Reflecting on a Year of Gamedev in Zig
93 comments
·May 2, 2025rob74
bmacho
> Am I the only one who feels this is a step back from platforms such as Stack Overflow?
I have a long-held belief that the single best community information database format is Stack Overflow. (As opposed to wiki, chats, user groups/mailing lists, flat/threaded forums.) An editable top-level question, answered by several editable top level answers with different weighting on different things. (And then chat or small comments, there should be a place for some busywork that should carry no relevant information ideally.) Not necessary the website, but the format.
The second is blogosphere, allow users to write articles and engage in comments. For some reason basically no communities have these. They tend to have discord, and maybe a wiki.
zamalek
I previously worked at a fintech and they, understandably so given the domain, blocked Discord. Think for a second what wonders that does for participating in open source (which the same employer had no problem with).
Discord is not a suitable platform for developers. I think Matrix is pretty good but, fuck it, I'll take IRC over Discord - it's still perfectly functional.
sebtron
I agree, Discord is a terrible user experience for me. The UI is confusing and any piece of information older than a few hours disappears forever into the void.
tobyhinloopen
SO is a very beginner-unfriendly platform. Every time I used it, I either did not get any replies or had my question altered, deleted, removed, closed etc. I have no clue why anyone still uses it.
Buttons840
My 10 year old question was closed as a duplicate of a 7 year old question. I'm still both bitter and amused by this. (This was a few years ago, but it's still closed.)
https://stackoverflow.com/questions/10181706/working-with-a-...
Night_Thastus
It's very hostile. Sometimes I just want to ask a broader, conceptual question. I don't need someone to produce working code. Just discuss.
Sometimes I can't make a working minimal example because I work with info that can't be placed online and writing something ground up that is similar enough and decoupled from the hundreds of files and hundreds of thousands of lines would take me a week or two and may not end up even working out.
mixmastamyk
They decided to not support those kinds of questions—their prerogative I suppose. They do have meta and software design, architecture sites.
rkangel
Yes, this is an unfortunate direction that SO took in the last 5 or so years. In the first 5-10 years of its life it was much more welcoming and a great resource. Now it feels like a static database of (potentially slightly out of date) FAQs.
vfclists
SO has always been this way, which is why I prefer to ask on reddit, and resort to SO when the question is unanswered or quite technical.
robertlagrant
The two options are: 1) no one has had their questions answered on stack overflow, or 2) your questions were duplicates or needed a lot of rewording to make them useful to people in the future.
QuadmasterXLII
Asking a question on stackoverflow that is long term useful to the community isn’t a beginner-difficulty task.
rob74
Yeah, I guess if you have a very specific problem that you need help with, it's better to ask about it on Discord. But for the general questions that many noobs are likely to face that the blog post mentions ("How do you make for loop in reverse?" or "What allocator to use in WASM?"), I still think SO would be the better platform, if only to avoid trying the patience of the helpful people on Discord too much...
firesteelrain
It’s beginner adverse site today. It’s got a lot of great existing questions and answers though
tobyhinloopen
It is.
charlie-83
I always found SO really slow. Often I would get stuck on something, ask a question and then have to wait an hour for a response (which would often become a back and forth of comments with large periods of time between them). Even on discords for less popular languages/libraries you generally get an answer in minutes.
I think this is likely due to the incentives SO creates. I have never actually answered a SO question because it feels like you need to answer perfectly or get torn apart. I will often contribute to a discord question even if I don't full know the answer just because I think I can add something useful.
lelanthran
> Even on discords for less popular languages/libraries you generally get an answer in minutes.
On SO, at its peak, you got an answer even faster because it was public (i.e. the answer was already there, you just searched for your question).
So you're comparing minutes to instant.
tester756
But zig's discord is also public ;o
Lyngbakr
> I have never actually answered a SO question because it feels like you need to answer perfectly or get torn apart.
milesrout
[dead]
dragonelite
I agree i rather see all this discussion etc take place on dedicated forums or other text based options so search engines can index the stuff..
karel-3d
You can still ask your Zig question on Stack Overflow, or they have a semi-official Discourse forum Ziggit (which I find preferable to Discord) where I got answers quickly...
