DDraceNetwork - developer

i get a crash on client shutdown in sdl joystick stuff

00:10

i have a joystick connected which i dont use

00:10

in tw*

If using this character is bannable why do we allow it to exist https://discord.com/channels/252358080522747904/779761780129005568/987422395671392276

fokkonaut

i get a crash on client shutdown in sdl joystick stuff

any crash logs for @Robyt3 ?

Tater

If using this character is bannable why do we allow it to exist https://discord.com/channels/252358080522747904/779761780129005568/987422395671392276

theres no system to ban clantags

if my pointer is pointing to 0xfffffffffffffffc thats bad i guess? Its a bit high ._.

yeah doubtfully used as a real memory address, even if its virtual

07:21

r u subtracting from a nullptr or smth xd

so who pooped in ma pointer? -._

07:21

yes i am

07:21

OMG YES I AM

07:22

omg jopstar u so smart

07:22

send new trading advice i lost this months paycheck and rent on stonk market

07:23

aaaaaaaaaaaaa finally

07:23

ok i guess i was just tired yesterday night

07:23

such a nobrainer haha

dunno hard to say, rtx 4090 might be next hype. else there isnt smth special rn xd

nvodia

07:29

i feel there is something fundamentally wrong in my strat of writing tests

07:29

i run the program and save the output as test xd

07:29

now i broke my tests but the output looks actually better :D

07:30

its more a insurance of not changing the output than it being correct

07:36

how did neither asan or clang complain that i was substracting from a nullptr?!

07:36

jupstar go fix asan

do we ignore nullptr claculations?

doubt

07:37

its probably asan being bad and waiting for ur pr

null:json.c maybe the filter is wrong?

ou

or shift-base:pnglite.c is it probably

lemme try in 1 min

static unsigned const short *spShort = nullptr;
spShort -= 1337;
printf("spShort=%p\n", spShort);

ah cmn clang tidy could warn there

07:46

ah wait asan does!

07:47

maybe i didnt see it earlier in the flood of opengl errors -.-

how bout not using opengl < 3.3

old but gold

the old backends suck

07:47

i'd love to remove this bloat

idk i got some freezes

07:48

and i panic downgraded again

vulkan is legend

omg poggies finally setup matrix in weechat lets see how it performs

07:58

07:58

user list looks a bit cursed :D

def-

[ddnet-web:master] 1 new commit

df78ecf video - def-

ah nice i can see github notifications here!

08:06

#990 :justatest:

08:06

meh still no bot

08:06

a

08:13

ez rank1 ddnet fork https://zillyhuhn.com/cs/.1655539965.png

ChillerTiger? xd

its not my mom i swear

I wonder if pressing ctrl+c on ./scripts/fix_style.py could delete code if you kill it while its writing. Or is ctrl+c gracefully enough?

probably yes

yes wat

08:48

god bad?

maybe u cannot even prevent it with a sigkill

so robust

maybe it sends one buffer to kernel and then its not in user space anymore

yea its python u cant kill that

(i am not sure about it, just speculation)

08:49

well i mean clang-format

08:49

not the script itself

fat python probably takes longer to realize a sigkill is coming than the clang takes to fix the format axaxax

ChillerDragon

[ddnet/ddnet] Issue opened: #5451 Editor: save maps to tmp file instead of overwriting them

Currently map files are overwritten in place when being saved. So when the save process does not finish you end up with a corrupted map file. That actually happend to me in my windows blue screen days. I was working on a big map that took around 1 second to save. Since I was aware of my pc being unstable I developed the habit of saving non stop. And one time my pc crashed during the save of the file. Which broke the map file. It would be nice to save in a .tmp file first and then move t...

