ash_window loses access to Surface Instance

[dthelegend]

Hi, I am new to Vulkan and was looking through the docs for ash_window::create_instance, the docs say that end-users should destroy the surface before destroying the parents, but this method explicitly drops the instance while orphaning the surface. How should I safely destroy a surface acquired this way?

[StarLederer]

I am in the same situation and also wondering this. The example from the repo linked in the readme is outdated.

[MarijnS95]

ash-window does not have a create_instance() function, this only exists on Entry in ash.

In case you meant ash_window::create_surface(), that only takes &Instance as a borrow without destroying anything, i.e. the VkInstance is definitely not destroyed before we return the created VkSurfaceKHR.

Instead, if you meant the loaded pointers for the extensions, like let surface_fn = win32_surface::Instance::new(entry, instance);, that is a little inefficient if multiple surfaces are created for the same RawDisplayHandle type but not invalid because no ownership is represented here either. It's something I would like to improve before landing #774 though, which also needs to reload the function pointers again to provide a generic fn get_present_support() ¹.

---

Regardless, ash is a low-level wrapper that does not implement ownership semantics, i.e. we do not call any vkDestroyXxx() function on Drop. All VkHandles are simple Copy types.

The only way to destroy a surface is by loading the instance function pointers for VK_KHR_surface (ash::khr::surface::Instance::new()) and calling destroy_surface() via it.

---

Maybe the only takeaway from this issue is: should we rename fn new() on extension function pointer loaders to fn load(), cc @Ralith?

Footnotes

1. Checking presentation support on a physical device + queue index before creating a surface. ↩

[Ralith]

should we rename fn new() on extension function pointer loaders to fn load()

I'm in favor of this; I think it's a readability win, and not the first time we've had confused users around here.

[StarLederer]

I found that this repo has an up-to-date example and I was already able to destroy the surface but thank you anyway for the explanation!

I think this issue is mostly asking for help or possibly misunderstanding what is happening behind the scenes so for me personally it is solved. I do like the changes proposed in #1019 and they are in line with what I was thinking create_surface() should have been. I will post more thoughts in the PR.

I don't think renaming ash::khr::surface::Instance::new() is a good idea, it might actually do the opposite because while new() is a generic name that suggests "figure it out, we believe in you", load() suggests that there is something special going on and you are probably supposed to unload() or that we shouldn't load it while it is already loaded, but we actually have to, like when using the current version of create_surface()