Teeworlds - general

that'd make sense

well I guess teeworlds performs decent enough that we don't need to optimize this

08:21

If we did, we could first split the map into chunks, that'd be simpler to implement then a quadtree

@heinrich5991 remember the guy with nazi symbols? apparently he is rly bored

1

1

@Learath2 you could split the map into chunks with a quadtree basically doing the same

@Assa an actual quadtree is more work to implement then splitting the entire map into 32x32 chunks

11:44

a quadtree is denser at some regions

you will never need to go under 2x2 and parts of the map can be ignored like walls

:0 50x50 map

You don't get the point, it's not that a quadtree wouldn't be better, it's that it'd be more difficult to implement so no need to resort to it unless we need to

There are no zero-cost abstractions

@Dune hm, wouldn't divvying the map up into chunks (either a quadtree or just fixed size chunks) be zero cost?

13:14

Worst case scenario with a horribly unbalanced tree it'd perform the same as now

I know, that there is overhead, you know characters and projectiles move through the tree/chunks? It's just a possible way, that might lead to improvements in the overall performance - I don't say anyone should implement this

@Learath2 the cost is not always in performance

we had examples of zero-runtime-cost abstractions

13:58

e.g. std::unique_ptr, std::vector

can you explain why the unique_ptr is zero cost?

well what is the cost?

it's a zero-cost abstraction over managing the pointer yourself with new/delete or malloc/free

13:59

the compiler inserts the calls for you, you don't have to care about it

every time you use the pointer you need to derefence two pointers, that's the cost

that's not true

oh, if the compiler solves it for you then you are right

no you don't, the compiler can optimize that out

a unique_ptr has the same amount of dereferences as a normal pointer

There is however a memory cost to unique_ptrs with custom deleters

@Learath2 depends on how you use it

I learned to increase performance by sacrificing memory - ofc a tree would take some mem

huh, can you get a custom deleter without memory cost?

yes, template arguments IIRC

ah, no it's EBO

EBO?

You can pass a "function like object" as the deleter, so you just pass a struct with only overloaded ()

yes

an instance of it is constructed but because it's empty it won't use any storage at all

https://www.youtube.com/watch?v=rHIkrotSwcc&t=1058 uses the example of unique_ptr

EBO = empty base optimisation btw

can you summarize the arguments? I don't like videos

but in general, I argue that abstractions have a cost and must be sparingly used

I think you overgeneralize - there are abstractions (and dataformats), that improve the overall performance

https://godbolt.org/z/EMTeoQ https://godbolt.org/z/1MCTIa

Compiler Explorer - C++ (x86-64 clang (trunk))

void bar(int* ptr); // Takes ownership void baz(int* ptr); void foo(int* ptr) { if (*ptr > 42) { bar(ptr); *ptr = 42; } baz(ptr); }

Compiler Explorer - C++ (x86-64 clang (trunk))

void bar(int* ptr); // Takes ownership. void baz(std::unique_ptr ptr); void foo(std::unique_ptr ptr) { if (*ptr > 42) { bar(ptr.get()); *ptr = 42; } baz(std::move(ptr)); }

14:09

@Assa you cannot improve performance with abstractions, anything you do on an abstract level, you can do on a lower level

that is incorrect

14:09

or at least misleading

14:09

there are things that you cannot reasonably do in C

(obviously it's more tedious

)

yes

14:09

but if that means that it won't be done

but you can't say that abstracting improves performance

14:09

of course, that's why we use abstractions

there are abstractions that improve performance in reality

but we should keep in mind that they all come with a cost - sometimes less than the benefits, of course

the same goes for coding low-level

14:10

that also comes with a cost

14:10

that might be performacne (because you don't use a hashtable with a proper hashing function, because you can't be assed to write it)

14:10

like Teeworlds, for example

I'm only arguing against abstracting for the sake of abstracting

Teeworlds does have the problem of too little abstraction in a few places

14:11

e.g. rolling their own, very bad, trivially DoSable hash function

14:11

that wouldn't have happened in a higher-level language

Teeworlds pushes the ideology quite towards the "litte abstractions" side, and suffers from it on some aspects, yes

interesting talk, didn't think about the implications of exceptions and the temporaries constructed

(note that the example that @Dune posted is a bit nonrealistic, I think)

It's a difficult slider, maybe Teeworlds does it too little, maybe other applications do it too much

you don't usually pass unique_ptr around

14:13

@Dune now that's just relativizing everything

14:13

that's an empty hprase you can say about everything

remember to program in reliability - you might lose some performance, but every second you don't have to spend at some code anymore is time you can spend on other maybe more usefull parts - so if you look at the overall efficiency of the development progress instead of the performance of teeworlds itself