08:57

Robyt3

[ddnet/ddnet] Issue closed: #5451 Editor: save maps to tmp file instead of overwriting them

Robyt3

[ddnet/ddnet] Issue opened: #5452 Crash on client close when joystick is connected

``` Thread 1 received signal SIGSEGV, Segmentation fault. 0x000000006d62af4d in SDL_DYNAPI_entry () from ddnet\build\SDL2.dll (gdb) where #0 0x000000006d62af4d in SDL_DYNAPI_entry () from ddnet\build\SDL2.dll #1 0x00007ff7a4f0cf67 in CInput::CJoystick::Close (this=0x1cccf62da60) at ddnet/src/engine/client/input.cpp:179 #2 0x00007ff7a4f0cee4 in CInput::CloseJoysticks (this=0x1ccce64be70) at ddnet/src/engine/client/input.cpp:171 #3 0x00007ff7a4f0c88b in CInput::~CInput (thi...

comp=-194 'client2: what??!?!?' 'client2: what??!?!?' any idea why the two single quoted string have a str_comp of -194? xd

09:35

dbg_msg("chiller", "comp=%d 'client2: what??!?!?' '%s'", str_comp_nocase("client2: what??!?!?", aBuf), aBuf);

probs how far the characters are "away" from each other mathematically

yes

09:37

but arent those strings identical?

09:38

str_endswith(aBuf, "?") returns 0._.

09:39

i feel like there is corrupted memory or shit but then it should print something funny

print both strings as hexa codes then we know

any issue with this? aBuf[str_length(aBuf) - 1] = '\0';

09:40

is there some str_hex in tw?

09:40

there is

looks wrong yeah

09:40

strlen should be without termination null already

it somewhat worked well for me so far to chop off the last char

09:42

until i tocuhed it again

09:42

OMG THERE IS A C2

09:42

who dis bitsh xd

09:43

https://paste.zillyhuhn.com/z7

09:43

hexorman

ez

ou

and guess what

09:43

0xc2

is ¿ maybe 2 bytes?

= 194

09:44

yes

09:44

probably unicode already

omaagwd

09:44

fak

09:44

can str_len count that as 2?

09:44

lets see xd

09:45

indeed

09:45

wow @Jupstar ✪ the hex was sooo good advice

09:45

fixed

char aPuncts[][4] = {"?", "!", ".", ",", "¿", " "};
	for(int i = 0; i < 10; i++) // strip up to 10 alternating punctuations
		for(auto const &Punc : aPuncts)
			while(str_endswith(aBuf, Punc)) // cut off punctuation and spaces
				aBuf[str_length(aBuf) - str_length(Punc)] = '\0';

my goal is todo "what?!?!?" -> "what" and it works. Not sure if thats a rly bad implementation xd

we probs also have str_lenutf8

Robyt3

[ddnet/ddnet] Pull request opened: #5453 Fix segfault in `SDL_JoystickClose` on client quit

Instead of closing the joysticks manually, use SDL_QuitSubSystem(SDL_INIT_JOYSTICK) to quit the entire subsystem, which will also close all joysticks correctly. The engine input destructor is replaced with a Shutdown method so we can control when it is called, i.e. before calling SDL_Quit, which forcefully quits all subsystems. CJoystick::Close is removed as we don't need to close joysticks manually anymore. Closes #5452.

Checklist

[X] Tested the change ingame
[...

What c++ standart ddnet use?

17 currently

@Jupstar ✪ appearently he found it :D

Jupstar ✪

17 currently

starting from what version?

Mʎɹ シ

starting from what version?

dunno, 15 maybe

12:20

smth around there

Ok, thx

nice github down

12:24

thanks microsoft

told ya gh has more downtime than ddnet

13:16

ddnet just has the bigger/better devops team than micronobs

Your dummy then? (<@749222324980416602_=5bquakenet=5d=20=43hiller=44ragon>)

[quakenet] ChillerDragon#0000

its not my mom i swear

C0D3D3V

[ddnet/ddnet] Issue opened: #5454 CRC Erros because of Pickups!?

When I tried to understand our snapshot system a little bit more, I found something interesting on this map (probably related to the findings I reported in #5368): test_2.map.zip The snap function to send the pickups should be correct, so the error is in the snapshot system. If you join the server with that map, then stand on this spot: ![grafik](https://user-images.githubusercontent.com/14315968/174446897-74b7173d-7...

https://github.com/ddnet/ddnet/blob/cd8706a84ab928a37d11b2c9a18ca4e91c5007e5/src/engine/client/client.cpp#L1946 Am I wrong or do I miss something. But I think this check is partially wrong, because CSnapshot::MAX_PARTS is 64 but m_SnapshotParts[Conn] is unsigned and therefor only 32bit? (edited)