Zig Discord is also just semi-official... there is also Slack, there is also mailing list, libera IRC, Telegram, QQ, Zulip...
koakuma-chan
> Zig Discord is also just semi-official
Is the Zig creator not there? I'm pretty sure he is.
kristoff_it
No, Andrew deleted his Discord account a while ago. Regardless, having the creator present does not make a space official, it's a matter of who owns the space and what the general policy is.
The general policy of the Zig Software Foundation is to not have blessed community spaces outside of spaces used strictly for coordinating the development of Zig (those currently are the Ziglang GitHub Organization and the ZSF Zulip), so all communities are equally non-official.
The main Zig Discord server is owned by community members and is actually not even the only Zig server on Discord.
wmedrano
I thought the Zig creator abandoned Discord due to invasive ads sometime last year.
atiedebee
That wouldn't necessarily make it official though
Zambyte
In the context of Zig, Ziggit has been pretty nice also. I use that instead.
IshKebab
I tried Zig a bit and found it quite nice, but it is very low level. Like in Rust you concatenate strings like this:
let result = format!("{a}{b}{c}");
const allocator = std.heap.page_allocator;
const parts = [_][]const u8{"a", "b"};
const result = std.mem.concat(allocator, &parts[0], 2) catch @panic("allocation failure");
defer allocator.free(result);
bgthompson
My game has a bunch of strings, and my string code currently looks very different to this! For example, copying two lines from source:
var buffer : [64] u8 = undefined;
const level_name_cstring : stringnt = std.fmt.bufPrintZ(&buffer, "{}. {s}", .{icon_index + 1, level_name}) catch unreachable;
The reason is, most of the strings in my game are bounded, and actually, known ahead of time. None of the levels have names that approach anything as long as 64 bytes, so I can actually do a fair bit of string manipulation on the stack before needing to use / save it to an allocator. (At least for now before localization XD)
So it depends on the use case. Sure, string manipulation in general can be tiresome in Zig, but in many cases it's also simple.
IshKebab
Isn't `unreachable` in Zig like in C? So very bad things can happen if you overflow.
This sort of code makes me nervous about Zig.
defen
In safe build modes (ReleaseSafe or Debug) it's an immediate panic, equivalent to calling `.unwrap()` on a None in Rust. In unsafe build modes it's undefined behavior.
MawKKe
There's also `std.fmt.allocPrint()` which functions similarly to `format!()`. Although I'd argue its rather poorly named, like many of the functions in the stdlib...
For wanting to avoid fiddling with multiple repeated alloc/defer-free, it's often convenient to use the arena allocators that allow you to release everything with a single `defer arena.deinit()`
unwind
I would (coming from a C background) guess that `allocPrint()` owes its name from the C standard library function(s) `as(n)printf()`[1]. At least that would make sense, the naming is modernized by being made longer, to gain clarity.
belst
Discord (Chat in general) is really nice if you are the first to have a problem.
If anyone else has the same problem, they can't find it by just googling tho. So they need to ask the same question again which gets annoying pretty fast unfortunately
Zambyte
Not sure why this was flagged. It was an interesting read, thanks for sharing. My biggest concern with using Zig has been the breaking changes that you mentioned, but I have been pushing forward with it for personal projects because it just seems like people are able to work around them easily, like you mentioned.
bgthompson
You're welcome! My experience with the breaking changes in Zig has not been bad at all. For brevity in the blog I didn't get into what I had to do to get things working again, but the process itself was straightforwards. Sometimes the names of some standard library functions had changed, sometimes it was some extra fields in the build system. Whatever the case, the changes were almost always documented in the release notes anyway, so after reading the release notes I basically knew the (small) amount of work I had ahead of me.
thunkingdeep
All posts not fellating Rust are eventually flagged here. It’s not really a neutral venue for more academic conversations about PLT.
jvanderbot
First class SIMD vector support is awesome. The complaint about that not being first class matrix support kind of misses the mark: it's all vectors all the way down.
This is one glaring omission from Rust. Their SIMD integration is library specific and patchwork but improving.
CollinEMac
2. Zig has good builtin support for vectors, but not for matrices.
Oh boy do I have the language for you.
n_jd
If you poke around you will find the author is already familiar with Odin. Might be interesting to hear why they moved to Zig.
