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 2019-05-24 00:00:00Z and 2019-05-25 00:00:00Z

[ddnet-maps:master] 1 new commit

e383b4d M Momentum, M run_2Rockets - ddnet-maps

[ddnet-web:master] 1 new commit

852f905 Update link - def-

@heinrich5991 here?

kinda

08:43

fun, irc showed the notification before discord did

quick question, do you know what AuthInit is in teehistorian files? I don't see any of them, I thought they were supposed to be in all

AuthInit is only done when the map changes and someone was still logged in

08:44

it's probably quite rare

doesn't seem to be what i thought

08:45

do we save keyslots and their names somewhere?

08:45

also do we record the keyslots on rcon commands?

you're probably looking for AuthLogin

08:45

that's the login events

08:46

AuthInitial is also interesting if you want to correlate rcon commands with key slots

Meh, so I have to keep state to figure this out

yes

08:47

you should generally expect to need to keep state if you want to make sense of teehistorian files

08:47

they're append-only and try to save the least data possible

what does authlogin have in it btw?

08:48

cid, level and authname?

08:48

yep seems to be it

08:49

let's see if I know enough rust to code this

you can also write it in whatever you want by using the output of teehistorian_dump

that sounds horribly inefficient

yes

I'd rather hack teehistorian_dump at that point

08:50

if you get a tool of sensible code quality, you could PR it to the libtw2 repo

well I'll PR it anyways, you can help me get it up to quality later

such a weird language...

it's not just another language with no new concepts ^^

they should make a nice to use version of c++ instead of all these new languages that are so confusing

I think rust is the safe to use version of c++, and probably a good attemtp at being nice at the same time

I don't think I like all these Somes and expects

those are actually a good thing to learn IMO

09:06

Option types generally make your code more robust against error cases IMO

sure, but I don't think that's something that needed improving

I think it is something

09:06

you see how people are complaining about hitting NPE in java in their code? that's the solution to that

09:07

(null pointer exception)

09:07

because it moves the error to the location where the wrong assumption happens

I mean you shouldn't dereference a null pointer

not the location where the null pointer is used

it's not rocket science to add a null check, we've been doing it for quite some time now

yes. but if you have a wrong assumption, it can blow up in a completely unrelated part of the program

09:08

not so for option types

It's a fairly easy thing to figure out where the null comes from if you have coded for more then a week

09:08

a simple backtrace gives you where it comes from

not in my experience

09:09

you don't know where the null initially came from

09:09

you see where the null pointer was dereferenced

09:09

but not why the variable was null to begin with

at that point, you do a bt

then you see a backtrace of the place where the null pointer was dereferenced

if you were calling the function with a null it's obv

e.g. you see it came from a member variable

if you were dereferencing a null that came from this function, it's quite easy to see where it is becoming null

09:10

it's obviously somewhere you are modifying it and don't have a null check

now you don't know why. did the function I call earlier return the null errornously? was it becasue I passed null to a function errnously? etc.

09:10

this doesn't happen with option types

In my experience I've never had a hard to debug null pointer issue tbh

09:11

it's one of those problems that are extremely obvious

09:11

for me atleast

I feel like you're misunderstanding what I'm saying. of course it's easy to find where the null pointer was dereferenced

09:12

but it's not so easy to see how the null pointer propagated through the program to that point, if it's not in the call stack anymore because it went through several variables or functions no longer in the call stack

I guess I've never had an issue in finding where a null originated from

anyway. even from a theoretical point of view. if you were designing a new system, you'd probably not say: oh, let's add a sentinel value to every pointer type

09:13

even C and java have nonnull annotations because of that

09:14

(or popular C compilers do)

but it's not so easy to see how the null pointer propagated through the program to that point, if it's not in the call stack anymore because it went through several variables or functions no longer in the call stack

rr can help you with that btw.

I've never felt the need to use such an annotation, I guess it depends very much on how you think while coding

@deen rr sounds cool

yup

09:24

unfortunately, I never looked into it so far :/

At work I played around with UndoDB recently, works similarly, but unfortunately not free/open source

I never thought there'd be a closed source debugger

09:50

sounds like a very hard to sell product, as developers most likely won't be used to it at all

And now you know why we had an UndoDB workshop

09:52

rr can't deal with shared memory well and we use quite a bit of that, so it's the only choice

09:53

and it's really easy to use, just record and then you can step backwards in a regular gdb interface

@deen is a time-travel debugger that useful that you'd use a debugger you aren't familiar with?

