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 2023-04-04 00:00:00Z and 2023-04-05 00:00:00Z

it's the same

As far as I'm aware there's none

00:00

First one seems uglier to me

there is no difference

00:00

it's just preference

00:02

stroustrup thinks the former makes more sense in cpp because TEXT is a pointer to a char (edited)

char *text is more used, but thats the same (edited)

[ddnet/ddnet] Pull request opened: #6494 Bring every Font Icon unicode into a namespace

For readability and cleanliness inside the code.

Checklist

[x] Tested the change ingame
[ ] Provided screenshots if it is a visual change
[x] 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
[x] Changed no physics that affect existing maps
[ ] Tested the change with [ASan+UBSan or valgrind's memche...

potentially silly question but pred is in ms right

how to warn at runtime

Ewan

maybe lacks context but lmk what you think

kinda almost there

Image attachment

can you use CScrollRegion in a horizontal orientation?? i would hope so but i'm not finding a flag or member to do so & a red flag i see is that there's only m_ScrollbarWidth in CScrollRegionParams, no height (edited)

You probably cant since I don't remember any horizontal scroll regions in the code and we usually only implement whats used

Dang

06:16

That puts a wrench in things

You'll have to take a detour and implement ScrollRegionH

i went through & added conditionals to all x/y/w/h accesses & based it off a new flag from CScrollRegionParams

06:43

should suffice

06:43

if not i'll just rethink the UI

06:43

2 components doin the same thing is dumb

I would probably have gone for some compile time variation just to avoid the branching

caching can minimize the branching to init time

06:46

but that'll only happen after a refactor

06:47

which will only happen if this works

got it working. i had some stupid and entirely foreseeable OOO issues

07:49

i had 2 instances along the lines of this m_ContentH <= m_ClipRect.h that i changed to this m_ContentSize <= m_Horizontal ? m_ClipRect.w : m_ClipRect.h that evaluated as this (m_ContentSize <= m_Horizontal) ? m_ClipRect.w : m_ClipRect.h when it should have been like this m_ContentSize <= (m_Horizontal ? m_ClipRect.w : m_ClipRect.h)

07:51

but i was blazing through on the first pass so i don't really blame myself

07:51

this particular combination of operators is not something that gets me often

07:54

but yeah when i refactor i'll probably just do like this

float *m_ClipRectSize = m_Horizontal ? &m_ClipRect.w : &m_ClipRect.h,
*m_ClipRectPos = m_Horizontal ? &m_ClipRect.x : &m_ClipRect.y;

...
// whatever
m_ContentSize <= *m_ClipRectSize
// instead of
// m_ContentH <= m_ClipRect.h

07:55

so it's just as fast being a pointer deref instead of member access

07:55

or if it isn't, it's not as expensive as checking each time

doesn't make too much sense to optimize this kind of code, the compiler will rewrite it for you anyway if it isn't too complex

there are dozens of these little m_Horizontal ? checks

07:57

and they are not necessarily slow but it's better to optimize anyway if it ends up making the resulting code cleaner. which it will if done this way

07:57

sooner or later it'll all be yanked anyway and replaced with a 2 axis scroll area, so i won't lose sleep over it (edited)

code clarity is always a good argument though

yea i mean doing it like that from the start would have saved me 20 mins

07:58

the rare objectively better approach

Ewan

but yeah when i refactor i'll probably just do like this

float *m_ClipRectSize = m_Horizontal ? &m_ClipRect.w : &m_ClipRect.h,
*m_ClipRectPos = m_Horizontal ? &m_ClipRect.x : &m_ClipRect.y;

...
// whatever
m_ContentSize <= *m_ClipRectSize
// instead of
// m_ContentH <= m_ClipRect.h

still a bit weird to store a pointer to float, I would have just copied the float float is 4 bytes and a pointer is 4 or 8 bytes, so your pointer indirection is actually an additional cost

it's across multiple functions whose timelines aren't entirely clear

08:00

so they are prone to changing under my feet

08:00

pointer is best

I highly doubt the compiler is smart enough to know this is ui code and getting rid of all these branches by splitting it in two is the best. But it's fine, we can worry about performance later

