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 2020-06-18 00:00:00Z and 2020-06-19 00:00:00Z

noby

Hey guys, i have a question: is there a thing like TWRequest.php thats like newer and better

01:34

Piepow made a teestalker webapp long ago so they told me that someone probably has a better thing like it

01:36

I know there is <status.tw> but i dont like the fact that its not open source and their api is not really good so i wanted to know if anyone can help

01:37

And of course it's 3:30 AM, no one is gonna answer at this time

@Alexandar for DDNet we have a hacky solution in https://github.com/ddnet/ddnet-scripts/blob/master/servers/scripts/ddnet.py#L722 based on https://github.com/ddnet/ddnet-scripts/blob/master/servers/scripts/ddnet7status.py for 0.7 and https://github.com/ddnet/ddnet-scripts/blob/master/servers/scripts/teeworlds.py for 0.6

[ddnet/ddnet] Issue opened: #2271 Crash in sql_connector.cpp:25

Caused by the new thread-safe SQL probably. @Zwelf ``


Core was generated by

./DDRace64-Server_sql -f servers/block.cfg'. Program terminated with signal SIGSEGV, Segmentation fault. #0 0x000055dd4e07e217 in CSqlConnector::ConnectSqlServer (this=this@entry=0x7fa8e60dcc50, ReadOnly=true) at /home/teeworlds/src/master/src/engine/server/sql_connector.cpp:25 25 for (int i = ReachableServer, ID = ReachableServer; i < ReachableServer + NumServers && SqlServer(i % NumSe...

06:16

[ddnet/ddnet] Issue opened: #2272 Crash in CSqlScore::RandomMap

Caused by the new thread-safe SQL probably. @Zwelf ``


Core was generated by

./DDRace64-Server_sql -f servers/8314.cfg'. Program terminated with signal SIGSEGV, Segmentation fault. #0 std::shared_count::operator= (r=..., this=) at /usr/include/c++/8/bits/shared_ptr_base.h:1078 1078 class shared_ptr [Current thread is 1 (Thread 0x7ffaff2fcdc0 (LWP 11112))] (gdb) bt #0 std::shared_count::operator= (__r=..., this=) at /usr/include/c++/8/bits/shared_ptr_ba...

06:18

[ddnet/ddnet] Issue closed: #2272 Crash in CSqlScore::RandomMap

06:18

[ddnet:master] 1 new commit

6fcb8ab Initialize m_SqlRandomMapResult (fixes #2272) - def-

pepeH

i should change a nickname eh, 2nd time i get pinged by accident

lol

Does ddnet use multiple SQL read/write server?

it seems so

08:33

I just looked into the config file

[ddnet/ddnet] Pull request opened: #2273 Make SaveID DEFAULT NULL

@Zwelf

Zwelf: Where we had the crash there was indeed just one SQL server, and connection to it kept failing

@ChillerDragon if we have the same default skin, we already map that

yes i just realized :D

10:31

sorry

10:32

ok but on unique i at least see my own greensward

10:32

and 0.7 to 0.7 should work without problems

10:32

supporting all custom skins

10:32

oh okay i just realized i now see my own skin after chaning it

10:32

but not when i rejoin

0.7 to 0.7 works without problems already

10:33

atleast from what I could test, it's really hard to test with more then 2-3 clients when my laptop would rather become a grill

not for me

10:33

only one client needed for me to test it

10:34

when i join i have default skin

OH that's the bug

10:34

It happened to me like 2 times

10:34

could never reproduce it again

i can consitantly

10:34

thats not how u spoell it huh?

consistently

ty

10:34

spell* also :D

10:35

yea that should be fixed

can you add some dbg_msg s in teeinfo.cpp? see why tf your skin is getting mapped to default

but also it would be nice if the 0.7 body is "greensward" that it shows greensward in 0.6

10:35

any functions or lines you have in mind?

actually not teeinfo.cpp

10:37

in gamecontext.cpp:L1410

10:37

well L1420

10:37

before and after the fromsixup

Embedded image

10:45

that looks good

10:45

afk for a sec

print Info.m_apSkinPartNames[1] instead afterwards

also good

ChillerDragtroll

omg burn it with fire https://github.com/MrEAlderson/Rekit

MrEAlderson/Rekit

Teeworlds server ported to Java. Contribute to MrEAlderson/Rekit development by creating an account on GitHub.

