DDraceNetwork - developer

DDraceNetwork

Development / developer

Development discussion. Logged to https://ddnet.org/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 2024-06-05 00:00:00Z and 2024-06-06 00:00:00Z

even if you only save 6 bits per mouse int you save 192 bits over the whole bundle which is more than the 32 for the zoom value

[ddnet/ddnet] Pull request closed: #8395 Refactor client input event handling

05:45

github-merge-queue[bot]

[ddnet:master] 3 new commits

248225c Refactor client input event handling - Robyt3 0d9f673 Include types.h instead of system.h in input.h - Robyt3 5eda5e9 Merge pull request #8395 from Robyt3/Client-Input-Event-Refactoring - def-

Tater

and then just assume that clients won't change their zoom value extremely fast, a mouse position with wrong magnitude is still much better than no input at all (edited)

multiview is constantly changing the zoom level though. though i guess in this case mouse events aren’t being sent in the first place

I don't think input relability is a concern when you're spectating

well i definitely do not want any misinterpretations of my inputs

06:02

and i do change zoom as i play

06:03

with the scroll wheel so it’s many smaller changes

[ddnet/ddnet] Pull request closed: #8448 Remake of Ukrainian l10n

06:04

github-merge-queue[bot]

[ddnet:master] 2 new commits

97623c3 Remake Ukrainian l10n - EGYT5453 c7dc7b6 Merge pull request #8448 from EGYT5453/uapdte - def-

Ewan

well i definitely do not want any misinterpretations of my inputs

yeah but it's only when your network is lagging or dropping packets (edited)

06:34

the less unnecessary data we send the more protection you can get from lag

06:35

under regular conditions nothing changes

i think there is a good way to do it without factoring zoom in

ok how so?

the bitpacking and xoring then compressing methods seemed promising

I'm doing that already

and is that not enough

06:36

ultimately this is not a lot of data

well it is

06:36

kind of a lot of data

Gm tater

cat_woah

1

late for us

06:37

gm teero

1

presumably deen is going to tell me that we can't have 2000ms of input bandwidth but I want to get as much as we can

06:38

the current bandwidth is 0.5kB-1.5kB/s. if you send a full 100 inputs per tick then it goes up quite a lot to like 15kB-20kB/s

06:38

(for upload)

hmm

06:38

currently we are doing half that and there’s not a huge need to go higher afaict

06:39

tho i missed a lot of the convo

wdym

06:39

half what?

servers run at 50 tps

06:39

does client send 100 and it processes 2 frames in 1 tick? (edited)

no I mean every tick the client would send 100 inputs if you were doing a full 2000ms of input bundling but that's not really nessecary anyway

oh i misread

so it would go from 50 inputs sent per second to 5000

06:40

but obviously we don't need 100x

06:41

300-400inputs per second is enough to get very good relability

06:42

I think targeting a max of 500ms (25 buffered inputs) is reasonable. any bigger than that is completely unplayable (edited)

so what does this rly end up looking like

06:44

like is bandwidth rly the root of most of ddnet’s connection problems

idk

06:44

the root of the connection issues is the lack of any relability in the critical input packets

06:45

which is what this is trying to fix

it’s rare to see connection upload speeds that can’t handle 1 or 2 clients with ease and outbound player data can be sent at a way lower frequency by the server

06:45

and already is afaik

wdym handle 1 or 2 clients

dummy

it's about the server's bandwidth not the client

are ddnet servers not run on servers with 100s of Mbps connections

06:46

idk

having all clients send 20kB/s to the server would be a max of 10.24megabits/s which is sort of a lot

Ewan

are ddnet servers not run on servers with 100s of Mbps connections

idk

06:47

deen implied they are not

i mean you’re right that we should get the bandwidth as conservative as possible

06:48

but not at the cost of playability like i think factoring zoom might do

06:48

another variable

it will strictly only improve your playability over the current version

06:48

it can never make it worse than it already is

well actually how can it even cause problems if zoom is factored in

06:49

is zoom being sent with each packet or as it’s changed

the issue is that the mouse position is stored as 2 32 bit ints which are large and change on nearly every tick

Ewan

is zoom being sent with each packet or as it’s changed

in this hypothetical scenario

it will be sent only once per packet and applied to all inputs in that packet instead of for each input

06:50

that way the mouse positions can be 12 bits instead of 32

ah

this is just in theory I'm not sure that's the best idea yet

maybe the client should keep track of a separate zoom for the server then

the zoom that the client already sends to the server is totally unrelated to this

06:51

it won't be touched

06:51

this is a new zoom value

06:51

only for mouse positions

right

06:52

but if the zoom can’t change between packets then there won’t be a problem

06:52

client can render whatever zoom

well likely you are not spamming zoom in/out more than 3 times per second very often

ehh

and even if you do the only difference is that the server will use a mouse angle which is millimeters wrong which makes the desync very small and probably not noticeable (edited)

06:54

you still get the same "angle" it's just slightly different because it gets rounded to the nearest integer position

06:54

where each pixel on your screen is an int

06:54

so its literally wrong by less than a pixel

in some cases the value can actually be floating point iirc

06:55

i think SDL sends floats

and it's only wrong if you drop packets and then change zoom while your network is dying and then a later input fills in the server buffer in that time

yeah

Ewan

in some cases the value can actually be floating point iirc

it's not

06:56

it uses ints

maybe ddnet does

your mouse pos is rounded to an int every time it moves afaik

06:56

that's why the aim bind works

Tater

you still get the same "angle" it's just slightly different because it gets rounded to the nearest integer position

