programming

[opaque wharf]

For one person

[opaque sigil]

My x86-64 is kinda rusty, what is rbx used for here

[stark needle]

will y'all be sad if i explode

[sage crag]

rbx is usually a general purpose register for indirect memory access

the main thing here is the amount of popping from the stack that is happening

i do not trust it

[opaque sigil]

I mean

[sage crag]

this is the program

[opaque sigil]

Doesn't the calling convention say things should be put on the stack

Past a couple arguments

[sage crag]

yes, but the a[b] = x should compile down to (a.ptr + b).* = x

which it isnt

[stark needle]

Turing completeness of transformers

[sage crag]

compiler optimisations are nonexistent

[opaque wharf]

As the wise beard said, try other profiler then. Maybe intel v-tune would help?

[opaque sigil]

Idk I haven't slept yet my brain is kinda mush rn

[sage crag]

i barely have enough symbols to objdump

[opaque sigil]

gl though

Also take more samples smh

[sage crag]

speaking of objdump, here is the objdump of the binary

[sage crag]

i have the polling rate set to 50000

not my fault the program is too fast

...i guess i should increase the loop length

[opaque sigil]

100

[sage crag]

i forget im not in a vm anymore

[sage crag]

i changed it to 100 million

m data

3.3kb binary is ok for this

still smaller than zig hello world

[tender river]

is it possible that it measures the next instruction?

i mean i guess you'd think they tested it properly and stuff

[opaque sigil]

Now this looks a lot more reasonable already

[sage crag]

i would hope perf works correctly

if not its mildly concerning

[tender river]

no idea how polling works in perf

[sage crag]

this is true

its only going to give a rough heat distribution

i set the polling rate to 50000 when the default is like 1000

so theoretically its accurate

interestingly there is a lot of time spent in Arena.realloc too somehow 🤔

i guess that makes sense if im copying a hundred million elements

[opaque sigil]

Constant resizing would do that yeah

[sage crag]

im already multiplying capacity by 2 each time

amortisation already optimal

[opaque sigil]

Preallocate

[sage crag]

either way its not really a big deal to worry about this while the compiler is doing nothing

[sage crag]

i have a function for this in fact

very smart

without preallocation

[opaque sigil]

[sage crag]

with preallocation

[opaque sigil]

I wonder, how does one even figure out which instruction to attribute a sample to

[sage crag]

lots of guessing

i like looking at it even though its not perfectly accurate

[opaque sigil]

I guess there's nothing stopping you from inserting some information

And treating each instruction as a block

[sage crag]

382 mb vec

[opaque sigil]

Are the other 4 functions from libc

[sage crag]

nope, libc is not being linked

i do not know what they are

probably some syscall or page fault handler

[opaque sigil]

Yeah I guess

[sage crag]

maybe they are in the objdump

[tender river]

i only worked with stuff like that on windows

[opaque sigil]

Only one way to find out

[sage crag]

or more likely its just absolute jumps being inserted in the compiler i guess

[tender river]

but i could literally inspect every register in the API

[sage crag]

you can do that with perf stat but its not as useful as perf record

[tender river]

so just record the eip

[opaque sigil]

Hmmm

[sage crag]

m

im not interested in making the vec faster for now i think

inserting 100 million elements in 220ms is surely enough for anyone

also notably goes up to 280ms if you dont preallocate

[opaque sigil]

What's next

[sage crag]

useful things that arent simple data structures hopefully

file I/O for one

[opaque sigil]

[sage crag]

unicode characters & strings for another

im procrastinating it a bit though

[opaque sigil]

unicode

[sage crag]

its ok i have most of the useful codepoint handling stuff written already in ableos

for font loading lol

[opaque sigil]

Ah nice

[tender river]

i wrote some cursed code that converted stdin and stdout to be nonblocking to play with epoll but ran into some compiler panics with inlining

[sage crag]

why is there async in the most sync language

[loud lion]

guys, how good are you at coding you would say?

[opaque sigil]

No

[tender river]

no

[sage crag]

nuh

the scale is too large to be meaningful

and too vague

[tender river]

async is secretly just state machines

[loud lion]

can you say hack into X headquarters?

[sage crag]

yeh totally

[tender river]

i cannot confirm nor deny having already done that

[sage crag]

stateless async when

[loud lion]

SeemsGood

[sage crag]

need a smarter compiler

[tender river]

stateless code too

[sage crag]

stateless binary

[tender river]

eax? we dont do that here

[sage crag]

e�x and r�x are my favourite registers

[loud lion]

guys, do you know lean4?

[tender river]

i only know lean the proof checker

[loud lion]

yep

[sage crag]

i know of it

[tender river]

oh it looks like lean4 is its implementation

[loud lion]

want to learn how to contribute to the mathlib in lean

to get started https://adam.math.hhu.de/#/

[opaque sigil]

Idk if I should be relieved or disappointed that this wasn't a deez nuts joke

[loud lion]

I wouldn't dare xD

[opaque sigil]

Missed opportunity smh

[loud lion]

Do you know DN? 👀

[tender river]

something relevant to hblang https://github.com/leanprover-community/mathlib4/blob/master/Counterexamples/Girard.lean
(not actually relevant because hblang doesnt make any guarantees that are broken by type: type)

[loud lion]

What is this?

[tender river]

a proof that type being the type of itself allows you to construct a contradiction, thus proving anything

[sage crag]

ill look in a minute

[loud lion]

so they made it into a theorem?

something something they can't add it into an axiom

[tender river]

its hard to prove

[tender river]

you dont have to

[loud lion]

Aren't there big foundational problems with set theory already with universes?

[rigid snow]

why are we talking about type theory again

[tender river]

i wouldnt say there are foundational problems, set theory still works fine for many applications, it's more that you have to be careful of how you use it

[olive sable]

does this both execute the indented code?

[opaque sigil]

Yes

[olive sable]

huh

interesting

[loud lion]

Does type theory solve all the issues present in set theory?

[rigid snow]

it's basically labels and the switch statement is goto

that's how i understand it

[ruby timber]

Yeah

Each case isn't its separate scope, that's why when once case executes, if it's not followed by break, the next also executes

[opaque sigil]

helps that this is often how it ends up being implemented too

[olive sable]

hmmmm.
ok

[opaque sigil]

treat it as jumping to the corresponding value and executing from there on

[dry charm]

Also the reason why in multiple languages even though you declare a variable in a case branch, it still complains about duplicate variable name declaration

[rigid snow]

match statements in c++ when

[dry charm]

switch(something) {
  case "a":
    int num = 4;
    break;
  case "b":
    int num = 3; //ERROR: Already declared above
    break;
}

Fun

[olive sable]

whats the diffrence between match and switch?

[rigid snow]

match is what you expect switch to be, each case is a separate scope

[olive sable]

ahh

okay

so is it better to use match?

[opaque sigil]

it doesn't exist

you don't have a choice

[rigid snow]

(in c++)

[olive sable]

oh

fuck

ok nvm

[dry charm]

which lang

[olive sable]

C++

[dry charm]

[opaque sigil]

you can make it work but for the sake of everyone's sanity let's say you cannot

[rigid snow]

macros?

