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 2020-07-06 00:00:00Z and 2020-07-07 00:00:00Z

[ddnet/ddnet] Issue opened: #2453 cl_default_zoom is broken

cl_default_zoom is broken/ignored. I kinda noticed it a few weeks ago before smooth zoom, but it was playing around with smooth zoom that made me actually confirm it. (btw smooth zoom is hawt!)

04:30

[ddnet-maps:master] 1 new commit

70e3b3e A Onyx - ddnet-maps

@Learath2 but I want to load a CSaveTee after the map changed

[ddnet-scripts:master] 1 new commit

2322d93 Add teehistorian info to ddnet-setup - def-

@fokkonaut I had some of that done in safe_reload. The idea would be to allocate space for the CSaveTee's std::move or std::copy them over, reconstruct the gamecontext, then std::move them back

06:33

But the harder part is to move the tees into the correct teams

why?

06:36

just saving all information, + team

Problem is that as soon as the gamecontext is constructed, I think the players haven't joined yet

07:20

So you can't use a variable inside Reset and as the gamecontext is being destroyed, you cant store anything in that legally.

07:20

So there is no good place to store this

so you need a variable inside CGameContext instead

@heinrich5991 can't, it's getting destructed

that you set in Reset

Well you could store it in a variable in Reset, then move it

07:33

Then you wait for all the players to join

07:33

Move to their teams as they join

[ddnet/ddnet] Pull request opened: #2454 Add basic XDG support (fixes #2431)

GER2 Novice Perma server dont show up in ddnet tab

https://dev.to/web/design-github-profile-using-readme-md-8al

Design Github profile using README.md

You can Design your Github profile using Username/README.md repo, this amazing update is under deve...

omg thats crazy @Soreu

indeed

is this planned or active?

16:24

https://github.com/ChillerDragon/

ChillerDragon - Overview

Fulltime github influencer artist. ChillerDragon has 67 repositories available. Follow their code on GitHub.

16:24

ah its working

16:24

not sure if i like it tbh

16:24

you can use it to create big image banners lol

I don't see it working yet

it is

16:26

Embedded image

there's one of example repos: https://github.com/katmeister/.github

katmeister/.github

Config files for my GitHub profile. Contribute to katmeister/.github development by creating an account on GitHub.

oh indeed in private browser i do not see it hm

so it's just not tricked by mobile's desktop mode

16:27

Image attachment

16:27

:p

nah seems to be only for the account owner during beta

Also, you created repo named as your nick, while they created one named .github

16:28

so...

ah interesting

guess there are two ways and both are tested

16:30

imo the .github version makes more sense, but for sure there will be people voting for both

i also prefer .github

16:31

sounds more future proof and obvious what it is

16:32

microsoft is going wild feature wise and yeeting on availability

16:32

hope github doesnt end like minecraft ... feature bloated and unusable unstable

16:32

or like skype :troll:

since microsoft github has more downtime (edited)

16:38

greenthing

[ddnet/ddnet] Issue opened: #2455 0.7 client, in /pause there is no Free view listing

in /pause there is no Free view listing (Spectate players still work normally, only every other spectate way is broken) !obrazek

ye in the beginning i was more joking about it but slowly its really noticeable

17:19

idk maybe i might also be spending more time on github nowerdays

imagine not using ddnet client

can't, I'm using it since the very beginning

gamerclient>DDNet client kek

kek

@deen heyo, how the /r works? How I can enable it?

@fearxito everyone in ur team has to write /practice first

Oh

19:22

I get it

ur rank wont count

Yeah, I know thanks, well I don't know that everyone has to write /practice first

I dont think it has to be everyone

19:55

Iirc it was majority voting so you only need 50% + 1

only >50%

Can anyone help me with a 128 player mask? i dont really understand why this doesnt work correctly: https://github.com/fokkonaut/F-DDrace/commit/fa74501da03b4087fed1876d98f054b57294c0f0

19:57

also, i dont know how to do this part: https://github.com/fokkonaut/F-DDrace/commit/fa74501da03b4087fed1876d98f054b57294c0f0#diff-df1a1bb3f0853fbdb44420570856cebfL708

~ negates all bits

20:28

you use Mask128() both for 0 (all bits disabled) and -1 (all bits enabled), at least one of these usages is wrong

oh, so

20:29

ah that makes sense i guess

20:29

and how do i solve the one with the ~

you need to flip all the bits of the whole mask

20:30

you can overload operator~ to do this

doesnt work

why not?

Image attachment

Image attachment

20:30

usually looks like this

~ only takes 1 operand

20:31

|= 2

oh wait, yea, i removed the paramter

I think the error you get would even be decent for this, in C++ standards atleast

what does the error say? and you probalby shouldn't return Mask&

what should i return?

20:33

Image attachment

Mask128

20:33

Mask128& would mean that you return a reference

wait, heinrich, uhm

20:34

so setting the mask 0 means no one can see it and -1LL means everyone can?

20:34

for the long long

yes

20:34

because -1 is represented as 0xffff_ffff

20:34

all bits are set

-1 is only guaranteed to be all ones in C++ past a certain standard when they made the representation two's complement

20:36

(just an inconsequential fun fact, there are no sign and magnitude implementations out there to worry about

)

@heinrich5991 would this be correct?

20:40

Mask128 operator~()
    {
        m_aMask[0] = ~m_aMask[0];
        m_aMask[1] = ~m_aMask[1];
        return *this;
    }

no

why?

if I say int i = 3; ~i; that doesn't change i

20:41

your implementation changes the operand of ~

?

?

20:42

that's not a question

i dont understand

These are not the questions of a programmer, my grandmother reacts like this when I try to teach her how to create contacts on her phone

20:47

What part do you not understand?

