03:03 <+bridge> naming of source files in ddnet source doesn't seem to be consistent case-wise 03:04 <+bridge> there are some files that are bunched together (foobar.cpp/h) and there are some that are separated with underscores (foo_bar.cpp/h) 03:04 <+bridge> is either one recommended over the other 03:20 <+bridge> why/how is aio broken on windows. it looks like it just uses the existing thread stuff 03:21 <+bridge> i was gonna use it for the thing but if it's really broken on windows i need to either figure out why & fix/work around it or do a lower level approach 03:21 <+bridge> either way ig i can't rn 07:59 <+bridge> Throw on tsan and look if it's broken. Afaik it was always broken on old windows versions? 09:13 < ws-client> OMG, today is friday for futurust.. almost forgot 09:13 < ws-client> but i have no topic 09:16 < ws-client> i have smth. yesterday i found a crate with an asshole maintainer xD 09:16 < ws-client> "friday for futurust" that doesnt really work phoenetically 09:17 < ws-client> !rust he used MIT/Apache for all his libs, but then changed one line of code to use a MPL license for that code line. causing all libs basically to be MPL bound 09:17 < chillerbot1> @Ryozuki 09:17 < ws-client> <:rust:920765596289880074> 09:17 < ws-client> that's how evil this license and thought is 09:17 < ws-client> jupsti u can now use the ping all rusters command 09:17 < ws-client> one last time heinrich told me to remove the spammy commands :c 09:17 < ws-client> what spam xd 09:18 < ws-client> pinging 5 ppl at once 09:18 < ws-client> sounds efficient 09:18 < ws-client> stuff like !c and !rust and !jup 09:18 < ws-client> heinrich hates everything that makes fun for others 09:18 < ws-client> xd 09:18 < ws-client> he big boss im scared he ban ma ass 09:18 < ws-client> so when i get time ill kick the commands 09:19 < ws-client> no 09:20 < ws-client> don't bow for dictator heinrich 09:21 < ws-client> he runs the bridge to my only social interaction he got me at the balls 09:22 <+bridge> idk 09:22 <+bridge> agpl best 09:22 <+bridge> idc* 09:22 < ws-client> but asshole move non than less 09:22 < ws-client> bsd best 09:22 < ws-client> better than agpl at least 😂 09:22 <+bridge> changing license after fact is a bit meh ye 09:22 <+bridge> and one line is being troll xd 09:23 <+bridge> chiller put some rust at the balls 09:23 <+bridge> so atleast they are memory safe 09:23 < ws-client> e wot xd 09:23 <+bridge> :poggers2: 09:23 < ws-client> balls of steel 09:23 < ws-client> get rusty 10:59 <+bridge> I'm sorry for hosting and maintaining the IRC bridge? 11:02 <+bridge> idk why i got the ich to implement my own hashmap in rust and c 11:03 <+bridge> are there existing stack based hashmaps on rust? 11:03 <+bridge> can be fun 11:04 <+bridge> like smallvec but smallmap 11:10 < ws-client> @heinrich5991 ... 11:11 <+bridge> what is it? 11:11 < ws-client> your comment :D 11:11 < ws-client> nobody complained about you hosting the bridge 11:11 < ws-client> its the first time i seen him express some kind of sadness 11:11 < ws-client> i appreciate that 11:12 < ws-client> we r on a good way 11:13 <+bridge> https://crates.io/crates/micromap 11:13 <+bridge> apparently it's just an array without any hashing 11:13 <+bridge> but thinking about it, it makes sense 11:14 <+bridge> if it's small, you can forego the hash part of the hashmap 11:14 <+bridge> true 11:14 <+bridge> does