92 releases

0.47.0 Apr 19, 2022
0.46.0 Dec 31, 2021
0.45.0 Nov 26, 2021
0.44.0 Jun 28, 2021
0.6.0 Jul 12, 2016

#41 in Graphics APIs

Download history 100/week @ 2023-11-18 110/week @ 2023-11-25 40/week @ 2023-12-02 112/week @ 2023-12-09 87/week @ 2023-12-16 92/week @ 2023-12-23 17/week @ 2023-12-30 74/week @ 2024-01-06 91/week @ 2024-01-13 53/week @ 2024-01-20 40/week @ 2024-01-27 58/week @ 2024-02-03 87/week @ 2024-02-10 109/week @ 2024-02-17 172/week @ 2024-02-24 105/week @ 2024-03-02

487 downloads per month
Used in 27 crates (19 directly)

BSD-3-Clause

305KB
5K SLoC

crates.io License

A simple, type-safe and opinionated graphics crate

luminance is an effort to make graphics rendering simple and elegant. It is a low-level and opinionated graphics API, highly typed (type-level computations, refined types, etc.) which aims to be simple and performant. Instead of providing users with as many low-level features as possible, luminance provides you with some ways to do rendering. That has both advantages and drawbacks:

  • On one side, because the API is opinionated, some dynamic branching and decisions are completely removed / optimized. Some operations breaking state mutations or invariant violation are not statically constructible, ensuring safety. Because strong typing is used, lots of runtime checks are also not needed, helping with performance.
  • On the other side, if you want to do something very specific and very low-level, you will find luminance not to be friendly as it doesn’t like, most of the time, exposing its internal design to the outer world — mostly for runtime safety reason.

A note on safety: here, safety is not used as with the Rust definiton, but most in terms of undefined behavior and unwanted behavior. If something can lead to a weird behavior, a crash, a panic or a black screen, it’s considered unsafe. That definition obviously includes the Rust definiton of safety — memory safety.

What’s included?

luminance is a rendering crate, not a 3D engine nor a video game framework. As so, it doesn’t include specific concepts, such as lights, materials, asset management nor scene description. It only provides a rendering library you can plug in whatever you want to.

There are several so-called 3D-engines out there on crates.io. Feel free to have a look around.

However, luminance comes with several interesting features:

  • Framebuffers: framebuffers are used to hold renders. Each time you want to perform a render, you need to perform it into a framebuffer. Framebuffers can then be combined with each other to produce effects and design render layers — this is called compositing.
  • Shaders: luminance supports five kinds of shader stages:
    • Vertex shaders.
    • Tessellation control shaders.
    • Tessellation evaluation shaders.
    • Geometry shaders.
    • Fragment shaders.
  • Vertices, indices, primitives and tessellations: those are used to define a shape you can render into a framebuffer with a shader. They are mandatory when it comes to rendering. Even if you don’t need vertex data, you still need tessellations to issue draw calls.
  • Textures: textures represent information packed into arrays on the GPU, and can be used to customize a visual aspect or pass information around in shaders. They come in several flavours — e.g. 1D, 2D, cube maps, etc.
  • Control on the render state: the render state is a set of capabilities you can tweak to draw frames. It includes:
    • The blending equation and factors. Blending is the process of taking two colors from two framebuffers and mixing them.
    • Whether we should have a depth test performed.
    • Face culling.
    • Etc.
  • And a lot of other cool things like GPU commands, pipelines, uniform interfaces and so on…

How to dig in?

luminance is written to be fairly simple. There are several ways to learn how to use luminance:

  • The online documentation is a mandatory start for newcomers.
  • The “Learn luminance” book. Ideal for newcomers as well as people already used to luminance, as it’s always updated to the latest version — you might learn new things!
  • The luminance-examples project. It contains lots of examples describing how to do specifics things. Not adapted for newcomers, you will likely want to consult those examples if you’re already familiar with graphics programing and to look for how to do a specific thing. Still, for newcomers, the hello-world example might be a good read.

