1
Fork 0
mirror of https://github.com/RGBCube/Site synced 2025-07-30 20:47:46 +00:00
Site/site/blog/2024-04-15-nix-iceberg.md

39 KiB
Raw Permalink Blame History

title description unlisted tags
Explaining the Nix iceberg And revealing how cursed Nix is. true
nix/os

Everyone who has ever interacted with Nix tooling knows that it keeps some secrets. It even keeps some so well hidden that it hinders its adoption, giving it a reputation of being arcane and hard to use.

I'll be explaining the contents of the following iceberg chart, which includes some truly arcane examples of Nix code.

Some knowledge of Nix is required, you may get confused with the terminology if you've never used Nix.

The Nix Iceberg

Let's start:

Tier 1: btw I use NixOS

IFD blocks evaluation

IFD (Import-From-Derivation) is when you import a Nix expression from a derivation in the Nix store.

For example:

let
  pkgs = import <nixpkgs> {};

  myNixExprDeriv = pkgs.runCommand "my-file" {} ''
    echo '{ a = "b"; }' > $out
  '';

  myAttributes = import myNixExprDeriv;
in myAttributes.a

This will evaluate to "b".

So, what are we doing in this snippet?

  1. Importing <nixpkgs> and getting the packages out of it.
  2. Creating a derivation that runs an echo command, which writes a Nix expression to the output file.
  3. Then we import the output file, forcing the derivation to be realized as we just accessed the contents of it.

Wait, what does realization mean?

It means to actually build a derivation, using the builder, arguments and inputs described within.

Nix does not realize derivations until you access the contents of them or force them to be evaluated using the :b command in the Nix REPL, see these two examples:

nix-repl> pkgs = import <nixpkgs> {}

nix-repl> pkgs.runCommand "foo" {} "echo 'bar' > $out"
«derivation /nix/store/h27fzbivcxw0cc1bxyyyqyivpw9rsz6k-foo.drv»

Here, it created a .drv file, which is how derivations are represented. But that's it. There is no /nix/store/AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA-foo with contents bar to be seen.

nix-repl> :b pkgs.runCommand "foo" {} "echo 'bar' > $out"

This derivation produced the following outputs:
  out -> /nix/store/rxz2bswgx6wlkdxnrcbsb503r9a67wc2-foo

And here we force the derivation to be realized, which produces the output.

Where were we again? Right, the 3rd point: Then we import the expression, forcing the derivation to be realized as we accessed the contents of it.

The 3rd point is the important part. A typical Nix expression does not depend on the output of any derivation, which in turn makes evaluating a Nix expression not require realizing any derivations whatsoever.

But with IFD, you have to realize a derivation to even finish the evaluation of your Nix expression. This will block Nix evaluation for a long time, as Nix is evaluated on a single thread and realizing any derivation takes a non-trivial amount of time.

TL;DR: IFD blocks evaluation because:

  1. Evaluation is single threaded, so naturally everything blocks it.
  2. You're trying to access a derivation output, so obviously you need to realize (build) it first.

nix-shell and nix shell are completely different

nix-shell is the legacy version of nix develop, which enters a devshell created by a Nix expression. It was (and still is) very useful.

People then realized getting a devshell by passing in the packages you wanted as command line arguments was really convenient, which resulted in the creation of the --packages/-p argument for nix-shell

nix-shell -p is similar to nix shell. But they are not the same.

nix-shell -p creates a shell using the nixpkgs stdenv (and thus depends on nixpkgs) by calling pkgs.mkShell, which includes all packages in the nixpkgs stdenv plus the ones you specified. It also places you in a Bash shell, and is integrated with Bash only.

nix shell only appends the packages you passed in to the PATH environment variable. It is much lighter, as a natural result of not using nixpkgs or its stdenv. It also doesn't have as much of a questionable implementation, as it is in C++ and in Nix natively instead of being a Perl script that uses string interpolation to produce Nix expressions.

However, nix shell does NOT replace nix-shell, the nix-shell replacement is nix develop, which does not expose a --packages option or assume anything about the way you are creating the devshell passed into it. It also uses flakes.

hydra is 17 000 lines of perl

Hydra, the Nix-based continuous build system is almost 17,000 lines of Perl.

