programming

[real sierra]

oh does the new version of the C api work for you?

or are there still issues

[glass flower]

you pushed a new version?

[real sierra]

17 min ago yea

[glass flower]

[real sierra]

now everyone on windows gets illegal moves

but they do get moves

how fun

[glass flower]

[real sierra]

is that with rust or py?

[glass flower]

rust

[real sierra]

language-independent bug

is it a problem with the c library...

[tender river]

thats better arguably

[real sierra]

not better

it doesnt happen on linux

[glass flower]

a language independent but system depended bug

[real sierra]

my respect for the windows software environment is plummeting tonight

[opaque sigil]

Do you understand why wsl2 is so beloved now

[midnight sigil]

shibo rn is like a corpa pushing and announcing untested products to public

[real sierra]

it was tested

...on linux

[midnight sigil]

[real sierra]

and in my defense, 0 bugs have occurred on linux

[midnight sigil]

0 bugs have discovered :mhm:

[real sierra]

if you dont notice them, thats the same as them not being there

[midnight sigil]

:neuroTrue:

[opaque sigil]

Are you calling fork anywhere by any chance shiro

[real sierra]

nope

there's a thrd_create for making the UCI thread

thats it

[glass flower]

it also seems like there is no patterns to when the illegal move happens... so good luck figuring it out

[opaque sigil]

Hmm

[real sierra]

bugs that make no sense and only happen on windows

[glass flower]

are you initializing every variable you use? i think in windows it can be garbage but on linux it gets cleared. or the other way around idk

there was definitely some memory mismatch that can make it work on one system vs another

[real sierra]

i found one potential problem in the C api

[midnight sigil]

[real sierra]

-# ( @sage crag )

[real sierra]

did not fix

[glass flower]

[midnight sigil]

VM under 1024 tokens

would be crazy

[opaque sigil]

Have you tried manually clearing before free'ing

[real sierra]

the issue isn't uninitialized data

im almost certain that freeing the moves array that C returns is freeing the copies i made in python too

which is really confusing because i made sure to very explicitly make copies

[opaque sigil]

Hmm

How are you making the copies

[tender river]

problem in rust too

so no

[real sierra]

its kinda ugly but i dont think it should cause bugs

[tender river]

[real sierra]

gotta consider that it could be a shared flaw in both apis

    def __init__(self, copy_from: Optional[Move]):
        if copy_from is not None:
            # awkward copy ctor
            self._from = _BitBoard(BitBoard(copy_from._from))
            self._to = _BitBoard(BitBoard(copy_from._to))
            self._promotion = c_uint8(int(copy_from._promotion))
            self._capture = c_bool(bool(copy_from._capture))
            self._castle = c_bool(bool(copy_from._castle))

you can make comments on the lack of elegance but these should definitely be copies

[sage crag]

https://cdn.discordapp.com/attachments/1067638175478071307/1406337271241904158/ezgif-3f552200e37817.avif

[real sierra]

caibi interested in vm

[tender river]

i can assure you its fine in rust

[opaque sigil]

Hmm looks fine to me

[real sierra]

valgrind time

i see no leaks

i guess i can comment out the free command on python and see if the bug vanishes

didnt fix

[tender river]

shiro just test the c api instead

[real sierra]

so different bug

but the api works on linux

[tender river]

but its windows

[real sierra]

i dont wanna mess with cmake again

[tender river]

[glass flower]

thats why its the c api thats broken and not python

[real sierra]

are ints different size

[glass flower]

[tender river]

cmake can create a makefile or msvs project and then you just build using the same command every time

you dont have to cmake every time you change the source code

[opaque sigil]

I can't wait for someone to try and run this on macos and everything blows up again

[real sierra]

but its only set to build the dll right now

[real sierra]

similar enough to linux

[upbeat hollow]

🍉

[opaque sigil]

POSIX will carry

[real sierra]

in posix we trust

[tender river]

i win

[real sierra]

time to paste the entire source into an LLM and ask it to find the bug

[glass flower]

funnily enough i was gonna suggest that....

but i thought... nah no way

also chat gpt 5 free can't take the chessapi.c file

[real sierra]

in grok we trust

[sage crag]

@tender river i have made peak vm execution

[tender river]

[real sierra]

in grok the do not trust

[glass flower]

time for gemini

[sage crag]

problem

PRESERVE_NONE void run_goto2() {
    void **pc = (void*[]){&&op2, &&op2, &&op1};
    while(1) {
        goto *(*pc++);

        op1:
            printf("op1 (should terminate)\n");
            return;

        op2:
            printf("op2 (should continue)\n");
            continue;
    }
}

it requires program being in terms of addresses to handlers

which is definitely not portable

unless i do some kind of pretranslation step, but that somewhat defeats the point

[tender river]

tangentially related

when i was reverse engineering uh

something

[sage crag]

[tender river]

i really liked the obfuscation scheme

[glass flower]

[tender river]

they split each function into small chunks, added some bogus chunks, and had a single switch statement per function

then each function looped a few times setting next "opcode" to run a different code chunk and update the next opcode

the thing is the next opcode was not just set, it was dynamically updated

this is not impossible to solve with symbolic execution

but in practice it just means you're gonna have to write custom control flow analysis for reassembling the program instead of just looking at the code

and i didnt feel like it

[sage crag]

nyum

[real sierra]

why bug

cannot figure it out

[obsidian mantle]

What bug

[real sierra]

on windows

all of the bindings seem to return illegal moves

not make sense

[obsidian mantle]

Cant you run it in debug mode and see where it comes from

If both chess api and chess bot have source

[real sierra]

i dont have the c version of the chess bot on windows, and bindings don't give stack trace into the api

and it only occurs when i use it through the cute-chess application, not when i test in terminal

