Generalized VkStruct?

[Rotaerk]

Have you considered, or do you see any problem with, the idea of making a generalization of VkStruct, one that's not Vulkan specific? It just seems to me that the ability to read/write a byte array as a struct with named and typed fields would be useful elsewhere. Particularly in C-library bindings.

[achirkin]

The funny thing is that this is exactly what PrimBytes achieves :) I did not aim at this, but it arose from our discussion of how to get field offsets for a user-defined Vertex type. I was thinking whether I should split PrimBytes from easytensor though.

[Rotaerk]

Ah, nice! I do think breaking it off into its own library would be useful. It could be used by both easytensor and vulkan-api (if you were willing to give it a dependency).

But I'm looking to make or modify a binding to assimp, and I noticed the existing one is defining simple record types and Storable instances. I realized it might be nicer to define VkStruct like things instead...

[Rotaerk]

Though, whatever was extracted into its own library would need to support arrays of structs, which is what your dataframes fundamentally are... so maybe you should have a generalized primbytes+dataframe library, and then just put tensor-specific stuff in easytensor?

[achirkin]

Yes, this would make sense! Except there is no tensor-specific stuff in easytensor yet :-D
I consider PrimBytes more as just an alternative to Storable and thinking of packing it as such. But I am not sure that many people will use it outside easytensor/vulkan-api anywhere, so I don't hurry with this decision so far.

[Rotaerk]

Well the tensor-specific stuff is your Matrix and Vertex stuff, for example. DataFrame is more general than being used as a tensor, since it can be treated as just a typesafe array of data. At least that's how I see it currently.

[Rotaerk]

And regarding people using it outside vulkan/easytensor, it seems like it would be very useful for FFI libraries, and code using them. Besides that, I can't think of why you would need it... If you don't need a pointer to precisely laid out data, you can just use standard ADTs.

[achirkin]

NB: there is some duplicated functionality in PrimBytes and VkStruct. If we were to split PrimBytes from easytensor, it's worth reusing it here.

https://github.com/achirkin/easytensor/projects/2