multiplayer

[grizzled stirrup]

since the game won't allow that as it is simulating the game itself on server

[exotic wasp]

aimbot is tragically just not counterable without anticheat since it mimics player input

it's still perfectly possible to spinbot in CS and valorant

but having a single match with a cheater has long term consequences with things like rank so it's a lot more important that EVERYTHING is perfect

[grizzled stirrup]

Yeah anything like the method that shootergame tried to do (cull based on visibility) sadly doesn't work with latency as they pop out of nowhere

[exotic wasp]

most games still do that

for performance and to negate wall hacks

ideally never send a client more information than it needs at that exact point in time

[grizzled stirrup]

Interesting, I've seen recent warzone hacks that show full body silhouettes through walls and at the very least many games have a rough bounding box of a player to simulate footsteps and such

The unfortunate side effect of needing to give legit players lots of information on who is around them

[exotic wasp]

well that's a massive map so it doesn't surprise me that they don't do that

could also literally be simulating animation based on audio locations

[shrewd ginkgo]

When I package the game, the create advanced session gives the output "On Success" and opens the lobby level. When my friend searches for the game, the find session also gives the output "On Success", but the result length is 0 and it does not see the server. Steam overlay is activated.

[/Script/Engine.GameEngine]
+NetDriverDefinitions=(DefName="GameNetDriver",DriverClassName="OnlineSubsystemSteam.SteamNetDriver",DriverClassNameFallback="OnlineSubsystemUtils.IpNetDriver")

[OnlineSubsystem]
DefaultPlatformService=Steam
bHasVoiceEnabled=true

[Voice]
bEnabled=true

[OnlineSubsystemSteam]
bEnabled=true
SteamDevAppId=480

[/Script/OnlineSubsystemSteam.SteamNetDriver]
NetConnectionClassName=OnlineSubsystemSteam.SteamNetConnection

[/Script/UnrealEd.CookerSettings]
bCookOnTheFlyForLaunchOn=False

[tardy fossil]

if you print the number of elements in the results array after the Find Sessions node, does it consistently say 500?

oh wait i didnt read the last part of your message lol

i do notice you're using create advanced session but not using find session advanced.. not sure if thats your issue but its worth a try

[shrewd ginkgo]

It was working fine before but it broke down for some reason I don't know.

[mortal blaze]

Hello, thank for the answer.
I started looking into gamestate, and it seemed promising for what I wanted to do.

However, when I empty the array with my system via the client, the server still has access to all the numbers in that array. However, when I empty it from my server, the client always returns "0."

I feel like I missed some information about servers with gamestate, but I can't find it on the forums.

[sinful tree]

Replication only happens one way, server -> clients. If you clear a variable on clients it'll only ever clear it locally. You can have the client request that the array be cleared for everyone by having that client send a server RPC and then having the server clear the array.

[mortal blaze]

Replication with gamestate is only server -> clients ?
A listening server is not seen as a client?

[sinful tree]

Replication = Automatic sending of stately replicated variables out to others. Doesn't matter what actor it is, replication only happens from server -> clients.
Setting a replicated variable on the server will make it so that value is replicated out to others.
Setting a replicated variable on a client will only set it locally on that client.
A listen server where the player is playing and hosting the server in the same game instance is still considered the server.

Clients can send the server RPCs (Remote Procedure Call), which is effectively asking the server to do something for them. You can have the client send the server values through inputs on the RPC and then have the server do something with those values. Clients can only call these RPCs on actors or components of actors that they are currently the owner of - typically their player controller, palyerstate and their controlled pawn

[mortal blaze]

Oh, thanks for the explanation. I think it's pretty clear, but it's still a problem for what I want to do. I'll try another way.

[arctic mist]

Just bumping this so it doesn’t get lost! Would like an answer please!

[vivid seal]

For the example you gave, you'd probably want a replicated variable since all clients should have the same collision profile when it changes. In general my rule of thumb is RPC for events (a thing happened and i want everyone to know about) and replicated variables for state (a variable changed and will stay changed until I change it again). In your example, if you multicast to change the collision profile, it might work right now, but if (for example) another player joins the game later on, on their screen the collision won't be set correctly, since they weren't there to receive the RPC originally.

collision is a weird example for late joining because movement is handled in its own little world, but if you were changing the player's team color or something for example, that would be something persistent that you want everyone to know about, including people who didn't receive that RPC when it happened.

[arctic mist]

oh okay, but if i had a bunch of unreliable RPC functions it wouldn't slow down the game dramatically though right? was my main question

or should i start using replicated variables more often?

[exotic wasp]

anything important to gameplay should be reliable

rpcs sadly aren't a replacement for replicated properties since RPCs don't change state for new players (and other issues)

[devout sonnet]

I removed all widgets and I still can't trigger that click event. The game is super small right now, i don't think i have anything consuming input. And i don't know why it doesn't work only in a joined session.

It's not just the click events either. it seems that all input event's (Like pressing enter) doesn't work in this joined session. Is there a setting i'm missing?

I forgot to set input Mode Game and UI lol

[amber vale]

Hey, can anyone explain why there's a discrepancy between the local client and the server when reading PlayerState::ExactPing? The server's estimate of a connected client's ping is usually 10-20ms higher. What's the technical reason for that?

[grizzled stirrup]

Been a while since I looked at it but AFAIK ExactPing isn't replicated and is in bytes so you have to multiply GetPing by 4 to get actual ping

int32(PlayerState->GetPing()) * 4)

[amber vale]

That's CompressedPing. ExactPing is replicated between the server and local client. I think it just has something to do with processing time or tickrate, cause it's consistently around 15ms higher than the client's version, regardless of what the client's ping actually is

[grizzled stirrup]

At least on UE4 it isn't but maybe it is now in UE5

[dark edge]

either reliable or sent often enough that it doesn't matter

my aim direction is important for gameplay but it doesn't need to be reliable because there's an absolute fire hose of aim directions flying at the server over time

a few missed ones is fine

[nova wasp]

it can optionally include the delta time of the frame

this is probably where the 15ms comes from

[dark edge]

https://github.com/EpicGames/UnrealEngine/blob/2d53fcab0066b1f16dd956b227720841cad0f6f7/Engine/Source/Runtime/Engine/Private/PlayerState.cpp#L41

Looks like CompressedPing is the replicated one

[cold temple]

Hi, I want to control all the item-spawners in a multiplayer setting, should I create a actor component to solely do this and then attach it to the gamestate? or is there a better way?

[thin stratus]

Component on GameState sounds like a good start

[twin cliff]

if i want somewhere i can hook into to persist data to a database when a player leaves the server or when the server shuts down gracefully, what are the recommended places for that?

Logout() on the game mode seems like the place, although the pawn is already destroyed (i guess i should track the state i need for it in a subsystem or something instead?)

[regal narwhal]

Hi there, I have a problem when GrantAbility via AbilitySystemComponent. My AbilitySystemComponent has a GrantAbility function that also bind to input via enhanced input system. I cached InputSystem when BeginPlay. but granting bility works on server side and binding ability to a input doesnt work since EnhancedInputComponent is invalid and only valid on Client. how could I fix the probnlem?

    {
        Spec.InputID = GetOrAssignInputId(InputAction);


        if (CachedInputComponent.IsValid())
        {
            BindInput(Spec.InputID, InputAction);
        }
    }


    return GiveAbility(Spec);```

[sinful tree]

I believe you should be able to override OnGiveAbility() in your UGameplayAbility to then handle the binding.

[regal narwhal]

that will most probably work. thanks for tip. I ve checked and see the inputcomponent is valid

[lost inlet]

yep, GetPingInMilliseconds will also return the corrected value, or the precise ping if you're a local player

[sinful tree]

APlayerController::PawnLeavingGame() is where the pawn gets destroyed when the playercontroller is destroyed. You can override this and it should be good for handling the pawn data when a player disconnects for whatever reason.

[meager aurora]

How would you approach avoiding the issue of replaying ragdoll physics, for late joiners during death event and showing the final death pose? SaveSnapshot? Would that replicate?

[lost inlet]

that seems like a solution waiting for a problem

[limber gyro]

why do you even need to replicate a ragdoll? i have a feeling you are going at it the wrong way, but i have been wrong before

[meager aurora]

@limber gyro I just wanted to simulate physics on death event, bad idea?

[limber gyro]

its not a bad idea, you are just probably over thinking it if it doesnt afect gameplay

a player that did not see other player die wont care that a ragdoll animation isnt there

[meager aurora]

No, that's not the issue.

Player 1 kills Player 2, Player 2 enters ragdoll state.
Player 3 joins later, sees the ragdoll just taking place, not the final pose.

[limber gyro]

i would do the ragdoll stuff localy only and not even show it to players joining

not sure how that would look like for you tho

if you dont have a camera following the ragdoll you could just spawn one locally at the time of death

[meager aurora]

Spawn what exactly ?

[limber gyro]

the skeletal mesh, and then make it ragdoll

spawn a copy of the skeletal mesh of the character that died

and set the character that died to invisible/no collision and get rid of it when you see fit

[shrewd ginkgo]

the game is properly packaged. when i start the game the steam overlay comes up. i install the server and it goes to the lobby menu. when the second computer searches for the server the find session result length is 0 and it says no session found. it was working fine until recently but I probably changed something unknowingly. can someone help

https://blueprintue.com/blueprint/df_3ozzt/

https://blueprintue.com/blueprint/s9enyufx/

there is what I do in Server UI and GameInstance

[sinful tree]

in the open level node you need to put "?listen" in the options field.

[shrewd ginkgo]

ty Im gonna try it

[twin minnow]

Did you ever finda workaround for this?

[shrewd ginkgo]

questionmark doesnt work I put it but I got same result

[shrewd ginkgo]

I found the solution

I closed Warp

[keen slate]

Is this generally safe to call on the server, or does it only work on the controlling client?

APlayerController::IsPrimaryPlayer();

[thin stratus]

What kind of Server?

That function was pretty cursed a couple years ago already.

It will use APlayerController::IsSplitscreenPlayer, which will check the ULocalPlayer of the PlayerController.

That won't be valid for anyone but the owning client

[keen slate]

Yes, I tried looking through the implementation, but it didn't look very comprehensible - some branches use net id.
But it's fine, I just moved the logic to a client RPC for safety.
Server is Listen..

Server does not have access to clients' UlocalPlayers? There's a base variable in PlayerController of type UPlayer, and it looks like some kind of networking is involved..

[thin stratus]

It shouldn't have access to the ULocalPlayer.

At least, iirc.

[hallow cliff]

Hello, in my lobby system, players are selecting characters, so the character is not spawned. Where can I install voice chat in this lobby section? I wanted to install it on the player controller, but it does not accept the component that the settings will attach to. Does anyone have a suggestion for where I should install it?

[vivid seal]

Having an issue using ServerTravel(“?Restart”) where the client is getting a connection timeout after 20 seconds and booted back to menu. I have both InitialConnectTimeout and ConnectionTimeout set to 60 seconds in my DefaultEngine, so I’m not sure where this 20 second value comes from.

[jade heart]

would anyone know why when i sprint it is glitchy and not smooth? my sprinting code > https://blueprintue.com/blueprint/l-mv2kee/

and here is a side by side of the server running and the client running (client is the one with all the ui)

the console error was just cause the server was running btw

[dark edge]

Because a true multiplayer sprint isn't really possible in BP only

[dark edge]

https://www.youtube.com/watch?v=4w8vSss4YiI

[errant mulch]

hello guys, excuse the interruption

im currently trying to pass a reference of the player character on client A, to the client B, this is to compare a variable inside the playerstate that indicates on which team are each player, so when client A calls for the function, it executes on client B but not Client A.

what would be the best option to pass the references? using rpcs like in the example image?

is this even the correct approach to execute a function on one specific client but not the other?

[dark edge]

in the big picture, what are you trying to do here?

When Client A takes this action, XXXXXXX should happen

[errant mulch]

So, what im trying to do here is basically

1. Someone from (lets say red team) use a power that its main goal is to play a music for the rival on the blue team

2. to do that, i have on playerState the teams, red being 0, and blue being 1

3. i pass onto the server the reference (self) of the player that used the power, so then i can pass it along to a multicast to the rest of the clients.

4. when it arrives at the multicast, i get the reference of the player, and get the team that originally used the power up, so it will only affect the ones that are NOT on the red team

thats my idea on how to get the reference from the original client to the rest of the clients

[dark edge]

Is anything spawned here or do you want to do this just with function calls, not spawning any MusicPlayer or whatever

[errant mulch]

but as im seeing, its prrobably wrong

nope, no spawns

unless you consider this node a spawn

[dark edge]

ok so assuming you have some event that plays music, give it a parameter of the same type as your team.

MulticastPlayForTeam(Team)

then on the clientside end of that, check if the LOCAL PLAYERS team matches that team

What actor does all this code run in?

[errant mulch]

on the Character player

[dark edge]

oh in that case you don't even need to pass anything over

[errant mulch]

i dont?

[dark edge]

MulticastEvent -> check if local players team = this pawn's players team -> play audio if not

the pawn has an associated playerstate so you can compare that team with the LOCAL PLAYERS team

you need to get the local player

In Pawn:
Multicast -> if self.playerstate.team != localplayerstate.team -> play audio

that simple

multicast code runs on every computer for which that actor is relevent

[errant mulch]

local players team... thats the one on the playerstate right? dosent the player state share its variables among all clients?

[dark edge]

so you want to compare the local playerstate on that computer with self (the pawn's) owning playerstate

[dark edge]

everyones playerstate is on every computer

if there are 4 players in the game, on your computer your playerstate will be local and the others won't

[errant mulch]

yeah, ive seen this diagram, its hard to keep track sometimes where is everything

[dark edge]

everything in the middle is on every machine

so if you and are in in the game, and i did the multicast, on my machine, my local player and my pawns local player would be = so no audio

on your machine, your local player and my pawns player teams would not be equal, so play audio

The "rule" to play audio is if the audio playing pawns team and the local players team do not match

[errant mulch]

im about to get it, but im still doubtfull about the definition of local player and pawns local player

[dark edge]

local playerstate vs a pawn's playerstate

[errant mulch]

every player has his own playerstate, that i know

[errant mulch]

god i feel really stupid but i think i get it

let me try something

also, another question, could i change the team directly on the player state with something like this?

in this case, it will change the team on the local playerstate, correct?

[dark edge]

setting the team will have to happen on the server

[marble magnet]

Does anyone know why using a listen server, crouching (using the built-in function) and making the server lag (outbound loss mainly), other clients see the server's character's location and crouch state "desync" ? If the same thing happens to clients, this issue doesn't happen, I remade the c++ function myself replicating it through rpcs and the same thing happens, even tho it replicates correctly normally, but the craziest thing is, if done in bp, the behavior changes (??), it stops desyncing but it may jitter for 1 frame. Then I checked on lyra, using the same functions I used, it doesn't desync. I'm completely lost on all of this, has anyone run into this and know wtf is going on ? It looks like the server starts sending its location while missing his crouch/capsule size, so it "floats" in the air as the center of the capsule went up

[errant mulch]

this is probably the key part i needed god

[errant mulch]

thats another thing i cannot grasp, i always assume that replication will take care of the variable going into the server

[dark edge]

no

it never will

variable replication is server to clients ONLY

and the only way to send info to the server is run on server events

read the compendium

[errant mulch]

i have to re read that alot these days yes, something are clicking, like, i knew that the only way to send info to the server is with a rpc, but i never thought it applied also to variable replication

[dark edge]

The only way to tell the server anything is through an RPC

just drill that in your head

[errant mulch]

golden rule drilled

[marble magnet]

tab triggers the default crouch function, caps lock trigger the c++ function (that is basically copy and pasted from the og one but without relying on the vars the og one sets), and left shift is full bp implementation

[nocturne quail]

Reliables needs with validation?

