multiplayer

[hybrid zodiac]

Thanks both! I'll investigate Chaos mover and see how much work is required to migrate over

[exotic wasp]

my experience with chaos mover half a year ago is that it crashed when I would open the bp

probably advanced since then if fortnite is using it?

if it's good enough for hundreds of players in a map for fortnite, I have no qualms with physics based movement

[thin stratus]

They did commit a bunch of stuff the past 6 months for Chaos Mover.

But the point remains that Mover in itself is not even close to the feature level and robustness of the CMC.

Sadly stuff like this only ever comes to light when a project is already on the way for a few months or something more complex is needed.

[exotic wasp]

I just wish it was easier to mesh with GAS prediction

[meager spade]

no we still use CMC atm, but im working on switching as we only have very basic movement

and in my tests mover is faster than CMC for what we need

[meager spade]

It has started making progress though, they implemented crouch and stuff, and basically it works just like how CMC works in its base form

ofc it requires a lot of different thinking, cause inputs are not how you do it via CMC

you have to send the state snapshots in

like WantsToCrouch and not already crouching

[meager spade]

im not sure they are using it in Battle Royale

iirc it was used in Lego side

im pretty sure battle royale still uses CMC..

[quasi tide]

Yeah, BR is still using CMC most likely.

But FN Creative is moving to Mover

I seen't it at UE fest

[meager spade]

yeah creative is like the playground

you dont want to break the main Battle Royale

thats the money maker

kinda sad that they gave up with Save the world though

as that was what Fortnite was

[quasi tide]

Yup

[meager spade]

but BR exploded (thanks to Pubg at the time) and ofc Fortnite with its less serious play style to PubG exploded.

[quasi tide]

RIP STW and Paragon and UT

[meager spade]

Paragon was a shame

UT was doomed as soon as fortnite exploded

cause they pulled the team into Fortnite

for BR right?

[quasi tide]

Yup

[meager spade]

yeah and most of them left now

i always wanted to take UT4 and carry it on, but getting permission from Epic

cause its all there stuff, you cant jus tmake a game using UT5

UT4*

[thin stratus]

It's the little things though. CMC has a lot of smaller functions and chunks of code that handle different edge cases. Such as jumping on and off a platform or all sorts of things against cheating.

Mover, at least when I started using it, literally crashed if you destroy the actor while it is being moved by based movement.

You can do the basics, but it's just not "battle tested" and is lacking the stuff that the CMC received based on noticing problems.

[obsidian ermine]

I 100% agree with this ^^^^
Mover is nowhere near production use ready.

[hybrid zodiac]

We are making an FPS-Z title inspired by Tribes, so we have walking, jumping, jetting, skiing and we also added the ability to ski on walls and wall-jump. So far it's all working great with Mover, and I do like how modular it is.

The main issue that inspired me to ask the original question was that I noticed that simulated proxies could sometimes look quite juddery at high speeds (you can hit 200mph if you're good with movement), and my hypothesis which was later confirmed by you guys is that NPP isnt properly simulating movement when forward predicting, so it's constantly having to get updates from the server which becomes noticeable when movement per tick is so high

So my experience with Mover is pretty positive overall, its just frustrating that it isn't well documented or, as in the case with NPP, finished

Hopefully the Chaos mover can fill that missing piece

[verbal ice]

Tbh moving at such high speeds is kinda niche, you might end up having to handle your own smoothing/simulation of those pawns even on another framework

[grand kestrel]

Yeah mover 2.0 is just their prototype. Any engine movement component also is going to be less suitable the more you move away from generic movement. I built a movement component from scratch for my platformer because getting Mario movement physics from CMC or another option wasn't going to happen

It's so much work tho, makes you appreciate CMC all the more

[blazing bear]

IIRC First Person Rendering fixes that, it's production ready in UE 5.7. It's just really good as it provides you with many other features you'll need in your case like accurate self shadows in first person, there is an Unreal Fest video talking about it somewhere.

[compact flame]

How do I fix the jittering of chaos vehicles on client side?

[pliant cypress]

What is IIRC?

[blazing bear]

If I recall correctly

[pliant cypress]

Oh okay

[pliant cypress]

am i setting up correctly right?

overloaded GetMesh()

USkeletalMeshComponent* ASoldierCharacter::GetMesh(int32 Index) const
{
    if (Index == 0) Super::GetMesh();
    else if (Index == 1) return FPS_Mesh;
}

[nova wasp]

what is this overloading?

GetMesh() has no parameters and is not virtual

you should probably just make a separate function for getting the other mesh

If you do need something that gets it based on the local perspective it should consider being the local view target (which is cached on pawns)

[pliant cypress]

just fixed it yeah

[pliant cypress]

doing it

[nova wasp]

APawn::IsLocallyViewed for example

this is how OwnerNoSee and OwnerOnlySee work, they use the view target to figure out if they are a component attached to the view target or not

This is nice because changing view targets will change what you see

[pliant cypress]

i was following this tutorial

https://www.youtube.com/watch?v=rCGrZvKrWzM&t=478s

[pliant cypress]

i dont understand how can i use that ?

[nova wasp]

you can call functions and use their return value to change what the code does

I was just saying for if you needed to figure out which mesh was the "active" one dynamically

[pliant cypress]

how can i fix this?

void UCombatComponent::EquipWeapon(AWeapon* WeaponToEquip)
{
    UE_LOG(LogTemp, Warning, TEXT("SoldierCharacter: %p"), SoldierCharacter);
    UE_LOG(LogTemp, Warning, TEXT("WeaponToEquip: %p"), WeaponToEquip);
    if (SoldierCharacter == nullptr || WeaponToEquip == nullptr)
    {
        return;
    }
    
    EquippedWeapon = WeaponToEquip;
    EquippedWeapon->SetWeaponState(EWeaponState::EWS_Equipped);
    const USkeletalMeshSocket* HandSocket = SoldierCharacter->GetMesh()->GetSocketByName(FName("RightHandSocket"));
    if (HandSocket)
    {
        HandSocket->AttachActor(EquippedWeapon, SoldierCharacter->GetMesh());
    }
    EquippedWeapon->SetOwner(SoldierCharacter);
}

void UCombatComponent::OnRep_EquippedWeapon()
{
    EquippedWeapon->SetWeaponState(EWeaponState::EWS_Equipped);
    const USkeletalMeshSocket* HandSocket = SoldierCharacter->GetMesh()->GetSocketByName(FName("RightHandSocket"));
    if (HandSocket)
    {
        HandSocket->AttachActor(EquippedWeapon, SoldierCharacter->GetMesh());
    }
}

void ASoldierCharacter::ServerEquipButtonPressed_Implementation()
{
    if (Combat)
    {
        Combat->EquipWeapon(OverlappingWeapon);
    }
}

[pliant cypress]

should i replace GetMesh with FPS_Mesh?

