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 2021-09-07 00:00:00Z and 2021-09-08 00:00:00Z

done

@Learath2 why not return 0 here?https://github.com/ddnet/ddnet/blob/master/src/engine/shared/snapshot.cpp#L648

[ddnet-web:master] 1 new commit

9d82e93 adcn - def-

fokkonaut

@Learath2 why not return 0 here?https://github.com/ddnet/ddnet/blob/master/src/engine/shared/snapshot.cpp#L648

if you return 0 then snapshots for 0.7 clients are incorrect (you can't see the character for example), but that code looks weird. shouldn't it return pObj->Data() instead of pObj? (edited)

why should they be incorrect?

dunno

they should be correcter

I justed tested

18:07

I can't see the character if I return 0

18:08

but returning pObj->Data() works like pObj (maybe better/worse in some hidden way)

return 0 is wrong there

why?

18:09

it should prevent from sending that object

negative netobj types are ones I reserved for "pass-through"

ah, okay, because on my server it works, i probably did something different than u

Hm actually that's the block above

i just dont want weird memory to be written

Ah, that's to let Obj_SixToSeven drop some netobjs

yeah, then return 0 should be correct

18:11

thats how i handle it too

18:11

CNetXXX = Server()->SnapNewItem()

18:11

if that returns 0, it wont start writing the memory

18:11

if that doesnt return 0, it will

If that returns 0 the snap functions will stop snapping

which for dropped objects, it should not

18:12

no, why?

leading to missing items

why does it stop

Look at anything that snaps more than one obj. It will do SnapNewItem and if the return is null snap itself will return

void *CServer::SnapNewItem(int Type, int ID, int Size)
{
    if(!(Type >= 0 && Type <= 0xffff))
    {
        g_UuidManager.GetUuid(Type);
    }
    dbg_assert(ID >= 0 && ID <= 0xffff, "incorrect id");
    return ID < 0 ? 0 : m_SnapshotBuilder.NewItem(Type, ID, Size);
}

18:12

oh

18:12

you mean it like that

18:13

well, true, but we have to make sure not to write to that memory somehow

18:13

we need to get rid of the returns then, and start using {}

Well it is a bit wasteful, but only a teeny tiny bit. I'll accept a PR if you test it very well

18:14

(There is no UB there btw, it's just writing to memory that won't be sent)

am i incorrect? because i am currently trying to fix a bug on my server that i dont understand at all

18:14

the one i linked earlier

return 0 there would be correct, I think when I made the sixup patch I just wanted to change the least amount of code

client somehow receives faulty characterobjects

faulty?

18:15

faulty how?

yea, they have m_Tick = 4, which is extremely low and CAN NEVER happen at all, leading to running the Evolve extremely slowly

18:15

it frezes the client for several seconds

18:16

and i dont know how on earth this would happen

18:16

i have a demo, which i take for debugging

18:16

because it only happened once for me at all

18:16

others are reporting it more frequently

is it just one field that ends up wrong?

no, the rest seems to be 0

18:16

and somehow, weird positions occur too

18:17

one at 0,0, few more a bit down of 0,0

maybe you corrupt the TypeAndId field of a snap object?

how?

when debugging snaps for sixup I added dbg_msgs on NewItem and UnpackDelta, if you can reproduce the issue you can check the snap sent and the snap received

i cant reproduce it, it just randomly happens

18:19

i only have the demo to reproduce clientside

fokkonaut

how?

Well if you underflow the Data pointer given to you you could have corrupted the typeandid

what is underflow?

e.g. accessing Data()[-1] in that case

mh

18:20

i have no idea tbh xd

FWIW that entire thing is a bit of meh design, C++ doesn't have flexible array members, so we allocate some space, create a new snap obj there, then we assume there is nothing in the memory after the member data

but i dont see where this issue could happen

well it sounds very hard to debug, I wish you good luck

:/

18:23

thx tho :D

fokkonaut

@Learath2 why not return 0 here?https://github.com/ddnet/ddnet/blob/master/src/engine/shared/snapshot.cpp#L648

does returning ->Data() here help maybe?

18:25

(or does that at least not have any other bad side effects?)

Robyt3

but returning pObj->Data() works like pObj (maybe better/worse in some hidden way)

Well at that point I'm just returning a piece of memory the requester can have fun with, the next snapnewitem will overwrite it anyway

ye

so that memory is only temporary?

18:27

I was wondering why m_DataSize was not being increased on the "return pObj" branch

yep, it's just scratchspace so I don't have to modify the classes themselves, the aim was to keep the huge patch atleast somewhat reviewable

@Learath2 could you somehow help me a little? i am getting frustrated already, its been weeks since i am trying to fix this

18:34

i can send you the demo

18:35

and all info i have :D

18:35

i am debugging client with demo

I don't know what else I can tell you, a 0 object with a corrupt field sounds like a corrupted type

are you using CNetServer::SendMsgs? because I kinda think that's broken.

No, i dont even have that in my source, i am 0.7 based

Robyt3

are you using CNetServer::SendMsgs? because I kinda think that's broken.

iirc it isnt, why do you think so?

18:37

(i can invite you to my repo)

at least that's where I ended up when debugging the issue with vanilla 0.6.5 and "sv_vanilla_antispoof 1"

Debugging this with a demo sounds unlikely, the issue seems to be the server sending a broken snap

but maybe that's just due to the weird map being used

fokkonaut

(i can invite you to my repo)

not sure how much I can help, as I don't understand the networking very much

Learath2

Debugging this with a demo sounds unlikely, the issue seems to be the server sending a broken snap

yea

with all the 128p stuff you have maybe you are snapping a player that already left or sth in place of someone that took over the id?

I'd first try to isolate the issue and find out what kind of snap is invalid

Learath2

with all the 128p stuff you have maybe you are snapping a player that already left or sth in place of someone that took over the id?

still, it could never get a m_Tick = 4

18:39

which will then take seconds to get to 400k in a while loop

if you snap garbage memory it could

if a player left, i can not snap him

18:40

character doesnt exists

18:40

snap is not even called

18:40

and it cant take over someone else, because i am explicitly checking while exchaning ids, and making sure there are no doubles

18:40

it just all doesnt match up for me

18:41

let me invite you at least, then you can take a look at these more obivous ones

sure

if pragma- were a ddnet contributor he'd tell you the irc debugger is offline

xD

18:43

https://github.com/fokkonaut/F-DDrace/invitations

18:43

both of you are invited

can anyone help with portable teeuniverse map specialization?

19:07

i want to add int data to my zone type but didn't find anything in teeworlds-ptum

ew closed source

19:19

stallman is angry

im angry too

Exported 109 message(s)