IshKebab
> After compiling the game with the flag -Dcpu=baseline, the binary contained only x86-64-v1 instructions, allowing my friend to play the game.
Feels like the startup code should actually verify that it's running on a suitable CPU rather than just crashing.
I don't think C/C++ or Rust do this either but they should!
tuetuopay
pretty much no language do this by default. however, most languages target the -v1 instruction set in 'release' or optimized mode. (e.g. Rust does this, and AFAIK GCC and clang both do this for C/C++)
sgt
Zig is really HN's new love story, isn't it. Or quickly becoming...
Rust is still popular but it turns out the developer joy is pretty low.
WD-42
I’ve been dabbling in OS development. I started with rust and it was… ok. Lots of what you’d expect: unsafe all over the place, weird borrow checker work arounds. It felt like I was bending the languages arm.
I recently started to re implement the (admittedly very basic) kernel in Zig and it’s been a breath of fresh air. The language seems much better suited for the level of abstraction that Osdev lives at. Major bonus is that all the existing C code is directly useable in a zig project without any wrapper nonsense or it can be easily translated.
I think I’ll stick with dig for a bit.
ultimaweapon
> Rust is still popular but it turns out the developer joy is pretty low.
Rust is one of the language I enjoy to use. The problem is you need to overcome its steep learning curve in order to enjoy it, which people tend to give up because it is too hard.
jvanderbot
Which to me is fine. It's not a great hobby language but it is a fantastic professional language, precisely because of the ease of refactors and speed of development that comes with the type system and borrow checker.
christophilus
For me, it wasn’t the learning curve that was the problem with Rust. It’s the compilation time. It’s just so slow. I’m used to OCaml, Go, and Typescript (via Bun or esbuild) with iteration times in the tens-to-low-hundreds of milliseconds. Zig still feels a wee bit slow, but it’s acceptable. Rust? It makes me want to toss my laptop into the fire.
To be honest, I haven’t touched Rust in years, so it may have improved.
jvanderbot
Coming from those, I can see how it'd be a downgrade. Coming from C++ it's like greased lightning.
steveklabnik
It has improved continually over the years, but on the order of 3x-5x, and by the way you talk about it, you’re looking for 10x-100x, so I doubt it’s not still an issue for you.
Buttons840
"Joy" is always highest for new languages, because there are only green fields and enthusiasts. "Joy" is also increased by languages that let you just do what you want, like dynamic typing, or other language features that delay seeing that first error.
jamiejquinn
Snap! I've been developing a roguelike in zig for nearly exactly a year and my experience is very similar. I especially appreciate your mentioning the positives that come with breaking changes.
Seems like you're also a C/C++ developer so I'm curious: how have you found zig's comptime compares to templates? (Personally I've found it a refreshingly simple experience, if occasionally annoying due to pre-1.0 bugs)
bgthompson
For my (relatively simple) use cases, I've had a significantly better time using comptime than templates; the syntax and the quality of the error messages are a lot simpler in Zig. While I've never attempted to do anything super fancy with templates in C/C++, my gut feeling is that comptime can provide most, if not all, of functionality that templates provide. (But C/C++ template experts please chime in if this is not the case.)
jvanderbot
Former Cpp dev here. If I never have to debug a template barf it'll be too soon. By the end of my time in C/Cpp land I'd written more C than anything, with occasional "structs with con/de-structors". Happily the next gen of systems languages fit about in that niche already!
andreldm
I think point 3 is completely off. That's not cherry-picking, that's comparing apples to oranges, those generated Makefiles are not meant to be readable, if you want to compare, look at configure.ac and Makefile.am. I also dislike CMake, but just saying it sucks is poor argumentation. Finally, in the title Ninja and Meson are mentioned, just to be completely ignored. I'm having a great experience with Meson, I would like to learn how the Zig build system is better than it, this piece unfortunately does not provide that.
tuetuopay
I would go even further: showing the generated cmake file is like showing the disassembly from the zig snippet shown.
I still think the point stands though: zig's build system is nicer. Still bad IMHO as I despise build systems where you describe the build in the same language. Most of the time, it's much better to have a declarative rather than imperative build system. All in all, Ninja/Meson/Cargo/etc are much better in this regard.