[dry charm]

#define match switch

[olive sable]

nah

[opaque sigil]

you can have a variadic match function i guess

no macros needed

[rigid snow]

oh hell nah

[olive sable]

im having issues cuz the tutorial did everything in 32 bit but i need it in 64 bit

need to update my sdl stuff

[tender river]

there are ways of solving set theory's problems while staying within the framework of set theory, but in a sense yes, some type theories were designed in a way that allows you to avoid paradoxes, imo you should prefer one over the other based on their other properties (like that type theories are easy to use for automated theorem proving, while set theories are convenient for encoding arbitrary relations or predicates)
in the first place, it depends on what you want to reason about

[opaque sigil]

ts-pattern calls

[tender river]

this wont even work unless you're in C# because chances are something's address doesn't match "a"'s address

[dry charm]

Most of the languages use Equality check for switch no?
I'm 100% sure Java and Kotlin works

[tender river]

well it wouldnt work in C

[olive sable]

uh

[fast pagoda]

rip

[hard shale]

what issues?

[noble zodiac]

uhm....

[rough bloom]

rip previous conversation

[olive sable]

what got deleted?

[noble zodiac]

that was a little too aggressive

[fast pagoda]

20 messages lolol

it DID remove the spam

there were only 11 of those tho

[sour harness]

some aggressive cello pounding got deleted

[olive sable]

idk wht that means but sure

[tender river]

rip

i think i'm clear to resend this? https://blog.pragmaticengineer.com/builder-ai-did-not-fake-ai/

[olive sable]

is that automated or did a mod do that anually?

[safe path]

it wasn'tm e

[tender river]

and there was also https://math.stackexchange.com/questions/5074503/can-pa-prove-each-goodstein-sequence-can-be-proven-in-pa-to-reach-zero/5075056#5075056 sent

[noble zodiac]

very good read, also contains lisp

[olive sable]

lisp sounds like the programminglanguage of snakes

[noble zodiac]

nah thats lisssssssssP

[rough bloom]

speaking of programming languages, Google released their incident report about how they took down half the internet yesterday
https://status.cloud.google.com/incidents/ow5i3PPK96RduMcb1SsW
it was (probably) Go's fault, they didn't handle an error and dereferenced a null pointer
wouldn't have happened with Rust

-# there's also a few other factors at play there but let's ignore those and just blame Go

[noble zodiac]

no language will save you if you just dont handle errors

[rough bloom]

but they can try to save you from accidentally not handling errors

[fast pagoda]

the reports i read did mention that natasha was a real tool but that it was not well liked or something
could be good based on the description though
works a lot like how i want my kubernetes agent thing to work lol

[rigid snow]

i hate golang error handling with a passion, how do people think it’s somehow better than exceptions

what do you mean you manually have to check for each error

[fast pagoda]

This policy data contained unintended blank fields. Service Control, then regionally exercised quota checks on policies in each regional datastore. This pulled in blank fields for this respective policy change and exercised the code path that hit the null pointer causing the binaries to go into a crash loop.

wish i were a fly on the wall in the room this button was pressed

[olive sable]

its a opengl
idk why it went to 1.0 tho, it should be 2.0

[fast pagoda]

classic pink default

[rigid snow]

compiling straight to wasm

who needs native anyway

[fast pagoda]

For some customers, the monitoring infrastructure they had running on Google Cloud was also failing, leaving them without a signal of the incident or an understanding of the impact to their business and/or infrastructure.
double OOF

[rigid snow]

also if you think using gcp is a good idea you have to know google does not dogfood off gcp at all

what they use internally is very different

that’s why no google products had an outage

maybe gemini

[fast pagoda]

i would imagine that's largely because most of it existed before gcp

[rigid snow]

if i was a gcp customer i would imagine they offered their existing developments on gcp

[fast pagoda]

they definitely have it integrated at least partially

[opaque sigil]

that's a feature not a bug

[fast pagoda]

but the backend for gmail for example wouldve been in place for years

not much reason to onboard it to the "platform" as a whole

probably just fr that reason too lol

backup

i know for a while amazon's services were separate from aws as well

they have been trying to do otherwise with new stuff

i think theyre mostly all aws now though
will issue not skill issue

[shadow sinew]

The problem is that you're declaring num twice. What you probably want is to declare int num; before the switch statement and then assign either 4 or 3 to it with just num = 4 etc.

BTW you can't switch on strings like this in C++ since you'd just compare the string pointers not the content. This might seem to work in trivial code if you for instance assign a string literal "a" to something since the compiler will figure out that it can use the same memory region for both "a" literals, but it'll fail if you for instance read user input into a buffer and try to compare.
The easiest way to deal with this in C++ is most likely to do an if-else with std::string comparison.

[dry charm]

That is the point I wanted to make, that it does not create a sco0e, hut it is a goto underneath.
In higher level languages it does work, causeost of the use equality function check instead of address checking

[noble zodiac]

switches in C++ support fallthrough so you can end up in multiple different cases

[opaque sigil]

they should just add pattern matching in c++29

[noble zodiac]

but you can just create a scope yourself if you really want to so its wahtever

[dry charm]

Yes, cause it is a goto, not a scoped if thingie

[opaque sigil]

okay but that's boring

[dry charm]

In C# I just do the cursed scoping:

switch (something) {
  case "a": {
    var num = 4;
  } break;
  case "b": {
    var num = 3;
  } break;
}

[opaque sigil]

hey, if it works it works

pretty comfy to type too

[tender river]

comfy

then again its C# i'm expecting too much

when i made some unity mods i did start with C# but it very quickly pushed me out to F#

[dry charm]

it became natural as well

it's like a weird lambda

[noble zodiac]

often times thats exactly what it is

[tender river]

explaining lambdas to imperative programmers

okay imagine a code block with lazy evaluation

[noble zodiac]

bold of you to assume that people know about lazy evaluation

[tender river]

its ok not all of us understand that the principle "you can use this value as many times as you want" can naturally be extended to "you can pretend this value never existed if you want"

[winged kraken]

hello

[sage crag]

ut8 string

[opaque sigil]

[sage crag]

just to be clear i am actually parsing it and not just pretending

[opaque sigil]

i wonder

would it be possible to detect the base of the numbers you're trying to print

probably not right

well, at least not with help from the compiler

[winged kraken]

who like the web developer and use javascript ?

[olive sable]

i use JavaScript, but i dont write it so

[knotty current]

i wrote javascript once and i dont want to write it again

undefined is so wrong

why is it not a NameError

[sage crag]

technically

i guess so

but it would mean wrapping them

so, not happily

log.print is mean to take format arguments but i havent done it yet

lossy conversion

[winged kraken]

actually Iam trying to learn it

[olive sable]

le triangle
i forgot how much shit you need to do with the vao's and vbo's.
pyOpenGl had those but i switched to zengl for web support and zengl doesnt have those. i need to relearn them

[noble zodiac]

my condolences

[olive sable]

nooo i have bad parenthesis in my code

i shouldnt have copy pasted the vertices

[opaque sigil]

if you have questions about the language whenever feel free to just ask, there's plenty of people here who know more about it than they probably should/would want to

[winged kraken]

how do you learn , do you use websites or youtube?

