Add RequiresAllOf, automatically enable required extensions and features

[Rua]

Changelog:

### Additions
- When creating an instance or device, you only need to specify the extensions and features you actually care about. Any extensions and features that are required by the extensions that you specified are now automatically enabled too.

This adds the ability to specify multiple requirements that must be enabled together, using RequiresAllOf. This is needed for some parts of the API, such as vkAcquireNextImage2KHR and vkGetPhysicalDevicePresentRectanglesKHR. RequiresOneOf now contains a slice of RequiresAllOf, which in turns contains a slice of Requires enums. In other words, requirements now follow a DNF structure.

Additionally, I added the automatic enabling of required extensions and features. This is purely a convenience for the user, but hopefully one that makes dealing with extensions less frustrating, by letting the user focus on what they actually need.

[marc0246]

Why did you revert #2205?

[Rua]

I was confused that Debug seemed to print the same thing as Display, when the norm is for Debug to show the internal structure of the object. What do you think the output should say?

[marc0246]

I think it should work like it did before, with Debug printing a nicely-formatted message. See the PR for an example, and/or try it for yourself. Nothing is saying Debug must be derived.

[Rua]

But how should it be formatted, now that there's multiple layers of requirements?

[Rua]

Actually looking at the code, I'm even more confused. There is a custom Debug implementation for ValidationError already.

[marc0246]

If you want I can do the formatting, I have a clear picture in mind. There's some more serious issues I think I found here that you should address.

[Rua]

Yes please.

[marc0246]

I was trying to debug it but you'll know better than me. If you look at the build output of DeviceExtensions::check_requirements it's all borked.

[Rua]

Do you get a compile error? It looks normal on my end:

pub(super) fn check_requirements(
        &self,
        supported: &DeviceExtensions,
        api_version: crate::Version,
        instance_extensions: &crate::instance::InstanceExtensions,
    ) -> Result<(), crate::ValidationError> {
        if self.khr_16bit_storage {
            if !supported.khr_16bit_storage {
                return Err (crate :: ValidationError { problem : "contains `khr_16bit_storage`, but this extension is not supported by the physical device" . into () , .. Default :: default () }) ;
            }
            if !(api_version >= crate::Version::V1_1
                || instance_extensions.khr_get_physical_device_properties2)
            {
                return Err(crate::ValidationError {
                    problem: "contains `khr_16bit_storage`".into(),
                    requires_one_of: crate::RequiresOneOf(&[
                        crate::RequiresAllOf(&[crate::Requires::APIVersion(crate::Version::V1_1)]),
                        crate::RequiresAllOf(&[crate::Requires::InstanceExtension(
                            "khr_get_physical_device_properties2",
                        )]),
                    ]),
                    ..Default::default()
                });
            }
            if !(api_version >= crate::Version::V1_1) {
                return Err(crate::ValidationError {
                    problem: "contains `khr_16bit_storage`".into(),
                    requires_one_of: crate::RequiresOneOf(&[crate::RequiresAllOf(&[
                        crate::Requires::APIVersion(crate::Version::V1_1),
                    ])]),
                    ..Default::default()
                });
            }
        }

etc.

[marc0246]

No there's no compile error, the checks it does are not correct.

[marc0246]

Will you mind if I push the formatting changes directly to your branch? I could also open a PR to your branch but lazy >.>

[Rua]

Yes, go ahead.

Are you talking about the device extensions required by other device extensions? Those checks are gone because they're automatically enabled now.

[Rua]

Ah never mind, I think I see what's going on.

[Rua]

I made a fix. Whenever a device extension was one of the alternatives, since we're always able to enable that extension, we can treat that alternative as true, making the whole test true. So now, the check is only emitted when there is no device extension among the alternatives.

[marc0246]

I added the Debug impl back.

Unwrapping a VulkanError looks like this:

called `Result::unwrap()` on an `Err` value: a validation error occurred

Caused by:
    something: is wrong

Requires one of:
    feature `feat1` + device extension `ext1` + device extension `ext2`
    feature `feat1` + feature `feat2`

Vulkan VUIDs:
    VUID-something-12345

I checked the Display implementation as well, using anyhow, and it looks like this:

a validation error occurred

Caused by:
    something: is wrong -- requires one of: (feature `feat1` + device extension `ext1` + device extension `ext2`) or (feature `feat1` + feature `feat2`) (Vulkan VUIDs: VUID-something-12345)

[Rua]

Is there anything else I need to do?

[marc0246]

Do you approve of the commit I added?

[Rua]

Yeah, it's fine.

[marc0246]

Alrighty, then it looks to me like this is good to go. Thank you for the work!