11:19

why would anyone port tw to java :D

11:24

@Learath2 seems like cammostripes is even sent over the wire if i understood that correctly https://zillyhuhn.com/cs/.1592479429.png

Embedded image

11:24

maybe the client just doent understands it

hm, it is entirely possible I'm forgetting a part of the protocol

11:30

I don't think I have to respond to a STARTINFO though, do I?

@deen so the only thing you have is a python version

11:34

Is it possible to integrate that into a website

11:34

Im not really experienced with py

@Learath2 i think on connection the client expects NETMSGTYPE_SV_CLIENTINFO which includes the initial skin and there currently part 1 and 2 are empty

11:39

from the client https://zillyhuhn.com/cs/.1592480333.png

Embedded image

11:39

https://github.com/teeworlds/teeworlds/blob/14d26d42925f472f34b76b97df9024640dc3a82a/src/game/client/gameclient.cpp#L1056

teeworlds/teeworlds

A retro multiplayer shooter. Contribute to teeworlds/teeworlds development by creating an account on GitHub.

11:41

Embedded image

BAN

1

So I'm not sending the correct clientinfo?

11:46

huh so weird

ok my wireshark was wrong that was a message from the client

@ChillerDragon can you check for a CNetMsg_Sv_ClientInfo on the client side where m_Local is 1?

[ddnet-web:master] 1 new commit

5e7019e Add donation by Shisui. - def-

12:23

[ddnet:master] 2 new commits

882185e Make SaveID DEFAULT NULL - def- 8c17e7c Merge #2273 - bors[bot]

12:23

[ddnet/ddnet] Pull request closed: #2273 Make SaveID DEFAULT NULL

Embedded image

12:35

there is only 1 tee on my server so its always m_Local when i print CNetMsg_Sv_ClientInfo

something with the skin colors is weird

I guess you should check what's sent

meh it happens if I build with release, doesn't if I build with debug

what about RelWithDebInfo?

trying that next

13:49

yep happens \o/

13:52

oh I see the problem

13:57

uh, I fixed it but the fact that it worked in debug mode concerns me

things that only work in dbg mode (read: without compiler optimizations) ive seen a couple times in the past and, i was told its likely due to failing to zero some memory somewhere, but the actual problems turned out to be related to casting different sizes of memory chunks to other types carelessly without regard for the signedness

13:59

and some of the optimizations it would do would cause it to break horribly there

But the way this one happened there is no uninitialized memory to go wrong :/

is it a signedness thing?

14:00

like doing *(unsigned int *)mem = 0xffffffff; if the mem was signed char type

14:00

or something along those lines

issue is that I was calling ToSixup for the skins of sixup players, which is obv not needed

14:04

but calling tosixup shouldn't destroy the skin completely, yet it does in the release build and doesn't in the debug

the things i saw before that work only in debug build are related to compiler optimizations and also sometimes generate a warning on some oses/gcc versions tahts like "dereferencing type-punned pointer will break strict-aliasing"

yes, UB is nasty. you could try ubsan, @Learath2

-fsanitize=undefined?

yes

never heard it as ubsan before

I ran this through ubsan before, didn't really find it :/

https://clang.llvm.org/docs/UndefinedBehaviorSanitizer.html @Xusep

14:07

@noby

yeahh

14:07

does it check for the same type of thing that causes the "dereferencing type-punned pointer will break strict-aliasing" warning?