[olive sable]

you can, but imo you learn the most from trying to use it and problem solving

it is handy to watch a video on it first to get an idea of the syntax and how stuff works

[opaque sigil]

i'm a big fan of jumping down the deep end and forcing myself to learn that way but that doesn't work for everyone

[olive sable]

ye same

i barely know C++ but im already trying ot get it to work

[sage crag]

which is why everyone should start with hblang

[winged kraken]

c++ they learn it to us at college

[olive sable]

tbh when lookign at hbland i couldnt figure out how it worked

[opaque sigil]

one downside of trying/using all kinds of languages is i keep getting the syntax wrong

[olive sable]

im not in college tho, and we'll be doing C# when i am next year

[sage crag]

why is this the same size as the hashmap code

main := fn(): uint {
    arena := lily.alloc.Arena.new()
    defer arena.deinit()

    // emoji := "😊asd"
    emoji := u8.['h', 'e', 0xC3, 0x28][..]

    string := lily.collections.Utf8String(
        lily.alloc.Arena,
    ).from_utf8_bytes_lossy(&arena, emoji)

    return 0
}

[opaque sigil]

wdym

[frozen igloo]

https://youtu.be/wm_YzdiAXm8?si=NiUMvNcSbgF3keVw

[nocturne olive]

Silly

[opaque sigil]

also i assume utf8 being under collections is only temporary

[sage crag]

binary size

[opaque sigil]

right

oh

stdlib?

idk how the compiler works, does it just throw in everything

[sage crag]

it should still be smaller

[sage crag]

no

well, a little bit

[winged kraken]

c# good language it's take fourth place globally

[sage crag]

utf8 codepoints maybe

for utf8 strings, they require an allocator, so they go in collections at least for now

[opaque sigil]

i guess fair enough

[sage crag]

not interested in fragmenting it much more than it already is

[opaque sigil]

can put it with a proper string type into its own module later though

[sage crag]

right now im just aliasing them

forget OwnedString it doesnt exist any more

[opaque sigil]

where's my 7(?) string types

i don't remember how many rust had

[olive sable]

ah yes. this may or may not be what the python code looke like on the right

[sage crag]

in hblang you have []u8

[opaque sigil]

probably more

[sage crag]

and Utf8String if you count lily

[opaque sigil]

[sage crag]

i will probably make OwnedUtf8String a separate thing again when i can be bothered

[opaque sigil]

if i see one more example of inheritance via animals or shapes i'm going to lose it

[knotty current]

kinda accurate

[sage crag]

because having the allocator in the type makes it a bit restrictive

[dry charm]

ah the classic Factory thing, which only enteprise level java devs follow

[rough bloom]

there's String/str, CString/CStr, and OsString/OsStr at least

[sage crag]

str is just the beginning since you can give it different indirection

Box, Arc, Cow, &

[opaque sigil]

found the image (it's incomplete)

[rough bloom]

yeah, but you can do that with all types, so I'm not counting those

you can probably nest Arcs too
I don't think that should count as infinite string types

[sage crag]

Arc<Arc<T>> is semantically equivalent to Arc<T>

[opaque sigil]

Arc<[u8]> should count at least tbf

[rough bloom]

not even a string, that's just bytes

Arc<str> is fair

but idk if that even works

[sage crag]

at least in rust, its not a string

[tender river]

your code doesnt work? go learn assembly

[sage crag]

str and String are guaranteed to be valid utf8, and OsStr and OsString use the encoding of your os

CStr is also (valid?) utf8, but its null terminated

Path and PathBuf are OsStr and OsString wrappers

but byte arrays have no compile time guarantees which rust loves so much

[tender river]

actually it isnt utf8

[sage crag]

just null terminated bytes huh

that's crazy

[rough bloom]

it's not UTF-8 but is usually converted to a normal UTF-8 string to be useful

[opaque sigil]

you can turn it into a rust string via utf8 validation iirc

yea

[sage crag]

i dont wanna think about small string optimisations

[opaque sigil]

https://www.tunglevo.com/note/an-optimization-thats-impossible-in-rust/

it's about as disgusting as you'd imagine it be

[sage crag]

random nerd sniping xkcd

[tender river]

sso is fun and easy
its not fun and easy when the standard string doesnt use sso and you have to find and replace a billion type signatures though

[rough bloom]

could you just replace the Rust std String implementation with this?

[tender river]

do NOT confuse it with single sign on though
single sign on is NOT fun and NOT easy

[tender river]

i dont think that theres anything in the standard library that mandates String to be a fat pointer with capacity

[rough bloom]

yeah, I think so too
could be interesting to try this out and see how it impacts performance for large builds with many dependencies that don't use this optimization

another item added to the "maybe do this someday" list

[tender river]

its possible to create an extension to the language that allows borrows that are independent of the owning object's moves 🤔

then sso would break that

like a handle that points to an object rather than a specific location, and you have to guarantee that the object lives for that lifetime rather than stays in place for that lifetime

[olive sable]

no branchless programming

[rare bramble]

branchless programming my beloved

[olive sable]

who needs branch predicition? just dont have branches

[rare bramble]

so true

[opaque wharf]

I read SSO immediately goes to Single-Sign On lol

[opaque sigil]

this is what big cpu doesn't want you to know

[olive sable]

if statements are the work of satan and should be banished

[opaque wharf]

SSO is easy. Choose a vendor and be done

[rough bloom]

I don't think this would matter?
AFAIK all that crates like strumbra or smol_str do is store the string as part of the struct itself, not inside of a Box which would put it on the heap
that struct can still be moved around as long as the reference to it remains valid, no?

[opaque wharf]

Because I don't want to touch the technology underlying SSO with a 9ft pole

[patent walrus]

[hoary lion]

million different :true: variations

btw gm

[olive sable]

which true is the one true true?

truetruetrue

ive said the word to the point that its lost its meaning

[hoary lion]

TRUE... True... true...
-# true...

[olive sable]

[rough bloom]

there should be a language where booleans are actually probabilities

probably exists already

[olive sable]

guys i think im gonna do it.
i think im gonna make this project have seperate files

not one monolith

[rough bloom]

clang will thank you

[olive sable]

may the clang be with you

[stone cedar]

But how many translation units?

[hoary lion]

today I finished d3pm paper

seems like that one is for the gemini diffusion

[olive sable]

idk what that means so 0

[nocturne olive]

Woah, unheard of invention
I expect to see you writing header files any time soon

[opaque wharf]

.c or .cpp file

As opposed to .h or .hpp

[olive sable]

idk

[sage crag]

in my experience replacing branches with branchless computation is often a really bad idea

[stone cedar]

Basically everything that the compiler sees on a single invocation is a translation unit.
So having a ton of headers and a single cpp you would still have a single translation unit.

[opaque wharf]

Smells heavenly

[sage crag]

#programming

[opaque wharf]

The white part is coconut milk, don't worry

[olive sable]

if you can do it in a well layed out method you probably should. but dont get too "creative" with solutions cuz thats gonna create more overhead than you're saving

[sage crag]

nah, you shouldnt even if its well laid out

the compiler can do it better

[opaque sigil]

i can't wait for sam to ditch emscripten

[sage crag]

and the code will be easier to read

[opaque wharf]

Sometime the compiler could be smarter than you sam

[opaque sigil]

don't try to be smarter than the compiler unless you can prove it, just don't make its life unnecessarily hard

[olive sable]

for glsl in my experience it does give some benefit to at least think about reducing branches

[sage crag]

additionally branches are incredibly cheap on modern cpus

[opaque wharf]

Only do optimization afterwards

After profiling

[opaque sigil]

gpus HATE branches

[olive sable]

im not talking cpu lol

[opaque wharf]

Like what konii do

[sage crag]

[opaque wharf]

Oh, fair

[sage crag]

to be fair then, gpus are about 2 decades behind on compiler opts

[opaque wharf]

WHAT?

