depot/third_party/nixpkgs/pkgs/development/misc/resholve/README.md
Default email 504525a148 Project import generated by Copybara.
GitOrigin-RevId: bd645e8668ec6612439a9ee7e71f7eac4099d4f6
2024-01-02 12:29:13 +01:00

12 KiB

Using resholve's Nix API

resholve replaces bare references (subject to a PATH search at runtime) to external commands and scripts with absolute paths.

This small super-power helps ensure script dependencies are declared, present, and don't unexpectedly shift when the PATH changes.

resholve is developed to enable the Nix package manager to package and integrate Shell projects, but its features are not Nix-specific and inevitably have other applications.

This will hopefully make its way into the Nixpkgs manual soon, but until then I'll outline how to use the functions:

  • resholve.mkDerivation (formerly resholvePackage)
  • resholve.writeScript (formerly resholveScript)
  • resholve.writeScriptBin (formerly resholveScriptBin)
  • resholve.phraseSolution (new in resholve 0.8.0)

Fair warning: resholve does not aspire to resolving all valid Shell scripts. It depends on the OSH/Oil parser, which aims to support most (but not all) Bash. resholve aims to be a ~90% sort of solution.

API Concepts

The main difference between resholve.mkDerivation and other builder functions is the solutions attrset, which describes which scripts to resolve and how. Each "solution" (k=v pair) in this attrset describes one resholve invocation.

NOTE: For most shell packages, one invocation will probably be enough:

  • Packages with a single script will only need one solution.
  • Packages with multiple scripts can still use one solution if the scripts don't require conflicting directives.
  • Packages with scripts that require conflicting directives can use multiple solutions to resolve the scripts separately, but produce a single package.

resholve.writeScript and resholve.writeScriptBin support a single solution attrset. This is basically the same as any single solution in resholve.mkDerivation, except that it doesn't need a scripts attr (it is automatically added). resholve.phraseSolution also only accepts a single solution--but it does still require the scripts attr.

Basic resholve.mkDerivation Example

Here's a simple example of how resholve.mkDerivation is already used in nixpkgs:

{ lib
, fetchFromGitHub
, resholve
, bash
, coreutils
, goss
, which
}:

resholve.mkDerivation rec {
  pname = "dgoss";
  version = "0.4.1";

  src = fetchFromGitHub {
    owner = "goss-org";
    repo = "goss";
    rev = "refs/tags/v${version}";
    hash = "sha256-dpMTUBMEG5tDi7E6ZRg1KHqIj5qDlvwfwJEgq/5z7RE=";
  };

  dontConfigure = true;
  dontBuild = true;

  installPhase = ''
    sed -i '2i GOSS_PATH=${goss}/bin/goss' extras/dgoss/dgoss
    install -D extras/dgoss/dgoss $out/bin/dgoss
  '';

  solutions = {
    default = {
      scripts = [ "bin/dgoss" ];
      interpreter = "${bash}/bin/bash";
      inputs = [ coreutils which ];
      keep = {
        "$CONTAINER_RUNTIME" = true;
      };
    };
  };

  meta = with lib; {
    homepage = "https://github.com/goss-org/goss/blob/v${version}/extras/dgoss/README.md";
    changelog = "https://github.com/goss-org/goss/releases/tag/v${version}";
    description = "Convenience wrapper around goss that aims to bring the simplicity of goss to docker containers";
    license = licenses.asl20;
    platforms = platforms.linux;
    maintainers = with maintainers; [ hyzual anthonyroussel ];
  };
}

Basic resholve.writeScript and resholve.writeScriptBin examples

Both of these functions have the same basic API. The examples are a little trivial, so I'll also link to some real-world examples:

resholvedScript = resholve.writeScript "name" {
    inputs = [ file ];
    interpreter = "${bash}/bin/bash";
  } ''
    echo "Hello"
    file .
  '';
resholvedScriptBin = resholve.writeScriptBin "name" {
    inputs = [ file ];
    interpreter = "${bash}/bin/bash";
  } ''
    echo "Hello"
    file .
  '';

Basic resholve.phraseSolution example

This function has a similar API to writeScript and writeScriptBin, except it does require a scripts attr. It is intended to make resholve a little easier to mix into more types of build. This example is a little trivial for now. If you have a real usage that you find helpful, please PR it.

{ stdenv, resholve, module1 }:

stdenv.mkDerivation {
  # pname = "testmod3";
  # version = "unreleased";
  # src = ...;

  installPhase = ''
    mkdir -p $out/bin
    install conjure.sh $out/bin/conjure.sh
    ${resholve.phraseSolution "conjure" {
      scripts = [ "bin/conjure.sh" ];
      interpreter = "${bash}/bin/bash";
      inputs = [ module1 ];
      fake = {
        external = [ "jq" "openssl" ];
      };
    }}
  '';
}

Options

resholve.mkDerivation maps Nix types/idioms into the flags and environment variables that the resholve CLI expects. Here's an overview:

