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 2021-07-15 00:00:00Z and 2021-07-16 00:00:00Z

[ddnet-maps:master] 1 new commit

cedaaec A Mountain Silence, M DarkCore 3, M Dead Darkness - ddnet-maps

https://github.com/fstd/pstd

07:34

- The paste IDs we hand out start as 2-letter permutations of [A-Za-z0-0]

0-0?

@TsFreddie: ?

07:53

oh 0-0

07:53

ye go fix

07:54

its a typo of 0-9 as you can see in this sample paste including a 9 https://paste.zillyhuhn.com/69

07:54

hihih 69

tee_thinking

when make pr @TsFreddie ?

how about never

ass

07:59

thenn ill do it

ye. glory awaits

08:04

you did it

, proud of you

ofc

@Learath2 what rust sv r u in?

09:17

i couldnt find u on the official one

09:17

pepeH

@Ryozuki https://discord.gg/rust-lang-community

7. If you come here for the game Rust, you will be kicked without warning.

add that to ddnet rules

7. If you come here for the game Teeworlds, you will be kicked without warning.

11:30

monkalaugh

[ddnet/ddnet] Pull request opened: #3954 Fix the possibility of tees stacking by joining teams

This would make some startline skips harder to exploit.

Checklist

[x] Tested the change ingame
[ ] Provided screenshots if it is a visual change
[ ] Tested in combination with possibly related configuration options
[ ] Written a unit test if it works standalone, system.c especially
[x] Considered possible null pointers and out of bounds array indexing
[ ] Changed no physics that affect existing maps
[ ] Tested the change with [ASan+UBSan or valgrind's memcheck](ht...

@heinrich5991 hm, could moving the tee a bit be acceptable?

I guess the idea is that being blocked from changing the team due to this is annoying

12:01

I agree with that

12:02

I can't immediately see problems with it, but it sounds icky, and it sounds like something that might be exploited on its own

12:02

I could reduce the range where it blocks you from joining a team, e.g. only block exactly stacking tees

Learath2

@heinrich5991 hm, could moving the tee a bit be acceptable?

I solved this by adding 0.2f to the y vel

12:13

minus*

Adding to velocity sounds dangerous somehow, but nudging the position should be relatively safe?

yea

the trait Encode<'_, Sqlite> is not implemented for u64

If only sqlx was half as mature as it pretended to be

15:27

Guess I'll just store everything as strings instead

@Learath2 i think sqlite doesnt have unsigned

15:31

Image attachment

15:31

i guess thats why they dont impl it

15:31

15:32

Image attachment

15:32

see

15:32

its deliberate

15:32

Image attachment

15:32

hmm

15:32

xd

15:32

idk

hm, sqlite3 C interface clearly allows you to get 64b values and store them

15:33

sqlite3_value_int64

thats not uin64 tho

15:33

uint64*

15:34

Image attachment

15:34

sqlite doesnt have unsigned values at all

Hm, you can store an 64 bit unsigned in a 64 bit signed. Sqlite doesn't even support different types at all internally

everything is a i64

The only value types in sqlite internally is 64b signed integer, 64b ieee float, string, blob and null iirc

15:35

Anyway, they can indeed implement a u64 one, I think it's just an oversight

i guess

15:35

open a pr

15:35

or issue

15:36

the good thing of open source

15:36

monkalaugh

I will give it a go after I finish up this mess

https://github.com/launchbadge/sqlx/issues/499

[SQLite] 64bit unsigned integer type support · Issue #499 · launchb...

It seems that sqlx currently has no support for unsigned 64bit integer values (u64). Especially for contents like file sizes, timestamps etc. this would be very favourable as SQLite has native supp...

15:38

@Learath2 isnt it a bit pointless to do everything async and paralel to end up using sqlite, which is not paralel at all from my understanding

15:38

concurrent*

The plan wasn't to use sqlite, that's another limitation of sqlx that sort kicked me in the gut

?