Btw scaling target pos with zoom is still broken af

wdym

You can have targetxy == 0 at some zooms

06:57

And generally zooming in is broken with mouse angles

06:57

Because everything is truncated to integers

this might make hookline even less reliable lol

Ewan

this might make hookline even less reliable lol

It already is xd

targetxy == 0 is explicity forbidden in the code

Ewan

are ddnet servers not run on servers with 100s of Mbps connections

European servers are mostly severed with high bandwidth like 100 to 1000 mbit/s. But in asia bandwidth is very expensive, esp. china

Tater

targetxy == 0 is explicity forbidden in the code

Only mouseposxy == is forbidden

not true?

targetxy does not care

Image attachment

06:59

it's removed on the client and server multiple times even

07:00

the idea is just you send the targets before doing this calculation, then do it on the server

Image attachment

Tater

the idea is just you send the targets before doing this calculation, then do it on the server

This can lead to targetxy being 0

how

??

??

07:01

it wouldn't even matter because it gets sanitized by the server as soon as it recieves the input (edited)

(int)(1 * 0.6) == 0

Tater

it wouldn't even matter because it gets sanitized by the server as soon as it recieves the input (edited)

But it messes but the hook line smh When I put my mouse in the centre the hook line disappears and I hook to the top

07:02

When zoomed in

I am not even planning to send hookline data in relability inputs

07:02

we do lots of corner cutting so you get more reliability

07:03

these inputs will only get used if the server didn't recieve your other ones

just zooming in or out can make some edges possible or not That is bs

that's fine

07:04

it won't change that

You are disadvantaged when playing with more zoom than 10

this is a seperate conversation

ye xd

07:05

I'm just ranting about the inputs scaling by zoom dw

also you're not disadvantaged, you still only get as many positions as there are pixels on your screen. they just have slightly different angles than at different zoom levels. you're actually disavantaged if you use a low resolution monitor

Tater

also you're not disadvantaged, you still only get as many positions as there are pixels on your screen. they just have slightly different angles than at different zoom levels. you're actually disavantaged if you use a low resolution monitor

Why is that? The positions stay the same not dependant on res

hmm maybe that's true actually

07:06

I guess the game assumes you use 1920 res

Tater

also you're not disadvantaged, you still only get as many positions as there are pixels on your screen. they just have slightly different angles than at different zoom levels. you're actually disavantaged if you use a low resolution monitor

If you zoom to zoom 15 you have literally less angles for your hook to be in since integer precision is limited

I wonder if you can see the cursor jumping multiple pixels when you move it slightly on a higher res screen

Tater

I wonder if you can see the cursor jumping multiple pixels when you move it slightly on a higher res screen

It will be the same distance on the monitor but yes

Tater

I wonder if you can see the cursor jumping multiple pixels when you move it slightly on a higher res screen

Yes

Imagine you have a max mouse distance of 400. When zooming in 2x you would have the same precision as someone playing with mouse max distance 200

why’s that

07:11

oh yeah,

07:11

hmm

Tater

the idea is just you send the targets before doing this calculation, then do it on the server

.

not sure if you can do anything about that without changing physics a lot

Tater

not sure if you can do anything about that without changing physics a lot

???

??

It was just added lmao

what?

07:12

when

The zoom scaling

I thought the game sends mouse positions as int forever

Tater

I thought the game sends mouse positions as int forever

Yes

oh

07:12

it was?

Targetxy only gets scaled for the tp to mouse position cmd

07:13

That's the only reason

hmm

Tater

hmm

You could just send your zoom to the server and the server can handle it.

would need to put it somewhere

07:14

it should be in however the server knows your screen window

07:14

for removing entities

Oh that's true. So it could already be done?

heinrich said it doesn't send the zoom value directly only the window position

What's happening here?

07:15

The client sends it's cursor in zoomed state nowadays

yeah

It used to not do it

yeah

exactly

Tater

heinrich said it doesn't send the zoom value directly only the window position

I guess you can figure that out

fokkonaut

It used to not do it

the idea is it could just send the non-zoomed value along with m_pClient->m_Camera.m_Zoom then the server can get the exact position by itself

07:17

not that I'm suggesting it should do this (edited)

07:17

just that it could

Might be good, but if you plan on doing something in these regards, then please make sure dyncam will be supported properly, because it isnt. Neither with the way it was before nor now with the zoomed cursor being sent

wdym

You can't get the correct position of dyncam on the serverside, only if the server knows the client uses dyncam (e.g. via a command to enable)

weird

07:19

you could steal it from checksum with a lot of effort I think

1

fokkonaut

You can't get the correct position of dyncam on the serverside, only if the server knows the client uses dyncam (e.g. via a command to enable)

How long has this been a thing?

