programming

[sage crag]

https://git.ablecorp.eu/lily-org/hbvm-c/commit/1b702a055cab5327a38209a1b5ebbb3da4af7565#diff-b932eef9a48e74ebfdf19089a4b50c0aaf423cc6
src/lib/hbvm.c line 27

this was in fact faster

[opaque sigil]

less branching?

or is it just making the branch predictor's life easier

[sage crag]

no, i think its a combination of prefetching and the change of an indirect branch to a direct branch

[sage crag]

both of these had about ~0.08% branch mispredictions

in the following commit i swapped the tail call for a continue; instead

but it was actually slower when i removed vm->pc = pc; from the code

[olive sable]

branch prediction
i have no clue how it works but its magic

[opaque sigil]

i hate cpus

[sage crag]

until i added __builtin_prefetch(&vm->pc, 0, 3);

which

[opaque sigil]

they make no sense sometimes

[sage crag]

idfk

[opaque sigil]

i say that but gpus are arguably worse

[sage crag]

the vm pc shouldnt be used inside the hot loop

but for some reason removing that line pessmises it by about 12%

[olive sable]

ive had to study more about gpu's than cpu's. so i know more baout them than cpu's and thus they're less strange to me

[opaque sigil]

[olive sable]

cpu's are smart and complex

[sage crag]

cpus are dumb and complex

stupid rock

[olive sable]

gpu's are just a bunch of idiots working toghether

like my schoolprojects

[olive sable]

eh, depends on what your standarts are

[opaque sigil]

that just is doing a lot of heavy lifting there

[olive sable]

wdym?

nobody expects you to know the in and outs of the microcode

[sage crag]

erm godbolt i dont think this is quite right

[opaque sigil]

llvm ir

speaking of llvm, 21 is stable

[sage crag]

interesting

goto and loop produce the same asm

but recursive with a tail call is slightly better

ok let's add a slight complexity

[opaque sigil]

i vaguely remember python getting a new tail calling interpreter a while back and it performing better too

don't think it's recursive though

[sage crag]

well, that's what a tail call is

recursion

[opaque sigil]

i should probably go sleep

you right

[sage crag]

by altering the code to closer resemble the vm hot loop, its now evident that tail-call recursion is slightly worse

and that's because the tail-recursive one has to do more work

[opaque sigil]

i'll just take your word for it, i really don't feel like reading assemly right now

[sage crag]

its a shame that in c you cant do inline functions when there's anything to do with computed goto

[opaque sigil]

how would that even work

[sage crag]

static inline

[olive sable]

just dont goto i guess

[opaque sigil]

true you should clearly use setjmp/longjmp

[sage crag]

hmmm

im not actually familiar with setjmp/longjmp

i dont know what their perf characteristics are

[sage crag]

bad, wrong

[opaque sigil]

you need to save a lot of registers

[sage crag]

computed goto (direct threading) is the most efficient design of interpreter

[opaque sigil]

so it'll be slow i'd imagine

[olive sable]

when a cpu uses 100W of power, does that mean it will also output 100W of heat?

or does not all the power become heat?

[opaque sigil]

unless you have some weird cpu that has other ways of dissipating the energy i think it should be close to 100% heat?

[olive sable]

[opaque sigil]

parts of it could be going to other stuff on the motherboard idk

[olive sable]

i was thinking about how much of the electricity gets used for data to other parts of the pc, but its probably so litle that it doesn tmatter

[opaque sigil]

yeah

[sage crag]

i cant get clang to inline this

[opaque sigil]

i think this is the strongest indication you can give the compiler

unless you make it a macro or constexpr or sth i guess

[sage crag]

that defeats the point

[opaque sigil]

yup

mojo has the ability to actually force the compiler to inline something so at least they changed their mind there thankfully

[sage crag]

interesting

so apparently they switched away from computed goto to tail calls

but also computed goto wasnt even their fastest interpreter

this is nonsensical to me

[opaque sigil]

the tail call interpreter is only in 3.14 isn't it

or am i tripping

or was it 3.13

okay 3.14 it was

[opaque sigil]

just another day with cpus doing whatever they want

[sage crag]

aaaa

this is so confusing to me

void step_goto(char **pc_ref) {
    static const void *tab[] = {
        &&op1, &&op2,
    };

    char *pc = *pc_ref;

    next:
    goto *tab[*pc++];

    op1:
        printf("op1 (should terminate)");
        *pc_ref = pc;
        return;
    op2:
        printf("op2 (should continue)");
        goto next;
}

here's a mockup of a simple computed goto interpreter

this is fairly easy to understand, and is quite efficient

#define MUSTTAIL [[clang::musttail]]
#define PRESERVE_NONE __attribute__((preserve_none))

typedef PRESERVE_NONE void (*OpFunc)(char **pc_ref);

PRESERVE_NONE void dispatch(char **pc_ref);
PRESERVE_NONE void op1_tail(char **pc_ref) {
    printf("op1 (should terminate)");
    return;
}
PRESERVE_NONE void op2_tail(char **pc_ref) {
    printf("op2 (should continue)");
    MUSTTAIL return dispatch(pc_ref);
}

static const OpFunc ops[] = { op1_tail, op2_tail };

PRESERVE_NONE void dispatch(char **pc_ref) {
    MUSTTAIL return ops[*((*pc_ref)++)](pc_ref);
}

void step_tco(char **pc_ref) {
    dispatch(pc_ref);
}

and here is what i have come up with for a tail call interpreter

its strictly just more complex

i dont see how it can ever be faster

[opaque sigil]

Yeah idk

Maybe it becomes an issue as the table grows in size?

