Survey of Config Languages

What is a "Config Language"? What Belongs on this Page?

Config languages let you provide parameters to software, usually at runtime. For example:

• What flags should we pass to an executable?
  • --port 80 to a web server, or --port 8080?
  • -O0 to a compiler for debug builds, or -O2 for release builds?
• What environment variables? PATH, PYTHONPATH, etc.
• What kind of container or VM should a program run in?
• How should a program be scheduled? (both locally and remotely)

This page is organized by least expressive (languages for string data) to most expressive (using general-purpose languages for configuration). This doesn't imply anything about what solution you should use!

Oil is on the right of that spectrum: Hay Ain't YAML - Custom Languages for Unix Systems. It's the only config language embedded in a shell! It uses a staged evaluation model.

This page is editable, so feel free to add links below. And feel free to add notable usages, especially under Who Uses it?

Languages For String Data

• The informal INI file format expresses key-value pairs in sections.
  • It's used by git for .git/config, and I believe Mercurial.
• XML started as a document language
  • But it's often used in IDE "project files", like Eclipse and Visual Studio.
  • It has a model of tags and attributes, but no arrays, integers, booleans, etc.
• YAML is (surprisingly) the de facto control plane language of the cloud!
  • Although it has JSON-like types, it's arguably string-ish because it has type confusion, e.g. the boolean no vs the string "no".
  • Almost all continuous integration / build services use it: sourcehut, Gitlab CI, Github Actions, Circle CI, Travis CI, etc.
  • Google App Engine uses it for app.yaml
  • Kubernetes uses it, often with Go Templates (e.g. Helm)

Languages For Typed Data

These languages allow you to express structured and dynamically typed data, but they aren't programmable.

• JSON is derived from JavaScript, but it's now available in almost every language. It's used by node.js in package.json.
  • JavaScript Object Notation is a lightweight data-interchange format. It is easy for humans to read and write. It is easy for machines to parse and generate.
• TOML looks like an .INI file, but it has JSON-like types. It's used by Rust in Cargo.toml.
  • A config file format for humans. TOML aims to be a minimal configuration file format that's easy to read due to obvious semantics.
• sdlang -- appears to be used by Oracle, Bank of America, etc.?
  • SDLang is a simple and concise way to textually represent data. It has an XML-like structure – tags, values and attributes – which makes it a versatile choice for data serialization, configuration files, or declarative languages. Its syntax was inspired by the C family of languages

Programmable String-ish Languages

• Go templates are for arbitrary text and HTML.
  • They are often used to generate variants of a YAML file.
• M4 is used by autotools to generate shell and Make.
• CMake generates Make and Ninja files.

Programmable Typed Data

• UCL (Universal Configuration Language)
  • UCL is heavily infused by nginx configuration as the example of a convenient configuration system. However, UCL is fully compatible with JSON format and is able to parse json files.
  • Who uses it?
• HCL (Hashicorp Configuration Language) is used by Terraform for AWS. Similar to UCL.
  • HCL is a toolkit for creating structured configuration languages that are both human- and machine-friendly, for use with command-line tools. Although intended to be generally useful, it is primarily targeted towards devops tools, servers, etc.
• UCG (Universal Grammar for Configuration) is designed to solve the "templating" problem for JSON, YAML, etc.
  • Templates can be difficult to manage without introducing hard to see errors in the serialization format they are generating. Most templating engines aren't aware of the format they are templating
  • Who uses it?
• Nix Expressions Language is used to write Nix package definitions.
• Jsonnet - The Data Templating Language.
  • Generate config data; Side-effect free; Organize, simplify, unify; Manage sprawling config
• Starlark
  • Starlark is a [dialect of Python] intended for use as a configuration language. It was designed for the Bazel build system
• Cue
  • Validate, define, and use dynamic and text-based data
  • The Logic of CUE -- Types are Values; The Value Lattic
  • CUE History -- Although it is a very different language, the roots of CUE lie in GCL, the dominant configuration language in use at Google. It was originally designed to configure Borg, the predecessor of Kubernetes.
  • Who uses it?
• Dhall
  • Dhall is a programmable configuration language that you can think of as: JSON + functions + types + imports
  • Note: Dhall isn't turing complete, but this restriction has no useful engineering properties.
    • Unconstrained side effects are the thing that matter for configuration, not computation.
    • Analogy: a language being context-free has no useful engineering properties. But being regular has a very useful engineering property: it can be recognized in constant space and linear time (with a NFA/DFA).
  • Who uses it?

Proprietary / non-open-source: Google's BCL

Internal DSLs - Using a General Purpose Language for Configuration

• Hay Ain't YAML - Custom Languages for Unix Systems. The only config language that's embedded in a shell! It uses a staged evaluation model.
• G-Expressions in Guile Scheme are used by the Guix distro
• Ruby blocks in Vagrant, Rake, etc.
• Tcl commands can also be used to define data.
  • Data Definition and Code Generation in Tcl (2003, PDF)

Links

• Related surveys
  • Alternative Shells
  • Alternative Regex Syntax
• Notes on Oil: Config Dialect / Hay
• lobste.rs thread on Hay, with more comparisons