programming

[real sierra]

its even stdlib

just like malloc

[obsidian mantle]

oh i can just #define int my_int and access my_int locally its all useless

[real sierra]

ok so question

C submissions will be compiled using gcc -Wall -std=c23 -O3 -o bot_exec bot_source.c -L. -lchess -lm (not looking for suggestions)

[opaque sigil]

i wonder if it's because of internal state that needs to be kept

[real sierra]

for C++ submissions is it sufficient to just replace gcc with g++

or are the options different

[opaque sigil]

and weird platform specific implementations

though i guess the 2nd part doesn't really apply since with malloc you have different ways of getting memory pages too

[rigid snow]

gotta see all the warnings

[opaque sigil]

-Wall -Wextra -Wpedantic or bust

if it fails to compile you get disqualified

[clear sedge]

afaik they're pretty much the same but you have to change out -std=c23 for -std=c++23

[rigid snow]

dq if it produces any warning with these flags

[real sierra]

oh right oops

i can probably leave that off

(though it would be funny to see)

[opaque sigil]

-std=gnu++2c

[clear sedge]

-std=rust

[real sierra]

rust++

[clear sedge]

https://tenor.com/view/dog-sad-why-why-are-we-still-here-just-to-suffer-gif-17485467

my c++ ain't c++ing

heavens forbid i try to return a class that derives from another class, why of course it should naturally just try to copy the base class by default! i think???

[real sierra]

your first mistake was using c++

[clear sedge]

why is the base class behavior taking precedent over the overriden one

someone kill me please

[real sierra]

is it actually an override?

[clear sedge]

it is

[real sierra]

double check the signature

[clear sedge]

i even slapped a big ole override onto it

[real sierra]

no idea then

[clear sedge]

that's what i'm sayin

gdb isn't helping :(

[tender river]

maybe you arent using dynamic_cast somewhere?

[real sierra]

1032 tokens

just need to find 8 more

[clear sedge]

i saw nothing

[real sierra]

true

i can no longer afford to have my increment and decrement on separate lines

must move into the conditions

1024

[clear sedge]

i figured it out oh my lord

unique pointers to the rescue

[real sierra]

uh oh

broke my bot

[opaque sigil]

[clear sedge]

undefined reference to typeinfo fuck my life

[opaque sigil]

Welp that didn't last long

[clear sedge]

i figured it out again :D

[real sierra]

something is wrong with the tree reroot fun

after the first turn it starts solving for the wrong side's pieces

[sage crag]

phrr

[real sierra]

fren

[sage crag]

:happie:

[opaque sigil]

Hii konii

[tender river]

who

https://imgur.com/a/NJHvyBH

[sage crag]

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

ye me

[tender river]

[real sierra]

i cant figure out whats going on here

[sage crag]

mambo, mambo, omatsuri mambo
tabu tabu yura yura, omatsuri mambo

[real sierra]

im

so confused

[sage crag]

[real sierra]

its just spitting out garbage

is this a library issue

like

none of these make sense

at all

these are not only moves for the wrong player

but none of them are legal

wtf happened

[sage crag]

[real sierra]

im super concerned

might be valgrind time

[sage crag]

https://www.youtube.com/shorts/ygYI79_c-9Q
SLOP

[real sierra]

antikythera

ok it was a problem with my code not the api

[rigid snow]

alein technolodg comprute bottom ocean anceint.

[real sierra]

code trying to checkmate itself again

[sage crag]

ye funny, the mechanism itself is real though

ye ye https://en.wikipedia.org/wiki/Antikythera_mechanism

[rigid snow]

i am not saying it's not

is ye you or yes or both

[sage crag]

[rigid snow]

or a controversial public figure

[clear sedge]

i physically pogchamped

string parsing is theoretically complete for my json parser

!!!

god bless std::wstring_convert
no idea what they were smoking with std::codecvt tho

[opaque sigil]

I'm lazy I just use simdjson

[clear sedge]

see i'm lazy in a different direction that leads me to doing more work

[real sierra]

bot made illegal move

move finder problem? or an edge case when my bot is low time

no way of knowing

[sick owl]

Okay, so Magistral Small 2509 is cracked in my testing

I believe the benchmarks indicating it beats Medium 1.1

It also uses very few tokens for simpler queries compared to Qwen 30BA3B

[real sierra]

im so lost

so the bot explores this line in a position i gave it

info time 660 nodes 9729 info pv c2c4 d3c2 d2b1

in this position, c2c4 is mate

?

to confirm this, i tried adjusting the test input to make the move c2c4 before giving the board to the bot

and sure enough the bot says there's 0 legal moves

sooo

..?

