programming

[hoary lion]

searching it up, seems like all languages are some kind of mix

[tender river]

anyway the important thing is thinking about how your data flows

[amber fractal]

Not hard to make anyways, that one is just IPC sockets which have a lot of libs in any language.

[tender river]

any program fundamentally just moves a bunch of data around small-ish functions

[knotty current]

which is what makes these kind of languages beautiful

[knotty current]

true

[tender river]

and if the data dependencies are simple the program is also simple, and if they are some cyclic self referential hellhole they become hard to think about

[opaque sigil]

chat, is there a way to invert ONLY the pdf page in chromium's viewer as opposed to the entire viewer like you can with pdf.js (without running pdf.js inside chromium)

[hoary lion]

very interesting

[dense cosmos]

morning

i wonder how to make language bindings

[tender river]

anyway instead of listening to me go learn stuff (like a good programming textbook i always recommend is SICP since it teaches a bit of everything) and build stuff (the more different kinds of projects you do the more you will learn, for example i work on languages, networking, low level stuff, each area has different methods of thinking about stuff)

[dense cosmos]

or use them

[hoary lion]

i was like super mad today because rocm and cuda's stablediffusion results are different
and eventually i ended up looking PRNGs, but most of them are like in C++ or Rust
too harsh for smooth brain ML only person

[tender river]

i know ML
you know ML
we are not the same

[opaque sigil]

depending on the language it's just declaring a bunch of functions, that's it

which is why it's often automated

[dense cosmos]

also wondered why not make a build system that uses c++

[hoary lion]

52 year old language???

damn

all the good names are taken fr

[dense cosmos]

oh, okay

[opaque sigil]

https://github.com/tsoding/nob.h may i interest you in nob (c not c++ but still)

[dense cosmos]

as in, not written in c++, but written with cpp

[dense cosmos]

ooh, wonder how hard it would be to extend it to work with cpp

[opaque sigil]

one extern c block probably

[dense cosmos]

I should try that later

anything but cmake

for now, minecraft with friends

[opaque sigil]

have fun

but yeah, iirc nob targets c89 so it should work pretty much everywhere which is nice

i haven't used it for massive projects but the few smol ones i've used it with it's been really nice

[dense cosmos]

is there any reason you couldn't use it on big projects?

[opaque sigil]

i just haven't had any new ones

it should work just fine realistically, unless you throw something gigantic at it maybe

[dense cosmos]

I mean, why use cmake over nob

if it works

interop?

[opaque sigil]

cmake lets you fetch dependencies automatically if you're into that

[dense cosmos]

oh, hmm

[opaque sigil]

and is generally just more extensive

tbf there's nothing stopping you from just using curl and doing that with nob too

you are just writing c/c++ after all

[dense cosmos]

true

[hoary lion]

speaking of cuda stuffs, i should reinstall torch

cuda was not working, but with now the blessing of uv, there is automatic setup

[opaque sigil]

how would uv help with cuda shenanigans

[hoary lion]

i saw this from other server related to lms, let me forward it

[opaque sigil]

oh damn

that's pretty dope

i sure love how installing torch results in a 6.3GB venv

thanks nvidia

[amber fractal]

Most useful feature honestly

[olive sable]

Goodnight

[amber fractal]

good eep to you

[opaque sigil]

https://devblogs.microsoft.com/directx/cooperative-vector/ ?

[hoary lion]

cuda install is taking forever

[amber fractal]

neural rendering, hmm

[knotty current]

did i just fall victim to sudo pacman -Syu

[opaque sigil]

One thing you could use this for, as far as I can tell it just allows you to define some blas operations in shaders

Wdym

[knotty current]

ok maybe i just need to reboot

[amber fractal]

Why not just render the entire thing using AI

[knotty current]

ok thank gosh i dont need to recompile my kernel to get samba support back

[opaque sigil]

[hoary lion]

i think my regular deletion of temp and random msis made me cooked

wish me luck chat

[knotty current]

[hoary lion]

VISUAL STUDIO ARGHH

[knotty current]

InabaXDD IDEs all suck just use terminal

[hoary lion]

oh you pure soul

you do not have experience with windows

and its cuda

and c++

[knotty current]

never touched windows in my life

briefly tried to use python on windows

gave up

[hoary lion]

yk just fuck it we ballin straight to cuda 12.8

[opaque sigil]

Do yourself a favour and just use wsl

[hoary lion]

speaking of that, i should really remove wsl and start clean too

its like beyond tangled

[opaque sigil]

[maiden geyser]

bro unironically uses visual studio instead of msys or something

[maiden geyser]

but it's one of the best things on windows

[hoary lion]

i had very bad experience with msys

its just classic windows cuda

[knotty current]

fsr i couldnt do pip directly and back then my dumb ahh brain just gave up

besides i dont understand windows at all

i have been with unix my entire life (macos and linux)

-# teerible experience with macos btw

[olive sable]

Goodmorning

[hoary lion]

did you just eeeped 50 minutes

[maiden geyser]

morbing

[olive sable]

Yes

More like 30 tho

[real sierra]

every goddamn time I see this thing it's in the last place I expect

[olive sable]

The answer is yes

I can

[real sierra]

who remembers the complexity

gotta be at least one algorithm enjoyer here

[amber fractal]

My brain is failing to solve it without something to write down on

Don't actually know the algorithm

[real sierra]

https://en.wikipedia.org/wiki/Tower_of_Hanoi

[maiden geyser]

where is the Loss

[ruby timber]

Good morning

[amber fractal]

It's the loss of braincells of people that know what the problem is

[maiden geyser]

but there is no problem

all disks can be moved to tower 3

[tender river]

i dont "remember" it but i derived ||O(2^N)|| just now

[olive sable]

Something like that ye

Each layer seems to be more moves

[stark needle]

I keep getting these and "are you gay" tests all over youtube for some reason

[tender river]

you misinterpreted it, its задача not проблема in this context

[tender river]

are you though

[stark needle]

no

[hoary lion]

hmm

i hate tower of hanoi

[tender river]

[hoary lion]

i forgo the complexity of recursive solution

its probably somewhere in my notes

[tender river]

discord took a billion years to send it just pretend i sent it after shadows message

[hoary lion]

discord is not healthy

i though it was my internet

[tender river]

that'll teach me not to quote reply]

[tender river]

not to not quote reply

to quote reply

that's the one

[hoary lion]

[tender river]

my brain is cooked all i can think of when looking at negation is logic and continuations

[hoary lion]

only able to speak in not operations

[stark needle]

There's a way to algorithmically solve this problem?

Btuh

Bruh

[hoary lion]

recursive

[stark needle]

I wouldve just implemented bruteforce approach

[hoary lion]

isn't recursive quite brute force anyway

its hella inefficent

[tender river]

||if you find a way to move N disks, you can move N+1 disks||

[hoary lion]

is there a non recursive way tho

very interesting question

[ruby timber]

I doubt it- unless with a ginourmous code base

[maiden geyser]

there is a recursive way?

