noCompletion isn't respected when using a build macro

[ciscoheat]

In Haxedevelop 5.2.0.3 with Haxe 3.4, @:noCompletion isn't respected for hiding fields if the class is created with a build macro. In the following example, both start and start2 are displayed.

Main.hx

@:build(Builder.build())
class Main 
{   
    static function main() {
        new Main();
    }

    public function new() {
        this.| // both start and start2 are displayed
    }

    public function start() {
        trace("start");
    }

    @:noCompletion public function start2() {
        trace("start2");
    }
}

Builder.hx

import haxe.macro.Context;

using Lambda;

class Builder
{
    public static function build() {
        var fields = Context.getBuildFields();
        var startField = fields.find(function(f) return f.name == "start");
        startField.meta.push({name: ":noCompletion", params: [], pos: Context.currentPos()});
        return fields;
    }
}

haxe --display Main.hx@0 returns <list></list>, so that seems to work.

[Gama11]

I don't think FD's internal completion engine supports noCompletion yet... (0 usages when doing a find in files across all FD files) - so that's probably what you're seeing. That would mean FD never gets to displaying compiler completion in this case, for whatever reason.

Interesting observation: when I add a function foo() {} at the end of Main.hx, it works as expected. Very strange.

[Gama11]

Oh, I think I know what's going on... If FD gets back an empty list from Haxe, it "doesn't trust it", so to speak, and continues to display internal completion instead. Very interesting.

[Gama11]

Hm.. I was about to make a PR with the change (it's just about removing a list > 0 check), but it admittedly looks quite strange (almost buggy) to have completion appear briefly and then immediately disappear because there are 0 results :/

[Neverbirth]

Yeah, it's strange heh, but I would probably go with it.