[rigid snow]

bun ffi does not fucking support structs

[sage crag]

!felloff

[stiff mica]

[real sierra]

!cookie

[rigid snow]

[nocturne olive]

Funny numbers
I have no clue what that's meant to mean

[tender river]

@real sierra remember the functions i sent you?

they should probably be added to the general api

[sick owl]

It means its remarkably token efficient for a reasoning model

[tender river]

even if its only used for bindings

[real sierra]

which

[tender river]

ones that take values instead of structs

[nocturne olive]

Silly silliness
I assume it's like really big or something or just useless for me

[rigid snow]

it should probably just be a node binding in the first place

[sick owl]

24B parameters :Smadge:

[real sierra]

do you mean

passing a board by value

[sage crag]

vani fix your bot

/tldr

[real sierra]

or something

[stiff mica]

🍪 Om nom nom
You've given me 33 cookies! | I've received 245356 cookies total!

[nocturne olive]

Suspiciously 3090 sized model
Not like I have a tool to use it

[tender river]

Move

[rigid snow]

boards are passed as pointers which is fine

[real sierra]

i dont remember the functions you mean

[sick owl]

Mistral specifically targets 24GB cards for their releases these days

Something I very much appreciate them for

[nocturne olive]

Very silly

Well I still don't have enough need for LLMs to warrant programming an inference interface

[sage crag]

126k members means each member has given nuro an average of 2 cookies

[real sierra]

[sick owl]

I mostly just use Mistrals new stuff for Twitch chat translation personally

Offline and it beats Google Translate/DeepL in quality

[rigid snow]

i still don't have enough need for any llms let alone any locally runnable ones. i put $60 on openrouter on a whim half a year ago and i have only spent $6 since then

[nocturne olive]

I don't need Twitch chat translations
Or Twitch chat

[sick owl]

Poor twitch chat

[rigid snow]

and i use actual sota models not the bad local sized ones

[nocturne olive]

All they do is for (i in 1..Long.maxValue()) print(getRandomEmote())

[sick owl]

I mean local sized models aren't bad these days unless you're using them for programming

[rigid snow]

yeah sure but i prefer to talk to humans and use llms for technical stuff

the first part they may do well

[sick owl]

See for me its mainly quick info retrieval and translation like I said earlier, I do use cloud for programming

But for simple questions, translations and web search I'd rather have the privacy of a local model and save some money in the process

[nocturne olive]

Cloud models are too pricy for me, I simply just don't vibecode

[rigid snow]

web search
<insert search engine> is right there

[nocturne olive]

Eh, LLMs can usually parse the data into a form where you get a more direct answer to specifically your problem

[sick owl]

I don't vibe code but I do use cloud models as a second opinion on stuff and to generate simple functions quickly to cut down on dependencies

[sick owl]

I've been giving package managers the side eye of late

[nocturne olive]

Silly
I personally don't use LLMs in my code editors

[sick owl]

👆 What Super said

If I just want to grab the release date of something quickly for example or get a word definition its much nicer

[sage crag]

let rand = rand::rng();
loop {
  let emote = rng.random::<Emote>();
  chat.iter().for_each(|chatter| chatter.emit(emote));
}

rand should be mutable i have failed

[sick owl]

I have free Github Copilot as a student which helps, I probably wouldn't pay for the privilege

[nocturne olive]

Rust be silly

[rigid snow]

if openrouter ever brings code completion apis those $60 will evaporate fast
somehow they still don't have code completion but have audio input and image generation

i do miss tab complete from copilot

[tender river]

"average neurocord member has given neurobot 2 cookies" factoid actualy just statistical error. average neurocord member has given neurobot 0 cookies. @cloud horizon who gave neurobot 126053 cookies is an outlier and should not have been counted

[nocturne olive]

If I need an LLM I just use the free ChatGPT website

[sage crag]

@cloud horizon

[sick owl]

If I want a fancy one outside vscode I just use lmarena

[sage crag]

[sick owl]

I don't mind that my stoopid questions might make their way into training data if it means free access to GPT-5-High for the rare query that really needs it

[cloud horizon]

[sage crag]

ye give nuro a cookie right now

[cloud horizon]

done

[sage crag]

WRONG

ye need /cookie

[nocturne olive]

Silly
I've never used another model website than ChatGPT myself
If I wanted a different LLM I would program something to run locally, but I can't be bothered to do that when I don't need it

[sage crag]

126270 member

126054 cookie

[tender river]

[cloud horizon]

!felloff

[stiff mica]

[cloud horizon]

canvas first

[sage crag]

average user have give nuro 0.94 cookie

[stiff mica]

🍪 Om nom nom
You've given me 6 cookies! | I've received 245358 cookies total!

[rigid snow]

[ivory plinth]

🍪

[stiff mica]

🍪 Om nom nom
You've given me 4 cookies! | I've received 245359 cookies total!

[rigid snow]

i am less generous

my cookie worth more

[tender river]

cookie fungible

[real sierra]

debugging hell

[hollow spruce]

Yo

[real sierra]

my janky code is breaking spectacularly

and i have no idea why

everything ive tried is working right

[hollow spruce]

Show me ?