[olive sable]

glsl shaders

[sage crag]

hyperbole

but its typically a lot harder to optimise a shader than a cpu program

[opaque sigil]

the things we give up for parallelism

[opaque wharf]

I'm no graphics programmer so I don't know the state of GPU programming lol

[sage crag]

mainly due to parallelism, yes

[olive sable]

ye..

[opaque sigil]

you can just think of it as

cpu = few smart cores
gpu = many stupid cores

[tender river]

i'm talking about a hypothetical language extension

[stone cedar]

Also depends on what you branch, branching on constants or shader input behaves very differently

[opaque wharf]

All I know about gpu world is just from the funny internet man lol

[noble zodiac]

don't remind me of the nonsense I had to do for branchless code in cuda C

[tender river]

if you could borrow a string's data while still allowing the caller to move the string itself without destroying or mutating it

[opaque sigil]

[olive sable]

gpu cores are like tadpoles. there are many but they have no clue what is happening and most of them will die if left alone

[sage crag]

do not do what i do when i start trying to make things faster, it's bad for you

[noble zodiac]

we even used X macros

[opaque wharf]

Luckily I do not need extreme performance most of the time

[sage crag]

me neither im just a numbers junky

[olive sable]

i need more power!!!!!!!
4k fps is for chumps

[rough bloom]

yeah, I got that
but I don't see how a small string optimization would be hindered by this more than any other struct would be
where the struct is actually stored does not matter, there is no explicit reference to the stack anywhere

if it's just about performance, then you can already kill the optimization by just putting it in a Box

[sage crag]

unicode garbage is my favourite kind of garbage

[rough bloom]

not CJK?

or are those characters just not rendered properly

[tender river]

because when you borrow the string as a slice, it returns a pointer to heap if its stored on the heap or stack if its a smol string

[winged kraken]

do you build any projects with c# ?

[sage crag]

"�A��&0�"�b���"

[tender river]

F# is cool C# god forbid (at least its not java i guess)

[noble zodiac]

I only use C# when I'm forced to

[sage crag]

this is the unicode replacement character U+FFFD which i insert in place of an invalid byte when converting bytes lossily to utf8

[winged kraken]

I know the basic of c# it's more simple than c++

[tender river]

its not hard to be simpler than C++

[sage crag]

true...

[rough bloom]

except it doesn't really return a pointer, it returns a reference, which for string slices is AFAIK implemented as a fat pointer
with your hypothetical extension, this reference would have to instead be a handle of some kind that remains the same even if the data it points to moves

[tender river]

with my extension, the reference could still be a reference as long as parent object's moves dont invalidate it

in fact, a String's moves don't invalidate the &str

[sage crag]

i suppose similarly moving Vec<T> doesnt invalidate the &[T] provided T allows it

[stone cedar]

CSM coming along nicely, now I just need to combine them
-# Camera frustrum is shown in yellow, and sun frustrum shown in grey

[stark needle]

[olive sable]

I just pass the depth to the fragment shader along with a uniform buffer that holds the thresholds to combine them

For optimization each depth is on a singl colourchannel in an rgba image

That way you dont need to bind 4 image buffers

[noble zodiac]

wait, since when can clippy emit something as an error?

[tender river]

yup, it can

[stone cedar]

I have also seen many modern games just use a giant image and reserving different areas for different maps, same difference though

[noble zodiac]

since when? never seen that before

[tender river]

i think clippy::never_loops uses that

probably

[olive sable]

True.
Im doing webgl tho so im limited to 4096x4096 to keep phone compatibility

Otherwise i could do 32K x 32K

[stone cedar]

I have the leisure to not target anything other than the hardware I own

[noble zodiac]

I'm 95% certain that clippy is wrong, but I'm also 50% certain that I'm often times not right when I say 95%

[tender river]

its fine just allow it

[noble zodiac]

let me actually verify it before crying foul

[tender river]

no no why would a code analysis tool know better

[noble zodiac]

as expected, clippy is right

[olive sable]

the amount of includes is already getting out of hand

i have turned the gl context into a seperate file now

idk what the diffrence is between c, cpp, h and hpp but it seems to work

[noble zodiac]

they are for c and c++

[olive sable]

so what about c vs h?

[opaque sigil]

just conventions (that are kind of enforced by the compilers)

c source code vs header

the idea is that the header files sort of act as an api spec

whether that's internal or public for others to use your library

[olive sable]

okay

is it normal to have this many includes?

[stark needle]

H files only have the definition

[noble zodiac]

depends on the library but yea, it can be like that

[stark needle]

Not the declaration

Otherwise u get annoying import loops and stuff

[opaque wharf]

Normal for complex project

[stark needle]

Which is why u also put the IFNDEF or something I don't remember

[rough bloom]

other way around
definitions have the fn body, declarations do not always have the fn body

[noble zodiac]

what you can do is create a prelude.h, put commonly used header in there and then only include that in your actual source files

[stark needle]

But ye u typically never wanna import the .cpp file

[opaque sigil]

codegen

[olive sable]

the file looks like this. so its just the definition yes

[opaque wharf]

Nope

That includes the fn body

[olive sable]

?

[opaque wharf]

Definition is just the function name, return type, and parameters/arguments

[olive sable]

ah

im used to the python meaning of def

[rough bloom]

*declaration
stop mixing up the words

[noble zodiac]

bold of you to assume that sdl_createwindow won't give you a null pointer

[olive sable]

it works so

[opaque sigil]

not like you can do anything without a window

[stark needle]

in the .h file instead of defining the function body with {} u just do ;

[stark needle]

Then in the cpp file u import the .h

[opaque wharf]

I'm sorry okay, my mind is already full of stuff I better not know about lol

[stark needle]

And write the same function again but with body

[opaque sigil]

reject separate headers and source files, embrace single header libraries

[olive sable]

why woudl i write it twice?

[rough bloom]

the joys of the C programming language

[stark needle]

Circular imports

[olive sable]

what the point of having it in a seperatew file if i have to put it in the main anyways?

[stark needle]

If u have for example

[noble zodiac]

ah C and forward declarations, my beloved

[olive sable]

aslo, why are yall neuroClueless'ing me? it works so whats the problem?

[opaque sigil]

it won't when sdl decides not to give you a window

but as i said it's not like that matters in this specific case

[noble zodiac]

wdym, its error checking, you do that in case something doesnt work

[opaque sigil]

you'd have to panic either way realistically

[rough bloom]

pretty sure that it will stop working once you have more than 1 file that uses this header

[noble zodiac]

there are a million reasons why sdl might fail

[stark needle]

mylib.h Which imports mylib2.h

And mylib3.h imports mylib.h and mylib2.h

And then have main.c which imports mylib3.h

U are importing mylib2.h twice

[opaque wharf]

Did he not use #pragma once or include guard?

[stark needle]

And a function can't be defined + declared twice

[rough bloom]

no

[olive sable]

if it doesnt work then just let it crash and burn. no reason to keep a game-engine running that wont dispaly anything

[opaque wharf]

NICE

[stark needle]

Pragma once my beloved

[noble zodiac]

the point is that you are suppose to notice and tell the user: hey, this went wrong. In fact, you tell it to yourself since some random dude will tell you UHHH it just crashed and you have no idea why

[olive sable]

fine

lemme backread the other part of the convo

[rigid snow]

erm clearly you have to be forward looking and handle that gracefully so you don’t have problems when you decide to add a multiplayer server feature in the future

[opaque wharf]

...he will do multiplayer? In C++ and webgl?

