Raylib with `-define:RAYLIB_USE_LINALG=false` does not work with recent addition of raymath

[laytan]

An API like raymath was recently added to the raylib package that calls into the linalg package. But with -define:RAYLIB_USE_LINALG=false this obviously fails to compile.

[laytan]

2 solutions that come to mind:

1. Drop the RAYLIB_USE_LINALG configuration, always use linalg instead
2. Put raymath in a when block over this config and bind to the actual raymath procs in the false case

[gingerBill]

It's probably better to just use linalg for everything.

[laytan]

Agreed, but, the linalg types still don't work right, I made a gist porting a raylib example here: https://gist.github.com/laytan/8c1dbe6c96d78045f95524da26237765

It contains a raymath.odin that should replace the vendor/raylib/raymath.odin, it just adds a when block around it and loads the needed procedures from raylib.

The example looks like this (correct) with -define:RAYLIB_USE_LINALG=false:

https://github.com/odin-lang/Odin/assets/20369598/456c29fe-66ed-4343-b7d2-b4d315f103da

But, without that define, it looks like this:

https://github.com/odin-lang/Odin/assets/20369598/d3483500-5e8f-4be9-a84d-52710c701461

[karl-zylinski]

I think the above might be due to more Matrix alignment issues? DrawMeshInstanced uses a multi-pointer of rl.Matrix. Will the multi-pointer also be affected by the same alignment issues that we saw in those structs you fixed previously @gingerBill?

[thePHTest]

Note that USE_LINALG also causes issues with rl.DrawMesh, which also has a rl.Matrix parameter.

[gingerBill]

I don't understand how alignment would effect anything. If something is 16 byte aligned, it also means it is 4 byte aligned.

[thePHTest]

It differences are from differing matrix conventions.

With linalg:

t := raylib.MatrixTranslate(3.0, 0.0, 0.0)
fmt.println(t)
fmt.println(transmute([16]f32)t)
matrix[1, 0, 0, 3; 0, 1, 0, 0; 0, 0, 1, 0; 0, 0, 0, 1]
[1, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1, 0, 3, 0, 0, 1]

Without linalg

Matrix{m0 = 1, m4 = 0, m8 = 0, m12 = 3, m1 = 0, m5 = 1, m9 = 0, m13 = 0, m2 = 0, m6 = 0, m10 = 1, m14 = 0, m3 = 0, m7 = 0, m11 = 0, m15 = 1}
[1, 0, 0, 3, 0, 1, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1]

[gingerBill]

I have figured out the problem now.

Raylib uses row-major matrices internally, but transposes at the last moment before uploading to the GPU.

It's kind of crazy the decisions that they made, but oh well.

The "solution" is to heavily document everything and tell people to only use linalg procedures if they remember to linalg.transpose the matrices, or do row_vector * row_matrix everywhere.

[laytan]

You are correct, reimplementing the functions I needed in the example to do the same thing but transposed made the example work correctly, so raymath.odin needs to be changed to do this everywhere.

[laytan]

Works now, with the recent addition of #row_major matrices 🚀