15:39

sqlx supports psql and mysql too

The plan was to use the Any backend and I was going to use sqlite to test locally so I don't have to set up database servers

15:40

However, the Any backend has very very weird interaction with the query! macro, as in it doesn't respect the DATABASE_URI at all

15:41

So I ended up scrapping any hope of keeping this generic and just finished it up in the interest of getting the immediate need for a working bot satisfied

15:42

The new version I'm planning will probably use diesel, which works a little better in the genericity department but suffers from having very meh async support

very meh as in it doesnt have

there are 3rd party crates to help with it atleast

https://github.com/NyxCode/ormx

bringing orm-like features to sqlx. Contribute to NyxCode/ormx development by creating an account on GitHub.

15:44

doesnt seem very maintained but it is there

15:44

xd

15:44

i personally use sqlx and generate stuff with this https://github.com/edg-l/modelizer

edg-l/modelizer

Contribute to edg-l/modelizer development by creating an account on GitHub.

15:44

it is very tailored to my needs tho

I might also say fuckit and just do all of the concurrency by hand

why not, in rust u can do it without fear

15:45

greenthing

15:45

u just have to take care of possible deadlocks

Anyway, I need to finish this first, after we have a working bot I can work on making a clean bot

async is for bitches

15:46

poggers

I'm actually on the fence with async aswell, I'm sure it utilizes the cpu better than I could by hand but it is completely missing some of the context we as programmers have

15:47

I wonder how a task just hopping across workers affects performance, how it plays with cache locality etc..

idk

15:48

for me async backends are a black box

You should read up on how rust does async, it's actually fairly interesting how the runtimes manage their tasks

i should

hi, why after that when such information as the IP address, the list of servers, etc. disappears? It is updated anew as if it has been cleaned

Image attachment

wut is strstr

15:52

so long since i dont touch ddnet code

15:52

Returns a pointer to the first occurrence of str2 in str1, or a null pointer if str2 is not part of str1.

@Ryozuki Q: When you want to store a type without an encode/decode implementation in sqlx, what do you do? Do you implement it for yourself or do you convert to a type it understands?

@Learath2 i never found myself in that situation, is this for a type made by a crate u dont control?

σℓí♡

hi, why after that when such information as the IP address, the list of servers, etc. disappears? It is updated anew as if it has been cleaned

I'm not quite sure how your code relates to your question, I don't really get the question either, can you try to rephrase?

Ryozuki

@Learath2 i never found myself in that situation, is this for a type made by a crate u dont control?

One type I created, one for a crate I don't control

can u show me the type u made

15:54

usually types are made of other types and those simpler types impl the encode decode right? so u should just derive fromrow or smth?

It's an enum of two types from a crate I don't control

hmm

I commented on this part when, and there were no problems, but if I use this, then everything works, but there is a problem. I go to the menu and there is no selected server and all the information is loaded again

The types from the crate I don't control cleanly map to u64 and I don't mind discarding the type information on the enum while storing, I can derive it again when loading

σℓí♡

I commented on this part when, and there were no problems, but if I use this, then everything works, but there is a problem. I go to the menu and there is no selected server and all the information is loaded again

Image attachment

i guess i would impl Decode but maybe its not worth

15:57

then u would have to do select x as "x: MyType" (or _) if u use the macros iirc (edited)

σℓí♡

Click to see attachment 🖼️

Mh, it doesn't really happen for me. If I join a server and check the ingame browser the selected server is still there

15:59

Did you change something in the code that could break this?

