Zero instance for TransformMatrixKHR

[sheaf]

The current Zero instance for TransformMatrixKHR doesn't abide by the specification:

instance Zero TransformMatrixKHR where
  zero = TransformMatrixKHR
           (zero, zero, zero, zero)
           (zero, zero, zero, zero)
           (zero, zero, zero, zero)

• The first three columns of matrix must define an invertible 3x3 matrix

If possible, the instance should be changed to

instance Zero TransformMatrixKHR where
 zero = TransformMatrixKHR
          (1,0,0,0)
          (0,1,0,0)
          (0,0,1,0)

However, if for any reason that's annoying, I think it would be better to remove the instance entirely.

[expipiplus1]

Removing it would cause VkAccelerationStructureInstanceKHR to lose its zero instance too which is sad. It's not too annoying to implement like that but I think it's a bit weird! There are probably other zero instances with the same problem, like struts with non optional arrays (which are currently empty)

…

On Sat, Dec 12, 2020, 3:58 AM sheaf ***@***.***> wrote: The current Zero instance for TransformMatrixKHR doesn't abide by the specification: instance Zero TransformMatrixKHR where zero = TransformMatrixKHR (zero, zero, zero, zero) (zero, zero, zero, zero) (zero, zero, zero, zero) - The first three columns of matrix must define an invertible 3x3 matrix If possible, the instance should be changed to instance Zero TransformMatrixKHR where zero = TransformMatrixKHR (1,0,0,0) (0,1,0,0) (0,0,1,0) However, if for any reason that's annoying, I think it would be better to remove the instance entirely. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#240>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAGRJXBMYQPIBJPII4EXACDSUJ2VNANCNFSM4UXFDLGA> .

[expipiplus1]

I think the zero instance should be left as it is. But we could consider adding an 'Identity'/'One' class for this and some other things (such as Quaternionf http://expipiplus1.github.io/vulkan/openxr-0.1-docs/OpenXR-Core10-Space.html#v:Quaternionf); probably best done in a utils package.

[sheaf]

Well I don't really see the point of the Zero instance as it is, as one would never want to use it. I suppose the perspective of "zero should mean all the bits are 0" justifies it, but I don't find it very convincing. I think zero should provide a default value whenever there exists a unique possible default value; in this case we want a multiplicative identity as opposed to an additive identity.

[expipiplus1]

Yeah, you're right, it's actually only useful for satisfying that law which is pretty flimsy anyway, shouldn't be too hard to make the change.