Custom Renderer

Dioxus is an incredibly portable framework for UI development. The lessons, knowledge, hooks, and components you acquire over time can always be used for future projects. However, sometimes those projects cannot leverage a supported renderer or you need to implement your own better renderer.

Great news: the design of the renderer is entirely up to you! We provide suggestions and inspiration with the 1st party renderers, but only really require processing Mutations and sending UserEvents.

The specifics:

Implementing the renderer is fairly straightforward. The renderer needs to:

  1. Handle the stream of edits generated by updates to the virtual DOM
  2. Register listeners and pass events into the virtual DOM's event system

Essentially, your renderer needs to process edits and generate events to update the VirtualDOM. From there, you'll have everything needed to render the VirtualDOM to the screen.

Internally, Dioxus handles the tree relationship, diffing, memory management, and the event system, leaving as little as possible required for renderers to implement themselves.

For reference, check out the javascript interpreter or tui renderer as a starting point for your custom renderer.

Templates

Dioxus is built around the concept of Templates. Templates describe a UI tree known at compile time with dynamic parts filled at runtime. This is useful internally to make skip diffing static nodes, but it is also useful for the renderer to reuse parts of the UI tree. This can be useful for things like a list of items. Each item could contain some static parts and some dynamic parts. The renderer can use the template to create a static part of the UI once, clone it for each element in the list, and then fill in the dynamic parts.

Mutations

The Mutation type is a serialized enum that represents an operation that should be applied to update the UI. The variants roughly follow this set:

enum Mutation {
	AppendChildren,
	AssignId,
	CreatePlaceholder,
	CreateTextNode,
	HydrateText,
	LoadTemplate,
	ReplaceWith,
	ReplacePlaceholder,
	InsertAfter,
	InsertBefore,
	SetAttribute,
	SetText,
	NewEventListener,
	RemoveEventListener,
	Remove,
	PushRoot,
}

The Dioxus diffing mechanism operates as a stack machine where the LoadTemplate, CreatePlaceholder, and CreateTextNode mutations pushes a new "real" DOM node onto the stack and AppendChildren, InsertAfter, InsertBefore, ReplacePlaceholder, and ReplaceWith all remove nodes from the stack.

Node storage

Dioxus saves and loads elements with IDs. Inside the VirtualDOM, this is just tracked as as a u64.

Whenever a CreateElement edit is generated during diffing, Dioxus increments its node counter and assigns that new element its current NodeCount. The RealDom is responsible for remembering this ID and pushing the correct node when id is used in a mutation. Dioxus reclaims the IDs of elements when removed. To stay in sync with Dioxus you can use a sparse Vec (Vec < Option

>) with possibly unoccupied items. You can use the ids as indexes into the Vec for elements, and grow the Vec when an id does not exist.

An Example

For the sake of understanding, let's consider this example – a very simple UI declaration:

rsx! {
	h1 { "count: {x}" }
}

Building Templates

The above rsx will create a template that contains one static h1 tag and a placeholder for a dynamic text node. The template contains the static parts of the UI, and ids for the dynamic parts along with the paths to access them.

The template will look something like this:

Template {
	// Some id that is unique for the entire project
	name: "main.rs:1:1:0",
	// The root nodes of the template
	roots: &[
		TemplateNode::Element {
			tag: "h1",
			namespace: None,
			attrs: &[],
			children: &[
				TemplateNode::DynamicText {
					id: 0
				},
			],
		}
	],
	// the path to each of the dynamic nodes
	node_paths: &[
		// the path to dynamic node with a id of 0
		&[
			// on the first root node
			0,
			// the first child of the root node
			0,
		]
	],
	// the path to each of the dynamic attributes
	attr_paths: &'a [&'a [u8]],
}

For more detailed docs about the structure of templates see the Template api docs

This template will be sent to the renderer in the list of templates supplied with the mutations the first time it is used. Any time the renderer encounters a LoadTemplate mutation after this, it should clone the template and store it in the given id.

For dynamic nodes and dynamic text nodes, a placeholder node should be created and inserted into the UI so that the node can be modified later.

In HTML renderers, this template could look like this:

<h1>""</h1>

Applying Mutations

After the renderer has created all of the new templates, it can begin to process the mutations.

When the renderer starts, it should contain the Root node on the stack and store the Root node with an id of 0. The Root node is the top-level node of the UI. In HTML, this is the <div id="main"> element.

instructions: []
stack: [
	RootNode,
]
nodes: [
	RootNode,
]

The first mutation is a LoadTemplate mutation. This tells the renderer to load a root from the template with the given id. The renderer will then push the root node of the template onto the stack and store it with an id for later. In this case, the root node is an h1 element.

instructions: [
	LoadTemplate {
		// the id of the template
		name: "main.rs:1:1:0",
		// the index of the root node in the template
		index: 0,
		// the id to store
		id: ElementId(1),
	}
]
stack: [
	RootNode,
	<h1>""</h1>,
]
nodes: [
	RootNode,
	<h1>""</h1>,
]

