Firefox is slow, Lodash is fast
Look at those colors! Arenāt they shiny?
Theyāre super shiny (unless youāre color blind), but what do they mean? Iām glad you asked. Thatās a speed comparison chart of 6 ways to clone JavaScript objects, run in 5 browsers, on 2 devices: my laptop and my iPhone 5SE.
You can try the benchmark yourself:
>Click here. < Iād make an iframe, but it freezes my browser for many seconds at a time. It even freezes the CSS animation on that React logo.
Dangerous business those benchmarks.
Firefox is the only browser that decides something weird is going on and throws a warning. Everyone else happily blocks JS, CSS, and UI.
Now, is this benchmark fair? I donāt know. Running benchmarks on a computer thatās doing a bunch of other stuff is never really fair. Maybe a different tab just tried to do something, or Spotify downloaded a song, or Dropbox ran a metadata update on my entire hard drive.
A bunch of things can affect these results. Thatās why you can run it yourself. But I did my best to ensure fairness as much as I could.
- Each test runs alone, asynchronously
- Each test is re-run 20-times
- Each test uses the same source data
- Each test produces the same deep-ish cloned dataset
I say ādeep-ishā because weāre cloning an array of some 81,000 objects. The objects are shallow, which means we can cut corners.
const experiments = {
"lodash _.cloneDeep": _.cloneDeep,
".map + lodash _.clone": (arr) => arr.map((d) => _.clone(d)),
".map + lodash _.assign": (arr) => arr.map((d) => _.assign({}, d)),
"JSON string/parse": (arr) => JSON.parse(JSON.stringify(arr)),
".map + Object.assign": (arr) => arr.map((d) => Object.assign({}, d)),
".map + React's update()": (arr) => arr.map((d) => update({}, { $merge: d })),
};We use _.cloneDeep without understanding context. This is a little bit unfair because it tries to do too much. We run _.clone, _.assign, Object.assign, and Reactās update in a loop. They benefit from not trying to work in the general case. JSON.parse/stringify is on the same level as _.cloneDeep: naĆÆve, complete, works for anything.
Iām gobsmacked that for datasets this big, youāre better off converting to JSON and back than using Lodashās cloneDeep function. I have no idea how thatās even possible. Maybe JSON benefits from an implementation detail deep in the engine?
But then why is _.assign faster than Object.assign? They both make a shallow copy of an object, but Object.assign is a language feature, and _.assign is implemented in pure JavaScript.
I think ā¦ how else? I hope @jdalton can shed some light on this.
You can see the entire test runner on Github here. There are a few comments, but the interesting bit is this runner function. It ensure fairness by isolating timing to only the cloning method.
runner(name, method) {
let data = this.state.data;
const times = d3.range(0, this.N).map(() => {
const t1 = new Date();
let copy = method(data);
const t2 = new Date();
return t2 - t1;
});
let results = this.state.results;
results.push({name: name,
avg: d3.mean(times)});
this.setState({results: results});
}The runner itself is called asynchronously via setTimeout(foo, 0), and the async library ensures tests donāt happen in parallel. Inside runner, we iterate through N = 20 indexes, take timestamp, perform clone, take another timestamp, and construct an array of time diffs. Then we use d3.mean to get the average and add it to this.state.results with setState.
It runs in React because Iām lazy and create-react-app is the quickest way to set everything up :)
Maybe thereās some unfairness in how Babel compiles this code? Thatās possible ?
My conclusion is this: Use the appropriate algorithm for your use-case, and run your code in Chrome.
Filed under: Technical