Server RPC missing 'WithValidation' keyword in the UPROPERTY() declaration statement.  Required for security purposes.

    UFUNCTION(Server, Reliable)
    void ServerToggleJump_Implementation();

this the first time i encounter this error

[fossil spoke]

Why have you appended _Implementation to your declaration?

The implementation function will already have that, you dont need to add that to the declaration

WithValidation is mandatory AFAIK

Might have changed in UE5 🤷

[nocturne quail]

i always do it and it was working fine

[nocturne quail]

I downgraded to 4.19 which is more suitable for my project

[fossil spoke]

    UFUNCTION(Server, Reliable, WithValidation)
    void ServerToggleJump();

Thats what it should look like

[fossil spoke]

That would explain it

[nocturne quail]

its working now Thanks

[nocturne quail]

oh, its not working and again gives the same error

[fossil spoke]

You need to add the _Validate implementation function

bool UMyClass::ServerToggleJump_Validate()

[nocturne quail]

yeah this is what the error says but why that needs validate if i don't want to validate the state change

[fossil spoke]

Its just a requirement, you generally will always return true from the function.

Unless of course you want to validate it

Returning false will cause the connection to be closed.

[nocturne quail]

I validated in the past when i need to, and it was not necessary afair

maybe its 4.19 specific?

[fossil spoke]

As i said earlier, in previous versions of the engine it was mandatory

[nocturne quail]

ah, my bad english i just translated mandatory

Thanks i will validate it as it needs to 🫠

[fossil spoke]

👍

[marble magnet]

Still about this, anyone knows why lyra and default ue5 differs in how the server replicates its position if it's desynced with the client (here intentionally setting the capsule taller on the server than the clients) ? Default ue5 tries to snap it back to the "real" location (center of the capsule) while lyra keeps it lower, this is not running the custom classes of lyra, default character and all

[torpid lantern]

How does Mass Entity hold up for multiplayer?

I see "MassReplication" is a thing, but I can't find much in terms of documentation or tutorials.

[quasi tide]

Not well supported. You'll roll your own.

[dark parcel]

to add on what is already said.
Incase it hasn;t click.

Server RPC is THE ONLY WAY for Client to communicate with server.

[cloud crown]

    bool PickupItemByInstance(USI_ItemInstance* InItemInstance);```
```bool USI_InventoryComponent::PickupItemByInstance(USI_ItemInstance* InItemInstance)
{
    if (!InItemInstance) return false;
    USI_ItemInstance* LocalItemInstance = InItemInstance;
    InItemInstance= nullptr;
    return InventoryList.AddItemByInstance(LocalItemInstance);
}```

bool FInventoryList::AddItemByInstance(USI_ItemInstance* InItemInstance)
{
if (!InItemInstance)
return false;
FInventoryListItem& Item = Items.AddDefaulted_GetRef();
Item.ItemInstance = InItemInstance;
MarkItemDirty(Item);
return true;
}

its been a month not able to fix this bug

after pick up i destrouy the item actor not the instance the actor have the instance ref
but in client the item instance in the inbventory becomes null if the item sactor is destroyed

{
    bool WroteSomething = Super::ReplicateSubobjects(Channel, Bunch, RepFlags);

    WroteSomething |= Channel->ReplicateSubobject(ItemInstance, *Bunch, *RepFlags);
        
    return WroteSomething;
}```

[chrome bay]

The instance is a replicated subobject of the actor, if you destroy it, the instance will go with it.

No different to an actor component

You also don't need to copy the item pointer in PickupItemByInstance(), that does nothing

[grizzled garnet]

I asked something in #online-subsystems but haven't had a response, so I'll ask here, a separate but related question

The issue I put in #online-subsystems is that there's an error preventing either players connecting to clients, or players starting up games, with the Steam Sockets protocol.

After doing some research, it would seem that Steam Sockets and the default Steam Networking net drivers are incompatible, which is the root of my issue. I want players to be able to connect via manual IP to games, meaning the null subsystem should be active simultaneously to the steam subsystem. However, I cannot set the fallback driver to IPNet, and IPNet doesn't get set when -nosteam is passed if the Steam Sockets is active (in this case the driver error I showed is reported).

Questions:

• What is the best way to swap subsystems at runtime? Or more specifically: how can I specify the NetDriverDefinitions used at runtime? I have a plugin that can check if Steam is enabled, so if I can select which NetDriverDefinition to use at runtime then I could choose IPNet when steam is disabled

• Is there some way to connect over IP to a game while Steam Sockets is active? Steam Sockets uses a relay thru the Steam network which is great for games found in session searching but not if I want a direct IP connection. In UE's documentation there's a parameter OnlineSubsystemSteam.bAllowP2PPacketRelay, it sounds like setting this to false would establish what I am looking for?

[nocturne quail]

creating onrep's using static is a good practice?

    UPROPERTY(ReplicatedUsing = OnRep_IsCrouch) bool bIsCrouch;
    UFUNCTION() static void OnRep_IsCrouch();

Static void function is declared as static, its scope is limited to the translation unit (e.g., the specific .cpp file) where it's defined. It can't be called directly from other files. The function exists for the entire duration of the program.

Normal void function, without the static keyword, has external linkage by default. This means it can be accessed from other translation units (files) within the program.

since this is a replication function and will have a tree connected to other classes, will it not break replication at some point?

[pseudo wagon]

Does it even work?

[nocturne quail]

yeah rider did compiled it, but yet not tested in the editor

if it does compiled it means it will work

and resharper in visual studio also suggested me to make it static or even const and both did compiled

I'm on UE 4.19 in case if for you it will not work in UE5

[nocturne quail]

are these virtuals not reduntant in the unreal API?

https://github.com/EpicGames/UnrealEngine/blob/release/Engine/Source/Runtime/Engine/Classes/GameFramework/Character.h#L677

it should be like

ENGINE_API void UpdateNavigationRelevance() override;

if i'm not mistaken

[lost inlet]

a static OnRep is extremely cursed

though that whole indentation scheme is also cursed

[nocturne quail]

agree, its not yet finished

[lost inlet]

using an ancient engine version for what I'm sure are Missty-exclusive reasons is also cursed

[nocturne quail]

I finished with unreal engine 5, it doesn't have the real needed things which it should have if i take into account the memory it eats not only in editor mode which i can afford, but in shipping builds there is 50% difference in terms of optimizations
talking about the same project which i moved form 4.26 to 5.4, and now downgraded to 4.19

this solve one of my mystery question that why is bigg aaa companies not switching to ue5

actually ue5 is also ancient, it has nothing new we really need... we developers don't need fancy interfaces

[errant mulch]

hot take: i preffer the ue4 interface and is not just nostalgia, everything is easier to find there

[nocturne quail]

someone told me it has plugin like landscape patch... i am telling them who need this plugin for flat surfaces, this plugin can't even detect a surface if you use it with a slopped object like road

[lost inlet]

I knew I was going to get something non-sensical

[nocturne quail]

not just to find there, look at to the zoom steps in ue4, makes it very easy to organize assets and view ports

in ue5 it jumps like crazy and have only 4 or 5 steps with big jumps 😄

[nocturne quail]

there is the finished one 🙂

[lost inlet]

U-macro on the same line as the declaration is still cursed, and I suppose since you're on an old version you're forced to have WithValidation and there's no TOP. Component references shouldn't be Blueprint writable either

[nocturne quail]

if it works that's fine?

because its working 😄

ah, bp read write should not be used in private members 🤔

so is it not possible to make it at least editable in the child bp class?

i mean having in private

[sinful tree]

Make it protected

[nocturne quail]

yeah seems like i have no other easy option

[prisma snow]

Is it possible to have a virtual function for OnRep?
Have a base class with a replicated variable and I want to trigger some code once it's replicated depending on the derived class

[lost inlet]

yes

[prisma snow]

I'm setting up player bots for our RTS, probably the central class for the bots will be a server-only PlayerController derived class. I wanted to ask, is it possible to spawn those controllers AFTER all human player controllers have been created etc by GameMode? It seems that StartPlay() get consistently called AFTER RestartPlayer() but Idk if I can rely on that

[latent heart]

Surely you want to derive it from AIController?

[prisma snow]

I looked into it, and for now I don't think that's a necessity - we don't plan to use any of the built in AIController functionality

But if you have a good reason to do it I am all ears, it's the first time I work with AIControllers

[latent heart]

Why not just use AController then?

What part of the Player part do you need?

[prisma snow]

Ahhh because we have some common functionality between human player controllers and AI player controllers (mainly the interface and helpers to send and replicate orders), plus some entity control stuff.
So we have a base class for both human and AI players that derives from APlayerController

[latent heart]

Components

[late vector]

Hey! Sorry to interrupt your conversation, I'm currently learning multiplayer in C++, and I have a question. If I don’t set up any replication, should the client still see exactly what the server is doing? My grab system works on the server, but the client doesn’t see it correctly, and I’m not sure why. Any insights would be really appreciated!

[nocturne quail]

Clients won't automatically see updates or changes happening on the server

[late vector]

sooo I need to do all the client and multicast stuff ?

so like logical stuff on server and visual stuff on server + client ?

[nocturne quail]

server don't need any visual stuff if it is dedicated server

[late vector]

and if i'm in a P2P game ?

[nocturne quail]

in short you request the server to update something on client

[nocturne quail]

i'm not sure if unreal uses p2p

[late vector]

I mean listen server

[nocturne quail]

while Unreal supports P2P, the more common approach is the dedicated server model for multiplayer games

[lost inlet]

The server can do whatever it wants, it's the server. But you have to implement replication if you want clients to see or do anything

This is not optional

Multiplayer is not trivial

[late vector]

mmmhh alright because for instance I try to do that before where I grab with my client (I do the line trace etc.. on client then server check if line trace is valid and also execute SetTargetLocation etc...) and it seems liek the client struggle to keep in sync and idk what to do next

[late vector]

well I make a game to play with friends onn steam so I don't need a dedicated server

[lost inlet]

The only difference a listen server makes is that one player is a server

And they're the authority and can do whatever they want

For everyone else, it's pretty much indistinguishable from being connected to a dedicated server

[nocturne quail]

BP_MessageActor
│
├── Variables
│   └── Message (String, Replicated)
│
├── Events
│   ├── Event BeginPlay
│   │   └── (Superclass BeginPlay)
│   │
│   ├── Custom Event: SendMessage (Input: NewMessage (String))
│   │   └── Set Message = NewMessage
│   │
│   └── Event: OnRep_Message
│       └── Print String (Message)

AMessageActor
│
├── Variables
│   └── Message (String, Replicated, using OnRep)
│
├── Functions
│   ├── AMessageActor::AMessageActor()
│   │   └── bReplicates = true
│   │   
│   ├── AMessageActor::BeginPlay()
│   │   └── (Superclass BeginPlay)
│   │
│   ├── AMessageActor::SendMessage(const FString& NewMessage)
│   │   ├── if (HasAuthority())
│   │   │   ├── Set Message = NewMessage
│   │   │   └── OnRep_Message()
│   │   └── (End of Function)
│   │
│   ├── AMessageActor::OnRep_Message()
│   │   └── GEngine->AddOnScreenDebugMessage(-1, 5.f, FColor::Green, Message)
│   │
│   └── AMessageActor::GetLifetimeReplicatedProps(TArray<FLifetimeProperty>& OutLifetimeProps)
│       ├── Super::GetLifetimeReplicatedProps(OutLifetimeProps)
│       └── DOREPLIFETIME(AMessageActor, Message)

this is the structure unreal uses for replication

[late vector]

alright and for instance this should work with client and server right ?

[sinful tree]

Unreal doesn't support P2P out of the box. Listen server = still client/server and it's a misnomer to call it P2P. Whether you use dedicated servers or listen servers is entirely up to what your intentions are with your game and there is no "common" approach as this varies based on the type of game you're making.

[nocturne quail]

yeah, i already told them unreal uses server -> client model, but it can support p2p and there even few plugins i just searched

[nocturne quail]

server instance is always authority and all connected clients are in the controll of server

you can code the server to update clients

[late vector]

well I try to do that indeed but lets say to do that in c++ I need to be on the server of course and them like execute a UFUNCTION(Client) or UFUNCTION(NetMulticast) ? or i'm dumb idk lmao

[nocturne quail]

client is for only owning client+server, multicast is for all clients and server

[lost inlet]

if you want true P2P then you need to implement custom networking for that. "P2P" in things like EOS is just a way of accommodating listen servers without having to forward ports on your router

[nocturne quail]

this message should be pinned for future users, short and very informative

[prisma snow]

I'd read through the Network Compendium pinned here and prepare yourself for learning a lot, multiplayer in Unreal is pretty difficult initially although very powerful later on

[nocturne quail]

how can we both talk about pinned post at the same time 😄

identical brains

[quasi tide]

I would say MP in UE is the easiest of all off-the-shelf engines.

[late vector]

Yeah, I know but even after reading the network compendium, it all seemed understandable at first, but in practice, it's a whole different story haha

[nocturne quail]

here is the code version:

UCLASS()
class YOURGAME_API AMessageActor : public AActor
{
    GENERATED_BODY()

public:
    AMessageActor();

protected:
    virtual void BeginPlay() override;

public:
    UFUNCTION(Server, Reliable)
    void SendMessage(const FString& NewMessage);

    UPROPERTY(ReplicatedUsing=OnRep_Message)
    FString Message;

    UFUNCTION()
    void OnRep_Message();
};

AMessageActor::AMessageActor()
{
    bReplicates = true; // Enable replication
}

void AMessageActor::BeginPlay()
{
    Super::BeginPlay();
}

void AMessageActor::SendMessage(const FString& NewMessage)
{
    if (HasAuthority())
    {
        Message = NewMessage; // Update the message on the server
        OnRep_Message(); // Call OnRep manually to notify all clients
    }
}

void AMessageActor::OnRep_Message()
{
    GEngine->AddOnScreenDebugMessage(-1, 5.f, FColor::Green, Message); // Print the message on all clients
}

void AMessageActor::GetLifetimeReplicatedProps(TArray<FLifetimeProperty>& OutLifetimeProps) const
{
    Super::GetLifetimeReplicatedProps(OutLifetimeProps);
    DOREPLIFETIME(AMessageActor, Message);
}

[lost inlet]

I'm not sure how this random ass replicated actor is relevant

also the mindset of multiplayer is the hard part rather than the UE framework part

[sinful tree]

By that logic, Unreal can technically support everything ever coded ever (which is technically true) which ends up being that stating Unreal supports something is a meaningless statement.

An example... I wouldn't say my fridge supports heating up food too, which could technically be true, but it would require me to insert a microwave into it in order to do so.

[late vector]

waiit so what the difference between doing OnRep and call a Multicast function instead here ?

[nocturne quail]

i would use the compressor of Your fridge to heat up the food 😄

[sinful tree]

We're not making heat pumps here dammit.. I'm using a plugin.

[quasi tide]

Ohhhh - Datura cursed

https://tenor.com/baZvE.gif

[sinful tree]

Ehh.. it's like the 5th time I've said damn (now 6th)

😛

[nocturne quail]

onrep is more reliable, in terms of if later clients joined they will automatically be getting the previous updates , and many more benifits

in short, multicast is when you call it, dies forever after execution, but on rep will be re executed if needed automatically

[late vector]

ohhhh yeah make sense tbh !

[sinful tree]

Multicast = Fire and Forget. OnRep = It can trigger the function whenever the client receives an update.

[dense sleet]

I'm wondering what the difference is between launching a client with "-server" (where it will run headless) vs building a dedicated server via engine source?

I see official unreal documentation suggesting you must build from source + have a custom engine version built from source (mostly going off the setting up dedicated server for Lyra example - https://dev.epicgames.com/documentation/en-us/unreal-engine/setting-up-dedicated-servers-in-unreal-engine) but it's not ideal to need to maintain a custom engine build for my whole team, when I'm not actually modifying the base binaries at all.

I assume when you launch with "-server" it's effectively just a headless client... But if so, then what makes the dedicated server so different?

[late vector]

Alrightttt, thanks for your time and help. I think it's time for me to just dive in and start trying things out...

[nocturne quail]

another example:
you need to fire weapon using multicast, but equp weapon using onrep

if you equp weapon using multicast, new player will see you without weapon they will not be notified you have equipped weapon when they were not in the match

[exotic wasp]

dedicated server build let's you strip things from it

[late vector]

yeah yeah so like onRep to check if a player have any grab item right now, right ?

I mean to update

[exotic wasp]

a client build running headless is the same size and code as a normal client

[nocturne quail]

it can update anything that needs to be

[nocturne quail]

if you destroy vehicle tire and new player join, they need to know that vehicle tyre is destroyed, so onrep will notify them if you destroy the car tire using onrep

[sinful tree]

What they may not need to see is the tire popping animation... Which would be a multicast.

[prisma snow]

Agreed partially and I really like the system but it's tricky to understand initially

[dense sleet]

Is this the only difference? I understand closer to shipping this is absolutely worth the trouble, but does a server build otherwise function the same as a client running headless?
In regard to how information is handled

[nocturne quail]

dedicated server supports splitting server code out of client build too for security reason, but this is a very difficult task becasue sometime server code should be existed on client too

[dark edge]

speak for yourself

I don't know how anyone would make a heat pump mechanic fun but I'd be open to suggestions

[sinful tree]

Just get the microwave plugin. 😛

[dark edge]

I think air simulation can be fun for vehicles but not air conditioning lol

[nocturne quail]

forcing this enum to be non bp type will optimize it? because i'm thinking to have a seperate base C++ class for each weapon , and all these classes will be derived from another master weapon base class.
so i can set this type in c++.

[dark edge]

Why would you have an enum for weapon types?

What makes an AKM different from an AUG?

[nocturne quail]

each weapon has unique animations

[dark edge]

Different mesh, different stats, different animations, sure sounds like they should be data assets to me

[nocturne quail]

i want to blend their offsets using the type

[nocturne quail]

like character picks akm, it will notify the animbp class to start blending the akm animations

[dark edge]

So why the enum?

why not pick up akm -> get anim set from data asset or data table entry -> blend that anim set

Seems like you're spreading the definition of your weapons out all over the place

So adding a weapon = add 1 data asset, instead of modifying stuff all over the place

[nocturne quail]

You mean each weapon data asset will have the BlendSpace asset and it will be passed from character to the character anim bp to blend?

its not one or two animation, each weapon comes with a full set of blend spaces, aim offsets for all poses crouch, prone, stand

[nocturne quail]

this is akm

if the type akm matches, the animation start to blend these

[exotic wasp]

holy shit heat pump

space station 13 has really "fun" atmospherics simulation

[dark edge]

I'm not doing anything with condensation etc that you'd need for a heat pump but you could probably technically make one with what I have. It's more for engine plumbing. Turbo/supercharger/intercooler/intake type stuff

Basically instead of an engine being a thing that reads throttle input and RPM and makes torque, it eats air based on throttle and RPM to make the torque, based on first principles stuff

[exotic wasp]

that's really cool

definitely complex if you account for the rest of the vehicle

[late vector]

Sooo it's me again and I try to implement what you say but as you can see on the video it works in "local" but struggle to keep in sync in the other client. In the video it's (client 1 and client 2) and not (server and client 1).

Here my code, I try to add in the server like Multicast or OnRep but everytime, the clients see the object lag like crazy

Input -> CheckGrab

CheckGrab()
{
    Grab(GrabLookAtComponent, lPos, lRot, CustomPlayerCharacter->PhysicsHandle);
    if (!CustomPlayerCharacter->HasAuthority()) Server_Grab(GrabLookAtComponent, lPos, lRot, CustomPlayerCharacter->PhysicsHandle);
}

void UInteractionComponent::Server_Grab_Implementation(UPrimitiveComponent* Component, FVector Pos, FRotator Rot, UPhysicsHandleComponent* PhysicsHandle)
{
    Grab(Component, Pos, Rot, PhysicsHandle);
}

void UInteractionComponent::Grab(UPrimitiveComponent* Component, FVector Pos, FRotator Rot, UPhysicsHandleComponent* PhysicsHandle)
{
    if (!CustomPlayerCharacter || !Component) return;

    if (Component) PhysicsHandle->GrabComponentAtLocationWithRotation(Component, TEXT("None"), Pos, Rot);
    CustomPlayerCharacter->GrabHandle->SetWorldLocationAndRotation(Pos, Rot);
}

[thin stratus]

Post code properly formatted please:

```cpp
Code here
```

[late vector]

oh shit yeah

[heady finch]

I am changing the mesh of a static mesh component on actor through an Interaction interface message that is called through a Server event, then in the item I interacted with I call a reliable multicast event that changes the value of a rep_notify bool that then changes the mesh on the component. The problem I'm facing is that if the server is too far away or is not looking at the actor the client is interacting with, then the server won't see any changes.

[thin stratus]

btw, UObjects and childclasses of those should be checked via IsValid(..)

Cause they can be != nullptr but pending kill

[thin stratus]

Video doesn't really show that something is going wrong, or what exactly do you mean?

Grabbing them on both sides and having physics run on both sides is never gonna stay in sync.

You won't get around only doing that on the Server and replicating the Transform back + smoothing it.

[late vector]

Ohhh okay I didn't know I will fix that ! thanks

[thin stratus]

Or, if only one client can interact with it at a time, do it locally and tell the Server the Transform, then have it replicate, skipping the owner and smoothed on everyone else.

[late vector]

well when client 1 grab and move around at a high speed, client 2 see the object lag sometimes (when there is a spike)

[thin stratus]

I assume you set the Actor to Replicated and ReplicateMovement

That's just gonna apply the Transform directly when it is received

Again, you won't get around writing smoothing code for it

[late vector]

Yeah of course

[late vector]

mmhh so what you suggest exactly ? 😅

Like doing it in local and then send the transform to the server ?

or humm idk

and disable like gravity I believe ?

[nocturne quail]

try this

UPROPERTY(ReplicatedUsing=OnRep_GrabbedObject)
UPrimitiveComponent* GrabbedComponent;

UPROPERTY(Replicated)
FVector GrabbedLocation;

UPROPERTY(Replicated)
FRotator GrabbedRotation;

UFUNCTION()
void OnRep_GrabbedObject();
void UInteractionComponent::OnRep_GrabbedObject()
{
    if (GrabbedComponent)
    {
        // Set the position and rotation on the client side
        GrabbedComponent->SetWorldLocation(GrabbedLocation);
        GrabbedComponent->SetWorldRotation(GrabbedRotation);
    }
}
void UInteractionComponent::Server_Grab_Implementation(UPrimitiveComponent* Component, FVector Pos, FRotator Rot, UPhysicsHandleComponent* PhysicsHandle)
{
    Grab(Component, Pos, Rot, PhysicsHandle);

    // Update the replicated properties
    GrabbedComponent = Component;
    GrabbedLocation = Pos;
    GrabbedRotation = Rot;
}

[late vector]

I tried doing something similar but it didn't work. Let me retry with that again

[nocturne quail]

maybe you will also need to call this in your server function to ensure proper replication

    // Mark the properties as dirty to ensure they replicate
    MarkPackageDirty();

[kindred widget]

Lol?

What in the fuck? Are you throwing around GPT stuff again? That's not how you mark properties for replication even using a pushmodel?

[nocturne quail]

that function they will implement themself according to their data

[marble magnet]

bump on this ?

[late vector]

well I tried what you say and now they are async but with lag ahah like client 1 is the same as client 2 with lag

maybe I should also add that I do that LMAO :

[nocturne quail]

try interp for setting transforms, it will add smoothing

[late vector]

Begin Play:
GetWorld()->GetTimerManager().SetTimer(InteractionLoopTimer, this, &UInteractionComponent::CheckLookAt, 0.05f, true);


void UInteractionComponent::GrabUpdatePhysicTransform()
{
    if (CustomPlayerCharacter && IsValid(CustomPlayerCharacter->PhysicsHandle->GetGrabbedComponent()))
    {
        CustomPlayerCharacter->PhysicsHandle->SetTargetLocation(CustomPlayerCharacter->GrabHandle->GetComponentLocation());
    }
}

[late vector]

mhhh I maybe can but I think its more of a replication issue like coding issue

[prisma snow]

Would be cool to keep track of the growing percentage of nonse from LLMs arriving to this chats

[late vector]

(BTW: I do that in client, try to do it on server, onrep, multicast etc.. doesn't work i'm so dumb bro)

[kindred widget]

Bruh I always mark my packages dirty to replicate them. Like I just send the client the entire game, they don't even need to install it.

[nocturne quail]

disable colliison on moving objects, they may interact with character physics/collision and making bad behaviour

[prisma snow]

LOL

It's like when I was writing a faster version of project world to screen and ChatGPT's test was 10x slower than the original xD

[late vector]

also try it before and retry it right now except a crash same issue

i start to think my system is just broken as f

[nocturne quail]

also you don't need to simulate physics if you are directly playing with world transforms

[late vector]

the issue is that if I set simulate physics on false after grab it, it won't work because the PhysicsHandle need it

maybe I should just make my own physicsHandle without simulate physics

[nocturne quail]

or try to decrease the physics update rate

i think gravity = 0.05 will fix that lag issue, try it 😄

[late vector]

it is the project settings or idk lmao

like the physics update rate first

nvm I find it

[nocturne quail]

there are two functions you can controll this in code

primitive->setlineardamping
primitive->setangulardamping

[late vector]

I try to increase, decrease everything same the lag is still here

I think i will eat unreal engine

[nocturne quail]

try standalone game with two clients

[late vector]

well in standalone it work like a charm

[nocturne quail]

you selected this one?

[thin stratus]

That's wrong, please don't do that.

[late vector]

ah no I did not do that

let me try

[nocturne quail]

this will start a real dedicated server and connect 2 clients to it

[thin stratus]

@late vector The usual way of smoothing stuff like this is to interpolate from LastReceivedData to NewReceiveData over time (Tick).

[late vector]

well yeah same issue

and WTF MY ITEMS FLYING

[thin stratus]

With the small addition that if data replicates faster than you interpolate, your "LastReceivedData" would probably want to be the CurrentTransform instead, as you otherwise suddenly jump the remaining part.

[late vector]

mmmhh let me try

[thin stratus]

tbh, this isn't much of "let me try" thing. At least not in the sense that you should throw 5 lines of code together and click some checkboxes.

You need to learn how to do this and then implement it.

[late vector]

wait idk you tell me I should interp last and new in tick its not what you say ?

[thin stratus]

If only one player can move the Actor at a time, send the Transform to the Server on Tick (not reliable) and have the Server set it to an OnRep variable.
If more than one player can move the Actor, that you'd need to handle that a bit different, but it would still set the OnRep variable on the Server.

Whenever the OnRep function calls for the Transform, you need to update the variables you are interpolating.

If only one player can move the Actor at a time, then you'd probably want to skip that one locally in the OnRep. This can be done in a smart way by setting the Owner of the Actor to that player and having the OnRep variable set to SkipOwner. Otherwise, you'll need to somehow keep track of who is holding that Actor and then check that "Holder" if they are local.
If more than one player can move the Actor, you don't need to skip anyone I guess.

The smoothing would also happen on Tick, no RPC though of course.
You always smooth from a "Last State" to a "New State" over time. How fast is something you need to configure to fit what you think looks best.

The smoothing can be done with a simple Interpolate on the Location and the Rotator.

The "Last State" and the "New State" are usually updated whenever the OnRep calls (which isn't every frame, hence the requirement for Smoothing).

The OnRep usually always counts as the "New State". The "Last State" will be whatever the current Transform is (the latest result of your smoothing).

There is a case where data doesn't come in fast enough, which means the Actor would stand stationary for a bit before continueing. You could "extrapolate" in those cases.

Extrapolation in this case is also relatively simple. If you consider the difference between Last State and New State to be 0 to 1.0 (percent), then you can always calculate the current state with CurrentState = LastState + DeltaState * Percent. Percent being the thing you slowly update over time from 0 to 1.0 over and over.

E.g. if you smooth from (100, 0, 0) to (150, 0, 0), then DeltaState would be (50, 0, 0). Now if you didn't get any data in time, if you just increase Percent further, beyond the 1.0, e.g. to 1.2, you would move to (160, 0, 0) ultimately. That's a gample though, cause it could be that the movement didn't continue in that direction.

After that you start tuning numbers.

Smoothing is simple, in theory, but you gotta code it after all.
Also if you do it this way, turn of "Replicated Movement", cause that will fight your locally smoothed value as well as the Client's predicted movement.

[late vector]

HOLY COW ! I understand now what you mean lmao.. yeah that not that simple indeed... How the f you even come with that 😭

Alright so It's very late where I live, I think I will try to do that tomorrow, and its a big try ahah but anyway thanks you A LOT for that ! I think I can do it with a lot of time !

(Also thanks Missty for you help ! ahah even if it didn't work)

lmao it's like the bottom of iceberg what you say here ahah

[lament flax]

sometimes i feel like I should pay to get answers in this discord lol

[crisp shard]

if i have a server event that changes a widget entry and it's relevant for anyone that would be looking at it, i know i could just send this updated info upon widget creation etc, but im curious if i want all potential viewers of that widget to get the update in real time, would i do a multicast and do the logic like that on the clients? or would i do the multicast plus a owning client ? i might be just recursively thinking but im not sure if the multicast would be best or if i should do the owning client event on a multicast (or repnotfiy now that i think about it)

[sinful tree]

If you want the change to be visible to anyone who may eventually have the widget relevant at any point in the future, then OnRep is what you want.

[pallid wyvern]

Hello fellow Devs, I came across a problem while creating a multiplayer game. It's PvPvE, and I need navigation system for Bots. As the map is big, baking the navmesh increases the build size y approx. 200mb. While this is fine for server, I don't want this for clients because it increase map load time for client as well.

Any leads on how I can avoid packaging map without nav mesh data? I am using world composition and world partition does not support baked lights.

[slim jay]

Hey, can't find any sources about advanced steam session ban system and AI gives overly complex solutions. Do you know any source about this?

[glossy wigeon]

If I want to have a quest system, and I want players to be able to accept quests, from a quest giver, I would need a Quest giver component, that replicates the offered quests and also have a "quest taker" component on the player. Since the player won't be able to call a RPC on the quest giver component.

So basically, the quest taker component will take a reference of the quest from the quest giver, and send a RPC.
This way we ensure, that the quest is taken by a player with authority over it, so that player A can't accept quests for player B.

Do I get the concept right, regarding validation and in general about authority?

[dark parcel]

Client make request to server by sending server RPC. ofc you have to do this in an actor owned by the client.
The server simply answer back to the client.

When you do a server rpc you are asking a server to Run a function on it's machine.
And ofc being in the server machine, you can get any object that lives in the server's instance.

---

Player ServerRPC -> Request to take Quest (id)
Server run the function -> Get the relevant component / w.e you need. Check if the quest can be taken. -> If can be taken, add the quest to the relevant player.
If you make the quest data a variable, you can use OnRep on the client to react the server's answer.

[serene copper]

Hi, I am not sure if this is the right place to ask, but here I go. I am currently having trouble replicating when the client reaches 0 hitpoints and "revives" back to 25% health. The players can upgrade a skill that lets them do that before the game. If the host does that, it works for the host, but the client somehow also revives even though they did not level that.
If the client leveled that, then the client does not revive. Anyone got an idea? I am kind of starting to lose my mind.

I have done the logic before a server event, in the server event and also in the multicast after the server event 😄 Help will be appreciated a lot.

Current attempt looks like this. The "Bonus receive level" integer comes from the save game that gets loaded on begin play. If value 0 no revive, if value 1 the player revives.

[dark parcel]

Why is restoring health not something that server exclusively run.

[serene copper]

I think I also already had it in the Server event

[dark parcel]

why are you calculating both in client and server?

client shouldn't have a say at all when it comes to restoring HP or dealing damage.

[serene copper]

You mean this one should not be before?

So rather like this?

[bright summit]

I will hookup onto this. So if I want to receeive the server data about object transform this is the way?

[glossy wigeon]

Yea I think that's the point at which I'm right now.

Coming from a web environment, it took my some time to wrap my head around, how "validation" is done.
Because you don't have a Client A made the request in a RPC, but you get this implicitly, because a client can only call an RPC on a component it owns.
That's how you do this kind of validation right, by knowing the Client owns this component, so it's him.

Also it means if a client want's to make a RPC, there need's to be something owned by the client, that can make RPC's.
Which in turn means for things like quest giving and taking, one most likely want's a pair of components, one that offers the Quest (owned by server) and one that takes the quest (owned by the client and communicates via a RPC on the server with the offer component).

It's a more conceptual question I know 😅

[outer birch]

Anyone done a transition to Iris? Been trying to figure out a Error: UNetSubObjectFactory::InstantiateNetObjectFromHeader issue. The actor in question has
bReplicateUsingRegisteredSubObjectList = true;
and after constructing the object, I do:
AddReplicatedSubObject(HealthSet); but still the client is unable to resolve the NetRefHandle

[quasi tide]

I know @thin stratus is doing it and @hollow eagle has been using it for awhile. @nova wasp as well. (sorry for the pings!)

[hollow eagle]

If this is a UObject (as opposed to a component or something else that already supports replication), make sure you've overridden RegisterReplicationFragments/IsSupportedForNetworking/GetFunctionCallspace/etc

Other than that, what you've posted is generally all you need

[outer birch]

It's a healthset, which is a subclass of UAttributeSet, which does implement RegisterReplicationFragments, do i need to do it for every subclass?

[hollow eagle]

no, you shouldn't

We're also using GAS, I don't think we've done anything special for attribute sets

not even AddReplicatedSubObject - I think the ASC already does that

[outer birch]

hmm

[hollow eagle]

yeah it does, in UAbilitySystemComponent::AddSpawnedAttribute

[outer birch]

weird, we're not calling any variant of htat

i wonder how our attributes are being found by the ASC