Next, Dioxus will create the dynamic text node. The diff algorithm decides that this node needs to be created, so Dioxus will generate the Mutation HydrateText. When the renderer receives this instruction, it will navigate to the placeholder text node in the template and replace it with the new text.

instructions: [
	LoadTemplate {
		name: "main.rs:1:1:0",
		index: 0,
		id: ElementId(1),
	},
	HydrateText {
		// the id to store the text node
		id: ElementId(2),
		// the text to set
		text: "count: 0",
	}
]
stack: [
	RootNode,
	<h1>"count: 0"</h1>,
]
nodes: [
	RootNode,
	<h1>"count: 0"</h1>,
	"count: 0",
]

Remember, the h1 node is not attached to anything (it is unmounted) so Dioxus needs to generate an Edit that connects the h1 node to the Root. It depends on the situation, but in this case, we use AppendChildren. This pops the text node off the stack, leaving the Root element as the next element on the stack.

instructions: [
	LoadTemplate {
		name: "main.rs:1:1:0",
		index: 0,
		id: ElementId(1),
	},
	HydrateText {
		id: ElementId(2),
		text: "count: 0",
	},
	AppendChildren {
		// the id of the parent node
		id: ElementId(0),
		// the number of nodes to pop off the stack and append
		m: 1
	}
]
stack: [
	RootNode,
]
nodes: [
	RootNode,
	<h1>"count: 0"</h1>,
	"count: 0",
]

Over time, our stack looked like this:

[Root]
[Root, <h1>""</h1>]
[Root, <h1>"count: 0"</h1>]
[Root]

Conveniently, this approach completely separates the Virtual DOM and the Real DOM. Additionally, these edits are serializable, meaning we can even manage UIs across a network connection. This little stack machine and serialized edits make Dioxus independent of platform specifics.

Dioxus is also really fast. Because Dioxus splits the diff and patch phase, it's able to make all the edits to the RealDOM in a very short amount of time (less than a single frame) making rendering very snappy. It also allows Dioxus to cancel large diffing operations if higher priority work comes in while it's diffing.

This little demo serves to show exactly how a Renderer would need to process a mutation stream to build UIs.

Event loop

Like most GUIs, Dioxus relies on an event loop to progress the VirtualDOM. The VirtualDOM itself can produce events as well, so it's important for your custom renderer can handle those too.

The code for the WebSys implementation is straightforward, so we'll add it here to demonstrate how simple an event loop is:

pub async fn run(&mut self) -> dioxus_core::error::Result<()> {
	// Push the body element onto the WebsysDom's stack machine
	let mut websys_dom = crate::new::WebsysDom::new(prepare_websys_dom());
	websys_dom.stack.push(root_node);

	// Rebuild or hydrate the virtualdom
	let mutations = self.internal_dom.rebuild();
	websys_dom.apply_mutations(mutations);

	// Wait for updates from the real dom and progress the virtual dom
	loop {
		let user_input_future = websys_dom.wait_for_event();
		let internal_event_future = self.internal_dom.wait_for_work();

		match select(user_input_future, internal_event_future).await {
			Either::Left((_, _)) => {
				let mutations = self.internal_dom.work_with_deadline(|| false);
				websys_dom.apply_mutations(mutations);
			},
			Either::Right((event, _)) => websys_dom.handle_event(event),
		}

		// render
	}
}

It's important to decode what the real events are for your event system into Dioxus' synthetic event system (synthetic meaning abstracted). This simply means matching your event type and creating a Dioxus UserEvent type. Right now, the virtual event system is modeled almost entirely around the HTML spec, but we are interested in slimming it down.

fn virtual_event_from_websys_event(event: &web_sys::Event) -> VirtualEvent {
	match event.type_().as_str() {
		"keydown" => {
			let event: web_sys::KeyboardEvent = event.clone().dyn_into().unwrap();
			UserEvent::KeyboardEvent(UserEvent {
				scope_id: None,
				priority: EventPriority::Medium,
				name: "keydown",
				// This should be whatever element is focused
				element: Some(ElementId(0)),
				data: Arc::new(KeyboardData{
					char_code: event.char_code(),
					key: event.key(),
					key_code: event.key_code(),
					alt_key: event.alt_key(),
					ctrl_key: event.ctrl_key(),
					meta_key: event.meta_key(),
					shift_key: event.shift_key(),
					location: event.location(),
					repeat: event.repeat(),
					which: event.which(),
				})
			})
		}
		_ => todo!()
	}
}

Custom raw elements

If you need to go as far as relying on custom elements/attributes for your renderer – you totally can. This still enables you to use Dioxus' reactive nature, component system, shared state, and other features, but will ultimately generate different nodes. All attributes and listeners for the HTML and SVG namespace are shuttled through helper structs that essentially compile away. You can drop in your elements any time you want, with little hassle. However, you must be sure your renderer can handle the new namespace.

For more examples and information on how to create custom namespaces, see the dioxus_html crate.

Conclusion

That should be it! You should have nearly all the knowledge required on how to implement your renderer. We're super interested in seeing Dioxus apps brought to custom desktop renderers, mobile renderers, video game UI, and even augmented reality! If you're interested in contributing to any of these projects, don't be afraid to reach out or join the community.