This is the pinning solution.

great channels vs pinning summary

Wonder what the consensus on channels nowadays?

Again, the entire comment is great.

Another good one.

This entire comment is gold

what

Is this still the case?

mull this over

If you would like to template yaml, then rather do it in Nix or Python

QUESTION: What exactly does nix-shell execute from the Nix expression (i.e., shell.nix, default.nix, etc.)?

ANSWER: Based on my current understanding, the answer is everything. It calls $stdenv/setup (see annotation below) to set up the most basic environment variables (TODO: expand on this), and "injects" the most common tools (e.g., gcc, sed) into it.

It also defines the phases (TODO: verify this) and builder functions, such as genericBuilder. For example, the default builder is just two lines:

source $stdenv/setup
genericBuild

TODO: pkgs/stdenv/generic/builder.sh is a mystery though.

QUESTION: Once dropping into nix-shell, how do I know what phases to execute by looking at a default.nix? (E.g., [..]freeswitch/default.nix)

ANSWER: As far as I can tell, one can override the phases in their Nix build expression (to build the derivation, see at the bottom), but they won't get executed as only the $stdenv/setup (see above) will get sourced, and no builders are called that, in return, invoke the phases (again, see above).

So if one is using nix-shell

• to create/hack on a package, the person has to manually invoke the builder or phases (TODO: still fuzzy on this subject)

• to set up an environment, then one doesn't even have to worry about builders/phases because we just use nix-shell to clear the environment and to inject tools that we need for a given task

QUESTION: When dropping into nix-shell, is this Nix expression (i.e., freeswitch/default.nix) executed? Or just parts of it?

ANSWER: As stated above, all of the input Nix expression is evaluated, but no builders and build phases are called; although, nothing prevents one to override the phases, in case they are creating/hacking on a package.

QUESTION:

The command nix-shell will build the dependencies of the specified derivation, but not the derivation itself.

What is the "derivation" here exactly? I know that it is a build expression, but does that mean the default.nix (or other Nix expression) nix-shell is invoked with?

^{This statement also seems like a contradiction with how `nix-shell` works (i.e., if one issues `nix-shell -p curl`, then `curl` will be available in that sub-shell), but `-p` acts like a shortcut to as if `curl` had been listed in `buildInputs` so this is not the case.}

ANSWER: I have the feeling my confusion comes from the fact that the term "derivation" is used ambiguously in the manuals, sometimes to mean multiple things (see list below).

TODO: Substantiate this claim, and make sure that it not coming from my misunderstanding certain topics.

• Nix build expression (such as default.nix) whose output is going to become the store derivation itself (see last item at the bottom about the Nix manual's glossary definition)

• store derivation.

Had multiple cracks at unambiguously define what a derivation is, and here's a list of these:

• What is the purpose of nix-instantiate? What is a store-derivation? (probably the best try yet)

• What is a Nix expression in regard to Nix package management? (feels sloppier, but commenter mentions ATerm, adding the possibility of making it very specific)

• Closure vs derivation in the Nix package manager (very short, and will have to be re-written, but adds closures to the mix)

• There is now a glossary definition of a derivation in the Nix manual; see this annotation why I find it problematic

QUESTION: What is the difference between nix-shell -p and nix-shell invoked with a Nix expression of mkShell (or other that achieves the similar effect)?

QUESTION: nix-shell does not create a sub-shell, so what does it do? (clarification: so nix-shell indeed does it; I confused it with nix shell)

Come up with an ultimate definition for what a "derivation" is.

So round up all the places where it is mentioned across Nix* manuals, and check out these:

• https://stackoverflow.com/questions/58243554/what-is-a-nix-expression-in-regard-to-nix-package-management (this also needs to be edited)

• https://stackoverflow.com/questions/31490262/what-is-the-purpose-of-nix-instantiate-what-is-a-store-derivation

• https://nixos.org/manual/nix/unstable/expressions/derivations.html

• https://github.com/justinwoo/nix-shorts/blob/master/posts/your-first-derivation.md

• look for more online

From Nix Pills section 6.1. The derivation function (see annotation):

A derivation from a Nix language view point is simply a set, with some attributes. Therefore you can pass the derivation around with variables like anything else.

So there is clearly an ambiguity between what derivations are perceived to be and what is stated in the Eelco Dolstra's PhD thesis. Or maybe I'm having issues with reading comprehension again...

