Rust Code Guidelines
We need a cross-platform key and transaction management component upon which platform-specific UIs can be built. We chose Rust for this component because of memory safety, low level control over key buffers, cross-platform portability and good blockchain library support. Rust is the only programming language that satisfies all these requirements.
Our Rust code is exposed to host languages that implement the app UIs via uniffi-rs.
Good performance of Rust code is desired, but it’s not a deal-breaker. We prioritize maintainability and agility over performance. In other words, feel free to clone stuff if it lets you ship features quicker or if it significantly reduces code complexity.
We are particularly weary to introduce any optimizations at this time, as we’ll have to introduce generic blockchain traits as we add support for more protocols (we currently only support Ethereum), and performance optimizations in Rust sometimes lead to rather unwieldy generics.
Since we are using uniffi-rs to expose Rust to host languages via FFI, we have two options to deal with concurrent operations:
- Make FFI calls non-blocking, do async in Rust, and pass results through callback interfaces to host languages.
- Make FFI calls blocking and let the host languages deal with concurrency.
The second option is simpler for both the Rust and the UI code, since the host languages (Swift and Kotlin at this point) have good concurrency primitives that UI developers are familiar with (as opposed to a custom callback scheme), and async Rust has some rough edges which we can avoid this way. For this reason, our Rust code is written as single threaded, blocking code and concurrency is handled by the host languages.
There are three exceptions:
- Some of our Rust dependencies only expose async interfaces. In this case we
block on them with the
async_runtimemodule that wraps a lazy-initialized global Tokio runtime.
- Since we have an async runtime anyway for our dependencies, if there is a function that we want to call multiple times concurrently during one FFI call (eg. to fetch multiple images), then we write that function as async, spawn instances of it on the async runtime and then block on the joined futures.
- While initially the in-page provider was implemented with the blocking approach, we discovered that it's surprisingly difficult to get a fixed-sized dedicated thread-pool in Swift and one request at a time is not good enough performance-wise, so we've refactored it to async in Rust and callback-based through FFI.