[rigid snow]

?

[olive sable]

are you not supposed to import something that in imported file already imported?

[rigid snow]

it’s not like i’m being serious but why not

[opaque wharf]

I thought you were serious lol

[rough bloom]

it wouldn't even necessarily be because of the missing include guard
even if the header is included only once per compilation unit, the symbols from the header are still present in multiple objects when linking

[olive sable]

im sure that by then i will fix this single file resposible for creating the context

[olive sable]

maybe

[opaque sigil]

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

[olive sable]

i want to create neuro monopoly eventually

[opaque wharf]

True

I forgot about that part

[rigid snow]

am i missing something here, what’s the problem

[opaque sigil]

there's nothing stopping you from just having a singular header with all definitions either and just including that

or not have a header at all but that doesn't work if anyone else is supposed to use your code from outside

[opaque wharf]

It's just hard to do I guess

[olive sable]

no its not

jsut need to do some websocket shenanigans

[rigid snow]

you’d have to use websockets but that’s it

yeah

or webrtc lmao

[opaque wharf]

Oh right, I forgot about websocket. Man it's been a long time since I did anything with that

[olive sable]

im not sure how i would do a server tho. ideally a player gets to host so i dont have to maintain a server myself but that invites hacking

[rigid snow]

probably actually better for a game

[tight tinsel]

damn you apple why cant i forever loop my audio files!!

[opaque wharf]

Thats where WebRTC comes into play

So you just need a server for signalling

Or become a relay in case of NAT

STUN and TURN respectively

[rigid snow]

fuck p2p based multiplayer that’s not what i meant

[opaque wharf]

Why?

[idle dune]

Anyone know a good tutorial channel for getting into learning JavaScript

[rigid snow]

because when a player’s potato computer becomes a server and i happen to join it i don’t enjoy the 10 game ticks per second

[olive sable]

another one

why are they all learning javascript?

[idle dune]

JacaScript is used by RPG Maker MV I’m sorry

[olive sable]

why?

fuck rpgmaker

[opaque wharf]

Because its ez

[rigid snow]

super widely used ig

[opaque wharf]

You already have a browser

You can already play with JS

[olive sable]

everything is easy once you know how its done?

besides like physical effort

[opaque wharf]

Yeah but the step required for JS is literally notepad and browser

[idle dune]

I just wanna make a fun retro inspired JRPG and I already own the software

[opaque wharf]

Do a test to determine which one to be the host lmao

But yeh, I can see why its annoying

[olive sable]

cout << "host error: potato pc, automatically rerouting host connections";

[opaque wharf]

I think the best place is just reading up documentation but thats just me because I already know programming concept

[idle dune]

I’ll try that

Thanks

[opaque wharf]

Yes. And insult them while you're at it lol

[olive sable]

ideally a core or 4 for the game itself an a core for hosting no?

so just choose the one with enough cores that also has the highest clock

[opaque wharf]

Ideally it is a separate process ngl. So you can have a dedicated server too

[olive sable]

you could still call the seperate procces from the main game tho i think?

altho that probably wont work in webgl ye

[opaque wharf]

Yes

[opaque wharf]

Maybe a web worker. But I don't know how feasible that is

[olive sable]

ill see

[opaque sigil]

Webgl is in fact available in web workers

If mdn is to be believed

[opaque wharf]

Good to know then. I don't read much the API availability that I don't really use

[rigid snow]

why aren’t threads a thing in wasm

there is a proposal tho

[opaque wharf]

Because JS lol

[olive sable]

they're not? damn

i guess you'l need to go through js then for a worker thread

[opaque wharf]

Yea that's why I said the best shot is probably web worker. A separate script run independently

[rough bloom]

it's a bit more complicated than that because you also have to consider things like latency to other peers and reliability of the connection
also you have to implement host migrations because the host might close the game at any time, but that's just an issue with P2P games in general

[olive sable]

true

[rough bloom]

can't draw to the screen, but can probably render to an offscreen canvas (or do compute with compute shaders, if WebGL has those)

[rigid snow]

or you just don’t handle host migration which is exactly what 99% of games do
“The host has disconnected from the session. Return to main menu?”

[olive sable]

for now i wont be doing p2p yet. the game engine is not even built yet, i have a single polygon on screen

[rough bloom]

great user experience, just randomly get kicked from the game

[rigid snow]

exactly

[olive sable]

my main takeaway from this is, i just wont be defining shit that gets defined in my imports

[winged kraken]

I have knowledge of the basics of c++ we took oop and data structure but to be honest data structure was not that easy , so now I want to start c# I want to ask if anyone use it too

[opaque wharf]

What if each peer is also a host so the system is basically byzantine general problem

[rigid snow]

actual nightmare

[opaque wharf]

Shhh, let Sam encounter it

[olive sable]

p2p is not yet planned

[rigid snow]

isn’t this how GTAO operates btw

[opaque wharf]

Interesting. I know some CRDT can be used to implement those. But I don't know how much of a task it will be

[olive sable]

what if 2 imported files use the same import? wouldnt the import happen twice then too?

[rigid snow]

i have no idea how they do it on a technical level but have you ever heard of what hackers can do there

[opaque wharf]

HE ENCOUNTERED IT GUYS

[rough bloom]

yes
that's literally what an include guard prevents

[olive sable]

ohh

i get it now

fuck

noooooooo

[opaque wharf]

Just use #pragma once

[olive sable]

https://tenor.com/view/meme-anime-gif-22734101

[rough bloom]

see? the joys of C++

[opaque wharf]

No, I do not play GTA after SA

[olive sable]

they cant seriously expect me to #ifndef every single import no??????

[opaque wharf]

They, in fact, could and will

[rough bloom]

yes they can, and they do

[olive sable]

who designed this?

fuck them

[opaque wharf]

Just use #pragma man

[opaque sigil]

[noble zodiac]

someone 50 years ago did

those were some dark times man

[rigid snow]

the list of those things include but aren’t limited to: spawn whatever you want, spawn cages around every player, teleport all players to a specific location, explode every player, spawn a giant fucking LOD mirror copy of the map above it, give any weapons to any players, fly in any vehicle, fly without one, etc.
should give you an idea

[olive sable]

yuove said this multiple times but that doenst help me not knowing what that is

[opaque wharf]

Ahhh, chaos. I love it

[opaque sigil]

pragmas are compiler hints

In this case once for telling it to only include the file once

[rough bloom]

putting #pragma once at the start of a header makes the compiler only read that file once
if it's already been read it gets ignored
so it does the same as an include guard except without the #ifndef stuff

[opaque wharf]

Treat everything that starts with # not as a code, but instruction for the compiler

#pragma once, if the compiler read it on a file, will only process said file once

[olive sable]

start of the main header, or every header?

[rough bloom]

every header

[opaque wharf]

Every

[opaque wharf]

Unless the header already have ifndef

Which probably true for the sdl and other library

[olive sable]

ye but i need to add it to my own ones

[opaque wharf]

Yep

[rigid snow]

except the game is a mmorpg and the majority of players are trying to grind and hackers are kids who enjoy preventing them from doing so

[noble zodiac]

technically #pragma once is a non-standard feature

[opaque sigil]

I love gcc extensions

[opaque wharf]

Aren't every pragma is?