[obsidian mantle]

I had same problem with arena because it restarted the bot and it forgot its board

[tender river]

did you check that start_chess_api() actually blocks

[real sierra]

it does

in a terminal at least

[obsidian mantle]

Does api need to stay in memory between moves

To remember the board or smth

[real sierra]

it will stay in memory because it's an application

the application isnt being closed

and even if it were restarted, that wouldn't be an issue

[obsidian mantle]

Oh so you run it manually and not just provide exe to the cute chess

[real sierra]

no i do give an exe to cute chess

but that exe is not being killed

because it's replying to cute chess

that's a separate exe with its own memory, it can't just have its brain removed by another application

[obsidian mantle]

Is it giving illegal move on turn 0

[real sierra]

not always

the turn is random

[tender river]

can you try testing in terminal but not manually

[nocturne olive]

How interesting
Pretty certainly not something that happens on Kotlin

[obsidian mantle]

What moves does it give

Does it give startgame pawn moves or

[real sierra]

with file input it's hard, the gui is not supposed to respond with more position or go commands until it hears back from the bot

[obsidian mantle]

Or startgame whatever the board was when it started

[real sierra]

good morning superbox

windows is horrible

[nocturne olive]

I assume because the copy is definitely made in JVM memory and cannot in any way be pointers to native memory

[real sierra]

and it wasnt a free issue like i thought

[tender river]

just write your own test harness at this point, better than guessing

[real sierra]

bug is present even if python api doesnt call free

[tender river]

then you can use ubsan/asan with that

[nocturne olive]

True Windows sucks

[real sierra]

its definitely some kind of memory issue though

[tender river]

depends on if you use msvc/clang too, they might exhibit different bugs good luck

[nocturne olive]

Are the move objects definitely being copied from the locations being pointed to into Python interpreter memory?

[obsidian mantle]

I have an urge to look into it but im at work why do timezones exist

[tender river]

thats not even the issue, it happens without python too

[real sierra]

something i do notice is that when i as a human play against it, sometimes it tries to play the move i just played

which means uninitialized buffer

[nocturne olive]

Weird

[real sierra]

agree

no clue what could be causing it, no part of this is platform-specific

[obsidian mantle]

Initialize it mhm

[real sierra]

valgrind sees no issues

[tender river]

ubsan/asan?

[nocturne olive]

I have not had anything like that with the KT API

[real sierra]

im as clueless to what these are as i was the first time you said them

[tender river]

not language related, os related

[real sierra]

[nocturne olive]

Hm, so Windows being Windows then

[tender river]

they're part of clang

[real sierra]

making me install clang

i should just preinstall every windows C build tool

why just have one when they each implement one crumb of the spec

[tender river]

what if someone tries to use the intel compiler

[real sierra]

banned from contest

next question

[obsidian mantle]

But it works with cute chess on linux? Or mac or whatever you have

[rigid timber]

hi yes hello will this work on Itanium

[faint sandal]

what if

[nocturne olive]

Getting what you want on Windows is so impossible
I downloaded like 25GB of garbage and still did not have C++ STD headers

[real sierra]

works on linux 100%

[obsidian mantle]

Maybe linux cutechess is different?

[tender river]

not cutechess issue

[obsidian mantle]

You cant run same exe on win and linux right?

[nocturne olive]

Everyone should just use Linux for real

[real sierra]

i downloaded only the build tools and it was still somehow 6GB
clion insists this still doesnt have threads.h but compiles anyway so

[nocturne olive]

Windows moment

[tender river]

intellij have their bespoke nonstandard tooling

[nocturne olive]

You mean JetBrains?

IntelliJ is the Kotlin/Java IDE

[tender river]

intellij is jetbrains's old name

intellij idea is the java ide, not intellij

[opaque sigil]

They're all intellij as a base anyway

[obsidian mantle]

100% solvable with debug messages

[tender river]

i just call jetbrains intellij at times, more descriptive

[real sierra]

this feels like such a no-win scenario

im trudging through the hell that is windows development to enable others to suffer the same fate

[rigid timber]

only C and C++ are hell on Windows tbf

[tender river]

shiro

this is a chess competition

[obsidian mantle]

But if you never worked on linux you will suffer on linux

[tender river]

you're supposed to say zugzwang

[sage crag]

#define PRESERVE_NONE __attribute__((preserve_none))

PRESERVE_NONE void **translate(uint8_t *bytecode, size_t len, void *table[]) {
    void **translated = malloc(len * sizeof(void*));
    for (size_t i = 0; i < len; i++)
        translated[i] = (void*)table[bytecode[i]];
    return translated;
}

PRESERVE_NONE void run_goto2(uint8_t *bytecode, size_t len) {
    void **translated = translate(bytecode, len, (void* []){
        &&op1, &&op2,
    }), **pc = translated;

    next: goto **pc++;
    op1:
        printf("op1 (should terminate)\n");
        free(translated); return;
    op2:
        printf("op2 (should continue)\n");
        goto next;
}

ye works

i wonder if there's any tangible performance improvement from this

[nocturne olive]

True just have everyone use either Linux or WSL

[sage crag]

i bet it sucks for programs without loops

[real sierra]

is this for your

your

[sage crag]

no

[real sierra]

bwa

[sage crag]

this is for a vm that doesnt have a 1024 token limit

[real sierra]

how will i fit it on my floppy drive

[sage crag]

if im translating the bytecode i may as well JIT it

also i'll have to load operands from the original bytecode anyway

so its just not worth it due to cache i guess

[tender river]

[sage crag]

which is a shame

[rigid timber]

can you even run cute chess under WSL?