only cursor + zoom is fucked iirc

hi fokkonaut

hi

I expect you would be the biggest proponent of the mouse zoom with your drawing feature on your server

https://github.com/fokkonaut/F-DDrace/blob/de3299d92ed85bd7c64c1f098ae7990d2b5155d6/src/game/server/gamecontext.cpp#L2937-L2945 https://github.com/fokkonaut/F-DDrace/blob/de3299d92ed85bd7c64c1f098ae7990d2b5155d6/src/game/server/player.cpp#L906C16-L906C26 https://github.com/fokkonaut/F-DDrace/blob/de3299d92ed85bd7c64c1f098ae7990d2b5155d6/src/game/server/entities/character.cpp#L3930-L3975

fokkonaut

https://github.com/fokkonaut/F-DDrace/blob/de3299d92ed85bd7c64c1f098ae7990d2b5155d6/src/game/server/gamecontext.cpp#L2937-L2945 https://github.com/fokkonaut/F-DDrace/blob/de3299d92ed85bd7c64c1f098ae7990d2b5155d6/src/game/server/player.cpp#L906C16-L906C26 https://github.com/fokkonaut/F-DDrace/blob/de3299d92ed85bd7c64c1f098ae7990d2b5155d6/src/game/server/entities/character.cpp#L3930-L3975

nice

ah

07:23

so you can get exact position without zoomed mouse?

https://github.com/fokkonaut/F-DDrace/blob/de3299d92ed85bd7c64c1f098ae7990d2b5155d6/src/game/server/entities/character.cpp#L3969-L3973

07:23

This is where it checks for the new clients

07:24

if they are new, we don't multiply it with the zoomlevel because they send it correctly

hmm

07:24

rather annoying that ddnet did not do this

For older clients (before that change) I manually calculated the zoom level

morning from gym

hi

fokkonaut

For older clients (before that change) I manually calculated the zoom level

That seems to be the correct way. Physics should use a non-scaled version

if you change it back then you need 3 ways to calculate mouse position

Ryozuki

morning from gym

Gm

Teero

That seems to be the correct way. Physics should use a non-scaled version

yeah I agree

Teero

That seems to be the correct way. Physics should use a non-scaled version

Yes, that's true. I was not really happy when I saw them merge this without issues

Tater

if you change it back then you need 3 ways to calculate mouse position

remove dyncam

please

what does dyncam have to do with this

fokkonaut

https://github.com/fokkonaut/F-DDrace/blob/de3299d92ed85bd7c64c1f098ae7990d2b5155d6/src/game/server/gamecontext.cpp#L2937-L2945 https://github.com/fokkonaut/F-DDrace/blob/de3299d92ed85bd7c64c1f098ae7990d2b5155d6/src/game/server/player.cpp#L906C16-L906C26 https://github.com/fokkonaut/F-DDrace/blob/de3299d92ed85bd7c64c1f098ae7990d2b5155d6/src/game/server/entities/character.cpp#L3930-L3975

.

it just needs more data send to the server it can stay

You need follow factor, deadzone etc

07:26

otherwise u cant calculate that pos

07:27

client should just send two

07:27

one zoomed and one unzoomed cursor pos

07:27

and unzoomed = physics

07:27

zoomed = do whatever u want

We just talked about how huge those 2 32bit ints for targetxy are that's why I was not suggesting this

ah

07:28

well, if u'd need to send follow factor, deadzone etc then it's not gonna be less

fokkonaut

well, if u'd need to send follow factor, deadzone etc then it's not gonna be less

that is also true I guees

Could send it once, or only on change

1

fokkonaut

well, if u'd need to send follow factor, deadzone etc then it's not gonna be less

But you only need to send them on joining and on changing them

but idk if thats worth it

07:28

yea

the zoom level changing edge hooks seems really bad actually?

07:29

like that was a bad change

Tater

the zoom level changing edge hooks seems really bad actually?

yes

yes

yes

07:29

do we have an issue for that?

i always have to play around with different zoom to hit bad steep hooks

yea :D

well in that case it gives you an advantage lol

07:30

before it was like you had zoom =10 always

Tater

well in that case it gives you an advantage lol

Well no xd

Teero

Well no xd

yes

if you play on zoom 10 it's the same as before

07:30

but if you line up an edge hook then zoom in/out and it goes away I feel that's not good

2

Tater

if you play on zoom 10 it's the same as before

I guess you can get lucky with the int truncation and then hit the edge hook lmao

lots of assumptions that everyone uses zoom 10 lol

[ddnet/ddnet] Issue opened: #8452 Don't send mouse target in zoomed state

!image Precision has been changed due to this. We should revert this change where the mouse position gets sent in a zoomed state. A fix would be to keep the normal mouse pos for everything physics related and either send required values such as follow factor, deadzone etc so the server can create the real position (also for dyncam, as that's currently not working correctly and gives wrong positions),...

1

fokkonaut

one zoomed and one unzoomed cursor pos

Where would you store the scaled version? In CNetObj_PlayerInput? Would that break compatibility?

Yes, that would break it

07:39

I think so

07:40

i'd create a new Ex object i guess ? havent really looked into all that in a while

So a new netmsg just for camera related changes so guess? (edited)

Before doing that we should probably wait on @heinrich5991 to approve smth in the issue

I will always happily waste my time doing fun stuff

troll

1

:D

@Teero you have to be careful because some clients will still be sending the scaled mouse inputs

07:49

that's why I am a little upset that this was ever allowed cause it's going to require permanent backwards compatibility code to be fixed (edited)

That's the reason I do not understand why @heinrich5991 let that one slip.