[hollow spruce]

Give me list of problem u have
{
i give the bot a chess position where there is a one-move checkmate on the board
however, i see that when it explores this position, it finds more moves after the checkmate
but if i make this move on the chess board myself and give it this board instead, it correctly states there are zero legal moves
}

[real sierra]

i give the bot a chess position where there is a one-move checkmate on the board

however, i see that when it explores this position, it finds more moves after the checkmate

which is not possible

but if i make this move on the chess board myself and give it this board instead, it correctly states there are zero legal moves

that's my issue

[tender river]

maybe the valid move cache is being improperly reused?

[real sierra]

it's zero'd on making a new board or cloning one

shouldn't be shared

[hollow spruce]

Maybe TRAINING
NAI TRAINING IS GOOD
MAYBE SYSTEM ALSO
MEM (HMM)
CACHE (THAT BE )
INSTRUCTORS (!)

[tender river]

the buffer is reused somewhere so thats just a guess

[real sierra]

was a logical place to start

[hollow spruce]

( is_checkmate() / is_game_over() and generate_moves()):

if (is_checkmate(board) || is_game_over(board)) {
return 0; // no moves after mate
}

int n = generate_moves(board, moves);
return n;

[hollow spruce]

This logic will works it is generation problem

LITTERY Convert cannot be run

[real sierra]

ah.

wait

[sage crag]

@real sierra i recently learn something

[hollow spruce]

What !

[real sierra]

[sage crag]

for (;;) { continue; }
while (1) { continue; }
loop: { goto loop; }

for bytecode interpreter, last one save two token

[real sierra]

goto carry

tho

actually i see

[rigid snow]

what is this forbidden knowledge blessing my neurons rn

[real sierra]

// Makes a new, blank board. Caller responsible for freeing.
static Board *create_board() {
    Board *board = (Board *)malloc(sizeof(Board));
    memset(board, 0, sizeof(Board));
    clear_board(board);
    board->can_castle_bk = false;
    board->can_castle_bq = false;
    board->can_castle_wk = false;
    board->can_castle_wq = false;
    board->fullmoves = 1;
    board->halfmoves = 0;
    board->last_board = NULL;
    board->en_passant_target = 0;
    board->whiteToMove = true;
    board->refcount = 1;
    return board;
}

// Creates a shallow copy of the given board
static Board *clone_board(Board * board) {
    Board *new_board = (Board *)malloc(sizeof(Board));
    memcpy(new_board, board, sizeof(Board));
    new_board->bb_black_moves = NULL;
    new_board->bb_white_moves = NULL;
    //new_board->last_board = NULL;
    if (new_board->last_board != NULL) {
        new_board->last_board->refcount++;
    }
    return new_board;
}

think i found part of prolbem

[sage crag]

because switch will fallthrough if you dont continue or goto loop out of it

[real sierra]

the assumption made was that refcount is initialized in create_board so clone_board no need worry

but then clone_board overwrites refcount with the value in board

during the memcpy

so it needs to explicitly be set to 1

[sage crag]

loop: switch (*pc++) {
  /*exit*/ op(return 0)
  /*print*/ op(__builtin_printf("hello\n"))
}

ye bytecode interpreter actually so easy people think it complicated

[tender river]

labels can also be reused across functions

[sage crag]

computed goto would be fun but ye so many token

[hollow spruce]

U made own bugs ?
Y/N

[sage crag]

phrr

[stark needle]

[hollow spruce]

Yo @stark needle

[stark needle]

I had a weird dream where i played chess with shiro in vrchat

[sage crag]

wpedantic

[rigid snow]

double ping to make sure it goes through

[hollow spruce]

I hate most google cloud shell

[stark needle]

[rigid snow]

why are you awake

why am i awake

[real sierra]

wtf

[rigid snow]

eu chatters go sleep

[sage crag]

nompe

[stark needle]

Sam in the dream was there and laughing at me cause i suck at chess

[real sierra]

im even worse at chess

my bot is already better than me

[tender river]

eu chatters wake up

[hollow spruce]

Eu ?
Asia ✓

[stark needle]

What I'm gonna do is

[tender river]

eu ✅

[sage crag]

eu chatters eu chatters

[stark needle]

Eat a whole package of gummy bears

[sage crag]

no

[tender river]

[sage crag]

bad

eat two package of gummy bear

[stark needle]

It's an extra large one

[sage crag]

dont care

[hollow spruce]

Bro heat waves
In raining season

[rigid snow]

diabetes

[sage crag]

yay

[stark needle]

My doctor said my liver values are fucked cause of schizo meds

[rigid snow]

are you actually schizophrenic or is that a figure of speech

[stark needle]

I am unfortunately

[rigid snow]

oh

[stray dragon]

want but the big bags are no longer for sale

:(

[stark needle]

It's not fully diagnosed yet cause it needs 6 months but my doctor told me it's obvious and unlikely to improve

[rigid snow]

[stray dragon]

rip

[stark needle]

