Strict WebGPU compliance mode

[teoxoy]

We seem to only issue a warning if we end up with a downlevel adapter/device.

wgpu/wgpu-core/src/instance.rs

Lines 303 to 313 in 3d33ace

	if wgt::Backends::PRIMARY.contains(wgt::Backends::from(A::VARIANT))
	&& !caps.downlevel.is_webgpu_compliant()
	{
	let missing_flags = wgt::DownlevelFlags::compliant() - caps.downlevel.flags;
	log::warn!(
	"Missing downlevel flags: {:?}\n{}",
	missing_flags,
	DOWNLEVEL_WARNING_MESSAGE
	);
	log::info!("{:#?}", caps.downlevel);
	}

We need a mechanism that errors in such an event.
I was thinking a feature flag.

[cwfitzgerald]

What's the use case of this? You can view the downlevel flags on the adapter and error out if you only want compliant adapters.

[teoxoy]

You can view the downlevel flags on the adapter and error out if you only want compliant adapters.

That's indeed a possibility but I think we'll need this distinction anyway to support the proposed strict_extensions field on the WGSL Parser (see #4410).

[cwfitzgerald]

Couldn't we automatically set that if all the native-only extensions are false on the device?

[cwfitzgerald]

Hmm, I guess the error messages change - "feature X not enabled" vs "unknown feature X"

[teoxoy]

We should also use this mechanism to avoid exposing features that are not implemented in naga.

For example, SHADER_F16 is currently exposed due to it being available through spir-v passthrough (but not via WGSL).

[teoxoy]

This can now be added to InstanceFlags.