Birbla

Join the discussion

Write your take first — we'll ask for email only when you're ready to publish.

Sign in
  • Hacker News
  • Author here.

    To be upfront about what this is: I'm not a Rust developer or a PHP internals person. This is an experiment in whether the "point the AI at the original project's test suite" methodology (the way Bun was driven against real-world suites) holds up when the human can't review the code. The oracle is php-src's own .phpt corpus, ~22k tests I didn't write. Current honest score: 3,844 passing (17.4%), with a realistic ceiling around 40-45% since the rest tests C extensions (GD, curl, intl, etc.) that are out of scope.

    "Renders WordPress" means: fresh install completes into SQLite, the front page renders with real posts, a real theme and /wp-admin/ renders without issues. The REST API is untested, and it's currently ~55x slower than PHP on the front page (a bytecode VM is in progress, micro-benchmarks are already at 1-3x of PHP 8.5).

    The scoreboard auto-generates into the repo after every run, whether the number went up or down.

    Happy to answer anything.

    by ekinertac
  • Compare with FrankenPHP?
    by pluc
  • This is a pretty cool experiment. Thanks for sharing!
    by adamtaylor_13
  • Will you answer questions yourself, or will you simply pass on what your LLM of choice writes for you?

    Edit: On further inspection, the blog design, the blog build, the blog articles and even the anecdotes used in the articles are entirely Claude generated.

    Stop being so lazy. Get Claude to do something interesting and use your own intellect to assess and challenge the work in your write up. Or the other way around. Inject some amount of human work, at least. Otherwise, what's the point in sharing?

    by bbg2401
  • Wow. Now did you try to check the setup with something like Claude Fable? Will it find issues, what kind of issues? Another question: how many tokens did this effort cost? Did you learn new prompting tricks?
    by MichaelMoser123
  • Ill preface my comment with saying: this might not be the best solution give the goal of your project to iteratively loop through and improve on the tests each round, and using deps would make that process longer/more complicated having to work potentially with another project.

    .....however.....

    mago, a static analyzer for php is written in rust and might be useful for gaining some "free" performance uplift: https://github.com/carthage-software/mago. iirc it splits out a far bit of its internals so they can be used by other projects (citation needed)

    by wsor4035
  • thanks, mago is a cool project. probably not as a dep tho, the parser isnt where the time goes (the 55x gap is all in evaluation, thats what the bytecode vm is for) and our parser is deliberatly tuned to match php's exact parse error messages, which is itself worth tests in the corpus. but using it as a second oracle to cross check my parser against theirs is actually a neat idea, same trick as the phpt suite but at the syntax layer.
    by ekinertac
  • > Here’s the part I need you to sit with

    no, i don't think i will

    by tensegrist
  • Use AI to make Wordpress secure and not suck as much
    by fuckinpuppers
  • Even an AGI can't accomplish the impossible.
    by lioeters
  • I’d be curious for a similar experiment converting frankenphp to rust.

    https://frankenphp.dev/

    by rbbydotdev
  • Maybe the takeaway is that 20% is about all the LLM can muster.
    by gamblor956
  • I mean, I got them to 100% using the official conformance suite on my copy-and-patch jit compiler/interpreter WASM VM...

    Saw that Salt Language article a day to two ago on how they do the static verification as part of the compilation process (or whatever they really get up to) and that's next on the agenda, tried that with a JavaCard VM I was poking at as its 'computation space' is much smaller but that was too much for my poor little laptop to handle but, apparently, this Salt thing is much different and actually tractable so, we'll see, still working out the details.

    by UncleEntity
  • > Maybe the takeaway is that 20% is about all the LLM can muster

    At this point there's a long list of projects that have used LLMs to rewrite a system in Rust including:

      - Bun (https://github.com/oven-sh/bun/pull/30412)
  - Valkey (https://github.com/ianm199/valdr)
  - Git (https://github.com/gitbutlerapp/grit)
  - Postgres (https://github.com/malisper/pgrust)
    With the exception of Bun, these projects were done pre-fable too, so I bet Fable will make these types of rewrites even easier.
    by malisper
  • Interesting read. Given what the process is producing it's probably quite cost-effective?
    by AmazingEveryDay
  • What do you mean? What's cost effective about this?
    by Chaosvex
  • Why is the AI only able to reach 17%?

