[MacOS] Missing `removeFromSuperlayer` causes memory leak

[jason-ni]

When investigating the multi_window memory leak issue in iced project(iced-rs/iced#3233), I see softbuffer(0.4.8) also has the same leak pattern, that is closing window doesn't release some memory resource. But unlike wgpu, it seems easy to fix in softbuffer. I just added

impl Drop for SendCALayer {
    fn drop(&mut self) {
        unsafe {
            self.0.removeFromSuperlayer();
        }
    }
}

in the cg.rs and then the leak issue solved.

But when I see that in 0.5.0 dev cycle, you are switching to IOSurface for more efficiency, I have concern that it may have similar issue.

[madsmtm]

Yeah that makes sense, we're the "owner" of the layer, but when calling root_layer.addSublayer(layer) we implicitly give ownership to the root layer (so we have to take it back in Drop).

Do you want to file a PR?

---

Why do you think IOSurface would exacerbate this problem though?

[jason-ni]

I'm testing your transparency PR branch in my tauri-plugin-iced(https://github.com/jason-ni/tauri-plugin-iced) today. I'd like to wait for the IOSurface PR and transparency PR both being merged to master, then have a test for leakage.
I have little knowledge about old solution of CALayer vs new IOSurface. But I do see in wgpu backend, the leakage pattern is IOSurface memory is ever increasing. So I have concern that the IOSurface solution may have issue with release memory, too.

[madsmtm]

#329 still uses CALayer.

The reason you're seeing IOSurface leaks is probably because CALayer internally contains those.