Let’s be honest: we all like smooth animations. It doesn’t matter if you’re looking at satisfying animations made in Blender/Houdini (or similar tools) or great code animations like the ones in the videos of the great Code Aesthetic channel on Youtube. Or smooth transitions found in some websites. At least I enjoy them and I’m regularly amazed of the possibilities CSS provides.
A few weeks ago I stumbled over shiki-magic-move, a package by the great Anthony Fu, which allows for very nice-looking animations of code like the following.
When I saw these transitions, I knew I had to implement them in this in blog in some way.
The journey to get to this result was both easier and harder than expected. Let me explain!
Trying the wrong approach
My initial idea was to use the low-level renderer that shiki-magic-move provides, in a plugin for the markdown files. The plugin should take a certain kind of markdown, parse the code in some way and then render the animation when the reader clicks a button.
As I am using expressive-code for the code-blocks, it’s actually pretty trivial to write plugins and to get the text of the code-block:
getLines()
returns an array of ExpressiveCodeLine
which I could then use to get the correct lines by looking for the before and after blocks.
In the next step I tried to pass these blocks to the MagicMoveRenderer
and to actually render. And this is where everything broke, because I did not think of a small detail, that is completely obvious. The preprocessCode hook runs on the server/at build-time, not client-side. And the render()
method of MagicMoveRenderer of course needs access to the document.
So, what do I do now? There are two options:
- Rewrite the plugin so that it does the same as the
render()
method, but withhastscript
instead of accessing thedocument
element. - Take a different route and build an interactive island.
Reverse Engineering
Either way, I have to take a look at the original repo and figure out how things are working. After some time and thinking, I decided to go for the interactive island.
There is a React wrapper for the MagicMoveRenderer
, but as I do not want to use React in this blog (nothing against React, but I want to learn new things and I’m using it at my job already), I want to use Svelte. The problem: there is no Svelte component (currently, there is a PR open to add it to the package that was opened shortly after I finished mine).
Time for a battle plan:
- I need to figure out what elements of the renderer I actually need as there are two different version, one being more lightweight than the other.
- How do I actually translate these parts from React to Svelte
- Find the best way to add both codes to the component
I’ll start with the most important thing I learned the hard way: if you try to reverse engineer something into a different framework, also double-check that you are adding the applied CSS. I did not at first, then put it into the wrong spot. Yeah.
The <MagicMoveRenderer>
Looking at the React version of the MagicMove
component, we can see that it calls the MagicMoveRenderer
component, so I guess I’ll start with this one. Before we add the logic, translating the JSX will lead to the following:
The main logic is in the useEffect
that mount the renderer, pass the options to the renderer, do pre- and post-functions if defined and call .render()
. Translated, this looks like this:
The actual MagicMove
component mostly creates the machine and passes the result to the MagicMoveRenderer
:
But I can’t really call this component in the .mdx
files yet. Time to build the wrapper.
The <AnimateCodeBlock>
I have to create a component that takes in the code to start the animation with and the code at the end. But first, we need the highlighter from shiki
since we have to pass it to the MaigcMoveRenderer
.
With the button, we can at least check if the animation works.
Looking good! Let’s get the correct code in here with props and change the logic a bit. And also, create a good looking button and move it to the same place the copy button is placed with expresive-code
.
So, if I want to call this component in my .mdx
files, it would look something like this:
client:only
Astro is, in its core, a non-interactive framework (which means very interactive with vanilla JS, but not with the ease of other JS frameworks like React or Svelte). But the one thing I love about Astro and why I use it here is the option to create interactive islands.
But Astro does not just know that this component is an interactive island - it would render the component, it’s just not interactive.
That’s why it is necessary to add the client:only
directive. If you use multiple frameworks, you would have to specify like client:only="svelte"
, but as I’m only using Svelte, I don’t have to.
In case you are wondering, there are several client directives:
- client:load: should be used for UI elements that are visible from the start and should be interactive as soon as possible
- client:idle: will be hydrated once the page has finished its initial load
- client:visible: as the name suggest, only load the element it is in the viewport
- client:only: behaves pretty similar to client:load but with the difference that the server does nothing with the component.
I need to use client:only as the MagicMoveRenderer needs access to the document
.
Adding Autoanimation
The last thing I want to add is the option to autoanimate the code block. But I can’t just animate the code when the element is mounted, as a lot of elements are probably not visible in the viewport.
What is the best way to check if an element is in the viewport? Enter the IntersectionObserver
.
This way the animation automatically starts whenever the code block is 200px off the bottom. And with the button, you, the reader, can still animate back and forth.
But there is one thing we have to keep in mind. Some people have problems with motion, they might get distracted at best or nauseous at worst. Therefore I added a check if the user prefers reduced motion (meaning if they set it in their system settings) and not do anything with the IntersectionObserver in that case.
This approach is not necessarily the best as it does not listen for live changes of the system settings, but users don’t often change this setting if at all.
Conclusion
This was a pretty fun exercise to do a bit of reverse engineering to add this feature. Thankfully, translating from React to Svelte is not too complicated.
Maybe at some point in the future I am curious enough to try to make this into a direct plugin for expressive-code
but until then, I am pretty happy with the results I got!