flecs VS legion

I've also been enjoying building My First Game™ in Bevy using ECS. The community around Bevy really shines, but Flecs (https://github.com/SanderMertens/flecs) is arguably a more mature, open-source ECS implementation. You don't get to write in Rust, though, which makes it less cool in my book :)

I'm not very proud of the code I've written because I've found writing a game to be much more confusing than building websites + backends, but, as the author notes, it certainly feels more elegant than OOP or globals given the context.

I'm building for WASM and Bevy's parallelism isn't supported in that context (yet? https://github.com/bevyengine/bevy/issues/4078), so the performance wins are just so-so. Sharing a thread with UI rendering suuucks.

If anyone wants to browse some code or ask questions, feel free! https://github.com/MeoMix/symbiants

Flecs does just that: https://ajmmertens.medium.com/why-it-is-time-to-start-thinking-of-games-as-databases-e7971da33ac3

I'm trying to optimize my workflow as much as possible, and came across this thing called an ECS. After doing a little bit more digging I found some decent guides on how you would make one, I also found one premade called FLECS. FLECS is nice and all, but I was looking for something more simple that just has the bare bones of what I need and is also configurable. I haven't been able to really find anything like that, so I was wondering if anyone had an example of maybe their way of implementing an ECS. I know how to go about it, but I'm unsure of exactly what the code would look like.

Since we wanted a common game simulation that would be on both the server and the client we looked into a few libraries that would fit our ECS needs. It was decided we were going to write this common part of our game in C++, but rust was considered. C++ was a familiar language for us so naturally EnTT and flecs came up right away. I had used EnTT before, writing some small demo projects, so our choice was made based on familiarity. In order to integrate with Unity we created a small C interface to communicate between our simulation code and Unity’s C#. Here’s close to what it looked like. I removed some parts for brevity sake.

Repository: https://github.com/SanderMertens/flecs

This is the project: https://github.com/SanderMertens/flecs (query engine implementation lives here: https://github.com/SanderMertens/flecs/tree/master/src/addons/rules)

There's also a performance question. While we can now use Blueprint nativization to convert Blueprints to C++ the result will be a fairly naive version, fast enough for most purposes but not if you're trying to push every bit of performance. This is where you're looking at making sure you're hitting things such as using the CPU cache as well as possible for an ECS system (Look at ENTT or Flecs if you want to see what they're about and why you'd want one), or a system needing to process massive amounts of data quickly such as the Voxel Plugin.

If I knew C++ and I'd heard about it before I started my current project, I would have been tempted to use this https://github.com/SanderMertens/flecs which can be built to WASM. Of course you still need JavaScript in the front end to link to the WASM part. I've recently been using esbuild to bundle my front end code, which does a pretty similar job to webpack, but is a bit faster.

When do think bevy will support entity-entity relationships ? https://github.com/bevyengine/bevy/issues/3742.

Flecs ECS already supports this: https://github.com/SanderMertens/flecs/blob/master/docs/Rela...

For a modern engine you’re probably best looking at Unity’s DOTS. You may also want to check out some of the different open source ECS libraries such as flecs and EnTT are two popular ones for C++, but there’s lots of them. Largely you’ll see lots of different approaches taken, all with their own pros and cons. Not all of them will be performant (some focus more on the design benefits) while others will be optimised for certain use cases. What you should prioritise will depend on your specific needs.

I know bevy, there is also legion from the amethyst game framework.

Wow... Okay. I see you didn't even bother watching the video in full. But, I'll respond in case your misinformed comment manages to deter people from watching it.

You somehow managed to ignore all the points of the talk. Like literally every one of them, and I'm not even exaggerating. And then you somehow both managed to focused on irrelevant details and to talk out of your ass (like, at least be right about the details?), all to make a tangential point about how "sUpErIoRlY sMaRt CpP dEvs ArE" because they know this really basic shit.

Let's start.