hm I'm usually good at spotting this kind of thing

14:08

yeah it's uninitialized memory, how did valgrind miss this even

@noby apparently not

ooo

I thought ubsan checked for all undefined behavior

hey learath how to use c# in ddnet server¿

it's a fast ub checker, some types require a massive amount of instrumentation

i guess i was accidentaly right before about uninitialized memory

14:09

but also assumed that it would be caught by these tools

@bubasio only C++ works with the server

@bubasio I mean you could probably add hooks for C#

ok ty

it will probably be more work than just using c++

C# is very dynamic, but that'll be a huge undertaking

@Learath2 that would require more knowledge of C++ than it would be to do it in C++

js runs in ddnet sv¿

no

lol

ok

The server is entirely in C++

u can extend the server code in c as well

14:11

getting it to work with other languages will be a lot harder i think

I did add a V8 engine at one point though but never really did anything except messing around with chat command hooks

14:11

same with lua

noath

1

14:12

Hm, I wonder how I should initialise this thing

justatest

https://reviews.llvm.org/D32199

I guess I can use inline initializers now that we are not stuck with C++03

this adds -fsanitize=type

Oh that sounds cool

I think it's not merged yet, but I can't quite figure out the UI

so that will catch the warning i was trying to describe?

yes, but I think it's not merged yet

cool lol

Btw, did you know that ASan allocates like 64 terabytes of ram?

did you know every haskell programs allocates a similar maount

Oh, what for?

for the GC I'd guess, for easy "allocation"

ah yeah, haskell is garbage collected

can you see whether this thing is merged already or not, @Learath2?

the TSan?

yes

looks merged

14:28

ah no it's not

14:30

The LLVM part and the Runtime still seems to be waiting on a review

14:30

and the last comment suggests some big change happened with llvm that these diffs need to be modified for

I'm very baffled, why would my default member initializer would get ignored for a constructor that doesn't set that field?

oh great I'm stupid

14:58

even better idk how to fix this

15:00

strncpy is undefined if the src and the destination are overlapping

15:01

okay I will do something very ugly and hope god will forgive me for it

hm we do have a bit of UB though

15:08

like unrelated to this patch

15:12

The amount of things ubsan finds in the client is also very scary

15:14

@ChillerDragon that should fix the skin thing, the only thing that remains there is the color conversion

15:15

which after 4 hours of staring at it, idk how to fix

we do so much division by zero

division by zero?

15:39

float? int should trap, right?

yep float

15:41

Oh no, I seem to have something wrong about the color code in ddnet

15:43

Oh actually I just forgot that the hsl cone has points that map to multiple points on the rgb cube

[ddnet:master] 1 new commit

7aed647 Allow larger map names - def-

16:34

[ddnet/ddnet] Issue opened: #2274 Points don't get updated when finishing Journey Man 2

@Zwelf As reported on Discord

[ddnet/ddnet] Pull request opened: #2275 Remove ugly color hack

This stateful color has been bothering me for a while. I tested this and it doesn't seem to break the colors but to avoid a repeat of the last time I'd appreciate it if a couple more brave souls try this.

We should get client QA people

16:59

@jao lmk when you roll out the bot with the github cog

weekend i guess

@Learath2 nice did you also add greensward? <3

hm, I want to support all ddnet skins, but I can't do that with this linear lookup

ok nice

hm, maybe I could, I think the current skin system also only uses linear lookups, can't be that bad

if its onle like 30ish skins it should be fast enough

Image attachment

19:17

github updating its design

[ddnet/ddnet] Pull request opened: #2276 Fix segmentation fault when creating sql tables failed

GameServer()->OnShutdown(true); (which would wait on running threads to exit) wasn't called if an error occured in the server init. Therefore CServer was freed before the init-thread finished executing, causing a use after free from CSqlConnector::ms_ppSqlReadServers pointing to CServer->m_apSqlReadServers. Fixes #2271 Note that the server exits when creating tables failed. Since the tables are created by default, it has to be disabled explicitly (e.g. `add_sqlserver r teewo...

@deen https://github.com/ddnet/ddnet/pull/2276#issue-436713977

Note that the server exits when creating tables failed. Since the tables are created by default, it has to be disabled explicitly (e.g. add_sqlserver r teeworlds record teeworlds PASSWORD "localhost" 3306 0) if the servers should start even when the database connection fails.