I just take it in a fun way cause what else am i supposed to do lol

[rigid snow]

good, embrace it

[sage crag]

@real sierra is your system little endian or big endian

this matters for my vm

[real sierra]

[sage crag]

im packing the bytecode into 64 bit literals to save space

but this means that endianness will affect the order the bytecode is interpreted

[real sierra]

will the vm be LE if my main system is LE

[sage crag]

[real sierra]

cuz i have no idea how to check on the vm

[rigid snow]

moderate endian

[real sierra]

actually

i have no clue how to check on my main system either

both systems are x86 i believe so

should be LE

on both

[sage crag]

vm 700 mb/s "hello!" throughput

pure C 842 mb/s throughput

83% of C performance

[real sierra]

probably faster than python

[sage crag]

yes indubitably

every vm ive written has been faster than python

[real sierra]

in this small of a footprint tho

super impressivbe

[sage crag]

not really, most of the work is still being done by c

the art of making a performant vm is to delegate more of the work to the host machine

bytecode is 0x0201, this is the whole vm

its not really a big deal to get to 83% of c printf performance when all the vm is doing is handling a loop and a native printf call

[real sierra]

wait

how does that switch work

[sage crag]

ye magic

[real sierra]

[sage crag]

#define op(...) case __COUNTER__: __VA_ARGS__; goto loop;

[real sierra]

[sage crag]

ye downside it mean that the declaration order is important for the bytecode interpreter

if i swap instruction by mistake whole program break

but save so many token

[real sierra]

ive identified the issue

somehow, i'm not undoing all the board moves im making

idk what i changed to cause that

its not the simulate function

somewhere else...

the spacing on something like tree ->moves[i] doesnt matter right

[sage crag]

whitespace is a suggestion in c

[tender river]

for (int i = 10; i --> 0;) printf("%d\n", i);

[real sierra]

my brain is so fried

idk where the issue is

it all loks correct

it's only sometimes that it changes too?

[sage crag]

what

[tender river]

[sage crag]

ohhh

you trick

[tender river]

[sage crag]

i-- >0

godbolt is complaining but my local compiler isnt

[stray dragon]

the i --> 0 part is kinda evil

also unnecessary

[sage crag]

[sage crag]

no save token

[stray dragon]

--i

[sage crag]

no no its different

[stray dragon]

?

why have >0

if anything over 0 gets treated as true

[tender river]

can change to --> 5

[stray dragon]

and 0 is treated as false

[sage crag]

i--:
9
8
7
6
5
4
3
2
1
0

[stray dragon]

i prefer --i tbh

[sage crag]

--i:
9
8
7
6
5
4
3
2
1

[stray dragon]

i swear it wasn't like that though

hmm

[sage crag]

noper

[tender river]

++i preincrement i++ postincrement

[stray dragon]

wait

AH

[sage crag]

--x evaluate to x-1

x-- evaluate to x

[stray dragon]

i see why i was thinking --i instead of i--

it was because of the oppositev version i used

[sage crag]

@real sierra what do, my program is cursed and compile only on some compiler

[stray dragon]

for(i=-1;++i<64;)

[sage crag]

[stray dragon]

in that case the ++i makes it go from 0 to 63 instead of 0 to 64

[sage crag]

i compiling with -std=c23 -Wall -O3

on both

both clang

but godbolt doing funny

[real sierra]

OH

i think i get it

[stray dragon]

(shiro did not get it)

[sage crag]

[real sierra]

while_ moves-- bclose
            chess_undo_move no_args_after_board
            __builtin_printf("undid move, need to undo %d more\n", moves);
        }

ive been using structures like this

no worky right

oh sorry let me unmacro all that

while (moves--) {
            chess_undo_move(board);
            printf("undid move, need to undo %d more\n", moves);
        }

but

this no worky

because even if moves is already 0

it has to evaluate the condition

which will drop moves to -1

[fast pagoda]

https://youtu.be/cFtymODJEjs?si=HDIt3k3TwWPTLDTc

[real sierra]

what a terrible title

was this video made just to spite me

W channel tho

[fast pagoda]

It was made to ruin your life

[real sierra]

anyway

i bet if i move the move-- out of the condition

problem disappear

tragic tho because now my token count go up by 1

[fast pagoda]

Regression progression

[real sierra]

actually it go up by 2

[sage crag]

try --move

[real sierra]

the semicolon

[sage crag]

[real sierra]

if move 0 entering loop

it still decrement move

which bad

because condition always evaluated at least once

[tender river]

theres do while or goto but maybe more tokens

[real sierra]

it fine

just gotta find token savings elsewhere

[tender river]

[sage crag]

while (move-- > 0)

waitge

lie

ye

while(move & --move)

[tender river]

while (moves--)
            chess_undo_move(board),
            printf("undid move, need to undo %d more\n", moves);

[real sierra]

wholesome

but i need moves = 0 after the loop

[tender river]

still less tokens

even if you move --moves back to loop body

[sage crag]

use do while

int main() {
    int moves = 0;
    do __builtin_printf("%u\n", moves); while (moves--);
}

print 0

int main() {
    int moves = 10;
    do __builtin_printf("%u\n", moves); while (moves--);
}

print
10
9
8
7
6
5
4
3
2
1
0

[real sierra]

good news

[tender river]

+2 not worth

[real sierra]

bot fix

bad news

1029 token

5 token where

[delicate dust]

I'm curious, what are you doing shiro?