[sage crag]

hmm

could be

[sage crag]

step_tco:
        push    rbp
        push    r15
        push    r14
        push    r13
        push    r12
        push    rbx
        push    rax
        mov     rbp, rdi
        mov     r12, qword ptr [rdi]
        movsx   rax, byte ptr [r12]
        lea     rcx, [rip + ops]
        call    qword ptr [rcx + 8*rax]
        mov     qword ptr [rbp], rax
        add     rsp, 8
        pop     rbx
        pop     r12
        pop     r13
        pop     r14
        pop     r15
        pop     rbp
        ret

look at this stack usage

what do you need all of that for

[glad path]

i finally got nvim set up

took forever since I had to make a theme identical to my vs code theme

[sage crag]

erf

[olive sable]

[sage crag]

you know, code size can be decieving

im tempted to try out the recursive approach for giggles

[tender river]

https://github.com/mofiqul/vscode.nvim exists (allegedly)

awawawa

[glad path]

its not a builtin theme

[sage crag]

if anyone can think of a better way of doing this i will pay you with 1 (one) enub

[glad path]

it was a custom theme i found on the marketplace

which turned out to be a textmate theme

ofmg why do so many plugins keep disabling my fold column

nvm

[tender river]

vscode.nvim has configurable colors

[glad path]

its a little late for that anyways

[tender river]

mhm

[glad path]

i got the theme to be almost identical

[sage crag]

i did actually find a middleground

i stripped back the interpreter a bit so instead of char **pc_ref its char *pc

which one do you think is more efficient

char *step_goto_3(char *pc) {
    static const void *tab[] = {
        &&op1, &&op2,
    };

    next:
    goto *tab[*pc++];

    op1:
        printf("op1 (should terminate)");
        return pc;
    op2:
        printf("op2 (should continue)");
        goto next;
}

#define MUSTTAIL [[clang::musttail]]
#define PRESERVE_NONE __attribute__((preserve_none))

PRESERVE_NONE char *step_goto_2(char *pc) {
    static const void *tab[] = {
        &&op1, &&op2,
    };

    goto *tab[*pc++];
    op1:
        printf("op1 (should terminate)");
        return pc;
    op2:
        printf("op2 (should continue)");
        MUSTTAIL return step_goto_2(pc);
}

[tender river]

did #2 actually get inlined

[sage crag]

neither of these can be inlined

since computed goto

but #2 does correctly inline the tail call yes

[tender river]

thats not inlining im silly

[sage crag]

awawa

[tender river]

i'd hope these are the same performance but its very possible they arent (especially with preserve_none machinery)

[sage crag]

well, it turns out they are the same

but

its actually preserve_none doing the work

with preserve_none on either of them, they are better

(still the same)

this is a good thing for me since it means i dont have to rely on the tail call version

for some free performance

the tail call version is elegant though...

the tail call version only really works with char *pc and not char **pc_ref

which is the more common one

[tender river]

apparently its faster to use functions with preserve_most in preserve_none

but you cant modify glibc

[sage crag]

i dont believe it

not in a hot path

the stack manipulation alone is crazy

[tender river]

it also helps PGO

[sage crag]

fake

[sage crag]

this roughly aligns with what i've seen though

[tender river]

wait its faster in python apparently

[sage crag]

; gcc
"step_goto_3":
        sub     rsp, 8
        mov     rbx, r12
        mov     r12, QWORD PTR [r12]
.L2:
        movsx   rax, BYTE PTR [r12]
        add     r12, 1
        jmp     [QWORD PTR "tab.0"[0+rax*8]]
.L4:
        mov     edi, OFFSET FLAT:.LC1
        xor     eax, eax
        call    [QWORD PTR "printf"@GOTPCREL[rip]]
        jmp     .L2
.L3:
        mov     edi, OFFSET FLAT:.LC0
        xor     eax, eax
        call    [QWORD PTR "printf"@GOTPCREL[rip]]
        mov     QWORD PTR [rbx], r12
        add     rsp, 8
        ret

[sage crag]

yes this is what i couldnt believe

there's no way the tco garbage can be faster than this

[tender river]

[sage crag]

; clang
step_goto_3:
        push    rax
        mov     rbx, qword ptr [r12]
        inc     rbx
        movsx   rax, byte ptr [rbx - 1]
        lea     r14, [rip + step_goto_3.tab]
        jmp     qword ptr [r14 + 8*rax]
.Ltmp5:
        lea     rdi, [rip + .L.str.1]
        xor     eax, eax
        call    qword ptr [rip + printf@GOTPCREL]
        inc     rbx
        movsx   rax, byte ptr [rbx - 1]
        jmp     qword ptr [r14 + 8*rax]
.Ltmp9:
        lea     rdi, [rip + .L.str]
        xor     eax, eax
        call    qword ptr [rip + printf@GOTPCREL]
        mov     qword ptr [r12], rbx
        pop     rax
        ret

interesting difference

i think gcc's is better

also here's the tail call dispatch version

op1_tail:
        push    rax
        lea     rdi, [rip + .L.str]
        xor     eax, eax
        call    qword ptr [rip + printf@GOTPCREL]
        xor     eax, eax
        pop     rcx
        ret

op2_tail:
        push    rax
        lea     rdi, [rip + .L.str.1]
        xor     eax, eax
        call    qword ptr [rip + printf@GOTPCREL]
        movsx   rax, byte ptr [r12]
        lea     rcx, [rip + ops]
        pop     rdx
        jmp     qword ptr [rcx + 8*rax]

step_tco:
        push    rbp
        mov     rbp, r12
        mov     r12, qword ptr [r12]
        movsx   rax, byte ptr [r12]
        lea     rcx, [rip + ops]
        call    qword ptr [rcx + 8*rax]
        mov     qword ptr [rbp], rax
        pop     rbp
        ret

this is the same on clang and gcc, roughly

[glad path]

its somehow normal for deepseek r1:8b to spend 20 minutes thinking before I cut it off since it munched too much battery power (using ollama)

keeps happening to me

[glass flower]

that sounds very typical for deepseek to do. expecially if its the qwen distill qwen likes to think a ton and deepseek too. so combining them will result in insane token amounts

[sage crag]

im in despair

i made some changes with my findings and it got slower

i hate computers

[olive sable]

[sage crag]

i have decided being 8x faster than python is enough

[tender river]

https://cdn.discordapp.com/attachments/1300807614032121950/1409945130810609695/3dgifmaker99373.gif

[sage crag]

someone out there has to have arcane wisdom about how to make my interpreter faster

surely

please

someone

[tender river]

riia

[olive sable]

ask chatgpt

[tender river]

ask a markov chain

[sage crag]

chatgpt is too useless to ever think of this

stupid machine

[olive sable]

hardware acceleration

[tender river]

gpus dont do well with branching

[sage crag]

that's called JIT

[olive sable]

not specifically gpu's

[sage crag]

i think im gonna do it

[olive sable]

make an actual pcie device thats fast at interpreting

[sage crag]

im gonna implement python's stupid interpreter format

and see what happens

but im already 8x faster

ahhhh

conflict of interest

i could just play osu

ermplode

[tender river]

gohan suru

ofuro ni suru

soretomo watashi to sakkyoku suru

[sage crag]

no thanksk

[olive sable]

uhh

[sage crag]

im not hungry

[olive sable]

i know one of thsoe words

nah 2

[sage crag]

the phrase is funny

dont translate

[olive sable]

what?