Fix segmentation fault when creating sql tables failed by Zwelf · P...

GameServer()->OnShutdown(true); (which would wait on running threads to exit) wasn't called if an error occured in the server init. Therefore CServer was freed before the init-thread...

yeah, but it shouldn't crash

I think that was just an additional note for our server set up

19:26

should I fix it?

[ddnet/ddnet] Issue closed: #2271 Crash in sql_connector.cpp:25

19:29

[ddnet/ddnet] Pull request closed: #2276 Fix segmentation fault when creating sql tables failed

19:29

[ddnet:master] 2 new commits

f5ac6ee Fix segmentation fault when creating sql tables failed - Zwelf 6f8f606 Merge #2276 - bors[bot]

[ddnet:master] 3 new commits

d06f194 Stop playing map sound when demo player is paused - def- 0d9d873 Reset hook & collision in prediction after kill - def- 13237ee Merge #2245 #2266 - bors[bot]

19:37

[ddnet/ddnet] Pull request closed: #2266 Reset hook & collision in prediction after kill

19:37

[ddnet/ddnet] Pull request closed: #2245 Stop playing map sound when demo player is paused

@heinrich5991 I have set the table create to 0 already on the affected servers

[ddnet/ddnet] Issue closed: #2274 Points don't get updated when finishing Journey Man 2

21:04

[ddnet/ddnet] Pull request opened: #2277 Minor code style fixes in sql_score.cpp

Can we have an Unfreeze entity on the Switch zone layer? If Deep and Undeep is there as well as Freeze, why not add Unfreeze for more possibilites for mapping puzzle maps or mapping different ways to approach races?

Huh 2263 is interesting, I didn't think a character could exist without it's core being a part of the gameworld

Freeze is only there so you can freeze for different times, switch number counts as number of seconds. For unfreeze this makes no sense

21:43

@Learath2 yeah, also surprised me

21:43

Seems like a wonder that we only had this crash with super and not with many more features

How do you get to that state btw?

with /spec

21:44

on spec-enabled maps

Oh when you spec your core is removed but the ccharacter still exists?

yes

Yeah interesting that we didn't hit this more often

21:45

I don't like having two sources of truth for super

@deen can you specify what makes switchable unfreeze "useless"?

The others are not switchable @Cellegen | HU

on switch layer, you can turn on and off with the right switch tiles: Freeze, Deep and Undeep

21:47

wdym its not switchable?

[ddnet:master] 2 new commits

2c7f4c7 Minor code style fixes in sql_score.cpp - def- 59e3009 Merge #2277 - bors[bot]

21:47

[ddnet/ddnet] Pull request closed: #2277 Minor code style fixes in sql_score.cpp

every single tile on the switch layer has it's potential to be turned off and on by the current switch tiles that do that (edited)

have you tried switching freeze tiles on and off?

i remember it working back then, some puzzle maps have that mechanic

Huh, I also thought those were only on the switch layer to adjust duration

do you have some map as an example?

game layer and switch layer are a big diff, idk where you came with that assumption

21:51

Puzzle box by Soapy sandwich and another map which i cant find

the problem with adding the seconds they should freeze for in the front layer is that the data structure doesnt really allow for that. meanwhile in the switch layer each tile stores 2 more numbers. this is probably why someone thought timed freeze tiles could fit in there, while not actually being switchable

but it's a well known puzzle map

21:53

So how exactly would it work if what i believed was totally incorrect?

21:54

nvm puzzle box does not use that (edited)

I honestly don't think switchable freeze exists

21:54

though it might be a cool idea

oh no

a different tileset would do the trick so it won't get replaced

sounds like either losing the kinda sensible layer structure or backwards compatibility

21:57

hm i guess the timed freeze tile would fit well in the speedup layer

@Learath2 i saw it the other way around, I'm not used to think that the numbers on the Switch tiles are indicators for the amount of seconds these are active. Tho having tiles with switchable capabilities might be easier if one of the features have that

I once wanted to add a lot more to the switch code, like basic logic operators and stuff

@Patiga i wouldn't call it well done on the Speedup layer as it's name stands different approaches

21:59

i would love to play with them in the editor once you're done

since there is only one tile in the speedup layer so far, i believe that it could be used for such diverse tiles

I never figured out how to make this user friendly at all

then it wouldn't be only speedups but switchable entities too

Imagine turing complete switch layer

22:00

maybe we could compile pacman into teeworlds switches

imagine tuning tiles...

