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 2021-05-18 00:00:00Z and 2021-05-19 00:00:00Z

Astramast

Bcd new ddnet steam comers dont know that they exist

i know that they exist

pro

Iza

i know that they exist

no u dont

ok, I know why my macos is compiling external/pnglite/VERSION now

09:10

i'm using a case-insensitive apfs, macossdk tries to include <version> then it just found external/pnglite/version

adding a BEFORE in target_include_directories when using bundled libs fixed it.

09:22

not sure if it is right way tho

Hey Devs, i bought a raspberry pi 3b yesterday and wanted to install a nas-sys and run tw beside. Do you think that the raspi 3b is good enough to run both task at the same time ?

09:47

i wanna run surely the native ver of ddnet client

just test it out, it should run, but maybe not good

Don't like it much, I'm not a big fan of pvp mods and think they are not so fun long-term. Also a tab makes no sense, since pvp is not a project

11:03

like DDNet or KoG

add community, put noby fng and kog there

11:05

but everything thats hosted by ddnet can stay in ddnet

Internet tab is plenty enough tbh

not untrue, official server are highlighted anyway

11:21

with https master servers especially

but I should change my repo name I guess

Image attachment

11:22

kek

lets create PVPNet.tw

why this shit tell me player.h does not exist when he exist :c

Image attachment

11:38

i tried on local and that work, i have this shit only in my vps (edited)

player entity. interesting

did u clone with submodules

TsFreddie

player entity. interesting

yeah i tried to move it but that doesn't work

Do random things until it works is my favourite approach to debugging aswell

dont buy ur compiler on wish

did you or why?

no bcs he cant compile with his vps XD

12:30

the joke behind it is, that things from wish are always low quality xD

when i was moving in US, i hired some helpers, and one of them asked me if i can get them something on wish.

13:53

didn't know what wish was at the time tho.

@Deleted User is there a player.h in src/game/server/entities/?

15:01

by default there's only one in src/game/server/

ye i know

15:02

I just did shit because I hadn't slept

someone can explain this shit? same code but on local and vps not the same result

smells like undefined behaviour

compiled with signed char?

15:35

on arm char isnt signed

15:36

-fsigned-char

has anyone run ddnet server on arm yet?

15:37

might find some bugs with it

i ran it on android

15:38

but worked fine and the toolchain also uses signed char there

15:38

ah but only client

15:38

thought he uses client anyway

[ddnet:master] 2 new commits

304d9f3 Fix UB with super (fixes #3830) - def- c6c0c1d Merge #3832 - bors[bot]

15:38

[ddnet/ddnet] Issue closed: #3830 UB when killing while super

15:38

[ddnet/ddnet] Pull request closed: #3832 Fix UB with super (fixes #3830)

but we still have alignment issues anyway, but i think they are mostly deadly for floats, and they were only in vote options

15:43

so might not be too critical

15:43

atleast on arm32 its deadly

Deleted User

but we still have alignment issues anyway, but i think they are mostly deadly for floats, and they were only in vote options

do we really have alignment issues left? thought we cleaned up ubsan

Learath2

do we really have alignment issues left? thought we cleaned up ubsan

in server i think so

15:44

client should be fine

15:44

there is still issues with destructors tho

15:45

soon in 20 years when tsfreddie rewrote all prediction code it might be solved xd

15:47

SUMMARY: UndefinedBehaviorSanitizer: undefined-behavior ../src/game/server/gamecontext.cpp:2785:12 in 
../src/game/server/gamecontext.cpp:2785:12: runtime error: load of misaligned address 0x63100001488f for type 'CVoteOptionServer *', which requires 8 byte alignment
0x63100001488f: note: pointer points here

@Learath2

I bet that's the weird memheap implementation we use to keep these allocations in a block

15:52

I'll add alignment support to it tonight

15:52

Hopefully, idk, I woke up way too late today

alternatively just align it to 8 bytes

15:52

instead of passing alignment everywhere

15:52

to pointer size

heinrich5991

alternatively just align it to 8 bytes

Could do that too, but why waste space?

15:53

I guess byte lifes don't matter to you young folk

I feel like this is applying the optimization at the wrong place

just remove these heaps xD

15:55

just use normal allocations

Deleted User

just remove these heaps xD

Tbf with modern arena allocators this would work fairly well

15:56

Actually no, it might end up in different bins. So we should probably put it in a vector instead

15:56

We iterate thru it a lot

heinrich5991

I feel like this is applying the optimization at the wrong place

Point was that not every type requires max alignment

yes. malloc also always returns maximum alignment

Because malloc needs to stay generic and support decades of C cruft.