it make sense to make one with hashing and stack based 11:14 <+bridge> but the stack doesnt need to be small 11:14 <+bridge> well i wanna make it for fun 11:14 <+bridge> can be useful to learn unsafe 11:14 <+bridge> then just do it I guess ^^ 11:16 <+bridge> i see the hashless aproach just works till 20 11:19 <+bridge> yea, just till some constant 11:44 <+bridge> @ryozuki if u like working on hashmaps and have no hobbies. I have a feature request for a create i use xdd 11:45 <+bridge> @ryozuki if u like working on hashmaps and have no hobbies. I have a feature request for a crate i use xdd 11:46 <+bridge> xd 11:46 <+bridge> I pay u 5€ 😬 11:46 <+bridge> i dont need 5€ 11:46 <+bridge> :ez: 11:47 <+bridge> 🥹 11:47 <+bridge> Then i need to do it, shit xd 12:07 <+bridge> @Jupeyy_Keks what's the feature request? 12:10 < ws-client> @heinrich5991 i use a crate called hashlink its basically a hashmap but with a linked list, which is nice to keep the order, even when calling remove. For the iterator i'd find it useful to have a "skip_to_key".. since its a linked list it could easily start the iterator at a different point in the linked list, which is ultimately nice if i want to drop an element and inform all elements after the element that there was a 12:10 < ws-client> change 12:10 <+bridge> ah, like the hashmap in python 12:13 <+bridge> I guess that would only be possible on iterator creation 12:14 <+bridge> although that's probably not an issue for you? 12:14 < ws-client> yeah its not an issue 12:16 <+bridge> ah! 12:16 <+bridge> the iterator implements `ExactSizeIterator` 12:16 <+bridge> that's probably not possible for yours 12:16 < ws-client> yes i know 12:17 < ws-client> but it could simply be another iterator 12:17 < ws-client> that doesnt implement it 12:17 <+bridge> yea 12:17 <+bridge> but that makes it a little more complicated to convince the maintainers to implement it 12:17 < ws-client> yeah :D 12:18 < ws-client> But if i wrap the iterator in a "skip" struct.. 12:19 < ws-client> similar to how map etc. to it with iterators 12:19 < ws-client> would maybe make the code dublication less 12:20 < ws-client> dup 12:46 < ws-client> !jup @Jupstar ✪ @heinrich5991 name ping commands are off now :c 12:46 < ws-client> !ping 12:46 < chillerbot1> pong 12:46 < ws-client> only this stayed for testing 12:49 <+bridge> !rust d 12:49 <+bridge> Mhh 12:49 <+bridge> Sucks 12:50 <+bridge> Now I have to type out ChillerDragons way too long name again xD 12:50 < ws-client> do @c on webchat axaxax 12:51 < ws-client> also my name is easier to type than fakin ✪ 12:55 <+bridge> Then map DragonBall to my name. Then we ChillerDragonBall 12:59 < ws-client> lol 13:03 < ws-client> !c test 13:10 <+bridge> ```rust 13:10 <+bridge> impl SmolMap 13:10 <+bridge> where 13:10 <+bridge> K: Eq + Hash, 13:10 <+bridge> S: BuildHasher, 13:10 <+bridge> { 13:10 <+bridge> /// # Panics 13:10 <+bridge> /// If length == N 13:10 <+bridge> pub fn insert(&mut self, key: K, value: V) -> Option { 13:10 <+bridge> assert_ne!