22:01

nope, switchable tiles are easier

We have tunezones

huh i guess tune layer would fit just as well

22:02

then it wouldn't be only speedups but switchable entities too

@Cellegen | HU my suggestion was that only the timed freeze tile should go into the speedup/tune layer

Oh, did anyone see that guy who compiled C to powerpoint?

also @Learath2 how well does the coordination system handles well detailed numbers like (0.00)

coordination system?

cuz i think TeeUniverse have that feature but ddnet editor doesn't

22:04

I'll demonstrate on pm (edited)

im also interested :)

Then I'll post here as a bug. This bug makes the coordinations on the DDNet Editor fucked up when modified, since when it updates, it cannot register numbers beyond [0.xx]. But I would like to have the possibility to make a more precise number editing on the DDNet Editor, so it won't be a hell of a trouble finding a perfect clipping point, so it won't bug out.

Image attachment

22:09

left is ddnet editor, right is TeeUniverse editor

I don't actually remember whether we save these as floats

22:10

If we do, it should certainly be possible to get the editor to edit them

it shouldn't but saves it anyways

22:10

a lot of people mapped maps on teeuniverse and got away with it without any obvious bugs

pretty sure parallax is not a float

22:11

ah its about position

@Patiga I sorta remember us storing these as fixed point actually

the teeuniverse exporting mechanism is as weird as these possibilities that even the ddnet engine can read it

I love how people instead of reporting this as an issue, used teeuniverse

exactly

22:13

teeuniverse isn't famous for the best GUI or for the best features it has, but rather the advanced tools it has

22:13

for animation and information gathering, editing and using, it's absolutely perfect

necropotame was working on quite interesting stuff, I wish he hadn't left

he did make it open sourced

22:14

that"s something

if you have any interesting features you'd like in ddnet, leave an issue. I kinda like doing editor patches

alright, will do

i am confused

22:17

so in my map parser i parse the position as integers

They are saved as integers

22:18

It could be that however they are fixed point numbers in reality

22:18

I think quads have them

that doesnt really explain this

22:18

i tested it on the map puzzlebox, the second group

22:18

this one has a position of (0, 25)

22:18

and my parser says (0, -25)

22:19

and if im not too tired float wouldnt explain that

22:19

*shouldnt

@Patiga so wait on that map, the second group has a float position?

no, its simply 25 on there

22:20

not float

22:20

(in the ddnet editor)

ah but your parser thinks it's -25?

yes

Huh

[ddnet/ddnet] Issue closed: #2262 Crash with super

22:23

[ddnet:master] 2 new commits

e38cd6b Fix /spec + super crash (fixes #2262) - def- 7d42867 Merge #2263 - bors[bot]

22:23

[ddnet/ddnet] Pull request closed: #2263 Fix /spec + super bug

for whatever odd reason pos x and y are negative but are negated when displayed

does anyone have map where there should be floating point positions?

doesn't look possible to me, teeuniverse must be doing some weird scaling trick to achieve it

22:27

@Cellegen | HU do you know map that uses precise coordinates?

xmas move on the Qshar gores server, MAIN category

22:27

the first place i found this

the quads do use precise values

all the animations are set on different intigers

yeah quads do have fixed points

and when you mess with them, its gone

22:28

obviously

animations are set on integers?

22:28

can you show what you mean?

well shit, i can record it but you can reproduce the bug by editing the map and try to change the values between the layers

22:29

all of them. gets fucked

hm, I was going to go to sleep but now I'm intrigued

the map if you're too lazy

XmasMove_3f11eba4.map

329.68 KB

@Cellegen | HU which layer should I take a look at?

