tofu is a protocol and an asynchronous Zig messaging library used to:
- Build custom communication flows.
- Create non-blocking systems.
- Enable peer-to-peer messaging between applications.
tofu is a completely new project. It is not a port of old code, and it does not use any C libraries. It is built 100% in native Zig. The core functionality uses only the standard library.
As a food, tofu is simple and doesn’t have much flavor on its own.
With tofu cubes, you can:
- Eat them plain as an easy snack.
- Add some spice to make them better.
- Cook up something really tasty.
As a protocol, tofu uses messages like cubes. By "cooking" these messages together, you can grow your project:
- Start with minimal setups.
- Build complex flows.
- Create full distributed applications.
Important
tofu is as good as you are a cook.
tofu did not come from nowhere.
The journey began in 2008 when I first built a similar system. I maintained and ran that system for many years in high-stakes environments. It powered everything from basic IPC to complex data transfers in a custom distributed file system.
I left that project a few years ago, but I haven't heard any complaints yet — the systems are still running strong.
Corporate lawyers can stay calm: I didn't take any code. I only took the "smell." (See the precedent case about paying for a smell).
By "smell," I mean the core philosophy:
- The Message is the API: The data itself defines the connection.
- Gradual Evolution: Start with something simple and grow it into a powerful system over time.
- The Mantra: "Connect your developers. Then connect your applications."
This tofu mantra is a paraphrase of Conway's Law.
tofu "expects" that development starts with a conversation (connection) similar to the one shown below.
Context:
- Two developers are discussing the message flow for a new Print Server.
- The first one is the Spool Server developer (S).
- The second one develops the RIP Worker Process (R).
- Don’t worry — RIP means Raster Image Processing, not what you might think.
- Some terms may be unknown — that’s fine. These two know exactly what they mean.
This dialog is shown without the usual jokes or side comments common in real programmer discussions — just the technical part.
S: I don't know the addresses of the workers, so you should connect to me.
R: I'll send a HelloRequest, because the worker can process only specific PDL types,
the PDL header will contain either PS or PDF.
S: Do I need to send you a HelloResponse?
R: No, just start sending me messages with PDL data.
S: As signals?
R: No, as multi-requests — each with a message ID equal to the job ID.
S: You forgot the Job Ticket.
R: Right. The first request should have a JobTicket header (JDF or PPD) and the
ticket data in the body. The following requests will have the PDL header
(PDF or PS) with the related content.
S: But JDF is usually used only for PDF...
R: Yes, but let's keep it flexible.
S: Can you process several jobs simultaneously?
R: It depends on licensing. Anyway, if I can, I'll send another HelloRequest —
working one job per channel looks cleaner.
S: I need a progress indicator.
R: No problem. I'll send signals with the same message ID — the Progress header
will show the range [N:M] for page numbers.
S: On job finish, send me a Response with the same message ID and processing status.
Also include the Progress header.
R: Why should I send an obsolete message? Are you expecting a graceful close?
S: Of course.
R: Then I'll send a ByeRequest with the same information, and you'll send me a
ByeResponse. After that connection will be automatically aborted.
S: That's enough for today. Send me a short text file with this protocol —
I'll save it in Git.
R: Deal. How about a cup of coffee?
I hope you got the point without long smart descriptions or advertising.
- Message-Based: Uses discrete messages for communication.
- Asynchronous: Enables non-blocking message exchanges.
- Duplex: Supports two-way communication.
- Peer-to-Peer: Allows equal roles after connection establishment.
- Stream oriented transport - TCP/IP and Unix Domain Sockets
- Multithread-friendly - All APIs are safe for concurrent access.
- Memory management for messages - internal message pool
- Backpressure management - allows to control receive of messages
- Customizable application flows - allows to build various application flows not restricted to request/response or pub/sub
- Simplest API - you don't have to bother with or know the "guts" of socket interfaces
Documentation and examples are available on the Tofu documentation site.
For the impatient:
- Karl Seguin — for introducing me to Zig networking
- tardy — I peeked into 2 files of the project (the author will guess which ones)
- temp.zig — helped me (and will help you) work with temporary files
- Gemini AI image generator — the only one out of six I managed to convince to seat Ziggy and Zero at the same table
- Zig Community Forums (in order of my registration) - for your help and patience with my posts
⭐️ Like, share, and don’t forget to subscribe to the channel !