nanoid/async/format.browser.js

// This file replaces `async/format.js` in bundlers like webpack or Rollup,
// according to `browser` config in `package.json`.

module.exports = function (random, alphabet, size) {
  // We can’t use bytes bigger than the alphabet. To make bytes values closer
  // to the alphabet, we apply bitmask on them. We look for the closest
  // `2 ** x - 1` number, which will be bigger than alphabet size. If we have
  // 30 symbols in the alphabet, we will take 31 (00011111).
  // We do not use faster Math.clz32, because it is not available in browsers.
  var mask = (2 << Math.log(alphabet.length - 1) / Math.LN2) - 1
  // Bitmask is not a perfect solution (in our example it will pass 31 bytes,
  // which is bigger than the alphabet). As a result, we will need more bytes,
  // than ID size, because we will refuse bytes bigger than the alphabet.

  // Every hardware random generator call is costly,
  // because we need to wait for entropy collection. This is why often it will
  // be faster to ask for few extra bytes in advance, to avoid additional calls.

  // Here we calculate how many random bytes should we call in advance.
  // It depends on ID length, mask / alphabet size and magic number 1.6
  // (which was selected according benchmarks).

  // -~f => Math.ceil(f) if n is float number
  // -~i => i + 1 if n is integer number
  var step = -~(1.6 * mask * size / alphabet.length)

  function tick (id) {
    return random(step).then(function (bytes) {
    // Compact alternative for `for (var i = 0; i < step; i++)`
      var i = step
      while (i--) {
        // If random byte is bigger than alphabet even after bitmask,
        // we refuse it by `|| ''`.
        id += alphabet[bytes[i] & mask] || ''
        // More compact than `id.length + 1 === size`
        if (id.length === +size) return id
      }
      return tick(id)
    })
  }

  return tick('')
}