DDraceNetwork - developer

DDraceNetwork

Development / developer

Development discussion. Logged to https://ddnet.tw/irclogs/ Connected with DDNet's IRC channel, Matrix room and GitHub repositories — IRC: #ddnet on Quakenet | Matrix: #ddnet-developer:matrix.org GitHub: https://github.com/ddnet

Between 2022-10-06 00:00:00Z and 2022-10-07 00:00:00Z

Ryozuki

yea

I hope you did some benchmarks

03:51

Else we can't compare

I didnt xd

another day on nixos

06:39

Image attachment

Wat nix can do

06:59

Sounds niche

Google it

//  Are you there god??/
??=define _(please, help)
??=define _____(i,m, v,e,r,y) r%:%:m
??=define ____ _____(a,f,r,a,i,d)
main(__)<%____(!_(-~-??-((-~-??-!__<<-
??-!!__)<<-??-(!!__<<!!__))+-~-~-??--~-~
-~-~-~-~-??-(-~-~-~-~-??-!!__<<-~!!__),-
??-!__))<%??>%>_(__,___)??<____
(printf("please let me die??/r%d bottle%s"
" of bee%s""""??/n",(!(___
%-~-~!!___))?--__+!___++:__+!___++,!(__-!!___)
&&___%-~-~!!___??!??!!(___%-~-~!!___??!??!__
-(-~!!___))?"":"s",___%-~-??-!!___<-??-!!___?
"r on the wall":"eeeeeeer! Take one down,pass ??/
it around")&&__&&_(__,___),"mercy I'm in pain")??<??>??>

justatest

1

07:23

valid c

07:23

gcc -w --trigraphs

try explaining this entire section to someone 200 years ago

Tolerance for weird errors

>

Note if you're going into more weird languages, you'll run into types of errors you almost certainly won't hit in the popular ones.

>

You'll get opaque error messages, especially if the language has invariants that aren't common in other environments (Rust's ownership checker is probably the most common example of "no analogue in a more common language," but this can also mean Mercury's determinism checks, or Pony's "reference guarantees"). To be clear: these parts are exactly why learning a new language is fun! This is what it costs to reap the reward of Perlis langauges (a language that changes how you think, from the quote: "A language that doesn’t affect the way you think about programming is not worth knowing.")

07:33

https://morepablo.com/2022/09/so-you-re-using-a-weird-language.html

So you're using a weird language

Tips and narratives around picking a weird language for your project

@Jupstar ✪ do you happen to know what the max amount of images is for maps?

11:11

I created one with too many and now it uses other textures from the game when it tries to display some of them

64

11:36

But i actually thought ddnet clamps

11:36

And not randomly access out of bounds

11:38

Well it clamps loading in mapimages.cpp

11:39

Maybe not at runtime dunno

11:40

Tbh the map should not even load if it's so buggy

huh I thought 64 was an arbitrary limit by the editor

Even if we clamp it will show wrong results

Patiga

huh I thought 64 was an arbitrary limit by the editor

Sadly no

11:40

Vanilla

ow

11:41

I just created a script to turn images/gifs/videos into menu backgrounds, but haven't found a suitable video file that is <= 2 seconds yet

11:42

hm, then I should probably also introduce that 64 limit in twmap

U could do texture atlas

11:45

8k should be supported by almost anything

hm true

11:46

that sounds like a big hassle though, I think I'll just stay with the limitation and drop remaining frames

11:47

didn't want to create complicated logic, only automate the current process

11:47

2022-10-06_13-33-26.mkv

1.61 MB

11:47

aww mkv doesn't get embedded

11:48

depending where I was in the menu a different texture was shown instead of the too many frames

11:48

this one showed me that those are indeed other game textures ^^

change the file extension to mp4 and it'll embed

fineee :p

11:49

poggers

2

We can must increase the limit for ddnet but check it in map realease tool

11:50

Just*

11:51

But yeah auto correction must is fine too

I mean my use case here isn't really something that should be accommodated to explicitly

11:51

but yeah I'll add the check I guess

Mhh 64 images are 260 mb

11:52

Guess 256 should work out xd

11:52

Ah actually double bcs mipmap

11:53

So 128

64 should already be 512MB i think

11:53

1920 * 1080 * 4 * 64 / 1024 / 1024 506.25

11:54

it adds up quite quickly

Yeah mipmap

11:55

So x2 for u

so 64 -> 1 GB

Maybe lower framerate

Patiga

so 64 -> 1 GB

Yes if you want fullhd res true

ah true

[ddnet-web:master] 1 new commit

4b109ee Add '^' to mods - def-

I just found this networking guide/rescourse https://www.gabrielgambetta.com/client-server-game-architecture.html. It's well written if anyone wants to check it out

13:15

It's for fast-paced multiplayer

13:19

It also has a cool demo showing the effect of lag, client predictions, reeconciliation, and interpolation https://www.gabrielgambetta.com/client-side-prediction-live-demo.html

it doesn't include any equivalent for anti-ping but for an FPS it's not needed because of lag compensation

