As I mentioned in my introductory post, this is a static site generated with Pelican. Now, for a new project for Lynn, a dog-trainer friend - who’s been kind enough to trade her dog-whispering services for design and web work - I wanted to create another generated static site.

The benefits are manifest:

  • Fast load times.
  • Free hosting with Github Pages.
  • A simple deployment stack that won’t go down.

However, I need this site to be easy for Lynn to edit. She needs blog-like features, but also needs to easily be able to add new testimonials from customers, add new images, change her class schedule and service pricing, update her contact information, etc. Editing a Github repository, rebuilding, and pushing to the gh-pages branch is out of the question.

So how does one create a CMS-like experience for a generated static site? Prose.io! With prose.io, we can edit Markdown files in a friendly web interface, auto-commit and push to Github. Then, with a Travis-CI job, the site may be auto-rebuilt and pushed to Github Pages. Even better, we can hide the non-content sections of the repository to avoid confusion. Awesomeness! Read more about this strategy here and here. There are also several sources for ideas on using Travis CI with prose.io.

One more problem, though: all the framework-y static site generators are heavily geared toward blogs… specifically, blogs for technies. It is painful to bend a blog data model to an arbitrary content type, much less have many different data models that are all organized and rendered differently.

This is where gulp comes in. After some consideration, I realized that gulp, which I have been enjoying using for client work and a personal project, would be a perfect fit. It provides a simple way to do processing on asset pipelines, which is exactly what a static site generator does. In addition, unlike most generators out there, I would have total control over the process and wouldn’t feel like I was working with a round-hole-square-peg blackbox. I could easily use SASS, npm with Browserify, whatever template language I want, etc. Plus, I could organize the project structure in the most sensible manner for this site and fully understand how the build process works.

Fortunately, I wasn’t the only one who considered this possibility: Sean Farrell shared his experiences creating such a pipeline, and I was able to steal some of his ideas. Specifically, the ideas Sean outlined that were useful to me were:

  • Use gulp-front-matter to parse YAML “front matter” from the content documents.
  • His method of iterating over content documents and applying front-mater and processing to each one.
  • Also, I decided to follow his lead and use Swig for HTML templating, which was a pretty natural choice for me due to its similarity to Django and Jinga2.

Beyond that, I wanted my own organization and workflow:

  • SASS for styling
  • CommonJS modules with Browserify
  • My preferred project layout
  • Travis CI deployment
  • Prose.io configuration

Getting Started

If you want to follow along with the codebase, you may do so here. The live site is here.

In the following sections, I’ll skip over some details of the build process and just focus on the big things that tie everything together.

The project structure looks like this:

  • assets/
    • images/
    • scripts/
    • styles/
    • templates/
  • content/
    • pages/
    • posts/
    • testimonials/
  • .travis.yml
  • _prose.yml
  • gulpfile.js
  • package.json

Compiling a Content Type: Testimonials

As you can see above, there are three content types in this project: pages, posts, and testimonials. Pages are just one-off things, like an About or Contact page. Posts correspond to blog posts, and every testimonial gets rendered to one testimonials page. Let’s take a look at how testimonials are processed by our gulpfile.js. This code is largely borrowed from Sean Farrell, and the same format is used for blog posts:

var frontMatter = require('gulp-front-matter'),
    marked = require('gulp-marked')

gulp.task('testimonials', function () {
    return gulp.src('content/testimonials/*.md')
        .pipe(frontMatter({property: 'page', remove: true}))
        .pipe(marked())
        // Collect all the testimonials and place them on the site object.
        .pipe((function () {
            var testimonials = []
            return through.obj(function (file, enc, cb) {
                testimonials.push(file.page)
                testimonials[testimonials.length - 1].content = file.contents.toString()
                this.push(file)
                cb()
            },
            function (cb) {
                testimonials.sort(function (a, b) {
                    if (a.author < b.author) {
                        return -1;
                    }
                    if (a.author > b.author) {
                        return 1;
                    }
                    return 0;
                })
                site.testimonials = testimonials
                cb()
            })
        })())
})

First, we iterate over all the testimonials, extract the front-matter, and compile the Markdown to HTML. Next, we create an array of testimonials, sorted by author name, and store it on the global site object.

Notice that we didn’t output anything here - we just collected data that we can render later. We will later pass the site object into our templates. To see how we render them, have a gander at the source for the testimonials page.

Now let’s take a look at how the pages are rendered:

var frontMatter = require('gulp-front-matter'),
    marked = require('gulp-marked'),
    merge = require('merge-stream')

gulp.task('pages', ['cleanpages', 'testimonials'], function () {
    var html = gulp.src(['content/pages/*.html'])
        .pipe(frontMatter({property: 'page', remove: true}))
        .pipe(through.obj(function (file, enc, cb) {
            var data = {
                site: site,
                page: {}
            }
            var tpl = swig.compileFile(file.path)
            file.contents = new Buffer(tpl(data), 'utf8')
            this.push(file)
            cb()
        }))

    var markdown = gulp.src('content/pages/*.md')
        .pipe(frontMatter({property: 'page', remove: true}))
        .pipe(marked())
        .pipe(applyTemplate('assets/templates/page.html'))
        .pipe(rename({extname: '.html'}))

    return merge(html, markdown)
        .pipe(gulpif(!DEBUG, htmlmin({
            // This option seems logical, but it breaks gulp-rev-all
            removeAttributeQuotes: false,

            removeComments: true,
            collapseWhitespace: true,
            removeRedundantAttributes: true,
            removeStyleLinkTypeAttributes: true,
            minifyJS: true,
            minifyCSS: true,
            minifyURLs: true
        })))
        .pipe(gulp.dest('dist'))
        .pipe(connect.reload())
})