ddnet/client.cpp at cd8706a84ab928a37d11b2c9a18ca4e91c5007e5 · ddne...

DDraceNetwork, a cooperative racing mod of Teeworlds - ddnet/client.cpp at cd8706a84ab928a37d11b2c9a18ca4e91c5007e5 · ddnet/ddnet

C0D3D3V

[ddnet/ddnet] Issue closed: #5454 CRC Erros because of Pickups!?

C0D3D3V

[ddnet/ddnet] Issue opened: #5455 Maybe a bug in the snapshot system

If a snapshot is spitted into multiple parts, the client receive them as NETMSG_SNAP identified by there part number. There can be up to 64 parts. And in this line is checked if all parts were received https://github.com/ddnet/ddnet/blob/cd8706a84ab928a37d11b2c9a18ca4e91c5007e5/src/engine/client/client.cpp#L1946 But I think the check is wrong, because m_SnapshotParts[Conn] is of type unsigned and as far as I know unsigned is only 32bit long. I think we should use uint64_t for ...

https://github.com/rui314/mold/releases/tag/v1.3.0

Release mold 1.3.0 · rui314/mold

mold 1.3.0 is a new release of the high-speed linker. This release contains a few new features and general stability/compatibility improvements as shown below. Note for those who create mold binary...

c0d3d3v

https://github.com/ddnet/ddnet/blob/cd8706a84ab928a37d11b2c9a18ca4e91c5007e5/src/engine/client/client.cpp#L1946 Am I wrong or do I miss something. But I think this check is partially wrong, because CSnapshot::MAX_PARTS is 64 but m_SnapshotParts[Conn] is unsigned and therefor only 32bit? (edited)

Only the implementation and god himself knows what size unsigned is. So if we are making assumptions about it that’s bad

God doesn't have programming knowledge @Learath2

If he exists he is supposed to be omnipotent and omnipresent. I’m sure he picked up some programming somewhere along the way

Where's the manual of that language

Bible, Quran or Torah depending on your favorite religion

hmmpf, can't comment on github...

@c0d3d3v btw, what did you mean by left/right and jump keypresses not being sent? I mean, cl_show_direction 1 will show those

did we change smth about skins lately?

18:49

another crash report that Graphics()->LoadPNG failed

trml

@c0d3d3v btw, what did you mean by left/right and jump keypresses not being sent? I mean, cl_show_direction 1 will show those

Clients only get the manipulated data from the server. So if the player is not allowed to press Right, the server will not send right to the clients... So If you are freezed the clients do not know what the player is pressing

we had this before but since LTO is active the crash logs arent really useful

Jupstar ✪

we had this before but since LTO is active the crash logs arent really useful

I see no real use for LTO anyway

oh, during freeze, yes thats true

trml

oh, during freeze, yes thats true

the clients also do know when you press fire...

18:51

but only when you are allowed to press

he uses skins.tw

AttackTick

how? attacktick is sent, but I think its only the last time you pressed

yes

i.e., you dont know if someone presses fire

thats more or less the same for jump. Because you can also not use the jump information to say that the player is really pressing jump. Just a little better then fire

Jupstar ✪

another crash report that Graphics()->LoadPNG failed

isn't that the new readfile function?

hmm, I thought one of the two jump bits indicated jump press

yes the first

18:54

but You are not always allowed to press jump (edited)

18:54

... but it is better then fire

18:55

I would also like to have all real input data in the clients, but on the other hand, that would not really improve gameplay

yeah, checked it again, it doesn't anymore after you used the last jump

Chairn

isn't that the new readfile function?

i assume its pnglite crashing bcs of an abitary skin or some weird file format

trml

yeah, checked it again, it doesn't anymore after you used the last jump

jup

skins.tw is a untested wildcard, and pnglite sucks

18:56

review #5385 and merge it for 16.2