[real sierra]

tldr im planning on hosting a chess bot contest here

[sage crag]

pain suffering

chess is side effect not main goal

[real sierra]

entries have a source code limit of 1024 tokens

ive put my bot together to help test and am trying to make it fit in the limit

it like fitting elephant in suitcase tho

[sage crag]

why is there a sticker on my face

[delicate dust]

so, like code golf?

[real sierra]

yeah

[sage crag]

code golf ye

token golf

[real sierra]

but tokens instead of source code len means you dont have to minify it to an illegible mess at least

[delicate dust]

got it, pretty interesting

[real sierra]

for example look at this beautiful function of mine

[tender river]

best_score_eq_best_score_gt_score_best_score_or_score_sc

[sage crag]

> counting tokens
> means you dont have to minify
> source code should be readable
>

[real sierra]

its readable

[real sierra]

#define best_score_eq_best_score_gt_score_best_score_or_score_sc best_score = best_score_gt_score best_score : score;

[delicate dust]

This explains the use of macros

[real sierra]

gets the go and bo from child and adds them

[sage crag]

[tender river]

thats not equality its assignment

[sage crag]

[real sierra]

when i read it aloud tho, i read it as equals

[sage crag]

a = b

a == b

[sage crag]

no you have to write as best_score_ass_best_score_...

[real sierra]

im a math major they're the same thhing to me

[tender river]

fake math major

[real sierra]

i wonder how long this is with the macros expanded

[sage crag]

ye real math major is actually programmer

[sage crag]

gcc -E src/main.c

[tender river]

[real sierra]

deletes my source code

ooo konii right tho

lets see

shiro@chess-box:/stg/backend/api/c$ cat test.c | wc -c
12011
shiro@chess-box:/stg/backend/api/c$ gcc -E test.c | wc -c
47791

[delicate dust]

I haven't touched c in so long, that seeing it makes me feel old

[tender river]

that includes headers

[real sierra]

[tender river]

remove and recount

[real sierra]

ill comment out the headers

shiro@chess-box:/stg/backend/api/c$ gcc -E test.c | wc -c
7845

smaller..?

[tender river]

ye wc

[sage crag]

ye wc

[real sierra]

[sage crag]

sym_super_symbol -> { } {}

[real sierra]

write to file and token count

[opaque sigil]

almost as if not including headers means those aren't preprocessed

[tender river]

| java -jar toknt.jar /dev/stdin

[opaque sigil]

[real sierra]

shiro@chess-box:/stg/backend/api/c$ java -jar /stg/backend/token_count/toknt-assembly-0.1.0-SNAPSHOT.jar ./test.preprocessed 
?

needs .c?

[tender river]

ye

how is it supposed to know the language

[real sierra]

shiro@chess-box:/stg/backend/api/c$ java -jar /stg/backend/token_count/toknt-assembly-0.1.0-SNAPSHOT.jar ./test.preprocessed.c
1587

not bad

i think thats a decent token compression ratio

can i compile the .so with -O3 ????

i havent been, i wonder how much performance it's cost me

@sage crag

Please note the warning under -fgcse about invoking -O2 on programs that use computed gotos.

you were sure you werent using those right

[opaque sigil]

[opaque sigil]

yes of course

[sage crag]

not in this vm

cost too many token

not worth performance gain unless i have token to spare later

computed goto cool technique

void *labels[] = {
  &&erms, &&ermies
}
goto *labels[*pc++];
erms:
  return 0;
ermies:
  printf("hello!\n");
  goto *labels[*pc++];

just cost lot of token

fast for bigger instruction sets but can be slower for tiny instruction sets since goto

computed goto have good performance characteristic generally especially for vm with simple instructions

since branch prediction

ye

[opaque sigil]

damn cpus with their annoying crutches

[sage crag]

nya nya nya

computed goto faster than function ptr LUT too

cant remember why

probably local stack + inlining

[tender river]

#define D x(erms, return 0;) x(ermies, printf("hello!\n"); goto *lables[*pc++];)
#define x(a, b) &&a,
void *labels[] = { D erms };
goto *labels[*pc++];
#define x(a, b) a: b
D

more tokens but less with more opcodes

[real sierra]

how do i get 5 more tokens

[sage crag]

not use seem painful

koni got more simple

less token

switch

invisible

[real sierra]

tokens

where are teh tokens

i cant find anywhere to save even one token

well thats a lie there's still stdbool

but thats permanent

[stray dragon]

why stdbool anyways

[tender river]

ye use char same thing

[real sierra]

stdbool was there when i needed it most

we have a bond

its ok i found a new macro

just had to rename some variables

[stray dragon]

anyways time to figure out how chayleaf's token counter works so i can set it up

[sage crag]

true false

[real sierra]

saved 3 tokens

[sage crag]

