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 2017-09-04 00:00:00Z and 2017-09-05 00:00:00Z

<deen> eeeee: interesting. very small pointer for me in firefox and feels laggy, I see tearing as well

07:55

<deen> @nub thanks, should be fixed. forgot to update the servers as well

<Learath2> eeeee: can't seem to connect to servers

<deen> the new info.ddnet.tw seems to work fine: avg time per request: 8 ms, 95th percentile: 70 ms

08:58

<deen> (and 0 errors so far)

hey guys I need a small project done in c++(not in teeworlds) and would pay for it. someone interested ?

If it Is an easy Project hehe

10:10

I can Build you an helloworld

I think, here are some gods of cpp

No joke i would help you. But i Need some Details If i can handle The task.

[ddnet:] 1 new commit

58d1294 oops - def-

<deen> And 47 people are STILL downloading https://hookrace.net/time.gif . I guess some bots don't have a timeout setting

<heinrich5991> deen: maybe their hdd will be full eventually ^^

11:49

<heinrich5991> wait

11:49

<heinrich5991> it doesn't stop downloading after 24 hours? :)

11:50

<heinrich5991> I thought it would ^^

<deen> nope, never stops

12:04

<deen> it would be wrong to stop after 24 hours since there are leap seconds sometimes

12:04

<deen> well, at 300 B/s it takes a while to fill the disk

12:05

<heinrich5991> :)

12:05

<deen> Only 9 GB per year

12:05

<heinrich5991> hm, that's slower than hdds will be added to a machine

12:06

<heinrich5991> so it can run forever \o/

12:06

<heinrich5991> I was away the last couple of days

12:06

<deen> welcome back

12:06

<heinrich5991> it seems you fixed the mac semaphore hting?

12:06

<deen> with Learath2

12:06

<heinrich5991> so we could close the PR?

12:07

<deen> but I'm not happy with the way semaphores are implemented in glibc

12:07

<deen> what pr?

12:07

<heinrich5991> oh, firefox loaded an old page from cache

12:07

<heinrich5991> nvm

12:07

<deen> so i wanted to use condition variables, but their semantics are weird

12:07

<heinrich5991> ah

12:08

<heinrich5991> what do you mean, semaphores in glibc?

12:08

<deen> the implementation of sem_wait in glibc source code

12:08

<heinrich5991> oh

12:08

<deen> too much busy waiting, syscalls all the time

12:08

<heinrich5991> spin lock for a bit and then wait using the os?

12:09

<deen> keeps syscalling futex() even when nothing happens

12:09

<heinrich5991> due to a timeout?

12:09

<deen> no timeouts either

12:09

<deen> try running strace DDNet and see it yourself

12:09

<heinrich5991> how can it call futex if there's no timeout?

12:10

<heinrich5991> futex should wait until it gets signalled, no?

12:10

<heinrich5991> k, I'll try

12:13

<deen> (the network can also be improved in the client I think, similar to server)

12:13

<deen> not to recv all the time

12:14

<deen> especially if you select over all the sockets we have open

12:14

<heinrich5991> hm

12:14

<deen> instead of blocking like in server

12:14

<heinrich5991> not much point there

12:14

<deen> probably

12:14

<heinrich5991> you have to push out a frame every 16ms anyway

12:14

<deen> font rendering is a lower hanging fruit

12:14

<heinrich5991> yes

12:14

<deen> and much slower

12:15

<deen> for the server it was a great improvement not to busy wait for network packets though

12:15

<deen> especially when you have 30 servers on 1 core, mostly idling

12:15

<heinrich5991> yep

12:15

<heinrich5991> most of them even empty

12:15

<heinrich5991> literally having nothing to do ^^

12:15

<deen> so strace ./DDNet-Server is much more reasonable

12:15

<deen> 3 syscalls per second

12:20

<deen> no one complained about performance problems on Windows, so I assume the semaphores work there as well

12:21

<deen> I switched some more code over to semaphores

12:21

<deen> btw, can we make CFetcher use CJobs instead of reimplementing the same thing?

12:21

<deen> Or is there a vital difference?

