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 2020-04-30 00:00:00Z and 2020-05-01 00:00:00Z

I would like to get my Map Intercepter rerated from 1* to 2* Brutal. Its very long and definitely does not fit a 1* map. This was already suggested right after release but didnt happen back then, i think because it didnt work somehow, so far i remember. Thanks

change OldschoolAdventure from 1* solo to 2* solo before u do anything else lmao

[ddnet-maps:master] 1 new commit

b55d8eb M Wave - ddnet-maps

I wanted to make a patch for ghidra, the integration tests have been running for the last 14 hour

07:17

s

07:18

apparently I need more cores then a macbook has

@deen you work on a large project which is probably mission critical, are your tests this extensive? There are more then 3500 integration tests here, it still didn't end

i think they add tests for everything they add

08:02

i saw it on a blog post

08:02

https://hookrace.net/blog/hana-cpp-development/

08:03

yep

08:03

@Learath2

08:04

"Hundreds of developers from all around the world are developing on about 10 million lines of C++ code and 15 million lines of Python tests."

I wonder what kind of beastly computers they run these kinds of testsets on

https://www.sap.com/dmc/exp/2014-09-02-hana-hardware/enEN/appliances.html#categories=certified&order=MemoryDesc

I doubt >16h is acceptable for tests

why wonder, when it is in the blog post

I guess I need a cluster

@Learath2 you wouldn't believe the amount of testing we do. we have a first stage of unit tests for our team (many thousands, takes a few seconds), a second stage of integration tests (takes 1-2 hours), a barrier to get into the main branches (takes 12-24 hours distributed among many systems with dozens of cores each) and then conti tests that only run weekly with the really extensive performance tests

09:21

the perf tests for the barrier run in parallel on a pool of quad socket 22 core machines (176 threads total) with > 350 GB of RAM. For larger tests we also have bigger machines with > 1 TB RAM (edited)

09:23

but yes, test times are a problem. removing tests is also bad, so you have to improve the tests to run faster and still test the relevant stuff

09:24

and yeah, customers have even crazier systems than we have internally, see @Ryozuki's list

so much ram

09:25

Image attachment

09:25

24 Teras of ram

it's always horrible when a customer has a performance issue on a large system, hard to reproduce those numa effects

numa?

non-uniform memory access

09:26

some memory is closer to one cpu than to the other, so it matters where you put your data

oh god

@Ryozuki There are larger systems as well, but maybe not off-the-shelf. At least PayPal already had a single 48 TB machine on HANA 4 years ago

justatest

@Learath2 why not just make a test push and let their infrastructure test it?

09:36

I never ran our full integration tests on my machine, always just the ones I think are relevant, and the rest can be done centrally

[ddnet/ddnet] Pull request closed: #2086 Remove setting ammo for freeze to 0

10:44

[ddnet:master] 5 new commits

d22a1a8 Remove setting ammo for freeze to 0 - fokkonaut 69a6e94 Fix two bugs - fokkonaut da72907 Fix ninja not getting removed - fokkonaut d460470 Move comment aswell - fokkonaut 5e213a2 Merge pull request #2086 from fokkonaut/master - def-

[ddnet/ddnet] Issue closed: #2153 add possibility to lock chat?

10:56

[ddnet:master] 2 new commits

76eda61 cl_showchat 2 to always show large chat area (fixes #2153) - def- d0cf8c0 Merge #2154 - bors[bot]

10:56

[ddnet/ddnet] Pull request closed: #2154 cl_showchat 2 to always show large chat area (fixes #2153)

[ddnet/ddnet] Issue closed: #2151 Can't whisper to players with quotes at beginning and end of name (eg. "nameless")

11:04

[ddnet/ddnet] Pull request closed: #2152 Fix whispering to players with quotes (fixes #2151)

11:04

[ddnet:master] 2 new commits

65a4eb7 Fix whispering to players with quotes (fixes #2151) - def- bd3b72e Merge #2152 - bors[bot]

They don't run their tests on a public CI, not even sure they run the tests until before a release

11:05

I tried all the relevant tests on my own computer and created a PR, guess they'll let me know

11:09

@deen when you have a minute you should rebuild the dmg file on ddnet.tw, it's still built with curl that has ssl disabled

[ddnet-web:master] 1 new commit

9755e47 Add donation - def-

@Learath2 I guess I should also rebuild the libs first?