[olive sable]

nah the sdl doesnt have pragma

[opaque sigil]

There's your include guard

[opaque wharf]

This bites me in the ass because apparently most math function isn't constexpr in the standard, but it is in gcc

[opaque wharf]

But it does have ifndef followed by define

[olive sable]

tru

[opaque wharf]

Thats include guard

[rough bloom]

I have consulted the C99 standard
#pragma itself is standardized but any parameters are not

[noble zodiac]

yea, once was considered but rejected

[olive sable]

it does not like pragma in the main.cpp

[rigid snow]

any semblance of good ux? in MY c??

[rough bloom]

oh, except #pragma STDC apparently
never heard of it

[opaque wharf]

.cpp or .c doesn't need it

[olive sable]

why not?

the scope?

[opaque sigil]

3 standardised, rest extensions

[opaque wharf]

Its very technical

[rough bloom]

you don't include .c or .cpp files

[maiden geyser]

because you don't import it

[opaque wharf]

Yeah, thats the simple answer lol

I'm bad at explaining shit

Without going on a tangent

[noble zodiac]

also with C++20 you could use modules instead. Not that I would recommend it

[maiden geyser]

imagine using c++20

[olive sable]

idk what the diffrence is bu sounds complicated

[opaque wharf]

They have a chance to make it good. Nope, they butchered it

[olive sable]

why do they have a system that does the same thing as the include system?

[opaque sigil]

Because includes suck

[maiden geyser]

because they're stupid?

[opaque sigil]

They're just copy paste

[noble zodiac]

they wanted to modernize it

[olive sable]

what?

[opaque wharf]

Tl;dr, we just treat module as if it doesn't ever exists

[rough bloom]

did you not know about the C preprocessor?

[opaque wharf]

I've said it before, #include is just basically saying "copy whatever is in the file here"

[olive sable]

why would i?

[opaque sigil]

You can include c files too

[rough bloom]

you are programming a C(++) application

[opaque sigil]

There's nothing stopping you

[opaque wharf]

The compiler will

I think

[opaque sigil]

It will not

[opaque wharf]

Fuck

[sour harness]

Could have used Rust where all that bs is not necessary :D

[opaque wharf]

No, the style is horrendous!

[rough bloom]

yeah, or any other remotely modern language really

[rigid snow]

no see, they prefer the bracket to be on the same line in the rust world, literally unusable

[opaque sigil]

Let's be realistic, rust would be even more painful

[rigid snow]

not with imports no

[opaque sigil]

At least here he can use opengl 1:1

[rigid snow]

oh that

right

[rough bloom]

they also have a formatter that allows you to configure the bracket to be on a new line if you hate yourself and don't like conformity
but that would require creating a .toml file with a few lines of text, absolutely impossible

[rough bloom]

surely there's a crate that abstracts differences between OpenGL and WebGL

[rare bramble]

you can use WGPU in rust

[olive sable]

that would mean no support for most browsers so nah

[rough bloom]

there is
https://crates.io/crates/glow

[olive sable]

its not that easy

[opaque wharf]

But not widely supported

But also, sam writes his own GLSL so maybe using the same language family is better

[rough bloom]

(this doesn't need WebGPU or wgpu btw, it's just bindings to OpenGL and WebGL and can use SDL2, same as the current stack on C++)

[olive sable]

webgl does not have geometry shaders, no tesselation, no CSM without seperate passes.
lower max buffer sizes, lower version of glsl supported, ....

and those are things that cant be just translated

you need to change the entire shader function

but besides that it should be fine

[noble zodiac]

ok, hear me out

[olive sable]

ia m

i have no way to stop you

[noble zodiac]

just generate one big <table> with colored cells that are exactly one pixel for every frame

[rough bloom]

[olive sable]

you mean pass that to the vulkan renderer?

[noble zodiac]

just raw dog that into html

[rough bloom]

-# this wouldn't actually help because rendering and presentation are different operations btw

[olive sable]

ok sure, but that would effectivly be wastin 75% of all the gpu's work since it always does 2x2

same reason you shouldnt have a lot of small trianlges

[noble zodiac]

and then you just make 1024x768 css variables to control the colors

gigabrain

[olive sable]

absolutly not

[noble zodiac]

it would be hilarious tho

[olive sable]

getting the gpu output and then passing it back to the cpu is a really bad idea

and im 99.999% sure that wont help with wgpu comaptebilit

[rare bramble]

i did this with my project,
I colored a 1600 x 900 texture pixel by pixel and I rendered that onto 2 triangles that cover the whole screen

[olive sable]

you're doing WHAT

[noble zodiac]

btw this was a joke. The dom parser will get an aneurysma

[olive sable]

ceace this behaviour emediatly

[opaque wharf]

I'm just sitting here, laughing my ass off

[rare bramble]

I didn't know how to do it better so it will stay that way

[rough bloom]

that seems fine
I think you could write directly to the framebuffer instead of using a texture but I'm pretty sure fullscreen quads are not that uncommon

[opaque wharf]

See, this is why webdev is fun

[olive sable]

https://tenor.com/view/steamed-hams-aurora-borealis-what-is-happening-fagetti-wtf-gif-16514726

[opaque wharf]

The world is your oyster

[olive sable]

its about 900 x 1600 times better than tsurai's quad per pixel, but not ideal

preferably just blit it

[noble zodiac]

you cannot put a value on art

[rare bramble]

if you want to see the monstrocity
https://github.com/juhotuho10/rust_GPU_raytracing

[olive sable]

in my experience copying all the data to another framebuffer when not needed uses 15+% of the total rendertime

you really shouldnt

[fast pagoda]

refresh running project
try new display for md
it crashed
how could i have anticipated this

[olive sable]

looks nice tho, good job

[noble zodiac]

also, this time clippy really is wrong, frfr

[rough bloom]

would be a small improvement if the texture is in VRAM, though as long as there's no shader it'll be slow regardless

[opaque wharf]

It says there dangerous

[fast pagoda]

i like to live dagnerously

[olive sable]

true

[opaque wharf]

Same

[olive sable]

dagne rously

[sour harness]

When unsafe {} is not enough

[fast pagoda]

they made it too tempting to use

calling it that

well i gotta try dangerously setting it

[tender river]

clippy is wrong, this is not a complex type

[sour harness]

You're only using standard rust generics, can't be that bad

Not even any trait boundaries

[opaque wharf]

What about when danger is not enough?

Are we entering cursed territory then?

[gritty dust]

good morning programming

[fast pagoda]

rm -rf $HOME

[rough bloom]

I can't find the issue here, it seems fine
the texture gets written directly from a compute shader, which then gets read from the render shader that displays the fullscreen quad
could probably just merge the shaders into the render shader, but if you want some form of asynchronous rendering then using a compute shader should work great

[opaque wharf]

[ $[ $RANDOM % 6 ] == 0 ] && rm -rf $HOME || echo Click

Play roulette instead

[fast pagoda]

lmao

[olive sable]

put the shader maker thingy also in a seperate file cuz it felt right.
im doing this purely based on vibes.

it is annoying how half these functions are underlined with red cuz they only work in web

in ms word you can atleast press ignore

[fast pagoda]

mine are underlined in red from all the import errors and total failures of awareness on my part

[opaque wharf]

you code in WHAT?