[hoary lion]

its the only way i know

call the previous n-1 with the same principle

[noble zodiac]

When something can be solved recursively it can always also be solved without recursion

[hoary lion]

but like it is so counter-intuitive

cant think of other way

https://www.geeksforgeeks.org/iterative-tower-of-hanoi/

eep

already exists

[noble zodiac]

but also a lil bit of recursion never hurt anyone

at least if you use tail recursion

[hoary lion]

it has been 50 minutes

any day now

this is not okay uv

[tender river]

strictly speaking not true as you cant always start from base cases, but more importantly is it really corecursion if you have to use a stack

[noble zodiac]

what example wouldnt work?

I dont read wingdings

[amber fractal]

looks like real osu code to me

[tender river]

right foldng a linked list

(yes you can technically do it with a stack)

[noble zodiac]

[hoary lion]

why steam game is installing its internal data to my c drive even when I explicitly mention it to download it on D

what the fuck is unity even about

[maiden geyser]

get C:'d lol

[hoary lion]

brother

🤦

[maiden geyser]

this guy stinks

[hoary lion]

very interesting behavior of unity found

love to shove everything in locallow

i dont know what is happening anymore

[tender river]

jump with continuation, but the continuation is linear

unless you have stack unwinding in which case it is affine

speaking of linear i should translate towers of hanoi to linear logic for fun

hashbrown just abstracts probing

is @inline broken?

[hoary lion]

Since when did torch or pip required admin terminal

well, anyway finally managed to install

[opaque sigil]

That doesn't seem right

I guess if you install it globally it would

[maiden geyser]

have you checked how long ssh takes to execute yet

[hoary lion]

its probably cause my d drive, it requires admin privileges sometimes

[amber fractal]

[knotty current]

20 minutes before i realized i was doing something wrong

[hoary lion]

the red squiggly lines and check mark should not belong together

wtf

[opaque sigil]

classic env mismatch

[rare bramble]

restarting the python LSP usually helps with that

[amber fractal]

Another common solution is starting the editor in the venv (for some reason)

[opaque sigil]

Yeah this sounds like the lsp hasn't realised the environment it's in

[amber fractal]

I love starting zed within my devenv

Best use of my time

[tender river]

btw, would a gc realistically live in lily? i feel like it shouldnt (nobody uses gc in systems programming even though it can save so much memory smh) but at the same time it will probably have some code for interoperability with lily so you can reuse lily's code for hashmaps etc

i wanted to chuck it into a "lily.experimental.gc" module or something but i guess no need since everything is subject to change for now anyway

like maybe 500loc rn

[tender river]

its around 300LoC but its wip

i read rust-analyzer-adjacent docs and i from what i read just a parser is not enough to fully utilize incremental compilation, the more complex the LSP code the more it benefits from good structure

[sage crag]

seems reasonable

[tender river]

but the cool thing is that if you implement your LSP code as pure functions it doesn't really matter how you pass data around since caching etc is separate from simple "A -> B" functions that you can plug wherever

[sage crag]

do you think it's worth trying to implement robin hood probing or should i just try and implement chunking

[tender river]

i mean i tried chunking and i didnt see an improvement, you're welcome to try though

[sage crag]

the 0x8080808080808080 pattern though

could be something

[tender river]

i remember writing something like occupied * 0x0101010101010101 in one of my attempts but idk how to do it without counting trailing zeroes

[sage crag]

https://en.wikipedia.org/wiki/De_Bruijn_sequence
possibly

[tender river]

you lost me

[sage crag]

scroll down

[tender river]

oh

wait i like this hash table approach

there are only 256 possible combinations of 1 and 0 bits

so you could easily write a functions that checks the hash table for invalid collisions

but my concern is that this would require a pointer addition and read and a bunch of other stuff and just doing it the normal way would be faster

[sage crag]

people online are suggesting that the branch predictor enables better performance on just using conditionals if the input numbers are uniformly random

[sage crag]

yeah

[sage crag]

what if i also store a 32 bit hash on the Entry struct

[tender river]

what for?

[sage crag]

just an extension of the swisstables principle, if you store the hash you need to do less computations on the input data

though i guess it doesnt really help locality at all

[tender river]

the main point is doing less reads not less computations

[sage crag]

this is true

perhaps if i really dont care about memory efficiency i can store keys and values separately for cache optimisation

[tender river]

metadata is already separate might as well

wont matter with arenas anyway

[sage crag]

ah quadratic hashing, i just realised, would want a prime-number length

[tender river]

also remove align(1) i dont think it helps at all (though didnt seem like it hurt either)

[sage crag]

will do

[sage crag]

actually with linear probing removing align 1 seems to improve it

very slightly

[tender river]

i never run benchmarks enough times to notice slight improvements so i didnt notice

[tender river]

your hashmap will be used for interning strings and sharing common ast subtrees in the lsp no pressure

[sage crag]

[sage crag]

interestingly ableos and x86 seem to prefer different load thresholds

[tender river]

did we get within an order of magnitude of rust?

[sage crag]

we are at 1.3s on x86 right now

for 5 mil elements

[tender river]

and rust is?

[sage crag]

0.2

[tender river]

so we did

[sage crag]

yes

[tender river]

i can probably implement an hbvm llvm target some day to compare rust and hblang performance on ableos its like not a priority at all though since you know lsp and all that

i feel like there should have been more commas in that message

[sage crag]

its readable

anyway these are what ive settled on

while im using linear probing at least

not good for memory efficiency

sorry that 1 should be a 2

[tender river]

isnt the former just self.entries.len >> 1

[sage crag]

so for x86 the load threshold is 0.75, and for hbvm, its 0.625

can probably write these nicer

[tender river]

so 3/4 on ableos and 5/8 on x86

[sage crag]

similar constants

but going far in either direction from those seem to pessimise the performance quite a lot

probably hasher dependant but i dont really care

[tender river]

also might depend on insert/remove patterns

[sage crag]

it will

im not considering average usage at the moment which is a mistake

but hey ho

[tender river]

just look at the other hashmap impls
the other hashmap impls: multiple 10ks of loc abstracting everything away (because they have real users with different needs unlike us)

[sage crag]

i also think they have unnecessarily large hashmaps though

they could get 90% of the performance within 300 cloc surely

[rigid snow]

good morning #hblang

[sage crag]

i feel like a big thing with this hashmap is going to be inlining

without inlining, you are jumping all over the place

[tender river]

probably, also inline hashing

though thats just one call per insert/remove so who cares

[tender river]

with hashbrown code i feel like 90% of what i read is just inline documentation

also all the different methods, entry api, etc, theres a lot that goes into "production-ready" hashmaps

even if hashmaps are super simple at their core

[sage crag]

yeahh

[tender river]

its nice when you can just ignore all that and write barely-documented code with the interface stability guarantees being "if you didnt want me to change it maybe you shouldn't have used it"