07:53

He himself said we have to discuss such things properly before.

justatest

08:04

gm

Tater

that's why I am a little upset that this was ever allowed cause it's going to require permanent backwards compatibility code to be fixed (edited)

Actually only 1 check for older clients that don't send scaled inputs. If the client doesn't send the camera netmsg we know it will be scaled except the version is lower than x

[ddnet/ddnet] Issue closed: #8436 "DDNet" tab name is translatable, while "Tee" is not

zhn

justatest

brownbear

Teero

Actually only 1 check for older clients that don't send scaled inputs. If the client doesn't send the camera netmsg we know it will be scaled except the version is lower than x

seems a little annoying to track if the client sends the camera netmsg

08:13

you need extra state somewhere right?

Tater

you need extra state somewhere right?

Ye Just some bool m_IsTargetScaled

idk where the server puts such stuff

And each client will have some Params such as deadzone, follow factor etc

I think you need to add it to some sort of reset function for handling swaps

08:15

idk it's not my concern :)

Teero

Actually only 1 check for older clients that don't send scaled inputs. If the client doesn't send the camera netmsg we know it will be scaled except the version is lower than x

inbetween versions(?)

zhn

inbetween versions(?)

wdym?

Teero

wdym?

scale thing wasn't there forever was it

08:17

i mean even older clients can play thia game justatest

justatest

08:17

or does ddnet check it already

08:17

latter i think xd

it's the already scaled clients who you need to check for

zhn

i mean even older clients can play thia game justatest

justatest

Older clients inputs just don't get scaled by the server. Since they are not scaled by the client it's fine.

so you propose to send additional data about camera parameters every input for newer clients

zhn

so you propose to send additional data about camera parameters every input for newer clients

Only on camera settings change and on join.

Teero

Only on camera settings change and on join.

hm i wonder if theres something you can manipulate server with

08:23

not on ddrace though

@heinrich5991 didn't you want to unsort #8448 before merging? The diff ended up being massive for no reason

https://github.com/ddnet/ddnet/pull/8448

Remake of Ukrainian l10n by EGYT5453 · Pull Request #8448 · ddnet/d...

I remade the Ukrainian translation, because its quality was "meh". Thanks to everyone who helped me on the way (especially @BlaiZephyr, @heinrich5991 and @Robyt3 ❤️) Checklist Tested th...

It's also the only language file with a different order of strings

I wanted to, but I forgot to do it yesterday and now deen has merged it

08:50

no way to undo, I guess

08:54

hmmm. we don't have a script to sort the translations, i seems?

08:54

weird oversight

08:54

@Learath2

08:54

I was thinking we could revert the commit and add it sorted again

08:54

but I can't find the sort function

I doubt there is a sort function, is it even sorted normally?

yes, seems to be sorted in other language files

Does anyone know if the automapper rules format is described anywhere? Or is this a "custom/yolo" format and I need to hope, that ddnet get's it right?

yolo format

xD okay, an example of what may break is the length of the automapper rule name, I probably need to extract limits out of the client (edited)

heinrich5991

yes, seems to be sorted in other language files

Sorted by what though? german.txt:L482 is Yes. L488 is Name Plate

hmmmmmm

Maybe sorted within the file it's grabbed from?

[ddnet/ddnet] Issue opened: #8453 Language files aren't sorted

This gives a degree of freedom, making diffs worse. A script should be added and sorting should be checked in CI.

i found a gigachad

09:07

https://ageinghacker.net/

Luca Saiu free software hacker, GNU maintainer and computer scientist

Luca Saiu is a computer scientist specialised in programming languages, free software hacker and GNU maintainer.

09:07

can u imagine me having this blog when im old

09:08

https://ageinghacker.net/talks/#jitter-gnu40-2023

when phd thesis

09:12

and when ego check

Ryozuki

can u imagine me having this blog when im old

the guy is old in that picture?

09:13

damn gigachad look

Ewan

and when ego check

xd

pepeW

owo

Ewan

when phd thesis

phd doesn't pay off in computer science, at least where I life saddo

saddo

(edited)

I can't imagine a PhD in anything but Chemistry paying

Biology, if you're lucky, making the next vaccine (edited)

ty for spelling

It looked so wrong, I couldn't even tell what was wrong

AssassinTee

Biology, if you're lucky, making the next vaccine (edited)

I doubt any PhD that worked on any of the vaccines got anywhere near fairly paid

probably true. At the end this is case-by-case, I also know a person who made a PhD in IT who now makes 6-figures, he is specialised in natural language processing with neural networks

09:22

but only, because he didn't stay in science and works in the industry (edited)

I don't think you need a phd for that

Machine learning people do get paid really well yeah. I wish I was more interested in that stuff

Learath2

I can't imagine a PhD in anything but Chemistry paying

medicine?

Ryozuki

medicine?

Oh yeah duh

heinrich5991

I don't think you need a phd for that

Of course you don't need a phd for that, you just need a computer and brain power. But the topic he was working on was very very specific and he was an expert in this field and he keeps working on it at his job

Learath2

Machine learning people do get paid really well yeah. I wish I was more interested in that stuff

for me the only interesting part of ML are compilers related to optimizing that stuff

09:24

MLIR

not interested in the hardware optimizing this stuff? Special TPUs and architectures?

https://mlir.llvm.org/docs/Dialects/MLProgramOps/

'ml_program' Dialect - MLIR

Multi-Level IR Compiler Framework

09:25

there is a new dialect in MLIR

09:25

https://mlir.llvm.org/docs/Dialects/IRDL/

'irdl' Dialect - MLIR

Multi-Level IR Compiler Framework

09:25

IR Definition Language Dialect IRDL is an SSA-based declarative representation of dynamic dialects. It allows the definition of dialects, operations, attributes, and types, with a declarative description of their verifiers. IRDL code is meant to be generated and not written by hand. As such, the design focuses on ease of generation/analysis instead of ease of writing/reading. Users can define a new dialect with irdl.dialect, operations with irdl.operation, types with irdl.type, and attributes with irdl.attribute.