(self.len, N); 13:10 <+bridge> 13:10 <+bridge> let mut hasher = self.state.build_hasher(); 13:10 <+bridge> key.hash(&mut hasher); 13:10 <+bridge> let start_idx = hasher.finish() as usize; 13:10 <+bridge> let mut iter_idx = 0; 13:10 <+bridge> 13:10 <+bridge> loop { 13:10 <+bridge> let idx_mod: usize = (start_idx + iter_idx * iter_idx) % self.storage.len(); 13:10 <+bridge> 13:10 <+bridge> if self.storage[idx_mod].0 { 13:10 <+bridge> if unsafe { self.storage[idx_mod].1.assume_init_ref() } 13:10 <+bridge> .0 13:10 <+bridge> .eq(&key) 13:10 <+bridge> { 13:11 <+bridge> let new_val = MaybeUninit::new((key, value)); 13:11 <+bridge> let old_val = std::mem::replace(&mut self.storage[idx_mod].1, new_val); 13:11 <+bridge> return Some(unsafe { old_val.assume_init().1 }); 13:11 <+bridge> } 13:11 <+bridge> } else { 13:11 <+bridge> let new_val = MaybeUninit::new((key, value)); 13:11 <+bridge> self.storage[idx_mod].0 = true; 13:11 <+bridge> ```rust 13:11 <+bridge> #[derive(Debug)] 13:11 <+bridge> pub struct SmolMap { 13:11 <+bridge> storage: [(bool, MaybeUninit<(K, V)>); N], 13:11 <+bridge> len: usize, 13:11 <+bridge> state: S, 13:11 <+bridge> } 13:11 <+bridge> ``` 13:11 <+bridge> `(bool, T)` → `Option` 13:11 <+bridge> would make it safe and potentially smaller 13:11 <+bridge> hmm i thought about ye 13:11 <+bridge> i guess in this case maybeuninit is not faster xd 13:12 <+bridge> but i wanted to play with maybeuninit 13:12 <+bridge> ill change it to option tho 13:12 <+bridge> `Option` is literally `(bool, MaybeUninit)` ^^ 13:12 <+bridge> xD 13:12 <+bridge> except saafe 13:12 <+bridge> except safe 13:12 <+bridge> hmm 13:12 <+bridge> you could put the bools into another array 13:12 <+bridge> then it's not useless anymore 13:12 <+bridge> SoA instead of AoS 13:13 <+bridge> whats better 13:13 <+bridge> SoA 13:13 <+bridge> so bools in another array? 13:13 <+bridge> ima try 13:14 <+bridge> yes 13:14 <+bridge> you could separate keys from values while you're at it 13:14 <+bridge> this also allows me to make new() const 13:14 <+bridge> pog 13:15 <+bridge> ```rust 13:15 <+bridge> pub fn new(hasher: S) -> Self { 13:15 <+bridge> let data = std::array::from_fn(|_| (false, MaybeUninit::uninit())); 13:15 <+bridge> 13:15 <+bridge> Self { 13:15 <+bridge> storage: data, 13:15 <+bridge> len: 0, 13:15 <+bridge> state: hasher, 13:15 <+bridge> } 13:15 <+bridge> } 13:15 <+bridge> 13:15 <+bridge> pub const fn new(hasher: S) -> Self { 13:15 <+bridge> Self { 13:15 <+bridge> storage: unsafe { 13:15 <+bridge> MaybeUninit::uninit().assume_init() 13:15 <+bridge> }, 13:15 <+bridge> tags: [false; N], 13:15 <+bridge> len: 0, 13:15 <+bridge> state: hasher, 13:15 <+bridge> } 13:15 <+bridge> } 13:15 <+bridge> ``` 13:15 <+bridge> before - after 13:17 <+bridge> ```rust 13:17 <+bridge> /// # Panics 13:17 <+bridge> /// If length == N 13:17 <+bridge> pub fn insert(&mut self, key: K, value: V) -> Option { 13:17 <+bridge> assert_ne!(self.len, N); 13:17 <+bridge> 13:17 <+bridge> let mut hasher = self.state.build_hasher(); 13:17 <+bridge> key.hash(&mut hasher); 13:17 <+bridge> let start_idx = hasher.finish() as usize; 13:17 <+bridge> let mut iter_idx = 0; 13:17 <+bridge> 13:17 <+bridge> loop { 13:18 <+bridge> let idx_mod: usize = (start_idx + iter_idx * iter_idx) % self.storage.len(); 13:18 <+bridge> 13:18 <+bridge> if self.tags[idx_mod] { 13:18 <+bridge> if unsafe { self.storage[idx_mod].assume_init_ref() } 13:18 <+bridge> .0 13:18 <+bridge> .eq(&key) 13:18 <+bridge> { 13:18 <+bridge> let new_val = MaybeUninit::new((key, value)); 13:18 <+bridge> let old_val = std::mem::replace(&mut self.storage[idx_mod], new_val); 13:18 <+bridge> return Some(unsafe { old_val.assume_init().1 }); 13:18 <+bridge> } 13:18 <+bridge> } else { 13:18 <+bridge> let new_val = MaybeUninit::new((key, value)); 13:18 <+bridge> self.tags[idx_mod] = true; 13:18 <+bridge> self.storage[idx_mod] = new_val; 13:18 <+bridge> self.len += 1; 13:18 <+bridge> return None; 13:18 <+bridge> } 13:18 <+bridge> 13:18 <+bridge> i guess this probs optimizes better 13:18 <+bridge> > `` /// # Panics 13:18 <+bridge> > /// If length == N 13:18 <+bridge> i dont even need len 13:18 <+bridge> oh 13:18 <+bridge> i do 13:18 <+bridge> xD 13:18 <+bridge> whats wrong? 