15:59

Modern allocators with an std::allocator interface correctly take and utilize alignment

use whatever fits

16:00

still better than own implementations that always make trouble xD

@Learath2 I guess you could make it a template argument, then it's not harder to pass

16:00

ah no, the usage doesn't line up with that

That is what I was planning, but I didnt look at the code yet. Putting it in a vector might be much easier

the usage is for structs that are larger than they actually are

16:01

i.e. ones that have dynamic data at the end, usually strings

16:01

I think consumers expect the pointers to remain stable

We have flexible array members??

yes

16:02

https://github.com/ddnet/ddnet/blob/c6c0c1dc2d9b0c4c6264959b5af55da238b8727e/src/engine/shared/network_conn.cpp#L137

I love how matricks coded these

16:03

we've only messed his nice memory layout out since we all started messing around with it

16:03

s/out/up

16:04

anyway, I'll do it however you prefer I do it, I honestly don't have the energy to argue anything

16:05

If you prefer I get rid of the flexible array member and shove it in a vector I'll do that

16:05

If you prefer I align everything to max alignment I'm also fine with that

do it how you like it, I don't htink there's much new usage of these classes

[ddnet/ddnet-discordbot] Pull request closed: #17 Check length of reason

16:49

[ddnet-discordbot:master] 1 new commit

f48cba5 Check length of reason (#17) - Learath2

@heinrich5991 sadly on windows dbg_curl doesnt seem to show the verbosity it does on linux

16:56

person checked dbg_curl was 1

[ddnet-web:master] 1 new commit

48e24b7 Tournament #57 - def-

@Learath2 i found something interesitng

Image attachment

21:14

https://stackoverflow.com/questions/34548762/c-isnt-that-hard-void-f

C isn't that hard: void ( *( *f[] ) () ) ()

I just saw a picture today and think I'd appreciate explanations. So here is the picture: I found this confusing and wondered if such codes are ever practical. I googled the picture and found an...

21:15

its actually a thing http://c-faq.com/decl/spiral.anderson.html

this is not flexible array member

class CNetChunkResend
{
public:
    int m_Flags;
    int m_DataSize;
    unsigned char *m_pData;

    int m_Sequence;
    int64 m_LastSendTime;
    int64 m_FirstSendTime;
};

21:16

this is flexible array member, but it's not supported in C++

class CNetChunkResend
{
public:
    int m_Flags;
    int m_DataSize;

    int m_Sequence;
    int64 m_LastSendTime;
    int64 m_FirstSendTime;

    unsigned char m_data[];
};

21:17

u can just malloc any type and run new over it tho so unsigned char* m_data;

21:19

any size*

21:19

and for simple structs ofc without constructing

yes, but that fake flexible array member wastes 4/8 bytes on that unnecessary pointer

well u can just use uint8_t

21:20

i mean in the end its just binary work

single byte pointer?

you just want an address of the starting pointer

21:21

rest doesnt really matter as you define how tall your struct is

21:22

so &m_data

what about

class CNetChunkResend
{
public:
    int m_Flags;
    int m_DataSize;

    int m_Sequence;
    int64 m_LastSendTime;
    int64 m_FirstSendTime;

    unsigned char m_data[1];
};

21:22

emulating flexible array member in C++

yeah

ah so you meant this

pointers of arrays are still the beginning of pointers i c++

21:22

so yeah

21:23

doesnt really matter how it looks like, u just need the start address

yes

*beginning of array

With heinrich5991's upcoming HTTP master server (https://github.com/ddnet/ddnet/pull/3772, https://master1.ddnet.tw/ddnet/15/servers.json) we now also archive the player statistics: https://ddnet.tw/stats/master/ Maybe someone has a good use for that data already.

Ryozuki

its actually a thing http://c-faq.com/decl/spiral.anderson.html

this is not the best way to read C declarations btw, there is only one rule and it's "declaration follows use"

i.e. try to imagine that the declaration tells you how to use it

22:51

(*abc)[10] pointer to array

22:52

*abc[10] == *(abc[10]) array of pointers

22:57

if I asked you what you'd prefer to write the HTTP masterserver API for servers in, you'd probably say go out of python, go, rust? @Learath2

22:57

go has the disadvantage (for me) that I don't know it well

@heinrich5991 I don't know go well enough to suggest it for anything tbf

23:30

I know this would be a decent usecase for it tho

23:31

I think rust is our best bet here

Exported 135 message(s)