    Surely it can just keep iterating until it implements the full test suite?

    by mgaunard
  • its still iterating, 17% is just where the counter is today. three weeks ago it was at 10%, two weeks ago 13.8%. i didnt post this as a final result, i posted becuase wp-admin rendering surprised me.

    but no, it cant reach 100%. around 55-60% of the suite tests C extensions, gd, curl, soap, intl, mysqli, ffi, sockets etc. passing those would mean writing all those extensions from scratch too (libcurl, ICU, an image library...) which is a completely different project. the realistic ceiling for a from scratch engine is around 40-45% and thats the number im climbing towards.

    by ekinertac
  • Money probably. This is a cash burn project.
    by hoppp
  • Is it astonishing you got to 17% with some vibe code? Sure.

    But most of the stuff I’ve vibe coded this year has been astonishing by 2025’s standards.

    If you got 100% I’d be genuinely blown away.

    by lawrenceduk
  • Does anyone know why we write code anymore? Why not pass through to an llm that generates the page on the fly (ssr)?

    Is it cost ?

    by pylua
  • Standards vary.
    by general_reveal
  • The article doesn't go into how they managed the AI context when implementing things but I would not be surprised if it was done in a methodical way, 80% - 90% of the test could have passed.
    by sdesol
  • What I suspect is this 17% is the exact sub-set it needed to hack together to make the goal (running some example website) a reality as this is what those dodgy weasels do if you let them. Then you get to spend 200x the time to fill in the rest of the "speculative features deferred due to no real consumer" on top of whatever dodgy system they made up, which is usually whatever is easiest/closest to the literature instead of the actual intended design. Lots and lots of fun to be had doing the full-pipeline refactors to add that last 2% which need support from tip to tail.

    It's all in good fun, though... probably?

    by UncleEntity
  • My boss asked me to set up a WordPress for a product landing page.

    I naturally won't do this; it's no more than a couple of weeks ago that some SQL injection landed in the search query function of this monstrosity.

    WordPress always was and always will be terrible.

    So I set up the landing page with a Hugo static site, and I've been vibe-coding a WordPress-like dashboard that operates on git repositories containing Hugo sites.

    I call it WorbPress (not released yet), and I'm sure that's what my boss told me to install, or I might've misheard.

    And yes, it's written in Rust (with Axum and Alpine.js), because why not?

    by sshine
  • what, no HTMX?
    by librasteve
  • > because why not?

    I'm not certain, but it seems like you're not being entirely serious here, however..

    If you aren't joking, or for other people in this position, I'd first wonder if the landing page required a search function that would hypothetically be subject to the vulnerability, then I'd wonder about what the normal nature of your business is and how much latitude you personally have in the allocation of billable hours to arbitrary technology choices and whether those do actually align with the deliverable, then if I was the boss I might wonder why you created a bunch of (potentially) out-of-scope random liability using unusual lesser-known tools based on a personal vendetta against WordPress.

    I've been in this position, conceptually if not literally, and I've probably been (in a way, rightfully) fired for it, but my country's labor protections are likely not quite as good as Denmark's.

    If there's a question about why money was spent on implementing a bunch of stuff nobody knows for a reason nobody cares about, especially for a very short-lived thing like a landing page, then it's a sticky situation if the answer is basically novelty. Something like this, if it does serve a purpose, should be planned for and a case made for it, but that also doesn't really seem like agency work.

    If I was asked for WordPress, which I have, and I delivered Rust, I don't think I'd keep that job, but mileage may vary.

    Most work is about solving problems as they are, not what we wish them to be, and if a 5 min job becomes a month long job that the customer didn't ask for, it's an extreme case of yak-shaving.

    by brailsafe
  • Why not use headless WordPress?
    by is_true
  • I feel like not choosing WordPress was a great choice but I'm not sure about the rest of the comment. A simple html file might make for a good landing page though.
    by sureglymop
  • Just to clarify: you think your vibecoded dashboard is more secure than WordPress? Not saying you're wrong, just wondering why you think you're right. Are you auditing the generated code, or is it a giant yolo?
    by kmoser