09:26

irdl.dialect @cmath {
  irdl.type @complex {
    %0 = irdl.is_type : f32
    %1 = irdl.is_type : f64
    %2 = irdl.any_of(%0, %1)
    irdl.parameters(%2)
  }

  irdl.operation @mul {
    %0 = irdl.is_type : f32
    %1 = irdl.is_type : f64
    %2 = irdl.any_of(%0, %1)
    %3 = irdl.parametric_type : "cmath.complex"<%2>
    irdl.operands(%3, %3)
    irdl.results(%3)
  }
}

09:26

looks pog

09:26

a ir to define irs dynamically

AssassinTee

phd doesn't pay off in computer science, at least where I life saddo

saddo

(edited)

rly?

09:26

i guess it depends what niche ur studying

AssassinTee

Of course you don't need a phd for that, you just need a computer and brain power. But the topic he was working on was very very specific and he was an expert in this field and he keeps working on it at his job

ah, I see

AssassinTee

probably true. At the end this is case-by-case, I also know a person who made a PhD in IT who now makes 6-figures, he is specialised in natural language processing with neural networks

like ml i was about to say xd

https://mlir.llvm.org/docs/Dialects/PolynomialDialect/

'polynomial' Dialect - MLIR

Multi-Level IR Compiler Framework

09:27

a dialect for polynomial operations

Solving any problem with a neural network always feels like a cop out to me. It's like saying the best minds of humanity couldn't figure it out. Just make some fuzzy guesses at it

// A constant polynomial in a ring with i32 coefficients and no polynomial modulus
#ring = #polynomial.ring<coefficientType=i32>
%a = polynomial.constant <1 + x**2 - 3x**3> : polynomial.polynomial<#ring>

// A constant polynomial in a ring with i32 coefficients, modulo (x^1024 + 1)
#modulus = #polynomial.int_polynomial<1 + x**1024>
#ring = #polynomial.ring<coefficientType=i32, polynomialModulus=#modulus>
%a = polynomial.constant <1 + x**2 - 3x**3> : polynomial.polynomial<#ring>

// A constant polynomial in a ring with i32 coefficients, with a polynomial
// modulus of (x^1024 + 1) and a coefficient modulus of 17.
#modulus = #polynomial.int_polynomial<1 + x**1024>
#ring = #polynomial.ring<coefficientType=i32, coefficientModulus=17:i32, polynomialModulus=#modulus>
%a = polynomial.constant <1 + x**2 - 3x**3> : polynomial.polynomial<#ring>

Learath2

Solving any problem with a neural network always feels like a cop out to me. It's like saying the best minds of humanity couldn't figure it out. Just make some fuzzy guesses at it

but it's working quite well ^^

09:29

it would be weird to not use it just because we can't explain its inner workings

maybe elon musk's ai to solve the universe is a cop out scheme xd but ml definitely has use cases

heinrich5991

it would be weird to not use it just because we can't explain its inner workings

Indeed. It just doesnt align well with how I think

there are also only so many great minds but millions of gpus

I'd guess there are at least billions of GPUs and at least millions of great minds ^^

In my mind AI is, by design, an imperfect system. It's not 100% accurate or even mathematically correct, but just a very very good guessing machine, because we can't do better then guessing good. You can throw an AI at problems, where it's known, that no perfect solution exists

10:04

That's also why chat AIs haluzinate, never learned the real answer? Make a best efford guess (which might be totally bs)

I guess you're talking about LLMs? (edited)

10:05

"AI" is quite imprecise

to be more specific, i am talking about neural networks

e.g. support vector machines are very explainable

you can throw a support vector machine at a problem, where it's known to to work so well. Same issue, it just makes the best guess

10:06

because a SVN normally just does linear operations even when the problem is parabolic in nature

Afaik it's mathematical fact that a neural network can learn to perform any computable operation perfectly if it has enough neurons

ye it can learn any function

yes, and that's called overfitting

no

AssassinTee

you can throw a support vector machine at a problem, where it's known to to work so well. Same issue, it just makes the best guess

I disagree that SVM have the same issues as neural networks

overfitting would be learning a different function

SVM are explainable, you know what they can and cannot do

10:07

NN are not explainable

10:07

the difference of explainability is what makes NN so annoying

heinrich5991

NN are not explainable

It's such a shame that they don't learn in any logical manner

I am not talking about explainability. A small NN is nothing different then a multilayer-perceptron, which is also explainable. I am talking about unperfect solutions

10:09

(a multilayer-perceptron is a small NN to be correct here)

i think he means explainable in a human context, not as in you don't know how it works under the hood

I mean explainable as in you know how it solves tasks, i.e. you know what failure modes to expect, etc.

10:10

if you have a tiny NN, it might well be explainable, yes

to be fair, i don't think that there is a 'correct' way to explain how to solve a task

louis

overfitting would be learning a different function

If your model has more parameters then the size of the problem, the model can simply learn every solution, like a map. You don't gain the generalization

AssassinTee

I am not talking about explainability. A small NN is nothing different then a multilayer-perceptron, which is also explainable. I am talking about unperfect solutions

sooo heuristics are bad?