[nova wasp]

It depends on the perspective

[pliant cypress]

i got it on FPS mesh

but now got stuck with full body mesh

updated code

[nova wasp]

I do not know what that means

[pliant cypress]

void UCombatComponent::EquipWeapon(AWeapon* WeaponToEquip)
{
    UE_LOG(LogTemp, Warning, TEXT("SoldierCharacter: %p"), SoldierCharacter);
    UE_LOG(LogTemp, Warning, TEXT("WeaponToEquip: %p"), WeaponToEquip);
    if (SoldierCharacter == nullptr || WeaponToEquip == nullptr)
    {
        return;
    }
    
    EquippedWeapon = WeaponToEquip;
    EquippedWeapon->SetWeaponState(EWeaponState::EWS_Equipped);
    if (SoldierCharacter->IsLocallyControlled())
    {
        const USkeletalMeshSocket* HandSocket = SoldierCharacter->GetFPSMesh()->GetSocketByName(FName("RightHandSocket"));
        if (HandSocket)
        {
            HandSocket->AttachActor(EquippedWeapon, SoldierCharacter->GetFPSMesh());
        }
    }
    else
    {
        const USkeletalMeshSocket* HandSocket = SoldierCharacter->GetMesh()->GetSocketByName(FName("RightHandSocket"));
        if (HandSocket)
        {
            HandSocket->AttachActor(EquippedWeapon, SoldierCharacter->GetFPSMesh());
        }
    }
    EquippedWeapon->SetOwner(SoldierCharacter);
}

void UCombatComponent::OnRep_EquippedWeapon()
{
    EquippedWeapon->SetWeaponState(EWeaponState::EWS_Equipped);
    const USkeletalMeshSocket* HandSocket = SoldierCharacter->GetMesh()->GetSocketByName(FName("RightHandSocket"));
    if (SoldierCharacter->IsLocallyControlled())
    {
        const USkeletalMeshSocket* HandSocket = SoldierCharacter->GetFPSMesh()->GetSocketByName(FName("RightHandSocket"));
        if (HandSocket)
        {
            HandSocket->AttachActor(EquippedWeapon, SoldierCharacter->GetFPSMesh());
        }
    }
    else
    {
        const USkeletalMeshSocket* HandSocket = SoldierCharacter->GetMesh()->GetSocketByName(FName("RightHandSocket"));
        if (HandSocket)
        {
            HandSocket->AttachActor(EquippedWeapon, SoldierCharacter->GetFPSMesh());
        }
    }
}

[pliant cypress]

wait, i'll send a ss

[nova wasp]

IsLocallyControlled should be IsLocallyViewed unless you plan to never have spectating/replays in first person

[pliant cypress]

no replay but spectating yes

[nova wasp]

you also appear to be using GetMesh in both cases for local and non locally controlled

[pliant cypress]

void UCombatComponent::EquipWeapon(AWeapon* WeaponToEquip)
{
    UE_LOG(LogTemp, Warning, TEXT("SoldierCharacter: %p"), SoldierCharacter);
    UE_LOG(LogTemp, Warning, TEXT("WeaponToEquip: %p"), WeaponToEquip);
    if (SoldierCharacter == nullptr || WeaponToEquip == nullptr)
    {
        return;
    }
    
    EquippedWeapon = WeaponToEquip;
    EquippedWeapon->SetWeaponState(EWeaponState::EWS_Equipped);
    if (SoldierCharacter->IsLocallyControlled())
    {
        const USkeletalMeshSocket* HandSocket = SoldierCharacter->GetFPSMesh()->GetSocketByName(FName("RightHandSocket"));
        if (HandSocket)
        {
            HandSocket->AttachActor(EquippedWeapon, SoldierCharacter->GetFPSMesh());
        }
    }
    else
    {
        const USkeletalMeshSocket* HandSocket = SoldierCharacter->GetMesh()->GetSocketByName(FName("RightHandSocket"));
        if (HandSocket)
        {
            HandSocket->AttachActor(EquippedWeapon, SoldierCharacter->GetMesh());
        }
    }
    EquippedWeapon->SetOwner(SoldierCharacter);
}

void UCombatComponent::OnRep_EquippedWeapon()
{
    EquippedWeapon->SetWeaponState(EWeaponState::EWS_Equipped);
    if (SoldierCharacter->IsLocallyControlled())
    {
        const USkeletalMeshSocket* HandSocket = SoldierCharacter->GetFPSMesh()->GetSocketByName(FName("RightHandSocket"));
        if (HandSocket)
        {
            HandSocket->AttachActor(EquippedWeapon, SoldierCharacter->GetFPSMesh());
        }
    }
    else
    {
        const USkeletalMeshSocket* HandSocket = SoldierCharacter->GetMesh()->GetSocketByName(FName("RightHandSocket"));
        if (HandSocket)
        {
            HandSocket->AttachActor(EquippedWeapon, SoldierCharacter->GetMesh());
        }
    }
}

[nova wasp]

I have no idea what I'm looking at here either, you have not described which images show which perspectives

if you change just two lines it might be better to just edit the message instead of pasting the giant code block again

[pliant cypress]

Top left is client and big one is server

i want the gun for Simulated Proxies to be attached with fully body

but for the client/server who have equipped the weapon to have gun attached to FPS Mesh

[nova wasp]

debug it I guess

this will silently fail if it can't find the socket because it does not ensure or log an error

[pliant cypress]

Socket exists.

[nova wasp]

prove the AttachActor is being called in all cases

add {,,UnrealEditor-Engine.dll}::GPlayInEditorContextString to your debug watches or just get GetDebugStringForWorld(World)

[pliant cypress]

i'm not that good with deep UE knowledge. kinda beginner-intermediate.

can u pls explain how can i do this?

[nova wasp]

it is REALLY important that you understand how to debug stuff in your IDE here

what this is is a global variable you can watch in the watch window

that shows the name of the current world

this is useful so you can figure out if you breakpoint on player 0, 1, 2, the listen server, etc

you can also just spam debug logs if this is difficult to set up. You can get this string by calling GetDebugStringForWorld

[pliant cypress]

ohk

[nova wasp]

just to be clear you are aware of what a breakpoint is, right?

[pliant cypress]

yes

just don't know what {,,UnrealEditor-Engine.dll}::GPlayInEditorContextString this is

[nova wasp]

yeah this thing is an unfortunate wacky prefix you need to use to tell your debugger about which module that type is in

it's weird... in an idea world we could just use "GPlayInEditorContextString "

some IDES might be smarter about resolving this than others

in my case (I forget why) my custom engine has renamed the base module which is annoying

yours should just be UnrealEditor-Engine though

[pliant cypress]

i am using Visual Studio

2022

[nova wasp]

just add it to the watch

most of the time you will only care about local variables/this but it can be useful to see other things consistently

[pliant cypress]

but what should i look into for this?

in single player

debugging is kinda easier even with breakpoints

this is for OnRep_EquipWeapon

@nova wasp I'm using IsLocallyControlled. If I also use NetRole == Simulated proxies, and attach it to full body mesh for simulated proxies.

Make sense or illogical?

[nova wasp]

I already said to use IsLocallyViewed

I suppose you may also need to swap between the two as well if view targets change

also debugging will be difficult if you are not using DebugGame Editor

[pliant cypress]

Changing configuration?

[pliant cypress]

Okay, but IsLocallyViewed placed on OnRep or Equip only?

[nova wasp]

that depends on what each one is supposed to do, I have no context for when it runs and on which sides

[pliant cypress]

Just see the game like COD

Cod MW2 (2009) to be specific

Each loadout prefixed per mission (level)
Basic multiplayer coop (Spec Ops)

Quest system (Missions / story progression)

[nova wasp]

I think I am asking about the code I am seeing, I don't really need to know everything else about the game

it helps I guess to get an idea of what you want but I am asking on which sides each function runs on

does EquipWeapon run only for the local player? from something else?

[pliant cypress]

The EquippedWeapon is Replicated Property but COND_OwnerOnly. OnRep is for all the clients/other player to update the property for that client so they can also see the weapon is equipped by the player.

