DDraceNetwork - developer

I think you're setting the wrong targets

00:00

I'm sure I can find a branch like this in the handcoded assembly game rct2, for example

00:00

which is an example for very good performance

if they need the branch, then it's fine

I'm sure I'll find unneeded branches in there

00:01

e.g. reducing the number of syscalls, that's a lot better target

00:01

these cost actual time

00:01

especially with spectre/etc. fixes

Ofc they are more significant

00:03

this is all beside the point though, it's only a single branch insruction in the silly no-op translation "table"

yes, but I find it ridiculous that you claim that that branch matters

00:04

I agree that adding a translation table is not worth it for no benefit

00:04

but claiming that a branch there matters, is absurd IMO

depends on how you define "matters"

matters for software performance

00:04

for making your software run faster on a '95 computer

that branch instruction will cost you a couple cycles, and it will have no gain, I don't see how anyone can argue that that is fine

00:06

If there was any value gain, fine there is nothing to argue, the branch is insignificant

say the gain is that I want every client id to be 0

00:06

that is the only reason

00:06

even then it's completely insignificant

if you need that, sure branch away

I can't state enough wishes of that form to make it matter

00:07

it's just that I find cid=0 aesthetically more pleasing

00:07

even this very discussion probably used up more cycles than the branch in my hypothetical tw server

00:07

by broadcasting it to many people, etc. etc.

00:08

so in a sense, arguing about that branch made performance worse already

I don't get the point of this, every decision you make comes with a cost and a reward, you try to maximize reward and minimize cost

except that you have a meta optimizer there, too

00:10

that makes you not care about details that don't matter

00:10

otherwise you'd spend too much time optimizing things that don't matter

00:10

as a matter of fact, this optimizer just called, sorry

Premature optimization is not a good idea, sure. Not being cognisant of the performance implications of your code is worse imho

00:16

in context of teeworlds we have the luxury of being lax with our code because we have many cycles to spare before the performance becomes an issue to players

00:17

can't apply the same attitude working with high performance software where you want to squeeze out as much performance as you can nor can you apply it to embedded programming where you don't have performance to throw out the window

00:20

branch mispredictions and cache misses have non zero cost, they don't matter much for something like teeworlds that lies in the middle of the spectrum, that doesn't mean we should shove a loop that branches a million times in the middle of the main loop

I agree with your previous messages

anyway this is a tangent, I don't think we should decouple the clients understanding of the gameworld from the servers

Having a seperate set of clients for dummy and main is just like having another client as dummy

00:24

it just makes sense imo (for 0.7, in 0.6 it is automatically due to snaps)

if you think clients shouldn't be receiving the same set of clients, yes, that is the correct approach

for mods you can never kniw

00:25

know*

I also still believe the player info belonged in the snap

Me too

apparently the new skin system made the snap a bit larger

Is it so much bigger?

idk, I never checked it

guess a few new entries is not too bad

still, delta compression should be dropping the entire playerinfo

wym?

we don't always send complete snapshots

00:32

most of them are diffs called deltashots

Why?

because it'd be a huge amount of traffic to keep sending entire snapshots?

Oh yea true

I'm very sleepy, I have an upset stomach so I can't sleep

00:36

so annoying

ddnet-maps

[ddnet-maps:master] 1 new commit

3f177ac A run_unbelievable - ddnet-maps

You need to enable the python language server in vscode to get proper intellisense...

ofc

not microsofts one, that one doesn't want to do type inference

12:34

you need to enable the jedi one

12:38

also protip: if any of you wants a windows 10 virtual machine, go for a windows 7 one instead it performs much smoother under virtualization

1

12:38

if you have to get a windows 10 one, boot the vm headless and RDP into the machine instead, microsofts rdp works better then all 3 virtualization platforms I tried

1

12:39

(their native display adapters that is)

can you rdp from linux?

hm, didn't test it from linux, macOS has an official client, so does iOS, I think linux has a couple unofficial alternatives that perform well but I can't vouch for those

Is there a tool better then bors? If not I'm thinking of doing https://github.com/bors-ng/bors-ng/issues/185

Take advantage of pull request status information when available · ...

Consider the situation where: Pull requests are built the same way as staging/trying. The tests pass on a pull request. The pull request has master as an ancestor. There is nothing else in bors&amp...

