Up until now I’ve used Lektor, a Python based static site generator.
Since discovering Tailwind CSS I’ve wanted to move as many projects to rely on that as possible. It’s an absolute joy to use, makes CSS more maintainable, faster and just easier to reason about. Since discovering Svelte, then Sapper, then SvelteKit (the new Sapper powered by Vite, similar to Next.js), I’ve started using Svelte[-Kit] for many of my projects:
SvelteKit is still beta but API is maturing quickly. Those who know Next.js know what it’s about. It helps you deliver a silky smooth user experiencing by giving complete control over what happens on the backend before a page is delivered, what happens when the page loads as well as classic component hierarchies for interactivity once the page has been hydrated in the browser. It’s the marriage of the SEO friendly server template HTML and more modern Single-Page Apps world — and one where it doesn’t feel you sacrifice anything to get the best of both worlds. SvelteKit has filesystem based routing which lends itself nicely to blogs since each file or folder becomes a route.
Internally SvelteKit uses Vite which in turn uses esbuild
to compile JavaScript to bundles
supported by modern browsers. SvelteKit does a bunch of other nice things like catch link events to
switch pages using an internal router, resulting in super fast navigation that still preserves
scrollbar position.
Recently Tailwind CSS has adopted (an optional for now) JIT compiler. Before JIT every single class from the Tailwind library was loaded in the browser in development. With the increasingly large scope of Tailwind this file was starting to reach the threshold of what browsers can comfortably handle for a single page. During the build phase a code scanner would detect the classes you made use of and purge the rest to decrease filesize. Now with JIT however, this purge step has been sped up significantly to the point of being comfortably performed on every file change - resulting in only the classes being used being loaded in the browser, even in development. This allows for an even bigger total composable class library, allowing all variants, even new ones, being available from the get-go, — and the build phase becomes much much faster!
Tailwind is an incredibly exciting project and new way to style.
Head on over to github.com/jokull/blog to inspect the repo. I’ll be going through a few things to explain how this came together. I picked up a number of things from the Gitpod website code they were kind enough to open source.
.
├── README.md
├── jsconfig.json
├── package-lock.json
├── package.json
├── postcss.config.cjs
├── src
│ ├── app.html
│ ├── app.postcss
│ ├── components
│ │ ├── PhotoCaption.svelte
│ │ └── blog-post.svelte
│ ├── global.d.ts
│ ├── hooks.js
│ ├── lib
│ │ └── prism-theme.postcss
│ └── routes
│ ├── $layout.svelte
│ ├── 2018-in-review.md
│ ├── 2019-in-review.md
│ ├── 2020-in-review.md
│ └── ...
├── static
│ ├── blog
│ │ ├── IMG_0567.JPG
│ │ ├── IMG_2528.JPG
│ │ └── ...
│ ├── favicon.svg
│ ├── keybase.txt
│ └── profile.jpg
├── svelte.config.cjs
├── tailwind.config.cjs
└── wrangler.toml
The config is a little bit more complex to add Svelte support to the code highlighter, but this might give you an idea.
/* svelte.config.cjs */
const preprocess = require('svelte-preprocess');
const adapter = require('@sveltejs/adapter-static');
const { mdsvex } = require('mdsvex');
const headings = require('remark-autolink-headings');
const slug = require('remark-slug');
/** @type {import('@sveltejs/kit').Config} */
module.exports = {
extensions: ['.svelte', '.md'],
preprocess: [
preprocess({
postcss: true
}),
mdsvex({
extensions: ['.md'],
layout: './src/components/blog-post.svelte',
remarkPlugins: [slug, headings]
})
],
kit: {
// hydrate the <div id="svelte"> element in src/app.html
target: '#svelte',
adapter: adapter()
}
};
Tailwind uses JIT mode which increases the number of variant classes available out of the box (for
example the last:
variant used to remove the bottom border from the last blog post in the index).
Two plugins are used; Typography that adds
a some default styling to make Markdown sections readable, and
Inter that tweaks sizing and spacing for the
super nice Inter font by rsms.
Markdown is handled by mdsvex which lets you mix Markdown, various Remark plugins to add features and Svelte components together. Highlighting of code blocks is something that comes with the default mdsvex configuration and works out of the box.
Markdown doesn’t let you do captions below photos, something I wanted to support:
/* src/components/PhotoCaption.svelte */
<script>
export let url;
export let caption;
export let wider = false;
</script>
<div class={`${wider ? 'wider' : ''} mb-4`}>
<img class="!mb-2" src={url} alt={caption} />
<div class="text-sm text-center">{caption}</div>
</div>
An example post might be:
---
title: My post
date: 2020-1-1
---
<script>
import PhotoCaption from "../components/PhotoCaption.svelte";
</script>
# Markdown header
<PhotoCaption url='/blog/biarritz.jpg' caption='Family in Biarritz.' />
Markdown paragraph
And the photo renders like this:
The blog svelte page used to render Markdown looks like this:
/* src/components/blog-post.svelte */
<script>
import '$lib/prism-theme.postcss';
export let title;
export let date;
export let image;
let dateDisplay = new Date(Date.parse(date)).toLocaleDateString(undefined, {
year: 'numeric',
month: 'long',
day: 'numeric'
});
</script>
<svelte:head>
<title>{title} - Jökull Sólberg</title>
</svelte:head>
<div class="prose prose-sm sm:prose page max-w-none sm:max-w-none">
<div class="py-8 sm:py-10 sm:text-center">
<div class="font-bold text-4xl mb-4">{title}</div>
<div class="text-sm">{dateDisplay}</div>
</div>
{#if image}
<img src={`/blog/${image}`} class="full-bleed" alt="Banner" />
{/if}
<slot />
</div>
You have may have noticed a custom Prism theme added there for the code highlighting library. I decided to make it a PostCSS file so I can add Tailwind classes where needed.