σℓí♡

hi, why after that when such information as the IP address, the list of servers, etc. disappears? It is updated anew as if it has been cleaned

I provided a screenshot of the code above, which is why this problem occurs, I change my color or the like, after which this happens

16:01

maybe because I change it directly g_Config.m_ClPlayerColorBody

https://dalek.rs/

dalek cryptography

16:02

omg

16:03

i love whoever made this

16:03

and they use rust

16:03

im a doctor who fan

16:03

monkaS

Image attachment

their ascii thing is messed up

σℓí♡

hi, why after that when such information as the IP address, the list of servers, etc. disappears? It is updated anew as if it has been cleaned

Ah I know what you broke

no its not

64 is way too much, playerskin is only 24 characters. Instead of 64 use sizeof(g_Config.m_ClPlayerSkin)

its some kind of wave

@σℓí♡ what are you making btw?

a bot

16:06

monkalaugh

16:06

idk why anyone not purple asking for code is always a botter

hopefully not, it'd be very sad if I helped a botter

its always ppl who are never active in this discord

16:06

probs using 2nd account maybe

skin stealer it's bot?

Looks like a remote controlled tee with the ability to change colors on messages, didn't look very dangerous to me

haha

16:07

u will look fishy to anyone if u use it

16:07

and they will probs call a mod and ban u

16:08

cuz u will be fishy

Eeeeeh, not really that problematic by itself but mods will be suspicious yes

actual bot clients already have that feature

might async be overkill for a discord bot?

i guess it depends on the scale

16:09

if its used for this sv i guess its overkill

maybe this is a more general question

but maybe the discord library is not async but implements sharding

when does async outperform spawning a couple of threads

idk

Yeah very tough question too, depends a lot on the runtime and how good it is

i think async threads are way cheaper

16:10

than os threads

16:10

tokio claims this

Learath2

64 is way too much, playerskin is only 24 characters. Instead of 64 use sizeof(g_Config.m_ClPlayerSkin)

big ty

@Ryozuki got a link?

Tasks are light weight. Because tasks are scheduled by the Tokio runtime rather than the operating system, creating new tasks or switching between tasks does not require a context switch and has fairly low overhead. Creating, running, and destroying large numbers of tasks is quite cheap, especially compared to OS threads.

16:11

https://docs.rs/tokio/1.8.1/tokio/task/index.html

tokio::task - Rust

Asynchronous green-threads.

hm, doesn't have numbers

16:11

I'd imagine it varies wildly by OS

I guess it's fair to say async can saturate a cpu easier than threads, since with enough tasks it can ensure every thread is working

16:14

Not really quite sure tbf, but async code is a good abstraction over the rather complicated issue of threading an application. Atleast in the sense that it makes it easier for a programmer to be efficiently using resources without having to think too much about it

I see

16:14

so the "only" downside is rust's immature async support?

thats how i feel about it too

heinrich5991

so the "only" downside is rust's immature async support?

Well I think so, I'm sure there are people far more qualified than me to comment on this.

Learath2

I'm actually on the fence with async aswell, I'm sure it utilizes the cpu better than I could by hand but it is completely missing some of the context we as programmers have

As I said ^^. I'm not quite sure if the runtimes are good enough at moving tasks around and inferring relationships between them.

ah, apparently async is also a memory usage optimization

I have a feeling I'm still holding a lot of rust wrong, but it is fun to learn more

nice

e.g. this https://paste.pr0.tips/hH just doesn't look quite right

I still find it weird that you use a RwLock<()> ^^

maybe it's the library interface just not being flexible enough, maybe it's me not really holding it correctly

heinrich5991

I still find it weird that you use a RwLock<()> ^^

Well if you have a minute we can try to discuss it, I really don't see how else to make this safe

I think we discussed it before, but I'm interested in why we didn't come up with a better solution

16:34

ah, was it that sqlx doesn't support transactions?

Well let me give you a concrete example I couldn't reason about and thus think I have to lock the entire table somehow

16:37

Tester A approves map M, Tester B approves map M at the same time. Task T_A launches and gets the row corresponding to the submission from the db. Before Task T_A gets to mark it as evaluated, Task T_B also gets the row. Now we have two threads trying to approve the same map

can you start transactions?