[hollow eagle]

AddAttributeSetSubobject maybe? It calls that too

[outer birch]

we do do that for other places but not this one, weird

anyway, thats prolly a different mystery this object should be replicating it correctly

got any good suggestions for extra ensures/error detection stuff

[hollow eagle]

Unfortunately not really, those errors are usually pretty simple in that the object just isn't replicating. Though the full error message may have more detail.

[outer birch]

LogIris: Error: UNetSubObjectFactory::InstantiateNetObjectFromHeader NetRefHandle (Id=1055):(RepSystemId=?): Failed to find static or stable name object referenced by SubObject: [NetRefHandle (Id=9):(RepSystemId=?)]/MapsDevelopers/UEDPIE_1_L_CombatGym.[NetRefHandle (Id=7):(RepSystemId=?)]L_CombatGym.[NetRefHandle (Id=5):(RepSystemId=?)]PersistentLevel.[NetRefHandle (Id=81):(RepSystemId=?)]B_DL_Pillar_C_2.[NetRefHandle (Id=1055):(RepSystemId=?)]PropHealthSet, Owner: RootObject B_DL_Pillar_C_2 (InternalIndex: 104) (NetRefHandle (Id=81):(RepSystemId=1)), RootObject: /MapsDevelopers/UEDPIE_1_L_CombatGym.L_CombatGym:PersistentLevel.B_DL_Pillar_C_2

[hollow eagle]

hmm

it seems to think the object has a static name? that's weird if you're instantiating it at runtime

it's trying to netmap the object - i.e. the object already exists on both client and server, it's just linking them together

[outer birch]

its a non default subobject thats being created at runtime optionally to reduce overhead

[hollow eagle]

that makes sense, the question is why it thinks such an object would be statically named

[outer birch]

[[2025.05.22-14.11.04:464][231]]LogIrisBridge: Verbose: ObjectReplicationBridge(0)::StartReplicatingNetObject Created NetRefHandle (Id=1055):(RepSystemId=0) with ProtocolId:0xc65decb for Object named PropHealthSet [[2025.05.22-14.11.04:464][231]]LogIrisBridge: Verbose: ReplicationBridge(0)::InternalAttachInstanceToNetHandle Attached: PropHealthSet NetRefHandle (Id=1055):(RepSystemId=0) to (InternalIndex: 193) [[2025.05.22-14.11.04:464][231]]LogIrisBridge: Verbose: ObjectReplicationBridge(0)::StartReplicatingSubObject Added SubObject PropHealthSet (InternalIndex: 193) (NetRefHandle (Id=1055):(RepSystemId=0)) tied to RootObject B_DL_Pillar_C_2 (InternalIndex: 42) (NetRefHandle (Id=81):(RepSystemId=0)) RelativeToSubObjectHandle NetObject None (InternalIndex: None) (NetRefHandle (Id=0):(RepSystemId=?)) [[2025.05.22-14.11.04:499][231]]LogIrisBridge: Verbose: ObjectReplicationBridge(0)::NetFlushDormantObject: RootObject B_DL_Pillar_C_2 (InternalIndex: 42) (NetRefHandle (Id=81):(RepSystemId=0)) [[2025.05.22-14.11.04:510][231]]LogIrisBridge: VeryVerbose: WriteNetRefHandleCreationInfo (new) for NetRefHandle (Id=1055):(RepSystemId=0) : Factory: NetSubObjectFactory_0 (id:1) Header: FNetStaticSubObjectCreationHeader (ProtocolId:0xc65decb):

these are the earlier logs indicating im registering it/whatever

[hollow eagle]

are you overriding IsNameStableForNetworking or similar?

looks like UAttributeSet has an extra condition it returns true for that, I think this should be false for your case.

[outer birch]

not exactly, but
void UAttributeSet::SetNetAddressable()
{
bNetAddressable = true;
}

and we set it true

[hollow eagle]

that might be the problem?

I don't think we set that anywhere

it's making iris think the object has a stable name when it doesn't

[outer birch]

hmmm, maybe a holdover from when it was a default subobject

it does have a stable name though, but maybe that means something different in iris land

heh, apparently there's some bullshit magic for how it registers,
// Create attribute sets (they need to be made before the ASC is initialized, or we have to manually register them)

[outer birch]

success!

legacy system hides many sins

[hollow eagle]

Having a stable name doesn't differ in iris vs legacy, but if you aren't also creating the object on the client with that same name it won't work.

[outer birch]

yeah that makes perfect sense

which would have been the case back it was created by default

[quasi tide]

Siliex - while you're here, I know you've been using Iris for quite some time now. Has you opinion on it changed at all since sometime ago when you listed out the highlights? Like is it still an overall improvement?

[outer birch]

thanks for knowing the people in the know

[glossy wigeon]

Is this right, if I have a component living on the PlayerState, I can't call an RPC on that component, since the PlayerState is not owned by the Client by default?

[sinful tree]

You can call an RPC to the server on the client whose playerstate it is and so long as that component is replicated.

[glossy wigeon]

Hmm then I need to investigate further, when I try to call the rpc from a widgets button click, I get thrown onto the default map 😦

[dark edge]

Show your code

[glossy wigeon]

Not sure if it's a code issue though 😅

// Header file
UFUNCTION(BlueprintCallable, Category=Quest)
void AcceptQuest();
    
UFUNCTION(Server, Reliable, WithValidation)
void ServerAcceptQuest();

// Cpp file
void UQuestComponent::AcceptQuest()
{
    if (!GetOwner()->HasAuthority())
    {
        ServerAcceptQuest();
    }
}

void UQuestComponent::ServerAcceptQuest_Implementation()
{
    bool abc = true;
}

bool UQuestComponent::ServerAcceptQuest_Validate()
{
    return true;
}

nvm I think I'm doing something else wrong

I really don't understand it, I feel like the component got tainted, because it got passed through widgets, or because it is not directly accessed from the player state but is a variable on a widget?

[sinful tree]

I don't think this is the relevant code. If you're getting pushed to the default map, that's an indication that you may be doing some other RPC with validation. Always returning true in this implementation means that it shouldn't be kicking you out.

[glossy wigeon]

Yea when accessing the Quest Component directly from the PlayerState of the widgets owner, it works.
Before I was using the same component, that was referenced as variable on a parent widget.

Not sure if this is really an issue, just wondering.

[proper vine]

Dealing with an issue where the client is unable to crouch. The player on the server is able to crouch and replicate crouching, but when I hit the crouch button as a client, they don't crouch, even though the function is getting called. It's as if the client isn't triggering the server side function. bIsCrouched always comes out as false for clients in the Validate function
https://i.gyazo.com/7043915a3eae0d98c7611296193df58a.png
https://i.gyazo.com/53c68f291fc50ff848d51d921990caab.png

[nocturne quail]

crouch and prone system

discord breaked the formatting, fix it yourself

[proper vine]

uhhh, ok. I'm kinda new to networking, I figured the problem might just a missing call or bad syntax or something, but I'll give that a try, thanks

[exotic wasp]

if you set a breakpoint in the SetPlayerCrouch function, does it get hit when a client crouches?

[proper vine]

I had some print statements that were successfully called by both the server and client players, and the names were correct, so I figured SetPlayerCrouch was successfully getting called by the client player as well. lemme try the breakpoint

[exotic wasp]

also, is current value ever what you expect it to be?

[proper vine]

current value? you mean bIsCrouched? it only is what I expect it to be for the server player, not the client player

[sinful tree]

If this class is based on the built-in character class, you shouldn't be RPCing the crouch, it's built into the CMC to be client predicted.

[proper vine]

so should I change SetPlayerCrouch to not be a server function and just call Crouch/UnCrouch in the OnCrouch function?

[sinful tree]

Yep

[proper vine]

that fixed it! Thank you!

[dark parcel]

No, why does the client need to make the rpc through the component? It can make a request from w.e it own. Including player controller or character.

Again client call the server rpc. It can be anywhere the client owns.

Read the pinned channel a dozen of times till it click.

You are still thinking of running the rpc in the same component on something the client potentially don't have ownership of.

[glossy wigeon]

Thats not at all what I wrote, but I think we are talking about the same thing, all good and thanks! 🙂

[dark edge]

Anyone have any idea why simulating physics on a replicated actor with an attached actor, seems to break the attachment? This all works on the listen server and I'm pretty sure it worked in UE4, but I can't really sort out WTF is going on now.

A is attached to B, then A has simulate physics turned on. Both replicate and have replicate movement turned on.
Expected: A and B fall as 1 body just like in single player.
Actual: A falls on client just the same as on server, but B acts as if it doesn't have an attach offset from A, and tries to be at the exact same location.

[dark edge]

It appears that the attach parent get cleared when it starts simulating physics

[meager heart]

Is there a way to force a UObject RepNotify parameter to broadcast its update to clients? If I just set its value to itself it'll call repnotify on the server but won't send that to the clients

• I'm trying to do this because I've changed some of the properties of the object at this point, but those aren't detected by normal repnotify so I'm basically trying to force it to update on clients

[regal narwhal]

Hi there. I m experienceing an issue while implementing a multiplayer interaction system. I have two components, one is InteractionComponent which grantAbility and the other is InteractableComponent which store InteractionConfig like GameplayAbility etc. I m trying to make it multiplayer works. so InteractionComponent is works locally. but I m having an issue since GrantingAbility should happen on the Serverside. I ve fix that but when I grantAbility, I need AbiltySpecHandle on the client to show some info. Since it doesnt work async I m stucked a bit. can someone maybe give any suggestion how to fix that issue

Would that make sense to convert GrantAbility with TFuture Tpromise? I m not sure if it works in mp btw

[lament flax]

Could you send a screenshot of the details panel with the var selected ?

But the ideal flow is to make the variables inside the upbject replicate

[pseudo wagon]

Rep notify will not call on the UObject but on the property of UObject, You should use that rep notify.

[dark edge]

Turns out I hadn't replicated the scene components

[lost inlet]

if this is an asset on disk that you are modifying properties of, then all that gets replicated is the path to the asset.

if you want a UObject that gets its properties replicated then you'll be looking at subobject replication

[river shore]

Hello guys!
I am trying to attach a weapon to my character mesh. The player who picks up the gun should attach it not to the third-person mesh but to the first-person mesh. But since I am handling the attachment through the server. it replicates the attachment to the local client, which attaches it to the third-person mesh for every player. Is there a way around so I can attach the weapon to the first-person mesh just for the client that picks up the weapon? Thx for any help ❤️

// Gets excecuted by the server
void ADefaultCharacter::AttachWeapon(ABaseItem* Weapon)
{
        // Marked replicated
    AttachWeapon = Weapon;
    OnRep_AttachWeapon();
}

void ADefaultCharacter::OnRep_AttachWeapon()
{
    if(!AttachWeapon || !GetMesh()->DoesSocketExist(AttachWeapon->SocketName)) return;
    
    if(IsLocallyControlled())
    {
                // Should attach to the first person mesh
        AttachWeapon->SetActorHiddenInGame(false);
        AttachWeapon->AttachToComponent(AttachArms, FAttachmentTransformRules::SnapToTargetIncludingScale, AttachWeapon->SocketName);
        GEngine->AddOnScreenDebugMessage(-1, 50.f, FColor::Red, "Is locally controlled");
    }
    else
    {
                // Should attach to the third person mesh
        GEngine->AddOnScreenDebugMessage(-1, 50.f, FColor::Green, "Is not locally controlled");
        AttachWeapon->AttachToComponent(GetMesh(), FAttachmentTransformRules::SnapToTargetIncludingScale, AttachWeapon->SocketName);
        AttachWeapon->SetActorHiddenInGame(false);
    }
}```

[dark edge]

You'd probably have to override OnRep_Attachment to handle it I'd guess

[river shore]

I already thought about it, but thought there might be a way I am missing...Thx though! I will probably do that then

[dark edge]

Maybe see if you can trick it by using a non replicated property

Attach to SkeletalMeshComponent* MeshToAttachTo which would be = the different meshes on different machines.

[river shore]

Oh

[dark edge]

Wouldn't hurt to try that

[river shore]

kinda unsafe but I guess the other way is so too, thx a lot for the idea

[river shore]

true

[dark edge]

If that doesn't work I'd maybe see about just swapping the mesh you use based on machine instead of hiding showing one of the 2 mesh components.

[river shore]

yeah, probably will try this first, seems the most safe ig...

[dark edge]

Is Character.Mesh a replicated component by default?

[river shore]

I do not know tbh

[nocturne quail]

What else I can do or change in this system to make it more robust and fast and optimized while keeping the security?

void ASurviveCharacterBase::ServerManageMovement_Implementation(const float NewValue, EMovmenetInputKey NewMovementKey)
{
    switch (CharacterPose)
    {
    case ECharacterPoseType::E_PoseStand: PredictedCharacterSpeed = CalculateStandPredictedCharacterSpeed(); break;
    case ECharacterPoseType::E_PoseCrouch: PredictedCharacterSpeed = CalculateCrouchPredictedCharacterSpeed(); break;
    case ECharacterPoseType::E_PoseProne: PredictedCharacterSpeed = CalculatePronePredictedCharacterSpeed(); break;
    case ECharacterPoseType::E_PoseJump: PredictedCharacterSpeed = CalculateJumpPredictedCharacterSpeed(); break;
    case ECharacterPoseType::E_PoseMax: break;
    default: break;
    }
    OnRep_PredictedCharacterSpeed();

    if(NewMovementKey == EMovmenetInputKey::EKeyForward)
    {
        ClientApplyForwardMovement(NewValue);
    }
    else if(NewMovementKey == EMovmenetInputKey::EKeyRight)
    {
        ClientApplyRightMovement(NewValue);
    }
}

void ASurviveCharacterBase::OnRep_PredictedCharacterSpeed()
{
    UCharacterMovementComponent* MovementComp = GetCharacterMovement();
    if (MovementComp)
    {
        MovementComp->MaxWalkSpeed = PredictedCharacterSpeed;
    }
}

void ASurviveCharacterBase::ClientApplyForwardMovement_Implementation(const float NewValue)
{
    if(!HasAuthority())
    AddMovementInput(Controller->GetControlRotation().Vector(), NewValue);
}

void ASurviveCharacterBase::ClientApplyRightMovement_Implementation(const float NewValue)
{
    if(!HasAuthority())
    AddMovementInput(FRotationMatrix(Controller->GetControlRotation()).GetScaledAxis(EAxis::Y), NewValue);
}

ofcourse i can pass the input key in ClientApplyto determine if client want to move f/r , but then it will cost... so i decided to have to seperate client rpc's

[sinful tree]

if this is using the character movement component, it's not properly predicted and will break with any amount of lag, and assuming ClientApplyForward/RightMovement_Implementation functions are RPCs to the client, the beginning of any movement inputs on a client will also feel extremely sluggish for all clients.

[nocturne quail]

it was very stuttering when i was using multicast to set the speed value, and using onrep it jitters very little on other clients but for the owning client it works fine.
is there any other technique to have the same server derived predicted movement without jitters?
yes I'm testing it with the default CMC

[sinful tree]

1. You have to create custom movement modes within the CMC itself so that they are inserted into the movelist with a timestamp so the server can validate when the speed changes took place. Look into any C++ tutorials on replicating sprinting with the CMC and that'll start getting you somewhere with it.