Here is the tokei output for its GitHub repository:

Language Files Lines Code Comments Blanks
Autoconf 2 38 37 0 1
C++ 8 4140 3068 360 712
C++ Header 5 768 492 75 201
CSS 3 505 388 35 82
JavaScript 6 343 270 37 36
JSON 1 24 24 0 0
Meson 10 328 293 9 26
Nix 48 2266 1948 84 234
Perl 127 17023 12258 663 4102
Python 1 35 25 1 9
Shell 24 371 279 35 57
SQL 85 1406 989 202 215
SVG 6 6 6 0 0
Plain Text 4 164 0 102 62
YAML 1 1137 1094 0 43
Total 349 30927 21210 3358 6359

nix pills

From https://nixos.org/guides/nix-pills/:

This is a ported version of the Nix Pills, a series of blog posts written by Luca Bruno (aka Lethalman) and originally published in 2014 and 2015. It provides a tutorial introduction into the Nix package manager and Nixpkgs package collection, in the form of short chapters called 'pills'.

Since the Nix Pills are considered a classic introduction to Nix, an effort to port them to the current format was led by Graham Christensen (aka grahamc / gchristensen) and other contributors in 2017.

inherit

inherit is a keyword in the Nix language that brings a variable into an keyed expression, such as an attribute set or let in.

Check out the Nix reference page that explains the keyword in depth.

nix-tree

nix-tree is a tool to interactively browse dependency graphs of derivations.

nix-diff

nix-diff is a tool to see how two derivations differ with colored output.

nix-shell -p gives you a compiler

As mentioned before nix-shell -p is the nixpkgs stdenv plus the specified packages.

And since the stdenv includes a C compiler, so does the shell you enter after calling nix-shell -p hello.

nix-output-monitor

nix-output-monitor, also known as NOM is a neat visualizer for Nix builds. See it in action:

nix-top

nix-top is a simple Ruby script to help people see what is building in the local Nix daemon.

The original source was deleted, so I've seeded it on radicle.

--debugger

The --debugger flag is used to halt evaluation and enter the Nix REPL when evaluating a Nix expression.

You set breakpoints using the builtins.break function:

let
  foo = 123;
  bar = "baz";

  # Nix will stop right here, just before
  # evaluating the attrset passed into
  # `builtins.break`. We are able to access
  # `foo` and `bar`.
in builtins.break {
  inherit foo bar;
}

Evaluate this expression with nix eval --debugger --expr/--file and see.

tvix

Tvix is an alternate implementation of Nix written in Rust.

It aims to have a modular implementation while also reusing already-written Nix crates in the Rust ecosystem. It is licensed under the GPLv3 license.

It has since slowed down in development, but the Snix, a fork of Tvix, still goes on.

eelco's thesis

Eelco's thesis is about The Purely Functional Software Deployment Model. Which also happens to be about Nix.

You can read the thesis here.

fixed-output derivations not rebuilt with changed URL

Fixed output derivations (also called FODs) do not get rebuilt even if you change any inputs passed to them (a URL string is also an input). The reason for this is simple.

Nix will see that the output is the same, and since there already is a derivation with the same output in the Nix store, it will assume it is cached and will use that derivation.

Try changing the URL in the following expression and building it:

let
  pkgs = import <nixpkgs> {};;
in pkgs.fetchurl {
  url = "https://raw.githubusercontent.com/NixOS/nixpkgs/56d6bf5daced702e0099e3a15f0b743363ae429d/README.md";
  hash = "sha256-/Lrhot+ejBBfXsPEyWtzScROLkCmdRjb4LBRcHHn+IE=";
}

Tier 2: package maintainer

github:boolean-option/true

The boolean-option GitHub organization allows flakes to be configured. Let's say you have a flake that provides a binary. Let's also assume you can run it with the following Nix CLI invocation:

nix run github:me/hello-world

This is great, you are able to run the binary. But, there is no way for a flake to accept any configuration arguments. If you wanted to run your program in debug mode, you have to create another output (like packages.x86_64-linux.{release,debug}). Same for compiling without support for X/Y/Z. This results in two to the N power of outputs, N being the feature toggle count.