(and yes, I think build.rs was a mistake, a necessary evil. Definitely not a feature)
jvanderbot
I have never once had to make a build.rs file, except to make python bindings for some of my own rust libraries. Are people out here using build.rs for everyday driving?
tuetuopay
Bindings, and generally any form of code generation, is the most common form of build.rs I see in the wild.
One such example of "daily driving" is for gRPC, where the canonical way to generate the protobuf and gRPC bindings using Tonic+Prost is through build.rs. Another is C-to-Rust bindings, for all of those -sys crates.
Though I've also seen it used to circumvent limitations of Cargo, e.g. for invocations of dependent cargo build steps, etc. Overall, it's less common (or more robust?) than it used to be 5-ish years ago.
amoss
On point 4, I don't use zig but a two second google leads to http://ratfactor.com/zig/stdlib-browseable2/math/atan.zig.ht...
I wonder about the accuracy of the rest of the article.
jamiejquinn
If you don't use zig, you might have missed why that error indicates a stdlib bug. The error in the article is down to a quirk of the type system where literals (like the literal float) have their own unique type (comptime_float) that can coerce to related types (f32, f64) but in this particular case, when atan switches on the type, it doesn't explicitly include comptime_float and fails. Seems like an oversight in the stdlib to me.
amoss
Missed all of that, thanks for the explanation. Why doesn't zig try and apply the coercion before the type switch - is this deliberate for safety?
bsder
The issue is that atan wasn't implemented for a comptime float.
A "comptime f32" is different from an "f32"
Ygg2
Not gonna lie, most of those things look like negatives.
- Zig main source of QnA is a black box (Discord)? One that might be soon enshittified to hell because we have to satisfy the IPO.
- Zig uses Gradle style approach to Maven. Having used Gradle, I honestly can't recommend it. Too much freedom in build pipeline leads to too many footguns.
- Zig changes in big exciting ways, sounds like euphemism for breaking changes. Having those on framework level is tiring, having those on language level is hell.
dns_snek
1. I agree
2. I guess this is a matter of perspective. I don't have strong opinions on build systems but as someone who only occasionally dabbles in systems programming, I know that Zig's build system is infinitely more approachable and coherent than the monstrosities of CMake and any other build system I've ever come across in various C and C++ projects.
3. That's true, but Zig isn't 1.0 yet so that's to be expected.
flohofwoe
> Zig main source of QnA is a black box (Discord)?
The Discord is quite okay-ish for getting a quick answer for things you're blocked on. For proper discussions and deeper questions, Ziggit is a better choice (which is also a 'proper' forum):
> Zig uses Gradle style approach to Maven.
Zig's build system is conceptionally further away from both Gradle and Maven than Gradle is different from Maven (having to wrestle with both from time to time, my personal opinion is: both are a huge pile of excrement, unfit for real-world usage - even cmake is better than what the Java world accepts as build systems, and that's some achievement ;)
At least the Zig build system looks promising in the way that simple things are simple and complex things are possible, even though the API still sometimes doesn't quite know if it wants to be imperative or declarative. But the basic idea is that the Zig build system and package manager are "just" part of the regular stdlib and that your build process is a regular Zig program that gets compiled and then executed to perform the build.
> Zig changes in big exciting ways, sounds like euphemism for breaking changes
...which is completely expected of a pre-1.0 status. OTH, the last couple of versions were not worse than fixing new warnings after a minor C/C++ compiler update.
Ygg2
> ...which is completely expected of a pre-1.0 status.
I fully agree, but I assume you want to ship your game (i.e. its a game not a toy). At least to Windows. Building your game on pre 1.0 language is like building a house on sand that's being transported by a truck.
The Zig Discord is a great resource for anyone learning Zig. At any given time, the zig-help forum is awash with questions from beginners like “How do you make for loop in reverse? or “What allocator to use in WASM?” Most get answered within minutes.
Am I the only one who feels this is a step back from platforms such as Stack Overflow? Discord is basically just a chat platform, and while it's nice that there are always people there who are willing to answer the same questions over and over again, you can't rely on that staying the same in the future. Whereas SO crowdsourced a "canonical" answer to a question, and if someone came up with the same question later (and didn't find the existing answer via the search function or Google), they could be pointed to that answer.