[nocturne olive]

I think it's time to add docs to Kotlin API

[opaque sigil]

Give it a try

[nocturne olive]

I have no idea, I guess depends if you can run X11 under WSL

[opaque sigil]

you can

[tender river]

biggest problem with jit is regalloc, if you have enough registers its not a problem

[sage crag]

the easiest way is to just back the registers in memory

[tender river]

ye surely l1 fast enough

[real sierra]

static Move *get_legal_moves(Board *board, int *len)

[sage crag]

i wonder if there's any way to get the best of both worlds

[real sierra]

    *len = len_moves;

    size_t len_moves = 0;

bad?

[tender river]

should be fine

[real sierra]

size_t to int fine?

[tender river]

ye

[real sierra]

[hollow spruce]

Are u modify chessbot.h ?

[real sierra]

chessapi.c

the .h is remaining unchanged

-# (i hope)

[tender river]

might be undefined if size_t is out of int range but its not gonna be so everything is fine

[real sierra]

memset(&add_move, 0, sizeof(add_move));

sizeof variable...

[opaque sigil]

I forgot what the static even does in this context

[real sierra]

doesnt export function

[tender river]

is fine

[real sierra]

maybe windows is just broken

[tender river]

makes the function or data local to the compilation unit

[real sierra]

my code would never have bugs

[opaque sigil]

[tender river]

test with clang on windows

[real sierra]

cruelty

🥀

[opaque sigil]

Would be funny if it's just msvc being broken

And unsurprising

[tender river]

if it works you have a clear path for msvc refugees at least

[real sierra]

adesi was using clang and doesnt work for them either

[tender river]

if it doesnt you have good debug tooling

(ubsan & asan)

[real sierra]

clear path for windows refugees: wsl2

if you want to submit a bot in windows please ask microsoft to write chessapi bindings

[sage crag]

wtf

the translated one is faster

cache randomiation

[tender river]

cpus are so weird

[sage crag]

ye on linear program it was about 5% faster consistently

[midnight sigil]

is llvm on windows tested

[sage crag]

and on the stdout throughput test it went from 1.2gb/s to 1.76gb/s

[opaque sigil]

I don't have a windows to test

[sage crag]

on a simple test vm, i should note

[tender river]

shiro will test it any second now (please notify shiro about it)

[sage crag]

so i just tested it

[opaque sigil]

I'm convinced cpus are secretly conscious and will vary their speed out of spite

[sage crag]

and its faster on 2 instructions

2

how

when instruction count small, previous one should be faster

its still slightly faster with 1 instruction

i hate computers

that literally makes no sense

[real sierra]

wait

how do i have clang-tidy

do i already have clang

????

[sage crag]

considering the second one calls malloc and free

[sage crag]

[real sierra]

why would i put that in a while statement

[tender river]

Atomically unlocks the mutex pointed to by mutex and blocks on the condition variable pointed to by cond until the thread is signalled by cnd_signal or cnd_broadcast, or until a spurious wake-up occurs.

[real sierra]

[tender river]

unsized buffer bad ye

scanf_s

[real sierra]

it fine

[sage crag]

the s stands for safe

[real sierra]

deprecation excessive

[sage crag]

[real sierra]

it blocks until signal and then wake up

so why would i put in while loop

[sage crag]

ok now to figure out how to handle operands with this new format

[tender river]

or until a spurious wake-up occurs.

[sage crag]

[real sierra]

it wont

[tender river]

[rigid timber]

would be fun if the bug was because it did wake up...

[real sierra]

but if it did

[sage crag]

exit

[real sierra]

i guess it doesnt break anything to make it a while loop

[tender river]

have separate array for data maybe? a bit scuffed though

[real sierra]

nevermind it apparently does break something

oh because i did it wrong

[tender river]

might want to make Semaphore* volatile not sure if the compiler can figure out that sem->locks can change after cnd_wait

[real sierra]

[sage crag]

compiler might optimise out your waiting

[tender river]

they wont optimize it out but they might optimize out the read

[tender river]

compilers are allowed to reorder reads and writes to happen anywhere as long as they are reasonably sure (for a definition of reasonably sure, refer to the standard) it wont affect program behavior

i'm not sure what c's definition of reasonably sure is

Semaphore* isnt restricted so it may be fine either way

[real sierra]

im getting like 1000 warnings now

the mtx and cnd functions dont like that Semaphore is volatile

[tender river]

fine undo it

[midnight sigil]

chat is using comic sans for programming cursed

[real sierra]

(ignore the "more than once" ones)

[obsidian mantle]

Warnings? Who cares

[sage crag]

could but locality

[obsidian mantle]

Its only bad if the code is bad

[sage crag]

cursed form of JIT where instead of actually jiting anything you just replace each instruction with the instruction handler

[obsidian mantle]

There was some safe function that behaved completely different for no reason relative to unsafe function. I hate safe functions

[sage crag]

[tender river]

i found this page also https://wiki.sei.cmu.edu/confluence/display/c/CON36-C.+Wrap+functions+that+can+spuriously+wake+up+in+a+loop

[real sierra]

that looks pretty much like what i have

[opaque sigil]

Is this about strcpy vs strncpy

[real sierra]

i resolved the warnings by only making Semaphore.locks volatile

is that sufficient

[tender river]

no

[real sierra]

[tender river]

i remember it

let me find it

[real sierra]

the problem is that the mtx_ and cnd_ functions ask for non-volatile pointers in their signatures

C doesn't like it when you typecast a pointer to something volatile, to a pointer to something non-volatile

[opaque sigil]

Good

[real sierra]

in fact, it might be undefined behaviour to do so

[tender river]

i think strncpy vs strncpy_s? the former pads with 0s, the latter one doesnt

[real sierra]

so i just cant make this volatile period

[tender river]

ye dont do it

who even suggested that

[real sierra]

i make Semaphore.locks volatile

maybe thats good enough

the loop is conditional on locks anyway

not on the mutex or condition

[sage crag]

me genius

[real sierra]

surely this fixes my issue

!

