DDraceNetwork - developer

An idea for fixing the weak hook issue would be to make strong hook on by default for everyone, but have a command to swap to weak. No idea if that's even possible or not.

it is possible, but not too great, will probably cause prediction errors for vanilla clients

Not sure if we should care about vanilla 0.6 since there is 0.7

strong hook by default means fixing the bug, after that it's very hard to make people weak

Not sure if we should care about 0.6 since there is 0.7

2

@ChillerDragon but that means that we can also not implement 0.7 support, where @Learath2 put alot of work in

0.7 needs a whole lot more work and atleast I won't be behind a direct port

09:45

Just porting our spaghetti over is pointless

Quote: "0.7 sux"

you suck

@fokkonaut no personal attacks please

1

its just annoying, he always trashtalks everything

no he doesnt trashtalk old ht

11:27

@fokkonaut that's no excuse for personal attacks

ok

yes u all pls stop attacking 0.7 personally

11:45

its good girl

11:46

Yea @fokkonaut the bridge thing is a nice idea but if implemented like on kog it might do more harm than good by serving a poor 0.7 expirence and advertising 0.6 clients to omprove it

11:47

improve*

yes

11:50

KoG did it very bad, I offered my help to improve it but qshar rejected it. Then I told him whats wrong (made a list) and still no improvements

same here :D

I just quoted feelsbadman

12:28

myself

Exactly what I intend to do, I'll offer a decent experience on 0.7 and as with all other subpar clients, we'll broadcast that our own client offers the best experience

12:33

If any of you have features in your clients you want supported we can get those in as long as we all decide on the common protocol, but I don't think we'll work on a 0.7 version of ddnet client until we have a 0.7 version of ddnet itself

12:36

If any of you want to put in the work to port the client to 0.7 we could maybe host that

People are forgetting that for a lot of maps you need weak hook to pass it.

We already considered this, that's why we won't do anything before taking a look

13:42

If it breaks maps it's probably not happening

How bad of an idea is it to put CNetObj_CharacterCore and protocol7::CNetObj_CharacterCore in a union? As long as vanilla doesn't touch the core "it should work(tm)" because of the common initial sequence rule

14:25

Or I can be a little more wasteful memorywise and just allocate an extra core

@Learath2 it definitely breaks maps. Except if you would turn weak hook on at the necessary parts etc

@n000b got an example?

I am bad in map names

I don't know of any except the 2 Konsti mentioned and those are speefly and shotgunrun, not certain those require weak

14:46

speedfly*

But there is are a few parts that when you have endless and hold the tee when it goes in arrow you need weak

C++20 Concepts would be so great for this project

Arrows =speedups

ah if you have strong the tee wont move?

14:47

I guess we could make the speeders weaker in those maps

Yeah or move to fast

14:48

Not sure how it was

14:49

And i know that there is a 3p map that need to hook one tee together with endless and get it through a maze. Not sure what you needed but this was also only possible with weak or strong. I guess strong tho

@heinrich5991 do you know what's going on in CCharacter::Snap The reckoning thing looks like a NOP except for clearing the core

14:57

oh, I see nvm

1way stoppers also have different physics with weak hook

how do stoppers and hook interact?

14:59

if you hook someone against the stopper?

^^ I presume

14:59

Hm I guess I'll ditch typesafety and use a template function in CCharacterCore::Write

did the character core change between 0.6 and 0.7?

15:00

perhaps you could just import it?

Well it didn't change but a cast between those is undefined behaviour

you could perhaps import the 0.6 one into the 0.7 one

hm, I'd have to mess with the generation though, do you think it's worth it?

15:02

I was hoping to keep the protocol folder the same

15:03

C++20 has a really nice std::is_layout_compatible that would let me do a std::bitcast

alternatively, you could memcpy it over

Oh I guess that's the same amount of unsafe and it's easier

I'd guess memcpy between layout compatible types is not UB

Layout compatibility is introduced in C++20 too iirc

hm, but de facto you already have to have it

15:07

because if you compile a header from two different source files, the struct layout must be the same

I mean no sane compiler would lay these out differently so I guess the implementation kinda guarantees it

15:08

Oh wow, that looks all kinds of broken

thanks @ZombieToad

Are there any maps abusing this?

idk ask @ᶰ°Konͧsti

Why would anyone use stoppers as knockoff speeders anyway?

15:09

@heinrich5991 are you sure the struct layout has to be the same even if the structs have different tags?

they can also have no tag at all, when they're typedef'd

15:11

it'd be interesting what the C standard has to say about this, but I can't imagine building a spec-compliant compiler trying to break that rule

15:11

i.e. you'd have to solve quite some issues before you're able to generate different struct layouts for structs with identical contents

I asked the ##C people but they didn't know whether there was a guarantee for this in the standard

if they don't know… ^^

It might be a little too early for the standardese people to be around though

ehm

15:13

you can put them into a union

If I put both in a union CIS covers it yeah

so they have to have the same layout?

Well idk how the compiler could avoid it tbh

oh, and we also have that the fields are laid out in the order we type them

With a straight cast it's a strict aliasing violation though

yes

15:15

because reinterpreting memory is not allowed

I don't think any sane compiler would optimize around that though

15:15

it's abused so frequently

llvm at least does it

eh, I'll put it in a union just incase

but special cases "obvious" cases

15:19

so that when you hit a bug concerning that, it's guaranteed that it's complicated

There is a map by Bixes Where one part requires weak

So stoppers and speeders are affected

Everything is affected

U