QUESTION: So when I see $out in a builder script, it refers to the default output path because the output attribute in the Nix expression has never been explicitly set, right?

That is, if an input attribute is a reference to a derivation in the Nix store, then

1. that derivation is built first (after a binary substitute is not found, I presume), and
2. the path to the built package (for a better word) is handed to the shell build script.

Organically related to the annotation regarding my nix-shell confusion.

The dissection of this definition to show why I find it lacking:

A description of a build action.

The first (couple) time(s) I read the manuals, this description popped up in many places, and I identified it with Nix expression every time, thinking that a derivation is a synonym for Nix expression.

Maybe it is, because it clearly tries to disambiguate between store derivations and derivation in the last sentence.

The result of a derivation is a store object.

Is this store object the same as a store derivation?

Derivations are typically specified in Nix expressions using the `derivation primitive. These are translated into low-level store derivations (implicitly by nix-env and nix-build, or explicitly by nix-instantiate).

QUESTION: So, the part of the Nix build expression (such as default.nix) where the derivation primitive is called (explicitly or implicitly, as in mkDerivation) is the derivation, that will be ultimately be translated into store derivations?

ANSWER: Start at section 15.4 Derivation.

QUESTION: Also, why is typically used here? Can one define derivations outside of Nix expressions?

ANSWER(?): One could I guess, because store derivations are ATerms (see annotation at the top), and the Nix expression language is just a tool to translate parameterized build actions into concrete terms to build a software package. The store derivations could be achieved using different means; e.g., the way Guix uses Guile scheme to get the same result))

I believe, that originally, derivation was simply a synonym to store derivation. Maybe it still is, and I'm just having difficulties with reading comprehension but I think the following would be less misleading (to me and apart from re-writing the very first sentence):

Derivations are typically the result of Nix expressions calling the derivation primitive explicitly, or implicitly usingmkDerivation`. These are translated into low-level store derivations (implicitly by nix-env and nix-build, or explicitly by nix-instantiate).

QUESTION: Does this refer to pkgs/stdenv/generic/setup.sh? According to 6.5 Phases in the Nixpkgs manual?

ANSWER: I'm pretty sure it does. It sets up the environment (not sure how yet; I see the env vars, but not the basic commands - sed, awk, etc. - that are listed below) and defines a bunch of functions (such as genericBuilder) but it doesn't call these functions!

"Documented" in the Nixpkgs manual under 6.1 Using stdenv.

Used the double-quotes above because I don't consider it well documted. Will give it a try too; worst case scenario is that I'll fail as well.

See annotations with the build-phases tag.

Why are the build phases not enumerated in the Nix manual? If the instructions on how to create a derivation (and thus, a package) then why not go all in instead of spreading out information in different manuals, making the subject harder to grasp?...

(By the way, it is documented in the Nixpkgs manual under 6.5 Phases; not sure why it is not called build phases when every page refers to them like that.)

This such a stupid move to go through a derivation example before introducing the language.

In addition to this rant, step 3. should be more generic, instead of tying it to Nixpkgs; at least, either show how to build your own Nix expression repo, or don't add this step, but it is not at all necessary to write a derivation. There is a Nixpkgs manual for a reason.

So a substitute is a built binary for a given derivation, and a substituter is a server (or binary cache) that serves pre-built binaries, right?

Update: in the next line it says that "it will fall back to using the binary cache substituter", so I guess that answers it.

this is another key topic. Also:

• substitute vs. substituter => this (I think)

See annotations with the substitute tag

binary caches tie in with substitutes somehow; get to the bottom of it. See annotations with the substitute tag.

Maybe this?

Another gem: who knows what a "closure" is.

[This highlight] (a couple lines below) implicitly explains it though:

The command nix-copy-closure copies a Nix store path along with all its dependencies to or from another machine via the SSH protocol. It doesn’t copy store paths that are already present on the target machine.

or this, also just a couple lines below:

the closure of a store path (that is, the path and all its dependencies)

That is a misleading statement. The latest version is where the master branch points, isn't it?

So a channel points to a Nixpkgs commit (on a branch named after the channel) where all packages inside are deemed stable, and all packages are built to have available binary substitutes by a (hydra) build farm.

Definitely avoid this, when a term is used but only introduced formally way later. (There is also a reference to "garbage collector roots" almost at the beginning as well.)

learn more about profiles; creating new profiles was new info

Explain the shit out of this one with tons of examples.