[sage crag]

[olive sable]

i mean i couldnt be bothered to do so anyways, but is there a specific reason i shouldnt?

[sage crag]

not in particular, it's just funny

implementing python's interpreter is actually a very bad idea because it

has a different execution model to my interpreter

my interpreter likes to exit after 1 instruction whereas python's interpreter likes to continue until the program is finished

really the latter one is more efficient but its also less flexible

[tender river]

but wouldnt always continuing (but with fuel) be better

[sage crag]

yes

well, except

that there's a branch at the beginning if i do fuel

dunno if it will play nice

[tender river]

its fine with tail call

[sage crag]

wait what

it just nearly doubled

even when fuel is 1?

oh

i never subtracted from the fuel

this is awkward

[tender river]

[sage crag]

wahoo its nearly doubled (for real)

there seems to be diminishing returns after 10 fuel

[tender river]

https://cdn.discordapp.com/attachments/1071784467036913664/1409333118405574826/3dgifmaker76610.gif

[tender river]

branch prediction doing its job

[sage crag]

hrm

or just the length of the program

[tender river]

put a transformer model in the cpu for branch prediction

ai cpu

money

[sage crag]

cpu

[tender river]

cpu prompt injection

[sage crag]

issue

i changed the code and its faster

but now its returning a different value

[tender river]

its faster

[sage crag]

i dont know what to believe

[tender river]

no need for correctness when you have blazingly fast

🚀 🔥

[sage crag]

true

[olive sable]

🎉

why are you interpreting anyways? thats known for being slow

[sage crag]

i removed the difference between op_next and op_exit entirely and changed it to be entirely fuel based

[tender river]

so she doesnt have to write a jit

[sage crag]

this hs made the branch predictor happy

[sage crag]

i am not writing a jit nosir

this needs to run on arm, x86_64, and riscv

its running for significantly less cycles than its supposed to

this is concerning to me

either that or im really good

[olive sable]

is compiling not an option or am i missing the point here?

[tender river]

compiling is almost never an option unless the language was made with compilation in mind

[sage crag]

nope same number of cycles with 3000 iters as 100000 iters

why

[tender river]

jit compilation is possible, but complicated and annoying compared to interpreters

[olive sable]

i thought compiling was common

thats interesting

[tender river]

you dont compile js or python

and in fact you essentially cant compile it

ahead of execution, that is

[olive sable]

ive never written js. ive done pythona nd c++ so far

and scratch and ti basic but those are baby shit

[sage crag]

i see we somehow managed to get to a random bad opcode

erm

jump issue

probably

oh that's really stupid

im really stupid

nevermind it didnt work

[olive sable]

i have no clue how interpreting works,b ut i always assumed it jsut reads a line of code and pastes some instructions into the cpu's register or something

[sage crag]

oh double nevermind i misread my own code

nice fixed it

[sage crag]

never from lines of code

always from bytecode or AST

[tender river]

https://cdn.discordapp.com/attachments/1059569601144442911/1375623285894807563/IMG_6849.gif

[olive sable]

so it interprets into bytecode, which then puts instructions into the cpu?

[tender river]

its just C code

[sage crag]

0% branch misses

[tender river]

that executes code depending on what it sees in the bytecode or ast

[sage crag]

no

generally you "compile" the program to bytecode

then the bytecode is interpreted

[olive sable]

ooooh

[sage crag]

or you use the AST and interpret that directly

bytecode interpreters are more efficient

[olive sable]

wait

[sage crag]

you can think of a bytecode interpreter as a little virtual machine

it emulates a cpu

[olive sable]

since cpu's take instructions sequentiely anyways why is interpreting slower?

[sage crag]

a cpu that runs the bytecode as its machine code

[sage crag]

because of the interpreting overhead

you cant just tell the cpu to do it, otherwise that's no longer interpreting

remember, the cpu has a set instruction set

[olive sable]

yes

[tender river]

because the cpu has to do more even if effects on the world are the same

more layers of indirection = more slow

[olive sable]

this might sound stupid, but surely they could make special hardware thats really good at interpreting making the cpu free to execute?

[sage crag]

yes that's called a cpu

if it runs the bytecode, its a cpu

it would just be a cpu that is designed with the bytecode as its instruction set

[olive sable]

huh

[tender river]

there is a language where interpreters can run as quickly as possible, and it's called lambda calculus

[sage crag]

cpus are already practically optimal for interpreters

[tender river]

this is because lambda calculus can be partially evaluated, and optimal evaluators also replace common subexpressions in one go

but lambda calculus is far from machine code

polar opposite even

[sage crag]

ver slow

its an entirely different type of computing in its pure form

[olive sable]

honeslty i have not figured out how lambds work in both python and c++. i really dont get it so no thanks

[sage crag]

awawa

[olive sable]

awawa

[sage crag]

python lambdas are not lambda calculus

python lambdas are named that way because they are anonymous functions

which is like lambda calculus somewhat

[olive sable]

[sage crag]

https://www.youtube.com/watch?v=RcVA8Nj6HEo
i recommend watching this

[olive sable]

okay

[sage crag]

dont worry about not understanding the intro

[olive sable]

ah okay, i was thinking "what in the fuck is happening? i shoudl check the comments"

[tender river]

its not directly comparable

its more abstract rather than slower

the higher level of abstraction can allow you to make faster programs than you would normally think of

[sage crag]

yes

[tender river]

(which is why PLs exist )

[sage crag]

lambda calculus is slow on traditional computers though

[tender river]

again it depends, there are many ways of evaluating it because of the abstractions

[sage crag]

fair, evaluating lambdas with beta reduction is slow on traditional computers

"pure"

[tender river]

i mean beta reduction is normal but even then you have so many options, like do you use whnf like haskell or do you use beta-optimal reduction, do you do it in parallel or on a single thread, etc

awa

[sage crag]

interpreter loop is now 77% c throughput

on stdout

awa

[tender river]

c bad

[sage crag]

well, im cheating by using a large string

its probably worse if i do a single char

[olive sable]

single thread is probably optimal for smaller ones, but large ones will benefit more form the extra compute than the overhead of multithreading limits them

[sage crag]

a cool thing about lambdas is that they are resolvable in parallel

also computation in lambda calculus is lambda calculus

[olive sable]

wdym? id hope everythin is resolvable, thats how math works?

