extra closure created when accessing members of an inlined function result value

[nadako]

Subj. Haxe example:

class A
{
    public var v:Int;

    public function new(v = 0)
    {
        this.v = v;
    }

    public inline function clone()
    {
        return new A(v);
    }
}

class Sample
{
    static function main()
    {
        var v = new A().clone().v;
    }
}

Generated JS:

Sample.main = function() {
    var v = ((function($this) {
        var $r;
        var _this = new A();
        $r = new A(_this.v);
        return $r;
    }(this))).v;
};

AST:

static main(method) : Void -> Void

 = function() = {
    var v = ({
        var _this = new A();
        new A(_this.v);
    }).v;
};

[nadako]

Probably this is the case when inlining should not occur but still does. I got another one here:

class Sample
{
    static function main()
    {
        var o:Dynamic = {f: true};
        if (Reflect.field(o, "f"))
        {
            trace(o.f);
        }
    }
}

js:

Sample.main = function() {
    var o = { f : true};
    if((function($this) {
        var $r;
        var v = null;
        try {
            v = o.f;
        } catch( e ) {
        }
        $r = v;
        return $r;
    }(this))) console.log(o.f);
};

[ncannasse]

We should not mark as inline Reflect.field if it contains a try/catch

[nadako]

Yeah, but shouldn't the inliner be smart here and disallow inlining in this case of usage?

[Simn]

We removed the JS-specific inlining restrictions recently, but I agree Reflect.field should not be inline.

@waneck-mode I have a filter for genc which removes "statements" in some places: https://github.com/waneck/haxe-genc/blob/genc/filters.ml#L111

The generated JS output with that would look like this:

Main.main = function() {
    var o = { f : true};
    var _g8;
    var v = null;
    try {
        v = o.f;
    } catch( e ) {
    }
    _g8 = v;
    if(_g8) o.f;
};

It still has too many locals, but that's a job for a separate filter.

[Simn]

Since the JS inlining restrictions are back in place this issue is "resolved".