I can, but I'm not completely sure what kind of guarantees that gives me. Afaik sqlite doesn't even have the kind of transaction I require

if I read the docs correctly, you can start a IMMEDIATE transaction, and that gives you exclusive write access

sqlx doesn't even expose this to me

16:41

So I settled for a rwlock that isn't really attached to any type and used the usual "advisory lock" pattern

but you can start a transaction, apparently

I can but I don't see any api for what type of transaction gets started

I don't know which of these transactions it is, but all three are okay for you

When I begin a transaction doesn't that lock the entire db though? If say I do an immediate one

it locks the DB for writing

16:42

not reading

16:43

reads will use the state of the DB before your transaction

And if someone wants to write?

what I'm more worried about is how to detect that you can write again

16:43

that will be blocked until your transaction finishes

The tables here are also independent, locking the entire db is sort of unnecessary, but I could do that for cleaner codes sake

I see

16:44

what does your sqlite library do with SQLITE_BUSY errors?

let me take a look

16:46

Ok so first of all https://github.com/launchbadge/sqlx/issues/481 means I'm right about sqlx not exposing the different kinds of transactions yet

Isolation level support · Issue #481 · launchbadge/sqlx

There are four transaction isolation levels in SQL language and most of database support them. But I can't find how to set isolation level for transactions in sqlx. Is it a missing feature?

16:47

I'll take a look at the source to see what they are doing with the SQLITE_BUSY

I also took a look

16:54

apparently waiting, and then returning SQLITE_BUSY to the user after (by default) 5 seconds

Hm, where did you get that? I was just about to conclude that it just returns an error like any other error from the db

16:58

I think the 5s might be on the sqlite side?

I grepped for busy (case-insensitive)

16:58

it asks sqlite3 to do the waiting

16:58

but it sets its own default value (5s)

16:58

grep for sqlite3_busy_timeout

ah, found it

17:00

Hm, that is quite annoying to leave to the poor user to handle, it's also something I can't wait on effectively

I guess you handle it like other errors?

17:01

what are you gonna do, wait longer? ^^

If I could start an immediate transaction I could busy loop until I get to start one, but there is no guarantee for me to get one, by sheer luck one task could starve forever

17:02

I actually wouldn't mind a task sleeping until the database isn't busy, if there was any wake up guarantee

17:04

Yeah I can't really drop these locks atleast until they expose immediate transactions

17:05

When they have immediate transactions I can get the row, immediately modify it to show the new state and commit so I don't lock too long, and T_B never ends up getting the exact same row as T_A

17:06

This does sound very wrong though, immediate transactions are a sqlite thing. How do people ensure this kind of thing doesn't happen in mysql and psql?

you have other things that write to the db, no?

17:06

then you can probably still get SQLITE_BUSY errors

17:06

also transactions, probably? why do you need an IMMEDIATE transaction, specifically?

A non immediate transaction means I can get a busy within the transaction

17:07

Busy within the transaction means I need wrappers for all my queries that busy loop until they can get their change in

17:08

With immediate transactions I can busy loop just once to get my exclusive access

I see

idk maybe there is a way turn this into a non-racy pattern, but I'm not seeing it

17:09

I basically need an atomic read-modify-write but in sql

does rwlock guarantee that you don't starve btw? probably not, how would it

17:09

but the transcation provides this(?)

17:10

you might just need to retry

tokio rwlocks ensure writers are not starved

ah nice

and they are fair due to using a fifo queue

heinrich5991

you might just need to retry

is this really acceptable? I'd basically need busy waiting, with no guarantee that I'd ever get through

This is in contrast to the Rust standard library’s std::sync::RwLock, where the priority policy is dependent on the operating system’s implementation.

17:13

I guess there should be some kind of "wait for transaction" sqlite3 command

17:14

SQLite is not terribly helpful when it comes to concurrency. By default, it does not even provide any fairness guarantees (although you can write them yourself as long as your concurrency is happening within the same process).

17:14

https://stackoverflow.com/a/25266277

I guess my rwlock qualifies as "write them yourself". An outside process is a bit of an issue but I currently don't have any

you'd need a write lock for the whole db though, not just this table(?)