[nova wasp]

if it's COND_OwnerOnly how are the simulated proxies going to see it

you are describing the complete opposite

[pliant cypress]

Equip Weapon run for local player (also via RPC if called on server) but OnRep is called as well whenever the Equip is called since property of EquipWeapon is updated.

[nova wasp]

is this onrep firing on simulated proxies or not?

[pliant cypress]

If I just set to Replicated Property, it'll be attached to all player

[pliant cypress]

Didn't implemented the firing rn

[nova wasp]

I am referring to the function being called

[pliant cypress]

Sorry for the bad english

[nova wasp]

all good, if I say anything confusing I can try to make it easier to understand

[pliant cypress]

Yes it is firing

The issue is, OnRep is actually attaching the gun to fps mesh which it shouldn't be

Basically, if you Equip a weapon. I'll see gun on your full body mesh not the fps mesh. While you see, then you'll see the gun on your fps mesh

[thin stratus]

That's a lot of text, quick question: Is this about attaching a WeaponActor to FirstPerson and ThirdPerson meshes?

[pliant cypress]

Yes

[nova wasp]

I am super lost because you are saying the replicating condition is owner only, but is firing on sim proxies

if there are TWO distinct properties we need to know that, we can only see what you show us

[thin stratus]

There is one key thing about this, which I'm not sure was mentioned, so I'm throwing it in:

Attachment in itself is replicated, which means whatever the Server attaches to will be replicated and attaches to on Clients by default.

[nova wasp]

ah, I should have asked if the weapon has replicated movement too

[thin stratus]

Of course with the little note that the Actor you are attaching has to be replicated.

[nova wasp]

that's a good point

[pliant cypress]

Wait, I'll public my GitHub repo

[thin stratus]

ReplicatedMovement isn't requited for this.

It happens by default for every Actor that is replicated.

[thin stratus]

/**
 * Used for replicating attachment of this actor's RootComponent to another actor.
 * This is filled in via GatherCurrentMovement() when the RootComponent has an AttachParent.
 */
UPROPERTY(Transient, ReplicatedUsing=OnRep_AttachmentReplication)
struct FRepAttachment AttachmentReplication;

[nova wasp]

Just click on the weapon actor and see if it's replicated

[pliant cypress]

UPROPERTY(ReplicatedUsing = OnRep_EquippedWeapon)
AWeapon* EquippedWeapon;

[nova wasp]

THE ACTOR ITSELF... not a pointer to it

also yeah I guess if this is replicating it and resolving it would have to be replicated

[pliant cypress]

void ASoldierCharacter::ServerEquipButtonPressed_Implementation()
{
    if (Combat)
    {
        Combat->EquipWeapon(OverlappingWeapon);
    }
}

void ASoldierCharacter::EquipWeapon()
{
    if (Combat)
    {
        if (HasAuthority())
        {
            Combat->EquipWeapon(OverlappingWeapon);
        }
        else
        {
            ServerEquipButtonPressed();
        }
    }
}

[thin stratus]

void AActor::GatherCurrentMovement()
{
    if (IsReplicatingMovement() || (RootComponent && RootComponent->GetAttachParent()))

[pliant cypress]

these are the code

[nova wasp]

we are asking about the weapon

not just something that points to it

[pliant cypress]

weapon is replicated

[thin stratus]

@pliant cypress The point is that your WeaponActor, if replicated, will automatically replicate Attachment Information and handle attachment in OnRep_AttachmentReplication.

[thin stratus]

If you additionally handle attachment in your own code, e.g. via some OnRep, you are fighting that.

You will need to override the OnRep_AttachmentReplication in your Weapon.

[nova wasp]

okay, then you are fighting attachment replication
you can simply override ::OnRep_AttachmentReplicationto not do anything as a simple band-aid fix

there is a special macro to disable replicating a property manually but I would rather not look it up right now

[thin stratus]

Correct. Other games simply don't use Actors for the Weapons, but have a WeaponMesh they attach, which doesn't have that issue.

But overriding the OnRep_AttachmentReplication is a valid solution imo

[nova wasp]

nah... you can still use an actor just fine

just making it replicated by a separate object

[thin stratus]

Yeah sorry, I meant for the WeaponMeshes and the Attachment part.

[pliant cypress]

i can overwrite but what should i do in overwrite for the solution? because i don't understand the solution

[thin stratus]

For starts you'd just leave it fully empty.

[nova wasp]

just leave it empty

you can just not call the super

why would it need to do anything here?

it might be valuable to ignore it only if it's for a certain actor maybe? I don't know

for example if you know it's for a character, ignore it but use it if it's replicating attaching to a vehicle

[thin stratus]

Your problem is that Unreal Engine takes the Attachment Information, replicates it, and applies it. Means that if your Server attaches the Weapon to the ThirdPerson Mesh, it will do that for Clients too.

But you want to do this manually in your own OnRep_EquippedWeapon, at which point you have to disable the default behavior.
And you do that by overriding OnRep_AttachmentReplication in your WeaponActor and just leaving it fully empty (also no Super call).

[pliant cypress]

i am on ListenGame so whenever a player attach weapon. For the player, it should be FPS mesh but for other including server if sevrer player didn't equipped, should be Third Person Mesh (full body mesh)

[nova wasp]

no... that is not the issue

listen server or not, it will still replicate

[thin stratus]

Yeah that's all good and you are doing that in your OnRep_EquippedWeapon already.

Just override that function we mentioned and leave it empty.

[pliant cypress]

okay, testing that now. Building the code

[nova wasp]

did you actually override it for the weapon?

[pliant cypress]

void AWeapon::OnRep_AttachmentReplication()
{
    // Empty
}

i did

[nova wasp]

I am a bit worried there might be one or two more places that can use attachment replication

but this should help

[pliant cypress]

@thin stratus @nova wasp Man, thank you for the great help

it did work

[thin stratus]

No worries. You can be happy that you are using C++, cause Blueprint peeps are lost on this :D

[nova wasp]

eXi figured it out, I was not even thinking about the weapon 😛

[pliant cypress]

LoL, i'm more good with C++ than blueprints unless it's about small work

[thin stratus]

I didn't even know what the problem was, I just threw in that the Actor does this crap by default :D

[pliant cypress]

one question

how to debug multiplayer?

single player is good, set breakthough and check

[thin stratus]

Gotta be more specific.

[pliant cypress]

what about client and server side debugging

[nova wasp]

I already told you to add the PIE debug name to watches

[pliant cypress]

yes

just changed configuration to Debug editor

[nova wasp]

this is a big subject and there are a lot of different things unreal can do to help here... for starters you can just literally click on things in the editor

[pliant cypress]

debugGame edtor?

[nova wasp]

yes, you need to be in debug/debuggame to debug c++

otherwise all code will be optimized away

you should read the pins in #cpp to get started

[pliant cypress]

ohk

[thin stratus]

When playing in the Editor and you are using a Single Process, which one almost always does, then you can't specifically choose if you want to breakpoint Server or Client. If a function is called on both, then you just have to check if the breakpoint hit on the the one you want to debug and if not, continue until it does.
The Watch that Megafunk mentioned can help with that, as it tells you, in most cases, which one caused the breakpoint.

Not sure how non-Single Process behaves here in Editor.

If you have a packaged build, then that problem somewhat resolves itself, as you'd attach your IDE to either the Client or the Server process anyway.
If you attach it to both at once, then you are back at the Breakpoint being able to call for either.

If you breakpoint functions that you 100% know are only calling on Server OR on Client, then the Breakpoint will also only hit on that one, so that part is straightforward.

For multiple Clients, you'd still want the Watch to know which Client caused it.

