#reactive #frp #signal

nightly mini-rx

bare-bones "reactive programming" (change propogation) using a central data dependency graph

2 releases

0.1.1 Jul 21, 2022
0.1.0 Jul 18, 2022

#642 in Algorithms

Apache-2.0

70KB
1K SLoC

mini-rx: Tiny reactive programming change propagation a la scala.rx

Cargo documentation

Example

use mini_rx::*;

fn example() {
	// setup
	let side_effect = Cell::new(0);
	let side_effect2 = RefCell::new(String::new());

	// The centralized data dependency graph
	let mut g = RxDAG::new();

	// Create variables which you can set
	let var1 = g.new_var(1);
	let var2 = g.new_var("hello");
	assert_eq!(var1.get(g.now()), &1);
	assert_eq!(var2.get(g.now()), &"hello");
	var1.set(&g, 2);
	var2.set(&g, "world");
	assert_eq!(var1.get(g.now()), &2);
	assert_eq!(var2.get(g.now()), &"world");

	// Create computed values which depend on these variables...
	let crx1 = g.new_crx(move |g| var1.get(g) * 2);
	// ...and other Rxs
	let crx2 = g.new_crx(move |g| format!("{}-{}", var2.get(g), crx1.get(g) * 2));
	// ...and create multiple values which are computed from a single function
	let (crx3, crx4) = g.new_crx2(move |g| var2.get(g).split_at(3));
	assert_eq!(crx1.get(g.now()), &4);
	assert_eq!(crx2.get(g.now()), &"world-8");
	assert_eq!(crx3.get(g.now()), &"wor");
	assert_eq!(crx4.get(g.now()), &"ld");
	var1.set(&g, 3);
	var2.set(&g, &"rust");
	assert_eq!(crx1.get(g.now()), &6);
	assert_eq!(crx2.get(g.now()), &"rust-12");
	assert_eq!(crx3.get(g.now()), &"rus");
	assert_eq!(crx4.get(g.now()), &"t");

	// Run side effects when a value is recomputed
	let var3 = g.new_var(Vec::from("abc"));
	let side_effect_ref = &side_effect;
	let side_effect_ref2 = &side_effect2;
	// borrowed values must outlive g but don't have to be static
	g.run_crx(move |g| {
		side_effect_ref.set(side_effect_ref.get() + var1.get(g));
		side_effect_ref2.borrow_mut().push_str(&String::from_utf8_lossy(var3.get(g)));
	});
	assert_eq!(side_effect.get(), 3);
	assert_eq!(&*side_effect2.borrow(), &"abc");
	var1.set(&g, 4);
	g.recompute();

	assert_eq!(side_effect.get(), 7);
	assert_eq!(&*side_effect2.borrow(), &"abcabc");

	// Note that the dependencies aren't updated until .recompute or .now is called...
	var3.set(&g, Vec::from("xyz"));
	assert_eq!(side_effect.get(), 7);
	assert_eq!(&*side_effect2.borrow(), &"abcabc");
	g.recompute();
	assert_eq!(side_effect.get(), 11);
	assert_eq!(&*side_effect2.borrow(), &"abcabcxyz");

	// the side-effect also doesn't trigger when none of its dependencies change
	var2.set(&g, "rust-lang");
	g.recompute();
	assert_eq!(side_effect.get(), 11);
	assert_eq!(&*side_effect2.borrow(), &"abcabcxyz");
	assert_eq!(crx2.get(g.now()), &"rust-lang-16");

	// lastly we can create derived values which will access or mutate part of the base value
	// which are useful to pass to children
	let dvar = var3.derive_using_clone(|x| &x[0], |x, char| {
		x[0] = char;
	});
	assert_eq!(dvar.get(g.now()), &b'x');
	dvar.set(&g, b'b');
	assert_eq!(dvar.get(g.now()), &b'b');
	assert_eq!(var3.get(g.now()), &b"byz");
	dvar.set(&g, b'f');
	assert_eq!(dvar.get(g.now()), &b'f');
	assert_eq!(var3.get(g.now()), &b"fyz");
	assert_eq!(&*side_effect2.borrow(), &"abcabcxyzbyzfyz");
}

Overview

mini-rx is a bare-bones implementation of "reactive programming" in Rust with 1 dependency. It uses manual polling and integrates well with the borrow checker by storing all values in a central data dependency graph, RxDAG.