09:54

oh is it also gdb-ish?

undodb is just a thin layer on top of gdb

that makes more sense

so all gdb commands work plus everything with an r prefix to d o it in reverse

rdelete production database

is time-travel debugging something you use often?

nope

09:55

recording is super slow, only one thread at a time

that'd be a very hard product to pitch to management

and requires huge amounts of memory to store

09:56

The thing is we have a few cases of really hard to reproduce problems

09:56

so you can run them in a loop for days on end while recording

09:56

and when you finally have them you can use the recording to figure out what happened previously

09:56

or a customer has a problem that only they can reproduce on their machines, but we can't on ours

09:57

because of the data set or whatever, so they can enable the live recorder, reproduce it, and send us the recording

but if it's very slow and takes a lot of memory is it feasible for a customer to run it?

but then you have the customer's data. that sounds like something the customer might actively want to avoid

only some

09:58

the customers often have productive systems and test systems, so they could run it on a test system

09:58

@heinrich5991 sure, they have to be aware of that. they also give us exports for reproducing problems

09:59

and we have to be careful with customer data of course

oh you can embed the recorder into your product aswell?

I think of reverse debugging as a last-resort solution for the problems hardest to debug, not something you use daily

10:01

yes

10:01

The alternative for us would be to add more tracing, release a new version, tell customers to use that and then send us the traces. Takes some time

yep, actually sounds like a decent last resort to debugging hard to reproduce problems

[ddnet-web:master] 1 new commit

2911bf9 Remove Greeen and just god, add Testers - 12pm

@heinrich5991 does the teehistorian header include max players?

I wonder if they actually tried hard to make the syntax unfamiliar to people

13:25

it's like I'm trying to decode moon runes

xd

like just try to have a guess at what pub identity: &'a [u8], could mean

wow

I think it's a reference to an array of 8 bit unsigned integers

13:31

lifetimes are such a foreign concept

okay how does this even make sense 1302 ConsoleCommand { cid: 0, flag_mask: 128, cmd: b"pause", args: [] }

14:29

pause should have the flag_mask 132 not 128......................

what's flag_mask?

the commands flags

14:30

when you register it in console

14:31

It's CFGFLAG_CHAT|CFGFLAG_SERVER for pause

14:31

which is 132, but the CFGFLAG_SERVER seems to be gone in the teehistorian file

oh

okay, the flagmask is set to only CFGFLAG_CHAT while executing chat commands

14:34

that makes more sense

ye xd

@Learath2 is it possible to create a background job to only do one action?

yes

How then? xd

take a look at CJob

k thx

@heinrich5991 on a scale of very horrible to horrible, how horrible is this? https://paste.pr0.tips/FO

man its beautiful

It's currently parsing all of the teehistorian files I have

18:29

it should finish in a decade or so

18:30

exporting this all to json then parsing the json to render graphs don't seem like the best idea performance-wise, but idk what else to export it as

msgpack, protobuf?

18:41

but I had no problems working with json files around 1 GB, there are performant libs around

Hey @deen @heinrich5991, how much time

hi Davide

18:41

time?

How much time i don't join here

18:42

I mean that

no idea

It was not a question, but an exclamation

18:43

Don't care, anyway there are one way to give strong to all @deen ?

18:44

Example via map config

@deen will see how performant js itself is with that much data

no

Davide: nope

Wtf

18:48

So, how kog do that?

they changed the source code

18:49

we had that change in ddnet too at some point, but the physics change makes some game play worse, for example hammerfly

Ok so there isn't way with config, only with code

I love how when we patched it everyone was angry and when kog patches it it's a revolution

so it was reverted quickly

18:49

kog doesn't have hammerfly and other tee interactions like in ddrace, right?

what dd the patch?

18:49

they*

@deen well you can still hammerfly with another tee

Learath2: yes, but it was worse than before

not really part of maps though

18:50

@deen I meant in kog

and now it would also invalidate all previous ranks

@Deen can use hammerfly but there aren't map require that, i think

Davide: you can ask qshar to make the source code public

I think he don't do it

https://github.com/ddnet/ddnet/commit/2b084645276baa37b7a31aafa3fab0e3a323201b

Same hook strength for everyone (by cinaera) · ddnet/ddnet@2b08464

Thanks @deen

We only had it for 10 hours :D

Why don't make a map-config for doing that? So you can choose what map can have it or not

18:52

So the mappers can use or not that config

because then players have to get used to two different kinds of physics

18:53

instead we have /spec

That isn't a problem if you're good

18:53

I know @deen

@deen little problem, the code you've linked me is very old, and the current code is very different

Oh man how strange is that

19:32

Image attachment

19:34

sometimes the name is wrong like that

19:34

its so weird

19:35

and sometimes

19:35

Image attachment

i'm lost xd

you can see how you got there in vs

tell me

20:17

please

call stack

ah i need to install something

omg who did that

20:34

Image attachment

@deen Solved

20:51

Thanks

@Learath2 i improved a bit the performance of the replay thing by putting the slicing operation in a separate thread, more than that i can't do xd

Exported 196 message(s)