12:25

<deen> the threaded logger is basically the same as well

12:25

<heinrich5991> not that I'm aware of

12:26

<heinrich5991> Learath2 said that you will murder him if he implements CFetcher atop of CJobs though

12:26

<deen> Only if he breaks the updater during that process :)

12:26

<heinrich5991> ah

12:26

<deen> I also removed the versionsrv entirely

12:26

<heinrich5991> well, doesn't matter

12:26

<deen> Using https://info.ddnet.tw/info?name=heinrich5991 instead

12:26

<heinrich5991> don't we run one, though?

12:27

<heinrich5991> ah

12:27

<heinrich5991> +1

12:27

<deen> yeah, we still run it for old versions

12:27

<deen> most notably, servers is sorted by how often you finished at a location

12:28

<deen> oh, and I had to use /info because otherwise we will run into problems with letsencrypt

12:29

<heinrich5991> why?

12:29

<deen> if I forward everything, then certbot can't put the file into the directory

12:29

<heinrich5991> yes

12:29

<heinrich5991> it needs .well_known

12:29

<deen> so we have no proof that the domain is still under our control

12:29

<heinrich5991> location /.well-known/acme-challenge/ {

12:29

<heinrich5991> alias /var/www/challenges/;

12:29

<heinrich5991> try_files $uri =404;

12:29

<heinrich5991> }

12:30

<deen> oh well, too late :D

what does "acme" means ?

12:33

i saw it a few times in different situations

12:33

it has special meanings?

12:33

or it just popular author?

<heinrich5991> acme is the protocol that letsencrypt speaks

12:35

<heinrich5991> to give you ssl certificates

12:35

<deen> https://en.wikipedia.org/wiki/Acme

Acme may refer to:

12:36

<heinrich5991> it's not actually in there

12:36

<deen> I only knew https://en.wikipedia.org/wiki/Acme_(text_editor)

Acme (text editor)

Acme is a text editor and graphical shell from the Plan 9 from Bell Labs operating system, designed and implemented by Rob Pike. It can use the Sam command language. The design of the interface was influenced by Oberon. It is different from other...

12:36

<heinrich5991> :o

12:36

<heinrich5991> https://en.wikipedia.org/wiki/Let%27s_Encrypt

Let's Encrypt is a certificate authority that launched on April 12, 2016 that provides free X.509 certificates for Transport Layer Security (TLS) encryption via an automated process designed to eliminate the current complex process of manual crea...

12:36

<heinrich5991> here's a reference to ACME

12:37

<heinrich5991> ah

12:37

<heinrich5991> it actually exists on the page, but I couldn't find it, despite me knowing it

12:37

<heinrich5991> that's a bad sign :D

12:37

<heinrich5991> https://en.wikipedia.org/wiki/Automated_Certificate_Management_Environment

Automated Certificate Management Environment

The Automatic Certificate Management Environment (ACME) protocol is a communications protocol for automating interactions between certificate authorities and their users' web servers, allowing the automated deployment of public key infrastructure...

12:41

<heinrich5991> deen: I think you should use POST instead of GET parameters

12:41

<heinrich5991> GET parameters show up in server logs

12:42

<deen> is that a problem?

so acme is over used naming !@

<heinrich5991> well, you now have a log of each ddnet player's IP address and nick name in your webserver log

12:43

<deen> I have the same already

12:44

<deen> from DDNet-Server

12:44

<heinrich5991> when they join the server

12:44

<deen> Yeah

12:44

<deen> So we should only log the name if there is a problem?

12:45

<heinrich5991> would you expect that your nick name lands on some server when you click a tab in a server browser

12:45

<deen> I find it useful to see if some name is especially slow for example

12:45

<heinrich5991> that would be better, I guess

12:45

<Learath2> could implement CFetcher on CJob but i'm scared it'll break that one remote ass possiblity that i didn't think about

12:45

<Learath2> no idea why i didn't make it a cjob at the beginning tho

12:45

<deen> heinrich5991: when that tab shows you your finished maps, yes

12:46

<heinrich5991> I don't know, I dislike it

12:47

<deen> The request or logging it?

