Lab 9: Animation

In this lab, we will learn:

  • What different methods do we have to transition and animate elements and when to use each?
  • What considerations should we take into account for using animations effectively?
Table of contents

Check-off

You need to come to TA Office Hours to get checked off for this lab (any of them, no appointment needed). Please fill in the check-off form at labs/9/checkoff before your check-off. Ideally you should fill in the form right before your check-off, but it’s ok if you fill it out in advance.

Filling out the form is a necessary but not sufficient condition to get checked-off. You still need to come to office hours in person for your check-off to be processed.

You could even fill it out before you finish the lab, since we won’t look at it until your check-off, but the closer to the end of the lab you fill it out, the more meaningful your feedback will be.

Slides

What will we make?

In this lab, we will go back to the Meta page of our portfolio, and convert it to an interactive narrative visualization that shows the progress of our codebase over time.

Step 1: Evolution visualization

src/meta/+page.svelte

In this step, we will create an interactive timeline visualization that shows the evolution of our repo by allowing us to move a slider to change the date range of the commits we are looking at.

Step 1.1: Creating the filtering UI

In this step we will create a slider, bind its value to a variable, and display the date and time it corresponds to. It’s very familiar to what we did in the previous lab, except we don’t need to worry about a “no filter” state.

First, let’s create a new variable, commitProgress, that will represent the maximum time we want to show as a percentage of the total time:

let commitProgress = 100;

To map this percentage to a date, we will need a new time scale, just like we did in Lab 7 that will map commit.datetime values to the [0, 100] range.

Once we have our scale, we can easily get from the 0-100 number to a date:

$: commitMaxTime = timeScale.invert(commitProgress);

We are now ready to add our filtering UI.

  1. Create a new <label> element with a slider input and a <time> element that will display the date and time corresponding to the slider value.
  2. Add some CSS to make the slider maximum width (flex: 1 in Flexbox, 1fr column width in Grid) and to place the time element underneath the slider (otherwise differences in output value length will move the slider, which is very jarring).
  3. Bind the slider value to commitProgress
  4. Output the selected time in the <time> element using commitMaxTime.toLocaleString() similarly to how we did in Lab 7 and Lab 8. This time we need to display both the date and the time.

Feel free to use any settings you like. In the screencasts below, I use dateStyle: "long" and timeStyle: "short".

If everything went well, your slider should now be working!

Step 1.2: Filtering by commitMaxTime

Let’s now ceate a new filteredCommits variable that will reactively filter commits by comparing commit.datetime with commitMaxTime.

Similarly, create a filteredLines variable that filters data in the same way.

We can now replace commits with filteredCommits and data with filteredLines in several places:

  • The xScale domain
  • The brushed() function that updates the selectedCommits variable
  • The {#each} block that draws the circles
  • The hoveredCommit variable
  • Your summary stats

Try moving the slider and see what happens!

Step 1.3: Making the circles stable

CSS transitions are already applied to our circles since Lab 7. However, notice that when we move the slider, the circles are jumping around a lot.

This is because Svelte doesn’t know which data items correspond to which previous data items, so it does not necessarily reuse the right <circle> element for the same commit. To tell Svelte which data items correspond to which previous data items, we can use a keyed each block, with a value that uniquely identifies the data item. A good candidate for that in this case would be the commit id:

{#each commits as commit, index (commit.id) }

Just this small addition fixes the issue completely!

Step 1.4: Entry transitions with CSS

Notice that even though we are now getting a nice transition when an existing commit changes radius, there is no transition when a new commit appears.

This is because CSS transitions fire for state changes where both the start and end changes are described by CSS. A new element being added does not have a start state, so it doesn’t transition. We could use Svelte transitions for this, but we don’t need to. We can actually use CSS transitions, we just need to explicitly tell the browser what the start state should be. That’s what the @starting-style rule is for!

Inside the circle CSS rule, add a @starting-style rule (and ignore Svelte’s “Unknown at rule” warning):

@starting-style {
	r: 0;
}

If you preview again, you should notice that that’s all it took, new circles are now being animated as well!

You might notice that the largest circles and the smallest circles are both transitioning with the same duration, which means dramatically different speeds. We may decide that this is desirable: it means all circles are appearing at once. However, if you want to instead keep speed constant, you can set an --r CSS variable on each circle element with its radius, and then set the transition duration to e.g. calc(var(--r) / 100ms). You can do that only for r transitions like so:

transition: all 200ms,
            r calc(var(--r) * 100ms);

src/routes/meta/+page.svelte, src/routes/meta/Scatterplot.svelte

src/routes/meta/+page.svelte has begun to grow quite a lot, and it’s only about to get bigger. It will really help make your code more manageable to start moving reusable functionality to components. One good candidate is the commit scatterplot.

