>>107610194
Failed to mention, I feed these documentation to the LLM for factual code generation.

If you want an LLM to write correct code, give it documentation.

For libraries, the language, the algorithm, etc.

For example, I have a text file called "~/aleph/txt-literature/literature-on-concurrency-and-garbage-collection.txt". It's a file that contains several books and papers on the titular subjects. I use `pdftotext` to convert them to text, concatenate them, etc. This is the Fish function that does most of the work:

# Defined in /home/chubakpdp11/.config/fish/functions/convdir2txt.fish @ line 1

function convdir2txt
    set -l ncat 0
    set -l nprod 1
    for f in *
        set -l bnm "$HOME/$argv[1]-$nprod.txt"
        ebook2txt $f $bnm
        set ncat (math $ncat + 1)
        if test (math $ncat % $argv[2]) -eq 0
            set nprod (math $nprod + 1)
        end
    end
end

PS: I got a tetromino in retarded new captcha! Wish I snapped a screenshot...

File: thinking-emoji-sk8t3r.png (1.05 MB, 1024x1024)

1.05 MB PNG

If you wanna *quickly* write a C compiler, heed my advice:

- Download the Fraser&Hanson lcc book, and the source code for lcc, feed it to several LLMs, ruminate, think, understand;
- Fire up a language with Tree-Sitter bindings, use it's C grammar to parse the language;
- Apply the lcc pipeline, *make it your own*;

The only thing that would be on your way is `lburg`, which is the "Code Generator Generator" for lcc, based on BURS algorithm.

https://dl.acm.org/doi/pdf/10.1145/131080.131089
https://dl.acm.org/doi/pdf/10.1145/203095.203098

But unless you're writing your C compiler in C (which is retarded, don't use C for anything but embedded code), then most languages support pattern-matching. So just use the pattern-matching facilities of that language to implement BURS, instead of using `lburg`.

Done, you now have written your respectably optimizing C compiler.

File: Screenshot_20251109_053324.png (18 KB, 106x121)

18 KB PNG

You know, it would be a journey to extract Knuth's LR C parser from CWEB's source code.

Anyone up to it?

>>107610194
Use case? Just a neet with too much time on his hands? Open up GCC and copy their implementation.

>>107610405
"Use case" meme uurgh.

As it says in the proposal, it's pedagogical.

Also, when implementing a C compiler, the most important thing you *don't* wanna do is to copy GCC. Massive piece of shit source code. I like their JIT though, and I plan on using it in my ECMASdcript implementation.


Also, I have a job, in program safety. It does not pay much, but it's at least PLT-heavy and most importantly, the furthest thing from webshit.

I'm thinking about writing a paper on the use of polyhedral analysis in UAF detection.

>>107610194
LLM psychosis

>>107610510
I use LLMs to expand my knowledge and learn new things. About complex subjects that people go to school for. Some people use LLMs to generate HTML templates, and these are people whose job it is to write HTML templates.

>>107610530
No, you use a sycophantic autocomplete engine as a feedback loop in your delusions of grandeur

>>107610543
Funny thing about 'sycophantic', because one of them (Grok) just shat over one of my ideas:

https://gapgpt.app/share/ea2ef490-624b-419c-838f-24f1ab1cb6ed

Just face it, anon, LLMs are tools, that are getting better over time. I apologize if your sysadmin shell scripting job is taken over by them, and you are no longer "The Big Cheese" because you can write two lines of ECMAScript. But dem's the breaks.

Use them to learn new, much more complex subjects. Something beyond webshit.