Well at the very least you'll have to rebuild curl

the output of configure should be telling you whether it is building with ssl

[ddnet/ddnet] Pull request closed: #2138 Some osx cleanup

13:04

[ddnet:master] 2 new commits

17d1bcc Some osx cleanup - Learath2 09fad44 Strip rpath before adding the correct one - Learath2

13:06

[ddnet/ddnet] Pull request opened: #2155 Don't require notify on Linux

Especially when only building server and not client

[ddnet/ddnet] Pull request closed: #2155 Don't require notify on Linux

13:28

[ddnet:master] 2 new commits

7fe8afe Don't require notify on Linux - def- e5442c4 Merge #2155 - bors[bot]

[ddnet-libs:master] 1 new commit

bde7315 Rebuild mac libs (10.9 osx version, curl with tls) - def-

16:31

[ddnet:master] 1 new commit

8c5de9b Update ddnet-libs (rebuild mac libs) - def-

16:35

[ddnet:master] 1 new commit

d1b7e84 Version 13.1 - def-

[ddnet/ddnet] Issue opened: #2156 Run eval `osxcross-conf` before compilation

I'm wondering if I'm doing something wrong, can't build 13.1 on Mac OS X: ``


+ cmake -DCMAKE_BUILD_TYPE=Release -DPREFER_BUNDLED_LIBS=ON -DCMAKE_TOOLCHAIN_FILE=cmake/toolchains/darwin.toolchain -DCMAKE_OSX_SYSROOT=/home/deen/git/osxcross/target/SDK/MacOSX10.13.sdk/ ../ddnet-master
CMake Error at cmake/toolchains/darwin.toolchain:2 (message):
  Run eval

osxcross-conf` before compilation Call Stack (most recent call first): /usr/share/cmake-3.17/Modules/CMakeDetermineSystem.cmake:93...

@Learath2

17:08

demo can save in fields>

@deen I added the osxcross_conf requirement, we used to hard code darwin15 which is wrong, it depends on with which version of the sdk you build osxcross with

ok, still doesn't work though

17:50

linker error now

17:50

@Learath2

17:50

So I should use darwin17 for my lib builds?

nope, setting darwin17 still fails the same way. any suggestions?

@deen

18:26

demo can save or no?

?

18:26

you mean in file based server? no

in field

what's a field?

in folder

18:27

without other programms

save as a video or as a demo file?

18:27

check out config_directory.bat

18:27

there is a demo directory

?

18:28

@deen

18:28

how see this

where you have the DDNet.exe there is also a config_directory.bat

ok

18:29

@deen

18:29

i cant

18:30

its in pause

18:31

i cant

18:31

wiew it

18:31

@deen

@Learath2 with this it builds, but not sure if it actually works: https://github.com/ddnet/ddnet/commit/e8bd8459a6f556594f48f33f4d145033bc89d46f

18:36

You can try https://ddnet.tw/downloads/DDNet-13.1-osx.dmg

Ah, @deen you need to build compiler rt in osxcross too

18:45

There is a script included in osxcross to build compiler-rt

oh

18:46

and then rebuild libs and everything?

Yes, that should resolve the undefined symbol without relying on weak linking

18:47

I think only libcurl uses the os version check

[ddnet-scripts:master] 1 new commit

180d2d9 Update build scripts - def-

@deen you could also use osxcross_conf so you dont have to hardcode dawrin17

it passes CFLAGS?

19:02

to configure calls?

Also isnt darwin17 10.15? I think we still cant update that far because of the sdl peoblem (it should be fine in sdl 2.0.13)

i have 10.13, so no idea

Ah 17 is 10.13, sorry the 5 different versionings get confusing

19:05

Re: osxcross_conf thing. I think its fine like this, nvm my earlier remark

opinion on whether I should just force-push ddnet-libs and ddnet?

19:06

don't want to add more useless megabytes of mac libs

I don't mind force pushing ddnet-libs

19:07

Do you need to force push ddnet too?

yes

Doesnt ddnet just track ddnetlibs through a commit id?

yes, and you have to make a commit to change that

Ye, but you can just push a new commit, but I guess that'll make the previous commit unbuildable

exactly

19:08

so I wouldn't do that

Eh, I think it's close enough to the previous commit

19:09

It shouldnt be too bad

19:09

Would you like me to test it before you push?

still the same failure with compiler-rt

19:11

ah no

19:11

this one linking error is gone

19:11

just 5 others left!

Can you show me, I probably already came accross them, I built this in osxcross

https://github.com/ddnet/ddnet/issues/2156

19:13

oh well, the notifications ones I can fix myself

19:13

just the ssl ones are interesting

@deen you could have a tag for the old tip of ddnet-libs

oh, I hadn't built it with notifications

19:14

can you show me the configure output from curl?