[sage crag]

dont need registers

therefore dont need operand

[tender river]

stack?

[sage crag]

or accumulator

[tender river]

one opcode for pushing every possible byte

[sage crag]

this would quickly turn the vm into esoteric territory though

since no immediate

[obsidian mantle]

Now the code is different and you are not sure it would work even on linux

[tender river]

you can use 2-byte encoding with opcode-operand or 4-byte with 1 byte opcode and 3 byte operand

8 byte opcode 24 byte operand

[obsidian mantle]

Make 128bit vm

[sage crag]

of course, i'll replace registers with addresses

overcome the indexing cost

[real sierra]

send bytes to my address

[sage crag]

at runtime of course

just precomputing the register memory address during the translation step

bwaa translation step being separate from main vm will double the program length

bwaa

bwa

[obsidian mantle]

Is read 1 byte = 2byte = 4 = 8 on 64bit systems?

In terms of time

Or how does it work

[sage crag]

ye koni make the fastest vm in the world TODAY

when i wake up

[opaque sigil]

8 byte opcode 56 byte operand, one cache line per instruction

[tender river]

thinking of cache lines already

thats how you know its fast

[opaque sigil]

[sage crag]

i'd rather do 24 byte operand and fit two instructions per cache line

since that makes it easier to pipeline

[tender river]

a 64-bit cpu except its a 64-bit opcode space rather than 64-bit address space

[obsidian mantle]

I mean how many bytes can a system read with 1 operation

[sage crag]

esoteric ISA design

its lower level than esolang

add1
add2
add3

[opaque sigil]

64 on most cpus

[tender river]

ye bring back room scale mainframes

[sage crag]

depends what you mean with 1 operation

[obsidian mantle]

See so 8 bytes read = 1 byte read

Or no

[sage crag]

technically an arm machine can read 2048 bytes with 1 operation due to wacky

[obsidian mantle]

Naah but its going to repeat it in cycles don't it

[opaque sigil]

In terms of speed? Yeah

[tender river]

technically x86 can rep stosb

[sage crag]

streaming vector extension

[sage crag]

ye i was going to mention that too

but that a loop in microcode

not fair comparison

[tender river]

[sage crag]

mostly yes, with caveats

typically alignment

[tender river]

in general you dont want to do a single operation on cpus

you want to do multiple operations

if the cpu cant figure out that you want to do multiple operations, everything is just going to execute one instruction at a time

[sage crag]

also in x86 ISA itself there is extra data to be encoded for 64 bit instructions than 8 bit instructions

[tender river]

where it really doesnt matter whether its 1 or 8 or 64 bytes

[sage crag]

0 performance impact i just wanted to mention

[sage crag]

which cpus are pretty good at still

especially with pipelining

[tender river]

and then other metrics like cache size and memory speed and prefetching and speculative execution come into play

[sage crag]

but you dont need to think about those

cpu magic rock

you programmer just throw bigger cpu or bigger library at it

until it work

[tender river]

i throw bigger brick

[sage crag]

[tender river]

[real sierra]

maybe i give up

[tender river]

go to sleep but do clang tomorrow

[real sierra]

i think i do have clang

[tender river]

ubsan/asan

[real sierra]

i get clang linter errors at least

[opaque sigil]

clang-tidy != clang

[real sierra]

dont have ubsan or asan

[obsidian mantle]

Debug the process

[real sierra]

[tender river]

ubsan and asan are clang things

not external binaries

[real sierra]

ok i dont have clang

[obsidian mantle]

Who just switches compilers randomly

[real sierra]

ive switched compilers like

3 times tonight

[tender river]

everyone

[real sierra]

im so damn tired

i have an assignment due on wednesday

[opaque sigil]

Any excuse to not use msvc

[real sierra]

msvc is the only compiler that actually makes a working dll for me at the moment

[tender river]

ye go sleep and i did warn you about rushing

[real sierra]

i dont wanna hear an "i told you so"

[opaque sigil]

Why are you doing this then

[sage crag]

[real sierra]

because it needs to be fixed

[sage crag]

funny me too

[opaque sigil]

The windows people can wait

[tender river]

either way good luck, cant help with windows stuff

[real sierra]

people wont wait that long

but they can use wsl in meantime

already added a note to the readme

letting people know that windows building/testing probably has to happen in wsl for the time being

[opaque sigil]

[nocturne olive]

I kinda wonder if KT API works with a Windows DLL

[real sierra]

i had to modify the python api to search for a DLL if on windows

[nocturne olive]

It's not explicitly looking for an SO file

[real sierra]

then it probably maybe works

[opaque sigil]

You're like 3 levels above it probably works

[real sierra]

thats where the maybe comes in

[nocturne olive]

It's just trying to Native.load("chessapi")

[nocturne olive]

From the ./native folder

[tender river]

maybe intended but the kotlin bindings only have a readme for now, letting you know just in case

[opaque sigil]

Assuming that uses loadlibraryex it should work I think

Assuming the name is right

[nocturne olive]

I haven't even pushed them anywhere yet?