12:47

<heinrich5991> first it was just logging it, but after thinking again, the request

12:48

<Learath2> class CFetchJob { CJob m_Job; allthestufffromCFetchTask; }; pass CFetchjob as userdata to the job. use CFetchjob as we used CFetchtask in the past

12:48

<deen> So we can add a setting "Use username to request finished maps and sort servers"

12:48

<Learath2> sounds sane ^^?

12:48

<deen> and if it's turned off, then the info is requested without name

12:49

<Learath2> how can you request that data without a name?

12:49

<heinrich5991> there's other stuff in there

12:49

<heinrich5991> more than just your records

12:49

<deen> yeah, still request it

12:49

<heinrich5991> it apparently replaces the versionsrv

12:50

<deen> but you just get empty records back

12:50

<Learath2> also replaces the servers json we had?

12:50

<deen> yes

12:51

<deen> and the news from versionsrv as well

12:51

<Learath2> well we could have a setting for "Show finished maps"

12:51

<Learath2> if it's not checked we don't need to request by name afterall

12:52

<deen> exactly what I just proposed

12:52

<Learath2> heinrich5991: ^^ reasonable?

12:52

<heinrich5991> if I added a box with information about this to the ddnet tab, would you consider making this opt-in?

12:52

<heinrich5991> so people will see the feature and won't request it unknowingly?

12:52

<deen> Well, people won't find that info

12:53

<heinrich5991> to the ddnet tab, near the place where the records would be showed otherwise

12:53

<Learath2> don't really have enough space in the serverbrowser to write an essay either, you'd only fit "Show finished maps"

12:53

<deen> even if you make it opt-out, they would click it by accident

12:53

<heinrich5991> see: vlc asks on first launch whether it can access metadata from a remote server for stuff you're currently playing

12:53

<heinrich5991> firefox also asks on first launch whether it can transmit data

12:54

<deen> https://info.ddnet.tw/info

12:54

<deen> there you go, no more name necessary

12:54

<deen> Ideally make it opt-in, but ask people at the first start

12:55

<Learath2> make it opt-in noone uses it, make it opt-out it's a privacy concern (a minor one imho)

12:55

<heinrich5991> Learath2: not if we don't hide it in the options somewhere

12:56

<Learath2> well i don't see anywhere to put it in the serverbrowser window so it needs to be a on startup thing like the nickname setting etc

12:56

<heinrich5991> yes, that would be bad

12:56

<heinrich5991> first launch questions are annoying

12:56

<heinrich5991> those that you can't skip

12:56

<Learath2> so POPUP_FIRST_LAUNCH or idk where to put :/

12:58

<deen> at least you can add it in the dialog at the start where you enter your name

12:58

<deen> it fits well there

12:59

<heinrich5991> true, it can be put into the same dialog

13:00

<Learath2> deen: do you want me to make CFetchTask->CFetchJob

13:00

<deen> but make an even bigger warning that their ip is broadcasted to every server when they open internet server browser

13:00

<deen> Learath2: only if it will have 0 bugs

13:00

<deen> becuase people still get ddosed from that

13:00

<heinrich5991> and if they deny that, don't show the server browser ^^

13:00

<heinrich5991> but actually, we could do that

13:01

<heinrich5991> fetch the servers ourselves and put them into your info.json

heartw

1

13:01

<Learath2> deen: you can't promise that, i can't promise that, i'm pretty sure even god would think twice before promising that

13:02

<Learath2> deen: eeeee has the teewebs serverlist available as json, could switch to using that instead of masters, calculating ping is not an exact science in that case but we'd get to upgrade to http masters :P

13:03

<heinrich5991> nah, we still need all servers

13:03

<Learath2> heinrich5991: i think eeeee has all the servers

13:03

<heinrich5991> what use would that be?

13:03

<heinrich5991> they don't support websockets

13:03

<heinrich5991> or webrtc

13:03

<heinrich5991> or whatever the current flavor of the day is

13:03

<Learath2> ask him wtf

13:04

<heinrich5991> or he proxies it

13:04

<Learath2> only supports connecting to ddnet servers but iirc he does keep a json list of all servers

13:06

<Learath2> deen: serverinfo amplification is still possible?