[olive sable]

i dont

[fast pagoda]

i used to write webpages in ms word

wordpad

[rough bloom]

fixing the errors: ❌
permanently ignoring the errors: ✅

[fast pagoda]

office 97

[opaque wharf]

Ahh yes, I forgot that MS Word can refers to two distinct product

[olive sable]

they're not really errors in the first place. it jsut doesnt work locally

[opaque wharf]

People mostly assume MS OFFICE Word

[olive sable]

basicly the opposite of "it works on my machine"

cuz it doenst

[rough bloom]

you mean it doesn't compile at all?

(without Emscripten I mean)

[olive sable]

nope

[opaque wharf]

You know, ideally it should not show error because the compiler should know you are compiling to some target

[olive sable]

it compiles with emcc src/main.cpp -o main.html -s USE_SDL=2 -s USE_SDL_IMAGE=2 -s USE_SDL_TTF=2 -s MAX_WEBGL_VERSION=2, but besides that it doesnt

[rough bloom]

if you want to target native then I'd consider that an error
if not then fix your VS Code settings

[fast pagoda]

i always fuck up my config somehow and end up with anything installed in the venv undetected by vscode

even tho i have it as the env for the editor too

[gritty dust]

Coding on vibes is based

[opaque wharf]

Nuke your user and system config dir and start over

[stone cedar]

Still waiting for #embed to finally be supported by all major compilers

[opaque sigil]

Look at bear on GitHub

[fast pagoda]

i did that and it was good for a while and then it started happening again so i said f it

[opaque sigil]

And point your ide towards that

[opaque wharf]

I think the error squiggle could be removed with proper cmake

[olive sable]

i do want native to work but im not big brain enough yet to know how to do

if sys.platform == "win32":
  import gl3.h
  ...
else:
  import emscripten.h
  ...

this shit in C++

[opaque wharf]

It's called compile flag

[olive sable]

[opaque wharf]

Basically you do this

#if PLATFORM_NAME == "Win32"
...
#else
...
#endif

[gritty dust]

Pog it worked in my setup at school yesterday btw, I got the vision, coefficients, and serial all working together

[opaque wharf]

And when you compile, you set the flag by doing

g++ -o main ./main.cpp -DPLATFORM_NAME=Web

[fast pagoda]

thanks for censoring the photos of me you keep on your walls at all times

[olive sable]

what about when compiling to wasm?

[gritty dust]

Lolll

[opaque wharf]

Then use the emcc compiler with the flag set accordingly

[olive sable]

ok

[opaque wharf]

This is why cmake and meson exists

It lets you setup complex compilation process

[olive sable]

i think im gonna have to make a seperate context.hpp file for local, since the current one has emscripten fuctions

[olive sable]

i just have a makefile

[rough bloom]

full command for wasm would be emcc src/main.cpp -o main.html -s USE_SDL=2 -s USE_SDL_IMAGE=2 -s USE_SDL_TTF=2 -s MAX_WEBGL_VERSION=2 -DPLATFORM_NAME=Web
for native it would be something like clang++ src/main.cpp -o main.exe <big list of include directories and objects from libraries here> -DPLATFORM_NAME=Win32

[opaque wharf]

Yeah, that works but tedious and brittle. Especially if you need to add more and more library or your source file grows in number

[olive sable]

im not expecting it to grow much more

just need something to read gltf and i should b good

[opaque wharf]

Yeah, for small project I think its still bearable to use makefile

[rough bloom]

yeah, build systems are a major pain

probably fine to skip them for now

[stone cedar]

Or just set up your project to be a natural unity build. Then a small batch/bash file will probably suffice

[olive sable]

i forgot this menu existed in win+v
(☆▽☆)

[olive sable]

this is a no unity household

[rough bloom]

surely you aren't referring to Unity the game engine

[stone cedar]

Not the game engine
https://en.wikipedia.org/wiki/Unity_build

[rough bloom]

bad

horrible

please nobody ever do that

[opaque wharf]

Even I, a chaos lover thinks that is not okay

[olive sable]

havinmg the word unity in my pipeline would rruin my image as a homemade game-engine

[opaque wharf]

I don't even know that paradigm existed tbh

Who comes up with that

[rough bloom]

never saw it in any actual project either

[opaque wharf]

If linker is an issue, use https://github.com/rui314/mold instead

[rough bloom]

must be ancient, from times when single-core processors were a common thing

[opaque wharf]

Yeah, that makes sense

[rough bloom]

you should try using Bazel
it's an anagram of Blaze, which is a very cool word

[opaque wharf]

Fuck bazel build system

[olive sable]

i hate how some programs have comments like //set atttributes right above a function named setAttribute(args)
like brother we can read we know

[stone cedar]

Depends mainly on linker speed yeah, since linking is single threaded and can be a bottleneck.
I am using it right now, but my project is currently only 4.5k loc so not that big. It is kind of reassuring to me to just hit a button and be certain the build system didn't fuck up but that might be just my msbuild PTSD.

[nocturne olive]

Yeah, make actually useful comments if you're gonna make some, otherwise just rely on self-commenting code

[olive sable]

[tender river]

bazel

might as well use maven for building c++

[olive sable]

might as well not build

just ship the code and let the user figure it out

[rough bloom]

Gradle

[opaque wharf]

Use mold my dude

[sour harness]

  Some tensors share memory, this will lead to duplicate memory on disk and potential differences when loading them again: [{'language_model.lm_head.weight', 'language_model.model.embed_tokens.weight'}].
 A potential way to correctly save your model is to use `save_model`.
 More information at https://huggingface.co/docs/safetensors/torch_shared_tensors```"Good job spending one hour quantizing that model. Oh, you want to store the result? Well, I guess sucks to be you then."
Also the proposed "save_model" function does not exist and the linked web page tells me nothing.

[opaque wharf]

ML is cursed ngl

[rough bloom]

at least it's using safetensors
pickles were even more cursed imo

[sour harness]

The fact that the ML eco system is built on Python is cursed

[olive sable]

it doesnt seem to like it

[rough bloom]

it should've been Julia

[opaque wharf]

Sorry, I was wrong and have since corrected the code lol

[olive sable]

ah

[tender river]

#if not #ifdef

[sour harness]

Storing the model in non-safetensor "works" but then loading fails :D

[opaque wharf]

Thats certainly is something alright lol

[stone cedar]

Why? I don't currently have a seperate link step

[opaque wharf]

Oh, right. The unity build...

[rough bloom]

on a multicore processor (all of them) it should usually be faster to compile multiple files individually and then link with mold than to compile all of them in a single process but skipping the linker

[stone cedar]

If it becomes unmaintainable and I feel the need to restructure I'll think about it. But I kinda want to test the waters and see how far I can go. It also looks like mold currently doesn't run on windows so that is also not working for me rn.

[rough bloom]

LLVM lld is also pretty fast and should work on Windows

[opaque wharf]

Then lld

But yeah, you do you

If it ain't broke, don't fix it

[olive sable]

couldnt get it to work, i have this now instead

[stone cedar]

Depends on how many files and how big, I guess. There is less parallelism but you don't have any io overhead. Let me check if I can find some sources that have some numbers comparing unity-builds with conventional c/c++ project structure.

[rough bloom]

I wanted to suggest something, but #elifdef isn't a thing until C++23

[opaque wharf]

#if PLATFORM=="Win32"
  #include "glcontext.hpp"
#elif PLATFORM=="Web"
#else
#endif

Platform is not reserved right?