1

well yeah and this isn't ending up in ddnet upstream

08:01

and if it is, it's in a very long time

08:01

my fork of a fork of a fork of ddnet

08:01

xd

If it's not for upstream then you have even less reason to care

fork maintainer would probably like it if i took it seriously

08:02

so i do

08:02

and i'd be happy to patch for ddnet but when i asked in here nobody was interested (edited)

08:02

so whatever

Ui performance doesn't matter anyway

08:37

Not like u have to react quickly. Responsiveness should be given tho ofc

what do u answer if ur offered senior level rust positions

09:07

but u dont think ur senior

09:07

im just 22 soon 23 lol

Just say yes

09:17

What do u have to loose

if someone knows, can i make rifle-sg shots and doors to full transparent?

Image attachment

U can remove the textures

Jupstar ✪

U can remove the textures

where?

09:22

i couldn't find :(

SG is in game.png i think. Laser is particles.png

Jupstar ✪

SG is in game.png i think. Laser is particles.png

all u called i've already made empty

09:26

pepePHONE

Mh then laser maybe not fully transparent xd

Jupstar ✪

Mh then laser maybe not fully transparent xd

Ncry

Blame whoever thought that color values do not contain an alpha channel is a good idea xd

Jupstar ✪

Blame whoever thought that color values do not contain an alpha channel is a good idea xd

yeah, weird

09:29

Sticker

09:32

LOL

09:33

deleted assets in roaming to check if smth gonna change, open the game it automatically create a folder and my previous Deleted xDDD (edited)

Yes but empty

09:34

It's just so the buttons in-game work

yeah but i mean it is just deleted all stuff i had before even from recycle bin

09:35

xDDD

Ryozuki

but u dont think ur senior

i got offered a senior cpp position last week and turned it down for the same reason

09:37

i just said something like "As much as I would love to humor it, I'm not there yet"

Jupstar ✪

What do u have to loose

if you're looking for work then you have time and effort to lose

Mh, depends on what u think a senior dev has to do i guess

yea

igot a job already tho

10:09

so its less that im looking more that they come to me

10:09

PES2_BasedGe

Ah thought same job xd

Ryozuki

but u dont think ur senior

Boss thinks ur senior, ur senior. That's how capitalism works

its not my boss tho its another company

12:34

im definitly not considered a junior

12:34

i seen some juniors in the company

12:34

some didnt even know what a trait is

HUH

12:48

And here I think I'm not qualified to be an intern even

@Learath2 there are juniors and then "juniors"

13:19

and then me

13:19

someone that knows more than some seniors

13:20

but well seniors are seniors cuz they are better at knowing how much x will take or how hard or idk

13:20

im probably a med or idk

13:20

a normal dev

I'd say a senior is someone who doesn't need to ask every 5 minutes

well idk if thats a metric

13:21

i barely ask

13:21

if i ask is because i dont know smth thats specific of the software

Then u a senior.. at least if u make progress with whatever u do

13:22

U can be a rust senior without being a senior in the field u are working on

i guess thats it

13:23

well i would say i lack the skill to manage people like others, altho i have no problem saying do x and i do y to someone

