Screenshot of my <article> template
A few months ago I started learning Jekyll and GitHub Pages. This was a severe limitation for me due to lack of database, ability to set headers, etc.

Initial impression of Jekyll

At first, this was just a single page portfolio. It was still hosted on GitHub Pages, so I decided that I may as well learn Jekyll.

Having already been a web deeloper for about 5 years (I’ve kinda lost track), getting started was pretty easy. Just break the page apart into components located in _includes/ and load them using {% include file.ext %}. This was pretty easy, since I was already making extensive use of <template>s and loading my projects via fetch.

I already knew Markdown pretty well too, so that makes blogging pretty easy.

Simple enough, and now I’d avoid writing something in plain HTML, given the option. Just being able to open _includes/footer.html instead of scrolling in a massive HTML file makes it well worth using Jekyll. And, if it happens to be more than a single page, Jekyll also makes it easier to update common parts of a page across the entire site at once.

Having worked in PHP for quite a while now (via require), this was nothing new, but it’s still nice to have that ability on static content.

Then I started making use of variables in “front matter” in posts and all of the things that I can do with _data/ files. Suddenly, Jekyll’s lack of database is not so much of a problem so long as I have no need for INSERTs or UPDATEs. In fact, for reading data, it’s better and easier in a lot of ways since MySQL doesn’t work so well with storing complex data. Sure, you can do JOINs or convert to and from JSON, but who would rather deal with that complexity over just working with YAML or JSON files?

Images and load times

Getting decent images and in the correct size wasn’t too easy either. Since this is using GitHub Pages, I have somewhat limited storage that I don’t want to fill up too quickly. I also do not have a WYSIWYG editor and the ability to resize and optimize images the way I usually would. Also, the typical markdown way of ![alt](/path/to/image.ext) only works for a simple image and won’t allow different sizes and formats. So, I created this to automatically create responsive images in a variety of sizes, including PNG and WebP formats.

<picture>
  <source srcset="..." sizes="..." type="image/webp" />
  <source srcset="..." sizes="..." type="image/png" />
  <img src="..." alt="..." class="..." crossorigin="anonymous" />
</picture>

I did roughly the same thing for Gravatar, though that doesn’t support WebP images.

Despite all of this, I don’t score so well according to Pingdom

Pingdom score screenshot
Despite all my work, I have no control over cache headers. Still, a load time of 678ms is not bad.

CSP and external resources

Getting CSP to work was less fun because of my inability to set headers and some limitations (such as disabling report-uri and report-only whe set in a <meta>).

Google Analytics doesn’t play well with CSP unless you want to allow unsafe-inline, and I wasn’t prepared to do that. Instead, I set data-google-analytics attribute on <html>, modify the inline script to create the <script> tag automatically, setting the src id using document.documentElement.dataset.googleAnalytics. This allowed me to use JavaScript that doesn’t ever need to be touched, even if the tracking code changes. This means that it will always have the same hash, which means that I can add sha256-qlev233W0WjthlK15TmbH0lVg2G2Wtv5ABDEiB2iDnE= to my script-src and be done with it.

var script = document.createElement('script')
var src = new URL('https://www.googletagmanager.com/gtag/js');
src.searchParams.set('id', document.documentElement.dataset.googleAnalytics);
script.src = src;
document.head.appendChild(script);
window.dataLayer = window.dataLayer || [];
window.gtag = function () {
	window.dataLayer.push(arguments);
};

window.gtag('js', new Date());
window.gtag('config', document.documentElement.dataset.googleAnalytics);

I applied a similar solution for Disqus comments recently, though my CSP still complains quite a bit. It works…

var disqus_config = function () {
	this.page = this.page || {};
	var commentsEl = document.getElementById('disqus_thread');
	this.page.url = commentsEl.dataset.disqusPageUrl;
	this.page.identifier = commentsEl.dataset.disqusIdentifier;
};
(function() { // DON'T EDIT BELOW THIS LINE
var d = document, s = d.createElement('script');
s.src = 'https://' +  document.getElementById('disqus_thread').dataset.disqusPageIdentifier + '.disqus.com/embed.js';
s.setAttribute('data-timestamp', +new Date());
(d.head || d.body).appendChild(s);
})();

Closing thoughts

For the most part, Jekyll is perfectly usable for me. I can’t create an API to setup full AJAX navigation and I’m pretty much forced to use Imgur and Disqus for images and comments, but I can deal with that.

I’d love to find something else for comments though because I really don’t like the delay due to the <iframe>. the use of data: and eval for JavaScript, or my inability to customize the comments section in any way other than choosing between a light and dark theme. As soon as possible, I’m planning on moving to a JSON API.

Lack of control over caching doesn’t affect speed for users so much due to my use of a service worker handling caching. Still, my scores on those tests used to be 98% or higher.

I could see myself using Jekyll for most simple types of websites. It is, of course, perfect for blogging, but I could also see using it for things like galleries, business directories, restaurants, event listings, and pretty much anything that doesn’t require login or purchases.

And, with use of oAuth and maybe Authorize.net, I could even deal with those limitations. The only thing Jekyll (at least on GitHub Pages) can’t be used for is paid content.