[nova wasp]

you can use conditional breakpoints if possible or even just add an if statement that switches on local network mode to make a breakpoint only hit for clients

sometimes temporary debug stuff like that is helpful

conditional breakpoints are unreliable

[thin stratus]

True, that can help. I'm not having much luck with Conditional Breakpoints, in general, though.

[nova wasp]

also a lot of debug printing in PIE can actually be sent FROM the dedicated server to display in other worlds but I forget the details

I would recommend just using the visual logger honestly... it can split categories for each instance in the options

[thin stratus]

If you want to print some value that you know can print on both sides or generally multiple players, etc., you'd want to also print the NetRole if you can (not always available).

[nova wasp]

just use GetDebugStringForWorld

no need to make a new thing

[thin stratus]

Ah, yeah that could also work. I'm just used to printing NetRole.

[nova wasp]

and yeah this stuff can just be hard... it can really help to make dedicated in-game debugging tools

that for example just debug draw where things are

so you can turn a cvar on or something and see more information for a certain thing

I make HEAVY use of visual logging when dealing with network movement code

it helps that way to be able to scrub through things

[pliant cypress]

🙂 great advices. Thanks for this too

[thin stratus]

Yeah I'm quite blessed with NPP/Mover having a dedicated Insights view that lines up the data to frames. I hate debugging network movement code when needing to check values due to ping..

[nova wasp]

yeah I should honestly try to jury rig my stuff into that

might be annoying though

[pliant cypress]

well

i'm making sprint now

in CMC as you suggested

[nova wasp]

yeah, stick with the cmc

you do not want to be fixing issues with mover...

[thin stratus]

Yeah, that Insights View is also not a reason to use Mover haha, stay with CMC:

[pliant cypress]

Sprint only in forward and forward diagonal. Not backward or right/left

[icy trench]

do you have to package your game everytime you want to test a new feature?

[thin stratus]

No, but you should package once in a while to ensure what you think works in Editor does work in a packaged game. Specifically shipping.

[icy trench]

so the dedicated server doesn't need to be packaged again before I test in the editor?

[thin stratus]

No, you shouldn't mix packaged builds and the Editor anyway.
You can play in the Editor with a dedicatd server by just selecting "Play As Client" in the settings.

If you test packaged builds, you should build Server AND Client and test them together.

[icy trench]

alright ty

[pliant cypress]

virtual FSavedMovePtr AllocateNewMove() override;

Is this removed from FNetworkPredictionData_Client?

i'm getting red lines on ths

class FNetworkPredictionData_Client_Soldier : public FNetworkPredictionData_Client
{
public:
    FNetworkPredictionData_Client_Soldier(const UCharacterMovementComponent& ClientMovement);
    typedef FNetworkPredictionData_Client Super;
    virtual FSavedMovePtr AllocateNewMove() override;
};

[nova wasp]

we can't see the red lines, also if you are just not able to compile just ask in #cpp

[pliant cypress]

problem on client side

can't understand?

not able to sprint

[nova wasp]

you need to reason about what is happening

it looks like corrections are occuring so you need to consider if the server version is doing the same thing

the cmc is ulitmately just playing the same saved moves back on both sides and trying to get close, if it doesn't get the same information it cannot resolve the same position

[rotund linden]

Hi! I want to separate the hands in the first-person view of my multiplayer game. In the local view, I attach the fishing rod blueprint to each player's FP hand, and if not, it attaches to the third-person hand. In the FP view, the fishing rod is in a different location than in the TP. How can I replicate the hook on the end of the fishing rod? When I replicate movement of the bp_hook, the hook is in the same place in all players, but because the fishing rods are in different positions, it appears incorrectly in either the TP or FP. Furthermore, because the TP has physical animation and locomotion animations, the fishing rod is constantly moving, while in the FP view, it remains stationary. Is there a way to do this?

[pliant cypress]

Server is working fine.

It's the client side

I did watch cmc architecture video to understand it, but a bit confusing

https://youtu.be/dOkuIvKCvpg?si=uWteKtfNfasL3I8Z

I was following his next part for sprint so after I can implement forward only sprint

Same problem as mine. You just need to override the FAttachment Rules

And leave it empty

So then you can implement your custom Attachment setup

Such as for simulated proxies, you can attach the rod to TP mesh and for player, attach to hand

[pliant cypress]

@rotund linden

Override this function in AActor so you can setup the rod attachment correctly

[rotund linden]

I'm attaching the fishing rod to FP and TP hands there's no problem with that. The problem is replicating the hook at the end of the fishing rod. Because the rod positions are different in the FP and TP

[pliant cypress]

Perhaps the Location of Hit isn't replicating well?

[rotund linden]

The hook is attached to the rod with a physics constraint. I use replicate movement on the hook, and the position is the same for everyone. But because the rod is positioned differently in different views, in third-person view, it looks like the hook is floating in the air. You can see it in the video. This is because there are animations in third-person, but not in FPS and the rod positions don't match.

Since the hook doesn't move in FP, it doesn't move in TP either

[pliant cypress]

Perhaps something blocking it?

Can't know until we see code.

[grand prism]

Bro, the problem has nothing to do with that. I’ll explain it to you in the simplest way because you didn’t understand. In FPS, the hook just stays in place and doesn’t move up or down. In TPS, since there’s an idle animation, it moves up and down, but because the hook is fixed, it looks like it’s hanging in the air. There’s nothing blocking it or anything like that

I think this is a bit too complex of an issue for you. Thanks for your help, but our problem doesn’t have a simple solution or logic

Lol, you’re just making the job look more complicated

[pliant cypress]

It's okay, I thought i could offer some help though.

Gotcha

Perhaps your FP hand is blocking the Rod since it cannot be seen, the collision exists

[rotund linden]

Thanks, I think there was a little misunderstanding

[rotund linden]

there's no collision

i turned it off

[pliant cypress]

Then sorry, it's really a complex for me.

[grand prism]

We’re already explaining to you what’s causing the problem. You don’t need to find it—the issue is already known and it’s written in the messages we sent you. Nothing is blocking the fishing rod

The problem comes from the animation difference between FPS and TPS, lol

[rotund linden]

No problem thanks for your efforts

[crisp shard]

I just discovered that The Finals are doing 100 percent sever side movement and logic and I’m baffled.

First question is how would I be able to even test this myself? Having movement fully sever side for example, and 2 how can a fast paced game like that feel good and be 100 percent server authoritative?

[crisp shard]

Hearing that they would most likely not even be using ue for the server. But this confuses me as I have no ideas how that would be setup?

[thin stratus]

Where did you hear that? I hardly doubt they don't have client prediction

[exotic wasp]

they definitely don't lol

it would feel absolutely terrible

there's no way to have server side only movement not feel laggy with any network latency

[pliant cypress]

can anybody please take a look?

[lament flax]

in the video the finals creative director (iirc the guy speaking on the devlog) said they used server AUTHORATIVE, this doesnt mean there is no client prediction

@crisp shard

[blazing bear]

Looks like your sprinting isn't implemented in CMC correctly

[pliant cypress]

This is my custom CMC. Just turned off my laptop.

Tbh it's bit hard to understand CMC

Prediction data and saveddata

[blazing bear]

The standard way of fixing this is to spawn a local non-replicated version of the weapon actor the server attaches to you, and keep track of it to destroy it when the actor is deattached/destroyed. You do this on OnRep of the currently held weapon property.

