ES6 Proxy 效能之我見
本文翻譯自https://thecodebarbarian.com/thoughts-on-es6-proxies-performance
Proxy是ES6的一個強力功能,它通過為 get/set
一個屬性 設定"陷阱"(函式處理器)讓我們可以攔截對於物件屬性的操作。比如:
const obj = {};
const proxy = new Proxy(obj, {
get: () => {
console.log('hi');
}
});
obj.a; // "hi"
Proxy被稱讚為現在已經被廢棄的Object.observe()
屬性的取代者。
然而不幸的是,Proxy有一個致命缺陷:效能。
更打擊人的是,Object.observe()
就是因為效能被廢棄的,而以我(原作者)對V8的理解,對於JIT(Just in Time,準時制)來說,Object.observe()
比Proxy容易優化多了。
Proxy到底有多慢?
我(原作者)在node v6.9.0中用benchmark簡單試了一下:
var Benchmark = require('benchmark');
var suite = new Benchmark.Suite;
var obj = {};
var _obj = {};
var proxy = new Proxy(_obj, {
set: (obj, prop, value) => { _obj[prop] = value; }
});
var defineProp = {};
Object.defineProperty(defineProp, 'prop', {
configurable: false,
set: v => defineProp._v = v
});
// 譯者注: vanilla js 指的就是原生js
suite.
add('vanilla', function() {
obj.prop = 5;
}).
add('proxy', function() {
proxy.prop = 5;
}).
add('defineProperty', function() {
defineProp.prop = 5;
}).
on('cycle', function(event) {
console.log(String(event.target));
}).
on('complete', function() {
console.log('Fastest is ' + this.filter('fastest').map('name'));
}).
run();
結果如下:
$ node proxy.js
vanilla x 74,288,023 ops/sec ±0.78% (86 runs sampled)
proxy x 3,625,152 ops/sec ±2.51% (86 runs sampled)
defineProperty x 74,815,513 ops/sec ±0.80% (85 runs sampled)
Fastest is defineProperty,vanilla
$
從這個簡單的benchmark中我們可以看到,Proxy的set
比直接賦值和defineProperty慢非常多(譯者注:ops/sec,每秒進行的運算元,越大越快)。
為防大家好奇,我(原作者)又在node 4.2.1測試了一下Object.observe()
:
$ node proxy.js
vanilla x 78,615,272 ops/sec ±1.55% (84 runs sampled)
defineProperty x 79,882,188 ops/sec ±1.31% (85 runs sampled)
Object.observe() x 5,234,672 ops/sec ±0.86% (89 runs sampled)
Fastest is defineProperty,vanilla
有些文章可能讓你覺得只要Proxy不用get/set
而是隻設定getOwnPropertyDescriptor()
的話,就比其他的快,於是我(原作者)又試了試:
var _obj = {};
var propertyDescriptor = {
configurable: true,
set: v => { _obj.prop = v; }
};
var proxy = new Proxy(_obj, {
getOwnPropertyDescriptor: (target, prop) => propertyDescriptor
});
不幸的是,反而更慢了:
$ node proxy.js
vanilla x 73,695,484 ops/sec ±1.04% (88 runs sampled)
proxy x 2,026,006 ops/sec ±0.74% (90 runs sampled)
defineProperty x 74,137,733 ops/sec ±1.25% (88 runs sampled)
Fastest is defineProperty,vanilla
$
用Proxy包裹一個函式並呼叫同樣比原生的包裹函式並呼叫慢非常多:
var Benchmark = require('benchmark');
var suite = new Benchmark.Suite;
var fn = () => 5;
var proxy = new Proxy(function() {}, {
apply: (target, context, args) => fn.apply(context, args)
});
var wrap = () => fn();
// add tests
suite.
add('vanilla', function() {
fn();
}).
add('proxy', function() {
proxy();
}).
add('wrap', function() {
wrap();
}).
on('cycle', function(event) {
console.log(String(event.target));
}).
on('complete', function() {
console.log('Fastest is ' + this.filter('fastest').map('name'));
}).