Hm, I guess it is actually possible that a two threads smack in the middle of a query collide and since sqlite doesn't allow concurrent writes even to discrete tables this would break

17:17

okay I guess my attempt at this isn't exactly safe either

17:18

eh, this is just a field I'm not well versed enough to do this properly

concurrency is hard who would have said

The documentation I read about this really exposes a very uncomfortable interface. With essentially each and every query having to be wrapped in a while(rc == SQLITE_BUSY) and that sounds absolutely insane to me

hm

17:23

I'd think it's the transactions that need to be wrapped in that

17:23

not individual queries

17:23

but still annoying

for an immediate transaction that is true, but for other ones I'm not so sure

there are cases where you need to redo your entire (also non-immediate) transaction

17:25

e.g. if another connection did writes to the DB since you read

um, how is that signalled?

apparently with SQLITE_BUSY_SNAPSHOT if extended error codes are enabled

17:27

otherwise SQLITE_BUSY

17:28

https://activesphere.com/blog/2018/12/24/understanding-sqlite-busy

Elixir Erlang Golang React scale web services Bangalore India

Handle SQLITE_BUSY errors correctly. Understand scenarios under which they occur.

This one actually sounds useful, in my usecase

17:30

One thing this entire adventure revealed is that database systems are much more strict than I initially thought

can you elaborate what you mean by "strict"?

Well in the case of sqlite it supports concurrency in a very specific way. Whether tables depend on eachother or not writing is only allowed for one connection at a time. Whether your transaction cares about other writes during it or not you are getting a SQLITE_BUSY_SNAPSHOT

I see, so strict in the meaning of "not supporting many options, e.g. many concurrency models"

Yes, I guess that's what I had in mind

@ScReeNy what do you mean by "unlock finishing"? (edited)

17:34

that you join team 0 after finishing in team?

yes if you finish in an unlocked team you join team 0 and can stack

and this can be abused?

by this you can stack tees in one tile and do a stack hammer

17:38

idk if this actually gets used, i only did once but it got fixed

17:38

doesnt work on many maps

ok, so that might need fixing, too

17:39

how about allowing people to join team 0 while standing on the finish tile?

17:39

and automatically letting them join team 0 if there's space during finish

17:40

hm, not much different than always allowing them to join team0 after finishing

17:40

after all, the last tee might just wait for everyone to get into position before finishing

InnoDB locking in mysql seems much more flexible allowing table and even row level locks

heinrich5991

after all, the last tee might just wait for everyone to get into position before finishing

yes thats how you do it

Learath2

InnoDB locking in mysql seems much more flexible allowing table and even row level locks

well sqlite is supposed to be rly simple

so only joining people to t0 that are currently unblocked and allowing the rest to join t0 whenever they want would work @ScReeNy?

Ryozuki

well sqlite is supposed to be rly simple

it is indeed, maybe picking one concurrency scheme and sticking to it does make it cleaner. afterall all the memory ordering stuff in atomics doesn't make it very easy to reason about them

heinrich5991

so only joining people to t0 that are currently unblocked and allowing the rest to join t0 whenever they want would work @ScReeNy?

if you count in unlock finish yes. This wont fix everything but at least shouldnt cause issues.

thanks. I'll post this to the issue if that's okay

sure

what do you call an identifier that lets you unambiguously identify which of two objects is newer, like a timestamp, but might be arbitrary data

17:52

version?

17:52

@Learath2?

serial is also common

interesting, haven't heard that before

well common is reaching, but I've seen it

I'm trying to find a name for the "serial" of the info object sent by the game server to the master server

17:54

so that old lingering requests do not cause the masterserver to distribute outdated server info

Ah I think serial is perfect for that, it even implies the number will be increasing

"info_serial"?

the first time I heard serial was in context of dns zones btw, bind9

ah nice

heinrich5991

"info_serial"?

sounds good to me

if it's used by bind9, it's probably the right term ^^

maybe the info_ is redundant? it's already part of the info after all, no?