But as long as OnRep_AttachmentReplication works for you, it's all good

[pliant cypress]

Yes

Now the sprint is currently my issue. I understand it's server side only

And also Crouch

[blazing bear]

There is a repo by Vaei pinned in this channel, you can study it and see how sprinting is implemented

Or just add it as a plugin

[pliant cypress]

Though I didn't overloaded or made a custom, just added Crouch and checked. Worked fine on server but it didn't crouched on client side. But a jitter like server rejected the client's Crouch validation

[pliant cypress]

Okay

[blazing bear]

This is a good read as well https://dev.epicgames.com/documentation/en-us/unreal-engine/understanding-networked-movement-in-the-character-movement-component-for-unreal-engine#performmovementandmovementphysics

[plain vapor]

Hi, could you please tell me how to fix an issue where, whenever any
client disconnects from the Listen server, the server crashes with this error?
What should I check? I really need help…

Engine version: 5.6.1
Plugins are shown in the screenshot.

The error in the crash reporter:
`Assertion failed: MappedClientConnections.Remove(ConstAddrRef) == 1 [File:D:\build++UE5\Sync\Engine\Source\Runtime\Engine\Private\NetDriver.cpp] [Line: 6918]

BackroomsLostRunners_Win64_DebugGame
kernel32
ntdll`

[pliant cypress]

Guys, I'm bit confuse between what things to do in C++ and in Blueprints.

After fixing sprinting and Crouch, I'll be working on ADS and Weapon Clipping. So I want to understand, since ADS is for local player but the crosshair or aim offset would be affected in future, should I do ADS in Blueprints or C++ ? How do you distinguish what to do when and when to use both together

Network related things to be for sure work in C++

[crisp shard]

Ah that might be it. This is where I heard that. And yes I thought the exact same thing, like there is no way they don’t have clients side prediction lmao. cc: @thin stratus @exotic wasp

[blazing bear]

I just do everything in c++, if you can do it in c++ then why not?

[crisp shard]

i don't know c++ but if i did, i'd bascially make all classes and base systems in c++ and then i'd extend things in blueprint, but this is just instincts as i only make structs and enums in c++ rn... lol

[lament flax]

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

[haughty ingot]

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

(slightly less) relevant.

[split ridge]

noob question: for listen-server multiplayer game using seamless travel from lobby->game, where does one idiomatically store game state ? Controllers/GameMode/GameState are not the same across levels. Feels weird using GameInstance since that feels super-global

(if you're kind enough to answer, please use discord reply 😄 )

[tender idol]

Yes game instance is the only thing that travels. So anything that needs to be traveled should be in game instance.

There are other ways too with blueprint objects and data table maps.

[lament flax]

see https://wizardcell.com/unreal/persistent-data/#persisting-data-across-seamless-travel

other stuff than game instance can "be transfered" between levels when using seamless travel
but as you will see some types are destroyed and recreated, but in c++ there are virtuals that you can implement to copy data

[meager spade]

You can keep actors across travel

Playerstate also has copyproperties

Can be done in BP alsl

Also

[tender idol]

Very cool

[split ridge]

cool thank you all. I suppose I'll identify the host user and keep important things in a struct or something on gameinstance until the session is destroyed.

i definitely read that article btw Fishy. It's amazing.

[meager spade]

You have subsystems

Should use them

[split ridge]

was also looking around in World .cpp to verify how GameMode copies stuff around

[meager spade]

It doesn't really copy stuff across world travels.

[split ridge]

yeah I'll use them eventually right now I'm super green with UE so just figuring things out and setting up the boring parts of a "game"

[lament flax]

things like player controllers are recreated tho

[meager spade]

Yes ofc

But it's fresh

[split ridge]

yeah and can override stuff to copy data there. but was thinking about gamemode stuff, where the "game" is the same but the mode class wil change

[meager spade]

Localplayer doesn't get destroyed

We use copy properties I'm playerstate for seamless travel

[haughty ingot]

Just root every object

[split ridge]

yeah feels lame storing non-player game data with the "server player" tho IMO

[meager spade]

Gamestate also had that ?? Can't remember

[split ridge]

(i'm coming from normal-ass day job programmer background)

[meager spade]

I don't progran

I vibe code

[split ridge]

[meager spade]

Have to be with the times

[haughty ingot]

It’s true, I’ve seen it

[lament flax]

you can travel and ask the world to change the used GM with Game URL option

If not absolute, you can set a new game mode by adding the following option to the URL :
?Game=Full/Path/To/MyGameMode

[split ridge]

been having fun digging around in UE5 source. been not having fun learning how fucked up C++ is as a language after not touching it for 15 years

UE++ is pretty cool tho

[haughty ingot]

Someone here is a huge C++ fan, arguably it’s biggest fan. His name is @quasi tide

Always ping him when you tell him how long you love c++.

[quasi tide]

https://tenor.com/bIqnS.gif

[split ridge]

was digging around in the delegatesImpl code for a bit yesterday 😵‍💫

[meager spade]

C++ is bad really. Es0eciallly compared to modern languages etc. Hence why they ate trying to make it more relevant.

[haughty ingot]

??

[lament flax]

i like c++

[quasi tide]

https://tenor.com/bo0Tj.gif

[lament flax]

(UE c++)

[haughty ingot]

I have a simpler one in my engine

std::move_only_function baby

[split ridge]

yeah UE C++ seems pretty decent tbf.

[meager spade]

It was stale for years last few tears they have really tried yo make it relevant. Instead of being stuck I'm 1990s

[haughty ingot]

Tell that to VS2026

[meager spade]

My phone auto correct sucks

[haughty ingot]

Still looks like a windows 95 application

[split ridge]

was trying to trace whether .Broadcast for dynamic multicast from C++ -> blueprint did any memory copy and utlimately got lost in the sauce

[haughty ingot]

If you’re dealing with Blueprints you’re not wrong like 99% of the time if you just assume there is some copying or memcpy in there somewhere

[split ridge]

yeah it was mostly curiosity. because I noticed the BP events didn't generate event params unless I declared as const TArray<foo> & and got to thinking whether that reference was maintained.

(way off topic here heh)

[latent heart]

I really wanted that to say 'socks'.

[lament flax]

Dump on some of my notes on interesting parts of UE networking

• General UE networking resources
• About NetDriver
• About the replication process
• About Push Model
• About Dormancy and relevancy
• About Channels
• About Fast Array
• Uncomplete list of net types

Keep in mind those are updated over time as I learn more

[split ridge]

oooh nice

[lament flax]

@thin stratus might be worth a pin

[split ridge]

Another followup, this section talks about using an AInfo transferred between different gameModes: https://wizardcell.com/unreal/persistent-data/#5-getseamlesstravelactorlist

[pliant cypress]

Actually UE was designed to work with C++ and Blueprint both together

[pliant cypress]

I've seen this

[blazing bear]

You're not forced to use it though, I mean there are things where it makes sense not to use c++ like in Anim bps. It's probably a matter of preference shrug

also bp nodes trigger my ocd lol

[pliant cypress]

Yes

Though my most of the work is done in C++

And I prefer it too

[pliant cypress]

I once made an escape room in blueprint only

End up making spaghetti

[thin stratus]

AnimBPs are a funny thing to mention as their EventGraph (not AnimGraph) etc. is so expensive that Epic added a checkbox to notify you that you should be doing stuff in its C++ parent instead.