20:54

Mask127 operator~() const is the signature I'd use and it'd tell you what's wrong with yours

it wouldn't tell the exact steps to fix it though

Yeah but it might give a hint as to what you meant I guess

isnt the random word generator for /save on an already existing code not good

20:56

its like if you tried to register for a website but it said "sorry, that password is taken"

it's not exactly a trivial problem to solve @louis, Zwelf tried and it broke quite a bit

ah ok

@Learath2 what do i do inside of the ~ function with the two halfs of the mask?

21:03

i got the rest working btw, the mask works

uf, I'm disappointed that you aren't even trying

c++
Mask128 operator~() const {
  Mask128 t;
  t.m_aMask[0] = ~m_aMask[0];
  t.m_aMask[1] = ~m_aMask[1];
  return t;
}

thats how i did it lol

Is it now?...

ye

It's not what you sent here

yess, true

and there is no operator~ in either of your commits

21:13

Turns out I'm not telepathic and can only see as much as you show

i know this sounds stupid, but i had it that way before, and then changed it, and right now i had this on my screen again

point is, inside of this operator, you are not allowed to modify this

21:15

hence the const there to protect you from mistakes

yea

I'd also probably add a constructor that takes two uint64_ts so I can just write return Mask128(~m_aMask[0], ~m_aMask[1]);

true

21:29

thanks alot

btw, I did this before and sadly compilers are not quite smart enough to optimize this well

21:33

they are smart enough to optimize std::bitset well though, except for MSVC

21:33

with msvc you can use the compiler builtin to do a 128bit shift, which optimizes great, so you can do that

21:34

(not that the performance matters that much in this case honestly)

optimize what?

21:39

the ~ should be well-optimized

21:39

is it the << ?

Checking and setting flags doesn't optimize well

21:43

The shifts optimized decently in clang, horribly in gcc and MSVC

I've been wondering why there is no processor support for any 128 bit operation except imul

21:51

does the ALU get too complex?

the question they're asking themselves is probably: is it work the die space?

worth*?

ye

possibly, I'd have guessed simd would benefit from these existing but I'm not very good with vectorization

https://godbolt.org/z/3op3jH

Compiler Explorer - C++ (x86-64 gcc 10.1)

struct uint128 { private: uint128(uint64_t a, uint64_t b) : a(a), b(b) { } uint64_t a; uint64_t b; public: static uint128 zero() { return uint128(0, 0); } static uint128 one() { return uint128(0, 1); } ...

21:55

doesn't look that nice, indeed

@Learath2 Isn't SSE 128bit, then AVX 256bit and AVX512 512bit?

21:55

oh, you mean for a single 128 bit number

Yep

but I also don't know how good it can be

I guess it's not very common when vectorizing to need a 128b + 128b

+ looks fine

21:57

https://godbolt.org/z/Uff_Kt

Compiler Explorer - C++ (x86-64 gcc 10.1)

struct uint128 { private: uint128(uint64_t a, uint64_t b) : a(a), b(b) { } uint64_t a; uint64_t b; public: static uint128 zero() { return uint128(0, 0); } static uint128 one() { return uint128(0, 1); } ...

or just use unsigned __int128 since our compilers probably support it anyway?

__int128 seems to optimize even worse

https://godbolt.org/z/iL8fVm

Compiler Explorer - C++ (x86-64 gcc 10.1)

struct uint128 { private: uint128(uint64_t a, uint64_t b) : a(a), b(b) { } uint64_t a; uint64_t b; public: static uint128 zero() { return uint128(0, 0); } static uint128 one() { return uint128(0, 1); } ...

22:08

dunno what you're talking about, unsigned __int128 optimizes slightly better for check_flag and is the same optimum for add

it's been a while since I checked it, I think it was optimized less consistently across compilers

22:11

(and isn't supported at all in MSVC)

22:14

besides https://godbolt.org/z/BSAXzu std::bitset optimizes better in all compilers, if all you want is checking and setting bits

Compiler Explorer - C++ (x86-64 gcc 10.1)

struct uint128 { private: uint128(uint64_t a, uint64_t b) : a(a), b(b) { } uint64_t a; uint64_t b; public: static uint128 zero() { return uint128(0, 0); } static uint128 one() { return uint128(0, 1); } ...

oops

22:15

hm

(atleast it optimizes to less instructions, I know that doesn't always mean better performance and it'd probably be impossible to predict with the insane pipelines of new cpus)

22:17

s/optimizes/generates/

nah, should be better

22:17

no branches

If you want to do arithmetic though, _int128 does work if it exists

arithmetic optimized fine in my own impl too

+ and - are trivial, / and * are the ones that are hard to get right

Most important DDNet dev change my mind

Image attachment

most important ddnet bug tester*

troll

Most important DDNet dev, change my mind

(edited)

22:24

the comma really helps

@heinrich5991 and funnily enough, clang optimizes both your add and the native add to the exact same instructions, which is shorter than the gcc one

hum

and without using compiler intrinsics, the msvc one just looks abysmal

22:30

https://godbolt.org/z/EgMHWm

22:30

even with the intrinsics the code generated is just not as clean as clang and gcc, but atleast it's readable

it sets up a stack

22:31

ah

22:31

maybe you need that due to SSEH

22:31

SEH*

It also actually calls the constructor

22:32

it should be a trivial constructor to inline, I wonder why it won't

22:35

lol there is no /O3 in msvc that's why

22:37

at /O2 it looks better, still not as cute as clang or gcc

found a bug where the score hud at the bottom right corner doesnt show i downloaded ddnet source with git clone --recursive https://github.com/ddnet/ddnet and ran it with the cmake gui for windows on visual studio 2017

Exported 190 message(s)