https://github.com/ddnet/ddnet/pull/5385

Add libpng, remove pnglite 2 by Jupeyy · Pull Request #5385 · ddnet...

Same as #3407 Rebasing gets too annoying. heinrichs concern: (he means pnglite instead of libpng in first sentence) "I think this needs a blacklist for formats that libpng can't load, at l...

that flag might still useful enough for cl_show_direction though (if they are shown if freeze at least)

c0d3d3v

I would also like to have all real input data in the clients, but on the other hand, that would not really improve gameplay

would probably be most useful showing in freeze and/or prediction. btw, is sending real input in freeze planned for next release? I guess it would require some alter changes to char/ddnetchar netobjects if not

19:03

additional*

when add 45 degree slides

trml

[ddnet/ddnet] Issue opened: #5456 Send fire keypress to the client

Since the other player inputs are now being sent in some form (i.e. currently effective keypresses, and input in freeze also being planned), it would be nice to also send fire presses, as it could potentially be useful in the future if someone for example wants to show fire keypress in freeze. One option could be as a flag in ddnetcharacter (in that case it might be nice to get it in alongside the new extension of ddnetchar), although it would perhaps have been a better fit in the vanilla ...

trml

would probably be most useful showing in freeze and/or prediction. btw, is sending real input in freeze planned for next release? I guess it would require some alter changes to char/ddnetchar netobjects if not

not planed for 16.2

21:29

I tried to rework the gamecore to be able to send the real inputs in the charactercore without manipulating them. But it turned out to be a bad idea, because to get the physics correct, so that it works as before is hard, if you do not want to have two diffrent gamecores (one for server, one for client)

trml

would probably be most useful showing in freeze and/or prediction. btw, is sending real input in freeze planned for next release? I guess it would require some alter changes to char/ddnetchar netobjects if not

The easiest solution would be to send the real inputs additionally. I had an PR open to send the real Key presses to make the show_direction option show the real presses. But heinrich wanted that I try to refactor gamecore so that we do not need to send them twice (the real and the manipulated)... But I did give up, maybe someone wants to try that in future, or we send them additional.

21:34

https://github.com/ddnet/ddnet/pull/5148

I actually try to do the exact same thing some time ago (and also avoiding the duplicated input)

however didnt quite finish it so never PRed

21:38

I could try to take a look again one day to see if I have any ideas (wanted to check if you were working on it atm first)

21:39

but its perhaps possible that the problems I ran into then are the same as you did now (I think it was getting jump to predict correctly)

21:39

tried*

21:40

as, a year or two ago, so don't remember exactly

21:41

as in*

I first tried to do it for jump, part of that is also in master. I did only a small step and thought this could not have changed anything xD And did have 2 physics changes in it, that I needed to fix afterwards

Its really critical to change something input related ^^and hard to test all physics. We would have to write unit tests before (edited)

absolutely. this is why I mostly only dare to touch client side stuff related to this

21:47

although in the long run it would be great to cleanup server physics code a bit

neither git operations nor the website are any stable and yet githubstatus flexes all operational status https://zillyhuhn.com/cs/.1655591461.png

22:32

so bad

ChillerDragon

[ddnet/ddnet] Pull request opened: #5457 Expose IStorage::GetPath()

the github status page is probably connected to SLAs, so they will only admit something is broken when they have to, otherwise lose lots of money

:brain:

23:47

but at some point they have to admit that their status is broken hrhr