13:08

<heinrich5991> yes, modulo rate limiting

why

14:05

on monster

14:05

there are

14:05

2 teamrnaks of NANAMAR

14:05

and Saro

14:05

Showing

<deen> someone might have played with moving ranks

14:09

<deen> or deleting ranks

14:09

<deen> or a bug

14:09

<deen> or the database link was broken for some time

14:10

<deen> apparently I didn't manage to fix the libfreetype on mac: https://forum.ddnet.tw/viewtopic.php?t=5654

14:10

<deen> it links to heinrich5991 directory, haha

14:11

<deen> heinrich5991: want to look into the mac library problem?

so you will fix it?

<deen> ok

<heinrich5991> I can try

<deen> thanks

[ddnet-web:master] 1 new commit

8234d5a Update settings&commands - def-

15:24

[ddnet-web:master] 1 new commit

7283278 Fix settingscommands style on mobile (Thanks Cr... - def-

15:25

[ddnet-scripts:master] 1 new commit

7808a5e Fix settings&commands - def-

<deen> heinrich5991: should remove the user agent as well as it is leaking information about the users?

15:36

<heinrich5991> mh no

15:37

<deen> Actually I didn't introduce an option for the ranks json thing because I noticed that we have way too many settings

15:37

<deen> and people set stuff by accident

15:37

<deen> or even another client changes ddnet settings and then we run into problems

15:37

<deen> like baumalein client allegedly changing the update server to their own

15:37

<deen> in settings_ddnet.cfg

15:42

<heinrich5991> deen: too many options is a valid concern

15:42

<heinrich5991> not for this thing, but: do people also set options accidently that are not exposed through the UI?

15:43

<deen> rarely

15:43

<deen> that's also why I removed some stuff from settings

15:44

<deen> I also thought the next step might be making info.json a log in ;)

15:44

<heinrich5991> oh so you want an account system? ^^

15:44

<deen> not sure

15:44

<deen> has advantages and disadvantages

<eeeee> deen: Learath2: wrt teewebs can you provide more details like browser and os version, gpu model

17:30

<eeeee> also what is "laggy", frame drops or network drops?

17:31

<deen> current firefox, arch linux x86-64, intel skylake gpu

17:32

<deen> 3840x2160 resolution, but 1920x1080 window

eeeee: spectator mode works weird on beta.teewebs.net

17:32

its always moving, if you move cursor in whatever direction

17:32

<deen> frame drops

anyway, does /kill is disabled

17:33

on spawn?

<deen> chromium works much better

17:34

<deen> oh, I have multiple processes enabled in firefox

17:34

<deen> text loads quite slowly

anyway, cursor shouldn't be disabled ?

eeeee: vivaldi, gentoo, intel integrated gfx

17:50

on second thought i should have used a more common browser to test

<deen> like qutebrowser

17:51

<heinrich5991> why not elinks?

17:52

<deen> you can only run DDNet-Server in javascript in elinks

17:52

<deen> client is a bit problematic

17:53

<deen> I'm not sure if it was elinks, but I tried editing one of links/lynx/elinks once, and the source code was full of polish comments only

17:53

<deen> or czech

<eeeee> ok yeah i did notice that high resolution is laggy on my broadwell macbook, maybe i'll dig into that

18:57

<eeeee> will fix the text loading once i settle on a way to get rid of freetype

19:01

<eeeee> Savander: specactor mode stops moving when you center your cursor. it's hard to make it work the same way as in desktop client because browsers don't let you easily switch between absolute and relative mouse positioning

19:02

<eeeee> Learath2: can you update your browser to the latest version and then copypaste the contents of developer tools console after it fails to connect to a server?

19:04

<eeeee> and what's qutebrowser, is that like konqueror?

What about disabling mouse ?

19:06

in previous teewebs you did it

<eeeee> yeah but if you compare the mouse pointer between the old and new teewebs you can notice new one has less latency

but its pretty hard to control i think

19:07

cause, i can click random icon

19:07

no web or close the window

Exported 254 message(s)