[tender river]

https://github.com/chayleaf/toknt
go to latest commit's ci job and it should have a jar artiffact

[real sierra]

just need 2 more

[tender river]

then java -jar toknt.jar file.c

[stray dragon]

...but why

[tender river]

[stray dragon]

1, 0

[sage crag]

101 also truthy

1 << 16 also truthy

bool give you true false

[real sierra]

i have no actual reason

i just like stdbool

[sage crag]

include stdbool in chessapi.h

so you dont have to include it in the main program

[real sierra]

WAI

YOU'RE RIGHT

[sage crag]

ye ofc

koni alway right

i already did that after all

[real sierra]

.

chayleaf...

why does removing #include <stdbool.h> save me 7 tokens

[tender river]

# include < stdbool . h >

[real sierra]

[sage crag]

# include < stdbool . h >

7

[real sierra]

i thought it was 2 or 3 tokens this whole time

includes are expensive

[sage crag]

you can also remove stdint this way

[real sierra]

i dont have stdint

just stdlib and chessapi

[sage crag]

well its in your chessapi.h anyway

[real sierra]

[tender river]

doesnt chessapi include stdbool

[real sierra]

good thinking konii

[sage crag]

ye

[real sierra]

[sage crag]

koni smart

[real sierra]

thats what konii pointed out

so i can remove stdbool

and still get my bools

ty chessapi

[sage crag]

koni already use

[real sierra]

1019 tokens

[sage crag]

btw shimo

[stray dragon]

not seeing a jar artifact anywhere in here

[sage crag]

uint8_t *pc = (uint64_t[]){0x201};

can you see if this compile

on your machine

it compile on my machine but not on godbolt

[stray dragon]

i go to latest commit i see files changed

no jar file

i go through entire repo

no jar file

[sage crag]

ci job

[real sierra]

i try

[stray dragon]

i would not have found it without that screenshot wow

[real sierra]

shiro@chess-box:/stg/backend/api/c$ gcc -Wall -std=c23 -O3 -o bruh bruh.c -L. -lchess -lm
bruh.c:3:15: error: initialization of ‘uint8_t *’ {aka ‘unsigned char *’} from incompatible pointer type ‘uint64_t *’ {aka ‘long unsigned int *’} [-Wincompatible-pointer-types]
    3 | uint8_t *pc = (uint64_t[]){0x201};
      |               ^

[sage crag]

why'mst

[real sierra]

different size?

[sage crag]

noe not problem me

[real sierra]

u use same gcc opts?

[sage crag]

main.c:38:10: warning: incompatible pointer types initializing 'uint8_t *' (aka 'unsigned char *') with an expression of type 'uint64_t[1]' (aka 'unsigned long[1]') [-Wincompatible-pointer-types]
   38 | uint8_t *pc = (uint64_t[]){0x0201};
      |          ^

[real sierra]

actually i have -Wall on

is that issue

[sage crag]

buildPhase = ''
tcount main.c
mkdir -p $out/bin
cc main.c chessapi.c bitboard.c -o $out/bin/${name}
'';

env.CFLAGS = "-std=c23 -Wall -O3";

[tender river]

[real sierra]

you may be answering this question a lot if there aren't instructions anywhere

[tender river]

[sage crag]

git ci

[real sierra]

or actually i guess im probably answering this question a lot

[sage crag]

basic

[real sierra]

idk i just build jar from source

[sage crag]

[sage crag]

me use clang

[stray dragon]

smh i don't work a job for this stuff

[tender river]

neither do i

[stray dragon]

my condolences

[sage crag]

work forever until your bones are dust

[stray dragon]

oof ouch owie my bones

[sage crag]

dus t

[real sierra]

why bot unironically bongclouding

on both white and black now

[tender river]

good move

[sage crag]

dumbot

no token left for opening book

[stray dragon]

NOOOOOOOOOOOOOOOOOO

i am crumbling away

[sage crag]

no you are t

[real sierra]

my brother

it is time to shed tokens

[sage crag]

t send me your source code

i will reduce token

[real sierra]

im so confuse

my bot still trying to lose as white

but won anyway

?

i think the checkmate finder is working but the rest isnt

[sage crag]

[tender river]

inting on purpose to reduce the opponent's guard and find a checkmate in their opening

[real sierra]

its a really weird checkmate too

[stray dragon]

behold my creation

[sage crag]