[olive sable]

nah, for this one i dont have to specify a variable in the build function

[olive sable]

idk

[opaque wharf]

Fair

[rough bloom]

if you want it to be slightly cleaner you could create a platform.h where you check for _WIN32 and __EMSCRIPTEN__ and define your own macros accordingly
then you can include platform.h everywhere and use those instead

[olive sable]

i could ye, but do you mean with the include or jsut a file with 3 if statements?

[opaque wharf]

Or you could also do

#ifndef TARGET_PLATFORM
  #ifdef _WIN322
    #define TARGET_PLATFORM "Win32"
  #elif defined(__EMSCRIPTEN__)
    #define TARGET_PLATFORM "Web"
  #else
    #define TARGET_PLATFORM "Fallback"
  #endif
#endif

[olive sable]

this is getting complicated

[opaque wharf]

Nah, just a suggestion

Its just over engineering really lol

[olive sable]

no indentation so i cant tell wtf you're even doing

[rigid snow]

mhm no platforms other than win32 and web exist

[olive sable]

im hoping it works on linux too, so fallback for that

i hope it crashes on imacs tho, out of principle

[rigid snow]

also indentation here is so cursed

[opaque wharf]

Hey, sam wanted it, not me. I wrote it without indentation lol

[olive sable]

are yo not supposed to use indentation there?
i like readablility

[opaque wharf]

For hblang I think that shit would be cleaner

[opaque wharf]

No, usually anything that starts with # does not get indented. Its a code style thingie like the bracket. But there are no hard rules

[nocturne olive]

Now that's silly

[olive sable]

im beginning to think yall hate making your files readable, and then complain that C++ isnt readable and go to rust

[rigid snow]

it's just convention

you can do whatever you want

[olive sable]

C++ convention is bad tho

[nocturne olive]

-# I approve

[opaque wharf]

And luckily there is no one complaining like rustfmt

[rough bloom]

yeah, that's what I mean
if you do this you can have your own TARGET_PLATFORM macro
or alternatively define separate macros like TARGET_PLATFORM_WIN32 and check if those are defined everywhere else

[olive sable]

i dont see why i would need to do that tho

[rough bloom]

you don't need to

[rigid snow]

it's useful to have for when you need it

[opaque wharf]

Its just when someone need to do cross compilation, usually its done that way

[olive sable]

what even is cross compilation?

[opaque wharf]

I should not have said that

[rough bloom]

you are currently cross-compiling from Windows to wasm

[olive sable]

yes

thast is true

[nocturne olive]

Compiling on one platform for another

I think

[olive sable]

but wouldnt just using #ifdef _WIN32 again work?
or is that bad for some reason?

[opaque wharf]

If someone want to compile from linux, it won't work

[sage crag]

see, hblang is very good for this:

// in lily
$match target.current {
  .hbvm_ableos => {},
  .x86_64_linux => {},
  _ => {},
}

// in the language itself
$if @target("hbvm-ableos") {}

and these get checked at compile time

[opaque wharf]

But then if you use the flag, the compilation can be configured by setting the flag manually

g++ <long ass command> -DTARGET_PLATFORM="Win32"

[olive sable]

hmmm
so you mean that the variable gets checked at compilation and not on the target system?

[rigid snow]

macros? at comptime?? boring, add runtime macros

[opaque wharf]

Indeed

[olive sable]

oh

[sage crag]

there are no macros here

[olive sable]

hmmm

[opaque wharf]

Anything with # is compile time only

[olive sable]

ok i see your point

[opaque wharf]

Lovely

[nocturne olive]

Anything starting with # is for the preprocessor

[rigid snow]

if you call them something else they don't stop being macros

[nocturne olive]

It happens even before the compiler actually

[sage crag]

no, there are literally no macros

[opaque wharf]

Meanwhile CodeParade adding generic template to C#

[rigid snow]

is the point that it compiles down to the correct @use directive

from this

the it being lib

[sage crag]

very close. modules are types in hblang, so you can return an @use() as a type.
but it does in fact only compile the correct @use(), yes

[tender river]

the point is that compile time code is a bit special because it can create types and use dynamic values (@Any()), but there are no type objects at runtime. @use returns a type because types are modules

[olive sable]

like this i guess?

[rigid snow]

so it is compile time code

[opaque wharf]

Yeah

[sage crag]

it is comptime, yes

not a macro though.

[rough bloom]

(except the _WIN322 typo)

[olive sable]

ah shit

ok

[rigid snow]

blah blah

[tender river]

it cant replace macros in the general case but it can replace a whole bunch of conditional compilation; the most complicated part is that parameter and return types are allowed to depend on function arguments (same as zig i think) but i wouldnt call it dependent types since its known at compile time

[opaque wharf]

Now here is the fun part. You see the NOT green one? That is your current active compilation target. With proper VSCode config, you can change it by setting the correct compile_commands.json. That means anything that's only valid on the web won't be underlined as error when you are not targeting it

[sage crag]

no code generation, just conditional compilation and inlining

[tender river]

also since the functions arent bijective there arent the analogs of infer in typescript

you use @Any() or you dont

[sage crag]

you probably could create an arbitrary binary using just the typesystem though

[opaque wharf]

WHAT?

[tender river]

hard to call it a type system when its just running arbitrary code javascript style

[rough bloom]

to add to this, it also means that you usually won't see compile errors for the Web target in VS Code because it's set to Win32 (or the other way around, no errors for Win32 when set to Web)
but such is life when using C or C++

[sage crag]

yeah

its a typesystem because its a system with types

[tender river]

also, the type system involves a highly sophisticated termination analysis algorithm - running the compile time code for a while and seeing if it terminates

[sage crag]

this is true

old compiler handled this better

but new one never recieved anything of the sort because we never write incorrect code right

[sour harness]

The Halting Problem (easy mode) solution

[tender river]

bet the computer is gonna feel really dumb when i unplug it

[sage crag]

i doubt it will feel when you unplug it

i think it is gone

[tender river]

in my experience unplugging computers doesn't make them disappear

[rough bloom]

what does the compiler need this information for?
to decide if it should log an error and skip that compile-time expression?

[tender river]

are you sure you're doing it correctly

[tender river]

its not really information, its just basically error handling

[sage crag]

so basically

if i want to implement arithmetic in pure types

i can only go up to the 8th fibonacci number before the compiler starts complaining

but before then, it works

[rough bloom]

[sage crag]

this is not a joke

[amber fractal]

[rough bloom]

so it is just to stop people from putting an infinite loop in the compile-time code

[rough bloom]

I did not perceive it as such

[sage crag]

yesh perhaps

[tender river]

i had to request an increase on compiler stack size from 1mb to 8mb in order to compile this https://git.ablecorp.eu/chayleaf/hblsp/src/branch/trunk/json.hb

[sage crag]

here is the src code

[rough bloom]

$N := Suc(Suc(Suc(Suc(Suc(Suc(Suc(Suc(Zero))))))))

[sage crag]

there's a certain beauty in whatever tf this is

at the same time

its uh

not useful

[tender river]

not useful in hblang

[opaque wharf]

I thought it was S U C

Oh wait, it is

[rough bloom]

compiler abuse

[opaque wharf]

Successor

[sage crag]

json parser is compiler abuse indeed

kinda rude to make a json compiler tbh

[rough bloom]

Suc(cessor)

[tender river]

but representing numbers as linked lists is very useful if you want something like actual dependent types