• https://hyp.is/ZHWETHDAEeuNGXcYXST5Ig/toraritte.github.io/saves/Nix-Package-Manager-Guide-Version-2.3.10.html

... and it takes AGES to complete

There is a lot of place in this manual (and probably in the others as well) where the prefix is referred to (usually with italics, such as "prefix/store"), so in the book

• this should be linked to this section (or the one in the book), and

• establish a clear and well-communicated notation to convey this

Why?

Answer is below in the manual: https://hyp.is/qRSFdnCJEeueY8NWtMIeHw/toraritte.github.io/saves/Nix-Package-Manager-Guide-Version-2.3.10.html

Would have been nice to link these to the install chapter where single- and multi-user modes were mentioned.

How would this look in a topic-based documentation? I would think that his chapter would be listed in the pre-requisites, and it could be used to buld different reading paths (or assemblies in DocBook, I believe) such as practical, depth-first (if there are people like me who want to understand everything first), etc.

https://en.wikipedia.org/wiki/Reentrancy_(computing)

Yeah, I there are several tickets and posts about how this is not entirely true.

• https://github.com/NixOS/nix/issues/1623
• https://github.com/NixOS/nix/issues/1402
• https://github.com/NixOS/nix/issues/458
• https://stackoverflow.com/questions/51929461/how-to-uninstall-nix
• https://stackoverflow.com/questions/443699/how-do-you-uninstall-in-nix
• https://apple.stackexchange.com/questions/170000/how-to-completely-remove-nix-package-manager

Traditionally, when a command should be invoked with sudo, it is either included in the example, or the shell indicator is # instead of $.

It should be noted in this section also that since nix 2.1.0, single user install is the default.

What is happening here exactly?

nix-shell's syntax synopsis always bugged because it looks like this

SYNOPSIS
nix-shell [--arg name value] [--argstr name value] [{--attr | -A} attrPath] [--command cmd] [--run cmd] [--exclude regexp] [--pure] [--keep name] {{--packages | -p} packages...  | [path]}

and the canonical example is nix-shell '<nixpkgs>' -A pan; what tripped me up is that path is usually the first in examples, and I thought that the position of arguments are strict. As it turns out, nix-shell -A pan '<nixpkgs> is just as valid.

Side note<br> Apparently there is no standard for man pages. See 1, 2.

'<nixpkgs>' path is the one specified in the NIX_PATH environment variable, and -A pan looks up the pan attribute in pkgs/top-level/all-packages.nix in the Nixpkgs repo.

Wouldn't hurt to tell folks that this is a convenience layer, and one could also just use the old package from the /nix/store, even though that path would be long and obscure; one could use symlinks of course.

Or, onc could just use nix-shell -p that specifies a specific version (that's already in the store), but, of course, it's not that simple...

https://github.com/NixOS/nixpkgs/issues/9682

See main annotation for the derivation topic

This is the gist of how Nixpkgs works (and how it is organized).

QUESTION: What is a fixed-point of a function?

ANSWER: See this video) at least, and the Fixed-point (mathematics)) wikipedia article:

In mathematics, a fixed point (sometimes shortened to fixpoint, also known as an invariant point) of a function is an element of the function's domain that is mapped to itself by the function. That is to say, c is a fixed point of the function f if f(c) = c. This means

f(f(...f(c)...)) = f n(c) = c

an important terminating consideration when recursively computing f. A set of fixed points is sometimes called a fixed set.

For example, if f is defined on the real numbers by f(x)=x^{2}-3x+4,}, then 2 is a fixed point of f, because f(2) = 2.

There is also the wiki article fixed-point combinator that actually plays a role here, but read through the articles in this order.

Then dissect the Stackoverflow thread What is a Y combinator?, and pay attention to the comments! For example:

According to Mike Vanier's description, your definition for Y is actually not a combinator because it's recursive. Under "Eliminating (most) explicit recursion (lazy version)" he has the lazy scheme equivalent of your C# code but explains in point 2: "It is not a combinator, because the Y in the body of the definition is a free variable which is only bound once the definition is complete..." I think the cool thing about Y-combinators is that they produce recursion by evaluating the fixed-point of a function. In this way, they don't need explicit recursion. – GrantJ Jul 18 '11 at 0:02

(wut?)