A dumb flake input like github:boolean-option/true fixes this, even though it is an ugly hack. You can do this in your flake:

{
  inputs = {
    nixpkgs.url    = "github:NixOS/nixpkgs/nixos-23.11";
    debug-mode.url = "github:boolean-option/false"; # Release by default!
  };

  outputs = { nixpkgs, debug-mode, ... }: let
    pkgs = import nixpkgs { system = "x86_64-linux"; };
  in {
    packages.x86_64-linux.hello = pkgs.callPackage ./hello { inherit debug-mode; };
  };
}

And override the debug-mode input like so, to run a debug binary instead:

nix run github:me/hello-world --override-input debug-mode github:boolean-option/true

nix-systems is the same idea as boolean-option, but for systems.

Example usages.

These hacks wouldn't be needed if Nix allowed users to put arbitrary values in inputs - in fact, there is an open issue from 2021 that is still being actively discussed - but here we are.

''foo''\n'' == "foo\n"

The Nix parser is very buggy, and this is one bug.

'' is the character set used to escape ${ in Nix indent strings (No, not multiline strings! All strings in Nix are multiline.):

''
  export BAR_OR_BAZ=''${BAR:-$BAZ}
''

This results in the literal string "export BAR_OR_BAZ=${BAR:-BAZ}", without string interpolation.

Nix will ignore an invalid \ escape after the '' escape in an indent string. Or if the \ escape is valid , it will just append the \ escape to the string, ignoring the '' escape.

(x: x x) (x: x x)

This expression is a way to make Nix recurse forever and overflow its stack. Nix can't detect it either, as the evaluated thunk is always different.

Derivations are just memoized execve

Derivations include all required information to build themselves. This also includes output directories (except when they are content-addressed, but that is for a future blog post!). You can dump a .drv file as JSON with the nix derivation show command, like so:

Long command output
 nix derivation show /nix/store/0aplz036lmggrryvx2xh87ci20hczijf-libsamplerate-0.1.9.drv^*