10:11

heuristics are used in everyday CS, say, starting from A*

10:11

or solving SAT problems

AssassinTee

If your model has more parameters then the size of the problem, the model can simply learn every solution, like a map. You don't gain the generalization

yes, but a NN can still learn any function

10:11

including the generalized function

yet, I wouldn't call modern SAT solvers bad, even though they can never tackle the general problem

10:12

they only tackle the practical problems

heinrich5991

sooo heuristics are bad?

no, but @Learath2 said he couldn't wrap his brain around why the best brains of humanity couldn't figure it out, so we do the next best thing: guessing good

same for SAT solvers ^^

10:12

SAT solvers are also about guessing good

It's not that I can't wrap my brain around it. It's that it feels like giving up on a problem and just throwing guesses at it instead

I guess that feelign would also apply ot SAT solvers then?

10:13

you can kill a lot of problems with them

I do understand these are incredibly hard problems that may even lack solutions

Well, now I can go on a lil vacation with peace in mind (edited)

Replying to @GitHub GitHub: > def- […]

The "gain" from NNs is, that they predict in a linear manner (just matrix multiplications, at least in the easiest cases), while you can technically throw them at problems way worse by nature (edited)

heinrich5991

I guess that feelign would also apply ot SAT solvers then?

Yes, but atleast these do something easy enough to understand guided by heuristics

@heinrich5991 are NN SAT solvers even used in any important context

SAT solvers do also guarantee you the correct solution if you use them correctly

It's closer to A*

or is it like a 'only false negative no false positive' thing

NNs generating computer understandable proofs also give you "no false positives"

A neural network is more like, you feed it input, you get output, there is some probability that the output is correct provided that your input is within some domain

I'd put that under explainability

10:17

since we have no idea how they work, we can't put error bounds on what they do

10:17

for a SVM, we can try to do that

heinrich5991

NNs generating computer understandable proofs also give you "no false positives"

This feels even more wrong

Bolt on even more to tame the inherent fallibility of them. Idk they just feel wrong to me. It's a feeling, you can't really explain it away

or a decision tree/forest

10:19

I bet there are NN guided SAT solvers out there

The randomness of it all just gives me a sense of general unease. Almost feels like in the future we won't be able to understand anything about computing at all. It'll just be throw data at the newest machine learning paradigm, if it doesn't work, oh well, nothing else we can do

it may just be my opinion but i feel like the growth of new ML techniques will die down somewhat soon (10 years)

10:23

there's only so many new things you can do with linear algebra, as of now it just seems like we're only increasing the training compute

It would be such a shame if we'd never see things as beautiful as say djikstras algorithm. There is no beauty in neural networks, they learn in a random manner, they improve with bruteforce conditioning

using ML for fields like mathematics or computing where you need accuracy is usually a losing game, i doubt it'll take over in that sense

louis

it may just be my opinion but i feel like the growth of new ML techniques will die down somewhat soon (10 years)

I do secretly hope that there is a mathematical great barrier somewhere along the line there. Like the halting problem for computing or the incompleteness theorem

just wait until openai releases chatgpt6 with humanoid girlfriends, you'll change your mind

I'll allow that one

10:27

Anyway, these are all very subjective opinions of mine. In reality I do understand very well that some of these problems are way beyond our current level. Perhaps even beyond a mathematical great barrier of their own making them only solvable by an approach like this

MTC guy DREAMS about this (edited)

Replying to @louis just wait until openai releases chatgpt6 with humanoid girlfriends, you'…

oh yeah our amazingly huge forum community lmao

Image attachment

kek

13:24

might want to update the random servermsg's

13:24

,/info should maybe display discord

https://github.blog/2024-06-03-arm64-on-github-actions-powering-faster-more-efficient-build-systems/

Larissa Fortuna

Arm64 on GitHub Actions: Powering faster, more efficient build systems

GitHub Actions now offers Arm-hosted runners with images built by Arm for developers to begin building on the latest and most sustainable processors on the market.

13:53

@heinrich5991 @Learath2 when ddnet arm test in ci

13:54

GitHub is ecstatic to unveil ArmⓇ-based Linux and Windows runners for GitHub Actions. This new addition to our suite of hosted runners provides power, performance and sustainability improvements for all your GitHub Actions jobs.

I don't even know how well ddnet runs on aarch64

ecstatic

when less corporate language?

We expect to begin offering Arm runners for open source projects by the end of the year

(edited)

13:54

oh rip

Learath2

I don't even know how well ddnet runs on aarch64

runs nice on my m1

why not "happy"? ^^

They are more than happy, elated even

@Robyt3 I'm not a terribly huge fan of your std::function use in #8395 they really don't optimize out very well usually and this is in the hot path, have you checked that this gets inlined properly? Perhaps a template ConsumeEvents would be more appropriate?

https://github.com/ddnet/ddnet/pull/8395

Refactor client input event handling by Robyt3 · Pull Request #8395...

Add IInput::ConsumeEvents function accepting a consumer std::function to replace the duplicate usage of the IInput::NumEvents, IInput::GetEvent and IInput::IsEventValid functions. Use an std::vecto...