Other resources in no particular order:

• google search for "fixpoint evaluation fixed point combinator explained"
• Cornell's CS 6110 S17 Lecture 5
• google search for "fixed-point of a function"
• Fixed-point theorem (wikipedia)
• How can I find the fixed points of a function? (math stackexchange)
• The Y Combinator (Slight Return) (see the "Fixpoint of functions" section)
• Clear, intuitive derivation of the fixed-point combinator (Y combinator)? (computer science stackexchange)
• Fixed-Point Combinators (stackoverflow)
• Fixed-point combinator (HandWiki)

QUESTION: How the hell did they come up with the idea of using this with Nix and package management? (..and who? I remember a video saved somewhere, but maybe that was about overlays)

QUESTION: ... and how does it work in this context?

ANSWER: Well, not an answer yet, but this may be something in the right direction:

http://blog.tpleyer.de/posts/2020-01-29-Nix-overlay-evaluation-example.html

QUESTION: What is ldconfig cache?

QUESTION: What is libc and why does Bash need it?

QUESTION: So the Nix store houses derivations and not the built packages?

QUESTION: Are the hashes in the Nix store not unique?

Doesn't really say anything but thought it important to highlight this first mention of derivations in the Nix Pills.

This is a very good pointer; I guess it refers to systemd services when it mentions "simple services".

Instead of #<output> it should have said channel-name instead at the top nix-channel example to keep it consistent.

TODO: Find other sources about this topic. I remember this mentioned already (and it makes) sense, but need to learn more.

TODO: What is a better alternative? An own repo? Flakes? Can cachix help?

It says right below that pinning can help but keep looking.

When package reproducibility become a major concern, as it is the case in this tutorial, it is preferable to refer to a pinned version of the nixpkgs repository instead — i.e, a specific commit of the repository or an immutable archived tarball. The ability to pin the version of nixpkgs is powerful, it will ensure that a package is always constructed from the same Nix source. As we go deeper in the tutorial we avoid using channels in favor of pinned environments.

And symlinks

nix-build <path> --run-env has been superseded by nix-shell. From Nix manual section C.12. Release 1.6 (2013-09-10):

The command nix-build --run-env has been renamed to nix-shell.

How?

It would be helpful to

1. formally describe the formats for NIXPATH, and
2. note the allowed angle bracket syntax accordingly

<path> will work with the prefixless format, but not with the prefixed one, and it may be helpful to spell this out explicitly.

0 [14:16:19] nix repl
Welcome to Nix version 2.3.10. Type :? for help.

nix-repl> :l <nixpkgs/doc>  
Added 40 variables.

nix-repl> :l <doc>
error: file 'doc' was not found in the Nix search path (add it using $NIX_PATH or -I)

I always saw a NIXPATH used with the prefix syntax so far:

$ echo $NIX_PATH
nixpkgs=/nix/var/nix/profiles/per-user/root/channels/nixos:nixos-config=/etc/nixos/configuration.nix:/nix/var/nix/profiles/per-user/root

but NIX_PATH documentation shows that the prefixless format is also valid.

$ export NIX_PATH=/home/toraritte:/nix/var/nix/profiles/per-user/root/channels/nixos

$ printenv NIX_PATH
/home/toraritte:/nix/var/nix/profiles/per-user/root/channels/nixos

$ nix repl
Welcome to Nix version 2.3.10. Type :? for help.

nix-repl> :l <nixpkgs>
Added 12439 variables.

TODO: Is this the same as lib.attrsets.attrValues in nixpkgs?

works both on strings and paths

Path concatenation as well: https://gist.github.com/CMCDragonkai/de84aece83f8521d087416fa21e34df4

Bonus: https://stackoverflow.com/questions/43850371/when-does-a-nix-path-type-make-it-into-the-nix-store-and-when-not

TODO Is this an alias to builtins.attrValues?

This is vague to the point of being useless, given the Nix expression language is untyped.

A snippet from freeswitch.nix:

      configTemplate = mkOption {
        type = types.path;
        # omitted
      };
      configDir = mkOption {
        type = with types; attrsOf path;
        # omitted
      };

• Where does types come from?

  See NixOS manual at "Option types".

• What is the difference between the path types? (I guess, the second one is attribute set consisting only of paths.)

