April 20, 2026
Speed, memes, and mystery compilers
How to Make a Fast Dynamic Language Interpreter
Nerd builds tiny code engine that rivals Python — commenters get obsessed with the weird tools
TLDR: A developer turbocharged his tiny homegrown programming language until it nearly matches big, mainstream ones, all without heavy-duty tools. But the audience is fixated on the oddball details: mysterious compilers, missing super-fast Lua numbers, and a codebase that’s mostly just HTML, turning a tech post into a mini drama thread.
A lone developer just showed how his homemade mini programming language, Zef, got sped up more than 16 times until it could keep up with big names like Python and Lua. That’s cool, sure — but the real show is in the comments, where the crowd immediately zooms in on the weirdest details.
One user is delighted that the project’s code is “99.7% HTML and 0.3% C++,” basically calling it the world’s smallest Ferrari engine hidden in a website. Another skips the tech talk entirely and asks the real question: what’s it like using this mysterious Fil‑C language to build it — is it actually useful, or just a science experiment with a cool name?
Then the drama kicks up: someone complains that regular Lua was tested but LuaJIT — the turbocharged version everyone worships — was left out, sparking the classic “you didn’t benchmark the right thing” energy. And maybe the funniest moment: a commenter admits that neither Google nor their AI assistant has ever heard of the “YOLO‑C++” compiler mentioned in the post, turning it into instant meme material — is this a real tool, an inside joke, or just the most chaotic compiler name ever invented? The tech is impressive, but the community is clearly here for the mystery, the naming drama, and the benchmarks-that-could-have-been.
Key Points
- •The author optimized a simple AST-walking interpreter for the dynamic language Zef to achieve about a 16x speed-up (up to 67x including an incomplete Yolo-C++ port).
- •Performance gains came from foundational techniques: value representation changes, inline caching, object model updates, watchpoints, and incremental optimizations like direct operators and RMWs.
- •A benchmark suite (ScriptBench1) with Richards, DeltaBlue, N-Body, and Splay was created to measure progress, with ports in Zef and other languages.
- •Experiments ran on Ubuntu 22.04.5 with an Intel Core Ultra 5 135U (32GB RAM), averaging 30 interleaved runs; Zef used Fil-C++ v0.677, Lua 5.4.7 used GCC 11.4.0, QuickJS-ng 0.14.0 was from GitHub, CPython 3.10 was from Ubuntu.
- •Most comparisons used Zef compiled with Fil-C++ versus other interpreters compiled with Yolo-C compilers, and the results brought Zef closer to QuickJS, CPython, and Lua performance.