59.12 KB

hm, I'm lost there we are building for 10.9 builtin_available should be dropping that call to _SSLCopyALPNProtocols

maybe the 10.9 build is not communicated properly from osxcross-conf

19:19

I'll try entering it manually in CMakeLists.txt

wait

19:20

is CMAKE_OSX_DEPLOYMENT_TARGET set correctly?

never seen that

19:20

where to set it?

well if you aren't setting it it's being set to 10.9

19:21

so that's not it

[ddnet:master] 1 new commit

c9dfcb2 Fix Mac OS X notifications linking - def-

okay building libcurl

sorry you have to do it again :/

19:33

I have no idea why this mac building is so hard

well you are not the one that has to be sorry, I'm the one that keeps breaking it

According to https://github.com/curl/curl/issues/2835 we need -mmacosx-version-min=10.13

19:36

or 12

19:37

or linking against Security.framework: https://github.com/microsoft/vcpkg/issues/7260

19:38

seems like we are already

19:41

Or use SDK 10.13.4: https://codereview.qt-project.org/c/qt/qtbase/+/241268

that call to _SSLCopyALPNProtocols is protected by __builtin_available, it shouldn't even appear in the final curl static library :/

19:44

when compiling, if the deployment target is lower then the builtin available, clang should just optimize that out

19:44

It was doing it just fine a couple weeks ago when you "fixed" that libcurl issue by changing the deployment target

then I should set the deployment target explicitly?

yeah maybe try setting it explicitly

with the cmake thing?

maybe cmake is messing up

19:45

the cmake thing should be setting to 10.9 just fine

19:45

maybe just pass --mminblabla like we used to?

19:45

but this is something wrong with the compilation of curl then ddnet I think

i tried that already, didn't work

19:47

I think this has the right explanation: https://github.com/curl/curl/issues/2835 Probably we use 10.13 SDK lower than 10.13.4

19:48

afk

hm, apparently clang doesn't optimize that out, even on macOS

20:14

oh, I'm an idiot, that's the whole point of @available, it's to be able to target old macOS but still use newer functions if available

20:15

but still, it shouldn't be trying to hard-link it

I think you are right that we need a newer sdk, that availability being wrong in the SecureTransport.h is messing with clangs mind

ok

20:24

can you provide one for me?

I can but I need to download the xcode image that'll take a while with the shitty internet here

no hurry

5gb wow

20:26

I'll go take a shower while that downloads

@deen I think c9dfcb2b should have instead renamed notifications.mm to notifications.m as it no longer uses any Objective C++ features, but either is fine

21:13

(as Objective C++ is a superset of Objective C)

I don't understand

21:14

It was .m before, that didn't link

21:14

I don't know details about Objective C/C++

Ah, we don't have it extern c so calling it from C++ has mangled names

21:17

I think this is fine in that case

right, I was always thinking about putting extern c

21:17

but we had some mm files already, so meh

21:17

could also rename system.c to system.cpp while we're at it

Well why mess with it now? We can do that rename when we actually want to use a C++ feature in system.c no?

because C++ has more reasonable semantics and Microsoft only supports C89 which is even worse

21:19

like declaring all variables at the start of scope

21:20

or the thing where in C a function "void x()" accepts any parameter

I keep forgetting msvc only does c89

[ddnet/ddnet] Pull request opened: #2157 Switch src/base from C89 to C++11

Disadvantage is of course losing Vanilla Teeworlds compatibility, so I don't care much either way. Haven't cleaned up much since I don't want to put work into it if it's not getting merged anyway.

One thing I'll miss from C will be the implicit cast of void *

yeah, I noticed

oh, I'm also close to getting my commit into ghidra, I'll have officially helped the NSA conquer the world by fixing a small UI oddity

nice

@Learath2 link?

https://github.com/NationalSecurityAgency/ghidra/pull/1824

Fix missing progressbar in TaskDialog. Close #1819 by Learath2 · Pu...

Not completely sure if this is fine, but calling initialize here instead of setMaximum gives the task dialog a chance to call installProgressMonitor even if the delegate was changed too late and Ta...

22:10

.tbd is a very interesting idea, instead of including the actual object to link to in the sdks, they only include the uuid of the library/framework, the install-name and the symbols it exports

22:11

the linker is then able to link things together as install-name is enough to link a mach binary

22:11

(there is no static linking in mach unless you really force it)

@Learath2 how many hours did this one-liner take? ^^

@Learath2 now the NSA knows about you justatest

justatest

22:57

justatest

@heinrich5991 around 5 hours to figure out what was going on, 3 to come up with a solution that's not shit

23:16

well also includes time to learn the ghidra codebase a bit

Exported 202 message(s)