10x Performance Regression on uPlot Benchmark since v3.0.0-alpha

[etimberg]

Expected Behavior

V3 dev should be as fast as V3 alpha

Current Behavior

V3 dev is 10x slower than v3.0.0-alpha

Possible Solution

• Downgrade to babel 7.8 @babel/plugin-transform-classses(createSuper)@7.9.x may have performance issue babel/babel#11356
• Use 'loose' mode and solve issues that arise

Steps to Reproduce (for bugs)

v3.0.0-alpha:

var Point =
function (_Element) {
  _inherits(Point, _Element);
  function Point(cfg) {
    var _this;
    _classCallCheck(this, Point);
    _this = _possibleConstructorReturn(this, _getPrototypeOf(Point).call(this));
    _this.options = undefined;
    _this.skip = undefined;
    _this.stop = undefined;
    if (cfg) {
      _extends(_assertThisInitialized(_this), cfg);
    }
    return _this;
  }

master:

var Point = function (_Element) {
  _inherits(Point, _Element);
  var _super = _createSuper(Point);
  function Point(cfg) {
    var _this;
    _classCallCheck(this, Point);
    _this = _super.call(this);
    _this.options = undefined;
    _this.skip = undefined;
    _this.stop = undefined;
    if (cfg) {
      _extends(_assertThisInitialized(_this), cfg);
    }
    return _this;
  }

Environment

• Chart.js version: https://www.chartjs.org/dist/master/chart.js
• Browser name and version:
• Link to your project:

[benmccann]

It sounds like loose is the way to go

The only other options would be:

• Drop IE11 support
• Differential bundling - create one bundle without babel for modern evergreen browsers and one with for IE11

Neither of those options sound super appealing to me though compared to loose. We should also be sure to document the loose option in the performance documentation for users who and applying babel themselves to our ES6 bundle