The current API model is GPUAdapter exposing limits and extensions, and in GPUDeviceDescriptor the user can opt into higher-than-standard limits by providing these structures.

The use case I'm worried about is people just passing the GPUAdapter capabilities straight into requestDevice regardless of which they need. Why this is concerning: developers tend to use more capable machines. They can easily hit a situation where something that works on their machine suddenly doesn't work on a user machine. This is a portability problem by definition.

Perhaps, if we just make it a little bit less straightforward to request all the limits and extensions (for example, have an API requesting a single limit/extension), then dev wouldn't be encouraged to do the wrong thing here.

(spawned from a Matrix discussion)