Despite the title, this blog post doesn't start with Rust. Instead, it starts with the tradgedy of the ongoing global COVID-19 pandemic. Numberphile had recently uploaded a video about the SIR model, which describes the spread of an infectious disease. The simple yet powerful nature of the model and its relevance to current events struck a chord with me: I had to code this up for myself.
Python seemed like the best candidate for the job, given the existence of numerical libraries such as NumPy and SciPy. My first iteration needed just 35 lines of code, a testament to Python's expressiveness and the well-designed library interfaces.
The next step seemed clear to me: how cool would it be to put this on the web? Sure, it's a simple model, but one of great relevance given recent events, so maybe it might interest some.
I weighed my options. I had a working implementation in Python, so reusing that made sense. I could host my implementation on a server somewhere and expose it via a REST API. Hell it might even be cheap to do this if I instead used something like an AWS lambda. This did leave some questions though, mainly around Python package support and the lambda taking too long to run and timing out.
Perhaps instead I could compile my Python implementation to WebAssembly and run it in the browser? Reading around, it sounded like this boiled down to having a WebAssembly implementation of the Python interpreter. Turns out there are a few, but after some experimentation I gave up because of the difficulty of integrating SciPy support into the project. If I was going to run with WebAssembly, perhaps I'd need to forego my existing Python implementation and use a language that better-supported WebAssembly.
Enter Rust. As a long-time C++ user, and a jaded one at that, I'd been aching for a decent project to apply Rust to for ages. Why not now? WebAssembly support in Rust is way more mature than Python, and the result would undoubtedly perform better without the need for Python's virtual machine.
I'm not going to lie: compared to Python, Rust certainly feels less expressive, but then it'd be hard not to. Compared to C++, however, Rust is just a dream. Variable declarations and type inference allow me to write code that just feels cleaner. Perhaps I'm superficial, but after C++ one of the biggest selling points is the standard library. It just feels more complete and well-thought-out than that of C++. I'm not trying to downplay the other features of the language. It was nice to know that I could write code without the risk of using a dangling reference or causing a segfault.
The biggest difficulty I found was finding a library to solve the SIR model differential equations. I initially experimented with the peroxide crate, but was quickly frustrated to find that the interface doesn't allow models that are expressed via closures, instead requiring a function pointer. This meant I'd either have to hard-code the model parameters or use global variables, both of which were enough to rule it out.
I instead settled for the diffeq crate, which permits closures and has great support for JSON serialisation of the solution it computes. My final implementation required only 45 lines of Rust.
The SIR model itself highlights the importance of social distancing: by cutting the rate at which the virus spreads through the population we can not only lessen the peak number of infections but potentially limit the number of people infected by the virus overall.