13:18 <+bridge> it should only panic if the key is not already in there, right? 13:19 <+bridge> true 13:19 <+bridge> if it needs to find a new slot 13:19 <+bridge> unused* 13:19 <+bridge> moved the assert to the else 13:19 <+bridge> hmm 13:20 <+bridge> wait maybe i should just loop N 13:20 <+bridge> not inifnite 13:21 <+bridge> are you guaranteed to hit every bucket like that btw? 13:21 <+bridge> i would guess yes 13:21 <+bridge> but idk rly xD 13:22 <+bridge> ok looping N is not enough 13:22 <+bridge> cuz it may need more loops 13:22 <+bridge> hm 13:23 <+bridge> ok it looks like it eventually hits every bucket 13:23 <+bridge> but i got a very long runtime xD 13:25 <+bridge> > Unfortunately, quadratic probing has the disadvantage that typically not all hash table slots will be on the probe sequence. Using p(K, i) = i2 gives particularly inconsistent results. For many hash table sizes, this probe function will cycle through a relatively small number of slots. If all slots on that cycle happen to be full, this means that the record cannot be inserted at all! For example, if our hash table has three slots, then records th 13:26 <+bridge> hmm 13:26 <+bridge> since i cant do resize, which is usually done when the load factor aproaches 0.75 (iirc) maybe linear probing is best used here 13:28 <+bridge> ```rust 13:28 <+bridge> pub fn insert(&mut self, key: K, value: V) -> Option { 13:28 <+bridge> let mut hasher = self.state.build_hasher(); 13:28 <+bridge> key.hash(&mut hasher); 13:28 <+bridge> let start_idx = hasher.finish() as usize; 13:28 <+bridge> let mut iter_idx = 0; 13:28 <+bridge> 13:28 <+bridge> while iter_idx < N { 13:28 <+bridge> let idx_mod: usize = (start_idx + iter_idx) % self.storage.len(); 13:28 <+bridge> 13:28 <+bridge> if self.tags[idx_mod] { 13:28 <+bridge> if unsafe { self.storage[idx_mod].assume_init_ref() } 13:28 <+bridge> .0 13:28 <+bridge> .eq(&key) 13:28 <+bridge> { 13:28 <+bridge> let new_val = MaybeUninit::new((key, value)); 13:28 <+bridge> let old_val = std::mem::replace(&mut self.storage[idx_mod], new_val); 13:29 <+bridge> return Some(unsafe { old_val.assume_init().1 }); 13:29 <+bridge> } 13:29 <+bridge> } else { 13:29 <+bridge> let new_val = MaybeUninit::new((key, value)); 13:29 <+bridge> self.tags[idx_mod] = true; 13:29 <+bridge> self.storage[idx_mod] = new_val; 13:29 <+bridge> self.len += 1; 13:29 <+bridge> return None; 13:29 <+bridge> } 13:29 <+bridge> 13:29 <+bridge> iter_idx += 1; 13:29 <+bridge> } 13:29 <+bridge> 13:29 <+bridge> panic!("no slot could be found") 13:29 <+bridge> final result 13:29 <+bridge> ill probs change it to a result 13:32 <+bridge> as the hash map gets fuller, it makes no sense to keep it as a hash map I guess 13:33 <+bridge> you'd have to manually ensure that it stays below the load factor 13:33 <+bridge> ye thats why resize happens 13:33 <+bridge> otherwise linear probing would be beneficial 13:33 <+bridge> without any hash 13:33 <+bridge> but i cant resize with stack based storage 13:33 <+bridge> yes 13:34 <+bridge> perhaps a tree based map would make more sense? 