Notice that we have two separate asset pipelines which are merged into one stream with the merge-stream module. Both pipelines, for HTML and Markdown, extract the front-matter from each document. All Markdown documents use the assets/templates/page.html template, but the HTML documents are expected to explicitly extend from a template. The HTML is rendered using the applyTemplate function, also borrowed from Sean Farrell. Next, the use of merge-stream allows us to apply common processing to both pipelines - specifically, HTML minification and output to the dist directory.

We can then follow the same process for other asset pipelines. For instance, we could populate a services page with a list of items with corresponding prices, availability information, etc, all populated with front-matter data and Markdown content. Who needs a database?

Deployment

Skipping over the dist task that compiles the entire project, let’s take a look at the deploy task:

var deploy = require('gulp-gh-pages')

gulp.task('deploy', ['dist'], function () {
    return gulp.src('./dist/**/*')
        .pipe(deploy());
})

gulp-gp-pages makes this pretty simple; we can use the defaults, but the documentation allows us to configure the git remote and origin, if necessary.

Travis CI Configuration

Now that we have a deploy process, we want to automate it. We should be pushing a built site every time a commit is pushed upstream. Here’s the .travis.yml that handles that for us:

branches:
  only:
  - master
language: node_js
node_js:
  - "0.10"
install:
  - npm install -g gulp
  - npm install
before_script:
  - git remote set-url origin "https://${GH_TOKEN}@github.com/danielnaab/wunderdog.git"
  - git config --global user.email "dan@crushingpennies.com"
  - git config --global user.name "Travis-CI"
script:
  - gulp deploy
env:
  global:
  - secure: "jTbRauX2+9E9WbSI6pu4oXO3P60d3KriWQr7sD39JArrXFqs3ZpeT0gdycmE4OlYS/t1MY7yzKFw2MPeyIO2tl5zIBRLx77GZRwqkKi0Y4Uu5nRNkOBiPsrVD7Iq5gLuknQGbLCHf2p+1MmtQbsuEVTSkV/FWzCxk2j0nRUm2ng="

Basically, we install our dependencies, set the origin with https:// rather than git:// (because we’re pushing), with an included Github auth token. Then, we run the deploy task.

To set GH_TOKEN, we first need a Github auth token. Github provides instructions for how to do that. Next, we need to encrypt the token so it’s not committed as plain text for the world to read; we can use the travis command line tool for that:

gem install travis
travis encrypt GH_TOKEN=<Github auth token>

This will output the secure: XXXXX line seen above. Add it to your .travis.yml, and you’re good to go! Commit, grab a beer, and your site will be updated in a few.

Prose.io Configuration

Now, onto prose.io. It’s easy enough to log in to the site, and at this point, we may make edits and things will just work. Remember, though, we want the data-entry UI to be as easy as possible and not include extraneous crud. We may create a _prose.yml to define the kind of interface prose.io should generate for us:

prose:
  rooturl: 'content'
  siteurl: 'http://blog.crushingpennies.com/wunderdog/'
  media: 'content/media'
  metadata:
    content/posts:
      - name: "published"
        field:
          element: "checkbox"
          label: "Published"
          help: "Uncheck to make this post hidden."
          value: true
          on: "true"
          off: "false"
      - name: "title"
        field:
          element: "text"
          label: "Title"
          help: "The blog post title"
          placeholder: "Enter title"
      - name: "date"
        field:
          element: "text"
          label: "Publication Date"
          help: "The publication date for this post."
          placeholder: "Enter date in the form YYYY-MM-DD"
      - name: "allowComments"
        field:
          element: "checkbox"
          label: "Allow Comments"
          help: "Allow users to comment with Disqus."
          value: true
          on: "true"
          off: "false"
    content/testimonials:
      - name: "published"
        field:
          element: "checkbox"
          label: "Published"
          help: "Uncheck to make this testimonial hidden."
          value: true
          on: "true"
          off: "false"
      - name: "author"
        field:
          element: "text"
          label: "Author"
          help: "The name of the author of this testimonial."
          placeholder: "Enter author name"
      - name: "authorLocation"
        field:
          element: "text"
          label: "Author Location"
          help: "The location of the author of this testimonial."
          placeholder: "Enter author location"
      - name: "authorUrl"
        field:
          element: "text"
          label: "Author URL"
          help: "A web URL for this author, if available."
          placeholder: "Enter author URL"
      - name: "signees"
        field:
          element: "hidden"
          label: "Signees/signatories"
          help: "If there is more than one author for this testimonial, enter them here."

Summary

gulp + prose.io + Travis CI = awesome. I’m pretty pleased with the gulp asset workflow, in particular, and intend to follow these patterns for a couple more content types and hopefully streamline the process a bit.

As this project is finished out with a finalized design and interactivity with Javascript, I’ll update this post with any new ideas. Anyone else out there using gulp for static site generation?