Jupstar ✪

oh yeah our amazingly huge forum community lmao

Done

1

[ddnet/ddnet] Pull request opened: #8454 Fix console search highlighting. Close #8451

Seems to have been just a small typo, it's very understandable given how we mix meanings of the word "character" throughout that code.

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
[x] Considered possible null pointers and out of bounds array indexing
[x] Changed no physics t...

Is there a possibility to start envelopes animation by touching switch tile? (edited)

no

16:20

maybe u could place a teleporter inside the switch that ports you to a part in the map that looks identically but has some animation. but it wouldnt really start the animation at that exact point

wtf is no login required option

I heard about map to json converter, so maybe it's possible to replace switch tile index by envelopes index xD (no prog sry)

cyberFighter

wtf is no login required option

when the horse needs to ask for permission to go to a server that is kinda a login. and if u dont want this u can filter those (edited)

ejjie

I heard about map to json converter, so maybe it's possible to replace switch tile index by envelopes index xD (no prog sry)

that wont do anything but nice try

Jupstar ✪

that wont do anything but nice try

Can you share that converter? I would try to experiment xd

Q: Do you think a strong root pw is still worth it for a private VPS where SSH does not allow pw login? I am kinda annoyed typing my pw when calling sudo commands lmao

ejjie

Can you share that converter? I would try to experiment xd

it wont work trust me

16:29

but the converter is called twmap

Jupstar ✪

Q: Do you think a strong root pw is still worth it for a private VPS where SSH does not allow pw login? I am kinda annoyed typing my pw when calling sudo commands lmao

I have passwordless sudo on servers

ejjie

Can you share that converter? I would try to experiment xd

https://gitlab.com/Patiga/twmap good luck xd

Patiga / TwMap · GitLab

A Rust library for parsing, editing and saving Teeworlds and DDNet maps

1

ejjie

Can you share that converter? I would try to experiment xd

the converter can only do the things the editor does

16:31

you won't be able to get new features out of ddnet using the converter

16:31

https://rosenzweig.io/blog/vk13-on-the-m1-in-1-month.html

Vulkan 1.3 on the M1 in 1 month

16:33

@Jupstar ✪ huh, did you know https://www.collabora.com/news-and-blog/news-and-events/introducing-nvk.html ?

Introducing NVK

Say hello to a brand new, open-source Vulkan driver for NVIDIA hardware in Mesa.

heinrich5991

@Jupstar ✪ huh, did you know https://www.collabora.com/news-and-blog/news-and-events/introducing-nvk.html ?

yeah nice how all those vk drivers pop up. I hope they can match performance of the closed source drivers at any time.

cyberFighter

wtf is no login required option

KoG's Authed servers were supposed to be filtered by that (even tho it has no effect right now because KoG is missing a flag :D)

[ddnet/ddnet] Pull request closed: #8454 Fix console search highlighting. Close #8451

16:38

github-merge-queue[bot]

[ddnet:master] 2 new commits

e9d568d Fix console search highlighting. Close #8451 - Learath2 1519562 Merge pull request #8454 from Learath2/dd_pr_fixsearchhilite - def-

heinrich5991

you won't be able to get new features out of ddnet using the converter

He probably means to convert the map to json to change all occurrences of a switch tile to smth else or just switch up some numbers (edited)

ejjie

I heard about map to json converter, so maybe it's possible to replace switch tile index by envelopes index xD (no prog sry)

I don't think so, see above

Oh I didn't see that

16:39

mb

ejjie

Is there a possibility to start envelopes animation by touching switch tile? (edited)

If you don't care about surroundings you can just use teleporters to teleport you to the position the quad is at

It would be nice feature anyway if someone can do it

Teero

If you don't care about surroundings you can just use teleporters to teleport you to the position the quad is at

Issue is freeze switch tiles does not give visual experience to player

ejjie

Issue is freeze switch tiles does not give visual experience to player

What?

16:42

Ohh

Bad lang lvl sry

almost forgot to remind of #8434 and #8404

Replying to egyt Well, now I can go on a lil vacation

ejjie

Issue is freeze switch tiles does not give visual experience to player

You mean to allow a passage of a map with a switcher

16:44

You can always use a bunch of doors I guess xdd

Ye but if I want to use deepfrz or tele tiles instead?

16:45

If player activates the switcher deepfrz tiles not hiding (edited)

16:46

There's still visible and just confusing

Learath2

@Robyt3 I'm not a terribly huge fan of your std::function use in #8395 they really don't optimize out very well usually and this is in the hot path, have you checked that this gets inlined properly? Perhaps a template ConsumeEvents would be more appropriate?

I don't know how to check that with the actual binary. According to quick-bench there's only little difference with GCC O1/O2. With clang 12.0, the std::function variant is faster, but with clang 17.0 the one without is faster. https://quick-bench.com/q/6oE7qyBNqcaSV_Vio2VY-Z4yrPE

That sounds absurd

yeah, seems like measurement error with this benchmark tool

Or perhaps compilers just evolved beyond my understanding, that's also possible

is it contructing std::functions on the fly?

17:58

or is it just calling them?

