Performance optimizations

[juanjoDiaz]

This undoes the performance optimizations that were introduced in #360 and #378.

It seems that v8 has improved significantly: support is a lot better and native implementation of the methods are consistently much faster in all browsers, Node, and Deno.

It can be easily tested using:

function flattenReducer(acc, arr) {
  try {
    // This is faster but susceptible to `RangeError: Maximum call stack size exceeded`
    acc.push(...arr);
    return acc;
  } catch (err) {
    // Fallback to a slower but safer option
    return acc.concat(arr);
  }
}

let start = performance.now();
Array(100).fill(Array(1000).fill(1)).reduce(flattenReducer);
let end = performance.now();
console.log(`Time: ${end - start} ms.`);

start = performance.now();
Array(100).fill(Array(1000).fill(1)).flat();
end = performance.now();
console.log(`Time: ${end - start} ms.`);

function fastJoin(arr, separator) {
  let isFirst = true;
  return arr.reduce((acc, elem) => {
    if (elem === null || elem === undefined) {
      elem = '';
    }

    if (isFirst) {
      isFirst = false;
      return `${elem}`;
    }

    return `${acc}${separator}${elem}`;
  }, '');
}

let size = 1000
let start = performance.now();
fastJoin(Array(size).fill('test'), '-');
let end = performance.now();
console.log(`Time: ${end - start} ms.`);

start = performance.now();
Array(size).fill('test').join('-');
end = performance.now();
console.log(`Time: ${end - start} ms.`);

for Node, you need to import performance doing const { performance } = require('perf_hooks');

[coveralls]

Coverage remained the same at 100.0% when pulling 7f357767311bc0c08f00704a5aaf843f479f7b3c on juanjoDiaz:performance_optimizations into 4c74c1e661e8bd0e4a5ff0efa2ba22a34ea9da35 on zemirco:master.

[knownasilya]

I wonder if we could add a perf test to see if things change with updates or just with node updates

[juanjoDiaz]

There is #365 about perf tests. I've just never found something like coveralls but for performance. To compare different engines and to avoid regressions. I'm open to ideas.

We should merge this anyway unless there is any blocker. I have some other PRs coming after this.