Create a new file, src/routes/meta/Scatterplot.svelte, and move the scatterplot code there. This includes:

  • The <svg> element
  • The commit tooltip
  • Any CSS styling elements in those
  • The JS dealing with dimensions, margins, scales, axes, brushing, user interaction with dots, commit selection, etc.

It should have two props:

  • commits (an array of commits)
  • selectedCommits (mostly used as output, but could be used as input as well)

We name the variable commits in order to make the scatterplot a bit more generalizable. As such, make sure to change any mentions of filteredCommits in your new file back to commits. VS Code allows you to do that safely in one go, by placing the text caret on the variable name, then pressing F2 (or right clicking and selecting “Rename Symbol”)

The final result should allow us to replace our entire <svg> and commit tooltip in src/routes/meta/+page.svelte with just:

<CommitScatterplot commits={filteredCommits} bind:selectedCommits={selectedCommits} />

Don’t forget to also move the necessary imports! I find it helpful to just copy all of them, then remove the ones VS Code highlights as unused.

Step 2: The race for the biggest file!

In this step we will create a unit visualization that shows the relative size of each file in the codebase in lines of code, as well as the type and age of each line.

Step 2.1: Creating a component for the unit visualization

src/routes/meta/FileLines.svelte, src/routes/meta/+page.svelte

To avoid bloating src/routes/meta/+page.svelte even more, let’s create a new component, src/routes/meta/FileLines.svelte, that will contain the unit visualization.

The component should take a lines prop, and will group lines of code into files internally. This means that from the outside, we’d just use it like this:

<FileLines lines={filteredLines} />

Eventually we want this to go after the scatterplot & pie chart, but for now let’s add it right after our filtering UI as that makes development faster.

Then, within the component we will group the lines by file and convert the groups to an array of {name, lines} objects like this:

let files = [];
$: {
	files = d3.groups(lines, d => d.file).map(([name, lines]) => {
		return {name, lines};
	});
}

Now that we have our files, let’s output them. While we can use D3 and generate an SVG for this, unit visualizations are one of the few cases where it can actually be easier to just use HTML and CSS, as we need to do a lot less work to manage the position of each element.

We will use a <dl> element (but feel free to make different choices, there are many structures that would be appropriate here) Let’s start simple, by just outputting filenames and number of lines:

<dl class="files">
	{#each files as file (file.name) }
		<div>
			<dt>
				<code>{file.name}</code>
			</dt>
			<dd>{file.lines.length} lines</dd>
		</div>
	{/each}
</dl>

Note that we used a keyed each block from the get go here. It’s generally a good practice to do this by default, especially when iterating over data that we expect to change, as it avoids a lot of the issues we saw earlier.

We should style the <dl> as a grid so that the filenames and line counts are aligned. The only thing that is a bit different now is that we have a <div> around each <dt> and <dd>. To prevent that from interfering with the grid we should use Subgrid:

& > div {
	grid-column: 1 / -1;
	display: grid;
	grid-template-columns: subgrid;
}

Then we can just apply grid-column: 1 to the <dt>s and grid-column: 2 to the <dd> as usual.

At this point, our “visualization” is rather spartan, but if you move the slider, you should already see the number of lines changing!

Step 2.2: Making it look like an actual unit visualization

For a unit visualization, we want to draw an element per data point (in this case, per line), so let’s do that. All we need to do is replace the contents of the <dd> element with another {#each} block that outputs a <div> for each line:

{#each file.lines as line (line.line) }
	<div class="line"></div>
{/each}

Seeing the total number of lines per file is still useful, so you may want to add it in the <dt>. I used a <small> element, gave it display: block so that it’s on its own line, and styled it smaller and less opaque.

And then add some CSS to make it look like a unit visualization:

.line {
	display: flex;
	width: .5em;
	aspect-ratio: 1;
	background: steelblue;
	border-radius: 50%;
}

Last, we want to make sure these dots wrap and are tightly packed, so we need to add some CSS for the <dd> elements to allow this:

dd {
	grid-column: 2;
	display: flex;
	flex-wrap: wrap;
	align-items: start;
	align-content: start;
	gap: .15em;
	padding-top: .6em;
	margin-left: 0;
}

At this point, we should have an actual unit visualization!