[sage crag]

i found a better threshold on both x86 and ableos

which is 77%

marginal gains

@ 200ms with 1m elements

near exact

[stray dragon]

fish

[tender river]

apparently swisstable impls often probe linearly within a bucket but use non-linear probing between buckets

well, groups, these arent exactly typical buckets

[sage crag]

makes sense

[tender river]

try at your own peril, woe is you https://github.com/google/cwisstable/blob/main/cwisstable/internal/control_byte.h

[sage crag]

bwa

[tender river]

just let go, you've done enough

from here on out its doubling the code for 20% performance

just nt worth it

you can use this as a template for the implementation of batching if you do want to go down that route

[sage crag]

i found a minor performance improvement by checking if meta is GE to tombstone

[tender river]

based on a8995f76fb134e2a61b97d58efc18264ab9b8751

[sage crag]

will read

[tender river]

it only implements batched delete and it doesnt use any of google's bit tricks

but i tried to make it as fast as i could (and failed, but i didnt make it noticeably slower at least)

[tender river]

also their hash mixer https://github.com/google/cwisstable/blob/main/cwisstable/internal/absl_hash.h

[midnight sigil]

I fixed my for loop

now it's the expected result

[tender river]

this reminds me of xkcd int randomValue = 3; // chosen by a fair dice roll

[midnight sigil]

it's kinda random though, still

unless you have perfect conditions everytime

[olive sable]

The github pagewasalmost fixed, but it compiled wrong on my laptop

Works on my machine ahh code

[gritty dust]

As long as it works on my device ahh mindset

[tender river]

@sage crag also a little thing you could do if you want to work on hashmaps a bit more is implement hashset (a thin wrapper over hashmap(K, void))

[stark needle]

This is how you develop a working SaaS

[tender river]

or a working just about anything that goes to prod, really, theres so much corner cutting

[sage crag]

they can just use HashMap(K, void)

im done with hashmaps for a while

[tender river]

good because they're beginning to scare me

[sage crag]

im getting tired of minor performance gains

i managed to get it to 1.29 seconds under optimal conditions

good enough imo

[tender river]

mhm especially considering no inlining, unoptimized asm, no autovectorization, etc

[sage crag]

40% performance improvement from when it started

when inlining comes out i'll test again

also when the foldhash bug gets fixed on x86

[tender river]

while using like two orders of magnitude less code than other impls

[sage crag]

only being 6x slower than rust on 5 million elements is pretty good

unfortunately the same cant be said of say, 6 million elements

hashmaps scale very badly

[tender river]

llvm + simd is a powerful combo

[sage crag]

mhm

and to be fair

the compilation time is sub 70ms

so, 6 times slower than rust but compiles 60 times faster

rust is a very fast language to begin with

[tender river]

the performance is close enough to leave a decent margin for unoptimal code (which i won't write of course ) so i think the interning part is covered now

[sage crag]

right

btrees are typically faster than hashmaps for insertion right?

[tender river]

erm

[sage crag]

maybe not

i dont remember

[tender river]

btrees are basically always O(log N)

for whatever you do

[sage crag]

by the way, this is 800ms just inserting

and 400ms removing

so approx 50% of the runtime inserting is rehashing

[tender river]

basically they're not as fast in the best case but they give predictable performance (and iteration order)

no hashdos and stuff

[sage crag]

i might write a special insert function for rehashing

[sage crag]

mhm

[tender river]

right rehashing is what i thought was the only remaining optimization without batching

but tombstones make it tricky since lower hash values may trickle up to the higher hash half and vice versa

[sage crag]

there were a few numbers optimisations

im also not doing optimal tombstone removal right now

so real world performance will suffer

[sage crag]

the compiler will probably handle that when inlining happens

is my hope

still though, even if you remove the cost of rehashing entirely you sit at 800ms runtime for 5 mil elements

still 4x slower than rust

pretty good though

[tender river]

just realized the interner will have to unintern stuff when GC removes it... i actually have no idea how to implement that cleanly but i'll figure it out i guess

[scarlet arch]

What the fuck :D

Also morning chat

It might be 1 in the afternoon, but ya know. Sleep cycle is fucked

[sage crag]

ripp

[tender river]

i did get an idea - weak references that use the relocation mechanism for setting their own address to null, permanently invalidating themselves, then iterating over the hashmap to remove all the nulls can be done at the end of a regular GC cycle (provided the hashmap doesnt move stuff around when removing which it wont with the current impl)

then this part is covered, and the rest is just operations on a graph (which is half of what language development is about anyway)

[midnight sigil]

I'M SO MAD RN CUZ MY BLENDER IS SO DUMB IT CAN'T DO 1 * 2 * pi * 1 * 1 / 64 PROPERLY(in a more complex way)

[sage crag]

actually i want to point out that on the same bench, rust's default hashmap is 457ms

though they are using siphash3

[tender river]

wait what did you compare to before?

hashbrown?

or just same hashmap with different hash?

[sage crag]

same hashmap with a different hash

here's running the same with the "equivalent" hasher to the one we use

[tender river]

hi shiro

[sage crag]

my implementation is poor though

[sage crag]

hashbrown hashmap is actually slightly faster than this

though it doesnt get faster if i swap the hash algorithm

so it seems its mainly an algorithm difference

in fact, its likely rust's hashmap is just hashbrown using siphash3

[tender river]

you can probably const'ize the length checks in the hasher but hashing is probably the least expensive part

[sage crag]

the hasher is fundamental to the speed of the hashmap though

if your hash x -> x

then your hashmap becomes O(n) insertion

[tender river]

yup if your impl doesnt match the original hasher then thats pretty broken

[sage crag]

i took an unrelated benchmark and slapped hblang in there

[real sierra]

#include catch_segfault.h

strlen_2(const char *s) {
 int a = 0;
 int b = 1;
 *(s+b);
 while (!segfaulted()) {
  a = b;
  b *= 2;
  *(s + b);
 }
 ptr = s + (a + b) / 2;
 while (segfaulted () || *ptr != 0) {
  if (segfaulted()) {
   b -= (b - a) / 2;
  } else {
   a += (b - a) / 2;
  }
  ptr = s + (a + b) / 2;
 }
 return (a + b) / 2;
}

thoughts

[sage crag]

if strlen is zero then it errors because you dont catch the initial segfault

[real sierra]

was assuming the header just did something to make segfaults set an error flag instead of crashing

I don't actually know any method to catch segfaults

[tender river]

i dont think its an exception its just a flag probably

[sage crag]

you can read the kernel sigsegv signal but you cant catch a segfault before it goes to the kernel

[wild wadi]

you can mask the segfault signal and define the error flag setting in a sigaction

[sage crag]

but a flag here does work

[tender river]

this is not running on linux though i think

[sage crag]

lol

[real sierra]

it's not platform specific I think

there must be a way to detect segfaults

and prevent your program from crashing from them

[sage crag]

there is a segfault interrupt on x86 cpus, but that's all you get

[tender river]

then it would be pretty broken because reading past end of string is perfectly legal and not at all guaranteed to segfault

[sage crag]

the OS handles it otherwise

and the handler for that interrupt is generally the OS

[real sierra]

yeah it's not guaranteed to work at all

[rough bloom]

fork and check if the child process dies

[sage crag]

ask me how i know

[real sierra]

if you put too much on the heap it becomes less likely

[real sierra]

wait genius?