13:34 <+bridge> whats that? 13:34 <+bridge> `std::map` in C++ 13:34 <+bridge> or `BTreeMap` in rust 13:34 <+bridge> it works by constructing a tree 13:34 <+bridge> you have a `<` relation on the key type 13:37 <+bridge> i see 13:37 <+bridge> well maybe next time xd 13:38 <+bridge> Self balancing rb trees, go go 13:53 <+bridge> i hear red black trees are pain 14:15 <+bridge> They are cool tho 14:16 <+bridge> <_voxeldoesart> red black tees 14:30 <+bridge> it seems my simple hashmap passes miri 14:30 <+bridge> but maybe i need more tests xd 14:38 <+bridge> ``` 14:38 <+bridge> smolmap get 1000 time: [7.6065 ns 7.6679 ns 7.7887 ns] 14:38 <+bridge> Found 10 outliers among 100 measurements (10.00%) 14:38 <+bridge> 5 (5.00%) high mild 14:38 <+bridge> 5 (5.00%) high severe 14:38 <+bridge> 14:38 <+bridge> hashmap get 1000 time: [8.9185 ns 8.9328 ns 8.9479 ns] 14:38 <+bridge> Found 1 outliers among 100 measurements (1.00%) 14:38 <+bridge> 1 (1.00%) high severe 14:38 <+bridge> ``` 14:38 <+bridge> pog 14:39 <+bridge> Another miri fan? 😄 14:39 <+bridge> well miri is nice to check if ur unsafe is ok 14:39 <+bridge> its the main reason it exists i think 14:40 <+bridge> ``` 14:40 <+bridge> The implementation of the Rust abstract machine - miri - stops execution of Rust programs when they exhibit undefined behavior. 14:40 <+bridge> 14:40 <+bridge> C, C++, etc. don't even have implementations of their abstract machines. They don't have one existing as a goal. And they are happy to make certain operations exhibit undefined behavior even if that implies that it would make an implementation of the abstract machine that traps impossible. 14:40 <+bridge> 14:40 <+bridge> This is why even if you were to combine valgrind with address sanitizer, memory sanitizer, thread sanitizer, undefined-behavior-sanitizer, and other existing C and C++ tools, there is still a lot of classes of undefined behavior that these tools can't detect. 14:40 <+bridge> 14:40 <+bridge> That's fine in C and C++, but not fine in Rust. In Rust, if we add a new type of undefined behavior, the constraint is that it should be (demonstrably) possible to extend miri to detect it, such that if a user doesn't know whether some program exhibits undefined behavior for some inputs, they can just run it under miri, and miri will precisely pinpoint which part of their code exhibited undefined behavior and why, and how their program execution go 14:40 <+bridge> ``` 14:40 <+bridge> i love this quote 14:40 <+bridge> I know, I filed a few bugs with it yesterday 14:41 <+bridge> where did you file bugs with it? 14:41 <+bridge> i guess his work 14:51 <+bridge> sad thing is miri doesnt work with futures 14:52 <+bridge> but well most unsafe code *tm is usually not in async 14:52 <+bridge> or u can isolate it 14:52 <+bridge> idk 14:52 <+bridge> async started working recently I think 14:53 <+bridge> epoll_wait has been implemented: https://github.com/rust-lang/miri/pull/2764, that was blocking it in my cases 14:53 <+bridge> oh 14:53 <+bridge> thats epic 15:20 <+bridge> @learath2 if i implement a red black tree can i get into google? 