{
  "/nix/store/0aplz036lmggrryvx2xh87ci20hczijf-libsamplerate-0.1.9.drv": {
    "args": [
      "-e",
      "/nix/store/v6x3cs394jgqfbi0a42pam708flxaphh-default-builder.sh"
    ],
    "builder": "/nix/store/bm0gsz7di3d4q0gw1kk2pa06505b0wmn-bash-5.2p26/bin/bash",
    "env": {
      "__structuredAttrs": "",
      "bin": "/nix/store/r3n9n5483q2zprrrjj0f442n723dkzyk-libsamplerate-0.1.9-bin",
      "buildInputs": "/nix/store/4rbkn1f0px39n75zbib2f43i851vy0ay-libsndfile-1.2.2-dev",
      "builder": "/nix/store/bm0gsz7di3d4q0gw1kk2pa06505b0wmn-bash-5.2p26/bin/bash",
      "cmakeFlags": "",
      "configureFlags": "--disable-fftw",
      "depsBuildBuild": "",
      "depsBuildBuildPropagated": "",
      "depsBuildTarget": "",
      "depsBuildTargetPropagated": "",
      "depsHostHost": "",
      "depsHostHostPropagated": "",
      "depsTargetTarget": "",
      "depsTargetTargetPropagated": "",
      "dev": "/nix/store/ajfrbfsqbmxb4ypnmp39xxdpg9gplxbx-libsamplerate-0.1.9-dev",
      "doCheck": "",
      "doInstallCheck": "",
      "mesonFlags": "",
      "name": "libsamplerate-0.1.9",
      "nativeBuildInputs": "/nix/store/xpah4lnaggs6qg87pg1rd9his89acprm-pkg-config-wrapper-0.29.2",
      "out": "/nix/store/55mwzr1k14mryxnhzz6z3hzaimhl8bpn-libsamplerate-0.1.9",
      "outputs": "bin dev out",
      "patches": "",
      "pname": "libsamplerate",
      "postConfigure": "",
      "propagatedBuildInputs": "",
      "propagatedNativeBuildInputs": "",
      "src": "/nix/store/9jnvkn9wcac6r62mljq9fa9vvriyib1i-libsamplerate-0.1.9.tar.gz",
      "stdenv": "/nix/store/jiz7bpw8vqzq8ncm6nn4v94qyqm9qc2p-stdenv-linux",
      "strictDeps": "",
      "system": "i686-linux",
      "version": "0.1.9"
    },
    "inputDrvs": {
      "/nix/store/356i9xqk710rnmq6y6308sv880m88r7k-pkg-config-wrapper-0.29.2.drv": {
        "dynamicOutputs": {},
        "outputs": [
          "out"
        ]
      },
      "/nix/store/gfybzgm5p0hh7w7mdrz5xkr29dlsriih-libsamplerate-0.1.9.tar.gz.drv": {
        "dynamicOutputs": {},
        "outputs": [
          "out"
        ]
      },
      "/nix/store/jkfhhkxlbkfhmqhaccpmqdna01wzlb42-libsndfile-1.2.2.drv": {
        "dynamicOutputs": {},
        "outputs": [
          "dev"
        ]
      },
      "/nix/store/zlf7fmxbnq4k2xgngk0p953ywjqbci6f-stdenv-linux.drv": {
        "dynamicOutputs": {},
        "outputs": [
          "out"
        ]
      },
      "/nix/store/zx3fgspv17raqfb859qkpqnql2fschm0-bash-5.2p26.drv": {
        "dynamicOutputs": {},
        "outputs": [
          "out"
        ]
      }
    },
    "inputSrcs": [
      "/nix/store/v6x3cs394jgqfbi0a42pam708flxaphh-default-builder.sh"
    ],
    "name": "libsamplerate-0.1.9",
    "outputs": {
      "bin": {
        "path": "/nix/store/r3n9n5483q2zprrrjj0f442n723dkzyk-libsamplerate-0.1.9-bin"
      },
      "dev": {
        "path": "/nix/store/ajfrbfsqbmxb4ypnmp39xxdpg9gplxbx-libsamplerate-0.1.9-dev"
      },
      "out": {
        "path": "/nix/store/55mwzr1k14mryxnhzz6z3hzaimhl8bpn-libsamplerate-0.1.9"
      }
    },
    "system": "i686-linux"
  }
}

nixos-rebuild --fast --target-host

The --fast flag in nixos-rebuild is an alias to --no-build-nix which is explained in the man page like so:

Normally, nixos-rebuild first builds the nixUnstable attribute in Nixpkgs, and uses the resulting instance of the Nix package manager to build the new system configuration. This is necessary if the NixOS modules use features not provided by the currently installed version of Nix. This option disables building a new Nix.

And the --target-host flag is also documented, like so:

Specifies the NixOS target host. By setting this to something other than an empty string, the system activation will happen on the remote host instead of the local machine. The remote host needs to be accessible over ssh, and for the commands switch, boot and test you need root access.

If --build-host is not explicitly specified or empty, building will take place locally.

You can include a remote user name in the host name (user@host). You can also set ssh options by defining the NIX_SSHOPTS environment variable.

Note that nixos-rebuild honors the nixpkgs.crossSystem setting of the given configuration but disregards the true architecture of the target host. Hence the nixpkgs.crossSystem setting has to match the target platform or else activation will fail.

nix supports floats

Yup, you heard it. Nix has floats, too!

Though, note that not every number in Nix is a float. Integers in Nix are stored as 64-bit integers. Floats are also 64-bit. Here's the Nix source code that denotes this.

nix-repl> 0.1 + 0.2
0.3

nix-repl> 0.1 + 0.2 == 0.3
false

nix-repl> 0.2 + 0.2 == 0.4
true

attrset ? key and attrset ? "key"

This syntax is a way to check for the existence of a key in an attribute set.

{ foo = 42; } ? foo evaluates to true. The same applies for { foo = 42; } ? "foo", which is just using a string identifier instead.

You can also do { foo.bar = 13; } ? foo.bar, though this isn't that well known.

flakes invented for Target Corporation

The development of flakes was partially funded by Target Corporation.