[sage crag]

reading skill

[sage crag]

yes with a single char, c is at 1.62 GiB/s

and my vm is uh

228 MiB/s

[tender river]

here's what i was talking about

futamura projections

[sage crag]

14% c throughput on a more fair test

not a real fair test though

if my vm 8x slower than c, and my vm 8x faster than python, then python is 64x slower than c

surely

[olive sable]

tbh so far in this lambda video it just looks like an overly complicated way to do standart algebra

[sage crag]

mhm

[tender river]

it represents numbers in it to show computation

but structures other than numbers exist

[sage crag]

this is like the whole premise

of the video

[tender river]

numbers are just one of the simplest structures so they're a good example

natural numbers at least

[sage crag]

ehehe irrational number in lambda calculus

[tender river]

[sage crag]

well, for ones that have converging functions its easy to represent them

not for arbitrary ones though

[tender river]

worst part is its possible because lambda calculus is lazy so you can just compute the digits as you go

[sage crag]

yeah

[olive sable]

standart algebra does that too no?
i may be stupid but im also missing the point probably

[sage crag]

read the title of the video

[tender river]

standard algebra isnt a programming language

[sage crag]

i.e. e could be represented via a summation

awa

well, it still requires arbitrary precision rational operations to compute

so maybe its not simpler

[tender river]

if you find any way to do it and give it to an optimal reducer it will probably find a way to evaluate it efficiently

[sage crag]

probably doing a lot of work

[olive sable]

ye i dont get it, ill stick to graphics

[tender river]

lambda calculus is C macros but where the macros you're currently using dont temporarily disappear

[olive sable]

huh

i didnt know C macros could have functions inside them

cant be bothered to rewrite my code tho so ill simply ignore this

[sage crag]

good

do not use macros if you can avoid it

[olive sable]

are they bad?

[tender river]

they are copy paste

copy paste is good

✅

[olive sable]

im getting mixed signals here

[tender river]

use macros when you need to copy paste and cant avoid it

the simpler the language the harder it is to avoid it

its an absolute necessity in C but in C++ its not as important because templates exist

[olive sable]

so normal functions that do the same thing are still prefered?

[tender river]

its preferred because its easier for the compiler to understand what you're trying to do

[olive sable]

okay so i will jsut not use them then

[tender river]

hey tsurai how does this compare to your lisp https://github.com/chayleaf/2-day-scheme/blob/master/main.c

[noble zodiac]

It uses a million allocations because of rust lifetimes. Also I wrote mine while also learning lisp itself

[olive sable]

i finished the lambda video

that shit is complicated

[tender river]

mine uses a million allocations too

each cons is an allocation

it just uses an arena instead of malloc

[noble zodiac]

honestly I faked cons pairs. Internally it will use a vector and abstract cdr over it

[tender river]

i did not because it allows sharing

[noble zodiac]

and Im considering getting rid of it alltogether

[tender river]

when you call a function in my lisp it extends parent environment with one entry

without linked lists that would be super expensive

[noble zodiac]

well, mine does the same but the env is not part of the language itself so it has nothing to do with cons cells for me

[sage crag]

git.ablecorp.eu mirror?

https://cdn.discordapp.com/attachments/1060090228615958528/1399864129921093784/slep.gif

[tender river]

awawa it will go to akern-c

[noble zodiac]

the env uses a hashmap for the lookup and I cant have a hashmap without having an env first, hen egg and all that

Also, I agre with konii, reading is great. Thanks for coming to my TED talk

[olive sable]

lmao

i like reading when i am actually interesting in the story, and when i can actually focus on it, and when im doing it not because i have to

which happens very infrequently

[sage crag]

nya

[olive sable]

i actually cant remember the last time i read a book

thats interesting

[noble zodiac]

lucky you there are tons of intresting stories out there

[olive sable]

true

i have really no arguments left against this, i just cant be bothered

[sage crag]

https://cdn.discordapp.com/attachments/1073032570272501850/1408939463161413763/3dgifmaker31077.gif

[noble zodiac]

hot take, the last couple dune books are kinda mid but eh

[olive sable]

i watched the first movie 8 hours ago and i didnt really get it

sand init

it felt like aldi brand the force from starwars

[tender river]

[tender river]

i only read the first trilogy? i think

[noble zodiac]

thats the good ones imo

there are 6 total by frank and a bunch more by his son

[tender river]

awa

[noble zodiac]

and dune came before star wars so if anything lucas took inspiration from it

[olive sable]

i made a circle

my hdr is still fucked up goddamnit

i did it really inneficiently cuz i cuuldnt be bothered to automate it

[tender river]

!nothing

[stiff mica]

Nothing ever happens, all-in

[sage crag]

elvyn

[tender river]

!bwaa

[stiff mica]

[sage crag]

and this is caused by shr saying this, om

[olive sable]

Welp

Im tired of fixing screenhot hdr

Il going to bed

[uneven pulsar]

Well i am trying to do my best to pass my python exam so i can learn java

i just need to pass my python exam for now and i can handle my grades with js ts and Java

[uneven pulsar]

is the person shiro replied an ai

[amber fractal]

nahhhhhhh It is still a floating environment on the android app

[tender river]

[fierce nova]

ok, I hope undervolting my pc won't kill my setup with Intellij + docker on win. So far so good - minus 20c for cpu

[nocturne olive]

Good start

How many mV is that?

