Marijn Suijten

Fair enough, `feature`s will have to be additive so we can only expose more API, not remove any. I.e. we would have to keep the original `RawFd`/`i32` functions and enable...

@nixpulvis Yup I've seen this as well (on Linux), the example is violating some semaphore rules that are caught in newer validation layers. It's not a problem with `ash` itself...

@filnet swapchain functionality doesn't support the usage of timeline semaphores.

Note also that `vk.xml` retroactively adds this attribute to existing structures from time to time, because it was initially forgotten. That would immediately break semver compatibility for us so we...

For completeness, note that upstream _does not_ set `returnedonly="true"` for structs that require the user to allocate storage, like https://registry.khronos.org/vulkan/specs/1.3-extensions/man/html/VkSurfacePresentModeCompatibilityEXT.html / https://registry.khronos.org/vulkan/specs/1.3-extensions/man/html/VkLatencySurfaceCapabilitiesNV.html.

> Hmm, seems like this should probably be an attribute on individual fields rather than entire structs. Perhaps that's something @oddhack can help us out with, but I can also...

Let us first link some relevant sources: The PR that made this `MaybeUninit::uninit()`: #798 The bit of code that you're bringing up for discussion: https://github.com/ash-rs/ash/blob/331724c216b62d2914479d44e16c977e685e0749/ash/src/instance.rs#L440-L452 Also relevant to mention that...

Now, given that the fields of `struct vk::PhysicalDeviceMemoryProperties` are safe to access, I don't know if `unsafe`ness of `Instance::get_physical_device_memory_properties()` allows us to say that the caller is responsible for only...

https://doc.rust-lang.org/std/pin/struct.Pin.html#method.get_unchecked_mut might be a nice example of this, where the safety docs don't specify any _preconditions_ to uphold, but do say how you have to handle the returned value to...

I don't want to close this prematurely as there's definitely something actionable on the Ash side, even if [the Vulkan specification specifically calls out](https://registry.khronos.org/vulkan/specs/1.3-extensions/man/html/VkPhysicalDeviceMemoryProperties.html) that the `*Count` fields define how...