Tier 3: assigned nix hacker at employment

#!/usr/bin/env nix-shell
#!nix-shell -i python3 -p python3

(taken verbatim from man nix-shell)

You can use nix-shell as a script interpreter to allow scripts written in arbitrary languages to obtain their own dependencies via Nix. This is done by starting the script with the following lines:

#!/usr/bin/env nix-shell
#!nix-shell -i real-interpreter --packages packages

Where real-interpreter is the "real" script interpreter that will be invoked by nix-shell after it has obtained the dependencies and initialised the environment, and packages are the attribute names of the dependencies in <nixpkgs>.

The lines starting with #!nix-shell specify nix-shell options (see above). Note that you cannot write #!/usr/bin/env nix-shell -i ... because many operating systems only allow one argument in #! lines.

For example, here is a Python script that depends on Python and the prettytable package:

#!/usr/bin/env nix-shell
#!nix-shell -i python --packages python pythonPackages.prettytable

import prettytable

# Print a simple table.
t = prettytable.PrettyTable(["N", "N^2"])
for n in range(1, 10): t.add_row([n, n * n])
print t

--accept-flake-config more like --pwn-me-mommy

The accept-flake-config Nix configuration variable or --option accept-flake-config true flag in Nix commands allows Nix to unconditionally accept flake nixConfig's.

This is dangerous, because this can enable builtins.importNative by enabling the allow-unsafe-native-code-during-evaluation option, which then allows Nix expressions to load arbitrary dynamic libraries, which can do anything as they are not confined to the Nix evaluation sandbox.

However, a malicious flake doesn't even have to go that far. It can define an evil substituter using the extra-substituters key in nixConfig, and you may get served malicious packages.

This is why you should generally be wary of using this option or typing in Y when asked to trust a substituter/enable a setting in interactive mode.

Zilch

ZilchOS is a decidedly tiny Nix-based distro. It is a great project to see how NixOS actually works behind the scenes without too much noise to distract.

It was created by t184256 on GitHub, here is the ZilchOS GitHub organization.

set.a or "meow" is set-specific

As mentioned previously, the Nix parser is weird and treats or as an identifier when it is not right after an attribute selection expression.

So, the or in set.key or default is the keyword, but in set or default it is not, and the latter expression is actually a double function application, where we apply or to set, and then default to the result of that.

builtins.toString [true false true] == "1 1"

I find it weird that this is in the 3rd tier. It's actually pretty simple:

Nix converts true to "1" and false to "" (empty string) when asked to convert a boolean to a string.

And when you convert a list to a string, it converts individual items and then joins them with a space character (0xA).

So builtins.toString [true false true] makes 1 1

__structuredAttrs

__structuredAttrs, when set to true in a derivation argument, will set the NIX_ATTRS_JSON_FILE and NIX_ATTRS_SH_FILE environment variables in the build environment to file paths to the derivation argument contents in the respective format.

Here is an example:

with import <nixpkgs> {};

runCommand "attrs.json" { __structuredAttrs = true; foo.bar = "baz"; } ''
  cat $NIX_ATTRS_JSON_FILE > $out
''

Build it with nix build --impure --expr/--file and then cat result, you will get something similar to this:

Long JSON output
{
  "buildCommand": "cat $NIX_ATTRS_JSON_FILE > $out\n",
  "buildInputs": [],
  "builder": "/nix/store/a1s263pmsci9zykm5xcdf7x9rv26w6d5-bash-5.2p26/bin/bash",
  "cmakeFlags": [],
  "configureFlags": [],
  "depsBuildBuild": [],
  "depsBuildBuildPropagated": [],
  "depsBuildTarget": [],
  "depsBuildTargetPropagated": [],
  "depsHostHost": [],
  "depsHostHostPropagated": [],
  "depsTargetTarget": [],
  "depsTargetTargetPropagated": [],
  "doCheck": false,
  "doInstallCheck": false,
  "enableParallelBuilding": true,
  "enableParallelChecking": true,
  "enableParallelInstalling": true,
  "env": {},
  "foo": {
    "bar": "baz"
  },
  "mesonFlags": [],
  "name": "attrs.json",
  "nativeBuildInputs": [],
  "outputs": {
    "out": "/nix/store/cw8gnrh2jwww459cbwig4y97an79qqnx-attrs.json"
  },
  "passAsFile": [
    "buildCommand"
  ],
  "patches": [],
  "propagatedBuildInputs": [],
  "propagatedNativeBuildInputs": [],
  "stdenv": "/nix/store/zykyv2faxz6s1l2pdn6i7i5hb5r5wri6-stdenv-linux",
  "strictDeps": false,
  "system": "x86_64-linux"
}