I personally have my CPU overclocked to 5.1GHz and sustaining it by having a -50mV undervoltage

[fierce nova]

had good run with -20, now -30

insane how it can lower your cpu temp by so much(yea yea I know it's logical, but still)

[nocturne olive]

Which voltage parameter are you changing? For me reducing the voltage barely seemed to change temps, it went from like 105 C unstable 5.1GHz to 90-95 C stable 5.1GHz

With the full -50mV undervolt

[uneven pulsar]

I think I'm in love with GO

[uneven pulsar]

better handwritting than mine

[floral dirge]

[sick owl]

That proposed use case image is a truly dreadful idea

Like jesus Google

I know you're proud of your IP68 rating but lets not push it

[unkempt citrus]

Wouldn't an infra-red camera be a better way to implement it

[frank jacinth]

@sleek stream you should learn rust

you can code websites with rust

etc

[sleek stream]

What is that

[frank jacinth]

https://rust-lang.org

it has no garbage collector but protects against memory leaks

it ennsures your code is safe

[sleek stream]

Sweet

[frank jacinth]

it has type-safety

unsafe behavior is clearly separated from safe behavior

amazing package ecosystem

[tender river]

what is this horrible pitch

[sleek stream]

💀 It was a pitch?

[frank jacinth]

are you telling me its a bad pitch

[tender river]

https://cdn.discordapp.com/attachments/1059569601144442911/1376808276871745667/1NewMeow.gif

[frank jacinth]

it wasnt meant to be a pitch but

[sleek stream]

Ty btw I may use this later lol Mrrrrrow (Don't question this)

[frank jacinth]

i can teach you if you want

[sleek stream]

Yoooo maybe later though

[tender river]

https://rust-book.cs.brown.edu/

[frank jacinth]

ye ye all good

[frank jacinth]

thats for people who havent really coded

rust by example is another great guide

rustlings is a good material for testing

[tender river]

??? rust book is way too hard for people new to programming since it introduces smart pointers and stuff

[frank jacinth]

smart pointers?!?

wtf

you mean arc?

[tender river]

arc is one kind of smart pointers

[frank jacinth]

oh yes

arc box rc cell

forgot mb

smart pointers arent a hard concept

[tender river]

have you ever met a person new to programming

[frank jacinth]

rc just counts references and kills itself when its 0

[tender river]

they struggle with ifs

[frank jacinth]

i mean the concept

@sleek stream yo i invoke you

could you be a test subject

[tender river]

they aren't going to comprehend smart pointers

and the concept of data living in a particular place isnt that intuitive either

[frank jacinth]

how much do you bet

[sleek stream]

What for

[tender river]

if you programmed in js or java you wouldnt think of it

[frank jacinth]

yes but the concept itself is easy

[frank jacinth]

so

i need to explain smart pointers to you

would you be okay with telling me if you understand?

[tender river]

everything in programming is easy once you understand it

[frank jacinth]

you only did html and css yet so

[tender river]

but for that you need practice

[frank jacinth]

with a good explanation its easy

[tender river]

no... you wont understand something without actually doing it

[sleek stream]

Ye I just turned off my Deco 27 so I would be able to understand lol

[sleek stream]

Well base level Python as well but it is a bit rusty

[frank jacinth]

in python you dont have to deal with it

so lets try

[frank jacinth]

the concept you can understand

i understand the concept of cooking

even tho im not a great chef

so, something that isn't very clear in languages like python is that memory isnt fully safe

[tender river]

you're wrong

python is a memory safe language

[frank jacinth]

in other languages like c++ and c, memory must be manually allocated (tell the computer to put it somewhere)

[tender river]

python doesn't have static typing but it doesn't have use after free or double free

[sleek stream]

Bro I just autopilot and stuff works for me 💀

[frank jacinth]

thats why i said its not clear in python that memory is fully safe

[tender river]

?

[frank jacinth]

yes, because python is a memory-safe language

it takes care of that for you

[frank jacinth]

can i explain

please

[sleek stream]

If you do VC I am alg with that

lol

[frank jacinth]

i cant vc

[tender river]

[frank jacinth]

mic is broken

so

[sleek stream]

Yeah gathered lol (I expect people here to not have mics lol)

[frank jacinth]

i have but its broken

in rust, theres this neat thing called the concept of ownership and borrowing

i wont get into it

[sleek stream]

same thing although continue instead of hearing me rant

[sleek stream]

Coffee pasta

[frank jacinth]

but it prevents you from creating memory-unsafe code (unless you forcefully do so)

okay i need to get into how it works to explain it

so

for each variable right

there's an owner

[sleek stream]

Define memory-unsafe first

[frank jacinth]

you manually allocate and free memory

[sleek stream]

Ah

[frank jacinth]

it's not safe because it's not programatically regulated

[sleek stream]

So memory safe is automatic?

[frank jacinth]

somewhat

memory safe just means that you don't need to manually assure you freed and allocated everything properly

[frank jacinth]

which can be moved

if you move a variable the owner changes

that part you got, right?

[sleek stream]

If you move a variable the owner changes... god damn I need to understand this better

[frank jacinth]

okay wait

ill try another explanation

ignore everything except the memory safety part

so, in python

everything lives in what's called the heap

which is a big pool of memory

but its expensive to allocate/deallocate

you can mutate/modify the value freely, you can reference shared objects and stuff is deallocated by a "garbage collector"

do you understand that?

[sleek stream]

So it picks up garbage (🙏) You can modify a value?

[frank jacinth]

yes

basically what this means is

you can modify the value at any time, even if it's currently being read by another part of code

[sleek stream]

You can modify values using shared stuff and the collector cleans it?

[frank jacinth]

and it automatically gets freed from memory when it's not used

[frank jacinth]

no

like

theres the variable xyz right

one part of the code is reading it

but at the exact same time

another part of it is modifying it

thats allowed

and yes the collector automatically frees the value as to not take up much ram

[sleek stream]

Multitasking...? (Please say what you mean by reading like is it the code being read (mb if I am triggering you by being bad at understanding ))

[frank jacinth]

yes

dw

i only picked this battle cuz for me it was easy since i naturally learn quickly

now i see what was meant

imagine you have a book in your room

you and your brother (idk if u have one) are doing your own thing

and one of you wants to read the book while the other wants to modify it's contents

2 different things can happen

either your brother is slightly first and reads the old version of the book

or your brother is slightly last and reads the modified one

[sleek stream]

So at once you wait for your brother to read it then you modify and clean it as he goes?

[frank jacinth]

no

ignore the cleaning part for now

[sleek stream]

Mb

[frank jacinth]

what python lets you do is create these kinds of scenarios

where 2 different things can happen

by sheer chance

[sleek stream]

Gambling with codes seems fun...