13:23

actually i do that

13:23

its more a exp based thing

Well that's not directly the job of a senior

true

That's a manager

project manager

13:23

but it feels weird to call urself a senior

Why

im a rust developer thats it

13:24

dont u always feel like ur less

13:24

or not deserving

13:24

i guess i have too high standards

13:24

a senior for me is someone like deen

13:24

learath would be my C senior

13:24

if i have a c question i ask him

13:25

kek

I feel like the best I can do is stuff i like, because I'm motivated for them. Skill is not an excuse then. So I feel less if I don't make progress in this field

yeah thats obs

13:25

obv

13:25

if u do smth u like u do it the best

13:25

im sure my php code would be bad

13:26

cuz i wouldnt be motivated for shit

Learath2

And here I think I'm not qualified to be an intern even

u r qualified dont joke urself

13:27

the job market is not prepared for nerds like us

Anyway. To me a senior is not equal to an expert. Learath probably knows more about c then most c senior

they require us to go through the stupid ladder

Jupstar ✪

Anyway. To me a senior is not equal to an expert. Learath probably knows more about c then most c senior

i guess thats a better view

13:27

for me senior always meant expert

13:27

but its prob not true xd

13:32

@Jupstar ✪ a good way to boost ur ego is to go to linkedin and see boomers posting screenshots of js code snippets saying "look this neat js feature totally you didnt know!"

13:32

BASEDDEPT

@Ryozuki a senior is someone who can write code that chatgpt can't write

Ryozuki

@Jupstar ✪ a good way to boost ur ego is to go to linkedin and see boomers posting screenshots of js code snippets saying "look this neat js feature totally you didnt know!"

Most ppl haven't created something revolutionary anyway. In fact as sad as it is. Most coders write lot of code and not so much high quality code. That's not bad, because that means progress for society. But the really hard stuff is probably often connected to math or physics

13:39

@Ryozuki would u say u have to think a lot and hard about what u code?

13:40

Or is it mixed? When I look at what I wrote during the years. The hardest stuff always was either related to a good multi threaded design, or math related

Jupstar ✪

@Ryozuki would u say u have to think a lot and hard about what u code?

lately yes, because im doing something i didnt before

13:40

writing llvm

13:40

its rly fun tho

13:41

well it makes sense

13:41

threading is one of the hardest if not hardest things in computing

13:41

and saying otherwise is a lie

13:41

hardware is racy

Yep especially if it's a design that doesn't target a single algorithm

Ryozuki

lately yes, because im doing something i didnt before

The question will be if u will be much faster once you learned out. Or if what u learn in itself requires you to rethink everything.. so basically never ending learning.. that's usually where u have to be innovative and that costs time to think and trial and error ^^

i think learning is always good

Ryozuki

and saying otherwise is a lie

Otherwise

and if the learning leads u to rethink

13:45

its good

Learath2

Otherwise

Liar

kek

13:46

i mean threading involves all the complexity of hardware, but multiplied by X

It's actually extremely annoying to reason about any order of execution in modern architectures. So I do kinda agree that it's very hard

because u do X things in paralel

13:46

@Learath2 whats another thing u thought?

13:46

thats as hard or more

13:47

tldr: why otherwise

Use no globals anywhere xd

use rust xd

Ryozuki

tldr: why otherwise

I just said it as a meme

ah lmao

13:47

xd

Ryozuki 10 iq

If I had to find anything as annoying it'd probably be proper benchmarking

oh true

It's actually deceptively hard to set up good benchmarks

is counting cycles a valid deterministic way of benchmarking?

13:48

i always wanted to have a realiable benchmark that can run in CI

What is my algorithm has many cycles but good effective speed

13:48

If

yeah

13:48

idk

13:49

my question is

13:49

is there a solution to have good benchmarks on machines with lot of noise

Eg Because of multi threading xd

Even at a higher level, you need to know what data sizes to bench with, what collection sizes to bench with, you need to know if there are any platform quirks ruining your performance

Learath2

Even at a higher level, you need to know what data sizes to bench with, what collection sizes to bench with, you need to know if there are any platform quirks ruining your performance

the solution tbh is live production telemetry

Ryozuki

is there a solution to have good benchmarks on machines with lot of noise

Just to it enoughmanytimes that the noise averages out I guess

13:50

do it*

Learath2

Even at a higher level, you need to know what data sizes to bench with, what collection sizes to bench with, you need to know if there are any platform quirks ruining your performance

