16/04/2020

# Finding Problems

In the past week, I was looking forward to getting really stuck into the rendering side of things, and finally get something more interesting on the screen. I was following along with some tutorials (opens new window), documentation (opens new window), and examples (opens new window) to further the progress. I started with the goal of loading and displaying a texturea simple task, I thought, couldn't take more than an hour — then to encounter a few roadblocks/sub-problems/side-missions which stood in my way, of which, I will explain now:

# Resource handling with Cargo

Since this is a Rust (opens new window) project, the idiom is to use Cargo (opens new window) as the build tool. Cargo has been great up to this point, but I found it lacking for this seemingly trivial task. I wanted to load a texture in my application at runtime, meaning, when I build my app I want that texture asset to be copied relative to the executable's output directory.

I thought this would be a single config variable in Cargo.toml, i.e., resource-dir = "src/assets". Woe is me, tis not. You see, Cargo only really deals with code, and after some 🦆quacking (opens new window) I found out that if you want to do anything else, well you'll need a build script (opens new window). And thus, build.rs was born.

// build.rs

use std::env;
use std::path::{Path, PathBuf};

fn main() {
    // get the source assets directory, which is found relative to cargo.toml
    let src_assets_dir: PathBuf = [env!("CARGO_MANIFEST_DIR"), "src", "assets"]
        .iter()
        .collect();

    // locate target directory by walking upwards from out directory
    let out_dir = PathBuf::from(env::var("OUT_DIR").unwrap());
    let target_dir: Option<&Path> = {
        let mut cwd = Some(out_dir.as_path());
        loop {
            match cwd {
                // it's probably the right directory if it ends with target
                Some(dir) if dir.ends_with("target") => break,
                Some(dir) => cwd = dir.parent(),
                None => break,
            }
        }
        cwd
    };

    // locate the destination asset directory, which is in the current build
    // profile directory, in the target directory
    let dest_assets_dir = PathBuf::from(target_dir.unwrap())
        .join(env::var("PROFILE").unwrap())
        .join("assets");

    // no need to explain this 🔥
    if dest_assets_dir.exists() {
        std::fs::remove_dir_all(dest_assets_dir.as_path()).unwrap();
    }

    // finally, straight up recursively copy every file using the copy_dir crate
    copy_dir::copy_dir(src_assets_dir, dest_assets_dir).expect("😢");
}

Now, this is a rudimentary way of solving this issue. In the future, if I have hundreds of megabytes of assets, I probably shouldn't delete everything and copy it all again on every build (especially when the vs code rust-analyzer plugin seems to run the script too). I may eventually have to walk the destination directory myself and check if each file was modified, and if so, replace it. But we'll burn that bridge as we cross it.

# Refactoring vertex arrays

So, now I had my texture asset loading from the correct directory and sent to OpenGL for rendering. Problem solved, nope, not problem solved. I still had to tell my sqaure shader to use the texture, meaning I had to bind the texture before drawing, which requires the fragment shader to have a sampler2D and texture coordinates, requiring the vertex shader to output texture coordinates to the fragment shader, requiring the vertex array object to have a vertex buffer with a layout including an attribute set to the correct indexed location and data type (and element count, stride, offset, and if to normalize)... which requires the vertex data to actually have texture coordinates in the first place.

As can be seen, that's a long string of dependent intrinsics to deal with. So, I had to refactor and encapsulate that code, otherwise I would lose my mind having to remember all of that when I add normals to the vertex data or try and load some arbitrary mesh.

My vertex data is now like this:

struct Vert {
    position: glm::Vec3,
    tex_coords: glm::Vec2,
}

And, I can easily reflect that layout:

let vert_layout = VertLayout::new(vec![
    VertAttr::new(VertAttrType::Float3, false),
    VertAttr::new(VertAttrType::Float2, false),
]);

This can then be passed to a vertex buffer and then a vertex array, which can calculate 90% of the data from that VertAttrType enum I'm using, e.g., a Float3 is 3, 32-bit floats, which gives me the count, size (in bytes), and type, then I use its index in the vector for the location, and calculate the offset from the size of the attributes before it. The stride is the sum of all attribute sizes.

In retrospect, this was a very necessary fundamental step, although annoying at the time because I just wanted to render a texture. 🤷

Full code here (opens new window).

# Some other wins

Last Updated: 10/31/2020, 5:02:02 AM