Kartoffeldruck: A hackable static site generator

14 Feb 2015 / posted by Nikku

There is probably around one thousand three hundred thirty seven static site generators out there. So why build a new one? Because it could be a 300 lines of code only, all-in-one, swiss army knife, hackable one. Kind of.

Static site generators are awesome. They take your plain text files, combine them with a layout skeleton and compile the whole stuff to HTML. Upload everything on GitHub pages or your favorite web server and you are good.

Must have Generator Features

There is only a few things a good static site generators needs to be capable of doing:

  • Markdown parsing
  • Meta data extraction, i.e. front matter support
  • Layouting / templating
  • Content grouping / tagging / categorization
  • Pagination

Checking the Node.js/npm ecosystem there probably exist a few tools that provide the mentioned feature set already. Good candidates are hexo or metalsmith. Why not use these? The reason is hackability.

As a programmer I would like to be able to control the way my pages are generated beyond JSON configurations and front matter annotations. In fact, I would like to have access to pages and meta data, too. Based on that I can easily implement grouping, categorization, table of content generation and the like.

A hackable Site Generator

Say hello to kartoffeldruck, a static site generator you program rather than configure. The library exposes only a single command line:

$ kartoffeldruck --help
generate a static site using the ancient kartoffeldruck principle

It will pick up a kartoffeldruck.js file in the current working directory. The file must expose a single function that receives the generator instance as an argument.

module.exports = function(druck) {

  // optionally initialize source, dest and template directory
  druck.init({ });

  // ... generate site

druck is an instance of kartoffeldruck and provides the crucial APIs to implement the actual site generation.

The Configuration

In its simples form a kartoffeldruck.js it will initialize the generator instance and generate posts from one folder into another.

The Kartoffeldruck#generate method accepts a globbing expression, allowing you to generate a number of pages on the fly. Specify :varName parameters in the dest option to generate destination file names based on source file meta data.

  source: 'posts/*.md',
  dest: ':name/index.html'

Meta Data Extraction and Filtering

You may also filter available pages yourself. This gives you access to each pages meta data.

var posts = druck.files('posts/*.md');

One option would be to filter posts based on certain criteria, such as a published field.

var unpublished = posts.filter(function(p) { return !p.published; });

// ... do stuff with unpublished posts

Aggregation / Pagination

Posts may be aggregated by passing them as a local variable to an aggregation template:

  source: '_drafts.html',
  dest: '_drafts/:page/index.html',
  locals: { items: unpublished },
  paginate: 5

The paginate variable decides on the number of posts to put on each page. :page/ will replace to an empty string on the first page, giving you nice overview urls.

The local variable items will be paginated based on the specified number of max items.

Tag Clouds

Tag clouds can easily be generated from the available posts meta data:

var tagged = {};

posts.forEach(function(p) {
  (p.tags || []).forEach(function(tag) {
    var t = tagged[tag] = (tagged[tag] || { tag: tag, items: [] });

The variable can be made available to all templates, i.e. to render a tag cloud in the side bar:

druck.config.locals.tagged = tagged;

It could also be used to generate overview pages, grouping posts by tag:

var forEach = require('lodash/object/forEach');

forEach(tagged, function(t) {
    source: '_tagged.html',
    dest: '_tagged/:tag/:page/index.html',
    locals: t,
    paginate: 5

Note how you are able to include you favorite utility library to perform matching, data extraction or object iteration, too.

Templating Capabilities

Kartoffeldruck supports Markdown files out of the box. For more complex page content and/or templating purposes it uses Nunjucks. Nunjucks is a templating language with a clean syntax, great expressiveness, impressive generation speed and extensibility. It should be sufficient for all the generation use cases out there. No need to include ejs, handlebars or insert any other obscure templating dialect here.

Local Variables

Templates provide access to all local variables, including default locals exposed through the configuration. The following configuration will evaluate to <h1>FOO</h1> in the generated HTML file:

title: FOO

# \{\{ title }}


Inside templates and Markdown files the relative helper is available. It inserts a link relative to the specified destination site, based on the location the current file is generated to.

Check the [about page](\{\{ relative('about') }})

It will always generate correct reference to the about page, independent of whether the page is currently aggregated or generated as a single page.

The render helper on the other hand allows you to include another page within the pages output. This is useful for aggregations.

\{\% for item in items %}
  <div class="item">
    <h1><a href="\{\{ relative(item.name) }}">\{\{ item.title }}</a></h1>

    \{\{ render(item) \}\}
\{\% endfor %}

Default Variables

The assets variable always points to a assets/ directory that is supposed to contain all non-HTML files.

![a picture](\{\{ assets }}/blog/some-picture.jpg)

You have to take care about copying the images yourself. kartoffeldruck only knows about HTML assets.


During pagination kartoffeldruck makes the page object available in templates.

It contains the links to the next as well as previous page.

\{\% if page.previousRef != null %}
  <a href="\{\{ relative(page.previousRef) }}">previous</a>
\{\% endif %}

\{\% if page.nextRef != null %}
  <a href="\{\{ relative(page.nextRef) }}">next</a>
\{\% endif %}

Template Inheritance

Nunjucks supports a sophisticated template inheritance scheme. Let us say a base template, i.e. base.html implements the general skeleton.

    \{\% block header %}\{\% endblock %}
    \{\% block body %}default body text\{\% endblock %}

Child templates can choose to extend from it, filling one or more placeholder blocks, i.e. to include custom scripts, meta data or page content:

\{\% extends "base.html" %}

\{\% block body %}
  Specialized body content
\{\% endblock %}

Implicit Layouts

When using front matter definitions, the layout variable defines the parent template to inherit from. This makes the above code equivalent to

layout: base

\{\% block body %}
  Specialized body content
\{\% endblock %}

Markdown pages will be automatically rendered into a item_body block.

There is More

This gave a quick overview about the basic concepts of kartoffeldruck. The project is available on GitHub and features an example project that generates a simple blog. To inspect a more elaborated example, browse the sources of this site.