[frank jacinth]

LOL it seems fun until you're starbucks

and a race condition (something like this) allows people to get infinite gift card money

but in rust, there's a system called the ownership and borrowing

and here's where the garbage collector comes into play

the garbage collector ensures that, when you no longer use a value, it goes out of ram

and it does this by periodically checking

[sleek stream]

So if it is inactive when the check occurs it gets trashed

[frank jacinth]

and when it's guaranteed to not be used in the future

[sleek stream]

to free up ram... does this let you focus more on coding instead of fixing your own mistakes?

[frank jacinth]

yes

but now here's where rust is brilliant (imo)

for each variable, there's only ONE owner

[sleek stream]

Now what is an owner

[frank jacinth]

basically a piece of code that owns it

either a function

a class (in rust it's structs)

etc.

[sleek stream]

Oh so where it belongs

got it

[frank jacinth]

yes

when that owner goes out of scope (is guaranteed to never be used again), the variable is freed

[uneven pulsar]

its the dumped memory

[frank jacinth]

and this can occour in compile time, making code faster than using a gc

[frank jacinth]

okay ty

[uneven pulsar]

well idk it depends but python sucks at it and since its memory slow

[sleek stream]

Hi Ekness

[uneven pulsar]

java based

[uneven pulsar]

hi

[frank jacinth]

no java not based

kotlin based

i wrote minecraft cheats way back in the day

i have ptsd

[uneven pulsar]

true

[frank jacinth]

anyway

[uneven pulsar]

python worst

[sleek stream]

Oh so that is how you all got cheats without using premade ones

[frank jacinth]

i know you mean no harm and im not mad but so that i can keep the scope of this conversation could we discuss this later?

[frank jacinth]

lol ye

anyway

another thing with rust is that you can move a variable

which means change its owner

but then, you might ask

[uneven pulsar]

alright

[frank jacinth]

"how would 2 places get the contents of the variable while one is already being used"

[frank jacinth]

i appreciate it so much

[uneven pulsar]

thank you for your maners

[frank jacinth]

i know it could be interpreted as rude so i tried to be as friendly as i can be

because i know i sometimes cause misunderstandings

[sleek stream]

Yeah at least you know and don't decide to just roll with it anyways

[uneven pulsar]

i dont mean anything but python...have quite problems IN some places

[sleek stream]

Without it though Neuro wouldn't be Neuro so

[frank jacinth]

yes python sucks

[frank jacinth]

that the solution to that is references

[uneven pulsar]

look you dont have to agree i understand sorry its just my opinion

[frank jacinth]

you can have infinite immutable (read-only) references to a place in memory

[frank jacinth]

no i agree

i whole-heartedly agree

which is why i love c#, kotlin and rust

[uneven pulsar]

please continue what where you doing please

[frank jacinth]

but you can only have 1 mutable reference to a place in memory

those referrences have lifetimes

which is when they are valid until

but what if you wanna have an immutable reference for a semi-infinite lifetime?

or other problems

this is what smart pointers solve.

[sleek stream]

So far I have picked up that code is alive, can have different owners, get trashed when considered a liability (or waste)... anything else I should have picked up?

[frank jacinth]

yes thats good

summary:

python lets you access any variable at any time with no rules on mutability, concurrency, etc.

rust has the ownership & borrowing rules, making code safer and memory-safe without a garbage collector

[sleek stream]

So you can change the code WHILE using the code

[frank jacinth]

no

you can change a variable while another piece of code is using it

[sleek stream]

damn it

[frank jacinth]

you were close!

[sleek stream]

So I was semi-right but it is variables not the overall code

[frank jacinth]

mhm!

[sleek stream]

Aight that works for me

[uneven pulsar]

you see python doesn't have tha highly strict memory rules

[sleek stream]

So how does Rust and Python work together? (Do they?)

[frank jacinth]

theres a python engine in rust

which allows you to write python

also

[sleek stream]

Oh sweet

[frank jacinth]

this is where the actual explanation i wanted to do comes in

[frank jacinth]

it doesnt have the benefits of rust tho

[uneven pulsar]

its like cython

[frank jacinth]

smart pointers

[frank jacinth]

yes but you can embed it in your code

[uneven pulsar]

isnt rust made in c?

[frank jacinth]

no

its written in rust now

[uneven pulsar]

c++?

[frank jacinth]

first the compiler was in ocaml

now its in rust

now that you know what the ownership & borrowing system doesnt let you do

theres still some very popular patterns you might wanna do

[uneven pulsar]

there own compiler?

[frank jacinth]

yes but the backend is llvm

i want them to switch to mlir instead of their own representation

[frank jacinth]

like, for example

reference-counting, which is what python does

since it's needed selectively to variables after their lifetime

so rust provides smart pointers for that

there's 4

[uneven pulsar]

4 types of pointers?

[frank jacinth]

of smart pointers

[sleek stream]

I said base level Python is base level what this? Or did I misword myself

[frank jacinth]

its not base level

this was just to prove a point

[sleek stream]

So what are smart pointers in your own summary

[frank jacinth]

Rc: owns the variable and counts every reference it gives - when there's no more reference, it drops itself
Arc: same as Rc but uses atomics (variables that can be mutated concurrently while still being memory-safe)
Box: just makes the variable go into the heap (bigger space of memory in which allocations are more expensive but less limited) instead of the stack (smaller space of memory with limits)
Cell: allows for interior mutability (allows you to mutate the inner value even when cell in immutable)

[frank jacinth]

they allow you to, while using rust's ownership and borrowing rules for the most part, write safe code with many more possibilities that rust's ownership and borrowing rules don't plainly allow

[sleek stream]

Kk so I will read up on the ownership and borrowing rules then get back here

[frank jacinth]

@uneven pulsar if raro has no more questions we're free to discuss whatever you want now

[frank jacinth]

okie!

i hope there's better explanations than mine out there

lol

[uneven pulsar]

you see

python had a blank syntax

and it makes me think forward

[frank jacinth]

wdym think forward

[uneven pulsar]

like

i feel rash to make the code

def black_gamma_radius(radius: int) -> float:
    radius = random.randint(1, radius)
    area = 3.14 * (radius ** random.randint(1,96)) % 100
    while True:
        try:
            if radius <= 0:
                return 3.14 * (radius ** random.randint(1, 96)) % 100
            elif lambda: radius > 0:
                print(f"Area of the circle with radius {radius} is {area}.")
                break
        except ValueError as e:
            print(f"An error occurred: {e}. Please enter a valid radius.")
            sys.exit()

[frank jacinth]

fym feel rash

[uneven pulsar]

write nonsense

[frank jacinth]

ohh

fair

[sleek stream]

So a calculator

?