First you talk about "borrow checker issues" she's demonstrating in the OOO design. The point here was to illustrate the amount of coupling and "explosion" in size with this approach. The point was "OOO is probably not a good design paradigm for game dev". Then she continues how Rust surfaces this bad design paradigm much earlier than it would've happened in other languages/ecosystems. That's all.

More, what you call a custom allocator with tombstoning/generations are just things that exist with a lot of ECS implementations (regardless of language), and it's just become "tribal wisdom" at this point that this table-esque storage backend work really well with ECS (I dunno which one came first; older game devs might know more about ECS history/evolution than I do).

Also not sure what you mean by "that can fail in most of the typical ways and that she's gained nothing over the cpp implementation while wasting effort to force Rust into allowing her to do something really basic"... Maybe I misunderstand what you mean, but it sounds to me like you're comparing working with `Option` as being the same failure mode as working with `void`. That's just utterly ludicrous nonsense.

Her ECS implementation is... Well, it was made to fit inside a few slides for a 40-minute talk. So, yeah it's not great; there's plenty of small issues with it, but none of them are with Rust, or with ECS itself... Did you expect a production ready ECS library you can copy off of slides and use them in your next game, lol? If anything, her implementation has too much passing resemblance with what a C++ implementation would look like (which is something she's very familiar with and her starting point) than a Rust one. So, I'm just bewildered why this is the bone you decided to pick.

If you're interested in what a production-ish ECS implementation in Rust looks like, check out Amethyst's Legion or Bevy's ECS. Although, Bevy takes takes ECS a few... staircases (rather than steps) further with its own ECS. I really* encourage everyone to read up on Bevy's ECS and check out the unofficial Bevy cheatsheet book; I'm certain at leat UE game devs will know how to appreciate how beautiful and ergonomic the design is, but others should as well). Aside: her ECS implementation is still safer than most C/C++ implementations used in published games, but that's a bit besides the point since that's the borrow checker doing its job.

Anyway, the irony is that she much better explains the pitfalls of her own implementation details that you do and she also explains why they're "fine" sometimes. And it was kinda implied that it's fine (from what I remember) because it's no worse than what you end up doing in C++ or Java when implementing ECS, but at least it's safe in Rust. She acknowledges that there's probably better ways to do it.

Here's a few links on some of the stuff I mentioned:

- https://github.com/amethyst/legion

- https://bevyengine.org/learn/book/getting-started/ecs/

Development on Amethyst (the engine) stopped, the Amethyst Foundation is now focusing on developing engine agnostic tools, like Legion and Distill

Though, Legion Github was not too active during this year, and with the recent shutdown repurpose of Amethyst Foundation its future looks even less promising.

Take a look at legion on how to compose data objects in Rust.

I'm trying to avoid that architecture even though that's the path of least resistance to work with existing ECS libs in rust.This discussion about FFI for legion touches on what I'm trying to accomplish. They use a u32 as a custom type ID to deal with the FFI boundary. The size param would be for each instance of that registered component, so the Vec wouldn't have knowledge of the size of each component (which might be 16 bytes for example). Ideally, components and systems defined in the scripting language are more 'first class' instead of scripts only being invoked by entities with a DynamicComponent. This discussion is very helpful btw, I appreciate your thoughts

In the end, I got something to work (also in issue 269):

I ended up with my own, a bit ugly, code. I'll quote it here, because even deserializing a JSON into Legion components is not 100% trivial task. Someone else filed an issue 268 in Legion, I have filed another one too.

Bevy was introduced in 2020 and the patent was filed in 2018. I don't think any of the ECS mentioned predate the patent filing (looks like legion introduced this model in 2019), so it's not as derivative as it seems in the current landscape.

I have been working on a game using Rust, SDL2 and Legion. I statically link SDL2, which mean that only the parts that I actually use gets included. Also i have been using the include_bytes to include my assets. This creates a single executable that is around 3mb (this is without compiler optimizing for size, it could go lower at the cost of runtime speed). Running (upx)https://github.com/upx/upx on it, it comes out to about 900kb.