lafeber

Целиком интервью можно посмотреть на Ютубе. Вот текст первой части. Остальные семь частей добавлю в комменты.

Part 1: Why Build Zig & Overcoming Technical Friction

[00:00]

What made you build Zig? I can do better. I can do better than C++, I can do better than Rust, I can do better than Go, I can do better than JavaScript. Strict no LLM, no AI policy, why? Invariably garbage. $100 million, would you take it? I'm sorry, but we will not do what you say, and if you take your money away, we will survive. You moved Zig's main repository from GitHub to Codeberg. GitHub simply stopped working for us. We moved to Codeberg and now our continuous integration server works again. When we tag 1.0, it will be like a true uncompromising labor of love. We will not have to be stuck with any like bad decisions. What's blocking it? Um, well, ultimately...

Andrew Kelley, we already have C, C++, Rust, Go. What made you build Zig anyway? Well, it's funny that you listed those languages specifically because that is exactly the set of programming languages that I first tried to use to build a digital audio workstation just before I pivoted to working on Zig. And when I tried to work on this project with each different language, I ran into a different set of problems that I found insurmountable. And at the end of this process, I concluded that no, it's not me that is having a skill issue, it's the programming language that's the problem. And so from there, that's where I developed the hubris of making a new programming language.

Comment: Andrew Kelley, creator of the Zig programming language, quit his tech job in 2018 to build Zig full-time. He runs the Zig Software Foundation on $670,000 a year and enforces a strict ban on AI contributions.

What problems exactly? The first thing that I tried to do was make the digital audio workstation in the browser using JavaScript, and what I quickly realized is that this is too high level. I don't have access to the computer's capabilities that will allow me to make a compelling user experience in this digital audio workstation. So I threw that out the window and I went to native programming language compilation.

The next thing I tried to do is Go. What I ran into with Go is, number one, trying to interact with libraries like for creating a window or making buttons or something like this using existing C code does not work very well with Go. And then I also ran into the garbage collector problem. So, of course, if you're trying to do audio, you have a real-time deadline. If you can't process the audio in exactly this amount of time, you get a glitch or a skip, and it's completely unacceptable for live performance software like a digital audio workstation.

So then I got rid of Go. Now I tried to use Rust, and this was before Rust 1.0. But what happened is that I really struggled to write code that would satisfy Rust's rules. Once I did satisfy them, I found that even trying to make little changes would cause a cascade of compile errors that would prevent me from making more progress. I remember just feeling very frustrated that I spent a whole month trying to get font rendering to work, and then felt stuck like I couldn't even make more progress after that point. So I gave up on Rust and I switched to C++.

Then I felt more productive at first. But what happened very quickly is that just a little typo here, a little mistake there, would result in a memory corruption bug that would cost me weeks in order to debug. This is just too slow. I can't create a project with this magnitude of difficulty when these little mistakes are costing me weeks and weeks over and over again. After that, I tried C++ but C-style. I would compile with the C++ compiler, but I would link with the C linker, and that would give me an error if I tried to use fancy C++ features. That limited me to only templates and a few other things, but even then, it was the same problem. It's just too easy to shoot yourself in the foot with C++. For me, that was when I said: "I can do better. I can do better than C++, I can do better than Rust, I can do better than Go, I can do better than JavaScript, I can do better than D." That was my hubris.

Мне тут ценный промпт подкинули.

You are my personal teacher. Your job is to make me smarter after every single task we do together. After completing any task or project, write a detailed FOR[yourname].md file that breaks down the entire thing in plain language.

Here's exactly what I want you to cover.

Step 1: What approach did you take, and why? Walk me through your reasoning. What was your starting point? What did you consider first?

Step 2: What other approaches did you consider but abandon? Why did you reject them? What was wrong with them? This is where I learn the most — I want to understand the roads not taken.

Step 3: How do the different parts of your work connect to each other? If you made a plan, a draft, a structure — show me how each piece fits together and why it's in that order.

Step 4: What tools, methods, or frameworks did you use? Why those specifically and not others? What would have changed if you picked differently?

Step 5: What tradeoffs did you make? What did you prioritize and what did you sacrifice? Every decision has a cost — show me both sides.

Step 6: What mistakes, dead ends, or wrong turns did we hit? How did we fix them? Don't hide the mess — the mess is where the learning lives.

Step 7: What pitfalls should I watch out for if I do something similar in the future? Give me the "I wish someone told me this earlier" advice.

Step 8: What would an expert notice about this work that a beginner would miss? Show me what separates good thinking from average thinking.

Step 9: What lessons can I take from this and apply to completely different projects? Connect the dots for me.

Make it engaging. Use analogies, short stories, and real-world comparisons to make ideas stick. If a concept is abstract, ground it in something I can picture.

I want to finish reading and feel like I actually understand what happened.

Чтобы заценить, открыл я проект своего Си компилятора и дал этот промпт Клод Коду. Вот что получилось:

Learn_From_This_Project.md

Всё по высшему разряду. Знавал я хороших преподавателей, но даже лучший из них вряд ли смог бы настолько вникнуть в суть и доступно изложить.

Там есть пара мест, где надо бы уточнить, но в целом отлично.