0 token

[stray dragon]

shh

i've removed everything unnecessary

can you make it negative pretty please

[sage crag]

ye just add 1.844674407370955e19 token

[stray dragon]

true

[sage crag]

nopy

i lied

that 1 << 64

[stray dragon]

oh

ok

[sage crag]

wormy

[tender river]

ye that nopy

[sage crag]

🐛

nopy worm

worm with leg

[stray dragon]

evil laugh for undisclosed reason

i wonder why it suddenly increased the tokens though

[tender river]

nopy 🐛

[stray dragon]

nopy 🐛

[real sierra]

are you including anything

[sage crag]

ye remove includes

[tender river]

🎷🐛

[sage crag]

woah

nopyphone

@real sierra

your chessapi.h doesnt have #pragma once

[tender river]

https://cdn.discordapp.com/emojis/1276149117717970964.gif?quality=lossless&name=nopy&size=64

[real sierra]

is problem?

[sage crag]

ye ye problem

[real sierra]

how would it be imported twice

[sage crag]

i invite you to

[stray dragon]

include paths are currently not set up properly because i have still never run this lol

[real sierra]

i c

[sage crag]

two file

two include

suddenly do twice

[real sierra]

pragma once non-standard

[sage crag]

nopie

[tender river]

clang gcc

[stray dragon]

YIPPEE multiline macros don't cost extra

[tender river]

msvc

[real sierra]

In the C and C++ programming languages, #pragma once is a non-standard but widely supported

[stray dragon]

i can make nice looking code

[real sierra]

--wikipedia

[tender river]

you can use ifdef but honestly just do pragma once

[sage crag]

can

#ifndef ERM
#define ERM erm

// library here

#endif // ifndef ERM

[real sierra]

[sage crag]

but

#pragma once

ye so easier

[real sierra]

i do

[sage crag]

nopy

https://cdn.discordapp.com/emojis/1276149117717970964.gif?quality=lossless&name=nopy&size=64

[sage crag]

by way

is updated header file anywhere

[tender river]

#ifndef ERM
#define ERM

// library code

#endif // ifndef ERM

[sage crag]

noe ERM macro

ERM macro for erm

[sage crag]

because when i compile

actually it updated source file but

sadgi

[tender river]

shiro doesnt use git he just updates the files in the repo manually

[sage crag]

bred do this too

[sage crag]

why header file on it own

badge

use submodule

explode you

[real sierra]

i dont know what any of that means

[tender river]

bred uses git at least

[real sierra]

all roasting me

[sage crag]

shimo

[real sierra]

i should just take my header back

[sage crag]

too late i already save

[real sierra]

scammed

[sage crag]

[tender river]

[real sierra]

everyone just ask for code so can bulli me

[sage crag]

https://cdn.discordapp.com/emojis/1276149117717970964.gif?quality=lossless&name=nopy&size=64

forehead sticker

[stray dragon]

ok interesting so with all the defines and includes commented out the difference is just 3 tokens between tcount and toknt

hmm

[real sierra]

this is so frustrating

when bot play white, is cracked

ish

but when play black, sucks

[stray dragon]

maybe it's the evaluation function?

[sage crag]

chess eval function evaluates black positions as negative

flip sign when playing as black

[stray dragon]

^

[real sierra]

already do

        rv_assign __builtin_abs(score_gap) >= 4 ? score_gap < 0 : -1;

[sage crag]

then unflip

[real sierra]

,

[stray dragon]

oh so -1 is considered 1 token in tcount, while in toknt it's 2 tokens

so all negatives are probably a bit extra tokens now

also noticed that "" went from 2 tokens to 1 token

[real sierra]

#define n1 -1

[sage crag]

sym

[sage crag]

@tender river fix

[stray dragon]

why care

is fine

[sage crag]

is literal

easy fix

[stray dragon]

little tiny change

[tender river]

what does the standard say

[real sierra]

so are all strings 1 token

[sage crag]

the standard say konii opinion correct at all time

[real sierra]

[sage crag]

[real sierra]

should i stop using <> in imports and switch to ""

[tender river]

[stray dragon]

wait you can just do that?

[real sierra]

i think it optional which you use

[stray dragon]

damn ok

[real sierra]

just convention to use <> for builtin

[sage crag]

noe

[stray dragon]

thought <> was for standard imports and "" was for local imports

[sage crag]

cant use <> for local import

[stray dragon]

oooh interesting

[sage crag]

can use either for standard import

[tender river]

konii wrong

[sage crag]

no no

prefixed or something

[real sierra]

konii right actually

[stray dragon]

anyways found a macro that was actually adding 2 tokens

smh

[real sierra]

"" will still check standard directory, it just checks local first

so switching to "" is better

[sage crag]

dies ire

[tender river]

standard says less work for me

correct standard

[sage crag]

standard say less more work for you alway

[tender river]

[sage crag]

last minute

canbas

[tender river]

[real sierra]

made debug output for bot

can see entire game tree now

wonder if can make output json...