run();
$ node proxy2.js
vanilla x 78,426,813 ops/sec ±0.93% (88 runs sampled)
proxy x 5,244,789 ops/sec ±2.17% (87 runs sampled)
wrap x 75,350,773 ops/sec ±0.85% (85 runs sampled)
Fastest is vanilla
無用的提升Proxy效能的方法
目前最有影響力的提升Proxy效能的方法是讓被修改的屬性的configurable設為false:
var _obj = {};
Object.defineProperty(_obj, 'prop', { configurable: false });
var propertyDescriptor = {
configurable: false,
enumerable: true,
set: v => { _obj.prop = v; }
};
var proxy = new Proxy(_obj, {
getOwnPropertyDescriptor: (target, prop) => propertyDescriptor
});
(譯者注:這段程式碼有些問題,enumerable
在configurable
為false時是無效的)
$ node proxy.js
vanilla x 74,622,163 ops/sec ±0.95% (85 runs sampled)
proxy x 4,649,544 ops/sec ±0.47% (85 runs sampled)
defineProperty x 77,048,878 ops/sec ±0.60% (88 runs sampled)
Fastest is defineProperty
$
要是這樣寫set/get
,還不如直接用 Object.defineProperty()
。
這樣寫的話,你就不得不設定每個你要在Proxy中用到的屬性不可配置(not configurable)。
不然的話,V8就會報錯:
var _obj = {};
Object.freeze(_obj);
var propertyDescriptor = {
configurable: false,
enumerable: true,
set: v => { _obj.prop = v; }
};
var proxy = new Proxy(_obj, {
getOwnPropertyDescriptor: (target, prop) => propertyDescriptor
});
// Throws:
// "TypeError: 'getOwnPropertyDescriptor' on proxy: trap returned
// descriptor for property 'prop' that is incompatible with the
// existing property in the proxy target"
// 攔截'prop'屬性返回的descriptor和target(原物件)已經存在的屬性不匹配
proxy.prop = 5;
Proxy 也要不行了麼?
Proxy比 Object.defineProperty()
有不少優點:
- Proxy 可以巢狀,而
Object.defineProperty()
的getter/setter
就不能巢狀,這樣你就不需要知道提前知道你要攔截的所有屬性 - 可以攔截陣列變化
但它效能太差了。
效能有多大影響呢?
以Promise和回撥為例:
var Benchmark = require('benchmark');
var suite = new Benchmark.Suite;
var handleCb = cb => cb(null);
// add tests
suite.
add('new function', function() {
handleCb(function(error, res) {});
}).
add('new promise', function() {
return new Promise((resolve, reject) => {});
}).
add('promise resolve', function() {
Promise.resolve().then(() => {});
}).
on('cycle', function(event) {
console.log(String(event.target));
}).
on('complete', function() {
console.log('Fastest is ' + this.filter('fastest').map('name'));
}).
run();
$ node promise.js
new function x 26,282,805 ops/sec ±0.74% (90 runs sampled)
new promise x 1,953,037 ops/sec ±1.02% (86 runs sampled)
promise resolve x 194,173 ops/sec ±13.80% (61 runs sampled)
Fastest is new function
$
Promise也慢了非常多,
但是 bluebird聲稱為Promise提供"非常好的效能",測試一下:
$ node promise.js
new function x 26,986,342 ops/sec ±0.48% (89 runs sampled)
new promise x 11,157,758 ops/sec ±1.05% (87 runs sampled)
promise resolve x 671,079 ops/sec ±27.01% (18 runs sampled)
Fastest is new function
雖然快了很多,但仍然比回撥慢不少。
所以我們要因此放棄Promise麼?
並不是這樣的,很多公司仍然選擇了使用Promise。我(原作者)雖然不是很確定,但是Uber好像就在使用Promise。
結論
Proxy很慢,但是在你因其效能而放棄它之前,記得同樣效能很差的Promise在最近幾年中被快速採用。
如果你想使用代理,很可能你不會感覺到效能的影響,除非你發現自己為了效能的原因改變了Promise庫(或者完全避開了它們)。
更新
2019.01, 在node v11.3.0中: Promise已經變得足夠好, Proxy還是那樣
vanilla x 833,244,386 ops/sec ±0.76% (89 runs sampled)
proxy x 28,590,800 ops/sec ±0.72% (88 runs sampled)
wrap x 824,349,552 ops/sec ±0.87% (86 runs sampled)
Fastest is vanilla,wrap
new function x 834,121,566 ops/sec ±0.82% (89 runs sampled)
new promise x 819,789,350 ops/sec ±0.76% (87 runs sampled)
promise resolve x 1,212,009 ops/sec ±40.98% (30 runs sampled)
Fastest is new function