2. Use the AddMovementInput on the client directly from their input. Calling AddMovementInput directly on the client without the server telling them to do so is part of the prediction process allowing client input to execute immediately which calls into the CMC to then make whatever RPCs are necessary to have the character move.

[nocturne quail]

Thanks for this detailed useful input, i think it's not that tough task and i will be implementing it next few hours

[nocturne quail]

I have CustomMovementComponent derived from default CMC in my another project, i think i can just take it and override these stuff and add new modes

[sinful tree]

Yeah, you'll need to subclass the CMC to handle the new movement.

[nocturne quail]

and have to Assign my custom cmc in ACahracter since i don't want to have two CMC's in a character class, right?

seems like i started to have my custom engine 😄

[dark edge]

I have a bunch of relationships between things, what's the approach to maintaining that over the network?

Of course I can have a replicated array of connections, but how to handle the connection making it to the client before the things being connected do?

[trail olive]

Hello, I am in a bit of a predicament whether I should let the Client Initialize their Inventory Component or make the server do that for them. What do you guys think ?

[dark parcel]

Totally depend on the game.

If you reference an existing game it is probably clearer on what approach to take.

[meager heart]

Thanks! (and @pseudo wagon @lost inlet ), here's the details. Understood about what you've said to use the repnotify inside the actual object, it's in a DataAsset so I think it's a bit rough but I'm already replicating the fields of the DA so I'll see if I can broadcast an event from the repnotify within the primary DA and see if that works

Was hoping there was an easier way to just force a repnotify to fire, but that solution should be a bit cleaner so I'll check it out

[lost inlet]

it doesn't work like that

like I said, a reference to an asset is only going to be replicated as an actual path to the asset

and it's not a subobject of a replicated actor either

you should really treat data assets as immutable

[meager heart]

It's actually a dynamically constructed DA

Looks like this is working actually, if I bind to this event dispatcher in the client it'll get updated by the repnotify

[nocturne quail]

I finished it, and now i have a small issue where i the default cmc is not replaced with my cmc

    GetCharacterMovement()->DestroyComponent();
    SurviveMovementComponent = CreateDefaultSubobject<USurviveMovementComponent>(TEXT("SurviveMovementComponent"));
    SurviveMovementComponent->UpdatedComponent = RootComponent;

or i have to do it directly in the ACharacter class to extend it and don't have a sub class?

then engine will need to rebuild which is not an issue with an ssd, but it will need to update then lots of references

ah nvm, it works 😄

[rare cloud]

you need to do it this way, using the CDO constructor :

AMyPlayerCharacter::AMyPlayerCharacter(const FObjectInitializer& ObjectInitializer)
    : Super(ObjectInitializer.SetDefaultSubobjectClass<UMyCharacterMovementComponent>(ACharacter::CharacterMovementComponentName))
{

}

[nocturne quail]

This looks super clean, is there any good sides of using it or is it just a good practice?

[rare cloud]

it's the way you use a Custom CMC with a character, destroying something to create another one isn't a good practice imo 😄

[nocturne quail]

totally agree, that approach is a new thing for me to learn, Thanks for sharing the knowledge

[kindred widget]

You can also just set it in the editor.

[latent heart]

Anyone got experiences with level instanes and multiplayer? Can they work? Is it a hopeless endeavour?

[nocturne quail]

TPair supports network reflection?

I don't remember I used it before, but not sure if it was working or not

in 4.19 specifically

this is my usecase:

UENUM(BlueprintType)
enum class EMovementType : uint8
{
    Stand,
    Crouch
};

UENUM(BlueprintType)
enum class EMovementState : uint8
{
    Walk_Forward,
    Walk_Other,
    Jog_Forward,
    Jog_Backward
};

USTRUCT()
struct FMovementSpeed
{
    GENERATED_BODY()

public:
    TMap<TPair<EMovementType, EMovementState>, float> SpeedMap;

    FMovementSpeed()
    {
        SpeedMap.Add(TPair<EMovementType::Stand, EMovementState::Walk_Forward>, 175);
        SpeedMap.Add(TPair<EMovementType::Stand, EMovementState::Jog_Forward>, 500);
    }

    float GetSpeed(EMovementType Type, EMovementState State) const
    {
        const float* Speed = SpeedMap.Find(TPair<Type, State>());
        return Speed ? *Speed : 0.0f;
    }
};

void USurviveMovementComponent::PhysWalking(float deltaTime, int32 Iterations)
{
    Super::PhysWalking(deltaTime, Iterations);
    
    //GetSpeed(..., ...); will be used to set value of `MaxWalkSpeed`
}

[outer birch]

@hollow eagle did yall end up replacing GAMEPLAYATTRIBUTE_VALUE_SETTER usages with one that invaldiates the attributes?

trying to decide if i should do it there, or PostAttributeSet

[hollow eagle]

don't think so, you mean marking it dirty?

[outer birch]

yeah

[hollow eagle]

I think the ASC already does that but I need to check

yeah, in FGameplayAttribute::SetNumericValueChecked

[outer birch]

ah hot damn

[nocturne quail]

What im doing wrong?

[outer birch]

Probably not including a header that defines the enum

[nocturne quail]

Enums are directly defined in the top of this header file, the issue was somewhere else in the character class where i renamed the Enum class and forgot to update the params in the function which was passed in the character class
for some reason it resharper was showing this CustomMovementComponent class with errors

[humble idol]

I wonder if this has anything to do with defining operator== and GetTypeHash for your TPair.

[crisp shard]

would i treat event dispatchers similar to interface messages in terms of how to replicate them? essentially, do a server rpc before the call event ?

example might be, character interacting w widget, presses button, server rpc to character and then call the event from there on whatever actor / bp ?

[nocturne quail]

currently it stops complaining after fixing the function params, I think it will work when i will test it after fixing some more issues in other places of the project

[dark edge]

What's the simplest way to keep a bunch of connections between things in sync, where there's really no actors involved at all. At least not fundamentally.

I have a bunch of components which add entries to a system when they spawn and remove entries when they end play. I can make connections between entries by index.

Assuming I have some way to guarantee indices or whatever I use as the address to be deterministic and stable, I guess I could just have a replicated array Connections, and in its onrep, make the connection if both targets exist. Then targets just have to retrigger connections when they start existing and every replicated connection will be made as soon as the 2 targets and the connection record make it to the client

The more I think about it the more this subsystem stuff should just be in components on GameState so I can use replication lol

[nocturne quail]

Why i'm not getting circular dependency? 😄

ACharacter: #include "CharacterMovementComponent.h"
CustomMovementComponent: #include "CharacterMovementComponent.h"
AMyCharacter: #include "CustomMovementComponent.h"

[thin stratus]

Why would you?

[exotic wasp]

if the movement component included the character it would be circular

[nocturne quail]

because I'm including in the character header file

[thin stratus]

Where are you including it though

You only wrote the class, not the file extension

[nocturne quail]

this function is the part of the character and the parameter ECharacterMovementMode passed is the part of MyCustomMovementComponent

UFUNCTION(Server, Reliable, WithValidation) void ServerUpdateCharacterMovementMode(const ECharacterMovementMode NewCharacterMovementMode);

[nocturne quail]

if not include, it won't compile

and that forward declaration not working, not sure why resharper is suggesting it 😄

[thin stratus]

Circular Dependency happens if you include A in B and B in A, either directly or over a chain.

[thin stratus]

Don't put structs and enums into headers that you don't want to include when you need the structs and enums.

/// Bad, unless you'd always include CustomMovementComponent.h anyway, and even then it's not good.
/// As soon as you only need the enum, or generally types of that header, it's bad.
CustomMovementComponent.h
  - Declare ECharacterMovementMode

MyCharacter.h
  - Include CustomMovementComponent.h

/// Good

CharacterMovementTypes.h
  - Declare ECharacterMovementMode

CustomMovementComponent.h
  - Include CharacterMovementTypes.h

MyCharacter.h
  - Include CharacterMovementTypes.h

And forward declare whenever you can.

[nocturne quail]

Perfect solution, but is it safe in terms of reverse engineering?
I mean if a hacker want to find those struct/enums to determine how is the main structure working....
finding data in a big compiled class can be difficult for them i guess but not sure 😄

[exotic wasp]

if you are just using a type to return, pass by reference, or by pointer, you can usually forward include

[exotic wasp]

a hacker will always win 🙂

[dark edge]

How many hackers are you expecting for your solo dev game lol

[nocturne quail]

not always, their journey is gonna end soon once nvidia released their AI anti hacks

[exotic wasp]

if you really care about people reverse engineering your game, obfuscate it

[modern garnet]

The responses aren't wrong about separating common types out into their own headers but I am pretty sure the problem with your snippet is the syntax isn't correct and not an include issue.

[exotic wasp]

just wait until ai hacks come out to counter the ai antihacks...

its an infinite arms race

the best you can do as a developer is make your game unexploitable. anything more is an utter waste of time

and also, who cares unless its a competitive game?

[modern garnet]

TPair is an alias for TTuple which has an implementation of GetTypeHash.

[humble idol]

Ah okay, that's good to know!

[nocturne quail]

it was already solved, the issue was in character class

[nocturne quail]

    float GetSpeed(EMovementType Type, EMovementState State) const
    {
        const float* Speed = SpeedMap.Find(TPair<EMovementType, EMovementState>(Type, State));
        return Speed ? *Speed : 0.0f;
    }

[thin stratus]

What makes you think that any of your code is safe?

Pretty wild assumption that having those types in any other files would make it any more or less safe.

Your game can be reverse engineered. Just live with that.

People have been reverse engineering WoW Servers.

[nocturne quail]

I asked cause I was expecting that U will recommend some strong hashing/encryption techniques )

[thin stratus]

I will recommend you to stop wasting your time on stuff like this and focus on the game itself.

[nocturne quail]

people were saying to me md5 is not that much hard decrypt

[thin stratus]

Missty. Your game will be decompiled, reverse engineered, hacked and cracked. Stop wasting your time on it.

The big companies fail at it, so will you.

It's a lost cause.

Make sure the game feels good and is fun to play. The people will buy it and enjoy playing it.

Everything else doesn't matter.

The ones who pirate it would most likely never have bought it.

It's not important.

The ones that hack will be found anyway.

And if you really end up with a popular game that suffers from cheating, then you can still solve that after the fact.

[nocturne quail]

Great constructive advice, Thanks

[thin stratus]

Yeah, it probably sounds harsh, but it is what it is.

Look at Nintendo that has huge problems with their Console being reverse engineered. Their N64 games decompiled, etc.

What are you going to do that such a company hasn't already tried?

The only thing they can still do is write some bullshit TOS that locks your Switch 2 if you do something with it that they don't want, even though you paid for the damn hardware.

[proper vine]

trying to fix a UObject not replicating properly to the clients. I have a custom UObject, HitData, that is failing to replicate on the client.
I know UObjects don't normally replicate without some extra effort, I followed this documentation to a T and it's still not working: https://dev.epicgames.com/documentation/en-us/unreal-engine/replicating-uobjects-in-unreal-engine
HitData's UPROPERTY is set to replicate, it's in DOREPLIFETIME in GetLifetimeReplicatedProps, I've overriden ReplicateSubObjects, the UObject has GetLifetimeReplicateProps and has overriden IsSupportedForNetworking, and still the client does not get the replicated UObject. The server receives the data just fine.

The code's a little long so I'll do my best to explain what's going on:

AGameTypeBase::AGameTypeBase()
    : Super()
{
    GameData = CreateDefaultSubobject<UGameTypeDataComponentBase>("Gamedata");
    HitData = CreateDefaultSubobject<UHitInfo_DataAsset>("HitData");

    // Set this actor to call Tick() every frame.  You can turn this off to improve performance if you don't need it.
    PrimaryActorTick.bCanEverTick = true;
    bReplicates = true;
    bAlwaysRelevant = true;
    bReplicateUsingRegisteredSubObjectList = true;
    GameData->SetNetAddressable();
    GameData->SetIsReplicated(true);

    if (IsValid(HitData))
    {
        RemoveReplicatedSubObject(HitData);
    }
    HitData = NewObject<UHitInfo_DataAsset>();
    // MyActorSubobject becomes a replicated subobject here 
    AddReplicatedSubObject(HitData);

}

void AGameTypeBase::GetLifetimeReplicatedProps(TArray<FLifetimeProperty>& OutLifetimeProps) const
{
    Super::GetLifetimeReplicatedProps(OutLifetimeProps);
    DOREPLIFETIME(AGameTypeBase, GameData);
    DOREPLIFETIME(AGameTypeBase, bReady);
    DOREPLIFETIME(AGameTypeBase, HitData);
}

bool AGameTypeBase::ReplicateSubobjects(UActorChannel* Channel, FOutBunch* Bunch, FReplicationFlags* RepFlags)
{
    bool bWroteSomething = Super::ReplicateSubobjects(Channel, Bunch, RepFlags);
    // MyActorSubobject becomes a replicated subobject here
    if (IsValid(HitData))
    {
        bWroteSomething |= Channel->ReplicateSubobject(HitData, *Bunch, *RepFlags);
    }
    // Add any other replicated subobjects here in a similar way
    return bWroteSomething;
}

void AGameTypeBase::OnTag_Implementation(UTagComponent* Tagged, const FHitResult& HitResult, AShootable* Shootable, const UWorld* worldRef)
{
    RemoveReplicatedSubObject(HitData);
    HitData = MakeHitInfo(Tagged, HitResult, Shootable, worldRef);

    bool bHitValid = false;
    if (HitData) {
        AddReplicatedSubObject(HitData); //temporarily replicate via the gametype as we process OnCustomTag
        bHitValid = OnCustomTag(HitData);


    }
    if (bHitValid) {
        GameData->AddValidHit(HitData);
        AnnounceTag(HitData);
    }
}

UHitInfo_DataAsset* AGameTypeBase::MakeHitInfo_Implementation(UTagComponent* Tagged, const FHitResult& HitResult, AShootable* Shootable, const UWorld* worldRef, float distMult)
{
    UHitInfo_DataAsset* data = NewObject<UHitInfo_DataAsset>(GetOuter());
    float baseValue = Shootable->GetBasePointValue() * Tagged->basePoints;
    float baseMultiplier = Shootable->GetBounceCount() + 1;
    float dist = Shootable->GetDistance();
    data->SetHitInfo(Tagged, Shootable, baseValue, baseMultiplier, dist, distMult, UPublicHelperFunctions::GetGameState(worldRef)->getGameTime(), HitResult);
    for (const auto& t : Tagged->ComponentTags) {
        //data->hitTags.Add(t);
    }
    return data;
}

sorry if all this might be too much but just to be safe, first its generated in OnTag via MakeHitInfo, after which its passed into OnCustomTag. In OnCustomTag, it's broadcasted via an event