So benchmarking for scalability.. yes that's very hard

for example

13:50

in my app i had a query that took 1 s

13:50

i knew cuz telemetry

13:50

so i fixed it

13:50

down to half

13:50

idk xd

Ryozuki

the solution tbh is live production telemetry

Yes. Almost always better than synthetic benchmarks

Image attachment

But a query is something specific. What if u run 10000 queries at once. What's the bottleneck then? ^^

When ddnet telemetry?

troll

rust has a rly good crate

13:51

actually i could do that

13:51

im motivated

Ryozuki

actually i could do that

Heinrich would literally shoot you for privacy reasons

the server just needs to offer a single http endpoint with prometheus formatted data

13:52

the telemtry isnt smth personal

13:52

more like

Maybe opt in telemetry?

Rust also static links. I could imagine that makes benchmarking easier if u want to inject it at compile time

average duration of X thing

13:53

m$$ rly stained the name of telemtry

13:53

@Learath2 oh i meant server side only

Oh that's boring

The problem with telemetry is the name itself

Ryozuki

Click to see attachment 🖼️

u think this telemtry is privacy related? xd

If I hear it I think they steal private. Information

We should make one of those profilers new engines have so we can track how much time we spend on what

The client version is also a kind of telemetry we collect

Learath2

We should make one of those profilers new engines have so we can track how much time we spend on what

thats the telemtry i thought of

13:54

but yeah

13:54

it would be cool to have gpu telemtry

13:54

detailed

Xd

legit for optimizing better

13:55

e.g draw call info

13:55

idk im talking bs

Imagine we'd know what config are used how often and what hardware, what drivers

13:56

And break backward compability lmao

13:57

@Learath2 btw what do you use for benchmarking if that's a topic u investigated your thoughts in already?

I only tried a couple libraries but havent used any in anything serious

13:59

I think Celero was the one I enjoyed the most

14:01

I was more talking about methodology than implementations though

Yeah but they depend heavily on your software requirement anyway

Googles attempt at is probably the best but it is as always rather heavy

What about human intuition?

I mean none of those help with the methodology, you still need to design the testcases. For which human intuition is indeed still the best we have

with rust i use criterion

If you can learn intuition, then maybe an ai can beat benchmarking software some day xd

we need property testing in ddnet

But it does occasionally fail you, so you need to make sure you explore all variables

its cool

14:07

https://altsysrq.github.io/proptest-book/intro.html

Introduction - Proptest

Usage documentation for the proptest and proptest-derive crates

14:07

Proptest is a property testing framework (i.e., the QuickCheck family) inspired by the Hypothesis framework for Python. It allows to test that certain properties of your code hold for arbitrary inputs, and if a failure is found, automatically finds the minimal test case to reproduce the problem. Unlike QuickCheck, generation and shrinking is defined on a per-value basis instead of per-type, which makes it more flexible and simplifies composition.

Modern cpus have some extremely unintuitive behaviours

Javascript too xdd

xddd

14:08

man

I think Ryozuki posted a very interesting one here once. Something with an array with a weird amount of elements performing better than a power of 8

wasm needs to take off

Learath2

I think Ryozuki posted a very interesting one here once. Something with an array with a weird amount of elements performing better than a power of 8

i know

14:10

@Learath2 https://en.algorithmica.org/hpc/cpu-cache/associativity/

14:11

Image attachment

14:11

There is no vectorization or anything, and the two loops produce the same assembly except for the step size. This effect is due only to the memory system, in particular to a feature called cache associativity, which is a peculiar artifact of how CPU caches are implemented in hardware.

14:11

why do i have such good memory

14:12

amazed of myself

14:13

that website itself is golden

14:13

rly good resource

lol I remember that

14:14

u sent it here once

ofc

14:14

im the messager /(idk the word kek) (edited)

Yep, so you really need to dig into your brain for anything you can think of. Almost everyone would use the nice even 256 there and might not even notice it is 10 times worse than 257

Xd

Learath2

Yep, so you really need to dig into your brain for anything you can think of. Almost everyone would use the nice even 256 there and might not even notice it is 10 times worse than 257

well this is why most SO answers say "profile before doing anything"

14:15

because its hard to determine

It's really easy to miss that even in a benchmark you might think is rather well designed. Say I tested incrments of 32, 64, 128, 256. I'd totally have missed it

That's where I'd say intuition is better.. just because this single loop is slower that doesn't mean your software must suck, if it overall has a good concept

14:17

If u want to optimize for a single hardware maybe

14:17

But usually only Learath do that in c

I mean, depends on the goal of your software. If you need the absolute minimum latency e.g. this might not be an acceptable loss

Yeah

Jupstar ✪

But usually only Learath do that in c

I optimize for an imaginary abstract machine in my head that isn't too hard to optimize for

inb4: nasa using 257 stride instead of 256 in the moon software

I mean I somehow actually like optimizations like these. But often u probably target a general purpose hardware

Ryozuki

inb4: nasa using 257 stride instead of 256 in the moon software

They use javascript xd

xd

BCS fast enough xdd

I target an old 16 bit cpu in my brain where ordered execution is still a thing

Isn't x86 strongly ordered?

ye

14:20

i think risc is weakly

But only the memory model. Dunno what Lea means

lea is an instruction

14:20

iirc

Eeeh. I don't honestly even know anymore. It's guaranteed that it'll appear as such. But cpu pipelines do reorder your instructions as they see fit

load effective address

14:21

i just figured learath name meaning

14:21

he was a assembly instruction all along

14:21

and one of the most used

Only the first 3 letters

It's a common girl name

PES_Blush

That implies he wants girls in his life desperately

Jupstar ✪

It's a common girl name

Was I a catfish all along?

ryozuki sounds girl too

14:22

but i cant change it

14:22

i just choose a japanese sounding name long ago

14:22

and stuck

14:22

it was cuz ryuzaki

14:22

from death note

Yours sounds more weeb than girl so that'll throw people off

kek

Ryozuki

from death note

Oh I always wondered if it's that xd

hmm Misutero Gosuto (edited)

14:50

or Gosuto-san idk I like the Mister though xD

Is that also anime? Xd

no xD

Mr.Gh0s7

or Gosuto-san idk I like the Mister though xD

It'd be misutā (ミスター) not misutero

oh thanks I was wondering xD

ミスターゴースト

today i had to download a goverment app to get a pin to do my taxes soon

15:13

cyberpunk 2077

15:13

probs can do without app but its way more complicated

15:13

https://clave.gob.es/clave_Home/en/clave.html

15:13

its this

I find newer interfaces to these legacy processes are indeed much easier

the gov doesnt even put a HD logo of spain there

15:14

lmao

Hd

15:14

Svg

Italy does the same. I keep thinking it's some scam

Sad

now my chinese phone has a spanish gov app

15:15

what else

Install anti virus

it has one

15:16

from the china phone

15:16

kek

15:16

it shows a ad after u install a app

Lmao

i wanted to install a custom rom but im 2 lazy

15:16

i dont use the phone much besides reading memes

15:17

MIUI

I want to install lineageos on my old phone but it's not maintained so I need to build my own

15:17

I honestly can't stand miui

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

MIUI (pronunciation: Me You I) is a mobile UI (User Interface) developed and maintained by Xiaomi exclusively for its smartphones. MIUI is based on Android Open Source Project and serves as the basis for other operating systems created by Xiaomi: MIUI for POCO, MIUI for Pad, MIUI for Watch, and MIUI for TV (PatchWall). There are different versi...

15:17

it has this

15:18

one day ill get a more open phone

I use HavocOs

I have that

but i dont want to spend $$$ on a phone

Huawei gang

i dont use much

or should I say, wawei

Ryozuki

one day ill get a more open phone

This is one thing there will never be good open alternatives in

is pixel good?

Oh if that is open enough for you yeah, it's decent

15:19

I thought you meant something like a linux phone

Learath2

I thought you meant something like a linux phone

i know its hard that

15:21

but yeah for me google is more open than china

15:21

for now

15:21

google atleast publishes the source code for the ROM iirc

15:21

huawei violates GPL

15:21

they dont publish it

Yeah. It's good enough for me too. You usually like extremes of freedom, so I just assumed

https://github.com/MiCode/Xiaomi_Kernel_OpenSource/issues/2506

15:22

https://github.com/MiCode/Xiaomi_Kernel_OpenSource/issues

Issues · MiCode/Xiaomi_Kernel_OpenSource

Xiaomi Mobile Phone Kernel OpenSource. Contribute to MiCode/Xiaomi_Kernel_OpenSource development by creating an account on GitHub.

15:22

check all the issues

Learath2

Yeah. It's good enough for me too. You usually like extremes of freedom, so I just assumed

is pixel 7 good

15:22

587€ is already enough

15:23

i dont like paying 900 for a phone

I haven't used a pixel in years. Idk if this generation is good or not, you'll have to research

my requirements arent high anyway xd

Yeah, I think this will be my last flagship phone. I'll downgrade next time I need to get a new phone

15:25

The prices are just insane now :/

Ryozuki

587€ is already enough

True. Phones sucks BCS no good os to write rust

default

or should I say, wawei

we don't say it "wawei"

but that's how you spell it

no

the "hoo" is silent

no, you have to spell it, they even do it

https://www.google.com/search?q=huawei+how+to+pronounce

15:38

hwaaway

yes, not waway, the h is pronounced, you can hear it

yea but I meant the "hoo"

15:40

can't you read

well u can kinda hear it too

germans pronounce it "hooawei" (edited)

15:41

while it gets pronounced more like wawei

15:42

and if you don't believe me, listen to German google translate (edited)

Image attachment

15:42

it is accurate

I already know

[ddnet-web:master] 1 new commit

a44ce33 Update Turkey - def-

[ddnet:master] 3 new commits

df48c44 Bring every Font Icon unicode into a namespace - VoxelDoesCode 90861d9 Get rid of unused warning - def- 89eb7e6 Merge #6494 - bors[bot]

17:25

[ddnet/ddnet] Pull request closed: #6494 Bring every Font Icon unicode into a namespace

^^

[ddnet/ddnet] Pull request opened: #6495 Fix swapping with dragger beams and plasmas

The client ID referenced by dragger beams and plasmas must also be swapped when two clients are swapped, as those entities should refer to the same character entity as before the swap. For dragger beams, swapping previously caused the active beams to switch target to another character. For plasmas, swapping previously caused already shot plasmas to not have an effect on their original target character. Closes #5865.

Checklist

[X] Tested the change ingame
[ ] Provided sc...

this has probably been talked about before but wouldn't it be cool if cl_nameplates_strong 1 would work so if you spectate someone you can see who they have strong/weak on. It's like that with the weapons hud etc aswell jaothinking

jaothinking

agreed

20:33

would also be super easy

20:33

and free view spec still showed who you have strong/weak on

Image attachment

20:51

ddnet 16.9 won't open

20:51

16.8 is ok

damn

lol

[ddnet:master] 2 new commits

aa08cfa Fix swapping with dragger beams and plasmas - Robyt3 e47f54d Merge #6495 - bors[bot]

21:23

[ddnet/ddnet] Issue closed: #5865 Swap change behavior of laser hooks and shooters

21:23

[ddnet/ddnet] Pull request closed: #6495 Fix swapping with dragger beams and plasmas

noby

Click to see attachment 🖼️

use windows u noob

ReiTW

use windows u noob

#developer rule #1: don't mention windows as usable operating system

23:17

Exported 400 message(s)