[tender river]

by legal i mean it causes undefined behavior which is allowed to explode your PC or go back in time and kill your grandfather before you were born

[real sierra]

worry not, I'm not using the values read

I'm just using whether or not they were read without a segfault

[tender river]

doesnt matter

they're allowed not to segfault if they want to

or, again, do anything else

[real sierra]

then why do they always segfault for me

I want O(log n) string length

[tender river]

for example, there is a possible optimization in which this program actually goes back to main and runs your program all over again

[stone cedar]

Does *(s+b); even survive optimization?

[sage crag]

it should survive

[trim valve]

the joys of UB

[midnight sigil]

LET'S FUCKING GOOOOOOOO

[trim valve]

it will do exactly what it always does until you want it to do something specific

[tender river]

the standard says "UB" because "well it may be different on different ISAs and we wouldnt want to force everyone to add a billion checks everywhere", but for compiler writers it means "if you can detect this in the code then this code won't actually run so we are allowed to emit whatever we want in there"

[real sierra]

can always put it in a variable

[sage crag]

strlen_2:
        push    rbp
        mov     rbp, rsp
        mov     QWORD PTR [rbp-24], rdi
        mov     DWORD PTR [rbp-4], 0
        mov     DWORD PTR [rbp-8], 1
        mov     eax, 0
        pop     rbp
        ret

asm identical with and without *(s + b);

[tender river]

may i interest you in

error.png

[trim valve]

[sage crag]

if you assign it to a variable it doesnt get optimised out

until you pass -o1

[midnight sigil]

[stone cedar]

doesn't help if you don't use it, then you just get a compiler warning on top

[sage crag]

then the compiler just converts your program to mov eax, ret

[rough bloom]

volatile variable should work, surely

[tender river]

#1288144271823601744 message

[midnight sigil]

[sage crag]

rotate rotate

[tender river]

volatile basically means "dont cache it" it doesnt mean anything else

[sage crag]

we need to get a SAT solver on the arg images

[real sierra]

malloc something and put the value in there

then free it at the end of the program

[sage crag]

or just black_box

[real sierra]

I'd like to see the compiler optimize that

[sage crag]

compiler can actually optimise mallocs

sorry

[tender river]

shiros these days dont know the wonders of stack all they know is eat hot chip and malloc

[real sierra]

C compiler as usual is my number one enemy

[sage crag]

if you malloc a buffer and then only use it for an int it compiles to a stack assignment

[real sierra]

im ok with that

that's best case scenario in fact

[sage crag]

no, because then it compiles it to nothing

[real sierra]

what

[sage crag]

dead code elimination

the compiler isnt oneshot, the static analysis keeps going

compilers are evil magic

[rough bloom]

isn't volatile mostly meant for I/O memory where reads and writes actually have to happen because there may be side effects to them?
I wouldn't be surprised if the C standard says otherwise but surely most compilers don't optimize accesses to volatile variables out

[real sierra]

ok uhhh

how about

[sage crag]

this is true

volatile is common for memmapped io on embedded systems

[real sierra]

malloc one byte, read as int, AND with 1, OR with 1, and then realloc to that length

[sage crag]

compiler will fix it

[real sierra]

bet the compiler can't fix that

[sage crag]

compiler knows the value throughout the entire time

it will just turn it to dust

[tender river]

i think its main purpose is stuff like spinlocks but yeah you're probably right, godbolt is needed to verify it though

[sage crag]

realloc and malloc are c library functions that have special cases in the compiler

if you call mmap directly it might not

[stone cedar]

You can probably safely read to the end of the memory page but that is very platform specific? And you would still need to scan for zeroes since there could be other data including strings after it but you night be able to do this in blocks of 4 or 16 characters at a time for a speedup?

[sage crag]

actually i know this case

c always ensures strings are 1 character longer than needed

reading 1 beyond the buffer is not UB but reading past that is UB

here it would depend if you strings are stack or heap

if they are stack it will probably work

[rough bloom]

gcc explicitly stores and loads a variable with volatile even at O3
https://godbolt.org/z/cEMv9zEeP

[sage crag]

heap strings will probably fail because the kernel will give you a full page, provided the kernel doesnt decide to memprotect outside the allocated region

[tender river]

i'm tryint to load the standard but my internet is dying as always (also i learned that hurricane electric is to blame but i'm not exactly a paying customer so oh well)

[stone cedar]

memprotect only works on full pages, i think, so that should be ok

[sage crag]

like you said though, architecture and system specific

@tender river segfault hashmap

malloc a ptr, which becomes the base offset

then, for each inserted item, malloc a whole page

[warped narwhal]

iirc volatile for local variables is different than those that are global or in structures, so maybe that is the reason?

[sage crag]

then when probing, if the index is a segfault, then its vacant

[tender river]

and if it isnt a segfault, then its not our problem

[sage crag]

yes

why am i hungry

[rough bloom]

shouldn't be
if anything I'd expect local variables to be more likely to be optimized out, so this was already kind of a worst-case test and gcc still didn't touch it

C99 standard says this

An object that has volatile-qualified type may be modified in ways unknown to the
implementation or have other unknown side effects. Therefore any expression referring
to such an object shall be evaluated strictly according to the rules of the abstract machine,
as described in 5.1.2.3. Furthermore, at every sequence point the value last stored in the
object shall agree with that prescribed by the abstract machine, except as modified by the
unknown factors mentioned previously. What constitutes an access to an object that
has volatile-qualified type is implementation-defined.
tl;dr there are some guarantees but it's mostly implementation-defined

[tender river]

yay i dont have to load it

[tender river]

• read according to the rules of the abstract machine
• write according to the rules of the abstract machine
• if the type is volatile idk your guess is as good as mine

[stone cedar]

Looks like GNU libc already does this, who knew...

[rough bloom]

yeah, it implies that reads and writes have to happen, but doesn't say what they actually mean for volatile variables
which makes sense, the C standard shouldn't be concerned with how memory works on a specific system at all, portability and all that

[tender river]

at one point i had an idea of making a super cursed standards-conforming C compiler but i realized it would be impossible to program anything in it

could still be a fun project... just not an easy one

[rough bloom]

why would it be impossible?