TODO: document lib.mkIf

Found a description in the NixOS manual:

The special function mkIf causes the evaluation of the conditional to be “pushed down” into the individual definitions, as if you had written:

config = {
  environment.systemPackages = if
config.services.httpd.enable then [ ... ] else [];
  ...
};

TODO: document lib.mkDefault

Not mentioned in any of the manuals (i.e., nix, nixpkgs, NixOS), only the NixOS options page has 1 mention:

Some promising search result (in the order of relevance):

• https://releases.nixos.org/nix-dev/2015-June/017549.html
• https://releases.nixos.org/nix-dev/2015-June/017549.html

From source:

  mkOverride = priority: content:
    { _type = "override";
      inherit priority content;
    };

  mkDefault = mkOverride 1000; # used in config sections of non-user modules to set a default

It seems that all the mkOverride calls set up a priority for certain actions.

What is an option?

From the link below to the source, it's just an attribute set with all the function parameters (see below) plus a _type attribute name to allow for type level operations (because Nix is untyped).

  mkOption =
    {
    # Default value used when no definition is given in the configuration.
    default ? null,
    # Textual representation of the default, for the manual.
    defaultText ? null,
    # Example value used in the manual.
    example ? null,
    # String describing the option.
    description ? null,
    # Related packages used in the manual (see `genRelatedPackages` in ../nixos/lib/make-options-doc/default.nix).
    relatedPackages ? null,
    # Option type, providing type-checking and value merging.
    type ? null,
    # Function that converts the option value to something else.
    apply ? null,
    # Whether the option is for NixOS developers only.
    internal ? null,
    # Whether the option shows up in the manual.
    visible ? null,
    # Whether the option can be set only once
    readOnly ? null,
    # Deprecated, used by types.optionSet.
    options ? null
    } @ attrs:
    attrs // { _type = "option"; };

How does it fit in a big scheme of things?

?

If there is recursiveUpdateUntil (i.e., mergeUntil) then mergeWith would also be welcome.

Would be nice to have an alias called merge.

Fix: add semicolon after last Any. (same above in the main type signature)

So the type signature

(String -> Any -> Any) -> AttrSet -> Any
                                     ^^^

should be

(String -> Any -> Any) -> AttrSet -> List
                                     ^^^^

instead, right?

See comment above.

Then why not call it collectVals?... (Following the distinctive naming convention between lib.attrsets.attrVals and lib.attrsets.attrValues.)

Especially because there is lib.attrsets.collect:

catAttrs :: String -> [AttrSet] -> [Any]
collect :: (Any -> Bool) -> AttrSet -> [Any]

(Call it filterVals? There are filterAttrs* functions but those return an attribute set, so no collision.)

This should read "Attribute Set Functions". Nowhere else is attribute sets spelled as "attribute-set".

edit: Alright, "attribute-set" is used in total of 4 times, whereas "attribute set" is used 112 times.

This can be written up in many other forms, plus the possibility of currying deserves to be pointed out:

$ nix repl
nix-repl> (sq: (x: y: sq y + sq x) 2 7) (x: x*x)
53

nix-repl> (sq: (x: y: sq y + sq x)) (x: x*x)     
«lambda @ (string):1:11»

nix-repl> (sq: (x: y: sq y + sq x)) (x: x*x) 2
«lambda @ (string):1:14»

nix-repl> (sq: (x: y: sq y + sq x)) (x: x*x) 2 7
53

nix-repl> (sq: x: y: sq y + sq x) (x: x*x) 2 7   
53

Or use the Nix package manager's nix-shell.

• https://toraritte.github.io/posts/2020-07-15-nix-shell-with-specific-path.html
• https://stackoverflow.com/a/62923700/1498178

There is hnix that is still actively developed (as of 2020/06/21).

It is kind of an anti-pattern using the angle bracket notation and the linked post also makes a suggestion:

{ pkgs ? import (fetchTarball https://github.com/NixOS/nixpkgs/archive/3590f02e7d5760e52072c1a729ee2250b5560746.tar.gz) {};
}:
# ...

It is kind of an anti-pattern using the angle bracket notation and the linked post also makes a suggestion:

{ pkgs ? import (fetchTarball https://github.com/NixOS/nixpkgs/archive/3590f02e7d5760e52072c1a729ee2250b5560746.tar.gz) {};
}:
# ...

It is kind of an anti-pattern using the angle bracket notation and the linked post also makes a suggestion:

{ pkgs ? import (fetchTarball https://github.com/NixOS/nixpkgs/archive/3590f02e7d5760e52072c1a729ee2250b5560746.tar.gz) {};
}:
# ...

