Skip to content

Iteration helpers that inline to native loops for performance

License

Notifications You must be signed in to change notification settings

planttheidea/inline-loops.macro

Repository files navigation

inline-loops.macro

Iteration helpers that inline to native loops for performance

Table of Contents

Summary

inline-loops.macro is a babel macro that will inline calls to the iteration methods provided, replacing them with for loops (or for-in in the case of objects). While this adds more code, it is also considerably more performant than the native versions of these methods. When working in non-JIT environments this is also faster than equivalent runtime helpers, as it avoids function calls and inlines operations when possible.

This is inspired by the work done on babel-plugin-loop-optimizer, but aims to be both more targeted and more full-featured. Rather than globally replace all native calls, the use of macros allow a controlled, opt-in usage. This macro also supports decrementing array and object iteration, as well as nested usage.

You can use it for everything, only for hotpaths, as a replacement for lodash with legacy support, whatever you see fit for your project. The support should be the same as the support for babel-plugin-macros.

Usage

import { map, reduce, someObject } from 'inline-loops.macro';

function contrivedExample(array) {
    const doubled = map(array, (value) => value * 2);
    const doubleObject = reduce(doubled, (object, value) => ({
      ...object,
      [value]: value
    }, {});

    if (someObject(doubleObject, (value) => value > 100)) {
        console.log('I am large!');
    }
}

Methods

  • every (MDN documentation)
    • everyRight => same as every, but iterating in reverse
    • everyObject => same as every but iterating over objects intead of arrays
  • filter (MDN documentation)
    • filterRight => same as filter, but iterating in reverse
    • filterObject => same as filter but iterating over objects intead of arrays
  • find (MDN documentation)
    • findLast => same as find, but iterating in reverse (MDN documentation)
    • findObject => same as find but iterating over objects intead of arrays
  • findIndex (MDN documentation)
    • findLastIndex => same as findIndex, but iterating in reverse (MDN documentation)
    • findKey => same as findIndex but iterating over objects intead of arrays
  • flatMap (MDN documentation)
    • flatMapRight => same as flatMap, but iterating in reverse
    • There is no object method, as the use cases and expected results are not clearly defined, nor is the expected outcome obvious
  • forEach (MDN documentation)
    • forEachRight => same as forEach, but iterating in reverse
    • forEachObject => same as forEach but iterating over objects intead of arrays
  • map (MDN documentation)
    • mapRight => same as map, but iterating in reverse
    • mapObject => same as map but iterating over objects intead of arrays
  • reduce (MDN documentation)
    • reduceRight => same as reduce, but iterating in reverse (MDN documentation)
    • reduceObject => same as reduce but iterating over objects intead of arrays
  • some (MDN documentation)
    • someRight => same as some, but iterating in reverse
    • someObject => same as some but iterating over objects intead of arrays

How it works

Internally Babel will transform these calls to their respective loop-driven alternatives. Example

const foo = map(array, fn);
// transforms to
const _length = array.length;
const _results = Array(_length);
for (let _key = 0, _value; _key < _length; ++_key) {
  _value = array[_key];
  _results[_key] = fn(_value, _key, array);
}
const foo = _results;

If you are passing uncached values as the array or the handler, it will store those values as local variables and execute the same loop based on those variables.

Aggressive inlining

One extra performance boost is that inline-loops will try to inline the callback operations when possible. For example:

const doubled = map(array, (value) => value * 2);
// transforms to
const _length = array.length;
const _results = Array(_length);
for (let _key = 0, _value; _key < _length; ++_key) {
  _value = array[_key];
  _results[_key] = _value * 2;
}
const doubled = _results;

Notice that there is no reference to the original function, because it used the return directly. This even works with nested calls!

const isAllTuples = every(array, (tuple) =>
  every(tuple, (value) => Array.isArray(value) && value.length === 2),
);
// transforms to
let _determination = true;
for (
  let _key = 0, _length = array.length, _tuple, _result;
  _key < _length;
  ++_key
) {
  _tuple = array[_key];
  let _determination2 = true;
  for (
    let _key2 = 0, _length2 = _tuple.length, _value, _result2;
    _key2 < _length2;
    ++_key2
  ) {
    _value = _tuple[_key2];
    _result2 = Array.isArray(_value) && _value.length === 2;
    if (!_result2) {
      _determination2 = false;
      break;
    }
  }
  _result = _determination2;
  if (!_result) {
    _determination = false;
    break;
  }
}
const isAllTuples = _determination;