__functor

__functor is a magic attribute that attribute sets can have which makes them callable. The lambda you assign to it must accept 2 arguments[^Technically, lambdas in Nix always take a single argument. But for clarity, I'll just refer to lambdas that return lambdas as taking N argument, where N is the lambda count.]. The first being the attribute set itself (commonly named `self") and the second being the argument that was passed in.

Here's an example:

let
  mulAll = {
    accum     = 1;
    __functor = self: arg: self // {
      accum = self.accum * arg;
    };
  };
in mulAll 1 2 3 4 5

This outputs the following:

{ __functor = <LAMBDA>; accum = 120; }

Oh no. We just emulated OOP in Nix!

--output-format bar-with-logs on old CLI

(later renamed to --log-format)

You know how the new nix-command CLI has that bar at the bottom, which looks like [4/0/804 built, 7.7/112.5 MiB DL] downloading '...'?

This option allows you to have that output format in the old CLI by passing in --log-format bar-with-logs.

traceVerbose

builtins.traceVerbose behaves like builtins.trace when you pass --trace-verbose to the Nix CLI. If you don't pass in that option, it completely ignores the first argument and returns the second one.

Tier 4: nix is easy we promise

let f = a: a; s = {f=f;}; in [(f == f) (s == s)]

This evaluates to [ false true ]. Why?

Normally, Functions in Nix cannot be compared. Comparing two functions will always return false, at least when done directly.

But if two attribute sets that are compared have the same memory location, Nix ignores this and does a pointer comparison, totally ignoring all members. This is a hack.

Here's the snippet:

bool EvalState::eqValues(Value & v1, Value & v2, const PosIdx pos, std::string_view errorCtx)
{
    forceValue(v1, pos);
    forceValue(v2, pos);

    /* !!! Hack to support some old broken code that relies on pointer
       equality tests between sets.  (Specifically, builderDefs calls
       uniqList on a list of sets.)  Will remove this eventually. */
    if (&v1 == &v2) return true;

This "temporary hack" was committed in 15 years ago. You can do whatever you want with this information.

nix plugins

As surprising as it sounds, Nix does indeed supports plugins. You can load plugins using the plugin-files configuration option.

From the configuration reference:

A list of plugin files to be loaded by Nix. Each of these files will be dlopened by Nix. If they contain the symbol nix_plugin_entry(), this symbol will be called. Alternatively, they can affect execution through static initialization. In particular, these plugins may construct static instances of RegisterPrimOp to add new primops or constants to the expression language, RegisterStoreImplementation to add new store implementations, RegisterCommand to add new subcommands to the nix command, and RegisterSetting to add new nix config settings. See the constructors for those types for more details.

Warning! These APIs are inherently unstable and may change from release to release.

Since these files are loaded into the same address space as Nix itself, they must be DSOs compatible with the instance of Nix running at the time (i.e. compiled against the same headers, not linked to any incompatible libraries). They should not be linked to any Nix libs directly, as those will be available already at load time.

If an entry in the list is a directory, all files in the directory are loaded as plugins (non-recursively).

Some example plugins are nix-doc and nix-extra-builtins.