17:58

hm, actually it is more metadata about the info than part of it

I put it next to the info

17:58

since the masterserver needs to interpret it

yeah, makes sense

17:59

oh, btw could you get a chance to look at the filtering issue @fokkonaut was having?

its basically fine again

ah true, should do that

it fixed itself?

i just need 128p support (for client info/clan/etc)

Learath2

it fixed itself?

Yes, i said that from the beginning. if i send as much client infos as there are clients online, then it works

Oh, that's not fixing itself, you fixed it

afk food. will read afterwards

fokkonaut

I fixed it by sending empty strings and 0's as playerinfo to fill the rest up

.

@heinrich5991 the master server doesn't really care how many players the server says it has now. Maybe it should? I think it's sane to allow incomplete ones, with maybe some ui indication that not all the info was sent

that sounds good

18:01

i guess

the masterserver does indeed not care about the number of players

18:02

it's just that the json (intentionally) has no player count field

no num_players and no num_clients

because that'd duplicate info from the players array

did you ever think about sv_server_info_per_second?

yes, I consider the current setup the backwards compatibility

But it's technically not duplicate info if we allow less playerinfo to be sent

heinrich5991

yes, I consider the current setup the backwards compatibility

sorry, but obviously not

that was just a hack (sv_server_info_per_second)

yea

18:03

but it fucks the info right now

18:03

client sorts it out

it'll get better

I'm even working on it

i dont doubt

what are you doing? the server -> masterserver part?

yes

how do i open a file with fopen? absolute or relative path?

18:04

i cant seem to open it

either is fine

doesnt work lul

18:04

returns 0

18:04

ah no

18:04

im dumb

18:04

wrong var

18:05

lol how can i see stuff like this only when i ask here

18:05

xD

it's the same thing that happens when you call a professional to take a look at something

yea, immediately its clear

18:05

or pops to your eye

I had intermittent issues with electricity that never happened when the electrician is here

xD

I've decided the sanest way to do concurrency in sqlite is to treat SQLITE_BUSY as fatal, let the user retry

18:08

With the delays introduced by the human element the conflicts should be rare enough that handling them just makes the code dirty

https://beets.io/blog/sqlite-nightmare.html seems I wasn't the only one to resort to locking at language level for sqlite

Oh I'm an absolute goof, my locking doesn't even work at all

18:59

Need to hold the lock longer, I guess I'll do a whole database lock and abstract over transactions

at that point you could put the db into the rwlock \o/

I could indeed, even better would be to have 2 connections, one for writing and a pool for reading and a rwlock coordinating so I never have to deal with SQLITE_BUSY at all

what do the cool kids use for multithreading in ddnet these days? i remember using thread_init and also the pain i experienced with it? i want to issue some requests asynchonously and if possible also get the result

jobs

(still painful though)

depends on your definition of pain a lot

19:08

I mean it's not as trivial as futures maybe, but it's not much more painful than threading in just C/C++

thanks, i will read into it

https://blog.crunchydata.com/blog/generating-json-directly-from-postgres

Cut Out the Middle Tier: Generating JSON Directly from Postgres

PostgreSQL has built-in JSON generators that can be used to create structured JSON output right in the database, upping performance and radically simplifying web tiers.

20:42

:o

Anyone does know java well? (edited)

ask your question

https://dontasktoask.com/

Don't ask to ask, just ask

20:52

this website fits so perfectly

20:52

and its always java

20:52

Image attachment

20:52

monkalaugh

Yea but

no yeas

Bit operations

20:54

I have 12 tasks

20:54

And 4 of them made my brain...

idk the others but im not here to do ur homework xd

Just wanna ask for some advices

then ask

How can i make binary number higher bit to 0, with simple bit operations?

20:56

Only simple (edited)

| 0x80000000

see, that doesnt even require java

Anyway

do you want to set the highest bit to 1?