Conditional / lazy scenarios

There are times where you want to perform the operation lazily, and there is support for this as well:

foo === 'bar' ? array : map(array, (v) => v * 2);
// transforms to
foo === 'bar'
  ? array
  : (() => {
      const _length = array.length;
      const _results = Array(_length);
      for (let _key = 0, _v; _key < _length; ++_key) {
        _v = array[_key];
        _results[_key] = _v * 2;
      }
      return _results;
    })();

The wrapping in the IIFE (Immediately-Invoked Function Expression) allows for the lazy execution based on the condition, but if that condition is met then it eagerly executes and returns the value. This will work just as easily for default parameters:

function getStuff(array, doubled = map(array, (v) => v * 2)) {
  return doubled;
}
// transforms to
function getStuff(
  array,
  doubled = (() => {
    const _length = array.length;
    const _results = Array(_length);
    for (let _key = 0, _v; _key < _length; ++_key) {
      _v = array[_key];
      _results[_key] = _v * 2;
    }
    return _results;
  })(),
) {
  return doubled;
}

Because there is a small cost to parse, analyze, and execute the function compared to just have the logic in the same closure, the macro will only wrap the logic in an IIFE if such conditional or lazy execution is required.

Bailout scenarios

Inevitably not everything can be inlined, so there are known bailout scenarios:

  • When using a cached function reference (we can only inline functions that are statically declared in the macro scope)
  • When there are multiple return statements (as there is no scope to return from, the conversion of the logic would be highly complex)
  • When the return statement is not top-level (same reason as with multiple returns)
  • The this keyword is used (closure must be maintained to guarantee correct value)

If there is a bailout of an anonymous callback function, that function is stored in the same scope and used in the loop:

const result = map([1, 2, 3], (value) => {
  if (value === 2) {
    return 82;
  }

  return value;
});
// transforms to
const _collection = [1, 2, 3];
const _fn = (_value) => {
  if (_value === 2) {
    return 82;
  }
  return _value;
};
const _length = _collection.length;
const _results = Array(_length);
for (let _key = 0, _value; _key < _length; ++_key) {
  _value = _collection[_key];
  _results[_key] = _fn(_value, _key, _collection);
}
const result = _results;

Note that in bailout scenarios that are used in a closure, the transform will wrap itself in an IIFE to avoid memory leaks from retaining the injected variables.

Gotchas

Some aspects of implementing this macro that you should be aware of:

*Object methods do not perform hasOwnProperty check

The object methods will do operations in for-in loop, but will not guard via a hasOwnProperty check. For example:

// this
const doubled = mapObject(object, (value) => value * 2);

// transforms to this
let _result = {};

let _value;

for (let _key in object) {
  _value = object[_key];
  _result[key] = _value * 2;
}

const doubled = _result;

This works in a vast majority of cases, as the need for hasOwnProperty checks are often an edge case; it only matters when using objects created via a custom constructor, iterating over static properties on functions, or other non-standard operations. hasOwnProperty is a slowdown, but can be especially expensive in legacy browsers or non-JIT environments.

If you need to incorporate this, you can just filter prior to the operation:

const filtered = filterObject(object, (_, key) => Object.hasOwn(object, key));
const doubled = mapObject(filtered, (value) => value * 2);

find* methods differ in naming convention

Most of the operations follow the same naming conventions:

  • {method} (incrementing array)
  • {method}Right (decrementing array)
  • {method}Object (object)

The exception to this is the collection of find-related methods:

  • find
  • findLast
  • findObject
  • findIndex
  • findLastIndex
  • findKey

The reason for findLast / findLastIndex instead of findRight / findIndexRight is because unlike all the other right-direction methods, those methods are part of the ES spec (findLast / findLastIndex). The reason thatfindKey is used for object values instead of something like findIndexObject is because semantically objects have keys instead of indices.

Development

Standard stuff, clone the repo and npm install dependencies. The npm scripts available:

  • build => runs babel to transform the macro for legacy NodeJS support
  • clean=> remove any files from dist
  • lint => runs ESLint against all files in the src folder
  • lint:fix => runs lint, fixing any errors if possible
  • prepublishOnly => run lint, typecheck, test, clean, and dist`
  • release => release new version
  • release:beta => release new beta version
  • test => run jest tests
  • test:watch => run test, but with persistent watcher
  • typecheck => run tsc against the codebase