[tender river]

possible, but there are like a billion things the C standard leaves up to implementation, byte size for one

[rough bloom]

oh

yeah that could be cursed

[sage crag]

does cstd support trinary

[tender river]

probably not

so much for "portable assembly" smh

[stone cedar]

The only thing I can see in the way of that would be the question of how a trinary XOR would work.

Theres probably more but this seems like an important one

[sage crag]

https://en.wikipedia.org/wiki/Three-valued_logic

[tender river]

i found the SETUN opcodes

will take a while to translate please wait

[warped narwhal]

you can always read to the end of your memory page, as that is the smallest size the cpu actually checks permissions on, however you will need to check/guess whether your application is using 4KiB pages or 2MiB pages as both can be used by the cpu at the same time.

(there are also hugepages at 1GiB but the OS needs to be configured to allow those specifically)

[tender river]

okay i wont bother but basically since each trit is positive, negative or neutral, its seems theres no xor built in but theres tritwise multiplication (basically "and")

i like that it has breakpoints (break instruction that stops execution and continues when you press a physical button)

debugging in 1959

[tender river]

there are no other tritwise operations btw, unless you count shift

[midnight sigil]

one more step ahead

time to think of a way to add every pixel on each column together

[stray dragon]

matrix multiplication?

[gritty dust]

What is that?

[midnight sigil]

it's 1D DFT before the summation

[trim valve]

huh neat

I should really look into how the dft works instead of treating it like a black box

[midnight sigil]

so e^(-2i * pi * x * y / 64) * ori[y]

where ori being the original image's first column

yuh

[opaque wharf]

Huh, does discord not support LaTeX?

e^(\frac{-2i\times{\pi}\times{x}\times{y}}{64})

Interesting

[midnight sigil]

cuz LaTeX can be exploited if Discord does not handle it correctly

[opaque wharf]

Yeah, but like, the basic notation are generally safe. Also, GitHub supports it

[amber fractal]

GitHub did have an exploit when it came off of LaTeX if I recall that cause correctly

[opaque wharf]

All I found is this
https://github.com/swisskyrepo/PayloadsAllTheThings/tree/master/LaTeX Injection

Oh, I found it
https://www.netizen.net/news/post/4541/mathjax-latex-exploit-github-users-use-css-injection-for-profile-customization

Then again, the mitigation already exists. Oh well

[stray dragon]

adding features that users actually want? discord would never

[tender river]

does it still count as boilerplate if its a bunch of stupid connectors but getting them right is super tricky and will blow you up

[trim valve]

@stark needle how is that arc a770 holding up

because I'm still stuck on my gtx 1660 and it really is showing its age

if anyone else has any affordable gpu suggestions that fit my annoyingly exact requirements feel free to tell me

basically looking for something that is:

• no wider than 2 slots (like the image)
• no more than 2 pcie power slots (barely considering 12vhwpr, I'd like to avoid it)
• not going to milk me for all the money in my bank account

[amber fractal]

I've heard decent stuff from the Intel B580 if you can find one, haven't daily driven intel dedicated graphics so take it with a grain of salt.

[trim valve]

current options seem to be:

• 4060 ti
• 9070xt reaper (unreleased, not sured if it properly works)
• either a b580 or a a770

[trim valve]

it is seemingly like an increasingly good option I just ideally want to have either cuda or rocm

[stray dragon]

my 6600 xt fits the bill i think

[trim valve]

oh I should also throw in that the card should be a reasonable upgrade from my current one, I don't want to have to replace it any time soon

and that seems like a vram downgrade at the very least

[stray dragon]

A770 seems to be a very small performance upgrade from the gtx 1660

hmm

[amber fractal]

Already had numbers on standby

6600xt seems to be 82% from this

[stray dragon]

i go by ratings from videocardbenchmark.net

6600 xt and B580 seem about equal, although B580 has more vram

[amber fractal]

I normally use techpowerup.com

[stray dragon]

as far as i know that site directly compares theoretical max performance to get those numbers

which seems limited

[trim valve]

the biggest issue with my current gpu is the way it runs modern games

it does manage ~60-70fps it just sits at 100% usage and gives me insane input latency

[amber fractal]

Ah, so always boasting numbers. Not the greatest for comparing day to day

[trim valve]

other than that I just need something that can:

• work nicely enough with blender
• encode / decode video

like 90% of my gaming is in tf2 I could run that on my igpu if I really wanted to

[stray dragon]

used 2080 supers seem to go for around $250 to $300 on ebay

[trim valve]

it looks like sunshine supports intel's hardware encoding so I'm in with a chance

[stray dragon]

from what i remember like a week or two ago the 6600 xt was going for around $250 on ebay

if you're wondering how those gpus compare:

vram is still limited though

hmm

[loud thicket]

Gonna be swapping an NVMe today

[trim valve]

honestly I might just go for a b580

[loud thicket]

Moving to dual boot linux

[trim valve]

it supports a decent amount of funny video encoding stuff, is very cheap and has 2x my current vram

[stray dragon]

fair enough

[trim valve]

probably not a perfect fit for the "never needs replacing" but unfortunately gpus just keep getting fatter and more power hungry

[loud thicket]

Shove a 980 in there

Those bastards live forever

[stray dragon]

6700 xt has 12gb vram and seems to be around $300 used

[trim valve]

I used to have a 970

[loud thicket]

Na must be the 980

[trim valve]

[loud thicket]

I still have one in my server used for minor transcoding

Still chugging from 2015

(Though its heavily modified from Nvidia)

[trim valve]

my only issue with 6xxx series (or older) amd is iirc they have pretty trash hardware encoding

[loud thicket]

Av1

[stone cedar]

One warning about the intel Arc cards, the legacy OpenGL performance is pretty bad under certain conditions, though that should not be relevant unless you use some truly ancient and demanding software.