I need to set the highest 1 to 0

20:57

With any number i scan

ah

20:57

you want to set the highest bit that is 1 to 0

20:59

?

Yes, how to set the lower i know, how to set the 2nd, 3rd etc ik too, but if number changes its higher 1 from one rank to another

sorry, I don't understand what you say

21:00

can you give an example?

Ehm

21:00

So

21:00

I mean, i need a code which inverts higher 1 to 0

21:00

For example

21:01

11000 i need to get 01000, and if number will be 10110000, so i need to convert a higher 1 as well, to 00110000

I see

21:02

so we can split up the problem into "find the index of the highest 1" and "set that 1 to 0"

Yea

21:02

But

21:02

Only with simple bit operations

how can you test whether the n-th bit is set?

21:02

say you have the number x

21:03

can you write java code that checks if the n-th bit is set?

If i know n i can use >>n

21:03

And using if, its 1 or 0

this expression doesn't contain x, it can't be the solution

21:04

static void checkNthBit(int x, int n) {
    [...]
}

21:04

fill the function so it outputs "1" if the n-th bit of x is set and "0" otherwise

void

monkaS

@Ryozuki I wanted it on stdout

ah

With void check yea, but what about using only ~, |, & or ^ and << >>

Ryozuki

void

monkaS

#define void int

use those

21:06

he asking u to do it

21:06

heinrich the good teacher

[ddnet/ddnet-web] Pull request opened: #137 Frontpage video

https://www.youtube.com/watch?v=dujUeC8tE8I

But i hadnt used english for a while, so im a bad explainer :(

whats ur main language

Rus

uh

21:07

sorry

java xd

xd

21:07

we dont have any rus dev that i know here

21:08

monkaS

Sad af

21:08

banzai is ukraine

21:09

BannZay*

[ddnet/ddnet-web] Pull request closed: #137 Frontpage video

21:09

[ddnet-web:master] 1 new commit

23df5a0 Frontpage video - murpii

how am I going to do the port forward check in the https masters?

21:27

I want to send a challenge that the game server has to include in its next register attempt

faceless можешь мне рассказать, я переведу

21:46

@Deleted User

Да уже вроде вспомнил как на англ разговариватт

ping me if u need help

heinrich5991

how am I going to do the port forward check in the https masters?

Hm, can we use the old protocol? or do you need something new?

need something new, the old protocol doesn't have challenge-response

@heinrich5991 okay, hm, this is a tad annoying to add but maybe just a large-ish packet similar to the 0.7 CONNECT?

probably add some bytes that can be set by the game server so that the game server can see that it comes from a legitimate masterserver?

why do you wnat to make it large? @Learath2

I wanted to discourage people from spamming it

it can include "authentication" in the form of bytes only known to the game server and the masterserver

and it'd lead to a worse than 1 reflection ratio

it wouldn't reflect at all if it doesn't include the "authentication"

@heinrich5991 it should definitely include authentication, my initial implementation included a secret sent with the info to make sure

22:27

You know what would be cute in rust? C++14s operator""s

can you convince me of that? ^^

String::from("Test string") vs "Test string".to_owned() vs "Test string".to_str() vs "Test string"s

22:29

One is much much easier to type than the others

22:30

Especially when you need a lot of them to initialize say a default value for a struct with many Strings in it

hm, not really convinced

22:32

sounds like bloat

22:32

can you do .into()?

I mean it's not getting shorter than ""s, into does work but it still looks meh imo

22:36

Why do you think it is bloat? When there is a prettier way to do things it usually gets in. E.g. the unstable try blocks

22:36

You can get their behaviour using a closure

try blocks are unstable since forever

22:38

also, there's something you can't do from a closure, returning from the outer function

btw rust already has literals, I don't really see why extending them to some heap types hurts

hm. dedicated syntax for .to_owned() for just strings is a bit underwhelming for a new language feature

22:39

also, rust tries to make less efficient stuff more verbose, String is less efficient than &'static str

22:40

(I realize you could theoretically use operator"" for other stuff, but I don't know any)

heinrich5991

hm. dedicated syntax for .to_owned() for just strings is a bit underwhelming for a new language feature

C++ thought so too, so they introduced user defined literals along with it. Maybe in that form it's more palatable to you

also it kinda does away with namespacing

I'm out of arguments, it looks pretty to me so that's enough for me

ok, about the masterserver forward check

22:45

it doesn't really have to be large if it includes some bytes chosen by the game server, right?

