I've got a data class ```data class MyVec2(val x:...

# kotlin-native

[Adam S]

I've got a data class

data class MyVec2(val x: Float, val y: Float)

and an equivalent C struct

typedef struct Vec2 {
    float x;
    float y;
} Vec2;

and a C function that takes a pointer to a vector

some_function(Vec2 *value)

When I call this function from Kotlin do I need a memory scope to create and pass a Vec2 pointer, or is it possible without? I want to be able to instantiate vectors without a memory scope (because most of the time passing a pointer is not necessary, so it adds some overhead/complexity that I'd like to avoid)

[Jan Holešovský]

I cannot imagine how it would work without memScoped 😉 - I believe yes, you need a memScoped. What is your concern about its usage?

[Adam S]

there are a few things.

• It's annoying to have to either require passing a memscope into the Vector constructor, or using an extension function (which then requires every function that wants to create a vector is also an extension function for memscope)
• from rough experimentation memscoped performance is not good when there are lots of vectors
• performance is better without copying to/from a memscope (but maybe using nativeheap everywhere will be fine?)

[Adam S]

yes it makes sense that a memory scope is required, but I was hoping there was some other way of exposing the primitive values as a cValue struct pointer :)

[Adam S]

For example: using nativeHeap.alloc<Vec2> { x = 1f; y = 2f } is ~5 times slower than MyVec2(x = 1f, y = 2f)

[Jan Holešovský]

Comes to my mind - would it be an option for you just to extend the C api to have also some_function(float x, float y), and call that directly? 😉 That might solve your problem (?)

[Adam S]

good idea, but I need to pass a pointer in because the C function will update the vector values

[Jan Holešovský]

Hmm, then what about if that some_function() with 2 params returns struct with the new values? To consume that, you don't need memScoped I think, maybe it'll perform better?

[Adam S]

possibly! But then I would have to write C 🙃. I think it might be too complicated to do it for all the functions, but I'll experiment and see...

[Krystian]

This is exactly why I gave up on Raylib binding. The amount of overhead and complexity just to get simple stuff done was tedious

[Adam S]

I think what I'll try to do is create a Vec2Var Kotlin class, similar to how there's a IntVar, specifically for passing a pointer into C functions and reading the value

[Adam S]

and I realised the ~5 times slower benchmark was run in debug mode - maybe the result is different in release mode!

[Jan Holešovský]

That might be good news 🙂