14:34

it would be cool to do server side rollback or "lag compensation" in teeworlds/ddnet. I tried to make it once but saving and loading world states is not as easy on the server as it is on the client right now (edited)

14:36

I guess it's probably unreasonable for an FPS client to run a full physics simulation on the client like we do

server rollback probably doesn't really work that well with entity projecties

how come?

even our hooks isn't instant hit like lots of completive FPS do

ah I should have prefaced that the serverside rollback in ddnet would only be for preventing lag spikes. if you tried to sync the players and the server in real time then it would be a nightmare yes (edited)

i'm talking more about bullets projectiles

Why wouldn't server rollback work for entity projectiles though?

well, server rollback implies that it favors client views, so it tries to match what client sees when dealing with bullet collision. that might make the party taking damage see bullet hitting them that looks like missed.

Wouldn't the same concept of applying lag compensation apply to rollback projectiles?

i think lag compensation and collider rollback are the same thing.

14:42

maybe there are other server rollbacks that i don't know justatest

justatest

14:42

if you want to deal with lag spikes then usually what people do is just to accept client states instead of only inputs. but that also introduces complexity in physics and cheat detection.

the idea is that you only do rollback on the server if you were missing client inputs from past ticks (it would actually be always but just for simplification) then when a client lags and drops inputs if the client also sends their past inputs on every tick then the server can go back and re-simulate with their inputs that it didn't have before. This is kinda what overwatch does (edited)

oh i see

14:43

but in that case i think most game would just increase the play ahead on the client-side and call it a day.

yes but that's not acceptable for ddnet xd

kek

14:44

ye

also you can still do that

14:44

you can use both

definitely should fix the O(n^2) physics before doing anything rollback tho.

I mean the client seems fine with it

14:45

it runs the same O(n^2) physics up to 200 times every tick I think (edited)

14:45

maybe its not the same physics?

i mean server cpu are usually weaker.

I suppose

14:46

even 2 ticks of rollback is already a 4x improvement in "play ahead" buffer over current ddnet

14:47

under normal conditions

14:48

I have not profiled a server with many players so idk if the physics are actually the most expensive part

14:49

probably is I guess

and maybe snapping as well?

dunno xd

14:55

idk how most physics optimizations work but I guess the issue is that it's hard to determine if it's impossible for 1 tee to interact with another tee since with tunes and all sorts of things you basically can never rule it out. it's much harder than just rigid body collisions

14:56

maybe it is possible but you would have to be quite careful before skipping some physics (edited)

justatest

i mean you can probably assume the tuning wouldn't be so ridiculous like zooming across the screen and collide with a tee on the other side of the map within one tick

idk I haven't looked closely at which parts are O(n^2) and which are not

14:59

I assume projectiles are just O(n) ?

n is all entities, which includes character and projectiles and pickups

14:59

a simple quadtree could be a huge improvement already.

ah true

i mean you can probably assume the tuning wouldn't be so ridiculous like zooming across the screen and collide with a tee on the other side of the map within one tick

meanwhile sg bug, giving infinite speed upwards happy

happy

justatest

no sure which bug is that, and are we keeping that kind of speed as features as well

yea, its when you are at the exact same position as the other tee (inside of each other), and shoot sg

I thought we have to keep all features xd

i see

division by 0 and so on

there are too many "features" to keep track of

yeah that's a large part of the issue I think (edited)

15:02

everyone afraid to touch physics

i mean does div by 0 even has the same behavior in different compilers

also no tests

yea its undefined I think. it was discussed a few days ago that it was still not replaced with explicit behavior

i couldn't see we keeping a feature that might behave differently in the future or on different hardware lol

oh wait its probably not undefined, no? float math is pretty well defined iirc

i have no idea (edited)

Patiga

yea its undefined I think. it was discussed a few days ago that it was still not replaced with explicit behavior

kek

sounds like we are just gonna if that one and assign it to Infinity or something.

ah, its defined by the ieee standard but its not enforced in the c++ standard according to https://stackoverflow.com/questions/42926763/the-behaviour-of-floating-point-division-by-zero

it should be easy to keep that bug since it only happens when both tees are at exactly the same position, so it's easy to check

ye, but I don't really think we should keep it tho.

15:05

maybe as a map bug compatibility or something.

idk, making any change to the physics will be controversial

15:07

but if the sg bug prevents doing a more optimized physics then I think there will be many other things that also prevent it

Iirc div by 0 is ub

15:07

In c++

TsFreddie

a simple quadtree could be a huge improvement already.

it needs to refactor everything as i got it

15:42

i tried to integrate quadtree buut...

15:42

nothing is done right it seems

it would be nice if there was a small number above the friend heart that says how many friends are in this one server

Image attachment

that's a really good idea

Why bother optimizing physics unless we are allowing higher player counts? We have enough headroom per tick to do whatever we want, unless it becomes too tight I really wouldn't bother with it

Iza

it would be nice if there was a small number above the friend heart that says how many friends are in this one server

Nice idea, could you open an issue on github for it?

