakes-arent-real)

The way I want to fix this in Cab is overridable expressions.

{ a = 1; } ==> a * a

assuming this is assigned to b, just evaluating b will get you 1. But overriding it like b <== { a = 2; } and then evaluating that expression will get you 4. The trick here is that ==> is not a function symbol (unlike :, which has the same behavior in Cab as in Nix)

So you will be able to do things like

inputs @ {
  packages = import <github:cull-os/packages> <== inputs;
} ==> {
  packages.bar = packages.buildCrate ./.;
}

import will import /${<github:cull-os/cab>}/default.cab, which is another expression like our default.cab. Then it will override it, with our inputs. The trick here is that our inputs were already overriden by the Cab CLI, which did <== { buildSystem = ...; targetSystem = ...; } so the build and target systems get passed into CullOS packages. Then we get a neat attribute set packed with packages configured for our target.

RGBCube (Jun 21 2024 at 18:42):

I also want to implement the "standard library" in Pure Rust, eagerly checking for types of things. This will also make it a part of the language instead of some unrelated repository like nixpkgs. And it will be faster, as a result.

RGBCube (Jul 13 2024 at 21:10):

cc @Jonathan Ringer

RGBCube (Jul 13 2024 at 21:11):

this would work without "overrideable expressions" in Nix too. But it would require you to call everything, which is fine I guess

RGBCube (Oct 26 2024 at 19:50):

Don't Do The Mistakes Of Nix

Link: <https://md.darmstadt.ccc.de/nix2>

Notable stuff:

Don't use ? for attribute bind defaults and attribute existence checking at the same time. Very unsure about the syntax for both of these things.
Support tabs in indented strings, but error out when they are mixed with spaces. We can't know the width of a tab, after all.
Remove .4 float syntax. Make 0x.3 invalid, too.
Line endings in strings will get normalized to \n no matter what. People can do string.replace "\n" "\r\n" if they desire.
Make multiline string normalization sane & add HUNDREDS of tests.

Tim DeHerrera (Oct 26 2024 at 21:16):

it might ne nice, although I'm not sure how feasible, if all functions can have default args, not just functions that take a set

RGBCube (Oct 27 2024 at 10:18):

Hmm, would be interesting. So you're saying

<pat> ? "default here" will default to "default here" if <pat> fails to match, ensuring the whole pattern always does?

RGBCube (Oct 27 2024 at 10:18):

This is actually pretty generic, I like it.

RGBCube (Oct 27 2024 at 10:20):

Though, you can't really use this to implement defaulting functions. How would you do first [1, 2, 3] and first 2 [1, 2, 3] with this? (not with type checking of the value, like \nOrXs :: list | number: if type nOrXs == number ...)

RGBCube (Oct 27 2024 at 10:22):

\first = \n :: number ? null:
  if n == null then # didn't match, but we can't access the list either... it's totally ignored
    \[x, ...]: x # now the user has to do `first [] [1, 2, 3]
  else
    getFirstN n;