[trim valve]

does solidworks count as ancient and demanding

[stone cedar]

And I mean like < 3.0 so every project started this millenia should be good, hopefully

[trim valve]

(I may not even end up using soldworks so its w/e in reality)

[stone cedar]

I work in CAD/CAM and our QA found out the hard way

[stray dragon]

this is NOT a bad deal at all

wow

[lost phoenix]

Very noice

[opaque wharf]

Ancient... depends on which version. But demanding, yes. VERY

But if you only need simple CAD, Fusion360 is free for personal use. BricsCAD has one-time purchase, and onshape also has free for personal use

[trim valve]

yeah I'm probably gonna toy around with f360 for the time being, given I can snag an educational license from them

but also I could nab a solidworks license from my uni and learn it for the shits and giggles

[opaque wharf]

The important thing is to learn the fundamental. If you've mastered that, even using FreeCAD would not be that impossible

[trim valve]

yeye fair

[opaque wharf]

But if you need to get a job, then get a certification. The industry (at least in my country) is still using old SolidWorks and AutoCAD

[trim valve]

a certification would be neat but I haven't super looked into it

[stark needle]

Works

[trim valve]

I have no real background in engineering I just happen to be less useless with it than the average person

[opaque wharf]

Ahh, then yeah. Fusion360 is super easy for beginner. Onshape also compete in that regards for more simpler design, and less resource hog since it runs on the browser

[trim valve]

I don't really mind a resource hog, plus the way onshape's free tier works sketches me out

I'll stick with fusion for now 🙏

[stray dragon]

there's also the 1080 ti with 11gb of vram

[trim valve]

I probably just need to find some kind of course / guide / cert / etc. to properly learn the software instead of just going off vibes though

[stray dragon]

from my experience with getting a solidworks certification all you really need to know is how to use a few important features and how to make simple parts in a limited timeframe

the courses for this stuff are usually "here's a sheet diagram of a part, go make it in CAD" and maybe a guide on how to use a feature that will be useful for that part

[sage crag]

t

[stray dragon]

hi konii

[sage crag]

hi t

[trim valve]

understandable tbh

from what I've heard CAD isn't super hard

[stray dragon]

it really isn't but you have no idea how much i've seen others struggle with it

[trim valve]

I am well aware from participating in a group project where someone vibed their way through fusion

its never a good sign when you manage to make a step file so bad onshape can't open it

[stray dragon]

genuinely how

[trim valve]

I am trying to find a screenshot of the model

but from what I remember they had some very messed up surfaces

best part was the triangle pattern they tried to cut into the surface which created ~1000 constraints and froze fusion every time you did anything

[stray dragon]

bruh

[hexed grove]

What's up yall

[trim valve]

don't question the design it never made it out of cad

but there has to be a better way to make that pattern in cad than copy-pasting a single triangle everywhere lol

[stray dragon]

that seems pretty normal?

[trim valve]

[stray dragon]

you are correct

[brisk laurel]

Can someone please check this isnt an IP grabber? i need it ASAP thank you

[stray dragon]

forgot what the feature is called but it can repeat a selection of lines at regular intervals

[trim valve]

my issue wirth this was the abuse of carbon fibre everywhere

[trim valve]

idk what you mean by that but I can open the image just fine

[stray dragon]

there's also a rotational version of it that rotates the shape around a point

[trim valve]

yeah I kinda assumed that would exist somewhere

[stray dragon]

ah it's the "linear pattern"

[trim valve]

no big gay pikachu :(

[stray dragon]

it's just a jpg? can you even fit malware of any kind on that

[amber fractal]

It's a jpg that isn't embedding

that's a red flag

[trim valve]

i mean it happens

[stray dragon]

you can open the link in browser

[trim valve]

discord is great at randomly not embedding stuff

[amber fractal]

I do agree it does

[rough bloom]

you could mess with the metadata and/or use a parser/decompressor vulnerability, but yeah, it's rare

[amber fractal]

Or it could not even be a jpg file, seen that before

[tight tinsel]

tf they gonna do with your ip even

[stray dragon]

i see

[trim valve]

it probably just didn't embed because of the insane resolution

[amber fractal]

Yep rez

[trim valve]

who tf decided a phone should take 200MP images

[opaque wharf]

Yet someone manage to send gif of full minecraft movie

[stray dragon]

thought it was a full video

in tiny resolution, to be fair

[opaque wharf]

Yeah, but its still around 65MB

[stray dragon]

it's probably laggier to embed a 50MB image with crazy resolution than to embed a 100MB video file in 200p or whatever

it's not like the entirety of the video is displaying at once

[trim valve]

discord gif handling got a lot better after they moved to react native iirc

[opaque wharf]

But to be fair, its due to how shitty gif is for actually moving stuff

[opaque wharf]

True

[trim valve]

I remember when any gif over like 1MB would bring the kotlin app to its knees

[stray dragon]

yet they still haven't figured out a way to have the gifs in favorites show up in a recognizable form instead of doing this until the gif is fully loaded

[trim valve]

thank you for reminding me I have saved gifs

but yeah its a pain

[stray dragon]

would be absolutely insane if they had thumbnails stored locally or something (they'll never do this because it would only benefit users and would take a bit of work from a developer)

[trim valve]

would be insane if they used an imageset that loaded a static webp first

[stray dragon]

that too

[amber fractal]

Samsung https://news.samsung.com/global/samsung-brings-advanced-ultra-fine-pixel-technologies-to-new-mobile-image-sensors

[trim valve]

I think the issue is probably due to the mildly scuffed way discord handles the saved gifs menu

[sage crag]

@tender river thought you would like this one

[trim valve]

not suprised, but still pointless on a phone

[amber fractal]

I find the 8k camera useless on this one, let alone higher

[trim valve]

my phone allegedly does 64MP but its just a waste of storage

I have zero faith in phone sensors anyways, pixel count barely decides their quality

[amber fractal]

I shoot 4k and RAW, because I do photography sometimes

[stray dragon]

clearly if you can't zoom in close enough on an image to see the textile pattern of a carpet it's too low-res to consider buying

yet it still can't show readable text properly LOL

[trim valve]

tbh samsung still handles text better than iphones in my experience

[amber fractal]

One of my previous phones had very good text quality, shame I can't use it.

[brisk laurel]

Okay cool

[brisk laurel]

What does text quality even do?

[stray dragon]

but yeah the image is just too big to load as an embed

[stray dragon]

makes text readable

[amber fractal]

in terms of camera quality

makes text readablw

[trim valve]

can't find a good comparison though 😔

[prime ridge]

64 mp is absurd

[stray dragon]

all i know is that my samsung A51 can take pictures of my homework that have clear and readable subscripts and superscripts

good enough

[trim valve]

my a52 is unsuprisingly very similar

[stray dragon]

no way

[trim valve]

though I'm pretty sure my OIS broke

[amber fractal]

If it could be used to average out stuff instead of raw res then I'd say it is slightly better.

[trim valve]

it probably is in the lower resolution modes

[amber fractal]

In hardware instead of software, less to process in theory.

I'd need to look into it if I want to make more points for or against it.