[real sierra]

i pushed the readme you sent

[nocturne olive]

Oh how silly

Well expect the rest of the things in a few more hours

[real sierra]

maybe i'll add a small note that it's under construction

[nocturne olive]

Once I push to a repo you'll want to definitely link that repo in your copy of the readme

[midnight sigil]

I waited 20 mins to download windows llvm installer

github you slow ass

[real sierra]

just put in a pull request with the link and remove the "UNDER CONSTRUCTION" header on the README.md when it's ready

i'll review it when i get a chance tomorrow or the next day

[nocturne olive]

Silly

[opaque sigil]

I had the pleasure of downloading the nix index for an hour the other way, was fun

[real sierra]

i hope it's working for linux users

it hasnt even been a full 24h so not every timezone has even seen the announcement

but there's still surprisingly few people that've come to ask questions

[nocturne olive]

Based on the Kotlin API development it works fine on Linux

[real sierra]

either people are really uninterested or many have figured it out

[nocturne olive]

We will see

[opaque sigil]

It's been a quarter of a day at night shiro

[real sierra]

and ive slept for none of it

[stray dragon]

wait the announcement happened?

[opaque sigil]

Me neither

[stray dragon]

do we have api

should i be awake

[opaque sigil]

Check #mod-announcements

Yes

[stray dragon]

oh god i need to be awake

[midnight sigil]

Canary Deployment

[nocturne olive]

Should I also add a wiki entry in that if I can figure it out or do you want to do that part?

[real sierra]

yes please

[nocturne olive]

I personally have no clue how to create one

[real sierra]

it's pretty simple but idk if that can even be done via pull req

if you cant just lmk and ill pop the md into a new page

its very easy

[nocturne olive]

Well I assume a fork of a repo contains the wiki

[real sierra]

honestly not sure

[nocturne olive]

Well I guess we'll see

[real sierra]

the wiki isnt in the normal file system of the repo

its odd how github does it

[nocturne olive]

If not I'll just give you the file

[real sierra]

no pressure

[nocturne olive]

We will see what happens

[real sierra]

alright im off to bed its 4am

good night

[nocturne olive]

That is indeed a bit late

[opaque sigil]

Imagine being up so late

smh

[stray dragon]

"late" is whenever i want it to be

i am in control here

[tender river]

its 3pm here

i should really be asleep by now

[opaque sigil]

[midnight sigil]

[stray dragon]

[midnight sigil]

I'm just here waiting for the typhoon impact

[stray dragon]

about 12 hours out

[obsidian mantle]

I'll try to help figure it out later today

When im home

[midnight sigil]

it's in the middle of changing its eye again

imagine bro succeed before impact

[stray dragon]

eyewall replacement cycle? i don't think it's gone through one yet

didn't see windspeed drop

at least, no drop + regain

looks like it's just been entering unfavorable conditions

[midnight sigil]

been one at least

[stray dragon]

oh damn you're right

[midnight sigil]

there's total of 4 cycles according to some taiwan guys

[opaque sigil]

Stay safe

[stray dragon]

not a direct hit and the storm is weakening, so probably fine

[midnight sigil]

it's trying to cycle again at this moment, this pic was 10am GMT+8 iirc

100+ km eye coming

[patent elk]

is that matthew

[worldly acorn]

Bartholomew?

[midnight sigil]

so I guess I didn't compile the dll correctly

[trim valve]

[midnight sigil]

I did clang -o libchess.dll -shared bitboard.c chessapi.c

is that not correct

[obsidian mantle]

How do you even link dll to main file
#include it on top?
Or need to do some project setting shenanigans

Never used dlls

[opaque sigil]

You include the header and link with the dll

[opaque sigil]

Chances are it doesn't actually export the functions by default

https://learn.microsoft.com/en-us/cpp/build/exporting-from-a-dll-using-declspec-dllexport?view=msvc-170

[midnight sigil]

[opaque sigil]

Can also write a .def file I guess

Might need to have clang generate the corresponding .lib too for the declarations who knows

[nocturne olive]

In Kotlin using Native.load("name")

[obsidian mantle]

Can I attach my visual studio to dll if i have source to debug it

Probably yes

[nocturne olive]

If you have symbols, I don't know

The DLL needs to be compiled for debug at least for sure

[opaque sigil]

It's definitely possible

[obsidian mantle]

Maybe reconfigure it to run raw without dll

Might be easier

[midnight sigil]

👍 you're correct

I did this and it at least can run now

and cute chess can run it

yes it's given illegal moves(and incorrect numbers of legal moves) mhm

[obsidian mantle]

Incorrect numbers like, it gave 20 moves but said its 3 moves?

Or does the structure of moves not work like that

Whatever i need to see it first
Speculating isnt so fun

[faint sandal]

taiwan

[dry charm]

I have a what moment

[rigid timber]

do you happen to have multiple node versions

[dry charm]

nop

I would use nvm then

but I use NixOS

that is why I have a what moment

WAIT

no shot

[rigid timber]

lemme guess, you figured out you do have multiple versions of node

[dry charm]

Nope

worse

[rigid timber]

that implies you only have one version of node but I don't see how that's worse

[dry charm]

LOVELY

NixOS @angular/cli ships own nodejs instead of using the installed one

[rigid timber]

lmao

tbh you don't really need it as a global package...

[dry charm]

the cli?

[rigid timber]

yeah

[dry charm]

it is not global, it is in a devshell

[rigid timber]

[dry charm]

basically, I can create development enviroments, where required packages are available when I enter the directory

and it doesn't dilute my global environment

I can specify versions even for different projects on demand

[rigid timber]

... but Angular projects already have @angular/cli as a devDependency, so why would you need an extra copy

[dry charm]

for ng command

dev dependencies do not extend for env

so you do not have ng command

This is why every angular install starts with npm install -g @angular/cli

[rigid timber]

true, you'd have to do npx every time or use something that does the env stuff for you automagically

and even then it'd only work for existing projects, not for scaffolding new ones bc npx can't run packages without installing them...