In OnCustomTag, it's passed into AddHitInfo like so:

if (bValidHit) {
    attackState->AddHitInfo(data, total * selfMultiply, total * teamMultiply);
}

then, AddHitInfo calls some functions which broadcasts the data:

void AUserPlayerState::AddHitInfo_Implementation(UHitInfo_DataAsset* HitInfo, const float playerPointsAdd, const float teamPointsAdd)
{
    if(GetPlayerGameData()) GetPlayerGameData()->Hits.Add(HitInfo);
    SetScore(GetScore() + playerPointsAdd);
    onPointsAdd(HitInfo, playerPointsAdd, teamPointsAdd);
    if (GetTeam()) GetTeam()->AddHitInfo(HitInfo, teamPointsAdd);
    //GEngine->AddOnScreenDebugMessage(-1, 10.f, FColor::White, FString::Printf(TEXT("POINTS: %f"), GetScore()));
    //GEngine->AddOnScreenDebugMessage(-1, 10.f, FColor::White, GetName());
}

void AUserPlayerState::onPointsAdd_Implementation(UHitInfo_DataAsset* HitInfo, const float playerPointsAdd, const float teamPointsAdd)
{
    PointsAddDelegate.Broadcast(HitInfo, playerPointsAdd, teamPointsAdd);
    onPointsAdd_Client(HitInfo, playerPointsAdd, teamPointsAdd);
    onPointsAdd_Multicast(HitInfo, playerPointsAdd, teamPointsAdd);
}

void AUserPlayerState::onPointsAdd_Client_Implementation(UHitInfo_DataAsset* HitInfo, const float playerPointsAdd, const float teamPointsAdd)
{

    PointsAddDelegate_Client.Broadcast(HitInfo, playerPointsAdd, teamPointsAdd);
}

void AUserPlayerState::onPointsAdd_Multicast_Implementation(UHitInfo_DataAsset* HitInfo, const float playerPointsAdd, const float teamPointsAdd)
{

    PointsAddDelegate_Multicast.Broadcast(HitInfo, playerPointsAdd, teamPointsAdd);
}

again, sorry if that's too verbose, but I thought I'd be thorough since I'm not really sure where the problem is

[crystal crag]

I am working on a PCG dungeon generator. I've gotten to the part where I want to spawn replicated actors that the players can pick up. Do I have to filter that logic from the clients, because then I will end up with duplicated actors (since SpawnActor is replicated)?

[dark edge]

Spawn the replicated actors on server only and they will replicate to clients

[crystal crag]

That's what I thought, thanks.,

[dark edge]

spawnactor is not replicated, spawning a replicated actor makes it also spawn on client

[crystal crag]

Ah, ok. Well, thanks for the extra clarification!

[torpid lantern]

Is there a trick to get smooth network movment while using Floating Pawn Movement?

I'm wanting bulk NPCs and I am trying to avoid using the CMC for NPCs.

I notice a very slight jitter on pawn movement, but if I reparent my NPCs from "Pawn" to "Character" the jitter is gone.

I believe this is due to the out-of-the-box network smoothing in CMC. How could I recreate that specific functionality using Floating Pawn Movement?

[thin stratus]

Might sound stupid but the answer will be: By coding it.

There is nothing outside the CMC available with smoothing. Despite maybe the projectile movement component.

[exotic wasp]

also worth mentioning, use the new registered subobjects list

its way easier to use

its on that same page

[proper vine]

thanks. we've found that the problem isn't that the data isn't being replicated, it seems that it's taking too long to replicate, so the message containing the relevant data comes out as null

[exotic wasp]

mmm classic

use ReplicatedUsing

[torpid lantern]

As a shareback and for future searchers -

Solution to resolve my NPC jitter for Pawns using FloatingPawnMovement was to create a "NetworkSmoothing" component.

Code: https://pastebin.com/8Q5FWLqb
warning: code is GPT generated

• Create a new cpp class "NetworkSmoothingComponent" and copy+paste the above code
• Replace YOURGAME_API with the correct info for your game
• Add the NetworkSmoothing component to your Pawn
• Disable movement replication for the Pawn
• On the Pawn, create a RepNotify ServerTransform variable
• On Pawn's SERVER tick, set the ServerTransform variable with the pawn's transform
• On the ServerTransform RepNotify, call NetworkSmoothing's "Apply Replicated Transform" function and feed it ServerTransform

Note: Component also exposes Interpolation Time as a blueprint variable for tuning

[exotic wasp]

why does the component even exist if you're having to change the pawn

[torpid lantern]

Was that a question for me? If so, I'm not changing the pawn. Objective here is to create a component that can smooth replicated movement for non-characters.

[exotic wasp]

arent you adding a ServerTransform property to the pawn?

[torpid lantern]

Yeah! Server tick sets a RepNotify transform var on the Pawn, which triggers the OnRep, which feeds the transform into the component; comp then interps the Pawn to the location/rotation locally.

I can disable "replicate movement" on the Pawn, because that's what it's doing anyway with the NetworkSmoothing component.

Do you see anything majorly flawed? I am not very good with cpp, but this is what GPT told me to do and it cleared my jitter issue with FloatingPawnMovement.

[thin stratus]

I can gladly review that code, if you want. Just not sure what's the best way to do that. Maybe throw it into GitHub as a PR/merge to some empty repo, so one can comment on it.

If we ignore redundant stuff and bad coding by ChatGPT, I would say that it's an okay start. The main thing I would try to improve are the way you smooth (doesn't need to be linear for example) and edge cases, such as the Transform smoothing having finished but the next update isn't coming in quick enough (e.g. ping increased). That would lead to your actor standing still for a few frames before it starts interpolating again

So basically some simple extrapolation is missing

And if you are working in C++ it's theoretically an option to use the ReplicatedMovement anyway and simply override the function where UE just sets the Actor Transform to the replicated one.

You might also want to consider the case that your actor could be attached to something, which would look strange if the location is then smoothed, as you'd rather want it to stick nicely to the other actor in case it's moving or animating.

So yeah, room for improvement is there. The general interpolation logic is usually simple. It's the edge cases that make it problematic

[soft condor]

Hey folks! I need some help with using delegates over the network. I have a delegate inside a Uobject(HolsterSystem) that I essentially call inside of a repNotify. I did some debugging and found that at the time when the repnotify is called the delegate has lost it's invocation list. My uobject is created on beginplay of the player. I also add a weaklambda like below. I'm pretty sure it gets added at the end of beginplay as debugging showed that as well. I suspect maybe it has something do with the delegate only existing on the server? I found out that delegates are not replicated so could it be that the delegate isn't on the client and therefore never had that invocation list added? Any thoughts on this would be helpful thanks:

  
DECLARE_MULTICAST_DELEGATE(FOnHolsterAssignedCallback)
class UHolsterSystem: public UObject {
//other stuff up here
FOnHolsterAssignedCallback OnHolsterAssigned;
}

//holstersystem.cpp
void UHolsterSystem::OnRep_PrimaryHolsters(){
for (int32 i = 0; i < PrimaryHolsters.Num(); i++){
   if (PrimaryHolsters[i].State != LastPrimaryHolsters[i].State){
    OnStateChange(PrimaryHolsters[i].State);//this method broadcasts delegate and it is here where it has empty list
      }
}
LastPrimaryHolsters = PrimaryHolsters;
}

//playercharacter.h
class AhordeGameCharacter : public ACharacter{
//stuff
UPROPERTY(Replicated)
TObjectPtr<UHolsterSystem> HolsterSystem;
}

//playercharacter.cpp
void AhordeGameCharacter::BeginPlay(){
if (GetLocalRole() == ROLE_Authority){
HolsterSystem = NewObject<UHolsterSystem>();
if (HolsterSystem){
HolsterSystem->OnHolsterAssigned.AddWeakLambda(this, [this](){
 TestClientCallback();
});
}
}
}//has invocation list at this point```

[thin stratus]

You are in fact only binding on the server.

You'll need to change the HolsterSystem variable to be ReplicatedUsing and then bind additionally in the OnRep for it

[nocturne quail]

I finally completed the system that You recommended me yesterday, and it is working as intended but a very very tiny small jitter still feels periodically like every 10 or 15 seconds. is there any thing incorrectly i'm doing or it can't be further improved?

by default character in standing mode

[thin stratus]

If you are calling an RPC to modify the CMC state and then modify it locally through the OnRep it will cause corrections

If the movement speed values are known upfront then you shouldn't need a Server RPC

bMoveForward seems wrong. You shouldn't need to set that. The CMC can figure that out on server and client by itself.

The AddMovementInput call will result in the Server and Client having an Acceleration value that points in that direction. So you can easily figure out what the client pressed by using that.

[nocturne quail]

all speed values exists in the movement component and applied by the PhysWalking based on the derived state from character.
if i change and update that state without rpc, then it extremely jitters, and setting it only by server also jitters extremely... so i changed the state by onrep and it works smoothly for first 15sec and then it jitters one and next 15sec again smooth

[thin stratus]

Yeah but all of the options you mentioned are wrong.

You need to use information that the server and client have available inside the CMC. You can't only set it on one side and you can use additional replication.

[nocturne quail]

it will be great, but i don't know how to figure it out directly in the PhysWalking without sending character state to it

[thin stratus]

The CMC already sends an RPC with information

[nocturne quail]

i think i need to access the controller somehow in cmc to figure it out which key and direction was pressed?

[thin stratus]

The acceleration variable you can access in the CMC should be the result of the Movement Input you added, multiplied by the acceleration setting the CMC has.

No. The acceleration already points into what the player pressed. In world space fwiw

Can be unrotated by whatever you use to define the direction

If you want it local for something

If you want to know if the player pressed forward (local forward, so e.g. W) then you can unrotate the Acceleration and check the X axis for being positive

[nocturne quail]

acceleration variable returns the actual maxwalk speed, or is it 1, -1 ?

[thin stratus]

Neither

Acceleration is your Input Value (0 to 1) multiplied by the WorldDirection you passed into AddMovementInput and then multiplied by the MaxAcceleration of the CMC

Could also be that it's called InputAcceleration

Would need to double check

But you can unrotate that vector by eg. The control rotation. To get it to point into your local input axis.

[nocturne quail]

this is an amazing information, this solved to avoid extra cache input var replication

[thin stratus]

Yop. As long as the information is known on both sides at all times you can simply use it inside the CMC. The only thing you can't do is use a speed value that only one of the two knows.

You should usually not have a speed value that only the client knows, cause sending that to the server allows cheating

And if you have one that only the server knows then you'll need to tell it to the client (e.g. via the OnRep) and more or less accept the one time correction

[nocturne quail]

yeah this is what actually i'm doing, the speed struct is initialized by the server at runtime

[thin stratus]

But if both know about it, e.g. you have a Sprint speed that you set on the CMC or character in the defaults before pressing play, then you are usually lucky and can just utilize whatever turns on sprinting inside the CMC via key presses send to the server along the existing server rpc

[thin stratus]

Once or multiple times?

[nocturne quail]

Once, the whole struct and then using it

[thin stratus]

Yeah that's fine. I assume this happens before the player even moves.

What I meant is something else. Like if another player casts a spell at yours and halfs your movement speed. You can't know about that on the local client. So when the debuff replicates you'll have a small time where the server moves you with the debuff and you locally predict it wrong until the debuff replicated

[nocturne quail]

I have a variable which enables the player input by server after 5 seconds of possession, and in this 5sec all critical data initialized by the server and then player can move, jump etc

[brazen anvil]

Is it possible to store a inventory component on the player controller of a character? The reason I ask is related to cheating. If the inventory was on the character, then everyone gets a copy of the character. If it was on the controller then everyone wouldn't get the controller.

[thin stratus]

Yes, that's possible. You can also place it onto the character and simply mark the properties to only replicate to the owner though

Which is what I would go for unless I would want the Inventory to survive the character being respawned

Which is rarely the case. And if it needs to then the character could also just be teleported and reset instead of fully respawned. That's cheaper anyway

[crisp shard]

i wouldn't consider this an issue per say, but i always am running into something im needing to solve for, which is when the character is interactin with a server actor and they ghen give the character a widget, once character has widget i typically will just rpc from character to do whatever. but there are plenty of times where i want to communicate back to that thing i just communicated with, and storing that as a ref on the charazcter for that singleuar event doesn't seem like a good idae, but im not sure what might be some other approaches to getting that communication to be efficient and proper.

i have ways to solve for this but i feel like they are all where i start to lose full confidence in the appraoch.

an example i have is when interacting with a "console" you get the console ref returned as an actor and i can communicate with it from there via interface functions etc, but this is a one off thing where it made sense to do that, but in most cases, like the one imcurruenlty working on i feel ike there has got to be a better appraoch to the communication between all things

[thin stratus]

Not sure I follow tbh. An actual example of where you struggle to "communicate" would be better.

[crisp shard]

yea hard to explain i guess, cuase tbh i can actually get everything to communicate but im doing overly extreme checks to get things to work correctly

i can share how things work currently but it's a bit much

• client interacts w actor (rpc)
• actor gives client a widget
• client does things, rpc's back to character toget to server

that part is fine, but from there i got from the cahracter to the game state where i have that actor stored (i hate this personally , as it has made the process of getting everything back and forth very difficult)

i go to gamestate, check all the possible actors that i just interacted with, and then i check those for another actor... it'sjust a mess lol and this is all on the gamestate, and so getting everything back to the widget / character / actors is turning into a nightmare

my instinct says that i could solve this by using event dispatchers to communicate better between the charcter and the actor and then that actor could communicate with the other actor via another event dispatcher

the real thing that i'd like to find a consistnecy for doing in terms of the logic, is how to communicate back to an actor that gave the client a widget after that client makes some changes.

so actor gives widget to client (the variables of which are from server) and client wants to communicate back to that specific actor.