The type of reactive programming is signal-based, which is similar to scala.rx but different than most libraries (which are stream-based) and maybe not technically FRP. Instead of manipulating a stream of values, you manipulate variables which trigger other computed values to recompute, which in turn trigger other values to recompute and side effects to run.

Key concepts/types

  • RxDAG: stores all your Rxs. Lifetime rules guarantee that they don't change while you have active references (see Lifetimes)
    • new_var: creates a Var
    • new_crx: creates a CRx
    • run_crx: runs a side-effect, will re-run when any of the accessed Vars or CRxs change
    • new_crx[n]: creates n CRxs which come from a single computation
    • recompute: updates all Var and CRx values, but requires a mutable reference which ensures there are no active shared references to the old values
    • now: recomputes and then gets an RxContext so you can get values. It must recompute and thus requires a mutable reference.
    • stale: Does not recompute but will not return the most recently set values unless recompute was called.
  • Var: value with no dependencies which you explicitly set, and this triggers updates
  • CRx: value computed from dependencies
  • RxContext: allows you to read Var and CRx values. Accessible via RxDAG::now or in computations (RxDAG::new_crx) and side-effects (RxDAG::run_crx)
  • MutRxContext: allows you to write to Vars. An &RxDAG is a MutRxContext. You cannot set values in a CRx computation because they are inputs.

Signal-based

The type of reactive programming is signal-based, which is similar to scala.rx but different than most libraries (which are stream-based). Instead of manipulating a stream of values, you manipulate variables which trigger other computed values to recompute, which in turn trigger other values to recompute and side effects to run.

You can simulate stream-based reactivity by adding a side-effect which pushes values on trigger, like so:

use mini_rx::*;

fn stream_like() {
	let stream = RefCell::new(Vec::new());
	let stream_ref = &stream;
	let input1 = vec![1, 2, 3];
	let input2 = vec![0.5, 0.4, 0.8];
	
	let mut g = RxDAG::new();
	let var1 = g.new_var(0);
	let var2 = g.new_var(0.0);
	let crx = g.new_crx(move |g| *var1.get(g) as f64 + *var2.get(g));
	
	g.run_crx(move |g| {
		stream_ref.borrow_mut().push(*crx.get(g));
	});
	
	assert_eq!(&*stream.borrow(), &vec![0.0]);
	for (a, b) in input1.iter().zip(input2.iter()) {
		var1.set(&g, *a);
		var2.set(&g, *b);
		g.recompute();
	}
	assert_eq!(&*stream.borrow(), &vec![0.0, 1.5, 2.4, 3.8]);
}

For more traditional stream-based reactive programming, I recommend reactive-rs

Lifetimes

You can't obtain a mutable reference to the value stored within a Var. Instead you call Var::set or Var::modify with a completely new value. This is because there may be active references to the old Var. When you call Var::set it doesn't immedidately change the old value, so those references won't change.

In order to actually update the reactive values and run side-effects, you must call RxDAG::recompute, or a function which internally calls recompute like RxDAG::now. In order to do this, you need a mutable refernce to the RxDAG, which you can only get if there are no active references to any of the reactive values.

Additionally, any compute function in the RxDAG must live longer than the RxDAG itself. This is because the function may be called any time while the RxDAG is alive, when it gets recomputed. So if you have values which you reference in CRx computations or side-effects, you must either declare them before the RxDAG or use something like a Weak reference to ensure that they are still alive when used.

Why? Signal-based Reactive programming 101

Here's a situation commonly encountered in programming: you have a value a which should always equal b + c. You don't want a to be a function, but when b or c changes, a must be recalculated.

Or here's another situation: you have an action which must always run when a value changes, which sends the updated value to the server.

You can chain these. Perhaps the value you want to send to the server on update is a. Perhaps b and c are computed from other values, d, e, f, and so on. Ultimately, for the theoretical folks, you have a DAG (directed-acyclic-graph) of values and dependencies. Modify one of the roots, and it triggers a cascade of computations and side effects.

Dependencies

~1.5MB
~33K SLoC