16:03

If not I'll try not to forget to create one after this lecture

ok i'll do

[ddnet/ddnet] Issue opened: #5911 Add a "counter" above the Friend heart in Server list

it would be nice if there was a small number above the friend heart that says how many friends are in this one server (see screenshot for reference) !DDNet_wnGq83UXPv

i find this notice is not nicely positioned, at least in german idk how it is with english, maybe find a better place for it

Image attachment

suggestion: when people are voting for a map change - maybe the server could /save your progress (if possible) if the map change vote goes through so you dont lose progress

Iza

i find this notice is not nicely positioned, at least in german idk how it is with english, maybe find a better place for it

or just make the proportions better with margins idk (edited)

[ddnet/ddnet] Pull request opened: #5912 Only show ingame hint in Game tab

Since it looks ugly overlaid over the other tabs

Checklist

[x] Tested the change ingame
[ ] Provided screenshots if it is a visual change
[ ] Tested in combination with possibly related configuration options
[ ] Written a unit test (especially base/) or added coverage to integration test
[ ] Considered possible null pointers and out of bounds array indexing
[ ] Changed no physics that affect existing maps
[ ] Tested the change with [ASan+UBSan or valgrind'...

Thanks @Iza ^

@Patiga where is the tool?

i think nixos is better suited for servers

17:38

actually its like a glorified docker xd

17:38

ill switch to gentoo without systemd

thonk

1

1

1

1

1

17:41

ill buy a 64gb usb3.2 and install this https://github.com/ventoy/Ventoy

GitHub - ventoy/Ventoy: A new bootable USB solution.

A new bootable USB solution. Contribute to ventoy/Ventoy development by creating an account on GitHub.

17:41

so i dont need to bother again

17:45

anyone know smth better than this https://www.amazon.es/Memoria-SanDisk-Extreme-velocidades-Lectura/dp/B01MU8TZRV

17:45

but at this price range

Indirect leak of 5447 byte(s) in 13 object(s) allocated from:
    #0 0x7f1467217817 in __interceptor_malloc (/usr/lib/gcc/x86_64-pc-linux-gnu/11.3.0/libasan.so.6+0xb2817)
    #1 0x7f1444275ff7  (<unknown module>)

17:51

I get these bizarre leaks, it smells like they are in external libraries, but why do they end up being from unknown_module?

Ok, if ever you get stumped by the output of ASan or LSan, report_objects=1 for LSan and fast_unwind_on_malloc=0 on ASan options helps a lot

18:52

(maybe that's how we should run it on CI too? it is slower, but it'll give stack traces that are actually useful in some cases)

18:56

I call every cleanup function properly and curl still leaks, maybe rust people were right, there are no good C libraries

Ok, actually I do cleanup properly, leak sanitizer just runs too early, hm

19:16

Why on gods green earth is leak sanitizer running before all global destructors happen?

why do u rely on exit time destructors

Learath2

I call every cleanup function properly and curl still leaks, maybe rust people were right, there are no good C libraries

good good

Ryozuki

good good

lmao, maybe read 1 sentence later

no need

19:27

BASED

Learath2

Why on gods green earth is leak sanitizer running before all global destructors happen?

just use valgrind

19:28

it also shows uninitialized values without needing an own compiled memsanitizer compatible os xd

19:29

@Ryozuki if u create a thread, do you also destroy it manually? or do you think it's better to leave it to the OS?

i mean a thread is destroyed by joining it

19:30

maybe with a timeout of 0 xd

yeah i mean a co thread u use the whole time

It's bizarre, it does do the leak check after the destructor, but it reports things that are guaranteed to be freed after the destructor as still leaks

Jupstar ✪

yeah i mean a co thread u use the whole time

i use a signal for that usually or smth

question is would a perfect OS, also trust its software so much that it doesn't care

Kinda off-topic what's teewolds connection sceme? What protocols are used?

need to make the perfect OS

Mr.Gh0s7

Kinda off-topic what's teewolds connection sceme? What protocols are used?

UDP

19:31

no "real" connection as in tcp

19:32

basically own connection logic

Jupstar ✪

no "real" connection as in tcp

We do a handshake and establish some sort of reliable transport on top of UDP, but yeah, it's not tcp

Ok thanks

@Jupstar ✪ https://agourlay.github.io/brute-forcing-protected-zip-rust/

Brute forcing protected ZIP archives in Rust

Yet another programming blog

19:34

i found this blog post some days ago

19:34

he uses threading

19:35

and no, he doesnt guess the password its not about that

Ok, I'm out of ideas and bored at this point

19:46

I allocate the exact same place I call curl_global_init and free where I call curl_global_cleanup, my allocation is not detected as expected, but stuff allocated by curl through global_init and freed in global_cleanup get detected as leaks

Exported 186 message(s)