It should look something like this:

Step 2.3: Sorting files by number of lines

Our visualization is not really much of a race right now, since the order of files seems random. We need to sort the files by the number of lines they contain in descending order. We can do that in the same reactive block where we calculate files:

files = d3.sort(files, d => -d.lines.length);

Step 2.4: Varying the color of the dots by technology

Our visualization shows us the size of the files, but not all files are created equal. We can use color to differentiate the lines withn each file by technology.

Let’s create an ordinal scale that maps technology ids to colors:

let colors = d3.scaleOrdinal(d3.schemeTableau10);

Then, we can use this scale to color the dots:

<div class="line" style="--color: { colors(line.type) }">

Much better now!

Step 2.5: Dot transitions

Notice that it’s a little hard to compare which lines of each file have been added as we move the slider. If we make new elements appear with a transition, it will be much easier to see what is happening. We can use one of the Svelte predefined transitions for this. I picked scale as that makes dots appear to grow from a single point.

If you don’t need to customize the duration etc, all it takes is adding in:scale to the <div> element!

Step 2.6: Consistent colors across visualizations

Notice that we have two visualizations that use colors to represent technologies, but they use different colors for the same technologies!

To fix this, we need to allow our components (Pie.svelte and FileLines.svelte) to accept a color scale as a prop, by prepending their colors declarations with export:

export let colors = d3.scaleOrdinal(d3.schemeTableau10);

Then we create the color scale on the parent page and pass it to each of them. For example, <FileLines /> would become <FileLines colors={colors} />.

That by itself is not enough. To get consistent colors each component needs to be looking up the color for a technology in the same way. Currently, <Pie> uses technology labels, while <FileLines> uses the raw ids from the data. We can make sure the data we pass to <Pie> include an id property with the raw id and then use that to look up the color. To make the component more flexible, we could even use both:

<!-- Rest of attributes omitted -->
<path fill={ colors(d.id ?? d.label) }>

Our visualization is now way more informative!

Step 2.7: Animated race

We can now use the animate:fn directive to animate the files when they move to a new position. Svelte provides a built-in animation called flip, which sounds perfect for this use case.

We import it at the top of our component:

import { flip } from "svelte/animate";

And then we apply it to the <div> that contains each file. If we are happy with its defaults, it can be as simple as this:

<div animate:flip>

However, if we preview the animation we get the result is …less than great:

We can add a long duration (3000) and a delay (1000) (`animate:flip={{delay: 1000, duration: 3000}}) to more clearly see what is happening:

From some further investigation, it appears that Svelte is prematurely caching the geometry of the elements (before and after), which is causing the animation to be incorrect.

But we can use a workaround! The part after the animate: is just referring to a function. There is nothing special about flip, it’s just a function we import. In fact, if we alt + click on it in the import statement, VS Code will take us to the Svelte module that defines it. Instead of using it wholesale, we can import with a different name:

import { flip as originalFlip } from "svelte/animate";

We can use a reactive statement to force flip to update whenever we need it to, e.g. whenever files changes:

function getFlip () {
	return originalFlip;
}
$: flip = getFlip(files);

This tricks Svelte into thinking flip depends on files, so it will re-run whenever files changes.

If you try it now (with a reasonable duration and no delay), you will see that the animation is now correct!

For the final result, you’d likely want to either specify a duration as a function (which depends on the distance travelled) or remove the parameters (since that is the default).

Don’t like the appearance of the items when they overlap? You can apply a semi-transparent white background and a glow with the same color to the <div> via something like:

& > div {
	background: hsl(0 0% 100% / 90%);
	box-shadow: 0 0 .2em .2em hsl(0 0% 100% / 90%);
}

Step 3: Pie chart transition

lib/Pie.svelte

In lab 6, we created a pie chart that shows the distribution of edited lines per technology, and responds to filtering. We even added a transition: 300ms to our pie chart wedges, which interpolates any property the browser can interpolate. This is useful when we select a wedge, but produces rather strange results when the pie data changes as you have no doubt noticed by now.

Let’s fix it!

It will help debugging to apply fill-opacity to the pie wedges, so you can see when they overlap. Ideally, we don’t want to have gaps or overlapping wedges during the transition. In the screenshots/screencasts below, I have applied fill-opacity: 75%; to the <path> elements. You may also want to use a more distinctive color palette at this point, like schemeCategory10 or schemeTableau10. The screenshots in this lab will use schemeTableau10.

