Tokamak: A web framework for Zig

15 Mar 2024

I’ve been recently extracting some parts of Ava into a separate open-source projects, and one of them is Tokamak, a web framework for Zig.

So first of all, why Zig? Zig is a very interesting language, it’s a low-level language but with a very powerful metaprogramming support. Hm… ok, I’m too far ahead, so maybe let’s start from the beginning…

Allocators

At first, I thought Zig is not a good fit for server-side programming, because you need to manage memory manually, and the code usually looks a bit more noisy for common tasks. But eventually, I’ve realized that all those allocations are short-lived, for the time of the request, so it’s not a big deal, you can just slap an arena.allocator() everywhere and you’re good to go.

This is actually awesome because you get some kind of predictability, you can identify memory leaks quickly, and you can be more confident about your application in general. If you’ve ever had to find a memory leak in an express.js application, you know what I’m talking about.

Not that it cannot happen but it’s way easier to reason about your application if you know which thing can allocate memory and which cannot. So I guess, it’s a win after all. And there’s something magical about looking at the memory usage charts and seeing that it’s pretty much constant, no matter how many requests you’re serving.

Still, the noise was a problem for me, but I had a scratch at the back of my head, I had a feeling that there has to be some easy way around it.

Dependency Injection

Zig is not my first language, I’ve been doing pretty much everything and I’ve spent a lot of time with Java, way more than I’d like to admit. But there are some good things about Java, and one of them is dependency injection.

Maybe let’s start with why is dependency injection so cool for Java. Java is a statically typed language, just like Zig, and despite the JVM which is actually a dynamically typed marvel, the language itself is, or was, way more statically typed and rigid than you’d like for a web application.

Let’s not get too much into details, but the point is, dependency injection makes it easy to provide (create & init) something at one place and use it from somewhere else. So in a way, you get a bit of dynamicity, without actually sacrificing the static type safety.

So, here’s a cool thing, the same pattern can work also for Zig, and it can solve the same problems. You can create a bunch of things at the beginning of the request, and then you can use them everywhere, without passing them around manually, including the allocators.

So your handlers can look like this:

pub fn createUser(db: *db.Session, data: User) !User {
    return db.create(User, data);
}

and you don’t have to worry about the allocators, because the db.Session will be created at the beginning of the request, and it will be destroyed at the end of the request, and it will pass its allocator everywhere it’s needed.

But now I’m getting ahead of myself again. How do you provide something at the beginning of the request?

Middleware

Apart from Java, I’ve also spent a lot of time with Node.js, and one of the things I’ve always liked is the middleware pattern. You have a function that takes a request and a response, and it can do whatever it wants with them, and then it can call the next middleware.

This is a very powerful pattern but I thought it’s not directly translatable to Zig, because you don’t have closures. But it turns out, you can do a lot of stuff in comptime, Zig is really a lot like JavaScript as long as you’re in comptime.

Ok, here’s a short example of what is possible:

var gpa = std.heap.GeneralPurposeAllocator(.{}){};
defer _ = gpa.deinit();

const handler = tk.chain(.{
    tk.logger(.{}),
    tk.get("/", tk.send("Hello")),
    tk.send(error.NotFound),
});

var server = try tk.Server.start(gpa.allocator(), handler, .{ .port = 8080 });
server.wait();

As you can see, it’s not that far from the express.js…

Oh, where was I? Right, we wanted to provide something at the beginning of the request. So, you can create a middleware that will create a bunch of things at the beginning of the request, and then it will call the next middleware, and at the end of the request, it will destroy all those things.

Something along these lines:

fn withDb(ctx: *tk.Context) !void {
    const pool = ctx.injector.get(*db.Pool);

    var session = try pool.get(ctx.allocator);
    defer session.deinit();

    return ctx.next();
};

And then you can use it like this:

const handler = tk.chain(.{
    withDb,
    tk.post("/user", createUser),
});

How does it work?

So, how does it work? It’s actually pretty simple, the tk.chain function just creates another function (in comptime), which will call ctx.runScoped(handler, next_handlers) with isolate chain and with dependencies reset when it returns.

Here’s a link to the source code.

Obviously, this can be used for many things, like logging, error handling, user authentication, etc.

Getting started

There’s also an example project in the repo which should work out of the box, so you can just make few changes and start hacking.

Future

It’s still very early, and there are going to be a lot of changes, but I think the general idea is sound and will not change. If you’re interested, you can check out the Tokamak repository, and maybe even help with writing docs :)