/bin/sh and sandbox impurity

By setting the sandbox-paths option to /bin/sh=/bin/sh, Nix will bind the /bin/sh path in the build sandbox (left) to the /bin/sh path in the host (right). This is of course impure, but is useful for bootstrapping from absolute scratch without copying impure binaries to the Nix store.

rec { a = 5; b = a + 1; __overrides.a = 6; }

There is a special field named __overrides in keyed expressions (attribute sets, let-in's and as secret third thing), which simply overrides the parent attribute set with the keys inside it. This is different from the update operator (//) because that will not override the self-references in the recursive attribute set.

rec { a = 5; b = a + 1; __overrides.a = 6; }.b will evaluate to 7, while (rec { a = 5; b = a + 1; } // { a = 6; }).b will evaluate to 6.

let __div = c: map (__mul c); in 2 / [ 1 2 3 ]

Previously mentioned in my HTMNIX blog post, Nix operators get desugared into normal function calls before execution. All operators have their "hidden" equivalents that they get desugared into (__div is for /, etc.), so you can override them using let in.

let __div = c: map (__mul c); in 2 / [ 1 2 3 ] is equivalent to map (x: 2 * x) [ 1 2 3 ] which evaluates to [ 2 4 6 ].

You can also check what a Nix snippet desugars into using nix-instantiate --parse --expr/--file

let __lessThan = a: b: b - a; in 1 > 2

As mentioned above, this expression will desugar into let __lessThan = a: b: b - a; in __lessThan 1 2 which will evaluate to 1.

__impure

With the impure-derivations experimental Nix feature, you can set the __impure attribute to true within derivations to mark them "impure".

What this does is:

  1. Let the derivation build have access to the network.
  2. Prevent the impure derivation from becoming a content-addressed derivation.

Impure derivations can also only be used by other impure derivations or fixed-output derivations (FODs).

Tier 5: normal and can be trusted with nix

let a = _: -1; or = 6; in [ a or 9 ]

The Nix parser is weird.

Normally, or is used for attribute path selection defaults:

{ foo = 123; }.not-here.not-here-either or 123

That above evaluates to 123.

But when parsing an expression that is not an attribute-select, or is treated as an identifier. This means that in the following let-in, we are passing or to a.

let
  a = _: -1;
  or = 6;
in a or

But there is another piece of weirdness. Function applications that use the literal or have higher precedence than the spaces when parsing lists, so these two codeblocks are not equivalent:

let
  a = _: -1;
  or = 6;
in [ a or ]

This evaluates to [ -1 ]

let
  a = _: -1;
  foo = 6;
in [ a foo ]

This evaluates to [ <LAMBDA> 6 ]

However, this behaviour might get removed in the future. But currently, in the Nix version that I am using which is 2.28.3, it prints this warning instead:

warning: at «string»:4:6: This expression uses `or` as an identifier in a way that will change in a future Nix release.
Wrap this entire expression in parentheses to preserve its current meaning:
    (a or)
Give feedback at https://github.com/NixOS/nix/pull/11121

eelco's home address is in nixpkgs

s/used to be/is in/g.

restrict-eval

From the Nix manual:

If set to true, the Nix evaluator will not allow access to any files outside of builtins.nixPath, or to URIs outside of allowed-uris.

nix2

nix2 is commonly used to refer to nix-<action> style commands, such as nix-build, nix-shell while nix3 is used to refer to nix <action> style commands, such as nix build, nix develop, and so on.

__noChroot

When the sandbox Nix configuration value is set to relaxed, fixed-output derivations (FODs) that have the __noChroot attribute set to true will not run in the Nix sandbox.

cloud scale hydra

Cloudscale hydra was Graham[^Graham "Determinate" Christensen] Christensen's previous venture.

It wasn't exactly a "failed" project, the reason it did not make it far was because Graham, in his infinite wisdom, realized that it wasn't particularly smart trying to manage 17kloc of Perl for a thousand or more users (yes, he did get >1000 pre-registrations!).

He then went on to create FlakeHub, which could be said is the successor to the ideals of Cloudscale Hydra.

It however is quite curious that the following links are the only non-automated mentions of the project on the open internet:

If you can't find the mentions in these pages, check the archives out.

Here is a screenshot of the Cloudscale Hydra landing page:

Cloudscale Hydra landing page sketch

(_:_) != (_:_) but (a:a) == (a:a)

Evaluating (_:_) == (_:_), we see that it is false, which means the two functions aren't equal to eachother, as we are comparing them directly and when compared directly, function comparisons return false.

But then why does (a:a) == (a:a) return true? Aren't we still comparing functions?