What is the difference between the vim and vim_configurable packages?

I believe the source for the latter is here.

This will still work on 20.03 but will show a warning as name = "root"; is deprecated, and luksroot should be used instead (see boot.initrd.luks.devices NixOS option) .

boot.initd.luks.devices = {
  luksroot = {
    device = "/nev/sda2";
    preLVM = true;
  };
};

The Nixpkgs pull request template has a checkbox "Determined the impact on package closure size (by running nix path-info -S before and after)" but there is only 4 instances of path-info in the Nix manual (and none in the Nixpkgs manual).

nix --help says

path-info        query information about store paths

so the command works at the bottom but what switches are available for example? From the examples, -r and -S is valid but where are they documented?

nix path-info -rS $(readlink -f $(which vim))

Docker based PHP development environment

Nix ecosystem introduction ("Talk at my company")

Declarative package management for normal users

What is the relationship between Disnix and NixOps?

NIX + BAZEL = FULLY REPRODUCIBLE, INCREMENTAL BUILDS

Web Hosting with IPFS and Nix

How we use Nix at IOHK?

Why is rec needed here, and not in the example under 2.6.1?

Based on what I saw with other examples, rec is usually included when a_package.override is used inside packageOverrides. But why?

The entire chapter 3 is worth reading. Great resource on what a package (or more broadly, a component) is in regards to Nix.

Specifically "3.1 What is a component?"

There is no officially prescribed reading order of the Nix manuals, but it's safe to say that one should start this, the Nix manual. Then it would be prudent to briefly describe what a package is in the context of Nix and/or (at east) link to the definition.

I like how Dolstra's thesis has an entire section on the topic (that is, on the more general concept of components).

It is clear now that there can be multiple versions of the same package in the store, but how does one call them (e.g., if is an executable application)? Simply by using the full Nix store path (and create and manage one's own symlinks, with stow or manually)??

haskell-overridez is a tool and library of nix functions that simplify the use of overrides while developing haskell projects with nixpkgs.

A nix-based Haskell project manager

Eelco Dolstra's seminal paper

"fixed-point", "fix point" seems to be most important concept in Nix, because overrides, overridePackages, overlays are built using it.

• Nix Pill - Chapter 17. Nixpkgs Overriding Packages (the first place I saw this concept properly described)
• Nixpkgs issue - Add pkgs.overrideWithScope#44196 (best high level summary of Nixpkgs ever read)
• How to Fake Dynamic Binding in Nix
• Comment by zimbatm on NixOS Discourse
• nixpkgs/lib/fixed-points.nix

• [Nix-dev] Introducing Nixpkgs Overlays
• Video - Initial announcement of overlays by Nicolas B. Pierron (nbp)
• Overlays implementation by Nicolas B. Pierron (nbp)
• https://nixos.wiki/wiki/Overlays
• Gabriel439/haskell-nix issue #58
• NixOS: The DOs and DON’Ts of nixpkgs overlays
• Nix overlays and override pattern Stackoverflow thread

Should packageOverrides be deprecated in favor of Overlays?

• Nixpkgs issue - Deprecate packageOverrides? #43266
• Nixpkgs pull request - [WIP] Deprecate packageOverrides #43560

Where is buildEnv documented?

• How buildEnv builtin function works? Stackoverflow thread
• mkShell vs. buildEnv?
• NixOS Wiki, Documentation Gaps, section What's the relationship between nix-env and buildEnv
• How to copy a nix-env profile using nix-copy-closure?
• https://news.ycombinator.com/item?id=11803558

Find out where stdenv.lib functions are documented.

builtins.elem x xs

Return true if a value equal to x occurs in the list xs, and false otherwise.

builtins.parseDrvName s

Split the string s into a package name and version. The package name is everything up to but not including the first dash followed by a digit, and the version is everything following that dash. The result is returned in a set { name, version }. Thus, builtins.parseDrvName "nix-0.12pre12876" returns { name = "nix"; version = "0.12pre12876"; }.

What is a package in this context? That is, the callback's pkg parameter. Is it a derivation?

If I understood it correctly, whenever referencing other packages as inputs, those are actually derivations, that are just attribute sets.

Here's a Rolling Stone article on the matter.

His Wikipedia page is here.