@Learath2 if you find a better tool, let me know

hm, I really expected more from pythons functional programming interfaces, you can't even easily chain

14:46

Java and javascript both handle this beautifully

itertools.chain?

chain as in chain higher order functions like filter, map, reduce

maybe I don't know what chain is

14:52

do you mean flatten?

list.stream().map(m -> sin(m)).filter(m -> m > 0).collect(toList()) is what it'd look like in Java

14:53

list.map(m => sin(m)).filter(m => m > 0) in js

[sin(m) for m in list if sin(m) > 0] pythonic

list(filter(lambda m: m > 0, list(map(lambda m: sin(m))) in python

14:54

ofc I can use a list comprehension, my complaint was about the higher order functions filter, map and reduce

hm

14:55

I guess python's stance on this is: use comprehensions instead of higher order functions?

hm, if that is your stance about something in your language, I don't think it belongs in the language

list comprehensions have some problems

I was thinking that list comprehensions are equivalent in functionality

14:58

yes, lambdas are too verbose for other things in python

I'm pretty sure they are

15:01

well javascript lacks it but java and rust have lazy evaluation for their streams and iterables respectively

put it in paranthesis in python to achieve that

put what in parenthesis?

the list comprehension

(for x in a)

(sin(a) for a in iterable)

hm, hadn't seen that before, that's useful

this makes a generator iirc

yes

fwiw I prefer the chained calls to the list comprehension syntax, it's just easier on the eyes imho

15:07

I guess one could argue that pythons list comprehensions are "easier" to read as it's basically english

15:07

oh and list comprehensions are just awkward to type

15:10

Overall, I think if I wanted a language that lacks a type system, I think I'd use javascript nowadays

15:10

Next up on the docket of things to try is perl

js has its weird things but its good

15:11

i use typescript if i can tho

I also go for ts for any project that's getting deployed

also the way it seamlessly connects with html when using react, jsx tsx

15:11

i love it

observables and promises aren't half bad constructs for doing asynchronous work too

15:19

nim's futures are also cool

15:20

maybe I can take a look at Go, it's supposed to be great for concurrency

tee keep crying now when you hold fire while frozen

15:44

is that wanted behavior?

no

lol, how did that happen?

the same commit as before, I guess

16:03

the same PR

idk what "as before" refers to in this context

16:04

also do you know of a python equivalent for an assignment in an if stmt?

https://github.com/ddnet/ddnet/pull/2086

Remove setting ammo for freeze to 0 by fokkonaut · Pull Request #20...

or if let in rust

they implemented the "walrus operator" recently

16:05

if m := re.match("abc", "abc"): (edited)

it's >= 3.8 sadly

ah

too recent

other than that, not aware of any

is the idiom really m = re.match() if m is not None:?

yes

16:06

or if m: also works

this language was made by the scope leak gang

there is only¹ function scope in python anyway

yeah, needs more scoping

16:07

scopes are nice to avoid bugs, like types

yep

16:07

try messing with lambdas and captured variables

16:07

that's horrible in python

16:08

return [lambda: i for i in range(10)]

16:08

gives you a nice list of ten functions returning 9

wait what?

the "fix" for that is: return [lambda i=i: i for i in range(10)], which is even worse

who decides the direction of python now that the BDFL is gone?

the community, probably? I don't know

16:14

ah look, JS has the very same problem

16:14

var array = 0; for (var i = 0; i < 10; i++) { array.push(function() { return i; }); }

16:14

gives you a list of ten functions returning 10

16:16

how to fix it in JS?

you want a list of 10 functions returning from 0 to 9?

yes

let arr = []; for(let i = 0; i < 10; i++) arr.push(() => i);

16:30

(arrow functions (lambdas) have lexical scope)

so the fix is replacing function() { return ...; } with () => ...?

yes

16:32

or () => { return ...; } if you prefer

16:35

hm, I don't get why the initial python example doesn't work

because it captures the variable i which changes later on

16:37

why does the JS example not work with functions but with lambdas?

16:37

or "what is lexical scope?"

16:37

and "what scope do function()s have?"

now that I've read it again, I don't think the scope was the issue here

16:42

ah an anonymous function would work too, it's scope creep

16:43

it's let i instead of var i that's fixing it

@heinrich5991 with let the scope of i is constrained to what's inside the loop body, at each iteration you get a new i, and the new anonymous function you create forms a closure around that new instance of i

16:50

with var it's scope is not limited at all so it suffers from the same issue with python

ah

js is such a minefield

16:59

but with ES6 we now have better control over scope (like let)

trml

[ddnet/ddnet] Pull request opened: #2161 Small prediction fixes/update

I looked through the latest changes in the server code, and made some updates to the prediction with it. This includes the hammer-out-of-freeze server fix and a small fix for stopper prediction. Also cleaned up some code, and fixed predicted smoke-trails slightly so they don't go through walls in certain cases.