One of my favorite xkcd comics of all time is Random Number. The image depicts a C function that supposedly returns a random integer, but in reality always returns 4. I was reminded of this wonderful comic a few days ago, due to a security incident where people accidentally used the number 4 (among others) instead of a properly generated sequence of random bytes. Satire has a tendency to become reality, I guess.
Since there is a clear demand among developers for the xkcd random number generator, I’m of the opinion that self-respecting languages should ship a built-in implementation. I’m here today to lead by example, as I unveil the glorious @xkcdRandom builtin for the Zig language. Can you guess what it does? Indeed, it returns the number 4! Now go found a multi-billion-dollar startup on top of this genius idea.
Why not? After completing the Ziglings exercises and brain-coding a toy minesweeper game, I felt like my grasp of the language was solid enough to attempt something more ambitious. I wanted to get my hands on production-grade Zig code, and play with it to see how the different pieces of the language fit together. Hacking on the Zig compiler sounded like the perfect idea!
\nA sane person might implement the xkcd RNG as a literal port of the original C code, which maps to the following Zig function:
\nfn xkcdRandom() i32 {\n return 4; // chosen by fair dice roll.\n // guaranteed to be random.\n}There’s one very big drawback to this approach: it’s boring. Don’t get me wrong, I like boring and low-maintenance code as much as the next person… but this time my goal is to learn, not to produce. I actually want to jump through hoops and make things overly complicated! Builtins are perfect for this. They are, as the name suggests, built-in constructs in the Zig language. Adding one means I get to fork and modify the compiler. Yay!
\nThe builtin source code is intertwined with the rest of the Zig compiler, so it’s not really possible to include it here. Feel free to inspect the diff if you are curious (it’s just 22 lines), or to clone my Zig fork and check out the 0.17.x-xkcd branch.
With these changes to the compiler, you can build programs like the one below. Isn’t that neat?
\nconst std = @import("std");\n\npub fn main(_: std.process.Init) !void {\n const someRandomNumber: usize = @xkcdRandom();\n const anotherRandomNumber: f32 = @xkcdRandom();\n std.debug.print(\n "The random numbers you ordered just arrived: {}, {}\\n",\n .{ someRandomNumber, anotherRandomNumber }\n );\n}The Zig developers have clearly put effort into developer experience. You can build Zig with a single command: zig build --zig-lib=lib -Dno-lib -Ddev=x86_64-linux. The fresh build took roughly a minute to complete on my laptop, which in my book is quite good for a compiler. Subsequent builds were faster.
As for the coding work itself, an LLM pointed me to the file where builtins were defined, and I took over from there: modifying things, checking for compile errors, fixing them, and so on. The code was easy to navigate, thanks to its clear structure and the Zig language server’s help. I was done way faster than I expected! Maybe I should find a real compiler issue to tackle next.
\nDid it work? Yes, it did! Once I had everything in place, I built the modified Zig, compiled the example program from above, ran it, and watched it output the 100% guaranteed random numbers:
\n$ ./zig-out/bin/zig build-exe example.zig\n$ ./example\nThe random numbers you ordered just arrived: 4, 4