This is awesome! As an enthusiast of 3D graphics, I appreciate how clean and easy to read the source is. We have very few examples of how the graphics pipeline works in code, tinyrenderer [1] by ssloy being one of the most famous and easy to follow. This should enter the awesome-graphics list as a must read code!
It's TypeScript! It is "a strict syntactical superset of JavaScript, and adds optional static typing to the language." [1], and compiles to standard JavaScript.
It's always "from one language to another", sometimes the destination is machine language. The output is generally not meant to be meant to be human-readable in any case. "Transpile" is an ugly and unnecessary word, please don't push it as "more correct".
"Transpilation" and "compilation" describe the exact same concept. From user's perspective, readable code is translated into runnable code. From theoretical perspective, code in a more complex language is translated into a less complex language while preserving semantics. From a compiler writer's perspective, it's a totem pole of compilers, with at least one middle-level language usually. Plenty of people use the term "compiles to JavaScript" to describe languages like TS or Elm. It's just some overly pedantic people insist on the term "transpile" when the target is also, in a completely unrelated way, used to write code directly by a significant number of contemporary programmers.
> "Transpilation" and "compilation" describe the exact same concept.
No I think transpile implies compiling from one level of abstraction to another similar level of abstraction, which is quite different in theory and practice to compiling to a much lower level of abstraction. It's a useful word to have in your communication toolbox.
I agree with your overall definition. In fact, the dictionary definition of compile does also suggest that you are transforming to a 'lower level' so, in strictness, there is a need for a word to fill the void that you suggest transpile provides.
But I do wonder about the actual usefulness of the distinction. If someone says that they compile Typescript to Javascript, I'm not sure that anyone is going to be confused about what is meant, even if violates the letter of the dictionary definition.
Ultimately, english is fluid and is always correct as long as those in communication are able to understand each other. For transpile to be a useful word in the toolbox, there must be a circumstance where compile cannot be used to convey the same idea. When might that be?
'This is a transpiler' gives me a massive amount of additional context compared to 'this is a compiler' - likely structure, implementation techniques, output. Not all transpilers will fit this usual structure, but if they do you can tell me by changing this one little word. Powerful communication tool, I think.
But neither 'this is a transpiler' or 'this is a compiler' provide any meaningful information without specifying the compatible inputs and outputs. Once the inputs and outputs are specified then it is already reasonably known if the operation is transpilation or compilation, making the distinction between transpiler and compiler redundant.
Fine, but by that logic, gcc with -O0 is a transpiler, and becomes a compiler again with -O2. Similarly, TS transpiles to JS, but, say, ClojureScript compiles to JS. And if a compiler is built with something like Nanopass, it's a sequence of transpilers that somehow become a compiler along the way.
Transpile is a superset or compile just like typescript is a superset of JavaScript. Transpile implies translating but not into a less complex language but rather onto a similar runtime. So typescript is dynamic just like JavaScript, and they share much of their semantics. So transpile means compile but onto a similar runtime with similar runtime semantic behavior.
In that case surely you'd agree that there are hardly any compilers in wide use these days - more often than not the first step of a modern compiler is translation into a similar but less expressive intermediate language.
Or does it only count as "transpilation" if the intermediate representation is an already existing language, thus making the distincion purely based on the history of a language, not the implementation?
It’s TypeScript. You can strip the annotations and type definitions and it’s pure JS. TS is typically ran on JS runtimes, hence why people consider TS programs to also just be JS program.
*not exactly, there’s a couple of features that map differently. But there's a 1:1 TS to JS conversion.
I think it's worded in a way that's odd to developers that usually write code for multiple target languages.
What you type is usually considered what it's implemented in, not what it's compiled/transpiled to. The later is the target language, which there may be many.
This differentiation is required if you want to modify/read the code. You would almost certainly modify/read it in the language it was implemented in, and make it clear which language this is when sharing.
Maybe "...implemented in TS, targeting JS" or at least something more obvious like "...implemented in JS using TS".
.ts comes from TypeScript, which is just JavaScript but with compile-time-checked types (and some other goodies) added on top. TS is really popular now, I don't think any sane dev would start a new project today in plain JS instead of TS.
That's huge hyperbole there. Plenty of projects are started in JS today. It's still wise to provide typings to your user base if you're publishing a library, whether or not you're using TS under the hood.
As someone with very little knowledge of computer graphics, are there any resources anyone would recommend in order to learn how to build something like this?
The next step up from that is to then render meshes using some form of projection (isometric or perspective). You'll need to figure out how to actually map the drawing to the screen at that point. (In high school we just used the built-in 2D methods in Java. Some of my classmates then went on to shading and textures, but I didn't quite get that far.)
Of course it has more to do with raytracing than with 3D projections (which is most often used for real time 3D rendering). If you want to learn about 3D projections, first you should understand what linear algebra has to do with projecting a 3D model on a 2D screen. This is probably a good start.
You should DEFINITELY write your own matrix library to learn the underlying principles, but once you want your renderer to actually do something with reasonable speed you'll pretty quickly be replacing your homegrown matrix code with something like Eigen that handles SIMD for you.
For rendering at interactive rates, maybe check out "Real-time Rendering", currently on its 4th edition. https://www.realtimerendering.com/
For rendering at rates measured in seconds per frame, consider the "Physically Based Rendering Textbook", currently on its 3rd edition. https://www.pbrt.org/
As far as I can see the terminal rendering is one of the main points that's showcased with this project. Sure you could do this with a raster framebuffer, but that's been done loads of times. The fact that it looks this good in a terminal is marvelous imho.
[1] - https://github.com/ssloy/tinyrenderer