[tender river]

C23 has _BitInt

_BitInt(32) -> u32

or i32 rather

[opaque sigil]

i wonder how gcc implements this

since clang presumably just uses the existing types from llvm

[tender river]

c already had bitfields

so i dont think its that much more effort

i don't do this properly since i treat preprocessor tokens as regular tokens

but i dont care

i disregard the standard when it says i need to do more work

[real sierra]

[tender river]

header name preprocessing tokens are recognized only within
#include and #embed preprocessing directives, in __has_include and __has_embed expressions,
as well as in implementation-defined locations within #pragma directives. In such contexts, a
sequence of characters that could be either a header name or a string literal is recognized as the
former.

shiro i'm not implementing this

[opaque sigil]

Right

[real sierra]

implementation-defined locations

screw trhat

[opaque sigil]

My favourite part about #pragma is the Pragma macro

[maiden geyser]

350 people in https://discord.com/channels/574720535888396288/1067400152698077194 is crazy

[real sierra]

canvas

hey

can someone sanity check me

(0xff & 0x010303050900 << P)

what does this return for P from 1 to 6

im starting to think this is wrong

actually this is wrong in so many ways

wtf

[tender river]

0 0 0 0 0 0 you're welcome

[real sierra]

(0xf & 0x9953310 >> 4*P)

potentially improved

[real sierra]

explains a lot

[real sierra]

is this one different

[real sierra]

nice

[tender river]

neural

[real sierra]

things like 0x14125fa9bul are one token right

[sage crag]

vm

[sage crag]

should be

[tender river]

[real sierra]

whats it doing

[sage crag]

(uint8_t[]){0x00, 0x01, 0xED, 0xB7, 0x91, 0x01, 0x02, 0x02, 0x01};

[real sierra]

good for it

konii do u see the emote in the big vc status

[tender river]

[sage crag]

its actually doing

ldr $1 2382829
stm [2] $1

where $2 contains an address to an 8 byte section of memory

and what im doing here is printing out:
*pc, regs[1], *ptr (where $1 == regs[1], *ptr == [2])

3 instructions

yay

already at 192 token

[real sierra]

:D

it work

[sage crag]

[real sierra]

[sage crag]

havent optimised tokens at all though

[real sierra]

yea ur gonna have to compress those instructions big time

[sage crag]

also im encoding in bytes not 64 bit literals for now

[real sierra]

ya that'll already be an 8fold increase in code space

[tender river]

9/2 fold

[sage crag]

also also havent implemented the fancy bytecode format

ideally i would be using the top one for space but its so much effort

[real sierra]

if you're really careful with code gen, you could try to generate repeat sequences of instructions aligned to the 64-bit word size

by padding with no-ops or something

[sage crag]

codegen

[tender river]

no its 8fold ye

[real sierra]

then you could #define multiple 64-bit values at a time

[sage crag]

im writing it myself

[real sierra]

[sage crag]

#define leb128_r ({uint64_t o = 0, s = 0; uint8_t b; do b = *pc++, o |= (b & 0x7F) << s, s += 7; while(b & 0x80); o;})

[real sierra]

[sage crag]

will c automatically zero variables if i leave them undefined

[tender river]

only globals

[sage crag]

no it will not

drat

[real sierra]

i fixed bot again

now its wicked good as white

lets see how it plays as black

i find it amusing that if my bot runs out of time it just plays the opponent's last move

results inconclusive but it looks like it plays quite well as black now, idk if it can find mates tho

5head play holy

the rook is hanging but if the knight moves, it dies because the only tile it can escape to is guarded by bishop

[sage crag]

nerd play

[real sierra]

🐎 🐎 🐎 🐎 🐎 🐎 🐎 🐎 🐎

👑 ?!

[sage crag]

#define leb128_r ({uint64_t o = 0, s = -7, b; do o |= ((b = *pc++) & 0x7F) << (s += 7); while(b & 0x80); o;})

[tender river]

i disap pear

https://tenor.com/view/teto-kasane-teto-teto-plush-kasane-teto-plush-crush-it-gif-16166927453889012815

[real sierra]

why you doin that

im actually pretty sure i know why

but it dont make sense either way

[sage crag]

#define leb128_r ({uint64_t o, s = -7, b; while(b & 0x80) o |= ((b = *pc++) & 0x7F) << (s += 7); o;})

im genuinely sorry for this

this code is vile

and it works

b and o are uninitialised

is that safe

[real sierra]

no

i dont

i dont know how that works

[sage crag]

variable size integer

read it

from memory

that is macro to read it

[real sierra]

im scare

[maiden geyser]

there must be a worse way to do this

[stray dragon]

features GAINED

[real sierra]

LOW TOKEN

[wispy tundra]

This is a better place to ask.

[real sierra]

im not sure unfortunately, i imagine there's some kind of web api

@lament igloo know anything?

[lament igloo]

there was an endpoint /current

i would NOT want to know more cus cursed code and network requests

no wonder why the site couldn't handle that much ppl