Crate reactive_rs
source ·Expand description
This crate provides the building blocks for functional reactive programming (FRP) in Rust. It is inspired by carboxyl, frappe and bidule crates, and various ReactiveX implementations.
Overview
Purpose
The main use case of this library is to simplify creating efficient computational DAGs (or computational trees, to be precise) that operate on streams of values. It does not aim to replicate the entire galaxy of ReactiveX operators, nor does it attempt to delve into futures/concurrency territory.
What is a computational tree? First, there’s the root at the top, that’s
where the input values get fed into continuously. Then, we perform
computations on these values – each of which may yield zero,
one or more values that are sent further down. Some downstream
nodes may share their parents – for instance, g(f(x)) and h(f(x)), where x is
the input and f is the intermediate transformation; in this case, we want
to make sure we don’t have to recompute f(x) twice. Moreover, this
being Rust, we’d like to ensure we’re not copying and cloning any values
needlessly, and we generally prefer things to be zero-cost/inlineable
when possible. Finally, there are leaves – these are observers, functions
that receive transform values and do something with them, likely recording
them somewhere or mutating the environment in some other way.
Terminology
- Observer - a function that accepts a value and returns nothing (it will most often that note mutate the external environment in some way).
- Stream - an object can be subscribed to by passing an observer to it. Subscribing to a stream consumes the stream, thus at most one observer can ever be attached to a given stream.
- Broadcast - an observer that owns a collection of other observers and transmits its input to each one of them sequentially. A broadcast can produce new streams, subscriptions, each one receiving the same input as the broadcast itself. Subscription is a stream that adds its subscribers to the broadcast’s collection when being subscribed to.
Context
Streams, broadcasts and observers in this crate operate on pairs of
values: the context and the element. Context can be viewed as
optional metadata attached to the original value. Closures required in
methods like .map() only take one argument (the element) and are
expected to return a single value; this way, the element can be changed
without touching the context. This can be extremely convenient if you
need to access the original input value (or any “upstream” value) way
down the computation chain – this way you don’t have to propagate
it explicitly.
Most stream/broadcast methods have an alternative “full” version that
operates on both context/element, with _ctx suffix.
Examples
Consider the following problem: we have an incoming stream of buy/sell price pairs, and for each incoming event we would like to compute how the current mid-price (the average between the two) compares relatively to the minimum buy price and the maximum sell price over the last three observations. Moreover, we would like to skip the first few events in order to allow the buffer to fill up.
Here’s one way we could do it (not the most ultimately efficient way of solving this particular problem, but it serves quite well to demonstrate the basic functionality of the crate):
use std::cell::Cell;
use std::f64;
use reactive_rs::*;
let min_rel = Cell::new(0.);
let max_rel = Cell::new(0.);
// create a broadcast of (buy, sell) pairs
let quotes = SimpleBroadcast::new();
// clone the broadcast so we can feed values to it later
let last = quotes.clone()
// save the mid-price for later use
.with_ctx_map(|_, &(buy, sell)| (buy + sell) / 2.)
// cache the last three observations
.last_n(3)
// wait until the queue fills up
.filter(|quotes| quotes.len() > 2)
// share the output (slices of values)
.broadcast();
// subscribe to the stream of slices
let min = last.clone()
// compute min buy price
.map(|p| p.iter().map(|q| q.0).fold(1./0., f64::min));
// subscribe to the stream of slices
let max = last.clone()
// compute max sell price
.map(|p| p.iter().map(|q| q.1).fold(-1./0., f64::max));
// finally, attach observers
min.subscribe_ctx(|p, min| min_rel.set(min / p));
max.subscribe_ctx(|p, max| max_rel.set(max / p));
quotes.send((100., 102.));
quotes.send((101., 103.));
assert_eq!((min_rel.get(), max_rel.get()), (0., 0.));
quotes.send((99., 101.));
assert_eq!((min_rel.get(), max_rel.get()), (0.99, 1.03));
quotes.send((97., 103.));
assert_eq!((min_rel.get(), max_rel.get()), (0.97, 1.03));Lifetimes
Many Stream trait methods accept mutable closures; observers are
also essentially just closures, and they are the only way you can
get results from the stream out into the environment. Rest assured,
at some point you’ll run into lifetime problems (this being Rust,
it’s pretty much certain).
Here’s the main rule: lifetimes of observers (that is, lifetimes of
what they capture, if anything) may not be shorter than the lifetime of
the stream object. Same applies to lifetimes of closures in methods
like .map().
In some situations it’ tough to prove to the compiler you’re doing
something sane, in which case arena-based allocators (like
typed-arena) may be
of great help – allowing you to tie lifetimes of a bunch of
objects together, ensuring simultaneous deallocation.