Yep

how do I best format the challenge/response in JSON?

22:54

should it be a zero-terminated string in the packet and a normal JSON string?

22:54

or a byte array in the packet and a array of numbers < 256 in JSON?

byte array in the packet and a base 64 encoded string in the json maybe?

22:56

If you don't want to bother encoding your first suggestion would be what I do

what about a hex string vs base64? ^^

22:57

hex is already implemented in teeworlds ^^

Sure that works too, b64 is just shorter is why I had it in mind

I mean, time to write a b64 encoder/decoder I guess

a byte saved here and there doesn't really matter anyway for this, we don't do fwchecks all the time

actually, we do (at least currently)

22:57

ah, but I'll drop it, since the game server can continue to provide the proof

hm, do we? I thought we only did simple heartbeats after the first fwcheck stuff

heinrich5991

ah, but I'll drop it, since the game server can continue to provide the proof

how will the gameserver prove it is connectable if not continuous fwchecks actually?

I'm assuming it won't go away immediately

22:59

by replaying the secret I sent in the port forward challenge from the masterserver

So wait the protocol is the server sends info with secret, master sends secret + challenge, the server sends response to some other endpoint on the master?

the server sends info with secret + challenge @Learath2

23:16

(at the end)

Ah, okay. But you still need to keep challenges to make sure the server is still alive, no?

23:17

keep sending*

hm, every few minutes might be enough, no?

Yep, sounds good

the server sends its heartbeats more often

23:19

(I think the port forward check isn't relevant to security, it just helps people figure out that their servers are unreachable. in the end, the game server could just respond to port forward checks and nothing else, and it would be the same result for the user as elaborate "port forward faking")

Yeah, sounds sane to me

23:28

https://paste.pr0.tips/P2 does this not look very cute?

looks good

23:29

maybe only retry on BUSY?

I was considering that but there are 3 issues with that: 1) I want to retry on LOCKED too, 2) sqlx doesn't really expose it in a way I can match easily, maybe I can hard code a numeric code there, 3) it wouldn't be cute

^^

Maybe when I'm making a cleaner version I'll directly use the database library instead of an abstraction

hi, i added a field to the NETSOCK struct, and for some reason it gets mem_zero'd some times... any ideas why? cant find any explicit mem_zero calls

no idea

23:48

you could try setting a memory breakpoint in gdb

i was trying to dodge that, but it seems like i have no choice

thanks

doesn't seem too bad to me?

never done that with gcc...

23:50

learning something new every day

watch <address>, I think

it's actually extremely simple and a very useful tool to learn to use

how do i get hold of the address though?

break somewhere where you have access to it

23:51

probably in CNetClient initialization or so

You can do &sock.mynewfield gdb will figure out the address of the field for you

(if sock is in scope)

what was the cmake debug flag again?

23:53

sorry xd

-DDEV=ON

23:54

(for teeworlds/ddnet. in general it's -DCMAKE_BUILD_TYPE=Debug, I think)

-DCMAKE_BUILD_TYPE=Debug is what I use, what does -DDEV=ON do?

it's ddnet specific. saner defaults for developing IIRC

23:55

the default configuration is for packaging

@heinrich5991 I have another request for rust, compound literals

is that "abc" "def"?

No C99 ones (struct foo){"foo", 3}

hm, rust has these, no?

23:59

you need to name the fields though

consider using a let binding to create a longer lived value

This is the issue I'd like to work around

Exported 522 message(s)