Step 3.1: Use consistent ordering

The first step is to make sure that the pie chart uses a consistent ordering of technologies, which also makes it easier to compare results across different sets of commits. By default, d3.pie() sorts the wedges by descending value, which in dynamic pie charts, is almost always undesirable. Instead, let’s sort by the data labels. This way, as long as the data labels remain consistent, the order of the pie wedges will also remain consistent.

You may notice that there is a pie.sort() function. Can’t we just use that to sort by labels? No! If we do that, the arcs would be out of sync with our data! Instead, we will disable sorting in the pie by adding .sort(null) after the call to d3.pie():

let sliceGenerator = d3.pie().value(d => d.value).sort(null);

and then we will sort pieData itself:


$: {
	pieData = d3.sort(data, d => d.label);
	// ... calculate arcData and arcs as before ...
	pieData = pieData.map((d, i) => ({...d, ...arcData[i], arc: arcs[i]}));
};

It is important to sort the data before we calculate arcData and arcs, otherwise they will be out of sync!

If you try it now, the order should already be correct, though the transition is still suboptimal:

Step 3.2: Eliminate flashing by using a keyed each block

Our technologies are now sorted, but this made the transition involve a lot of flashing. Why is that? The pie wedges are always in the same order, so nothing should be changing color!

This is because Svelte doesn’t know which data items correspond to which previous data items, so it does not necessarily reuse the same <path> element for the same data item in the new data. To tell Svelte which data items correspond to which previous data items, we can use a keyed each block, with a value that uniquely identifies the data item. A good candidate for that would be the label:

{#each pieData as d, index (d.label)}

Just this small addition fixes the issue completely!

Step 3.3: Fixing the shape transition using d3.transition()

We may have fixed the issues discussed above but the transition is, how should we put it, …suboptimal. Our shape transitions are still rather weird, especially when transitioning between pies that have different technologies. Change the transition duration to something longer (e.g. 3s) so you can more clearly see what is happening:

This is because paths are interpolated naïvely, by just matching up path segments and interpolating their parameters as numbers. Especially when it comes to arcs, that is almost never what we want!

Play with this demo to better understand how the browser’s naïve interpolation works (or rather, doesn’t work) for arcs. Open in new window ↗️

This is actually a bug in how native CSS interpolation works, and there is an open issue to fix it (opened by yours truly while authoring this very lab).

Unfortunately, there is no way to fix this with CSS transitions, and fixing it with CSS animations would require so many keyframes that it would be impractical. Instead, we will use the d3-transition module, which trades simplicity for control, by allowing us to compute the intermediate states ourselves, with JS.

The first step is to disable CSS transitions for d, so that they don’t interfere with our JS transitions. We can do this by restricting which properties the browser can auto-transition:

transition: 300ms;
transition-property: transform, opacity, fill;

When using d3-transition, together with a reactive framework like Svelte, there are several questions to answer:

  • How do we get a reference to the DOM nodes for each element we are transitioning?
  • When do we call d3.transition() on them?
  • How do we read the previous state so we can transition between the old and new states?

Let’s take them one by one.

How to get a reference to the DOM elements we are transitioning?

Because d3-transition works with DOM elements, we will need to obtain a reference to the DOM element for each path. As we know from previous labs, this is done with bind:this. The only thing that is different here is that we are binding elements inside an {#each} block, so instead of a single variable, we will need to use a data structure to store the references (array or object). The main consideration here is that we need to be able to go from pieData to the element reference and vice versa easily. Since both are uniquely identified by the data label, let’s use an object with the data label as the key and the element reference as the value:

let wedges = {};

Then, on the <path> element, add bind:this={ wedges[d.label] }.

You can check that this is populated correctly by using a $: console.log(wedges) reactive statement.

When to call d3.transition()?

Now that we have the element references, how and when do we call d3.transition() on them? Because d3-transition is based on function calls, we need to pick the right time to call it ourselves.

Let’s create a new function, transitionArcs() that will take care of the transition for us:

function transitionArcs() {
	// TODO Transition all arcs
}

Then, in the reactive block that sets pieData, add a call to transitionArcs() at the very end, so that it runs whenever the data changes.

How to read the previous state?

