Leaving Rust gamedev after 3 years

Well the problem with Jonathan's argument here is that he's spent the past decade mostly ranting about Rust and working to make a perfect game programming language, instead of making games.

So it turn out that even if he's opinion on Rust is correct, he would still have been much more productive using it than trying to build his own language for a decade…

(But he already shipped his masterpiece and he's a millionaire so he gets to chose his full time hobby as he wishes)

> but picking up C# or gdscript given that you already know programming should be straightforward compared to re-implementing all of those things yourself in Rust.

Right, this is practically my point. I suspect that the tools available from those languages mean that learning one of them would more than pay for itself in the course of developing a (single) game. Many many times over really.

Like, yes, I've dealt with both sound, basic graphics programming (though I'd need to learn a bit more to make a modern looking 3d game), networking, ... in rust. If I had to program my sound system and graphics engine from scratch myself I'd do it in rust (and I believe I'd be more productive in rust than I'd be in while doing so). But I don't have to do everything from scratch, and the best not-from-scratch versions aren't in rust, and the cost of switching to something I don't know just isn't that high.

Also OP is definitely right that rust has some anti-features that would be pain points for game development.

Now they need to try blockchain :)

Excellent article. Rust is my favorite language for several uses, but I'm becoming less optimistic about it as a versatile language long-term. An important point regarding the article is that it mixes rust shortfalls, and shortfalls of any language other than C++ for games. I've personally used Bevy for a 3D wave function renderer, but moved away from it in favor of a custom WGPU-engine due to Bevy's complicated ECS DSL.

I am worried because:

  - Games seem like a no-go, as articulated in the article
  - Web usage has been dominated by Async, with no signs of reversing. I have no interest in this. And, there is no promising Django analog or ORM.
  - Emebedded support on Cortex-M, and to a lesser extend RISC-V is good at its core, but the supporting libraries have a mix of A: the game failures listed in the article (Driven by hype, mostly makers, serious companies are not using it), and B: Also being taken over by Async.

This is disappointing to me, because IMO the syntax, tooling, and general language experience of Rust is far better than C and C++.

As someone who's become a core contributor to Bevy lately, while also doing contract work in Unity on the side, I obviously disagree with the idea that Rust isn't up to the task of game dev. The grass isn't greener on the Unity side, with a mountain of technical debt holding the engine back. (They're still using Boehm GC in 2024!) Bevy is a breath of fresh air just because it's relatively new and free of legacy. Using Rust instead of C++ is just one part of that. Bevy has a more modern design throughout: for instance, it has a relatively straightforward path to GPU-driven rendering in an integrated system, without having to deal with three incompatible render pipelines (BiRP, HDRP, URP).

What I find more interesting is the parts of the article that boil down to "Rust isn't the best language for rapid development and iteration speed". And that may well be true! I've long thought that the future of Bevy is an integrated Lua scripting layer [1]. You don't even need to get into arguments about the suitability of the borrow checker: it's clear that artists and designers aren't going to be learning Rust anytime soon. I'd like to see a world in which Rust is there for the low-to-mid-level parts that need performance and reliability, and Lua is there for the high-level logic that needs fast iteration, and it's all a nicely integrated whole.

Long-term, I think this world would actually put Bevy in a better place than the existing engines. Unity forces you into C# for basically everything, which is both too low-level for non-programmers to use and too high-level for performance-critical code (unless you have a source license, which no indie developer has). Unreal requires C++, which is even more difficult than Rust (IMO), or Blueprints, which as a visual programming language is way too high-level for anything but the simplest logic. Godot favors GDScript, which is idiosyncratic for questionable gain. I think Rust and Lua (or something similar) would put Bevy in a Goldilocks spot of having two languages that cover all the low-, mid-, and high-level needs well.

As for the other parts of the article, I disagree with the ECS criticism; ECS has some downsides, but the upsides outweigh the downsides in my view. I do agree that Bevy not having an official editor is an ongoing problem that desperately needs fixing. Personally, I would have prioritized the editor way higher earlier in Bevy's development. There is space_editor [2] now, which is something.

[1]: https://github.com/makspll/bevy_mod_scripting

[2]: https://github.com/rewin123/space_editor