To make posts available to the index page I’m adding it via a SvelteKit session hook (stolen from Gitpod):
/* src/hooks.js */
/** @type {import('@sveltejs/kit').GetSession} */
export const getSession = async () => {
const markdownFiles = import.meta.glob('/src/routes/*.md');
const posts = await Promise.all(
Object.entries(markdownFiles).map(async ([path, page]) => {
const { metadata } = await page();
let pathComponents = path.split('/');
const filename = pathComponents.pop();
const slug = filename.split('.md', 1)[0];
return { ...metadata, filename, slug };
})
);
posts.sort((a, b) => Date.parse(b.date) - Date.parse(a.date));
return {
posts
};
};
… which then becomes available in src/routes/index.svelte
:
<script context="module">
export const prerender = true;
export async function load({ session }) {
const posts = session.posts;
return { props: { posts } };
}
</script>
/* trimmed ... */
GitHub Pages is a great service and I love that it comes free with all repos. The gh-pages
utility
makes it very easy to push a build directory (usually .gitignored on the main branch) to a
gh-pages
branch, which GitHub automatically uses for the static build. By default static sites are
deployed into a /{repository-name}
subpath which can easily break links and references assuming a
root /
deployment.
Another issue I’ve had is with custom domains. The combination of DNS lag, CNAME top level file, having to opt out of Jekyll support, gh-pages purging dotfiles by default, the flakey GitHub Pages settings tab … it’s all a bit wonkey. When it works it’s nice, but when it doesn’t it’s very frustrating.
Cloudflare Worker free-tier to the rescue! What CF Workers allow you to do is host code on hundreds of edge servers with response times measured in milliseconds. It gives your pages an extremely snappy feel. SvelteKit even comes with adapter allowing you to ship backend code to hydrate pages and perform other backend tasks. Some people call this serverless, but this is really all about pushing computing - and not just static assets - to edge servers to deliver a fast user experience.
Even though SvelteKit comes with an adapter to host sites on Cloudflare Workers, I opted for the static adapter to output a build that is then pushed to Cloudflare. That’s mostly because I don’t have the need for any backend processing, although it’s tempting to test the Cloudflare Worker just to see how it works.
Setting up a site on Cloudflare means transitioning DNS hosting to their servers. I was confused as to how you point a CNAME to a worker.
Apparently you just set a CNAME pointing to 100::
— an IPv6 address. 🤨
Then in the Clouflare wrangler.toml
project file set the route
thusly:
name = "blog"
type = 'webpack'
account_id = '...'
route = 'www.solberg.is/*'
zone_id = '...'
usage_model = ''
workers_dev = true
target_type = "webpack"
site = {bucket = "./build",entry-point = "workers-site"}
The site.bucket
key points to ./build
which is the default static output directory of the
SvelteKit static adapter.
This GitHub Action performs the build-and-deploy dance so all I need to do is push new .md files to the main branch:
name: cloudflare-publish
on:
push:
branches:
- master
workflow_dispatch:
jobs:
deploy:
runs-on: ubuntu-18.04
steps:
- uses: actions/checkout@v2
- name: Setup Node
uses: actions/setup-node@v2
with:
node-version: '16.x'
- name: Get npm cache directory
id: npm-cache-dir
run: |
echo "::set-output name=dir::$(npm config get cache)"
- uses: actions/cache@v2
id: npm-cache # use this to check for `cache-hit` ==> if: steps.npm-cache.outputs.cache-hit != 'true'
with:
path: ${{ steps.npm-cache-dir.outputs.dir }}
key: ${{ runner.os }}-node-${{ hashFiles('**/package-lock.json') }}
restore-keys: |
${{ runner.os }}-node-
- run: npm install
- run: npm run build
- name: Publish on Cloudflare
uses: cloudflare/[email protected]
with:
apiToken: ${{ secrets.CF_API_TOKEN }}
You’ll need to copy your Cloudflare API key to the GitHub repo secrets for this to work.
I wanted Prettier to hard-wrap Markdown files, but not touch double-whitespace at the end of lines
since that can indicate a <br>
— just set proseWrap: "always"
.
{
"useTabs": true,
"singleQuote": true,
"trailingComma": "none",
"printWidth": 100,
"proseWrap": "always"
}
Adding an RSS XML is easy in SvelteKit. Simply create a routes/feed.xml.js
file that exports a get
function that returns a {body: xmlString}
. Mine looks like this:
import getPosts from '$lib/getPosts';
const siteUrl = 'https://www.solberg.is';
const renderXmlRssFeed = (posts) => `<?xml version="1.0"?>
<rss version="2.0">
<channel>
<title>Jökull Sólberg</title>
<link>${siteUrl}</link>
${posts
.map(
(post) => `
<item>
<title>${post.title}</title>
<link>${siteUrl}/${post.slug}</link>
<pubDate>${new Date(post.date).toUTCString()}</pubDate>
</item>
`
)
.join('\n')}
</channel>
</rss>`;
export async function get() {
const feed = renderXmlRssFeed(await getPosts());
return {
body: feed,
headers: { 'content-type': 'application/rss+xml' }
};
}
npm add --save-dev [email protected]
- 3.0 does not work
with mdsvex..prose
class on your top level
element that stores the Markdown output in your DOM. There is currently no neat way to have other
Tailwind classes (such as ones on the PhotoCaption
component) take precedence. I resorted to
using !important
hack, which the JIT plugin makes easy to do actually: class="!mb-2"