←back to thread

Fil's Unbelievable Garbage Collector

(fil-c.org)
597 points pizlonator | 5 comments | 05 Sep 25 00:55 UTC | HN request time: 0s | source
Show context
kragen ◴[05 Sep 25 04:55 UTC] No.45135095[source]▶
Hmm, Fil-C seems potentially really important; there's a lot of software that only exists in the form of C code which it's important to preserve access to, even if the tradeoffs made by conventional C compilers (accepting large risks of security problems in exchange for a small improvement in single-core performance) have largely become obsolete.

The list of supported software is astounding: CPython, SQLite, OpenSSH, ICU, CMake, Perl5, and Bash, for example. There are a lot of things in that list that nobody is likely to ever rewrite in Rust.

I wonder if it's feasible to use Fil-C to do multitasking between mutually untrusted processes on a computer without an MMU? They're making all the right noises about capability security and nonblocking synchronization and whatnot.

Does anyone have experience using it in practice? I see that https://news.ycombinator.com/item?id=45134852 reports a 4× slowdown or better.

The name is hilarious. Feelthay! Feelthay!

replies(8): >>45135151 #>>45135324 #>>45135967 #>>45137459 #>>45139406 #>>45139586 #>>45139998 #>>45140957 #
CuriouslyC ◴[05 Sep 25 11:46 UTC] No.45137459[source]▶
With improvements in coding agents, rewriting code in rust is pretty damn easy, and with a battle tested reference implementation, it should be easy to make something solid. I wouldn't be surprised if we have full rewrites of everything in rust in the next few years, just because it'll be so easy.
replies(2): >>45137560 #>>45138530 #
kragen ◴[05 Sep 25 12:02 UTC] No.45137560[source]▶
I have had better experiences with LLMs translating code from one language to another than writing code from scratch, but I don't think the current state of LLMs makes it "pretty damn easy" to rewrite code in Rust, especially starting from garbage-collected languages like Perl or Lua.

Certainly it's plausible that in the next few years it'll be pretty damn easy, but with the rapid and unpredictable development of AI, it's also plausible that humanity will be extinct or that all current programming languages will be abandoned.

replies(1): >>45137599 #
1. CuriouslyC ◴[05 Sep 25 12:07 UTC] No.45137599[source]▶
In the last day I've rewritten two service hot cores in rust using agents, and gotten speedups from 4x to >400x (simd+precise memory management) and gotten full equivalent test coverage basically out of the gates from agent rewrites. So I'd say my experience has been overwhelmingly positive, and while I might be ahead of the curve in terms of AI engineering ability, this capability will come to everyone soon with better models/tools.
replies(1): >>45137874 #
2. kragen ◴[05 Sep 25 12:46 UTC] No.45137874[source]▶
That's great! What's a service hot core? Which agents are you finding most useful, and what's the best way to take advantage of them? I was recently surprised to see Antirez recommend copying and pasting code back and forth to a "frontier model" instead of using agents: https://antirez.com/news/154
replies(1): >>45138225 #
3. CuriouslyC ◴[05 Sep 25 13:18 UTC] No.45138225[source]▶
A service hot core is the part of the service where the vast majority of the computation takes place.

I have actually been on a similar workflow with frontier models for a while, but I have an oracle tool that agents can use which basically bundles up a covering set for whatever the agent is working on and feeds it to gemini/gpt5, and generates a highly detailed plan for the agent to follow. It helps a ton with agents derpily exploring code bases, accidentally duplicating functionality, etc.

I use claude code mostly because the economics of the plan are unbeatable if you can queue a lot of work efficiently. Most of the agents are pretty good now if you use sonnet, the new deepseek is pretty good too. I have my own agent written in rust which is kicking the pants off stuff in early tests, once I get an adapter to let me use it with the claude code plan in place I'm going to switch over. Once my agent stabilizes and I have a chance to do some optimization runs on my tool/system prompts the plan is to crush SWEbench :P

replies(1): >>45138426 #
4. kragen ◴[05 Sep 25 13:37 UTC] No.45138426{3}[source]▶
That's really exciting! Can you run DeepSeek locally?
replies(1): >>45138760 #
5. CuriouslyC ◴[05 Sep 25 14:06 UTC] No.45138760{4}[source]▶
Maybe if you quantize the everliving hell out of it, but I wouldn't. Depending on how much ram you have I'd run qwen coder, gemma3 or if you can go bigger (huge workstation with research class gpu) I'd go gpt-oss120 or GLM air.