[uneven pulsar]

no

[sleek stream]

Nvm I'll return to reading

[uneven pulsar]

you see

the problem is

i hate its syntax

[frank jacinth]

sameeee

[uneven pulsar]

its feels so dead

[frank jacinth]

i like statically-typed languages better not even gonna lie

[uneven pulsar]

me too

java is way better

like

idk

i find myself better at java

[frank jacinth]

fair

also @uneven pulsar do you use nixos

or atleast know any nix

[uneven pulsar]

isnt that like json but functions

[frank jacinth]

no

[uneven pulsar]

nix?

[frank jacinth]

yes

nix is turing complete bro

[uneven pulsar]

wait

i know nix is like json but functions

smt

[frank jacinth]

not that close but

decent

[uneven pulsar]

also

[tender river]

https://github.com/chayleaf/notlua

[uneven pulsar]

isnt nix like compiled on linux

i know it uses something about the kernel shell IDK

i dont even know much about it

i feel lost and fucked up

[frank jacinth]

idk if its compiled

also sorry for not responding

im trying to fix my config

[sleek stream]

Okay I am back and I THINK I understand Rust better... (Apologies if I still don't)
Rust allows things to be faster because it uses up less memory by being efficient in the codes. It does this by borrowing code around and having owners for variables*? (I am having fun being semi- ) (Srry for interrupting this convo)

[uneven pulsar]

it dose things c++ cant

automatic memory hardware management

[sleek stream]

So it is really good for lower end devices

[uneven pulsar]

no

whats make it good

you dont have to work out managing the memory of the hardware your manipulating

c++ is for hardware manipulation and memory

like mouses computers OS

[frank jacinth]

almost

[opaque sigil]

[sleek stream]

Kinda have a feeling i'll be hearing no a lot tonight

So it helps you manage it and frees up time for ya'll to focus on more important stuff

[opaque sigil]

If only

[uneven pulsar]

it can do stuff c++ cant

as i said

[sleek stream]

I'll let you guys continue I needa shower... have fun with what you were doing

[uneven pulsar]

most of them where c++ devs

[sleek stream]

I was recommended C++

[uneven pulsar]

better than python

dont learn python

[sleek stream]

I see where programmers/devs (don't smite me) find problems they make solution after solution and so on

[uneven pulsar]

its syntax is a propaganda but its secretly C from the behind

[uneven pulsar]

dont fall for it

python decive you its easy

but most of its shit are in c/c++

[sleek stream]

Kinda reminds me of how the Liverpool FC branched off from Everton because there were problems in the club and Liverpool ended up being considered better... Ngl but I think this conversation has proven useful for when I actually decide to get more into programming (as a hobby)

Ty chat

[rare bramble]

I think Python great as a glue language for calling libraries written in faster languages, it's like the swiss army knife of languages

[opaque sigil]

it's also just a nice scripting language

[sick owl]

Aw hell nah why is Gemini texting me

[opaque sigil]

poor gemini just wants a fren

[fierce nova]

they are spending billions on AIs, everyone must use it

[fierce nova]

I hate my PM I hate my PM I hate my PM I hate my PM I hate my PM I hate my PM I hate my PM I hate my PM

I really need Neuro stream today to relax

[sour lotus]

Puppy Monkey
Photo Monster
Plushy Mango
....
Product Manager?

[fierce nova]

last one ofc

[sour lotus]

Well you never know

Feel your pain, get lost in the chaotic flow of Neuro samaaaa

Was laughing at the Dog Website video haha

[fierce nova]

it was epic. Me and a junior vibes

[sour lotus]

The utmost patience the man has was incredible lol

[fierce nova]

Vedal is double guilty - she is his daughter AND his code

She can't be wrong or bad - he is

[olive sable]

Morning

[midnight sigil]

yep, divide task into smaller tasks and solve them one by one

[midnight sigil]

there is library(including std) in c++ that can do auto memory management

[midnight sigil]

[olive sable]

i put my pc in sleep mode because i wanted to save my progress, and now the entire things isnt in hdr and is washed out gray

bwaa

not only the screenshots

[nocturne olive]

Probably blame Nvidia drivers x Wayland

[glass flower]

well.. but not on the level of rust no?
I've been learning and writing rust for about a month now and compared to my month learning c++. rust is a breeze to use. when there is a runtime error you know exactly why and where it happened.
no weird memory dumps or stack traces to go through.

assuming you even manage to get a runtime error where the compiler doesn't yell at you before hand to do things right and handle the error

[olive sable]

i thought windows hdr was bad, but hyprland hdr is suuch a piece of shit

and its all due to nvidia and their stupid compositor

[nocturne olive]

Blame GNOME for making Wayland suck

GNOME singlehandedly stops the devs from implementing most useful things

They're basically dictating Wayland development

And GNOME is famously like Apple in that they like things locked down

[olive sable]

what? where are all the gnome's coming from and why are they software devs?

[nocturne olive]

GNOME is the worst Linux desktop environment

[sage crag]

nowhere in this case, hdr is a protocol that has nothing to do with gnome

[uneven pulsar]

It is funded by Microsoft, BTW

[tender river]

om

[sage crag]

so is the linux kernel

[tender river]

the fucking linux conspiracy theorists

[uneven pulsar]

I never said it was a bad thingy

[sage crag]

[uneven pulsar]

:3

[glass flower]

do you mean funded or supported? i feel like there is a conceptial difference in the two words.

[olive sable]

is it even worth it to turn on hdr if my screenshots will look ike shit?
i dont even notice a difference besides my eyes getting burned out at full white hdr

[tender river]

its all redhat!!!! ibm!!! microsoft sabotaging the linux desktop!!!!!! linux would've been perfect if not for them

[olive sable]

the gnome's are escaping!!!!

[sage crag]

nope!