[blazing bear]

Can I do the multithread thing in c++ though?

forgot what it's named

[thin stratus]

That's AnimGraph stuff

It's more about the GameThread EventGraph iirc

[blazing bear]

on my anim bp I got a few functions that access properties

ah I see

[blazing bear]

Never knew this though. I'll keep that in mind when I revisit my anim bp as it needs big refactoring

[narrow prairie]

i use states for my animbp and let that just do the work instead of animation blueprint update node

exi is a god so he probally knows more as me, but that seems to work for me

need to learn c++ for the animation sharing plugin though

my animations are not that important for the style im working on

[grand kestrel]

You might benefit from reading my wiki
https://github.com/Vaei/LocoTips/wiki

[fallen fossil]

is there way to make interface functions with Server / Remote flags?

🤔

[viscid heart]

Doing anim logic in c++ is so much nicer than BP. Just collect the data from GT in NativeUpdateAnimation and then do the logic in NativeThreadSafeUpdateAnimation and just use Live Coding when iterating. Wonder who started the meme that you need BP for animation logic and now every noob parrots it. Probably someone who doesn't even make games.

[viscid heart]

"Worker Threads update asynchronously away from the Game Thread." Thats not true. There is a single AnimThread. That will be bottleneck for GT for example when doing BP only procedural animation. Luckily C++ can solve it.

GT will be waiting for that slow ass BP.

[gilded vapor]

The slow-path for animbps truly is distinctly slower

[crisp shard]

Random Q, but is tile based movement possible in unreal? Similar to RuneScape? Point and click but there are set “tiles” for movement

[latent heart]

Everything is possible.

[crisp shard]

Lmao was pretty much typing this

Guess I’m asking “how” but that’s probably not a simple answer

[latent heart]

When you click, trace to the landscape (or whatever) and round to the nearest tile and tell your pawn to go there.

[crisp shard]

Seems the answer was simple lol

I guess you’d just have to manage tile size and really that’s that

[latent heart]

Pretty much

[crisp shard]

Not even manage, just decide tile size

[latent heart]

How you move the character and what you move them on is a bigger question, really.

[crisp shard]

Yea, like timeline it / lerp / etc to get to the desired position. Numerous ways all with probably pros and cons. And of course movement would be 100p server side at that point

[zealous wigeon]

lots of iris updates recently !!

[crisp shard]

curious about iris, but also don't want to go through the process of converting what i have to iris for replication... but also don't know how difficult or what that would entail

[zealous wigeon]

They've been updating the multi server plugin to work with Iris. Def the future.

I don't know enough about rep graphs to know.. I've only recently started learning unreal

[compact flame]

How do I replicate variables from player state to the server? sever does not read the correct client set info.

[blazing bear]

That's great! Noted. It's really the only thing I'm doing in bp which is ironic.

[dark edge]

Frame everything in terms of your tile coordinates, then convert to world dimension on demand for when something needs to be in the world

My units location is FIntVector2(12,32), that means it should be in the world at FVector(1200.0, 3200.0, 0.0)

Just have some functions to convert from world to tile and back and voila

the world to grid conversions already exist

but it's trivial, it's just a divide and round

[crisp shard]

yea thats a simple way to frame it, thanks for sharing that

are you doing this yourself ?

for your project/game ?

[dark edge]

https://github.com/AdrielKinnunen/SpectrelightPlugins/tree/main/Plugins/SLTilemap/Source/SLTilemap/Public

Kind of

not the tile based movement but an abstract tile representation that gets turned into 3d stuff

Could easily add movement stuff to this tho

[crisp shard]

bit of a personal take, but i am slightly rethinking all of my movement mechanics, as i feel realtime movement done properly is going to be a bit more difficutl to get correct in the way i'd like. i have solutions avaialbe like GMC and maybe getting into custom movement component stuff, but it's just out of my league at the moment and im kind of thinking of changin movement completely.... but i could just be tweaking

[dark edge]

oh btw if you want a nice queryable way to ask "what is at this grid location" try a map of IntVector to Thingy

then it can be sparse

My one recommendation if you go a grid based approach is to have the visuals FOLLOW the grid. That is, moving a thing should be setting its grid coordinates, which happens instantly, then the visual stuff happens in response to that.

A bit like a chess game, your piece instantly moves, but the cosmetics layer shows a nice sliding animation etc.

It's kinda like a model and a view

the model is the abstract "game" which is the grid representation etc. The view is the 3d actors and animations and timelines

[crisp shard]

hm, so you mean, i'd click location to move to and (on server) the player/character immediately is "teleported" there, and then the visual representation is essentially "faked" in whatever fashion

[dark edge]

Sure, assuming it's a game where that can be a thing. Otherwise it might step there 1 tile at a time, or whatever. Point being, the true state of the game is the grid representation, not what's visible.

The query "where is my character" shouldn't return some partial grid coordinate because they're animating right now

as far as the Game™ is concerned, your character is at an integer grid coordinate.

Anything else is just cosmetics on top

[crisp shard]

yes, i actually love this appraoch, in theory. it would just be something i'd have to decide on, cuase it would be a significatn change

[dark edge]

What's the game?

[crisp shard]

dealing with latency and movement w combat is just ... very challenging to get to feel good

[crisp shard]

like the game im working on ?

[dark edge]

yeah

[crisp shard]

tbh, i couldn't really answer this question. which might be a bigger problem lmao. but it's an open world with the standard RPG like mechanics, location based states (safe zones, world zones, and a "wilderness"). pvp combat, you can start different game modes as well, a duel arena mode (1v1, 2v2, etc) and another mode i call "release", essnetially like capture and release but 2 sides and they both have their own objectives. these games can be started in the open world and it will "block off" those areas during game

i think i've gottent o the point tho, where i need to scale back a lot of things, and hone in on what i want . but movement is something that really bothers me, and i dont know if it's the movement or like i said above, im just tweaking lol

it's set in a "suburban" like area tho

[dark edge]

How much of that works right now

[crisp shard]

everything i mentioned works

[dark edge]

Yeah it all depends on what the combat is meant to feel like i guess

that doesn't scream grid based

[crisp shard]

there's some more than that, but they all dont connect yet, so hard to say they are importnat to the overarching gameplay as of yet

[crisp shard]

right. what i love the most is how i've built the scaling of levels and combat items, but they are super simple but everthing can be upgraded and you can levelup skills, and the combo of this does add some interesting dynamics to combat, but i feel like i've based it too much off of what i loved about runescapes combat, without it being runescape based combat. so, it's realtime, and i just don't think it feels that great ? lol

[dark edge]

maybe look at CRPGs

what's the viewpoint and controls?

[crisp shard]

it's third person pov, and characters are actually 2D sprites, but in a 3d world. controls are semi standard, although i do have flying... which of course would not work in a tile based movement setup lol

[crisp shard]

i'll have to peep what that is

[dark edge]

Baldur's Gate etc

[crisp shard]

ahhh, i've actually never played

[dark edge]

flying can work tile based

Position = 3,4,1 is flying

3,4,0 is on the ground

3,4,-1 is below ground (somehow)

[crisp shard]

lmao "somehow"

[dark edge]

or underwater

[crisp shard]

that makes sense tho, and if you saw my current world, it's probably not the most "tile based movement friendly"

as, im sure you'd probabaly want a semi flat world

[dark edge]

yeah I don't really feel like tile based is the play here tbh

click to move or continuous movement?