and again i had 1 solution that works but i don't want to scale that method (bascially putting an actor ref on the character when im interacted with it). not the bp , but an actor then i just interface message with it, but i think an event dispatcher from that charadter rpc, after changing the widget, i think it would maybe work, i just dont use them alot

This type of thing usually isn’t a challenge but I think it’s because I’m going from the character to an actor and then that actor creates another actor, and although I do have things “working” it’s just omega messy

I plan on redoing pretty much everything from scratch on this particular system tho so I can try again.

[soft condor]

Thanks for help! This makes sense and it worked 🫶🏿

[outer birch]

Also, don't put in things that silently erode gameplay if you detect a hacker. Like it's funny as shit when in the batman arkam games, they had his cape fail to deploy if they found you were using a pirated copy and youd constantly fall to your death. HILARIOUS. But then the priates would review bomb the game about how the game was broken and buggy.

When using iris, do you ever get down to 'nothing', or do you only ever get down to a point where during the pollpush pass of iris they check to see i guess if anything is dirty?

AFAIK, i've removed all polling from the healthset, but there's still 200ns of it doing...something. I assumed it would disappear completely and would add itself to a larger table when dirtied, but perhaps not.

[dark edge]

It also happens if character is not attached so I'm pretty sure it's not self-collision

Any thoughts on where to look for this jittery behavior?

[nocturne quail]

obviously something needs to tune-up in physics settings

[dark parcel]

If I got an array of a data type, lets say ItemFragment (Uobject).

The array is replicated but the ItemFragment is not.

Will adding / removing the array replicate the changes? I am guessing not?

[thin stratus]

Reminds me of Game Dev Tycoon (or one of those games at least), where they had a version of the game where you would always lose the playthrough cause your games were getting pirated. And iirc they uploaded that to the given pirate websites themselves and enjoyed the support posts about "How do I stop my company from failing due to people pirating my games?".

EDIT: Just googled it and it was Game Dev Tycoon. When the game launched almost 95% of the over 3000 running game sessions were apparently the pirated version. Can't make that shit up.

[dark edge]

I did some more investigating, totally at a loss here as to what the problem could be.

Found a difference. Component Replicates is ON for the jittery one. Turning it on for the BP one causes it to jitter too.

[thin stratus]

I can't share too much but my wife recently looked into performance improvements for Iris for our current customer and Iris has some pretty big flaws/bugs on the polling side currently

If you want to figure that out check how it handles dirtied subobjects as well as properties. You might come to the same conclusion. Fixing/altering this can improve performance of Iris drastically

There is a CL by someone (official epic commit) that tried to fix something on Iris and they ended up causing dirty subobjects to cause a huge overhead.

What you posted there about the health attribute set showing up even though it's not dirty should be exactly that. But I never looked at the code, the fix etc, I only got told about it.

[thin stratus]

If it's a runtime spawned uobject then I would assume no

[dark parcel]

https://dev.epicgames.com/documentation/en-us/unreal-engine/replicating-uobjects-in-unreal-engine

It seems NewObject will do the trick?

but maybe I need to make the UObject replicated too 🤔

testing right now, gonna take a while to write this.

Can't help but feel like I am overstepping this though. I am just working on interaction component in multiplayer. Trying to have modularity like how Lyra handle items, maybe I should have kept this simpler 🤷‍♂️ .

[dark edge]

So anyone got any insight as to why that could be the case? I'm guessing the component is replicating its transform in some way that's clobbering the actor movement replication.

[dark parcel]

https://forums.unrealengine.com/t/replicating-tarrays-crashes-game/21237/6

found this, gonna give it ago.

Seems like I have to mark both the array and the underlying Uobject as replicated.

OnRep will get called twice as well , one for the underlying property, the other for array size change.

[thin stratus]

I can have a look in a few minutes. I'm just making a coffee

[thin stratus]

That's always the case.

Even for Actor and Component Arrays

Without that you'd have a shitty time

[dark parcel]

[thin stratus]

Cause if the server spawns an actor and adds it to an array, you have no chance in guaranteeing that the actor fully replicated before the array of the other object does

[dark parcel]

gotcha!

[thin stratus]

The only case where the object doesn't need to be marked as replicated is if the object exists locally too and is stably named and netaddressable iirc

E.g. an asset in your content browser, a component of a replicated actor, an actor placed into the scene etc.

[thin stratus]

How do you not throw yourself out of the window with that sound constantly playing.

@dark edge Any reason you choose to fight Physics with this stuff? Are you simply accepting that you can't predict the movement on the client and will have shitty rubberbanding and corrections?

This is of course not relevant to the actual question, which I will have a look at in a second.

[dark edge]

I am sort of predicting, the sim runs everywhere

Without components replicated it works beautifully, I have a simple 1 actor vehicle which is butter smooth in multiplayer

[thin stratus]

Yeah, but there is no implementation for replaying moves etc.

Unless you make this client authorative I guess.

[dark edge]

Nothing is replayed, no prediction at all

Entire sim runs everywhere, stock physics replicating keeps it in sync if it gets too divergent

You do have ping lag between inputs and things happening but that's fine for my project

[thin stratus]

Hm well, okay. Mainly asking cause 5 years or so ago I was in a similar position and I choose against physics and handled in in the CMC instead, with Vector math.

[dark edge]

Nooooo this is physics at its core

and you have multi-player vehicles, prediction would be an absolute nightmare

[thin stratus]

But only one drives the multi player vehicles or?

[dark edge]

This is more garrysmod than Rocket League

[dark edge]

However it's set up, multiple drivers is totally doable

Anyways, where I'm at right now is that if Component Replicates is off on the root component of every part, it works beautifully, except I lose all the attachment replication goodies.
If it's on, attachment replication works great, but the physics replication is getting clobbered in some way.

[thin stratus]

Right, I coded that stuff for the "Hovering Drones" of Hoverloop (there is a gameplay trailer on YouTube, but the game never made it). It's of course not GMod or Multi-player vehicles, so was just wondering why the need for actual physics. But makes sense.

[thin stratus]

Checking the code atm

[thin stratus]

So the main Vehicle Actor is set to Replicates and Replicates Movement?

[dark edge]

Each part, yes.

There is no vehicle actor, just parts, and a single part is the one simulating physics

all others are attached to it with auto weld

[thin stratus]

About the RootComponent being marked as replicated:

In theory, most of the time you'll not need to do that, as the majority of properties would be driven through the Actors own replication.

USceneComponent replicates a shit ton of properties and has OnReps to apply them:

DOREPLIFETIME_WITH_PARAMS_FAST(USceneComponent, bAbsoluteLocation, SharedParams);
DOREPLIFETIME_WITH_PARAMS_FAST(USceneComponent, bAbsoluteRotation, SharedParams);
DOREPLIFETIME_WITH_PARAMS_FAST(USceneComponent, bAbsoluteScale, SharedParams);
DOREPLIFETIME_WITH_PARAMS_FAST(USceneComponent, bVisible, SharedParams);
DOREPLIFETIME_WITH_PARAMS_FAST(USceneComponent, bShouldBeAttached, SharedParams);
DOREPLIFETIME_WITH_PARAMS_FAST(USceneComponent, bShouldSnapLocationWhenAttached, SharedParams);
DOREPLIFETIME_WITH_PARAMS_FAST(USceneComponent, bShouldSnapRotationWhenAttached, SharedParams);
DOREPLIFETIME_WITH_PARAMS_FAST(USceneComponent, bShouldSnapScaleWhenAttached, SharedParams);
DOREPLIFETIME_WITH_PARAMS_FAST(USceneComponent, AttachParent, SharedParams);
DOREPLIFETIME_WITH_PARAMS_FAST(USceneComponent, AttachChildren, SharedParams);
DOREPLIFETIME_WITH_PARAMS_FAST(USceneComponent, AttachSocketName, SharedParams);
DOREPLIFETIME_WITH_PARAMS_FAST(USceneComponent, RelativeLocation, RelativeLocationParams);
DOREPLIFETIME_WITH_PARAMS_FAST(USceneComponent, RelativeRotation, SharedParams);
DOREPLIFETIME_WITH_PARAMS_FAST(USceneComponent, RelativeScale3D, SharedParams);
DOREPLIFETIME_WITH_PARAMS_FAST(USceneComponent, Mobility, SharedParams);

Any kind of replication of Transform will fight your local sim.

[dark edge]

The funny thing is it fights the built-in physics replication as well, it's reproducable with a single component actor

[thin stratus]

There is no replication for Physics though

[dark edge]

Sure there is

[thin stratus]

Right, maybe Chaos has it by now?

For the LEGO stuff?

[dark edge]

That's been in since early UE4

maybe since the beginning

it's just not predictive is all

[thin stratus]

Right, but that and whatever Chaos is doing for LEGO is both only meant for things that are in the scene, not directly controlled.

[dark edge]

I think maybe my solution is to either opt out of whatever is clobbering it, or make the system work with root component not replicated

[thin stratus]

The jitter is probably also going away (more or less) if you disable the local sim

[dark edge]

testing

[thin stratus]

The issue is that the Location locally is X, while you constantly get replicated locations from the Server that are X+-SomeValue

So the Actor/Component gets teleported around.

You also have the Actor ReplicatedMovement on, which in theory is also not needed, unless the Physics replication needs that.

ReplicatedMovement on the Actor just does this:

void AActor::PostNetReceiveLocationAndRotation()
{
    const FRepMovement& LocalRepMovement = GetReplicatedMovement();
    FVector NewLocation = FRepMovement::RebaseOntoLocalOrigin(LocalRepMovement.Location, this);

    if( RootComponent && RootComponent->IsRegistered() && (NewLocation != GetActorLocation() || LocalRepMovement.Rotation != GetActorRotation()) )
    {
        SetActorLocationAndRotation(NewLocation, LocalRepMovement.Rotation, /*bSweep=*/ false);
    }
}

That's the main reason why people who enable that see jitter.

Cause there is no smoothing built in and the packages the client receives from the Server aren't accurate.

And that's without even taking prediction into account.

Just moving a Block on the Server over Time and having that ReplicatedMovement enabled results in (expected) jitter.

[dark edge]

Yeah I think there's at least 2 and maybe 3 things having an opinion over where the thing should be

[thin stratus]

I assume the DOREPLIFETIME_WITH_PARAMS_FAST(USceneComponent, RelativeLocation, RelativeLocationParams); will do the same

[nova wasp]

you can disable replicated properties in a child component

[thin stratus]

void USceneComponent::OnRep_Transform()
{
    bNetUpdateTransform = true;
}

[thin stratus]

I know why Adriel has them replicated.

Because part of the replicated properties is also the AttachParent.

But in theory it's not needed to replicate the Component for that.

[nova wasp]

also many onreps are virtual and cna be changed to do whatever instead

[thin stratus]

Cause the Actor has logic for replicating the attachparent stuff already.

Discord, please.

void USceneComponent::PostRepNotifies()
{
    if (bNetUpdateAttachment)
    {
        Exchange(NetOldAttachParent, AttachParent);
        Exchange(NetOldAttachSocketName, AttachSocketName);
        
        // Note: This is a local fix for JIRA UE-43355.
        if (bShouldSnapLocationWhenAttached && !bNetUpdateTransform)
        {
            SetRelativeLocation_Direct(FVector::ZeroVector);
        }
        if (bShouldSnapRotationWhenAttached && !bNetUpdateTransform)
        {
            SetRelativeRotation_Direct(FRotator::ZeroRotator);
        }
        if (bShouldSnapScaleWhenAttached && !bNetUpdateTransform)
        {
            SetRelativeScale3D_Direct(FVector::OneVector);
        }

        // Check if this is a detach
        if (AttachParent && !bShouldBeAttached)
        {
            ensureMsgf(NetOldAttachParent == nullptr, TEXT("Local modification of AttachParent detected for replicated component %s, disable replication or execute detachment on host."), *GetFullName());
            DetachFromComponent(FDetachmentTransformRules::KeepWorldTransform);
        }
        else
        {
            const bool bOldShouldBeAttached = bShouldBeAttached;
            const bool bOldShouldSnapLocationWhenAttached = bShouldSnapLocationWhenAttached;
            const bool bOldShouldSnapRotationWhenAttached = bShouldSnapRotationWhenAttached;
            const bool bOldShouldSnapScaleWhenAttached    = bShouldSnapScaleWhenAttached;

            AttachToComponent(NetOldAttachParent, FAttachmentTransformRules::KeepRelativeTransform, NetOldAttachSocketName);

            // restore to what we have received from the server
            SetShouldBeAttached(bOldShouldBeAttached);
            SetShouldSnapLocationWhenAttached(bOldShouldSnapLocationWhenAttached);
            SetShouldSnapRotationWhenAttached(bOldShouldSnapRotationWhenAttached);
            SetShouldSnapScaleWhenAttached(bOldShouldSnapScaleWhenAttached);
        }

        bNetUpdateAttachment = false;
    }

    if (bNetUpdateTransform)
    {
        UpdateComponentToWorld();
        bNetUpdateTransform = false;
    }
}

That's what the SceneComponent does if you replicate it

Or rather mark it as replicated.

It handles Relative Transform and Attachment.

Relative Transform causes Adriel's jitter.

[dark edge]

Yeah that might be it

[thin stratus]

Turning replication off turns the Attachment off too

But you shouldn't need the SceneComponent attachment.

These Components are on Actors, or not?

[dark edge]

just a check if simulating physics before updating component to world?

Every vehicle part is an actor. A vehicle is some collection of parts all welded together

lego style

[thin stratus]

if Component Replicates is off on the root component of every part, it works beautifully, except I lose all the attachment replication goodies

This part is what I meant.

[dark edge]

Yeah I'll try override that onrep and just not update component to world if simulating

[nova wasp]

DISABLE_REPLICATED_PROPERTY(USceneComponent, blah)

[thin stratus]

The Actor VehiclePart, if you mark that Actor as replicated (doesn't need ReplicatedMovement), you should already have Attachment replicate.

[nova wasp]

overriding the onrep is far more simple though I guess, and lets you conditionally do the intended route when needed

[dark edge]

It breaks local transform somehow

[thin stratus]

Megafunk, I don't 100% get why that is relevant though.
Adriel has full control over turning it on or off already.
It's more about figuring out why the Component has to be replicated to begin with.

[dark edge]

if Component is not replicated, attachment offsets break

[thin stratus]

Yeah, and it shouldn't if the Actor itself is replicated.

Cause if you attach the Actor it will already replicate.

[nova wasp]

I guess my point is you can keep other replicated properties and remove some you don't need

its sometimes useful to only turn some things off and not all of them

[thin stratus]

Ah, :D then you might want to actively write that in one single response. Felt more like you are throwing in random things to try and help.

[dark edge]

It does, that was the first roadbump I encountered.
