Array binding provided zero elements validation error when using PARTIALLY_BOUND_BINDING_ARRAY

[JMS55]

Using the PARTIALLY_BOUND_BINDING_ARRAY feature and using BindingResource::TextureViewArray(&[]) I get a validation error Array binding provided zero elements when creating the bind group. I'm not sure if this is intended.

Currently I'm working around this by adding a dummy texture to my array to ensure it has at least one element.

CC @kanerogers

[kanerogers]

@JMS55 Sorry to hear you're having issues here!

Can I ask what backend you're using and which adapter features you've enabled?

[JMS55]

RTX 3080, Vulkan, all of them. Bevy enables all features by default. Let me know if there's more info you need.

[kanerogers]

Thanks so much for the prompt response!

To be totally upfront I'm absolutely not a wgpu expert; just someone who's contributing some work on bindless. That said, I'm more than happy to take a look for you in the next week.

[kanerogers]

Ah, one last question: how are you using the texture array in your shader?

At first glance what you've described should be legal, at least from a Vulkan perspective (as per: https://registry.khronos.org/vulkan/specs/1.3-extensions/man/html/VkDescriptorBindingFlagBits.html) but there may be something wgpu specific that requires at least one texture view to be present in the array.

[JMS55]

I'm using it as a binding_array<texture_2d<f32>> in my shader.

The issue is in bind group creation, though. It looks like wgpu unconditionally disallows zero-length arrays.

wgpu/wgpu-core/src/device/mod.rs

Lines 2299 to 2301 in bf8e6fe

	if num_bindings == 0 {
	return Err(Error::BindingArrayZeroLength);
	}

I'm wondering if we can relax that, so that I don't have to work around it in user code.

[kanerogers]

Hi, @JMS55 👋🏻

Thanks for your patience on this one. I've spoken with some of the wgpu maintainers and it looks like the solution here is something like the following:

• If each backend that supports PARTIALLY_BOUND_BINDING_ARRAY also supports zero sized bindings, then this restriction will be relaxed.
• Otherwise, the default behaviour will be to support the lowest common denominator, and a new feature, for example ZERO_SIZED_BINDING_ARRAYS will be created to identify which backends support this.

I can't give you a timeframe on when this will be done, but I hope this clarification explains this a bit further. If you do have some spare time and are looking to contribute, I would be happy to help you tackle this one yourself; wgpu is a very interesting piece of software and, given its complexity, is very approachable.

If you have any other questions or if there's anything else I can do to assist, please, just let me know.

[JMS55]

Hi, thank you for the update!

Unfortunately I don't have time at the moment to contribute myself. No worries about it though, the work-around is easy enough, it's not a major issue :).