Nope!

a:a is a legacy URL literal, which can be disabled using the no-url-literals experimental Nix feature.

de betekenis van @niksnut

TODO

let { huh = "?"; body = huh }

This is the legacy let syntax. Equivalent to let huh = "?"; in huh.

Tier 6: has meowed before

let { body = 1; __overrides.body = 2; }

This is a combination of __override for keyed expressions and the legacy let syntax.

function identity is load bearing on importing nixpkgs

Since attribute sets with function members compare function identities (memory locations), comparing any attribute set that contains a function is load-bearing on the function's identity.

The way this affects importing nixpkgs is that nixpkgs internally compares stdenvs, which contain functions, to determine whether if we are cross-compiling.nixpkgs internally compares stdenvs, which contain functions, to determine whether if we are cross-compiling.

Therefore, function identity really is load bearing on importing nixpkgs.

import <nix/fetchurl.nix>

This looks like we are importing <nix>, and getting the fetchurl.nix file in it.

Let's see if that is true:

nix-repl> builtins.readDir <nix>
{
  ".clang-format" = "regular";
  ".clang-tidy" = "regular";
  ".dir-locals.el" = "regular";
  ".editorconfig" = "regular";
  ".github" = "directory";
  ".gitignore" = "regular";
  ".mergify.yml" = "regular";
  ".shellcheckrc" = "regular";
  ".version" = "regular";
  ".version-determinate" = "regular";
  "CITATION.cff" = "regular";
  "CONTRIBUTING.md" = "regular";
  COPYING = "regular";
  "HACKING.md" = "symlink";
  "README.md" = "regular";
  contrib = "directory";
  doc = "directory";
  "docker.nix" = "regular";
  "flake.lock" = "regular";
  "flake.nix" = "regular";
  maintainers = "directory";
  "meson.build" = "regular";
  "meson.options" = "regular";
  misc = "directory";
  nix-meson-build-support = "directory";
  packaging = "directory";
  "precompiled-headers.h" = "regular";
  scripts = "directory";
  src = "directory";
  tests = "directory";
}

There doesn't seem to be a fetchurl.nix file here.

This is because <nix/*> actually falls back to corepkgs, which is a Nix path defined inside Nix itself.

Later, the fetchurl.nix path is defined in corepkgs and its contents are set to a generated C++ header.

You do not need to be in impure evaluation mode to use corepkgs, aka <nix/*>.

test suite of nix wasn't run

TODO

fixed-output derivation sandboxing

TODO

importNative

builtins.importNative allows Nix expressions to import arbitrary dynamic libraries to produce Nix expressions.

Of course, this is turned off by default as it is a security risk. You probably shouldn't use this.

chromium recompressTarball

TODO

more than 1 million chars of indents breaks things

The weird Nix parser hard codes 1000000 instead of SIZE_MAX when determining the minimum indent to strip in strings spanning multiple lines.

So when you have a line with more than a million spaces for the indent, it is ignored and not included in the minimum indent calculation.

Tier 7: wears animal ears to NixCon

nix-repl> builtins.fromJSON ''{"uwu\u0000": 1, "uwu": 2}''
{ uwu = 2; "uwu" = 1; }

TODO

(_: builtins.break _)

Historically, the builtins.break function used to not work reliably in some cases, such as let-in's and function calls.

This was fixed in this merge request, in February 8, 2024.

But before that fix, you would use (_: builtins.break _) or an equivalent instead of builtins.break itself.

multiplayer tic-tac-toe in nix repl

You want to read this blog post: Are Nix Expressions Pacman-Complete?

You can run it using the following bash/zsh/etc (any shell that implements $(( RANDOM ))):

git clone https://stuebinm.eu/git/playground nix-tic-tac-toe

# In first shell:
nix-build --argstr seed $(( RANDOM )) nix-tic-tac-toe/nix-turing/game.nix

# In second shell, in parallel:
move=0
while true; do
  read -p "move: " content
  echo "$content" > "/tmp/input-${move}"
  ((move++))
done

let e="e"; in [001.2e01e.30.4]

TODO

/__corepkgs__/

Already explained previously.

some-expr

TODO

__darwinAllowLocalNetworking

TODO

builtins.derivationStrict

TODO