Finally, how do we read the previous state so we can transition between the old and new states? We will create a new variable to keep track of the old data, and we will update it every time the data changes.

First, we create a new root level variable:

let oldData = [];

Then, in the same reactive block that we are setting pieData, we set oldData = pieData right before we update pieData (so that it stores the previous value).

Great! We now have all the pieces we need to transition the arcs.

The main idea is that instead of interpolating the arcs directly, we will interpolate the angles and generate new arcs from them. This is a variant of dataspace interpolation where data is interpolated rather than attribute or CSS property values.

It is recommended to prefer CSS properties for transitions when possible, so will use the d CSS property to output the arc as a string. It may look a little weird because its name is only one letter long, but it works the same way as any other CSS property (except it only works on <path> elements). Unlike the attribute, its value is not just a series of path commands, but a path() function that a string argument with the path syntax, e.g. the CSS declaration d: path("M 0 0 L 100 0 L 100 100 L 0 100 Z") is the same as the attribute d="M 0 0 L 100 0 L 100 100 L 0 100 Z".

You should also create a new top-level variable, transitionDuration, to control the duration of the transition. JS-based transition & animation APIs typically expect durations in milliseconds. Set it to something large for now like e.g. 3000 (3 seconds), to make it easier to debug the transitions. You can also export it as a prop, so that users of the component can optionally override it.

Adding an <input type=number bind:value={transitionDuration}> to your HTML can be very useful for debugging. It allows you to override the transition at runtime (e.g. to make certain problematic transitions very slow so you can inspect what is happening without slowing everything down.

The general structure of our function will be:

function transitionArcs() {
	let wedgeElements = Object.values(wedges);

	d3.selectAll(wedgeElements).transition("arc")
		.duration(transitionDuration)
		.styleTween("d", function (_, index) {
			let wedge = this;
			// Calculations that will only be done once per element go here
			return t => {
				// t is a number between 0 and 1 that represents the transition progress
				// TODO Interpolate the angles and return the path string that corresponds to t
			};
		});
}

To calculate the arc, we need the angles. To calculate the angles, we need the data (old and new). To find the data, we need the wedge label. All we have is the index, but that’s all we need:

let label = Object.keys(wedges)[index];

Now that we have the label, we can we can find the corresponding data items in oldData and pieData:

let d = pieData.find(d => d.label === label);
let d_old = // ...

Note that we may not have both. For now, let’s just not proceed in that case:

if (!d || !d_old) {
	return;
}

We now have all four angles (start and end, old and new), and we can interpolate them. Rather than interpolating them separately, D3 actually allows us to interpolate the whole object:

// Always clone objects first, see note in https://d3js.org/d3-interpolate/value#interpolateObject
let from = {...d_old};
let to = {...d};
let angleInterpolator = d3.interpolate(from, to);

d3.interpolate() will return a function that takes a number between 0 and 1 and returns an object with the same properties as from and to, but with all properties interpolated (we only care about startAngle and endAngle but we can just ignore the rest). We can feed that object into a function that generates the path string:

let interpolator = t => `path("${ arcGenerator(angleInterpolator(t)) }")`;

Finally, we can now replace the return t => {...} line with the interpolator:

return interpolator;

If we try it out, you’ll see that it works, at least when transitioning between pies with the same technologies:

Step 3.4: Transitioning between pies with different technologies

If you try transitioning between pies with different technologies however, you will notice that it all breaks down:

How can we fix this?

Right now, we simply exit the function if we can’t find both old and new data items for a wedge. What if we generated dummy objects with the right startAngle and endAngle that we want these wedges to transition from or to? We basically want an object where obj.startAngle and obj.endAngle are both equal to whatever wedge is drawn immediately before the place our wedge is appearing or disappearing from. The exact logic is a little tricky, so we’ll just provide the code:

function getEmptyArc (label, data = pieData) {
	// Union of old and new labels in the order they appear
	let labels = d3.sort(new Set([...oldData, ...pieData].map(d => d.label)));
	let labelIndex = labels.indexOf(label);
	let sibling;
	for (let i = labelIndex - 1; i >= 0; i--) {
		sibling = data.find(d => d.label === labels[i]);
		if (sibling) {
			break;
		}
	}

	let angle = sibling?.endAngle ?? 0;
	return {startAngle: angle, endAngle: angle};
}

Now, back in our tweening function, we can use getEmptyArc() as a fallback if we can’t find the old or new data:

let from = d_old ? {...d_old} : getEmptyArc(label, oldData);
let to = // ...