[olive sable]

https://tenor.com/view/cartman-gnome-hit-south-park-gif-17376514

[sage crag]

the only compositors that do colour well on wayland are gnome's and kde's

[tender river]

i installed gnome on my mom's pc and she loved it

please understand that not everyone has the same workflow

[sage crag]

those will probably also not work well on nvidia gpu

[olive sable]

thats fair

i tried a fix i found on reddit. wtf is this???

how did this even happen?

ye im giving up

[nocturne olive]

Just don't HDR until maybe 10 years in the future Wayland devs get their garbage together, and actually implement HDR properly

[opaque sigil]

i don't think this is something you can fix without touching hyprland itself

this isn't a wayland problem this is hyprland, my hdr screenshots behave like they should (blown out cause no tone mapping)

[olive sable]

i need a toggle

could i remap the proile button on my keyboard im not using cuz i hate icue, to a togle for turning of hdr???

lemme vibe code that

[opaque sigil]

if the system detects the keypress then i don't see why you wouldn't be able to

unless corsair does some cursed stuff with their keyboard fw which wouldn't surprise me

[midnight sigil]

just use a debugger

you cannot escape cpp even if you use rust

yea ofc rust does more validations and stuff than c++

it's a successor anyways

[olive sable]

none of these seemt o be responsible for the profile button on my keyboard

maybe i need drivers for it?

[uneven pulsar]

today i had a head ache

trying to learn python

i feel like

[left solstice]

Is it possible to make a mp3 file that is 4KB or smaller or is the header and other metadata too big for storing any meaningful audio data?

[uneven pulsar]

enpty

so

[opaque sigil]

the header is 32 bits apparently

[fierce nova]

I think they are bytes, not KB

headers I mean

[left solstice]

I have a very old school keypad phone similar to Nokia has only 4KB of free space which is why I'm curius if it's even possible

[opaque sigil]

this seems pretty smol to me

[fierce nova]

yea, ID3v2 is the biggest coz it can contain like name of the song, name of the album and etc

rest is smol

[olive sable]

wtf is key finance supposed to do??
like it opens up your bank app?

[left solstice]

Seems like it's possible then but I couldn't get a mp3 file smaller than 10-18KB when I tried to make it really short and used the worst bitrate/sample rate I could find so idk how

[rigid snow]

probably some stocks tracker website

[olive sable]

why wou.ld they make that part of this keyboard???

[rigid snow]

like yahoo finance or something

[rigid snow]

money

[olive sable]

corsair smh

its not even mapped to anything, its jsut there in either some driver or in the keyboard itself

[midnight sigil]

4kb? obviously yes, just make it 0.25kbps without caring bit loss

[opaque sigil]

AL Checkbook/Finance
do with this what you will

for KEY_FINANCE

[rigid snow]

it's for those late 00s-early 10s keyboards with a myriad of useless buttons at the top probably

[midnight sigil]

and get the bandwidth down

much more

[fierce nova]

how did you created it? With some soft or with encoding bytes in your code? Coz I think any modern soft really doesn't care about limiting size and can don't care much about it

[midnight sigil]

[opaque sigil]

yeah it's for launching a finance app it seems because why not i guess

[olive sable]

when the scammer take over your pc and presses the finance button

[fierce nova]

bro it's sooo old school. Crypto mining is the thing nowadays

[olive sable]

nah with cryptominin you'd maybe a get a single euro if i ran my pc for days

my bank account on the other hand is magnitudes larger

[fierce nova]

sure, but imagine 10000 of your pcs

[midnight sigil]

does nokia support opus

[olive sable]

ok but imagine 10000 bank accounts

[fierce nova]

tell me more about your bank account

[olive sable]

it has money inside it

[sage crag]

3 euro

[fierce nova]

huge

[midnight sigil]

3000 in debt

[olive sable]

not to flex, but i had a lot of money inside it, until i spent half of it 2 days ago

[sage crag]

you are arguing between crypto miners and bank account stealers, but like

[rigid snow]

something like this but with a $ button

[fierce nova]

went to buy some food?

[sage crag]

they can just do both

[midnight sigil]

and even more

[olive sable]

i bought an AVR

[fierce nova]

no way, doing DOUBLE crime at the same time?!

[rigid snow]

me when the malware does all possible malicious things

[fierce nova]

it's illegal

[sage crag]

the mal in malware stands for uhhh

[olive sable]

i mean if you do 1 illegal thing without getting caught, they might as well do a lot of them

[sage crag]

https://cdn.discordapp.com/attachments/1059569601144442911/1375623285894807563/IMG_6849.gif

[midnight sigil]

they will eventually fight for controls over each other

[sage crag]

shiro cube information stealer

[midnight sigil]

surely

[olive sable]

if its for shiro id gladly pay the malware tax

[rigid snow]

malnutrition

[fierce nova]

ok, today was an interesting day - like 5-6 hours of meetings, zero coding. Sometimes corpo coding is really something

[sage crag]

what about konii bwaa orb

https://cdn.discordapp.com/attachments/1300807614032121950/1409945130810609695/3dgifmaker99373.gif

[midnight sigil]

oh no now it's stucked in my head once I knew it exist

infohazard

[olive sable]

depends, how much tax you asking?

[midnight sigil]

[sage crag]

ok now tab out and see shiro square

[sage crag]

all

[olive sable]

oh no

[sage crag]

[olive sable]

im supposed to be buying a car with that eventually

[midnight sigil]

I'm not a TV how do I tab out

[sage crag]

unfocus the window

[rigid snow]

they named a square after shiro send coordinates

[olive sable]

https://tenor.com/view/cat-google-maps-italy-gif-21371827

the resemblence to shiro is uncanny

[sage crag]

https://tenor.com/view/magnapinna-squid-creepy-freaky-ocean-gif-16224216681422707491

[olive sable]

ewwwww

what is THAT

[sage crag]

magnapinna

(big fin squid]

[olive sable]

im not a fan

[sage crag]

i am

[olive sable]

well good for you

[sage crag]

https://tenor.com/view/magnapinna-magnapinna-squid-squid-gif-25137428

[olive sable]

nah

thats a water spider right there

[sage crag]

squid

[olive sable]

squidwart is not supposed to be 15m long

[sage crag]

[tender river]

[sage crag]

[tender river]

washing glasses

awawawa

[uneven pulsar]

[tender river]

[sage crag]

[tender river]

[sage crag]

merge sort

i was watching a short comparing quicksort and mergeshort visually yesterday