[trim valve]

tbh if I'm doing anything where I care about image quailty I'm not using a phone anyways

they lack behind so much in other ways

[amber fractal]

I don't have issues on the software side, just smol sensor

[trim valve]

yeah

and annoying focal lengths

[amber fractal]

Not much precision you can get at all

[trim valve]

still, they can take nice photos

but imo a proper camera would look better here too

[gritty dust]

Lads I got my stupid camera mount thing to work pog

Now I only need x and y not the z axis

[trim valve]

also these fuckass lens flares

like yes I know a real camera also gets lens flares but not to this degree

[trim valve]

[gritty dust]

yay, its a stool with a wood slab clamped onto it with the camera at the edge of the stool, I got my yolo model to actually run well so now it's finishing the coordinate translation code yippee

[stray dragon]

the sun is attacking

[trim valve]

solar flares if they were epic

[sage crag]

@tender river if you can submit some issues of things you have been experiencing that would be great

its hard to categorise them in just one channel

this in hblang repo

[loud thicket]

tf is that for

[scarlet arch]

Hi Wheelz

[noble zodiac]

îts 2025 and I'm implementing a insertion sort

[faint sandal]

https://localmess.github.io/

tldr your Android browser tells Meta-owned apps what websites are being visited

[opaque sigil]

Sick

[ruby timber]

[dry charm]

UPDATE: As of June 3rd 7:45 CEST, Meta/Facebook Pixel script is no longer sending any packets or requests to localhost. The code responsible for sending the _fbp cookie has been almost completely removed.
LMAO

disclosed and instantly pulled the plug

[gritty dust]

Does anyone have an image that I can use as the banner for my programming portfolio

[sick owl]

[stark needle]

nuclear reactor

[trim valve]

(you're gonna have to be more specific about what you want if you want anything meaningful)

[sage crag]

no computers

[trim valve]

oh

[sage crag]

i cant tell what the most insulting part of this is

ive never even touched a computer

[dry charm]

I love playing incremental games!

[tender river]

stateless
anyway seems like a question by a person who just learned what fp is and hasnt actually used any fp language, their question would be half answered if they learned a fp language (or even concepts like tail recursion and fold) and full answered if they actually did some programming that required lots of data processing (which is just about any program from UI to databases) and ended up having to write fp in whatever language they were using at the time, whether it be JS or C++
tl;dr the primary use for functional programming is writing neuro sdk mods in F#
(sorry if you wanted an actual answer and this was a bit handwavy)

[tender river]

gotcha, dont really remember anything but i'll keep that in mind

[sick owl]

[sage crag]

i already submitted the issues regarding those u8 binops

if anything else comes up, a test case or some way of reproducing bug is required

though you dont need to be told that

[stark needle]

codex

[sage crag]

what is codex

software engineering agent

that's me, im software engineering agent

[stark needle]

can you do my job for me for 20$/month subscription

[sage crag]

yes

[ruby timber]

so um

structs are just classes

in c++

that's

okay

[sage crag]

no inheritence

not classes, sorry

[ruby timber]

Ah

[sage crag]

[ruby timber]

Actually, that's better for me-

[sage crag]

structs are infinitely better than classes anyway

[ruby timber]

I'm coming from C like

[rough bloom]

IIRC they had some difference in the default visibility of attributes, but yeah, same thing
(referring to class and struct keywords in C++)

[ruby timber]

yeah, I would have been surprised if there weren't at least some differences

I don't know why structs with member functions feel so icky (I know why, C brainrot)

[rough bloom]

do they actually not have inheritance?
I'm quite sure I used C++ classes with inheritance at some point but maybe I'm just going insane

[sage crag]

structs do not have inheritence

classes have inheritence

[ruby timber]

@grok is this true?

[sage crag]

methods are too much for the c programmer's mind

all struct functions must take a struct ptr

[ruby timber]

yes

I discovered the concept of a static getter and felt like a caveman coming out of the lights cave

I'm still such a beginner it's scary sometimes

[rough bloom]

gcc disagrees
https://godbolt.org/z/acfqvsarY

[ruby timber]

the stackoverflow overlords have spoken

[sage crag]

c++ is clearly doing it wrong

not the first time

[ruby timber]

Time to switch to c--

[rough bloom]

[vestal prism]

WHAT IS THAT THING

[rough bloom]

D

[vestal prism]

kill it with fire

[sage crag]

ropher

[ruby timber]

don't worry about it

[ruby timber]

Isn't there E somewhere too

Oh oh

[sage crag]

and F and G

also H

[ruby timber]

One project of the advanced part at my school has us building a B compiler

[sage crag]

most of them exist

[ruby timber]

Alphabet stack soon

[ruby timber]

And the next one after that is a C compiler

The B spec provided is glorious

[rough bloom]

alphabet ouroboros quine

[tender river]

method pointers

[ruby timber]

function pointers

...

pointers

[tender river]

i remember some incredibly cursed things to get some particular kind of method pointers

[sage crag]

i have (programmatically) created a test that checks every binary op, every unary op, and every upcast under permutation

[tender river]

or was it because method pointers can't be trivially casted to function pointers in C++

[sage crag]

this is not a neuropoghd time

[tender river]

i'll go find that code

[tender river]

it is

[sage crag]

it has broken something

i do not know what

[tender river]

binsearch time

[sage crag]

ueaj

i found it

v52_0: u32 = 4
v52_1: u32 = 1
v52_0 >>= v52_1
if v52_0 != 2 return 52

[tender river]

no way my old C++ code uses tabs

[sage crag]

youre not supposed to be here

[tender river]

my fascination with programming languages started after writing this code and realizing it fucking sucks

this was uuh in 2018

this is real chayleaf lore

[sage crag]

wait it works

why does it crash the compiler

it literally works

[tender river]

too much stack values?

some specific register allocation may trigger it

[tender river]

this may be just too many variables in scope

[sage crag]

oops

is that a hard limit do you think

or is this just a compiler skill issue

[tender river]

you can replace appendAssumeCapacity with append

probably

or maybe you cant

[stark needle]

as an ai language model

[sage crag]

lets find out

[sage crag]

no, i dont have an allocator

[tender river]

surely just pass it in there nothing could go wrong

[sage crag]

this feels like a bad idea to change function args of

im just going to make an arena and not deinit it

[tender river]

add it to struct initialization

but i guess you might have to have it recursively which would suck

[sage crag]

:mhm:

i dont want to fiddle with this

[tender river]

just separate into multiple functions and report and move on

[sage crag]

i found a canary

got it working

i dont know

this is very annoying. all other types properly upcast with both lhs and rhs, but not i64 and uint

v361_0: u64 = 5
v361_1: uint = 3
v361_r := v361_0 + v361_1

there's no reason this cant upcast into u64

[tender river]

oh i didnt even know theres explicit u64

[sage crag]

yeah

uint and int are meant to be ptr width

u64 and i64 were added because we complained

i still prefer sN to iN though

s32 yes please

sint is less nice though

[tender river]

shorten it to sin

[sage crag]

[tender river]

if functions can be fun why cant signed integers be sin

[sage crag]

we have a function called sin

[tender river]

clearly this means we need contextual scopes this is not at all a bad idea

rust kinda has them btw, proc macro and trait can have the same name

[sage crag]

[tender river]

ml has them with module an type potentially having the same name

[sage crag]

what if i just alias ints as signed ints in lily

s32 := i32

[tender river]

but yeah not a good idea unless you design a language from scratch with that in mind

[tender river]

i've long realized that pointless aliases are best avoided

[sage crag]

yes im aware

it was not a serious suggestion

[tender river]

no you should do it

[sage crag]

no.

