Crux helps you share your app's business logic and behavior across mobile (iOS and Android) and web, as a single, reusable core built with Rust.
Unlike React Native, but like Kotlin Multi-platform Mobile, the user interface layer is built natively, with modern declarative UI frameworks such as SwiftUI, Jetpack Compose and React/Vue or a Wasm based framework (like Yew) on the web.
The UI layer is as thin as it can be, and all other work is done by the shared core. The interface with the core has static type checking across languages.
Note, that Crux is experimental and currently under active development (probably not ready for use in production apps just yet). However, the master branch should always be working well, and we will try to keep the examples and documentation up to date as we go. We do think that the API has now settled, so have a play! :-)
This readme describes at a high level how Crux works, but you can find more details in the book. Or watch the introductory talk at the recent Rust Nation 2023 conference in London.
You can also join the friendly conversation on our Zulip channel.
The fundamental architectural concept is the strict separation of pure computational tasks from tasks that cause side effects. This is similar to the way Elm works.
In the above diagram, the inner "Core" is compiled and linked to the outer "Shell" on each platform as a library:
- On iOS as a native static library
- On Android as a dynamic library using Java Native Access
- In a browser as a WebAssembly module
In fact, because WebAssembly (Wasm) is one of the compilation targets, the core must remain side-effect free, due to the sandboxed nature of the Wasm runtime environment.
As such, the core is completely isolated and secure against software supply-chain attacks, as it has no access to any external APIs. All it can do is perform pure calculations and keep internal state.
Following the Elm architecture, the core defines the key component types within the application:
Event
— anenum
describing the events which the core can handleModel
— describes the internal state of the applicationViewModel
— represents information that should be displayed to the user
The former two are tied together by the update
function, familiar from Elm, Redux or other event sourcing architectures, which currently has this type signature:
fn update(
&self,
event: Event,
model: &mut Model,
capabilities: &Capabilities,
)
The job of the update
function is to process an Event
, update the model accordingly, and potentially request some side-effects using capabilities.
The enclosing platform native "Shell" is written using the language appropriate for the platform, and acts as the runtime environment within which all the non-pure tasks are performed. From the perspective of the core, the shell is the platform on which the core runs.
Following the Elm architecture, the interface with the core is message based. This means that the core is unable to perform anything other than pure calculations. To perform any task that creates a side-effect (such as an HTTP call or random number generation), the core must request it from the shell.
The core has a concept of Capabilities — reusable interfaces for common side-effects with request/response semantics. There are already a few embryonic Capability crates (Http, KeyValue, Time, Platform, and the builtin Render) — and you can write your own if you need/want to.
This means the core interface is simple:
process_event: Event -> Vec<Request>
- processes a user interaction event and potentially responds with capability requests. This is the API for the driving side in the above diagram.handle_response: (uuid, SomeResponse) -> Vec<Request>
- handles the response from the capability and potentially follows up with further requests. This is the API for the driven side in the above diagram.view: () -> ViewModel
- provides the shell with the current data for displaying user interface
Updating the user interface is considered a side-effect and is provided by the built-in Render
capability.
This design means the core can be tested very easily, without any mocking and stubbing, by simply checking the Input/Output behaviour of the three functions.
The Foreign Function Interface allowing the shell to call the above functions is provided by Mozilla's UniFFI on a mobile device, or in the browser, by wasm-pack.
In order to both send more complex data than UniFFI currently supports, and enforce the message passing semantics, all messages are serialized, sent across the boundary, then deserialized using serde-generate which also provides type generation for the foreign (non-Rust) languages.
This means that changes to types in the core, especially the Event
and Request
types, propagate out into the shell implementations and cause type errors where appropriate (such as an exhaustive match on an enum check).
Three types of message are exchanged between the application and the core.
- Messages of type
Event
are sent from the Shell to the Core in response to an event happening in the user interface (the driving side). They start a potential sequence of further message exchanges between the shell and the core. Messages are passed on unchanged. - Messages of type
Request
are sent from the Core to the Shell to request the execution of some side-effect-inducing task. The Core responds with zero or moreRequest
messages after receiving anEvent
message (the driven side). - Response messages are sent from the Shell to the Core carrying the result of an earlier request.
Request
messages contain the inputs for the requested side-effect, along with a uuid
used by the core to pair requests and their responses together.
The exact mechanics are not important, but it is important for the request's uuid
to be passed on to the corresponding response.
A typical message exchange cycle may look like this:
- User interaction occurs in the Shell, which results in an event
- The Shell handles this event by constructing an
Event
- The Shell calls the Core's
process_event
function passing theEvent
as an argument - The Core performs the required processing, updating both its inner state and the view model
- The Core returns one or more
Request
messages to the Shell
In the simplest case, the Core will respond to an Event
by returning the single Request
- render.
This requests that the Shell re-renders the user interface.
When Render
is the only response from the Core, the message cycle has completed and the Core has now "settled".
In more complex cases however, the Core may well return multiple Request
s; each of which instructs the Shell to perform a side-effect-inducing task such as:
- Make a network call, or
- Fetch the current date/time stamp, or
- Perform biometric authentication, or
- Obtain an image from the camera, or
- Whatever else you can think of...
Many of these side-effecting-inducing tasks are asynchronous.
The Shell is responsible for passing responses back to the core (to the handle_response
function), which may respond with further requests.
This exchange continues until the core stops requesting further side-effects (typically the last side-effect requested would be Render
).
Refer to examples/counter README
Refer to the Getting Started section of the tutorials.
Crux is kindly sponsored by the following organizations. Your help is very much appreciated.
Red Badger is the digital product consultancy trusted by blue chips and global brands. Our product design and technical pedigree allow us to craft high-impact digital products customers want. We use modern engineering approaches to deliver sustainable change. And embed digital capabilities to power continuous innovation.
Zulip is an open-source modern team chat app designed to keep both live and asynchronous conversations organized.
Zulip sponsor Crux by providing our Zulip server — thank you Zulip!