@heinrich5991 well the guy says it's a pretty common/idiomatic pattern to pass ownership of memory

@heinrich5991 true

14:14

I don't really want to argue the list of benefits it has with Teeworlds and the list of inconvenients

https://godbolt.org/z/apjjM_ @Learath2 I don't see the performance penalty

Compiler Explorer - C++ (x86-64 clang (trunk))

void bar(int* ptr); // Takes ownership. void baz(std::unique_ptr&& ptr); void foo(std::unique_ptr&& ptr) { if (*ptr > 42) { bar(ptr.get()); *ptr = 42; } baz(std::move(ptr)); }

14:14

I see one extra instruction, haven't investigated it furtherly

14:14

(dunno why it wouldn't use && for transferring ownership)

it does use rvalue references though, no?

the original one? no

I'm not done with the talk yet btw, he is talking about those two instructions on the white lines

ah

the two white instructions there are memory loads, which is why he says it's worse then the move between registers in the raw pointer one (that one is only to match calling convention)

hm, c++11 masterrace

rust might be even better here, because it has proper move semantics (might fare better here than && references)

there was this thread about zero-cost abstraction in rust https://users.rust-lang.org/t/not-quite-zero-cost-abstraction/11514

Not quite zero-cost abstraction

(co-authored by @bugaevc and @YaLTeR) Zero-cost abstraction is one of the major selling points of Rust, along with memory safety and others. In fact, it is listed first in the Rust features list on the rust-lang.org frontpage. Zero-cost abstraction means abstraction without ...

from a quick glance, this is something different than the move semantics of unique_ptrs

it seems like vitalyd argues that abstractions are only zero cost if the compiler is smart enough

I don't see how the abstraction can never be zero-cost though, as long as you have a way to communicate your intent perfectly to the compiler it should be possible to get unique ptr to zero cost

I don't know rust well enough though

see how it got better with the c++11 example as you conveyed your intent better and better to the compiler?

As long as you do something that is not too smart for the compiler, I guess

@WingedTeam does that have any relation to teeworlds?

14:25

then please don't post it here without explanation

xd

@heinrich5991 I think ownership is indeed what's messing up the zero-costness of this abstraction

14:36

If you pass by value, you need to check whether the called function did indeed take ownership, you pass by rvalue and there is an indirection

14:39

Teeworlds would indeed benefit from a couple more abstractions though, like a proper hashmap

ease of development is worth some abstraction:0

obviously, otherwise we'd be writing assembly

хд

some more than now!

14:41

assembly masterrace

btw @Dune I think his argument is that you can't have "total" zero-cost abstractions, as in it'd either have a cost while building, at runtime or a human cost (e.g. readability or complexity)

Optimize when there is a need to optimize. Good ol' Donald

@Learath2 I think you're right, though in particular unique_ptr seems to have some more issues because of shortcomings(?) of the ABI

14:50

I think the point is that you shouldn't take anything, and not weight the benefits, because it would be zero cost

I'm not very fond of ownership mechanics to begin with, a little backwards that way. I prefer keeping track of that stuff myself

going back to @Assa's kd-trees, I think you shouldn't introduce heavy abstractions like that without considering the costs and the benefits

I would never go for kd-trees for teeworlds, too expensive to balance (edited)

wdym?

however a quadtree can indeed work decently well, or just divvying the map into chunks

oh

14:52

I thought quadtree was a specialization of a k-d tree

no, a quadtree just splits every dimension in half (in 2d = 2 splits = 4 chunks)

14:54

so a 32*32 chunk has 4 16x16 chunks, each of those has 4 8x8 chunks and so on

14:55

I said above, that I don't think you should implement this, I don't know if you gain something

Well the discussion is more theoretical at this point then practical, I don't think anyone is considering doing quadtrees for teeworlds

(Ignore me when i bullshit)

It could indeed be helpful to people like @fokkonaut who were trying to make 128 tee servers

with enough players you WILL gain something! (edited)

14:59

It's like comparing quicksort and selection sort - selection sort is faster on very small arrays

@Learath2 what?

