Asset memory is not free correctly because registry attribute is still binded

[lukasharing]

engine/src/asset/asset.js

Lines 490 to 515 in 771201a

	unload() {
	if (!this.loaded && this._resources.length === 0)
	return;
	this.fire('unload', this);
	this.registry.fire('unload:' + this.id, this);
	const old = this._resources;
	// clear resources on the asset
	this.resources = [];
	this.loaded = false;
	// remove resource from loader cache
	if (this.file) {
	this.registry._loader.clearCache(this.getFileUrl(), this.type);
	}
	// destroy resources
	for (let i = 0; i < old.length; ++i) {
	const resource = old[i];
	if (resource && resource.destroy) {
	resource.destroy();
	}
	}
	}

to

 unload() {
        if (!this.loaded && this._resources.length === 0)
            return;

        this.fire('unload', this);
        this.registry.fire('unload:' + this.id, this);

        const old = this._resources;

        // remove resource from loader cache
        if (this.file) {
            this.registry._loader.clearCache(this.getFileUrl(), this.type);
        }

        // clear resources on the asset
        this.resources = null;
        this.registry = null;
        this.loaded = false;

        // destroy resources
        for (let i = 0; i < old.length; ++i) {
            const resource = old[i];
            if (resource && resource.destroy) {
                resource.destroy();
            }
        }
    }

[lukasharing]

This also needs a proper destroy function for the AssetRegistry where the _asset list and the _cache list should cleaned

[willeastcott]

Yeah, I noticed the AssetRegistry wasn't being destroyed on the Application when I was writing unit tests last week:

https://github.com/playcanvas/engine/blob/dev/test/framework/application.test.mjs#L62

So what you're suggesting here looks correct to me, @lukasharing. Would you like to draft a PR? Or would you like somebody else to run with what you write above?

[lukasharing]

I would like that somebody else solves this for me :), because Im not used to tests

I did write this issue too:
#3866

[willeastcott]

Ah yes. Well, I have fixed #3866 now. We'll aim to address this one soon too. Thanks again for researching the solution - it helps so much!!