[nocturne olive]

Kotlin bindings for Chess thingy now just waiting for Shiro approval

[rough bloom]

why would it not
if it's an application then the interpreter is built into it, otherwise you have funny runtime dependencies
this is pretty much always the case in Nixpkgs AFAIK

if you want a different Nodejs then override it

[dry charm]

It makes sense, and obvious

Just at the time I was very confused why this is the case

[midnight sigil]

it should be 20 moves but instead gave 4 moves

4 illegal moves

[obsidian mantle]

So it counts them correctly. Good

[midnight sigil]

the counting mechanism is made by me, just doing len() on the legal moves given

I'm not sure if there's uci commands that returns the number of legal moves

[obsidian mantle]

There is none

Oh. I guess you are not using C bindings.
Thats why i was confused - c bindings give length right away with move array

And i definitely saw it somewhere before

[midnight sigil]

WHY WOULD A FREAKING DRAW DECREASE MY ELO

bruh

[noble zodiac]

but.... why tho

[worldly acorn]

Because chess is a language

It is also the friends we made along the way

[olive sable]

im home

tuesday is a short day

[hollow spruce]

It means that 🙂

[midnight sigil]

I can finally say "I speak chess" confidently

[hollow spruce]

🤨

[fast pagoda]

I have been asleep for 12 hours so I wake up to API drop hooray

[amber fractal]

[fast pagoda]

Also I like how hal 9000 is moving the chess piece from the observation side of the chess board implying he is not even a player @real sierra

[thorn chasm]

I told myself in the past to never go back into chessdev, but how can I resist

[worldly acorn]

You don't

It calls

[amber fractal]

Well, guess the month timer for chess has finally started.

[olive sable]

"i like the numpad, it has all my favourite numbers"

most insane quote from class

[obsidian mantle]

Whar

[amber fractal]

I thought that was your quote @olive sable

[olive sable]

[fast pagoda]

It has all my favourite numbers too

[obsidian mantle]

What about 21

[fast pagoda]

All the greatest hits in there

[olive sable]

how did you know? why are you in my walls?

[olive sable]

just press the 21 key

[amber fractal]

Just press the 10 key

[obsidian mantle]

Its not on the numpad

I hate numpad

[amber fractal]

(Why is the 10 key real)

There are so many keys that have no right to exist

Like the xf86 red, green, yellow, and blue keys.

[olive sable]

must not be a good numpad

mine does have 21, but you have to do a key combo

you have to press 2 and 1

[obsidian mantle]

???? Am i supposed to know the combo for every number higher than 9

Insane

[olive sable]

yes

this is the difficult technique i have mastered a century ago, i call it counting

its from the far regions in arabia

[obsidian mantle]

[amber fractal]

There are some that do deserve to exist, like the 30ish dedicated macro keys

[olive sable]

nobody uses the home, page up, page down and scroll lock keys

[obsidian mantle]

Ctrl + Shift + Alt + Win + L

[olive sable]

the other ones from that island above the arrow keys can stay

[amber fractal]

Can't tell which app that opens

[obsidian mantle]

I used pageup/down

[olive sable]

im on linux so idk

[obsidian mantle]

In some old game console it was the only way to scroll

[clear sedge]

[olive sable]

ok but i spend 130 bucks on a keyboard with nice switches, i dont want to spend more than i have to on buttons i never press

[obsidian mantle]

It opens linkedin page in default browser

Anybody touched the new token counter yet? Does it have problems too

[amber fractal]

Not as far as I know

[amber fractal]

Edit, found the site that I use to check keys

https://www.tcl-lang.org/man/tcl8.6/TkCmd/keysyms.htm

[midnight sigil]

2.7 seconds sounds suboptimal

[bleak depot]

Which keyboard you have

[olive sable]

corsair k70 rgb pro

with the cherry mx brown switches

[obsidian mantle]

cherry switches interesting

[olive sable]

wdym?

[obsidian mantle]

Sounds unusual

Oh its an term

[olive sable]

i think cherry is a pretty popular brand

[obsidian mantle]

Or is it a company

[trim valve]

surely implementing a bunch of transmutes in my chess api will have no negative consequences

[obsidian mantle]

I thought its colour

[olive sable]

cherry is the brand, brown is the type

[midnight sigil]

cherry mx brown 🥀

[amber fractal]

*prown

[olive sable]

why? i dont like clicky switches

[midnight sigil]

wait for you to hear the 78db smack

[olive sable]

?

i dont feel like my keys arent that loud

wait

double negative

[midnight sigil]

[olive sable]

i guess they are loud my typos says so

[amber fractal]

Fact checked infomation yep, see no issues with the statement

[midnight sigil]

and cherry mx brown is uhh

[olive sable]

tactille

[midnight sigil]

stucked in the middle of linear and clicky

fragmented

bad

[olive sable]

i dont care much for your opinion

[midnight sigil]

[olive sable]

i liek knowing when the key has activated, hence i like the tactile response

but i dont like audible clickeyness

[midnight sigil]

get magnet switches

[olive sable]

i aint psendign my money on all that

[midnight sigil]

smh

[trim valve]

what did I do this time

why did you make the same move again silly computer

[olive sable]

we love weird errors

[midnight sigil]

[obsidian mantle]

What does N mean

[midnight sigil]

Imma hook a debugger to see wha da fa is happening in chess_get_legal_moves

[midnight sigil]

Knight h6

[obsidian mantle]

you going to fix it before me

[midnight sigil]

no

I'm bad at programming

[obsidian mantle]

Every god programmer ever "im bad at programming"

[trim valve]

this may or may not be my fault

[midnight sigil]

[trim valve]

time to test the C bot

[midnight sigil]

are you on windows

[trim valve]

but I think I may have added a bunch of unsafe without thinking

no

[midnight sigil]

linux?