[crisp shard]

it's probably not, like i said about... prob just tweaking

[crisp shard]

currently? you can just move freely like any typical client controlled realtime movement game

[dark edge]

I'd just keep that and tune it if that's what the thing was designed around

click to move or grid based movement can work but they really need a lot more depth in the other systems to be attractive. Nobody wants to play grid based Mario for example

[crisp shard]

yea, i think i do need to fine tune the movement in general, which is compltetly standard as of now apart from sprinting and flying (probably incorrectly) done in cpp/custom movement component

[crisp shard]

yea 100 percent. best part about what i like about the tiled based setup is simply the fact that latency is a non-issue at that point. or in other words, there's not client prediction / "making it feel good" for all of things that the client "thinks" it's doing

and real time combat / etc in theory is more fun, but the output of dmg, and reciving dmg, is not as simple to make feel good . at least not like the games i've played

bascially struggling with the most common of issues, and it's nothing that crazy, but i do want to get movement to feel better and make events like combat be less about crazy movement and more about levels and items

and nothings broken, im just being critical i think

im realizing im semi venting here gentlemen, appreciate the discourse tho

[gray blade]

hey yall im looking for networking advice, tldr I have a randomly generated dungeon and im spawning and destroy room actors on the server and then looping them to replicate with a delay. I pretty much have put delays in everything but after it finishes it cause the server to freeze which I presume the host sending a backlog of actors to the clients. Im just looking for any advice to go about solving the freeze

[quasi tide]

Just replicate the random seed and let everyone generate the same level layout locally.

[gray blade]

id have to rebuild the whole system because its a collision based try and test generation sys

[fierce forge]

Can almost guarantee that it's a rewrite that's worth doing. Not only will you get rid of your current issue, there'll be a LOT less to replicate.
The only thing you need then is for each client to tell the server when they've finished their respective generation, so that the game doesn't start until they're all ready.

[dark edge]

I'm thinking about doing something like that, is there some common way to do that but get replicated actors?

That is, the seed results in some door actor being placed somewhere. Can it be replicated, just as if it was placed on a saved map in the editor?

[full knot]

Hello friends! Sorry this is probably extremely simple, and yet I hit a wall
When a player is eliminated, I want them to unposses their pawn and possess a spectator pawn instead. This has to happen from the server's gamemode, which decides which player should spectate or not depending on the situation

There seems to be no easy way to do it:

• Changing possession of the controller on the server does nothing, because the controller is not replicated
• Overriding player state in C++ to set "bIsSpectator" to true, changes nothing (do i have to do something after that?)
• Restarting the player after setting their player state to bIsSpectator does nothing either

[toxic cobalt]

i have a widget component on all bp_firstpersoncharacter that replicates and is designed to appear for all players. for some reason, the host cannot see these widget components, only clients are able to see it. the clients can see the hosts widget, the clients can see other clients, but the host cannot see any. any idea?

[meager spade]

you should setup proper tests

https://github.com/AdrielKinnunen/SpectrelightPlugins/blob/main/Plugins/SLTilemap/Source/SLTilemap/Private/SLTilemapBPFL.cpp this test is kinda silly 😛

[meager spade]

what do you mean?

server is the only thing that can change possessin

so server just tells the controller to possess X pawn

not sure what the issue is

[full knot]

The issue is it does nothing 🙂
On the gamemode script, I spawn a spectator pawn, then I make the controller of the client possess it - nothing happens from the clients POV. In standalone mode it works, but not in client mode

[meager spade]

then your doing something fundamentally wrong

i cant see your code/bp

the spectator pawn needs to be replicated also

[full knot]

Thank you for informing me I'm doing something fundamentally wrong. I will post my Blueprint shortly

[meager spade]

we have spectators working fine

(though we dont do it in the gamemode..

[full knot]

I'm sure many people do! I'm sure it's very simple to do, actually

This is why I'm confused I've been trying it for more than an hour by now and no matter what approach I take I don't get the expected result

This is a very simple example of what I did that didn't work - we are in the gamemode here
When the player "dies" the server calls this function

Note that I've succesfully broke in there with the debugger and can assure you the server is executing this code

In fact, in standalone mode, this works gracefully. In multiplayer mode, running a client, the client does not possess anything when this code is ran. Whatever possession occurs on the server for that controller, is not replicated to the client

You seem to have a lot of experience with Unreal, so I've read up about how Controllers are not replicated per se but their states are - so I've been supposing that calling possession on the controller is useless from the Serverside because the Server's controller instance is not the same as the Client's controller instance for a same client. This is why I've been trying my luck with PlayerStates instead as they are shared between client and server

However setting bIsSpectator & bIsSpectatorOnly on state, despite that the change is replicated, does nothing more than the BP I've shared above

By all means if you have a fundamentally right way to approach this problem that would make spectators work simply, please share it with me - I believe I've exhausted the first two pages of every google search I could find

[toxic cobalt]

i have a widget component on all bp_firstpersoncharacter that replicates and is designed to appear for all players. for some reason, the host cannot see these widget components, only clients are able to see it. the clients can see the hosts widget, the clients can see other clients, but the host cannot see any. any idea?

[fallen fossil]

It seems that you spawn this component locally

[haughty ingot]

Starting from the top, and I apologize if these questions seem basic, simple, or have been asked.

1:) Can you show that the server is the context that’s calling this function?

2:) Can you show that the player state you’ve passed in is the correct player state?

3:) Can you confirm if the spectator pawn is the one that’s included in the engine, or one of your own?

What actually happens when you call this function? Nothing? Does the viewport of the client change? Do you get errors, warnings, etc? It’s difficult to debug with the limited information you’ve provided

[full knot]

Thank you for your answer, no problem 🙂
Sorry however but since I've been at it for hours I've ended up finding another solution for my problem - I simply unpossess and set the view position on the client at the desired position (it's a static shot of the level). It was simply too much hassle to make it work and no resources I could find online provided relevant help 🙁

• It was indeed the server calling that function, because it's Gamemode - and that is only executed on the server (clients get side effects & state changes)
• There is only one player whose state gets passed so it couldn't have been another one - moreover i was monitoring the two instances of the game on the same screen so i would have seen if another player suddenly spectated
• The spectator pawn was one of my own but the one from the basegame also did not work

Hopefully if somebody else stumbles into the same problem they'll have more luck fixing it than me

[toxic cobalt]

shouldnt i?

i found the issue

when i set visibility for the widget, it doesnt work for the hostr

[fallen fossil]

if you spawn things locally, they are local, thats basic rule of replication

[toxic cobalt]

it should be local. each client should only be able to see the ones they can access

but even when the host is in the conditions where it can see the widget, the "set visbility" node doesnt work

i tested it, it runs through the code, its just that it ignores the set visibility for the widget

any idea why @fallen fossil

[tardy fossil]

default spectator class doesn't replicate by default, so when the server spawns it, its not spawning on clients so the clients cant possess it. need to make a custom class the inherits the spectator pawn class and set it to replicate

[toxic cobalt]

Hey friend any chance u can help

Struggling lots

[fallen fossil]

?

[toxic cobalt]

Ok, so

The widget is a component on every BP first person character

It only comes visible to a player when they are close to another player

when they’re far, the widget disappears

This works just fine for all players except the host

For some reason, even when the host approaches another player, the widget still won’t appear

[toxic cobalt]

I’ve narrowed the problem down to this node

The set visibility, it never takes place for the host

@fallen fossil is that helpful?

[fallen fossil]

Where do you run this code?
Where is widget spawned?
Where u want to see widget?

Do you have more than 1 widget?

[toxic cobalt]

No

[toxic cobalt]

The widget should be seen as an actor component so in the screen of the player that sees it

The widget is spawned as a widget component in BP-first person character

The code I sent, where visibility is changed, is ran inside actor

[fallen fossil]

uh, it though you meant components

[toxic cobalt]

“On player widget”

[fallen fossil]

is your widget even visible before hiding?

https://getgreenshot.org/
this app is great for screenshots

[toxic cobalt]

No

It’s not

It gets set visible

When under conditions

It works for clients

Even when I set it default visible, the host still cannot see it

Even when its default visible

[fallen fossil]

which net mode u run ?