[sage crag]

always something in my way

[tender river]

nope no idea how to fix that one, other than "just fix allocBytesWithAlignment surely that's the wrong one"

[sage crag]

what if i just remove the assert

surely its a good idea

segfault

im just going to upload this

and mlokis can figure it out tomorrow

[tender river]

yup its like a billion minimal repros

[sage crag]

maybe it will inspire him to actually test this in the compiler's fuzzer

[tender river]

a maximal repro

[sage crag]

its the minimax repro

[tender river]

now that i think of it creating a testing framework sounds like a good idea... it doesnt sound that hard (autogenerate code for every .hb file that has a tests := {) but would you want tests to live with the code or in a separate file

[sage crag]

i generally dislike unit tests being in source files

perhaps it seems silly, but i prefer the tests to be centralised

[tender river]

i dont mind either way

[tender river]

i think if theres a nice correspondence between modules and tests tooling is easier to develop

for example, i could add an lsp feature to run tests for a specific module on save or on demand

[rigid snow]

can someone explain to me why this works if it even does, i skipped math lectures

let digits = (value.abs() as f64).log10().floor() as usize + 1;

[sage crag]

log_10 of a number is the number of times you need to multiply 10 to get to that number

take the floor, which puts that to the nearest whole number

and you get the number of 10s to get to the nearest power of 10 to that number

which will be the number of digits in base 10

[rigid snow]

i know what a logarithm is

but

[olive sable]

my brother's laptop has a metal frame, and metal hinges. but lenovo in their infinite wisdom decided to make the piece connecting the 2 from plastic
it broke

[sage crag]

then you know how the algorithm works

[rigid snow]

yeah it makes sense actually i'm just slow

thx

[sage crag]

[tender river]

in general, you can use a logarithm of base N to get the number of digits in a number with base N

[sage crag]

base 0

[tender river]

this reminds me of the whole debate on whether 0 should be a natural number

[sage crag]

its true if you take log_0(x) as infinity

[tender river]

different countries use different definitions of natural numbers

[sage crag]

i like to include 0 in natural numbers

because it distinguishes them from positive integers

[sage crag]

an actual counterexample is base 1

111 (base 1) == 3 (base 10)
but log_1(3) is undefined

[tender river]

111 isnt a base 1 number

0 is the only base 1 number

[sage crag]

you are in fact correct

[tender river]

when did they teach you nondecimal bases btw

in our case it was primary school but i dont remember if it was 2nd, 3rd or 4th grade

[sage crag]

hm

[worldly plank]

"Nondecimal bases"?

[sage crag]

not base 10

[sage crag]

well, i believe the first time would have been IT

non-decimal bases are not really a concern for UK maths students until A-levels

[stark needle]

@sage crag 50M parameter llm schizo

[sage crag]

look at me im a 50m parameter llm

[rigid snow]

holy, did you go to a special school or what
binary as part of IT in like 8th grade for us

[sage crag]

meh

kids are able to learn it as soon as they know what exponentiation is

[worldly plank]

We had Roman numerals in 3rd grade, binary in 8th

[rigid snow]

don't worry nobody bothered actually learning it anyways and cheated on all tests

[sage crag]

roman numerals surely dont count

[tender river]

i mean i went to a montessori school so thats why i guess

[rigid snow]

oh yeah makes sense then

[stark needle]

hi

[rigid snow]

[sage crag]

yeah i pushed the test

im just gonna let mlokis deal with that

if there's something wrong with it he will yell at me about it

[tender river]

@sage crag what do you think of this

[sage crag]

its useful to some people

i dont think i would ever use it

though it would get used in ableos if it existed

since ableos is very gui-centric

[tender river]

well, adding it later shouldnt be a problem

[sage crag]

hm

i was thinking about something

but now its gone

[tender river]

head empty

[tender river]

do you want to standardize some kind of tests directory format for tests that live separately from the code

i can offer editor actions that run those at least

though, i dont use them myself when developing for other languages

theres a lot of lsp features but i kinda only use renames and error highlighting

hard to think about editor tooling when you barely use it

i'll just forget about fancy features for now and focus on building the dependencies of a proper lsp impl

[sage crag]

it does not matter

unit testing shall be done or shant be done whether or not the lsp has support for it

thanks for the consideration but the extra work is just not meaningful to anyone's usecases right now

it would probably be important if hblang was more popular

[tender river]

yeah i just thought i probably need quite a lot of tests for the libraries i write because of the language's current state

and that just writing the code and checking it works by writing something in main and moving on is not a good iea

[sage crag]

at the moment im taking the approach that documentation and testing for my code is less important than volume of code

the compiler is so unsound its impossible to use the results of your tests for actually testing things

[tender river]

i guess thats the way to weed out the most critical bugs

[sage crag]

mhm

it will likely require a rewrite at some point

[trim valve]

do I stay up late waiting 3 bajillion years for bitcoin to transfer or sleep

[sage crag]

you should download more bitcoin

[trim valve]

[tender river]

do you actually need to wait for it or are you just anxious for no reason

[trim valve]

I need to transfer it out as soon as it finishes

or well some of it

[tender river]

surely the recipient can wait

[trim valve]

join moving coinbase -> proton -> elsewhere because I'm getting sick of coinbase being up my ass on every button I click

(I'm sure there's better options)

[tender river]

proton as in protonmail/protonvpn? do they offer crypto stuff now, or is there some exchange or whatever named proton

[trim valve]

proton offer a bitcoin wallet via proton wallet

I'm sure I could use a local wallet instead

but this is nowhere near enough money for me to care about proton having the keys to my wallet

[tender river]

personally i just store my wallet in my keepass db which is replicated between all of my devices so its relatively safe in there

if you get access to my laptop while its unlocked you can steal 1 eth

[trim valve]

[sage crag]

can i steal your laptop while it's unlocked

pleaes

[rigid snow]

lol just remember all the words ez

[trim valve]

[tender river]

probably not too hard since i rely on swayidle for locking but you'd have to climb up to the third floor without anyone calling the police

[trim valve]

it went through

but somehow only on the mobile app and not the website yet lol

[sage crag]

no i mean like

just let me steal it

you know

[tender river]

sure, go on

[sage crag]

thanks

[tender river]

be warned, the accursed information stored in my laptop is a heavy burden the bearer of which shall pay a heavy price

[trim valve]

who would've guessed proton -> elsewhere happened in seconds compared to the literal 10 minutes coinbase -> proton took

[stray dragon]

is it a one-time payment or a subscription fee

[tender river]

one time but multiple times

[stray dragon]

just say subscription fee

[tender river]

no its multiple times but its bounded

[stray dragon]

😔

[tender river]

its not a subscriptio