[trim valve]

yeah

[midnight sigil]

oh so it's a common problem

[trim valve]

hm?

[obsidian mantle]

Shiro said before its only windows problem

[dry charm]

uhoh

its spreading

[obsidian mantle]

But it was about "on windows it always gives incorrect moves"

[midnight sigil]

he was wrong

[trim valve]

do not discount this being a problem entirely caused by me yet

[midnight sigil]

modify your bot to print the length of the returned legal moves, and run the bot directly then do

ucinewgame
position startpos moves e2e4
go

see if the result(printed number) is 20 or 4(or neither)

[hollow spruce]

New project update
Making new based on India theam games on pc
Max allocation gb is 302

[obsidian mantle]

Which one is correct? 20?

[midnight sigil]

yes

[trim valve]

ucinewgame
position startpos moves e2e4
go
legal moves: 20
info currmove g8h6
bestmove g8h6

[midnight sigil]

it's correct and legal move

[obsidian mantle]

It looks like its rare on linux and frequent/always on windows

[midnight sigil]

let's continue and do position startpos moves e2e4 g8h6, go

repeat until bug appear again

[trim valve]

legal moves: 30
info currmove e1e2
bestmove e1e2

[obsidian mantle]

Inb4 memory alignment changed because of print function and now it disappeared

[trim valve]

lemme test that theory

(it won't be memory alignment but could realistically be a skill issue on my part)

[clear sedge]

it is very suboptimal

my guess is heap allocations are fucking me over

[nocturne olive]

Whar you sillying?

[trim valve]

low allocation json deserializer

[nocturne olive]

I see a lot of Chess stuff

[obsidian mantle]

Doing research on false legal move bug

Damn my thumbs are too thick so many typos

[clear sedge]

i'm just gonna write my own allocator so i only need to do actual heap allocation a couple or so times per deserialisation

[trim valve]

fair enough

[stark needle]

rrrrrr

[midnight sigil]

the C version seems fine

is it only happening in the bindings

python one happens everytime

[obsidian mantle]

Shiro has it in c every time on windows i think

[midnight sigil]

that's funny

[fast pagoda]

tfw sleep for 16 hours

and now i have to work for 10

[final tinsel]

just woke up... they still havent made a move

[fast pagoda]

but i want chezz

[midnight sigil]

fucking 30878968564 nodes

your stockfeesh has analyzed 30878968564 chess positions

[final tinsel]

30.9B nodes

[midnight sigil]

great

👍

[final tinsel]

i didnt expect it to think this THIS long lol

[midnight sigil]

you are supposed to limit its movetime

or depth

[final tinsel]

well ik

but like

[trim valve]

new idea for the chess competition, move time inversely proportional to token count

[final tinsel]

i wanted to see how long itd take

[trim valve]

20 tokens -> 20 minutes, 1024 tokens -> 1 minute

[final tinsel]

i expected one move to be less than 10h at least

[trim valve]

surely that would end fine

[midnight sigil]

it'll go on until it reach the limit of your hash size

[fast pagoda]

me when "but let's just search to depth 9000"

[midnight sigil]

nah it won't surpass 300

even if we have infinite hash size

[fast pagoda]

https://tenor.com/view/doctor-strange-dr-strange-strange-benedict-cumberbatch-avengers-gif-22604801

[trim valve]

ig inversely proportional isn't the right term here

[final tinsel]

i was thinking the same thing lol

[midnight sigil]

"We have made an evaluation algorithm for chess that's down to 5 decimal places, and we're going to find the most optimal(for now) chess game"

[midnight sigil]

that many nodes and yet the engine will still tell you e4 is the best move

[silent cloak]

its calculating the best way to throw

[stray dragon]

the hell is -lchess supposed to be

[midnight sigil]

so uhh

it's returning len_move of 20

but returned an array with 21 elements

for the second move it only returns knight moves, with again an extra illegal move

third move and it's all nonsense with an extra a1c6, again

gotta look into that

[rigid timber]

Shiro should've just used a library

[midnight sigil]

I don't know about linux, but this is Windows 11, with LLVM 21.1.1(x86_64-pc-windows-msvc)

[nocturne olive]

With the correct passing method the Kotlin bindings at least work fine on Linux

[midnight sigil]

:neuroTrue:

[stray dragon]

very cool

[final tinsel]

it finally calulated another depth step

~3.5B nodes now

[stray dragon]

of course there's like no proper solutions on the first page of search results why would it be that easy

[glass flower]

do you have a old compiler?

[olive sable]

i thought you just did #include <thread>

thats how ive been doing it

[stray dragon]

that's how it works in C++

in C it's threads.h

[olive sable]

huh

okay

[stray dragon]

got it fresh today

[glass flower]

what compiler are you using?

[stray dragon]

gcc

the specific command used:

[glass flower]

i only tried clang... so good luck

and clang worked for me

[stray dragon]

followed the VScode guide on setting up a C/C++ environment to use, got that stuff installed, restarted computer, test the compile

[glass flower]

why would it not recommend msvc on windows...

gcc seems like a odd choice as the default compiler

atleast on windows

[stray dragon]

mostly using gcc because of this line in shiro's competition

i know precisely nothing on the differences between these compilers

[glass flower]

i see....

nowadays i don't think there is much difference. atleast on like the basic features

[stray dragon]

guess i try clang

[shadow sinew]

Any word on runtime/memory limits? Could you for instance set up a gigabytes-sized lookup table on the heap?

[stray dragon]

60 seconds on the bot's side in total for each chess game

[shadow sinew]

Any extra time for precalculations on startup?

