Incremental computation through constrained memoization.
[dependencies]
comemo = "0.4"
A memoized function caches its return values so that it only needs to be
executed once per set of unique arguments. This makes for a great optimization
tool. However, basic memoization is rather limited. For more advanced use cases
like incremental compilers, it lacks the necessary granularity. Consider, for
example, the case of the simple .calc
scripting language. Scripts in this
language consist of a sum of numbers and eval
statements that reference other
.calc
scripts. A few examples are:
alpha.calc
:"2 eval beta.calc"
beta.calc
:"2 3"
gamma.calc
:"8 3"
We can easily write an interpreter that computes the output of a .calc
file:
/// Evaluate a `.calc` script.
fn evaluate(script: &str, files: &Files) -> i32 {
script
.split(' ')
.map(str::trim)
.map(|part| match part.strip_prefix("eval ") {
Some(path) => evaluate(&files.read(path), files),
None => part.parse::<i32>().unwrap(),
})
.sum()
}
impl Files {
/// Read a file from storage.
fn read(&self, path: &str) -> String {
...
}
}
But what if we want to make this interpreter incremental, meaning that it only recomputes a script's result if it or any of its dependencies change? Basic memoization won't help us with this because the interpreter needs the whole set of files as input—meaning that a change to any file invalidates all memoized results.
This is where comemo comes into play. It implements constrained memoization with more fine-grained access tracking. To use it, we can just:
- Add the
#[memoize]
attribute to theevaluate
function. - Add the
#[track]
attribute to the impl block ofFiles
. - Wrap the
files
argument in comemo'sTracked
container.
This instructs comemo to memoize the evaluation and to automatically track all
file accesses during a memoized call. As a result, we can reuse the result of a
.calc
script evaluation as long as its dependencies stay the same—even if
other files change.
use comemo::{memoize, track, Tracked};
/// Evaluate a `.calc` script.
#[memoize]
fn evaluate(script: &str, files: Tracked<Files>) -> i32 {
...
}
#[track]
impl Files {
/// Read a file from storage.
fn read(&self, path: &str) -> String {
...
}
}
For the full example see examples/calc.rs
.
This crate is dual-licensed under the MIT and Apache 2.0 licenses.