We were discussing quadtrees, it could help with coding servers that handle more then 128 players

wow, linux have normal package manager

manjaro >:O

yep

heh yea most linux dists have package manager :P

my first linux dist

some even accept payments for not-free apps

wow

15:53

windows sucks

15:53

xd

yea (edited)

15:56

its hard for me to think of some windows feature that you cant get on linux

oh no...

wine it

1

no more server crashing now?

if you update or apply the commit as a patch, yes

when will yall fix the tees

21:59

feet are still off

21:59

its literally a 1 line fix just move the right foot 5 pixels left

was there a design change? @Zatline @Sonix

@gerdoe i call people insane who WANT to work with vs (exception for vs-code, heard its okay) - I had to work with it and i hate it, i hate it so much, i want it to burn in hell forever, in painfull purgatory

@Assa I can't understand it for the life of me either, but there are many people who enjoy it so I'm starting to think I'm the insane one

no, you are not. vs is the result of wanting a one-for-everything application (edited)

for teeworlds project it isnt good or bad, but as ide where you will make really big projects - yeah this is an peace of shit

22:07

qt creator much better

22:07

and

teeworlds is so simple, i can code it with the command line in linux

for teeworlds projects it isnt good or bad

22:08

vs code on windows always

22:08

always swear

@Assa doesn't mean you should

on linux - its much better because u can just download packages that you need and use (edited)

I prefer codeblocks, i don't need a run button or fancy project stuff, just autocompletion

22:09

@Dune yeah you are right

vs on windows == vs code on linux for me

@Assa the debugger is quite good

Also it reduces the human cost of learning/customizing two IDE :)

for a windows debugger, yeah

no, for a debugger

22:10

IMO

i still remember when i used valgrind on the MAIN product from that company i worked for

22:11

the first week i was just fixing memory bugs

simple debugger?

I have seen their code and my instinct told me: valgrind this bitch immediatly

eeeeh, the debugger has been quite inadequate in my experience, especially after gdb and lldb, it's so hard to do what I want to do on it

22:12

(it could be that I'm just not used to it, but I really had trouble debugging mostly at lower levels)

22:12

like stepping in assembly, observing memory, conditions on breakpoints

I'm not well versed in gdb actually

22:14

I can barely set breakpoints and step

22:15

conditions on breakpoints are beyond my level currently. if you have nice resources to read, please share

my worst windows experience came with microsoft azure

well it's literally condition <br#> <condition>

22:17

ngl gdb is not too easy to learn

22:18

maybe I could write something, I always wanted to get better at writing

how do you delete a breakpoint?

I gave up on VS while learning Unreal, a project of that size just makes VS cry

I always struggle with gdb, can't remmeber all the things and have to resort to google again

22:19

I use it too rarely, I guess

I do clear when I hit the breakpoint

22:20

but you can specify which breakpoint to clear the exact same way you specified the point to break on

ah

I never had used this but apparently you can delete them by breakpoint# too with delete

ah yea

22:22

that's the command I can't remember all the time

lldb is meh, but it's also very useful

never used it before, I think

Very close to the featureset of gdb

22:27

A bit saner of an interface :P

22:29

I should try a couple of gdb interfaces some day, cutter was nice to use for radare2

@Dune Not since 0.7 iirc

yeah, he's probably talking about a 0.7 change

They were moved a bit to make the tee look a bit leaning

22:33

If my memory is correct

22:33

@heinrich5991 do you have access to the website files?

yes

Oh nice, i thought only minus did

I think he gave me access because he didn't want to deal with this all the time ^^

Then i might send you a .css file tomorrow with some style improvements

22:35

uni starts tomorrow, might not have a 5min response time

Can We do it now then?

sure

Like actually starts? Is germany rolling back the lockdown?

22:37

Or do you get online lessons like all of us?

virtual

22:37

germany does want to roll back the lockdown a bit though

Ours started on schedule

they want to start schools again, soon

22:38

our semester started a week late

yeah, i am not proud of my country belonging this

@Dune notice anything different? https://www.teeworlds.com/?page=docs&wiki=client_settings (might still be cached lol)

oh yeah, looks much better

22:51

looks like a doc now (edited)

@Dune just completed the English -> Czech translation of teeworlds, hope it helps

When I update my code to the master now it still compiles but I cant start teeworlds anymore... last time I pulled was in january

@Fudgy what happens when you start it? what's your OS

wait I think it was the dll were not copied to the directory so it used the ones I had before

23:43

I copied them manually and seems like its working

23:43

it was crashing in Input->Init() when I debugged which is related to freetype

23:43

I'll say if another problem occurs..

23:43

also I dont remember how to force rebuild with bam

bam -c to clear

bam -c config for regenerating the config IIRC