If the std functions are getting properly optimized out, I guess it can just be measurement error and all equal

Jupstar ✪

is it contructing std::functions on the fly?

i assume constructing (edited)

18:00

then it might do a heap allocation

18:00

calling it can be optimized to a certain degree.. with loop unrolling

https://godbolt.org/z/5z5YW9WxT mh the stdfun does generate an extra layer of indirection, but in the specific case we have in the code I think we capture, capturing lambdas I think always entail a heap allocation, (though idk if the compiler can see through that too)

Compiler Explorer - C++ (x86-64 gcc 14.1)

class CEvent { public: int m_Flags; int m_Key; uint32_t m_InputCount; char m_aText[32 * 4 + 1]; }; uint32_t m_InputCounter; std::vector<CEvent> m_vInputEvents; void ConsumeEventsStdFunction(std::function<void(const CEvent &Event)> Consumer) { for(const CEvent &Event : m_vInputEvents) { if(Event.m_InputCount == m_InputCounte...

18:10

eh, idk, I'd be more comfortable with a template there, but it's not very important I guess, compilers do seem to be smarter with std::function nowadays

[ddnet/ddnet] Pull request opened: #8455 Fix wrong file extension being removed in demo cut popup

The .demo extension is supposed to be removed from the target filename when slicing demos, but anything after the last dot was being removed instead, e.g. test.abc.def was incorrectly replaced with test.abc.

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 possi...

19:30

[ddnet/ddnet] Pull request opened: #8456 Unify editor UI mouse handling with CUi

Remove separate handling of UI mouse position and delta in the editor and use the UI directly for this like in the gameclient. Raw cursor movements are redirected to the UI with the CUi::OnCursorMove function. The editor separately updates its world positions and delta after the mouse position was changed. The mouse world position for the editor is passed to the UI with the CUi::Update funtion as before, whereas the world position is unused in the gameclient. Use vec2s for mouse posi...

19:31

[ddnet/ddnet] Issue closed: #8451 Bug: Weird Control-Find Behavior in Console with Cyrillic Characters

[ddnet/ddnet] Pull request opened: #8457 Fix Ok-button of warning popup not being localized

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's memcheck](https://github.com/ddnet/ddnet/#using-addres...

Learath2

https://godbolt.org/z/5z5YW9WxT mh the stdfun does generate an extra layer of indirection, but in the specific case we have in the code I think we capture, capturing lambdas I think always entail a heap allocation, (though idk if the compiler can see through that too)

only a heap allocation wrt. std::function, not in the template

Those people will always exist xD https://github.com/Dealerik/F-DDrace/commit/9003e31a72251cf0a93348587bafbc4377e7fff6

reposittt · Dealerik/F-DDrace@9003e31

[ddnet/ddnet] Pull request closed: #8455 Fix wrong file extension being removed in demo cut popup

20:24

github-merge-queue[bot]

[ddnet:master] 2 new commits

ca89162 Fix wrong file extension being removed in demo cut popup - Robyt3 abc7d05 Merge pull request #8455 from Robyt3/Client-Demo-Cut-Filename-Fix - archimede67

20:24

[ddnet/ddnet] Pull request closed: #8457 Fix Ok-button of warning popup not being localized

20:25

github-merge-queue[bot]

[ddnet:master] 2 new commits

f8e8421 Fix Ok-button of warning popup not being localized - Robyt3 80228ca Merge pull request #8457 from Robyt3/Client-Warning-Button-Localize - archimede67

fokkonaut

Those people will always exist xD https://github.com/Dealerik/F-DDrace/commit/9003e31a72251cf0a93348587bafbc4377e7fff6

maybe you could relicense F-DDRace to a stricter licnese, the strictest would probably be AGPL

heinrich5991

only a heap allocation wrt. std::function, not in the template

Yes, for the template no mandatory heap allocation

heinrich5991

maybe you could relicense F-DDRace to a stricter licnese, the strictest would probably be AGPL

What would that mean ?

heinrich5991

maybe you could relicense F-DDRace to a stricter licnese, the strictest would probably be AGPL

would it really help tho? i guess this is kinda always legit to do, since his source is public and the lincense wasnt changed

AGPL means that everyone who hosts the mod must publish their source code

Jupstar ✪

would it really help tho? i guess this is kinda always legit to do, since his source is public and the lincense wasnt changed

there are licenses that require attribution

heinrich5991

there are licenses that require attribution

"F-DDrace is a modification of Teeworlds, developed by fokkonaut."

20:31

on the github

heinrich5991

there are licenses that require attribution

Usually not very specific about where the attribution is though

Mh

I thought there were some where attribution is required in the end product

That is hard to put into legalese too. Is just a string in the binary enough?

no. say it must be attribution to the people interacting with the software

apparently you can add extra terms

Does it have to be displayed or only if the user wants to see it? You could have it behind a weird command

APGL has a extra section for such things

20:34

there u can require that the string is never removed or changed

Learath2

Does it have to be displayed or only if the user wants to see it? You could have it behind a weird command

in the same way or more prominent than in the original

3.07 KB

https://youtu.be/IS5ycm7VfXg this is pretty cool

"Z2" - Upgraded Homemade Silicon Chips

influencers in 10 years: homemade rtx 5090

[ddnet/ddnet] Issue closed: #8438 Some demo errors are untranslatable

20:59

[ddnet/ddnet] Issue closed: #8438 Some demo errors are untranslatable

Image attachment

Replying to @heinrich5991 "egyt is the next contributor who contributes as much as Robyt3"

21:06

well

You just need to do the same size commits 7 times more

you want me to make 7 more translations?!

More Like 7000 more Lines of Text/Code (every month),

Exported 502 message(s)