The test ? iftrue : iffalse syntax is the conditional operator which helps us get one value or another based on a test, much more compactly than an if() {...} block.

We should now get a perfect transition regardless of the technologies present in the old and new pies, right? Except …we don’t. If we preview, we see that all our effort did not make an iota of difference. Why is that?

The reason is that when we transition between pies with different technologies, the wedges are added to the DOM or removed from the DOM after our code has run.

Thankfully, Svelte provides a feature dedicated to transitioning element addition or removal from the DOM: Svelte transitions! You can either use the same code for both states, via the transition:fn directive, or separate functions for entering and leaving, via the in:fn and out:fn directives. Svelte provides a bunch of predefined functions for common effects that you can just import, but in this case we’ll define our own function.

The general structure is quite similar to what we did for d3-transition, just with a different syntax:

function arc (wedge) {
	// Calculations that will only be done once per element go here
	// TODO use transitionArc() to get the data you need
	return {
		duration: transitionDuration,
		css: (t, u) => {
			// t is a number between 0 and 1 that represents the transition progress; u is 1 - t
			// TODO return CSS to be applied for the current t as a string, e.g. `fill: red`
		}
	}
}

and then use it, by adding transition:arc to our wedge <path> (we can also pass parameters, but we won’t do that this time).

There is a lot of the code we wrote in styleTween that we can reuse in animateArc. Let’s abstract it out into a new function, transitionArc() that accepts a wedge element and optionally its label (as a performance optimization):

function transitionArc (wedge, label) {
	label ??= Object.entries(wedges).find(([label, w]) => w === wedge)[0];
}

We can now move all our code from styleTween that calculates d, d_old, from, to, interpolator etc into transitionArc, which will just return an object like:

return {d, d_old, from, to, interpolator};

After d and d_old are calculated, we also want to exit early if they are not different:

if (sameArc(d_old, d)) {
	return null;
}

You should implement the sameArc() function. It should return true if either both arcs are falsy (e.g. one is undefined, the other null), or if both arcs have the same startAngle and endAngle.

Last, it’s useful to also add a type property to the object returned so we know what type of transition we are dealing with:

let type = d ? d_old ? "update" : "in" : "out";

We can now replace the code in styleTween() to use our new transitionArc() function:

let wedge = this;
let label = Object.keys(wedges)[index];
let transition = transitionArc(wedge, label);
return transition?.interpolator;

We can now use the same logic in our Svelte transition! Can you fill the arc() function so that it calls transitionArc() to compute the overall transition details and then use the returned interpolator to produce a CSS declaration for each frame? Remember that the interpolator that transitionArc() is returning gives us a CSS "path(...)" value, so all we need to do to turn this into what Svelte expects is to prepend "d: " to it.

In “out” transitions, i.e. where an existing element disappears, Svelte inverts the progression. Therefore, to get the right CSS string, you need to do do something like transition.type === "out" ? u : t instead of just t as the argument to the interpolator (i.e. transition.interpolator(transition.type === "out" ? u : t)).

The css function needs to return a CSS string, not an object! If you get an error like “Cannot read properties of undefined (reading ‘split’)” it’s likely because you are returning the wrong type.

If everything goes well, this is what you should see:

Step 3.5: Harmonizing easing across the D3 and Svelte transitions

We’re getting there, but there are some weird gaps. This is because we are trying to synchronize Svelte transitions with D3 transitions that have different parameters. While we did take care of applying the same duration across both, they use different easings! Svelte uses linear easing by default, whereas D3 uses easeCubic by default.

We can fix this in one of two ways:

  • By making the D3 transition linear as well, by adding .ease(d3.easeLinear) to the chain of function calls that make up the transition.
  • By making the Svelte transition cubic as well, by adding easing: d3.easeCubic to the object returned by arc().

Try them both and see what you prefer!

An easing function is just a function that takes a number between 0 and 1 and returns a number between 0 and 1. Therefore, you can use Svelte easing functions in D3 and vice versa.

Left: Linear. Right: Cubic.

Step 4: Scrollytelling Part 1 (commits over time)

So far, we have been progressing through these visualizations by moving a slider. However, these visualizations both tell a story, the story of how our repo evolved. Wouldn’t it be cool if we could actually tell that story in our own words, and have the viewer progress through the visualizations as they progress through the narrative?

Let’s do that!