22:31

or layer group

QFr (freeze) is the easiest to see the bugs

22:33

first you cannot change anything but the position only cuz thats the only one it has no float in it

22:33

anything that has float in it will get it fucked

22:34

sometimes it renders it completely transparent and gone

uhm what else is there to change about a quad?

you sure? look closer

Image attachment

22:35

guess which one i wasnt able to edit

but what did you do to id?

22:35

it*

opened it as i got it downloaded from the original source, edited it and this was natural for it to happen

22:36

i didnt do anything cuz i cannot do anything with it in the first place

did you place those quads yourself? i cant find those

animation seems fine cuz it supports floats but dragging the picture is not possible

22:36

drag some of them out, you'll see it

Honestly I don't get it, I moved those from the left and they move fine, I can also move their corners around

position is possible cuz its not float based, but i cannot change the shape of the picture

I can play with the color

weird

Image attachment

didnt manage to recreate that bug

Image attachment

22:39

howw

@Learath2 nice

22:39

so you found the one with the float indicator in it

[ddnet/ddnet] Issue closed: #1307 Crash in CRingBufferBase::MergeBack

which ones did you use?

No I didn't I can change the shape just fine

22:39

I'm the one who messed up the shape by hand

[ddnet/ddnet] Issue closed: #946 SQL stuff doesn't seem to do any locking whatsoever

Image attachment

22:40

this one @Patiga

@Cellegen | HU so if you try to move the corners of that it breaks?

i used all of those up but found one working deeper in the map

you cant change shapes, it wont move the pictures

22:40

you cant zoom in or out the pictures

Are you stuck with an old version maybe? I can move the corners just fine

i have latest

can you just record it please, I don't quite get it

22:41

I can even make this an image quad with smoke in is just fine

the ones a bit right of spawn kinda worked for me. there were 3 quads that had their middle point on the exact same spot tho, so im not sure if that couldve been a editor bug

the middle points are quite odd in this map yeah

i could stream my broken ones in the voice channel

sure

22:45

huh what an odd way to put quads

22:45

I guess it's generated though

thats everything i could find broken for me, the corners not matching up with the texture

ah.

22:46

good laggy shit computer cant handle anything

22:46

for some fucking reason

22:47

jesus christ next time imma prepare improving my graphic settings

22:48

@Learath2 i'll leave it as that for now, tmr i'll try to demonstrate the bug with more information in hand

@Patiga the corners not matching up is odd though

yes

22:48

trying to find those with my parser rn

22:48

yours looked a bit weirder tho

22:49

isnt the scale on yours wrong?

the more I messed with it the more it went off the corner

22:49

scale?

ah no

22:49

that looks just fine

actually could it be that at the start of the position envelope it's not at 0?

aaah

22:50

ye

22:50

that would make sense

22:51

yes, removing the position envelope fixes the position

but the x is 0 in the position envelope at time 0

i feel like im trolling all of you but this was not the case, i found it. the POS to window does that

Image attachment

22:53

Image attachment

22:53

monkaS

22:54

thats... still emberrasing

I actually forgot what Pos TO is

22:54

but that's what's offsetting the corners

time offset i think

22:55

in the envelope

aha that explains it

22:55

I wonder why teeuniverse exports these quads so weirdly

22:55

these are quite trivial even

teeuniverse does some weird stuff

22:56

it really messes up grouping

the main problem is still on

Well when you can show the main problem we can take a look

Image attachment

22:56

yes

22:57

we need this feature to be improvised

There is no way to save this in the teeworlds mapformat

22:57

Groups don't save any precision

well shit if nerco was here

He could probably tell us what that actually does

should i dare ask the official infection staff about it? kek

kek

22:59

they dont do shit

could you save such an map in the editor and upload it?

yes

23:00

how Qshar's server does it is still different

i can dump all the saved information in the group data structure quite easily

I don't see anything interesting in qshars map tbh

qshar's server instead

@qshar is there any floating point magic in your maps?

zzzz

Anyway, I require sleep

good night :)

gn8

night

@Learath2 what do you mean xD

Exported 407 message(s)