Implementation and architecture

luminance has been originally designed around the OpenGL 3.3 and OpenGL 4.5 APIs. However, it has mutated a lot to adapt to new technologies and modern graphics programming. Even though its API is not meant to converge towards something like Vulkan, it’s changing over time to meet better design decisions and performance implications.

The current state of luminance comprises several crates:

  • A “core” crate, luminance, which is about all the abstract, common and interface code.
  • A proc-macro crate, [luminance-derive], which is exported by luminance if you use the "derive" feature flag. That crate allows to implement various important traits of the core crate.
  • A set of backend implementation crates, implementing the luminance crate backend interfaces.
  • A set of windowing crates, executing your code written with the core and backend crate on native systems (most of the time, windowing platforms, but not limited to).
  • A special crate, luminance-front, a special backend crate that allows to combine several “official” crates to provide a cross-platform experience without having to pick several backend crates — the crate does it for you. This crate is mainly designed for end-user crates and should be a good fit for most users.

The core crate

The luminance crate gathers all the logic and rendering abstractions necessary to write code over various graphics technologies. It contains parametric types and functions that abstract over the actual implementation type — as a convention, the type variable B (for backend) is used.

Backend types — i.e. B — are not provided by luminance directly. They are typically provided by crates containing the name of the technology as suffix, such as luminance-gl, [luminance-webgl], luminance-vk, etc. The interface between those backend crates and luminance is specified in luminance::backend.

On a general note, Something<ConcreteType, u8> is a monomorphic type that will be usable only with code working over the ConcreteType backend. If you want to write a function that accepts an 8-bit integer something without specifying a concrete type, you will have to write something along the lines of:

use luminance::backend::something::Something as SomethingBackend;
use luminance::something::Something;

fn work<B>(b: &Something<B, u8>) where B: SomethingBackend<u8> {
  todo!();
}

This kind of code is intended for people writing libraries with luminance. For the more usual case of using the luminance-front crate, you will end up writing something like:

use luminance_front::something::Something;

fn work(b: &Something<u8>) {
  todo()!;
}

In luminance-front, the backend type is selected at compile and link time. This is often what people want, but keep in mind that luminance-front doesn’t allow to have several backend types at the same time, which might be something you would like to use, too.

Backend implementations

Backends implement the luminance::backend traits and provide, mostly, a single type for each implementation. It’s important to understand that a backend crate can provide several backends (for instance, luminance-gl can provide one backend — so one type — for each supported OpenGL version). That backend type will be used throughout the rest of the ecosystem to deduce subsequent implementors and associated types.

If you want to implement a backend, you don’t have to push any code to any luminance crate. luminance-* crates are official ones, but you can write your own backend as well. The interface is highly unsafe, though, and based mostly on unsafe impl on unsafe trait. For more information, feel free to read the documentation of the luminance::backend module.

Windowing

luminance doesn’t know anything about the context it executes in. That means that it doesn’t know whether it’s used within SDL, GLFW, glutin, Qt, a web canvas or an embedded specific hardware such as the Nintendo Switch. That is actually powerful, because it allows luminance to be completely agnostic of the execution platform it’s running on: one problem less. However, there is an important point to take into account: a single backend type can be used with several windowing crates / implementations. That allows to re-use a backend with several windowing implementations. The backend will typically explain what are the conditions to create it (like, in OpenGL, the windowing crate must set some specific flags when creating the OpenGL context).

luminance does not provide a way to create windows because it’s important that it not depend on windowing libraries – so that end-users can use whatever they like. Furthermore, such libraries typically implement windowing and event features, which have nothing to do with our initial purpose.

A platform crate supporting luminance will typically provide native types by re-exporting symbols (types, functions, etc.) from a windowing crate and the necessary code to make it compatible with luminance. That means providing a way to access a backend type, which implements the luminance::backend interface. However, platform crates are not supposed to be a replacement for the underlying platform system; you will typically still have to depend it as well.

luminance-derive

If you are compiling against the "derive" feature, you get access to luminance-derive automatically, which provides a set of procedural macros.

Vertex

The Vertex derive proc-macro.

