Improve performance by using for-of loop instead of iterate functions

[yhatt]

Use for-of loop instead of iterate functions like Array.forEach() or Array.reduce(). It receives better performance and readability of code.

Motivations

Marpit internal plugins traverse the parsed tokens from Markdown many times. We were always used JavaScript iterate functions because of restricted by ESLint rule from Airbnb. However, for-of statement has better performance.

Marpit has targeted to modern environments. (Node >= 6 LTS, Modern browser) Airbnb says that its polyfill (regenerator-runtime) is too heavyweight, but our target environments are not required polyfill. Thus, we have allowed for-of by overridden no-restricted-syntax ESLint rule .

One more motivation is the readability of code. Array.reduce() had well used to rearrenge tokens, but they have not-intuitive codes. Using Array.push() to the empty array looks like better than Array.reduce().

It is not means that the all of iterate function are disallowed. We're still using iterate funcs to the place that has less impact of performance and expects to lost code readabilty by replacing to for-of.

Benchmark

We have measured the time of 1000 conversions from example of uncover theme. Averages of 20 trials are below:

	Node v8	Node v6
Iterate functions	2079.95ms	2778.75ms
for-of	1211.00ms	2008.5ms

const request = require('request')
const { Marpit } = require('./lib/')

const average = arr => arr.reduce((p, c) => p + c, 0) / arr.length
const marpit = new Marpit()

request.get(
  'https://gist.githubusercontent.com/yhatt/a352e6087c2d5185113b9143f330fb68/raw/ca3ca25b3a022ccaae3836c0464b02693230dcef/_uncover.md',
  (_, __, body) => {
    const tries = []

    for (let t = -5; t < 20; t += 1) {
      const start = new Date().getTime()

      for (let i = 0; i < 1000; i += 1) marpit.render(body)
      const end = new Date().getTime() - start

      if (t >= 0) tries.push(end)
      console.log(`Try ${t >= 0 ? t + 1 : 'for idling'}: ${end}ms`)
    }

    console.log(`Average: ${average(tries)}ms`)
  }
)

[yhatt]

UPDATE: Add 5 times idling in benchmark code. We have re-measured and it seems to be about 40-70% faster.