Zig creator seeks 'uncompromising perfection' before blessing 1.0

Software

Andrew Kelley interview describes paying monthly for cloud-powered AI coding as an 'insane proposition'

Andrew Kelley, inventor and BDFL (benevolent dictator for life) of the Zig programming language, was interviewed by Vitaly Bragilevsky, head of the Rust ecosystem at tool vendor JetBrains. 

Zig is a general-purpose language that aims to be as performant as C but with "fewer footguns," in Kelley's words. It is a niche language, at 82 on the latest RedMonk Programming Language Rankings, but well-liked by its users; last year's Stack Overflow survey has Zig as the fourth most admired language, defined as Zig developers who want to continue using the language.

We found the interview disjointed, perhaps because of the way it was edited, but nevertheless it touches on many of the key questions developers face today, including AI, GitHub reliability problems, and programming language choices.

Kelley describes why he created Zig, when other options including C, C++, Rust, and Go already exist. He said he set out to develop a digital audio workstation. He tried Go, but found interoperability with C libraries difficult, and said the garbage collector caused audio delays. He tried C++, or coding C-style using a C++ compiler, but found that small mistakes led to memory corruption bugs that took weeks to fix. He tried Rust but "really struggled to write code that would satisfy Rust's rules," and spent a month trying to make font rendering work.

Kelley's solution was to go in a different direction and create a new programming language. Zig, he said, "does not give up any of the power that C offers, while improving the flaws and weaknesses that C has."

As far as we can tell, the digital audio project remains in its early stages.

The Zig project is known for its no-AI policy, set out in the code of conduct. The reason, he said, is that AI contributions are "invariably garbage" and consume code review time that the team believes is better spent on human contributions. His view on AI, though, is more nuanced than these first remarks suggest. One of his objections to AI is that it is unteachable, whereas the team values mentoring contributors so that they may later become better contributors or even part of the core team. 

Another negative for AI tooling is that it is non-deterministic and therefore its output always needs review, even for something as simple as refactoring the name of a function. Kelley prefers deterministic tools in which he can have full confidence.

Regarding vibe coding – delegating all coding to AI – Kelley said it is interesting but he does not wish to use technology "controlled by four companies." He also said: "I'm not going to go from using my own computer and my own electricity, in order to use closed-source programming on someone else's computer through the network, that I have to pay for monthly. To me, that is an insane proposition."

"I'm always hearing people say that AI code works surprisingly well. But to me, that is not the bar that I want to hold software to. The bar that I want to hold software to is uncompromising perfection."

This perfectionist attitude is also evident in the progress of Zig, which after 11 years has reached version 0.16 and releases are sometimes characterized by major breaking changes. "When we tag 1.0 it will be a true, uncompromising labor of love," he said. It will also be a backward-compatibility promise, he said, whereas in pre-release the team can continue to improve the language without that constraint. The aim, he said, is to create a language for the next 50 years.

Why did Zig move from GitHub? "GitHub simply stopped working for us," he said. "We're here to write software. If our continuous integration server doesn't work, we need to find one that does."

Zig migrated to Codeberg, which Kelley said was "essentially a clone of GitHub, so it was an easy transition to make." He also likes that Codeberg is a German nonprofit, on the grounds that "I find nonprofits to be a more stable business than startups or corporations." Zig itself is funded by the Zig Software Foundation, which is a US 501©(3) nonprofit.

A contentious aspect of Zig is a decision three years ago to "fully eliminate LLVM, Clang, and LLD libraries from the Zig project," though the Clang compiler will remain. According to Kelley, "you want to avoid a dependency for your core product. We have done this with LLVM, so we're in the process of rectifying this mistake."

If JetBrains was hoping for endorsement of its tools, it was disappointed. "I have never used the JetBrains products because it's closed source," said Kelley. He uses a terminal and Vim (Vi improved) for coding, he said. 

One thing we did not learn is when to expect Zig 1.0. According to the release notes, though, "the 0.17.0 release cycle will be short." ®