I've wondered about some kind of hal callback that gives you all the supported extensions and then allows you to push more into the vector.

I've discussed this with @cwfitzgerald recently, and we agreed that a callback is probably the best option.

In the mean time, you can create the vulkan device yourself and then pass it to wgpu. It's not that hard, mostly just copy pasting from wgpu_hal's device initialization code:

https://github.com/JMS55/dlss_wgpu/blob/main/src/request_device.rs

Some kind of callback is fine, though we need to mind various things:

• The lifetime needs to be such that the closure can put its own borrows in the pnext chain. The pnext members need somewhere to live.
• We need to document that we use the standard vk 1.N features, as users might need to search for those in the pnext chain.

We might also want to build a macro to make this easier, but that doesn't have to be in v1

I'm going to have a go at adding a callback, here is a rough idea of what we need to do for a callback from looking around the code:

• hal: Type added for what gets passed into the callback. (idk what to call this)
• hal: An optional function that gets passed into adapter.open.
• core: Moving validation for create device into it's own function call.
• core: Add an extra function call with a callback.
• wgpu: Add an extra function call with a callback.

Btw I would also like the same thing for instance-level vulkan extensions, not just device.

My initial idea is at https://github.com/Vecvec/wgpu/tree/create-device-hal-callback.