That proc-macro allows you to create custom vertex types easily without having to care about implementing the required traits.

The Vertex trait must be implemented if you want to use a type as vertex (consumed by Tess). Either you can decide to implement it on your own, or you could just let this crate do the job for you.

Important: the Vertex trait is unsafe, which means that all of its implementors must be as well. This is due to the fact that vertex formats include information about the structure of the data that will be sent to the backend, and a bad implementation can lead to undefined behaviors.

You can derive the Vertex trait if your type follows these conditions:

  • It must be a struct with named fields. This is just a temporary limitation that will get dropped as soon as the crate is stable enough.
  • Its fields must have a type that implements VertexAttrib. This is mandatory so that the backend knows enough about the types used in the structure to correctly align memory, pick the right types, etc.
  • Its fields must have a type that implements HasSemantics as well. This trait is just a type family that associates a single constant (i.e. the semantics) that the vertex attribute uses.
  • Each field's type must be different.

Once all those requirements are met, you can derive Vertex pretty easily.

Note: feel free to look at the Semantics proc-macro as well, that provides a way to generate semantics types in order to completely both implement Semantics for an enum of your choice, but also generate field types you can use when defining your vertex type.

The syntax is the following:

use luminance::{Vertex, Semantics};

// visit the Semantics proc-macro documentation for further details
#[derive(Clone, Copy, Debug, PartialEq, Semantics)]
pub enum Semantics {
  #[sem(name = "position", repr = "[f32; 3]", wrapper = "VertexPosition")]
  Position,
  #[sem(name = "color", repr = "[f32; 4]", wrapper = "VertexColor")]
  Color,
}

#[derive(Clone, Copy, Debug, PartialEq, Vertex)]
#[vertex(sem = "Semantics")] // specify the semantics to use for this type
struct MyVertex {
  position: VertexPosition,
  color: VertexColor,
}

Note: the Semantics enum must be public because of the implementation of HasSemantics trait.

Besides the Semantics-related code, this will:

  • Create a type called MyVertex, a struct that will hold a single vertex.
  • Implement Vertex for MyVertex.

The proc-macro also supports an optional #[vertex(instanced = "<bool>")] struct attribute. This attribute allows you to specify whether the fields are to be instanced or not. For more about that, have a look at VertexInstancing.

Semantics

The Semantics derive proc-macro.

UniformInterface

The UniformInterface derive proc-macro.

The procedural macro is very simple to use. You declare a struct as you would normally do:

use luminance::{shader::{types::Vec4, Uniform}, UniformInterface};

#[derive(Debug, UniformInterface)]
struct MyIface {
  time: Uniform<f32>,
  resolution: Uniform<Vec4<f32>>,
}

The effect of this declaration is declaring the MyIface struct along with an effective implementation of UniformInterface that will try to get the "time" and "resolution" uniforms in the corresponding shader program. If any of the two uniforms fails to map (inactive uniform, for instance), the whole struct cannot be generated, and an error is arisen (see UniformInterface::uniform_interface’s documentation for further details).

If you don’t use a parameter in your shader, you might not want the whole interface to fail building if that parameter cannot be mapped. You can do that via the #[unbound] field attribute:

#[derive(Debug, UniformInterface)]
struct MyIface {
  #[uniform(unbound)]
  time: Uniform<f32>, // if this field cannot be mapped, it’ll be ignored
  resolution: Uniform<Vec4<f32>>,
}

You can also change the default mapping with the #[uniform(name = "string_mapping")] attribute. This changes the name that must be queried from the shader program for the mapping to be complete:

#[derive(Debug, UniformInterface)]
struct MyIface {
  time: Uniform<f32>,
  #[uniform(name = "res")]
  resolution: Uniform<Vec4<f32>>, // maps "res" from the shader program
}

Finally, you can mix both attributes if you want to change the mapping and have an unbound uniform if it cannot be mapped:

#[derive(Debug, UniformInterface)]
struct MyIface {
  time: Uniform<f32>,
  #[uniform(name = "res", unbound)]
  resolution: Uniform<Vec4<f32>>,
}

Dependencies