15:20 <+bridge> :gigachad: 15:22 <+bridge> maybe if its a red, yellow, green, blue tree 15:23 <+bridge> lgbt tree 15:23 <+bridge> https://upload.wikimedia.org/wikipedia/commons/thumb/5/53/Google_%22G%22_Logo.svg/1200px-Google_%22G%22_Logo.svg.png 15:23 <+bridge> oh lmao 15:23 <+bridge> xd 15:23 <+bridge> i heard google asked to implement rd trees 15:23 <+bridge> the real g 15:23 <+bridge> and a guy who made something in javascript and was famous for it failed 15:23 <+bridge> and he complained in social media 15:23 <+bridge> xd 15:24 <+bridge> https://twitter.com/mxcl/status/608682016205344768?lang=en 15:31 <+bridge> Why would you want to be a corporate slave? 15:31 <+bridge> to get 500k$ per year 15:31 <+bridge> i just work 1 year 15:31 <+bridge> and retire 15:31 <+bridge> ok i can do with 300k 15:31 <+bridge> give or take 15:31 <+bridge> actually im half sure most of devs working for google arent that better than us here 15:32 <+bridge> they just got a degree in (famous name of uni here) and some contacts 15:32 <+bridge> but contacts are everything 15:33 <+bridge> He wrote homebrew, the worst package manager to ever be envisioned by humankind. I'd have rejected him just for that 15:33 <+bridge> ye xd 15:33 <+bridge> hn mentions homebrew sucked hard back then 15:33 <+bridge> they didnt even know how to manage ssl certts 15:34 <+bridge> > However, they don't usually remember that Homebrew at that point was a poor and incomplete reimplementation of other package managers, used to break installed software on updates, and the fact that the very team that was maintaining Homebrew was not competent enough to either manage SSL certificate for their website properly or at least admit that it was a bug. 15:34 <+bridge> They probably have bigger algobrains, able to golf 69 algorithms at the same time while brushing their teeths 15:34 <+bridge> im quite sure not 15:35 <+bridge> they just werent lazy enough and studied a bit to pass the leetcode exam 15:35 <+bridge> its literally the only entry bar 15:35 <+bridge> and maybe a good uni degree 15:35 <+bridge> but its how the world works anyway 15:35 <+bridge> there is a reason most interesting software is done by ppl in their hobby time 15:36 <+bridge> @learath2 when will u finish uni btw 15:37 <+bridge> if u ever get in a big company, give good word of me ty 15:37 <+bridge> September hopefully 15:37 <+bridge> "i know a crazy rust dev" 15:37 <+bridge> :justatest: 15:38 <+bridge> No way I get into anywhere decent finishing a 3y degree in 7 years. I'll be mopping floors at mcdonalds or living in estonia filling web templates at a "web design agency" 15:38 <+bridge> no 15:38 <+bridge> make a nice project in C 15:38 <+bridge> show it off 15:38 <+bridge> it worked for me 15:38 <+bridge> actually the barebones following a tutorial rust OS i made helped me find a job 15:38 <+bridge> it impressed some ppl 15:38 <+bridge> just by being interested in that 15:38 <+bridge> even if it sucked 15:39 <+bridge> anyway im sure u will be working rly early 15:39 <+bridge> it would be a waste 15:40 <+bridge> and absolutely mention ddnet 15:40 <+bridge> its a lifelong project, also telling ur story about it more or less helps 15:40 <+bridge> atleast it did for me 15:41 <+bridge> thanks for coming to my unsolicited advice ted talk 15:41 <+bridge> (also to me uni matters almost nothing, u will see why when u join workforce, there are devs, and "only" uni devs who have rly not much interest outside of work) 15:53 <+bridge> I think you'll be able to find a decent job given your experience 15:54 <+bridge>