Or rather, it breaks when you simulate physics.
#programmer-hangout message

[thin stratus]

Is it simple to reproduce your setup?

[nova wasp]

its both lol

[thin stratus]

Also, you have this BP vs C++ thing. I assume with your realizing the difference, you can break it in either setup?

[nova wasp]

my first reaction upon seeing how scene component rep works is questioning how anything works because there are 10 different properties that can show up at random frames

non atomic send = random shit shows up with 0 context

[thin stratus]

Yeah, kinda. I think in most cases it still runs through one and the same replication package due to the Actor

[nova wasp]

so you could for example receive what you are attached to and not RelativeLocation until a few frames later

[thin stratus]

There was docs page that explained the order in which this all works. But Iris might do it differnetly in the future i guess.

[nova wasp]

most cases is not good enough in my opinion here

[dark edge]

Try this:

[thin stratus]

UE version?

[dark edge]

If root component is not replicated, it works fine.
If replicated, it jitters.

[thin stratus]

I mean I can potentially just recreate the setup. Was mainly wondering if BPs is enough

[dark edge]

5.5.4

That's all it is

[nova wasp]

I forget, is this using fixed ticking physics?

[dark edge]

Nope just default

Iterations turned up but otherwise Chaos is stock

[thin stratus]

Gimme a minute. Have no active project locally. Need to quickly push together a character to actually walk with >.>

(EnhancedInput is so fun........)

[dark edge]

1PP template should work

[thin stratus]

@outer birch Wife said you should "look into the BuildPollList function where they propagate dirtyness".

[thin stratus]

Bold of you to assume that I have templates downloaded.

[thin stratus]

Assume that WorldLocation.Z math has some deeper meaning?

The -250 are a bit strange. :D

[dark edge]

just to get it to hover

just gotta be able to move around with physics as smoothly as possible

My actual project is tires and such but gotta smoothly hover before trying to smoothly have suspension and tire grip

[dark edge]

Investigated overrideing this, all those properties used inside of it are private LOL

and UpdateComponentToWorld is final

[thin stratus]

I think I'm doing something wrong. Mine just starts lifting off

Are you sure the MapRangeClamped is correct o.o

If location is positive, we add a positive force?

[dark edge]

if Target-Location is positive

[thin stratus]

Ah the 250 is a look ahead?

[dark edge]

That's the Z elevation to have zero force

below that you push up, above that you push down

This test rig has 4 hover points so it can get a little bit of wobble action going to accentuate any transform clobbering visually

[thin stratus]

Isn't that the same setup? o.o

Ah no

The minus is inverted

Aha, got it

[dark edge]

Once you can confirm it hovers and slides around if you bump it, test in multiplayer. You should see that it behaves very badly if Component Replicates is on for the root

[thin stratus]

Yeah I need to make that thing a bit smaller first.

So that's what I have for now.

[dark edge]

Looks good, now push it

Short clip showing test and results

[thin stratus]

Yeah just adding a bit of debug functionality

Could be smoother fwiw.

Pretty sure that's the Actor Transform correcting constantly.

Yeah it is

But without setting Replicate Movement on the Actor, you run into the problem that it can desync.

So you would still need to add some smoothing fwiw.

[dark edge]

That's always been on

[thin stratus]

Doesn't matter though, you are just not noticing it yet.

[dark edge]

The weird part is it's smooth with Component Replicates off, and a jittery mess with Component Replicats on

[thin stratus]

Lemme check how bad it gets with the scene compoent replicating

[thin stratus]

Yeah but it's the same concept basically.

The Server Transform replicates back to the Client

And constantly corrects it

[dark edge]

If you have stock physics replication settings they are really strict so you'll encounter more snapping than me

[dark edge]

Yes, but I think the root is that there's 2 systems doing that.
Physics replication is doing it nice and smooth with all sorts of tunability, and something is just directly setting component transforms when Component Replicates is on

[thin stratus]

Do you replicate only teh root of that hover actor?

Do I need to enable physics replication somewhere?

I usually never touch that shit due to the nature of it desyncing

[dark edge]

on by default

[dark edge]

Yes, turning on ReplicateComponent on the root of the hover actor is enough to clobber transforms

[thin stratus]

Replicated Movement + Replicated RootComponent

Starts jittering when hitting it a second time.

[dark edge]

Actor Replicates
Replicate Movement
Components Replicates = false

Behaves

Actor Replicates
Replicate Movement
Components Replicates = true

Jitter City

[thin stratus]

Mind you, I set the FPS to 60 flat for now. Lemme unlock that.

[dark edge]

That might be it

Might be rotation only

hard to tell

[thin stratus]

Hm na, looks similar even with 120 fps

[dark edge]

I think there's 2 systems trying to keep the root components transform in sync, and they're fighting.

[thin stratus]

ReplicatedMovement on the Actor Off
Replicated on the Component On

So it doesn|t really matter which of the two are on. As soon as you enable replication of the Transform of either Actor or Scene Component, you will get the "corrections"

I don't know why it's so much worse for you when you have both enabled

But I don't see a difference.

Lemme spawn two of them with the settings all exposed

[dark edge]

What engine you on?

[thin stratus]

5.3 tbh, cause I have no 5.5 project at hand.

But this should be relatively unchanged between versions

@dark edge Doesn't really seem like they are very different in behavior. Minus the one that doesn't replicate Movement and Component.

Also with both of those turned off, if Physics replication would be on by default, shouldn't it still sync?

[dark edge]

What are the red and green boxes?

[thin stratus]

I guess it's not on by default in 5.3?

[thin stratus]

Green is the Location of the Component on the Client, Red on the Server.

Makes it easier to compare.

[dark edge]

That's just prediction

[dark edge]

My only difference is my angle lerp is 0

[thin stratus]

Okay with that on I do see a lot more corrections

Lovely

The "none" one is still not replicating, so it will be tested somewhere

Too many systems fighting each other remains my guess.

The one that has no replication on movement or scene component still desyncs

Enabling Resimulation doesn't make it any better

Let's have a look how that boolean is even utilized

Okay so for one, that Physics stuff overrides the LinearVelocity on the RootComponent with the replicated one of the Server.

void AActor::PostNetReceiveVelocity(const FVector& NewVelocity)
{
    const FPhysicsPredictionSettings& PhysicsPredictionSettings = UPhysicsSettings::Get()->PhysicsPrediction;
    if (PhysicsPredictionSettings.bEnablePhysicsPrediction)
    {
        // required to keep client deterministic with server simulation
        if (RootComponent && RootComponent->IsRegistered() && (NewVelocity != GetVelocity()))
        {
            if (UPrimitiveComponent* PrimitiveComponent = Cast<UPrimitiveComponent>(RootComponent))
            {
                PrimitiveComponent->SetPhysicsLinearVelocity(NewVelocity);
            }
        }
    }
}

[thin stratus]

Hm, it seems to cause all sorts of shit. It also touches how the CMC works and moves the whole thing to an Async Tick.

[dark edge]

I think it's that raw UpdateComponentToWorld doing it

[dark edge]

yeah the new physics stuff isn't ready for prime time I don't think lol

[thin stratus]

I struggle to put my finger on what is going wrong in yours vs mine.

There is also bReplicatePhysicsToAutonomousProxy on the Component, which is true by default in 5.3

[dark edge]

Yeah that's on

[dark edge]

Maybe you're drawing boxes early

I see the mesh moving relative to the clientside box

[thin stratus]

I just don't see how it would keep the Server and Client in sync without at least one of the two Movement Replications being true

[dark edge]

I think the physics replication relies on Movement Replication or at least the actor being replicated

[thin stratus]

Could be

[dark edge]

That's fine, it works great like that. What breaks it is the raw component to world transform being updated when Component replicates

I think in the meantime I might just see if I can keep the attachment and relative transforms of attached parts in sync without ComponentReplicates, if I can do that then it'll work great

Just gotta give each part an AttachParent and RelativeTransform and they can hook themselves up clientside

To be sure, attachment works without ComponentReplicates, until you enable physics on the root. Then all the children claim to have no attach parent and get sucked into the root's transform. Something is busted.

[thin stratus]

if I turn angle lerp to 0 it starts breaking the one that only replicates the actor LOL

Actually it breaks a lot of stuff

That's probably why it#s smooth for you

You might want to debug your Server and Client positions like I do

Look at how smooth the yellow one is, that is the one that has the Scene Component turned off

The Physics Replication seems to only affect the Actor itself, and if you turn off the Component Replication and set the Angle Lerp to 0.0, you start avoiding corrections but you get a desync.

[dark edge]

just draw debug box?

[thin stratus]

Thing is, if your Parts are Actors and are replicated and you call Attach on them, they should replicate the Attachment. Actor has a ReplicatedAttachment logic already.

[thin stratus]

In the loop.

After the loop for the center mesh.

Extent can vary depending on how big your boxes are.

Also can someone please PR a change to Select nodes so PickA and True/False are not swapped...

From what I can tell, one would:

• Want AngleLerp to be > 0.0
• Want Position Lerp to be very small. Epic says "if not even 0.0", but not sure.
• Need to enable MovementReplication on the Actor
• Not need to enable Replication on the SceneComponent, as that fights with it
• Need to toy around with some of the numbers to get the jitter under control when having AngleLerp > 0.0.

Not sure if the Physics Prediction stuff does any smoothing, but I would say no.

I guess the Angle Lerp etc. is that smoothing

[dark edge]

This is with angle lerp 0

uploading

[thin stratus]

Yeah it does do smoothing, but man the smoothing is choppy

[dark edge]

the lerp is not smoothing, it's a hard snap per packet. The smoothing comes from the coefficients and error accumulation

[thin stratus]

Turning the Angle Lerp to 0.0 turns of my smoothing for it entirely though until it reaches the point where it really hard snaps

[dark edge]

wtb fiber internet, still uploadeing lol

[thin stratus]

:D Gotta love how different it can be

I could blame 5.3 vs 5.5

[dark edge]

I'm probably the only person who uses the built in physics replication lol. It had a typo for YEARS, it was "Physic Replication Settings"

[thin stratus]

I don't know why yours manages to sync up again. In my video you can see how it just desyncs the rotation when I touch the mesh.

[dark edge]

I have everything on except Component Replicates

[thin stratus]

Must be 5.3 vs 5.5 then. Lemme see if I can open this in 5.5

Now I'm curious

For the Actor Attachment, you'd need to actively show the problem.

I'm not sure I understand what is going wrong with that.

Probably the last thing I will do for now though. Can continue with this another time.

Which setting do you have on above the replication settingsß

[thin stratus]

Right, it does seem to behave like you showed in 5.5

That said, I don't see additional jitter with I enable the Component replication -.-

Are you shitting me... 5.5 runs like garbage compared to 5.3

The fuck did they do

[dark edge]

maybe some default option

[thin stratus]

I had 120 fps in 5.3 and now I'm at 62 or so

[dark edge]

maybe you had the editor viewport in live mode vs not

[thin stratus]

Oh wow, it was cause I had the BP open on my other screen

Man this Engine is so shit sometimes

[dark edge]

lmao

[latent heart]

Lol

[dark edge]

Unreal: The worst engine, except all the other ones.

Anyways, gotta get to bed. Thanks for the help, you gave me a few avenues to look into to fix this project.
This week I'll either sort out some way to fix the problems fundamentally or just live with Component Replicates turned off, and handle all the attachment and relative transforms in my own code. Thanks a bunch, and also thanks for the compendium.

[thin stratus]

You can't make this shit up

This is a freaking empty character BP, not even a mesh in it. 2 Input actions and 2 debug keys for the resetting and drawing.

Even minimized it still steals performance

[dark edge]

lmao

I go from 70 fps to like 240 when just launching without editor

[latent heart]

Slate is super efficient.

[thin stratus]

This should be more in line with what you have on your end.

Can't tell if the SceneCompo being replicated adds jitter though. My setup doesn't show that too well

I would say yes though

[dark edge]

network emulation or no?

[thin stratus]

No, but ping shouldn't matter and package loss shouldn't be set to != 0

Lemme try anyway

Welp, ping does matter apparently. It's gonna be that the PhysicsReplication logic does not touch the SceneComponent at all and is only integrated into the Actor Replication.

As soon as you turn on the replication of the SceneComponent you get a second set of Transform being replicated and applied

I think that's enough of testing though. Seems to be pretty clear now that for one 5.3 is not far enough to test/use this, and that the Physics Replication is only on Actor level, and if you start replicating the SceneComponent you add the RelativeTransform being replicated back to the Client into the mix.

@dark edge Don't have the time to look at the video anymore, but here is what I, for whatever reason, had to do to get the attachment working.

To save on coding things twice, I made use of OnRep calling on Server in Blueprints.

In C++ you'd need to manually ensure you attach on the Server too.

[dark edge]

Thanks, yeah I think that's the approach I go with for now

[thin stratus]

Replicating SceneComponent messes that up too btw ...

I like how the physics stuff makes the thing too heavy on my end now

Also, just fyi, attaching a lot of actors with a lot of scene components will quickly become a bottleneck due to the shitty way UE updates the transforms.

So try to keep the amount of scene components low.

[outer birch]

Did they try and then back it out?

[kindred widget]

Slate invalidation is fun. I'd wager 90% of that cost is the details panel.

[dark edge]

how low?
I picture 100-200 scene components per vehicle absolute max

[thin stratus]

Should still be there.

[thin stratus]

Probably.

[dark edge]

At least my device components aren't scenes. I thought about it but I just opted to have them piggyback on scenes for spatial stuff

[thin stratus]

I can't give you exact numbers. If you have a lot of components on a vehicle you gotta profile it.

But I'm sure 100-200 is gonna make you bleed ms.

Depends also on how many vehicles you have in the scene.

[dark edge]

10 max, probably about 50-100 scenes per on average

[thin stratus]

Yeah, that can get pretty expensive.

[dark edge]

I'm guessing the workaround is ISM eventually if it gets there?

[outer birch]

I assume your wife is referring to https://github.com/EpicGames/UnrealEngine/commit/88a2558657312ebc9f394ebb377f698c52fcafa5, the "// Currently we must enforce polling SubObjects with owner" seems like a crappy requirement. Any hints on the 'proper fix' ?