[stray dragon]

don't know but doubt it

[worldly acorn]

prob not though

heck if id know

[stray dragon]

apparently the bot's memory is not cleared after each game though

"Bots may not be restarted between tournament games, so be careful what you cache."

[worldly acorn]

dang

[real sierra]

[real sierra]

in theory yes, but if you take too long and your engine is deemed unresponsive by the Gui you will probably be disqualified. I would try not to let this exceed a couple seconds.
as well, no thinking during the opponent's turn (i.e. ponder is off)

[real sierra]

mem limit per bot is ~4GB but I wouldn't try to get too close to that limit

in fact I may drop that further

given how many bots there may be

I would not exceed 1GB, I'll add this to the FAQ

[midnight sigil]

you need windows sdk to be installed/linked/included

[midnight sigil]

cc @real sierra what's your thought

I haven't read the codebase yet, just took some tests

and ate mcdonald

:NeuroopaTalk:

[real sierra]

should be < len_moves not <= len_moves

your loop bounds are wrong

[midnight sigil]

so the extra move is explained

still returning incorrect numbers of legal moves tho

and incorrect legal moves

[real sierra]

how curious

windows?

[midnight sigil]

mhm

[real sierra]

naturally

[sage crag]

sync primitive mabe

[midnight sigil]

idk if it only happens on windows or not

[real sierra]

there's some windows bug bashing ongoing

[sage crag]

windows bashing

thats illegal

[real sierra]

not sure yet which operation I'm doing that's somehow platform dependent

[sage crag]

dont break windows glass is expensive

[real sierra]

maybe multiplication is too hard for windows

[sage crag]

[real sierra]

good info though, thank you

[midnight sigil]

according to bred, this also happens on linux

idk man

I don't have a linux on my hand can't test that out

I think he's using the python binding

[real sierra]

bred said they hadn't ruled out it was them but if the C bot isn't working on Linux either then that's concerning it works flawlessly on my system along with all bindings and nobody has reported issues until the contest started

[rigid snow]

should've developed the whole contest around docker containers

[real sierra]

there is a small concurrency change I haven't pushed yet for the C API because it didn't affect the outcome for me but it could be helpful

[rigid snow]

[midnight sigil]

"I can't replicate this bug on my end, I'll mark this as Unresolved/Need More Information."

[real sierra]

should've banned windows

[stray dragon]

windowsphobic

smh

[real sierra]

correct

work is ongoing to figure out why the API gets sad on windows

[midnight sigil]

because only 3~6 personals were able to access your chessapi before the contest announcement smh

[lime galleon]

What about macOS

[real sierra]

@fast pagoda you're up

[midnight sigil]

macos should work, it's POSIX

my bro Windows think it should be self-dependent

fuck POSIX

who tf cares universal ABIs

[rigid timber]

there are no universal ABIs

[glass flower]

i saw this yesterday.... idk what it means. with clang -v

why does it say thread model: posix

[midnight sigil]

it means it's using POSIX as the threading implementation for the clang compiler

[fast pagoda]

@real sierra mfw i can't tell what is hatemail and what is normal messages directed to you

[glass flower]

but wouldn't that mean pthreads should have been included?

and be possible to be used in windows

[midnight sigil]

the compiler is independent with your program

[glass flower]

[midnight sigil]

MSVC doesn't support POSIX, but the compiler works on POSIX, compiling a program with POSIX threading doesn't mean your program works with POSIX

[glad path]

nothing is wrong with macos

[midnight sigil]

sry meant APIs

[nocturne olive]

It appears that Shiro has awakened

[fair horizon]

For the chess stuff is there any limit on how long it takes to compute a move?

[obsidian mantle]

is it fixed yet? im finally free

[glass flower]

i see...

[fast pagoda]

@midnight sigil @rigid snow @glass flower It's been brought to my attention that there may be some issues with the API on your various machines and that there are some concerns regarding the process surrounding the API's development.

@real sierra has kindly directed me to handle any and all adverse messages regarding this issue and I would like to provide the following clarification that may help, it comes directly from on high:

[midnight sigil]

shibo just awake and he stated he's investigating

[glad path]

"if it works it works"

[lime galleon]

Happy to hear that cuz gamers roasts me for using it

[glad path]

they're just idiots

nothing is wrong with macos even for gaming

just get fusion or somethin g

or whisky

unfortunately crossplay is expensive

[rigid snow]

i have not done anything what issues

[lime galleon]

I just love unix

[midnight sigil]

bro speaks like a high-salary PM

[glad path]

real

[rigid timber]

can we fire this PM and get a new one

[glad path]

good question

[fast pagoda]

i can't tell what is hate mail to shiro and what is just normal messages to shiro ( theyre the same) so i just tagged whoever seemed to be talking about it

[worldly acorn]

W pfp

[glad path]

wait is it just me or is it the turret from portal but chopped up into an octopus looking thing

[worldly acorn]

Walle

[rigid timber]

no

[worldly acorn]

The steering wheel

[glad path]

what is it ._.

[worldly acorn]

https://tenor.com/view/wall-e-pixar-disney-earth-is-amazing-captain-b-mccrea-gif-16046020

[glad path]

OH

wait its that one thingy from walle

[rigid timber]

it's AUTO from WALL-E

[worldly acorn]

Looks just like it

[glad path]

(i somehow have never watched walle 😭)

[rigid timber]

[worldly acorn]

Their name IS auto, complete forgot about that

[fast pagoda]

wall-e is incredible

[worldly acorn]

Ong

[fast pagoda]

one of my favourite movies

[worldly acorn]

Watched it on repeat as a kid, same with cars

[fast pagoda]

same

[glad path]

i watched cars but dont remember a lot of it

[worldly acorn]

Life is a highway

[glad path]

i also like
dont remember moana nearly at all