Option Type Containing
scripts <list> scripts to resolve ($out-relative paths)
interpreter "none" <path> The absolute interpreter <path> for the script's shebang. The special value none ensures there is no shebang.
inputs <packages> <paths> A list of packages and string paths to directories/files to resolve external dependencies from.
fake <directives> pretend some commands exist
fix <directives> fix things we can't auto-fix/ignore
keep <directives> keep things we can't auto-fix/ignore
lore <directory> control nested resolution
execer <statements> modify nested resolution
wrapper <statements> modify nested resolution
prologue <file> insert file before resolved script
epilogue <file> insert file after resolved script

Controlling resolution with directives

In order to resolve a script, resholve will make you disambiguate how it should handle any potential problems it encounters with directives. There are currently 3 types:

  1. fake directives tell resholve to pretend it knows about an identifier such as a function, builtin, external command, etc. if there's a good reason it doesn't already know about it. Common examples:
    • builtins for a non-bash shell
    • loadable builtins
    • platform-specific external commands in cross-platform conditionals
  2. fix directives give resholve permission to fix something that it can't safely fix automatically. Common examples:
    • resolving commands in aliases (this is appropriate for standalone scripts that use aliases non-interactively--but it would prevent profile/rc scripts from using the latest current-system symlinks.)
    • resolve commands in a variable definition
    • resolve an absolute command path from inputs as if it were a bare reference
    • force resholve to resolve known security wrappers
  3. keep directives tell resholve not to raise an error (i.e., ignore) something it would usually object to. Common examples:
    • variables used as/within the first word of a command
    • pre-existing absolute or user-relative (~) command paths
    • dynamic (variable) arguments to commands known to accept/run other commands

NOTE: resholve has a (growing) number of directives detailed in man resholve via nixpkgs.resholve (though protections against run-time use of python2 in nixpkgs mean you'll have to set NIXPKGS_ALLOW_INSECURE=1 to pull resholve into nix-shell).

Each of these 3 types is represented by its own attrset, where you can think of the key as a scope. The value should be:

  • true for any directives that the resholve CLI accepts as a single word
  • a list of strings for all other options

This will hopefully make more sense when you see it. Here are CLI examples from the manpage, and the Nix equivalents:

# --fake 'f:setUp;tearDown builtin:setopt source:/etc/bashrc'
fake = {
  # fake accepts the initial of valid identifier types as a CLI convenience.
  # Use full names in the Nix API.
  function = [ "setUp" "tearDown" ];
  builtin = [ "setopt" ];
  source = [ "/etc/bashrc" ];
};

# --fix 'aliases $GIT:gix /bin/bash'
fix = {
  # all single-word directives use `true` as value
  aliases = true;
  "$GIT" = [ "gix" ];
  "/bin/bash";
};

# --keep 'source:$HOME /etc/bashrc ~/.bashrc'
keep = {
  source = [ "$HOME" ];
  "/etc/bashrc" = true;
  "~/.bashrc" = true;
};

Note: For now, at least, you'll need to reference the manpage to completely understand these examples.

Controlling nested resolution with lore

Initially, resolution of commands in the arguments to command-executing commands was limited to one level for a hard-coded list of builtins and external commands. resholve can now resolve these recursively.

This feature combines information (lore) that the resholve Nix API obtains via binlore (nixpkgs, repo), with some rules (internal to resholve) for locating sub-executions in some of the more common commands.

  • "execer" lore identifies whether an executable can, cannot, or might execute its arguments. Every "can" or "might" verdict requires:

    • an update to the matching rules in binlore if there's absolutely no exec in the executable and binlore just lacks rules for understanding this
    • an override in binlore if there is exec but it isn't actually under user control
    • a parser in resholve capable of isolating the exec'd words if the command does have exec under user control
    • overriding the execer lore for the executable if manual triage indicates that all of the invocations in the current package don't include any commands that the executable would exec
    • if manual triage turns up any commands that would be exec'd, use some non-resholve tool to patch/substitute/replace them before or after you run resholve on them (if before, you may need to also add keep directives for these absolute paths)
  • "wrapper" lore maps shell exec wrappers to the programs they exec so that resholve can substitute an executable's verdict for its wrapper's.

Caution: At least when it comes to common utilities, it's best to treat overrides as a stopgap until they can be properly handled in resholve and/or binlore. Please report things you have to override and, if possible, help get them sorted.

There will be more mechanisms for controlling this process in the future (and your reports/experiences will play a role in shaping them...) For now, the main lever is the ability to substitute your own lore. This is how you'd do it piecemeal:

# --execer 'cannot:${openssl.bin}/bin/openssl can:${openssl.bin}/bin/c_rehash'
execer = [
  /*
    This is the same verdict binlore will
    come up with. It's a no-op just to demo
    how to fiddle lore via the Nix API.
  */
  "cannot:${openssl.bin}/bin/openssl"
  # different verdict, but not used
  "can:${openssl.bin}/bin/c_rehash"
];

# --wrapper '${gnugrep}/bin/egrep:${gnugrep}/bin/grep'
wrapper = [
  /*
    This is the same verdict binlore will
    come up with. It's a no-op just to demo
    how to fiddle lore via the Nix API.
  */
  "${gnugrep}/bin/egrep:${gnugrep}/bin/grep"
];

The format is fairly simple to generate--you can script your own generator if you need to modify the lore.