• playcanvas.js: Transpiled to ES2015

I think this is mistake. As ES2015 (should be written as ECMAScript 2015), is ES6. And IE11 does not support ES6. It supports only ES5 (ECMA2009).

Tested few modern features, and babel seems to be weird with default parameters feature. It always (loose and not loose, with both targets) not outputing default parameters, while they are valid ES6, so should not be transpiled for ES6 target.

I've tested more of Babel and Closure, what it outputs, and for ES5 both seem to do a good job, but for ES6, they start to differ.

Example 1:

Source:

class Ray {
    constructor(origin = new Vec3(0, 0, 0), direction = new Vec3(0, 0, -1)) {
        this.origin = origin;
        this.direction = Object.values(direction);
    }
}

Closure Compiler ES6:

class Qd {
    constructor(a=new z(0,0,0), b=new z(0,0,-1)) {
        this.origin = a;
        this.direction = Object.values(b)
    }
}

Babel + Terser ES6:

class Ray {
    constructor() {
        var e = arguments.length > 0 && void 0 !== arguments[0] ? arguments[0] : new Vec3(0,0,0)
          , t = arguments.length > 1 && void 0 !== arguments[1] ? arguments[1] : new Vec3(0,0,-1);
        this.origin = e,
        this.direction = Object.values(t)
    }
}

Issues:

1. default parameters - are converted to ES5 format
2. use of arguments leads to unoptimized (unable to compile to machine code) code in most JS engines.
3. it seem to not mangling classes like Closure does, which leads to higher file size

Example 2:

Source:

containsPoint(point) {
    for (const plane of this.planes) {
        if (plane[0] * point.x + plane[1] * point.y + plane[2] * point.z + plane[3] <= 0) {
            return false;
        }
    }
    return true;
}

Closure Compiler ES6:

containsPoint(a) {
    for (const b of this.planes)
        if (0 >= b[0] * a.x + b[1] * a.y + b[2] * a.z + b[3])
            return !1;
    return !0
}

Babel + Terser ES6:

containsPoint(e) {
    for (var t of this.planes)
        if (t[0] * e.x + t[1] * e.y + t[2] * e.z + t[3] <= 0)
            return !1;
    return !0
}

Issues:

1. It seem to not like let/const.

File sizes:

Right now, I've tested only on engine with small part moved to ES6+ code, but it already produces different file sizes:

• Raw ES6+ + gzip: 387Kb
• Closure Compiler ES5 + gzip: 297Kb (76.7%)
• Closure Compiler ES6 + gzip: 294Kb (76.0%)
• Babel + Terser ES5 + gzip: 287Kb (74.1%)
• Babel + Terser ES6 + gzip: 303Kb (78.3%)

Size difference between them is insignificant. ES5 Babel + Terser is smaller than Closure Compiler ES5, for around ~2.5%. While ES6 version of Babel + Terser is bigger than Closure Compiler ES6 by around ~2.3%.

Size wise, both paths produce good results, but I assume that Closure Compiler will lead for ES6, as Babael + Terser sometimes transpiles valid ES6 code to ES5.

My only concern, is why Babel + Terser is transpiling some of valid ES6 features into ES5, when instructed to use ES6 target.

@Maksims Ha, yes, sorry, I get confused with all the JavaScript language naming. You're right - I have corrected my original post.

Issues:
default parameters - are converted to ES5 format

Fixed with this commit.

use of arguments leads to unoptimized (unable to compile to machine code) code in most JS engines.

I'm not seeing this. With the commit above, I'm getting the following (after beautifying):

class Ray {
    constructor(e = new Vec3, t = new Vec3(0, 0, -1)) {
        this.origin = e, this.direction = t
    }
    set(e, t) {
        return this.origin.copy(e), this.direction.copy(t), this
    }
}

it seem to not mangling classes like Closure does, which leads to higher file size

Yeah, I noticed that. I'm not sure why but I don't think it's making a significant difference.

Oh, and that let + const thing is also fixed by that commit.