Step 4.0: Making our page a bit wider, if there is space

Because our scrolly will involve a story next to a visualization, we want to be able to use up more space, at least in large viewports.

We can do this only for the Meta page, by adding a CSS rule where the selector is :global(body). :global() Tells Svelte not to rewrite that part of the selector, and that we will handle any scoping conflicts ourselves.

Then, within the rule, we want to set the max-width to 120ch (instead of 100ch), but only as long as that doesn’t exceed 80% of the viewport width. We can do that like this:

max-width: min(120ch, 80vw);

Step 4.1: Using a scrolly

I prepared a <Scrolly> component for you to use in this step, you will find the documentation here: svelte-scrolly. If you find any bugs, please file bug reports directly at its bug tracker.

There is an official Svelte package for this purpose: @sveltejs/svelte-scroller but it seems to only cater to scrollytelling cases where the narrative is overlaid on top of the visualization, which is not what we want here.

To use <Scrolly>, you first need to install it (via npm install svelte-scrolly), then import it at the top of your component:

import Scrolly from "svelte-scrolly";

Then you can use it like this:

<Scrolly bind:progress={ myProgressVariable }>
	<!-- Story here -->
	<svelte:fragment slot="viz">
		<!-- Visualizations here -->
	</svelte:fragment>
</Scrolly>

<svelte:fragment> is a special element that we can use when no suitable element exists, to avoid bloating our DOM with pointless wrapper elements. If there is an actual element that makes sense to use, you should use that instead!

Step 4.2: Creating a dummy narrative

Before you finish the lab, you should have something meaningful for the narrative. Don’t spend long on it; you can even generate it with ChatGPT as long as you check that the result is coherent, relevant, and tells a story that complements to the visualization next to it without simply repeating information in a less digestible format.

For now, let’s just create some dummy text that we can use to test our scrollytelling so that writing the narrative is not a blocker:

{#each commits as commit, index }
	<p>
		On {commit.datetime.toLocaleString("en", {dateStyle: "full", timeStyle: "short"})},
		I made <a href="{commit.url}" target="_blank">{ index > 0 ? 'another glorious commit' : 'my first commit, and it was glorious' }</a>.
		I edited {commit.totalLines} lines across { d3.rollups(commit.lines, D => D.length, d => d.file).length } files.
		Then I looked over all I had made, and I saw that it was very good.
	</p>
{/each}

Step 4.3: Creating a scroller for our commits over time

Move the story you just generated into the <Scrolly> component, and the scatterplot and pie chart into the <svelte:fragment slot="viz">.

Then, bind the existing commitProgress variable to the progress variable of the <Scrolly> component (<Scrolly bind:progress={commitProgress}>).

If you try it now, you should already see that the scroller is advancing the slider as you scroll!

Now that everything works, you should remove the slider as it conflicts with the scrolly, and it’s largely repeating information that the scrollbar already provides.

One thing you could do is show a date next to the actual browser scrollbar thumb, so that users have a sense of where they are in the timeline.

Step 5: Scrollytelling Part 2 (file sizes)

Step 5.1: Adding another scrolly

Create another <Scrolly> instance for the file sizes visualization, after the commit visualization. You can copy and paste the same narrative as a temporary placeholder, but as with the one about commits, you should replace it with something meaningful before you finish the lab.

The progress of this component is independent, so you will want to use a different variable for it (and similarly the corresponding maxTime variable and filtered data). Aside from that, the rest is very similar to Step 4.

To make it more visually interesting, we can place that story on the right, and the unit visualization on the left. To do that, you add --scrolly-layout="viz-first" to the component, which passes that CSS variable to it:

<Scrolly bind:progress={raceProgress} --scrolly-layout="viz-first">

You can also specify --scrolly-viz-width="1.5fr" to change the proportion of viz to story to 3:2 give it more space.

Step 5.2: Limit number of updates

Our scrolly currently looks smooth if we scroll relatively slowly, but breaks down if we scroll fast:

This is where throttling and debouncing come in. They are both methods of achieving the same goal: limiting the number of times a function is called.

  • Throttling enforces a minimum interval between subsequent calls to the same action.
  • Debouncing defers an action until a certain period of time has passed since the last user action.

The <Scrolly> component we are using supports both, via throttle and debounce props. Experiment with different values for these props (you don’t need to use both) to see what works best for your scrolly.

The final result looks similar to this:

Resources

Transitions & Animations

Tech:

Scrollytelling

Cool examples: