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 2021-02-04 00:00:00Z and 2021-02-05 00:00:00Z

[ddnet-maps:master] 1 new commit

0dc8441 A run_4tea - ddnet-maps

wow when did the spec speed changed? I did not notice anything o.O maybe because i prefer to play over spectating :D

1;3A

10:18

oop

@heinrich5991 can you explain your motivation for f8277267a0a8a4d2ac296cdf4b22cab00047c5e3 and why it is not in vanilla?

now that the percentage thing is shown at the rank. wouldnt it make sense to reform the rankpoints system? so you get rankpoints for every finish and it doesnt need to be in the top10

That would mean we have to clean out all duplicate ranks, instead of just in the top 10, way too much work

11:49

(by duplicate I mean same person finishing with multiple names)

ah right...

11:51

but still its hard to get in the top10 on many maps and its not a good motivation to do a speedrun

Could measure it as % slower than top time, but I'm not that motivated to change it

i did the stupid thing of shutting down while updating and it was updating kernel related stuff

justatest

1

13:16

lucky i had a live usb near

13:16

on windows u would be fucked, lucky on linux u have full control

Ryozuki too stupid for windows

not stupid

13:21

its simply that on windows u wouldnt be able to fix it

ChillerDragon: less unused code. that should be enough

13:23

ChillerDragon: I didn't want to engage in discussions whether an obviously unused feature should be removed or not, tw usually seems skeptical of changes like that

ok thanks

14:03

Ryozuki: i wouldnt be able to fix it on linux either and would have to buy a new linux xd

Cuz u nobo

ChillerDragon do u have a rpi with raspbian?

14:32

Heads up: In a significant screw-up, the Raspberry PI foundation pushed an update installing Microsoft GPG keys and repo secretly all Raspberry PIs running Raspbian Linux. They are currently silencing everyone on their forum who is pointing out this issue. https://t.co/wF6eUCoaOn

Microsoft repo secretly installed on all Raspberry Pi's Linux OS - ...

Microsoft repo secretly installed on all Raspberry Pi's Linux OS by Raspberry foundation's official OS, and many Linux/FOSS users are upset.

mac book pro 2020 44fps ingame

14:42

is it normal?

nope, should be much higher

14:48

maybe it renders at 4x resolution and then gets scaled down?

wtf Ryozuki yes i use it as router

14:50

that sounds fucked up

how 2 fix that btw ?

Install arch

n1 advice

it actually is

14:52

but its some work yikes

14:52

Embedded image

14:52

that sounds easier

14:52

but for the furture i might try arch

Skyrx

how 2 fix that btw ?

what are u graphics settings

cl_max_fps 44

14:53

xd

14:54

thanks a lot for the info Ryozuki

Image attachment

my try without opengl3

14:55

but thats just a random guess

already tried

14:55

nothin changed

how does the vanilla teeworlds client perform?

lemme test

hmm sudo apt-mark hold raspberrypi-sys-mods does not look good to me

14:58

also i want my router to not break so i have internet :D and debian based stuff is nice for not breaking

14:58

maybe i should run vanilla debian on my pi then

I run archlinuxarm on my router pi

14:59

no breaks so far

arent u even running arch on ur vps

14:59

and u didnt brick ur system in 8 yrs or somthing

14:59

i cant rly compare myself to u

300 fps

15:03

default client

15:05

300-400 fps default client

15:05

why wtf

yea i think vanilla does some things different and i also usually have higher fps there

15:05

but ddnet should still be higher im sure there is a fix for u but dont ask me how

try high dpi off i heard some ppl had issues with that

already tried

15:06

still nothin

did you try turning ur device off and on again?

15:07

joke ^^dont do it xd

15:08

maybe try linux

15:08

best gaming os

"facepalm"

its actually worth a try

15:09

in the meantime u can use F-Client which is a vanilla based client with ddr features

no antpiping i guess?

well vanilla has some antipingish thingys

15:10

https://fokkonaut.github.io/F-Client/

15:10

its not that advanced as ddnet tho

15:13

yikes that fucks with my OCD now it alsays says 2 packages can be upgraded. Run 'apt list --upgradable' to see them. and then does not installed them cuz i blocked ms :D

15:14

i guess that is the downside of linux getting more used is that companys like ms want to enter the market

companies*

U can use any other distro

what is this black magic? https://github.com/ddnet/ddnet/blob/cbf5aa6e36be1f1a28f7da64df6a5ef8b2ad1f21/src/game/client/components/menus_browser.cpp#L580

DDraceNetwork, a cooperative racing mod of Teeworlds - ddnet/ddnet

17:38

changing ui_server_address has no affect on the ui?

17:38

is the ui somehow constructed after connect and then stored?

17:40

i leaked my fokin password because i clicked a server in the ingame browser :/ and my pw manager thought i changed server

Mobo

17:50

Nobo

@QshaR how to reset pw on kog?

17:52

also why is /login not in /cmdlist

@qshar ^

https://qshar.com/reset_password.php

ty

18:03

omg google

18:03

how does the average user find this?

18:03

i could not find it on qshar.com and as i reported cmdlist was even worse

18:04

Embedded image

18:04

also fuck google

18:05

but thanks for the fast response :) when kog irc or open forum again

we have all info on discord now

open irc chillerbridge

oof i forgot the email of my dummy

18:19

oh lol one mail can setup both names? ah nice

omg failed to login is it possible that you have bad special character support? @qshar

18:27

or did i mess something up

18:29

if some passwords do not work maybe add a filter on the password reset form if the password wont work for login

how to get id of player from the

18:41

m_pCient

18:43

oh wait nvm its ClientID

18:45

wait no its not

Eh? Are you on the client or on the server? Which class are you coding in?

uhh different question

19:01

is there a better way to do this m_pClient->m_aClients[m_pClient->m_Snap.m_LocalClientID].m_Predicted.m_ActiveWeapon != WEAPON_GUN

19:01

in players.cpp

19:01

cleint

m_Snap.m_pLocalCharacter ?

got a 100% crash with 15.2.5 when joining right now ger1:8305 with Back In Time 2

19:28

on win32

19:32

voting the map on empty server also crash

win32 oof

19:34

can u reproduce consitantly?

100%

thats a good start i guess :D

nah its win64 but the client is 32 bits

why do u run 32 bit?

(it's still a bug)

ah ye sure

19:35

just wondering

64 bit client does not trigger graphic card switch on laptop

im trying to implement some derive macro and my brain is melting

19:36

@heinrich5991 have u ever tried this magic?

no

im trying to implement a trait Form which outputs a struct as a html form

Image attachment

19:36

monkalaugh

19:37

Image attachment

19:37

im there

19:37

but its hard

19:37

(this is the expanded macro)

19:38

im sure i can even avoid the for loop

19:38

#(#field),*

19:38

or some weird magic

can't help you rn, in a lesson

np

19:38

just wanted to show u

in 30min

ehm, gdb on windows doesnt show anything...

19:38

just say it exited with code 3...

what does that mean? it doesn't crash when running in gdb?

19:39

ah

it crashes, but doesnt give a backtrace or anything

maybe it doesn't crash but actually just exits for some reason

19:39

idk how to debug that in a good way though

nah, outside gdb i got window saying app has crashed

ah

19:39

hm

is there a win version with debug symbols?

19:40

This application has requested the Runtime to terminate it in an unusual way. Please contact the application's support team for more information. [Inferior 1 (process 4312) exited with code 03]

try building from source with visual studio and run it from there

Code de l’exception: 40000015 According to internet this means: STATUS_FATAL_APP_EXIT

[quakenet] ChillerDragon

try building from source with visual studio and run it from there

no way i install this big factory, my disk is almost full

xd

i use mingw

fair

19:43

im not sure how to debug with mingw

long time i haven't build ddnet on my computer

i think we dont compile in mingw

19:44

so it may be bugged

19:44

or maybe deen knows

releases are compiled with mingw I think

trying to compile tag 15.2.5 right now

make sure to pull submodules

rofl cmake error

make sure to pull submodules

git submodule update?

git submodule update --init --recursive

19:46

fixes all your submodules woes

master and 15.2.5 changed ddnet-libs iirc

19:46

due to discord

E:\Teeworlds_DDNet\ddnet\src\base\system.c:3392:10: error: 'uintptr_t' undeclared (first use in this function); did you mean 'intptr_t'?
 3392 |  return (uintptr_t)ShellExecuteA(NULL, "open", link, NULL, NULL, SW_SHOWDEFAULT) > 32;
      |          ^~~~~~~~~
      |          intptr_t
E:\Teeworlds_DDNet\ddnet\src\base\system.c:3392:10: note: each undeclared identifier is reported only once for each function it appears in
E:\Teeworlds_DDNet\ddnet\src\base\system.c:3392:20: error: expected ';' before 'ShellExecuteA'
 3392 |  return (uintptr_t)ShellExecuteA(NULL, "open", link, NULL, NULL, SW_SHOWDEFAULT) > 32;

(edited)

19:49

well, seems like some typedef are missing in mingw

#include <stdint.h>

only on mingw i guess

19:50

rip iirc

19:50

Image attachment

19:50

Image attachment

19:50

monkalaugh

heinrich5991

#include <stdint.h>

works

omg it took me 10 sec to realize thats a build err not a crashlog xd

/src/base/tl/base.h(14): assert!

19:55

wait, i transfer my config for this one

assert o.O that should be catched in the log

19:56

no need to debug that

19:57

i mean yea debug helps to see why it breaks tho xd

how do i get gdb prompt when it crashes, it doesn't accept any command, seems just like to hang

u should get one instantly

normally it should give you a prompt when it crashes, yes

#0  0x77239962 in msvcrt!memcpy () from C:\Windows\syswow64\msvcrt.dll
#1  0x005a96e8 in mem_copy (dest=0x0, source=0x4a9d0020, size=8)
    at E:\Teeworlds_DDNet\ddnet\src\base\system.c:262
#2  0x004633a1 in mem_copy_special (pDest=0x0, pSource=0x4a9d0020, Size=8,
    Count=14952016, Steps=12)
    at E:\Teeworlds_DDNet\ddnet\src\game\client\components\maplayers.cpp:397
#3  0x004652ac in CMapLayers::OnMapLoad (
    this=0x8d5620 <gs_MapLayersBackGround>)
    at E:\Teeworlds_DDNet\ddnet\src\game\client\components\maplayers.cpp:843
#4  0x004c63ad in CGameClient::OnConnected (this=0x76cfc0 <g_GameClient>)
    at E:\Teeworlds_DDNet\ddnet\src\game\client\gameclient.cpp:511
#5  0x00419589 in CClient::ProcessServerPacket (this=0x3bb0020,
    pPacket=0x28d6f0)
    at E:\Teeworlds_DDNet\ddnet\src\engine\client\client.cpp:1794
#6  0x0041c63d in CClient::PumpNetwork (this=0x3bb0020)
    at E:\Teeworlds_DDNet\ddnet\src\engine\client\client.cpp:2591
#7  0x0041d557 in CClient::Update (this=0x3bb0020)
    at E:\Teeworlds_DDNet\ddnet\src\engine\client\client.cpp:2858
#8  0x0041ea78 in CClient::Run (this=0x3bb0020)
    at E:\Teeworlds_DDNet\ddnet\src\engine\client\client.cpp:3242
#9  0x0042399f in main (argc=1, argv=0x1c71d78)
    at E:\Teeworlds_DDNet\ddnet\src\engine\client\client.cpp:4355

(edited)

20:02

Program received signal SIGSEGV, Segmentation fault. 0x77239962 in msvcrt!memcpy () from C:\Windows\syswow64\msvcrt.dll

ah, good old backtraces triggering github issue linking, now also on discord

Oh the bot got confused :P

soryr, wanted to do shift+enter to add a newline, just pressed enter

epic bot

20:03

@Learath2 i remember u saying to blacklist the first 10 issues

20:03

irc must be overloaded

20:04

+ 1 edit

so can reproduce 100% only on back in time 2 right now

mem_copy_special?

Sounds like a jank solution I'd indeed come up with :P

do we ahve that function? oO

with release from ddnet website and debug build from mingw

syswow64

20:04

much wow

im ded Ryozuki

I dont remember a mem_copy_special

can someone delete the bot message maybe?

ChillerDragon hi ded im dad

fix bot

is the github up to date

20:05

repo

https://github.com/ddnet/ddnet/blob/1c750a73ff329197ed7661b11ba7aa7ebff33f20/src/game/client/components/maplayers.cpp#L392-L400

DDraceNetwork, a cooperative racing mod of Teeworlds - ddnet/ddnet

yes, did a pull before building

its not a bot issue

20:05

its non format as code issue

ah ye

20:06

he forgot first time

the name mem_copy_special is big meme

my bad

#1

https://github.com/ddnet/ddnet/issues/1

Show time of player you're spectating · Issue #1 · ddnet/ddnet

Probably difficult since spectators don't have a character?

still

oh

20:06

nvm

@jao u used 2 backsticks

20:06

try 3

then actually bot issue

Looks like a very dangerous function :P

#2

#1

https://github.com/ddnet/ddnet/issues/1

Show time of player you're spectating · Issue #1 · ddnet/ddnet

Probably difficult since spectators don't have a character?

gg

an no its discord issue

20:06

xd

20:06

:D

$info

20:07

how was it

20:07

to get the url of the bot repo

20:07

$about

not here i guess

DDNet#9607

Discord bot for DDraceNetwork

Stats

95 Guilds 1702 Channels 7971 Users

Process

151.31 MiB 0.00% CPU 11 Threads

Bot

10d 23h 20m 9s Uptime 138.14ms Latency

Latest commits

5cc7c6f bump pps thresholds (4 days ago) 78f4216 . (6 days ago) 0ba2183 change expire timestamp format (6 days ago)

$help

its about

Anyway, Count looks way too high

20:08

That would be where I start investigating, the source of that count

95 Guilds

20:09

this bot is in lot of places

its going places

at 100 it will get banned

??

20:11

why

personal ID required by discord or so?

does this bot use sharding

Spy on the guilds with it

after that u need verification

ow i guess i can rls mine ^^

20:12

or can you just deactive the invite process?

did you try not using discord?

you have a discord bot @Patiga?

someone invite ddnet to 6 random discords justatest

justatest

nah i got a server where i invited the bot

Therefore, we are making verification a requirement for bots in 100 or more servers. Until your bot is verified, you will not be able to grow past 100 servers. If your bot is already in more than 100 servers, it will not be able to join any more.

20:13

u wont get banned

20:13

ppl will simply be unable to invite it

ah

ye

nice

but how to get it verified

20:13

they will decline it like they declined server verification

send personal data to discord I think

server is not verified?

maybe deen can do it

20:14

if he doesnt care

isnt owning ddnet.tw enought?

server can definitely get verified

we are not talking about the server

server got declined 2 times

lol when 2 times

or 3

that sucks i saw some 500 member server verified

half a year ago and 1 month ago

member count isnt the only requirement

I think they want social media presence and third-party sources

20:16

I could only find one third party src I think

python supports lookaheads right

yes

hmm idk if regex is the best way

lookaheads are bad because they make the language non-regular and thus no guarantees about performance anymore

r'(?:(?P<owner>\w+)/)?(?P<repo>[\w-]*)#(?P<id>[1-9]\d*)\b'

20:20

this is the current regex

[ddnet/ddnet] Issue opened: #3601 Segmentation fault on 15.2.5 when loading map "Back in Time 2"

Got it 100% on 32 bit release build from ddnet.tw as well as debug build from mingw32 on a win 7 64 bit. Output from gdb: ```shell Program received signal SIGSEGV, Segmentation fault. 0x77239962 in msvcrt!memcpy () from C:\Windows\syswow64\msvcrt.dll #0 0x77239962 in msvcrt!memcpy () from C:\Windows\syswow64\msvcrt.dll #1 0x005a96e8 in mem_copy (dest=0x0, source=0x4a9d0020, size=8) at E:\Teeworlds_DDNet\ddnet\src\base\system.c:262 #2 0x004633a1 in mem_copy_special (pDest=0x0,...

32bit

monkalaugh

64 bit client doesn't use graphic card even if i force it into nvidia control panel (edited)

20:24

so i have to use the 32 bits one

20:24

maybe this would help: https://stackoverflow.com/questions/10535950/forcing-nvidia-gpu-programmatically-in-optimus-laptops/14041061#14041061

Forcing NVIDIA GPU programmatically in Optimus laptops

I'm programming a DirectX game, and when I run it on an Optimus laptop the Intel GPU is used, resulting in horrible performance. If I force the NVIDIA GPU using the context menu or by renaming my

Optimus is awful at deciding when to turn on

20:27

Maybe try opengl3.3 that might get it going

nah, nothing got it working

20:27

it starts the GPU then fall backs to intel integrated graphics

20:28

whatever option i set in ddnet, and even if i force it into nvidia control panel

client uses too few resources so it falls back to intel?

20:28

is integrated too bad for ddnet?

but it works for 32 bits version, so i guess it might be a nvidia bug as well

20:29

64 bits client always use intel graphic whatever i do, 32 bits client does whatever i tell him to do (intel or nvidia)

the max allocation i see is 80mb at once

20:29

doesnt sound too horrible

should i also open an issue for graphic card selection (it's a minor defect that i work around)?

can u even do that in opengl?

apparently, you need to export a symbol in the binary

on windows?

yup

cant u rightclick ddnet.exe and select the GPU at start?

it ignores it on the 64bit, not on the 32bit

strange

We should probably debug the client with an allocator that randomly fails

Deleted User

strange

that's why i believe its probably non ddnet related

well malloc is allowed to return NULL

20:32

i just dont understand why it should for 80mb

on 32 bit there might not be enough address space

Maybe it's allocating way too much due to a bug?

contiguous

@Chairn can you add a dbg_msg before the malloc to see what UploadDataSize is?

what file?

but why shouldnt there, what does the compiler do

maplayers.cpp

so the heap is soo small

20:34

already did @Learath2

80 MB is 1/25 of the address space of a 32bit application (edited)

@Deleted User you did, can you even reproduce the issue?

no

20:35

for me the allocation doesnt fail

20:35

pdest is null so it must be the allocation

Yes, and how can an allocation fail?

no heap space

or not enough contiguous address space

as heinrich said, i just cannot imagine for a rather small allocation like this xd

80MB? that's 1/25 of the address space

20:36

that's quite a lot

mhh i dunno

20:36

doesnt sound alot

So I assumed the first one, I thought to myself how that could happen and one theory I came up with is UB leading to a massive UploadDataSize

imagine loading 25 shared libraries at random addresses

20:37

that will give you no remaining space for 80MB on the heap

20:37

idk how ASLR works on 32bit windows

we could use LARGEADDRESSAWARE flag

20:37

then its only 2.6% of the address space

UploadDataSize is 299040320

20:37

then it crashes

that's 280MB

previous size are 1006320 and 184160

oh then i usedw the wrong map appearently

okay, that"s a lot of memory, but should still not crash

20:39

there's enough memory available

a single 32bit process can only see 2GB of RAM

oh yeah that map is gigantic

20:39

it can use 3GB with largeaddressaware

largeaddressaware comes with some drawbacks

yeah cannot remember

ill print malloc return value

but i think i had a patch for a 32bit game that then could use more RAM

is there a perror function like on linux you think?

Are we even using 2g?

well, obviously malloc returns 0

20:41

when it crashes, memory usage is around 180MB

20:42

so it should be able to get 280MB more

As @heinrich5991 pointed out it might be just a ridiculous amount of fragmentation

btw thats the upload data, so i dont think it can be minimized

20:42

it shouldnt contain any 0 vertices anymore

So if we aren't using 2g (which sounds like way too much, we shouldn't be using even close to that) then it's likely we have fragmentation

maybe it could be uploaded in chunks

20:43

does it also crash in release mode @Chairn ?

Permission denied: Not enough space

We can optimize our heap usage, maybe use a different allocator

so, weird (edited)

Deleted User

does it also crash in release mode @Chairn ?

yeah, it crashes as well using the build from ddnet.tw

i mean 250mb is already nearly 3 times 80mb so i could understand it more xd

20:44

and still its not too much

20:44

i wonder if the compiler can just sort the heap address better

that's the runtime

20:44

and the allocator

20:45

and possibly the kernel

ok

probably the kernel

is there a way to see memory maps of process in windows?

You know what else could cause this? a corrupted heap

but would sanitizer show it?

20:46

or do u mean inside the kernel?

Would have to be quite unlucky but I've overwritten the stack return with a pointer to _start once

20:46

I believe everything is possible

maybe an out of bound array?

@Deleted User could be that it never happens on linux due to the way malloc lays it out there

but yeah @heinrich5991 u might be right, on linux 32bit it works

I'm not very familiar with memory debugging on windows, let me read up a bit to see what tools there are to observe allocations

20:48

Actually let me see if it happens for me

oh on wine it crashes in 32bit

vmmap seems to be the tool

i dunno how well wine emulates the win kernel xd

seems like a fidel reproduction

@Learath2 ^ i dunno if malloc layout wrapped in wine

yep consistently crashes, cool

we could just use non moved pointers for 32bit as workaround

20:53

i implemented both, uploading small chunks and big chunks

non moved pointers?

yeah, so you can delete the pointer directly

20:53

instead of keeping it alive for GPU upload

Using VMMap from windows, address space is not that fragmented

20:54

white is free

Image attachment

@Chairn wanna test quickly if changing int BufferObjectIndex = Graphics()->CreateBufferObject(UploadDataSize, pUploadData, true); to int BufferObjectIndex = Graphics()->CreateBufferObject(UploadDataSize, pUploadData, false); free(pUploadData);

20:54

works

what line?

20:55

oh found 850

yeah

no, still crash

ok so its the single allocation

20:56

not the previous ones

20:57

i wonder if the driver wouldnt crash

20:57

since it probs has to allocate the same amount of VRAM

What is this even?

Windows headers require the default packing option. Changing this can lead to memory corruption. This diagnostic can be disabled by building with WINDOWS_IGNORE_PACKING_MISMATCH defined

ah

20:58

its probs #pragma pack(...)

20:58

so you can ignore alignment for e.g. 64bit variables

20:59

or increase alignment for variables that dont use such high alignments normally

This is what happens if I try to build with msvc btw

20:59

didn't touch anything

oh thats strange tho

21:00

we dont use such stuff maybe a broken header?

not surprised for even a second that something about building on windows is broken

LTO would warn about mismatches in structures (edited)

u want to implement a new allocator to fix some 32bit client issues from the 2000s?

Here I go diving into the stupidities of MSVC again when I try to debug a small allocation issue

@Ryozuki i dont even know if thats fixable, since it has to be a single buffer on the GPU anyway

21:02

so even if its fixable on the software side we'd need to see if it works on GPU side

There is absolutely no way we can't fit a 280M buffer in the ram

^^

actually, this is what i got with VMmap when it segfaults without crashing vmmap as well

It being un-fixable is absurd

Image attachment

@Learath2 i'd say the same, but windows proofes me wrong

DX Calls: Any 3D game engine or DirectX application will trigger these calls

21:03

directx

21:03

windows at its best

so it definetely use way more data than process manager shows

There was a tool for high frequency sampling like perf on linux I remember that guy from chromium project with the nice blog was using it

21:04

Having trouble remembering the guys name

Learath2

21:04

trollet

@Chairn if u want to play that map u have to use opengl 1.x for now ^^

meh

not using opengl 3.3 justatest

justatest

using 32bits in 2021 is just ridiculous

not using graphic card for a game is ridiculous as well

@Ryozuki believe me in south america everyone uses 32bit xD

Us not being able to fit a game that should work on a gameboy into the 32bit address space is what's ridiculous

Chairn

not using graphic card for a game is ridiculous as well

i wonder if > DX Calls: Any 3D game engine or DirectX application will trigger these calls

21:06

means that it doesnt recognize opengl

intel graphics are not as stable, sometimes fps goes down and game is not really fluid

to trigger that

@Chairn can u just disable intel completly?

now the real question is why does it work for the 32 bit and not the 64 bit?

21:06

how i do that?

in device manager

21:06

u can disable components

well, intel graphics is used all the time except when i launch a game

i think i did that too some day

21:07

i just know that i could set opengl configs in the nvidia control panel

21:07

back then

memory view from vmmap when it crashes

Image attachment

Embedded image

21:08

like this

21:08

but dunno if that still exists

21:08

havent had nvidia since a longer time ^^

Chairn

memory view from vmmap when it crashes

See that looks more reasonable

21:10

I swear this thing requires more upkeep than a bleeding edge linux system

21:10

I don't touch it for a week and compilation breaks

but for some reason i never have problems with MSVC XD

21:11

its only ur setup

I literally have no setup

21:11

I downloaded VS, I cloned ddnet

21:12

I wouldn't want to set anything up on this sad excuse for a "stable" "ecosystem" atleast apple supports their single ecosystem perfectly

Deleted User

@Chairn can u just disable intel completly?

just had to reboot after this

21:12

it turned off the screen with no way to turn it back on

ah too bad

Use Metal and develop apps like they want you to and you will never have a single issue. You can compile for x86_64 and it runs perfectly under M1, you can do fat binaries, you can compile for ios and run it under macOS

Deleted User

Click to see attachment 🖼️

yeah, that's what i do, but no change on 64 bit client

21:13

only works on the 32 bit one

its really strange

21:13

did u install intel drivers manually?

yup

@Chairn you sure it's not using the gpu?

Learath2

Use Metal and develop apps like they want you to and you will never have a single issue. You can compile for x86_64 and it runs perfectly under M1, you can do fat binaries, you can compile for ios and run it under macOS

i better say nothing about MacOS ^^

yes, power led changes color when using GPU

21:14

blue for intel, orange for Nvidia. 64 bit client start, led goes to orange then fall back to blue

21:14

when looking at the log, it says intel graphic

21:14

32 bit client, says nvidia

btw i think newer clients also output the version of the driver capabilities [opengl]: Version string

21:14

then u should see if its nvidia

ypu

I'll probably be able to take a look after the 2gb update to my 20gb installation of visual studio downloads

[2021-02-04 22:15:47][client]: starting...
[2021-02-04 22:15:47][sdl]: SDL version 2.0.8 (compiled = 2.0.8)
[2021-02-04 22:15:47][gfx]: Created OpenGL zu.zu context.
[2021-02-04 22:15:48][opengl]: Vendor string: NVIDIA Corporation
[2021-02-04 22:15:48][opengl]: Version string: 4.6.0 NVIDIA 425.31

Vendor string being NVIDIA strongly suggests it's running on the gpu

21:17

Though optimus does weird things, who knows maybe there is a shim reporting the version?

21:17

Ah yes, I need to restart my computer to finish installing software updates that are completely unrelated to the OS

21:17

brb

xDDDDD

21:19

created opengl zu.zu context

21:19

is that mingw?

21:19

maybe we should use mingw stdio instead of mSvc broken c98 thing

nah, release build from ddnet.tw

yeah that should be mingw

That should never happen

21:20

smells very much like UB

[2021-02-04 22:20:14][sdl]: SDL version 2.0.8 (compiled = 2.0.8)
[2021-02-04 22:20:14][gfx]: Created OpenGL zu.zu context.
[2021-02-04 22:20:15][opengl]: Vendor string: Intel
[2021-02-04 22:20:15][opengl]: Version string: 3.0.0 - Build 10.18.14.5162

its bcs c98 didnt had it, and MSVC doesnt care about C

mingw build because i did not configure nvidia to turn on for this one

atleast not C std

Ah true we need to compile to ancientC

21:21

and ofc an update to vs completely broke the project in which I just cloned ddnet so I need to clone again

Deleted User

maybe we should use mingw stdio instead of mSvc broken c98 thing

there are a lot of warning to fix then ^^

we can just use c++ for system.c

21:23

just as we wanted always

21:23

why struggle with C

system.rs

iirc the idea was to keep compatible with tw

yeah but our codebase isnt really compatible anyway

deen did convert to C++ but closed his pr

we could just copy paste if they ever have smth useful

don't remember his reason but he said it here

yeah i know

21:24

i think its mainly heinrich being against it xd

I don't remember this guy's blog name either

i think _mingw.h changes behaviour of c98 stuff https://github.com/mirror/mingw-w64/blob/master/mingw-w64-headers/crt/_mingw.h.in#L420

https://randomascii.wordpress.com/

Random ASCII - tech blog of Bruce Dawson

Forecast for randomascii: programming, tech topics, with a chance of unicycling

21:27

ETW is the name of the tool

21:27

Windows Performance Toolkit (xperf) is the thing it's part of

21:28

It's about as intuitive to use as radare2 though

21:28

Okay ddnet does not compile on MSVC

mingw emulates most windows stuff tho

21:29

so msvcrt is compatible

21:29

or does it compile in some instructions?

?

error message?

Learath2

?

thought it was related to the xperf stuff

21:32

but u just meant i doesnt compile ^^

First of all it's not generating the protocol so let me figure that out but the main issue is: WINDOWS_IGNORE_PACKING_MISMATCH this thing

21:36

Okay it is generating just fine

21:36

I guess it's just intellisense failing to find them

21:37

Yeah it's

error C2338: Windows headers require the default packing option. Changing this can lead to memory corruption. This diagnostic can be disabled by building with WINDOWS_IGNORE_PACKING_MISMATCH defined.

21:37

while compiling backend_sdl.cpp

that sounds bad

Okay figured it out, how this compiles with anyone else is beyond me, but apparently including SDL_syswm.h messes with the packing

21:40

it's fixed in SDL 2.0.12

u should create an issue for that then

21:41

if we dont use 2.0.12

including syswm after windows.h "fixes" it

21:41

I'm eager to find what it breaks though

21:43

Let us all come together and invent the ideal OS

21:43

I'm sure between us we can come up with something that's not as shit as every option available right now

let's call it systemdos

teeworlds-os

microkernel?

21:46

lets build it with rust

I must say the CRTs heap debugging functions are quite nice

link/explanation?

Image attachment

21:49

I can see what kinds of objects are on the heap, I can see where the allocations happened, I can see the call stack leading to each allocation

nice

Okay, the heap seems completely fine, so no heap corruptio

21:56

n

vmmap really seems like the tool to look at this but ofc it can't do high frequency

[ddnet/ddnet] Issue closed: #3239 deepfly bind in controls

In other news we seem to leak a bit

[ddnet/ddnet] Pull request opened: #3602 Editor: keyboard support for fonts

Press T to enter text mode and place font letters based on keyboard presses Support custom number and letter offsets using json files. self made cursor.svg (CC0 or whatever u need)

Checklist

[ ] Tested the change ingame
[ ] Provided screenshots if it is a visual change
[ ] Tested in combination with possibly related configuration options
[ ] Written a unit test if it works standalone, system.c especially

...

Depending on the map size joining and leaving a server leaks some heap

22:14

I don't get it we don't use nearly enough heap for address space to become an issue :/

Well the error is ERROR_NOT_ENOUGH_MEMORY so I guess it is an issue

I managed to make it crash earlier, amazing

22:51

Preallocating an entire gb of heap memory doesn't seem to help either wtf

are the vectors maybe too big?

22:53

i think it allocates width * height, and then prepares the buffers for GPU upload

22:54

tho it "only" uses 700mb on that map

22:54

so in worst case it shouldnt be much higher than 1400MB

22:55

ah yeah it works on 32bit linux anyway

The tooling looks pretty but it's quite shit, VS snapshots quickly seems to be well integrated with the crt, but it has no fragmentation view. VMMap has the fragmentation view but it can only snap at 1 second intervals

forgot already

I don't know why/where we are hitting the boundry because I can't see the whole memory map because there is no way to make vmmap snap when I want

22:56

I guess people just abandoned 32bit

22:56

in 64 bit out of memory is just a joke anyway

22:56

atleast the address space kind

thats true, i just wonder what gcc does differently when compiling / or if wine actually simulates windows that accurate xD

22:56

when compiling for windows*

My theory is that our shared libs land in a shitty position in the address space so that there is no contiguous block of memory of 290m after the first shared lib

possibly

then heinrichs first guess would be right already

22:58

i wonder if we can tell gcc to try to maximize heap or smth

look at my gigantic heap

Image attachment

22:59

Still didn't save me

fails on a smaller allocation

23:00

There has to be a trick to move the stupid shared libs towards the end

32 bit = 4,294,967,296 bytes (4 Gigabytes) 64 bit = 18,446,744,073,709,551,616 (16 Exabytes) the difference is ridiculous

I wonder if vmmap is open source, I could modify it to snap sooner

my mobile is also 32bit, should test the map there xD

does it support opengl

im dissapointed in your dev team

why

zwelf doesnt know who nouis is

cuz he is not on this discord

yh just talked to him about biggernouis and he had no idea about it

23:02

and didnt even know nouis

@Ryozuki even OpenGL es 3.0

23:05

tested it few months ago on android

poggers

You know what could work? Preallocating a large chunk for this kind of stuff at the very beginning

but i'd call this map the worst case

23:08

2660x1100 or smth like that

23:09

with many layers

LARGEADDRESSAWARE works

23:11

though I'm sure we are breaking something with that so I'll enable top-down allocation to see what breaks

@Learath2 you can maybe disable ASLR to map libraries to places you want

23:13

probably not a good idea though

23:13

(but an interesting exploit target)

Anyway from 2gb you dont get to use much if at all for your heap

23:17

So keep that in mind next time working on 32bit stuff

what's the reason for allocating 250MB btw?

23:17

sounds like a lot

That you'll have to ask @Deleted User

23:18

Maybe there is a way to upload it to the gpu in chunks? I honestly wouldn't know

23:19

Or maybe a way to avoid making a copy of it

23:22

Yeah this allocation is just massive I see why it won't fit now that I've turned on LARGEADDRESSAWARE

pix

Anyway, I think this is a viable(tm) solution. Forcing malloc to return memory top-down didn't break any of our libraries either

LARGEADDRESSAWARE you mean?

well its allocating the vertices

LARGEADDRESSAWARE gets us well large addresses

23:24

forcing the top down allocation lets me make sure the larger addresses that can now be returned by malloc doesn't break any library

ah

This is the largest snapshot I managed to capture but it gets a little larger than this, no way to get faster than 1s intervals in vmmap

Image attachment

23:25

This is with large addresses, with small addresses the huge orange blocks don't fit above the first purple nor is there enough contiguous memory anywhere after the first purple

appereantly one layer is pretty full of tiles, didntr check tho if true

if this stupid tool was integrated into VS I could debug it even further

23:29

Anyway, with only small addresses this is the largest heap I could capture

Image attachment

23:29

That white block in the middle is 260M so I guess we are making a huge deallocation that's causing fragmentation

freezebg is 2415*1620 tiles each tile consumes 4 vertices each vertex 24 bytes

Nah, I just need more resolution, idk why that happens

23:36

There is enough space but that 260M block is just left empty for some reason

factorio only renders stuff near the player I think

23:38

(given that the world is unbounded, they have to do that in some way)

factorio is in a class of optimization of it's own, us mere mortals can't achieve that

I think they had a blog post about that

23:39

they render each layer near the player

the things they come up with to cut 1ms from their pipeline really should be inspiration to us all

and then just render new stuff as the player moves

I guess I can code tooling to figure out why the allocator is skipping that 260m block

23:40

Or maybe a high resolution memory map

23:41

Or maybe dig into some very old blogs to figure out what they did back in Windows XP

heinrich5991

factorio only renders stuff near the player I think

it has nothing todo with the rendering

23:44

pre creating the buffer is fastest, what else would u do

really? I thought you upload information about hte whole map up-front

runtime creation is slower

factorio doesn't do that, I think

only about tiles that exist

23:44

the map has a layer that filles all tiles tho

23:44

yeah but why should that be faster?

it would use less memory

preallocating everything is fastest obv, but has disadvantages as RAM usage (edited)

probably has a higher up-front cost, too

why

because you need to look at every tile at the start vs only looking at tiles near the player

we also only look to tiles near the player

23:46

the renderer doesnt render anything not visible

oh, so we only allocate that buffer for later usage?

yes

leaving it mostly empty?

no

23:47

we have 2 buffers

23:47

one that helps with finding offsets and one on the GPU containing vertices

and the one containing vertices contains info about every tile, not only near the player? so it has a higher up-front cost to build?

ah thought u mean at runtime

23:48

yeah, should have

23:48

but in tw u alsmost always spec the whole map probably

23:48

or huge chunks

23:49

so wouldnt really be worth optimizing it in chunks for RAM

Q: Isn't tmpTiles also on the heap?

we just had someone who couldn't run this game on that map ^^

he can run it

23:49

with gl 1.x

23:50

@Learath2 yes, but tmp xd

did we tell him?

@Learath2 in the end it worked on linux 32bit

23:50

so cant be out of address space

23:50

@heinrich5991 yes i did

nice

ASLR works differently on each platform

@Deleted User linux might allocate differently, might even be a different userspace allocator

So does malloc and the underlying kernel allocator

yes

23:51

thats what learath tries to understand since 2 hours isnt it

23:51

(how we can change it)

It quite obviously runs out of address space, the largest remaining free contiguous block in the address space is 260M which is not enough

so large address space works or what?

Large address space works and didn't break anything in my 10-20 minutes of testing

23:52

Though it's a shit fix, it would work for people running the 32bit binary by mistake or on situations like the one reported

23:53

It wouldn't work so well on an actually memory constrained 32bit computer

yeah ok

23:54

dunno the details of that

23:54

anyway i'd prefer to disable GL >= 2.1 on 32bit completly than to fix this

Figures

for me this is also a pretty worst case

23:55

the editor already warns at 1000x1000 tiles

sounds good. maybe we could also display an error message

this is like 3 times the size

You are okay with it crashing too I guess

i am indeed

instead of crashing I mean

xD

an error message to tell people to enable opengl1.1 sounds better to me

@heinrich5991 we are using unguarded mallocs everywhere anyway, getting to an error is tough without exceptions as teeworlds code is built for crashing and burning mostly

funny how slow pcs get a slow implementation

23:56

almost ironically xD

@Learath2 since it's only about address space exhaustion, we can guard that single big allocation

23:57

not about actual out of memory problems

Well I'd feel warmer about that

i think its better to not allow gl >= 2.x at all

23:58

completly disable it

23:58

most ppl with 32bit system have a old pc anyway

23:59

or maybe only on windows

That I don't quite like

i'll test with my mobile now, but since android is basiucally linux kernel, probs will work

Dooming people to worse performance because some very edge case massive maps crash sounds silly

Exported 737 message(s)