Try seeing widget without interaction first. did u attach them to viewpoer?

[toxic cobalt]

Not sure, but it works for clitents

Clients

[toxic cobalt]

Listen server

I’ll send a video one sec

[fallen fossil]

I mean,
Does Client1 changes visibilty?
Do you want to see it on Client1?

since its in components list, it should be on each instance

[tacit furnace]

hi how can i clients run "run on server" events on actor on world

[fallen fossil]

actor calling RunOnServer must be owned by this client

[tacit furnace]

I tryed set owner node but not working

[fallen fossil]

set owner must be executed on server ;d

also actors must have flag to replicate

[tacit furnace]

Owner should player controller or Pawn?

[fallen fossil]

its ownership chain, it depends if pawn is owned by client then its ok

[toxic cobalt]

@fallen fossil left is client, right is host

[tacit furnace]

In need for click events its fixs with player controller

[toxic cobalt]

See how it should be? And how it doesn’t work for the host

[fallen fossil]

Add print commands when u show them

I guess they show somewhere else xd

[toxic cobalt]

I did

I delted them now

[toxic cobalt]

But the print command confirmed that for host, this line of code runs, but doesn’t change viability

[thin stratus]

You should install ShareX to grab proper screenshots and short recordings.

Recordings and pictures like this make it really difficult

[toxic cobalt]

See in the corner the print text confirmed that it ran through the code

Just the visibility didn’t change

[toxic cobalt]

Didn’t know that was a thing, but I can

[thin stratus]

Would be appreciated.

[toxic cobalt]

Does this recording work for now? Or do I need to install the software. The whole point is just to show that for the host, it is activating the code to show the widget, but for some reason it stays invisible

[thin stratus]

What you need to do is share proper screenshots of your BP code that is relevant to this. From start to end.

The video just shows the problem, but without seeing what you are doing we can't really help.

Unless one of us has a 🔮 sitting around.

[toxic cobalt]

Will do. Gimme a moment to collect

[haughty ingot]

Darn, left it at the dry cleaners

[toxic cobalt]

thats most of it @thin stratus

some other small stuff but thats most ofg whats going on here

Or @fallen fossil if ur still available

[thin stratus]

What calls the Trace function?

@toxic cobalt

There doesn't seem to be any replication related logic involved here. Should each player only see the widget if they themselves get close to the actor or should it show up on all of the players if just one of the players gets close?

[toxic cobalt]

Only the involved player should see

[toxic cobalt]

Bp first person character

No sorry

Event tick in the AC_InteractionTrace

And tick is only activated by BP first person character

heres the logic @thin stratus

[fallen fossil]

where do you set Focused ?

[rancid current]

Hey guys,
I've got these physics in my game where getting bonked sends you flying a bit, how can I accurately replicate the position of where the skeletal mesh went? Eventually Clients go out of sync somehow after a few bonks.
Also any tips on making the physics nicer? Currently I am only simulating the pelvis on the server and the rest of the bone physics are done locally, which.. is not very nice to be honest, any one got any ideas to make this more wow? : )

So characters stand up after getting bonked, so they need to stand up for all players at the location where the ragdolled mesh is.

Should I hide my current mesh and spawn a different actor where the skeletal mesh is the root of the actor for specific replication of ragdoll location?

[toxic cobalt]

Uh let me check

this is in the arctor component its just a function inside of it

[nova wasp]

this is a tough problem... one way to deal with it is similar to what you have now where you have the pelvis/torso anchored via a constraint to a ball which is the thing that actually moves

storing the full bone positions would be fairly expensive bandwidth wise so getting the rough area is probably more important

#cpp message
an example of how a commercial game solves something similar

[split ridge]

learning how to wire up server travelling. Got some code running on the listen server to seamless travel.

GetWorld()->ServerTravel(TEXT("/Game/ThirdPerson/Lvl_ThirdPerson"));

Works in PIE, (both with and without enabling seamless in PIE). When I package dev build, I get a failure that it cannot load the map. What have I messed up? (Just using the third person sample map as the destination).

[2025.09.21-22.44.24:178][ 26]LogWorld: Error: Unable to travel to '/Game/ThirdPerson/Lvl_ThirdPerson' - file not found
[2025.09.21-22.44.24:178][ 26]LogGlobalStatus: Warning: UEngine::BroadcastTravelFailure Travel failed, type: ETravelFailure::InvalidURL, reason: "Invalid URL: /Game/ThirdPerson/Lvl_ThirdPerson?Name=Player?Listen"
[2025.09.21-22.44.24:178][ 26]LogNet: Warning: Travel Failure: [InvalidURL]: Invalid URL: /Game/ThirdPerson/Lvl_ThirdPerson?Name=Player?Listen
[2025.09.21-22.44.24:179][ 26]LogNet: TravelFailure: InvalidURL, Reason for Failure: 'Invalid URL: /Game/ThirdPerson/Lvl_ThirdPerson?Name=Player?Listen'```

[nova wasp]

can you open the level normally in the packaged build? are you certain it was packaged?

[split ridge]

i am not certain. I am very fresh and may not have set up things properly

[nova wasp]

also don't write in the level as a string in code if you can avoid it here...

[split ridge]

yeah I was thinking about wiring it up as a BP reference but to start I wanted to just test it out.

how do people normally reference levels to CPP code

[nova wasp]

TSoftObjectPtr<UWorld> I guess

you could also just use the asset registry to find all worlds and list them out, whatever

[split ridge]

do most just normally do this in BP graph?

[nova wasp]

If that's easier, sure? This kind of thing is totally fine to do in BP if it exposes what you need

[rancid current]

oh, thats a neat idea, imma look into it 😮 thx

[split ridge]

(btw turns out it wasn't being packaged, and for now I just explicitly added it in packaging settings)

[toxic cobalt]

@fallen fossil @thin stratus any help further or do you guys not really know what to do

[thin stratus]

Where is the setup being called from?

Pretty sure you call this from BeginPlay, right?

And that would be too early for your IsLocallyControlled check.

[thin stratus]

If it turns out that you are indeed calling it from BeginPlay, then I can tell you why it's affecting the Server but not the Clients.

1. For the Server and Clients, the Pawn/Character is first fully spawned by/on the Server.
2. Then it will be possessed by the PlayerController of the individual player, also on the Server.

Now, BeginPlay works slightly different for them.

• For the Server, BeginPlay will call after 1. but before 2. This will lead to the Pawn/Character not having a valid PlayerController on BeginPlay, so IsLocallyControlled will return false.
• For the Clients, BeginPlay will most likely call when the Pawn/Character gets locally spawned from replication, but the key difference is that the PlayerController will already be set, as the whole "Spawn + Possess" happened in the same frame on the Server and replicated down to the Client as part of the initial bunch.
  • There is a chance that this does in fact not guarantee that the Client will have a valid PlayerController in BeginPlay.
  • BeginPlay, on Clients, is controlled by the GameState being valid and having "InProgress" as MatchState replicated. That will trigger the general BeginPlay on the World. Actors that are newly spawned will check if the World has begun play already, and if so call their own. Otherwise, they wait on the MatchState kicking things off.

The result is, that your Server can't deduce if the Pawn/Character is their own or not.

Solution: Use "OnControlledChanged", or whatever the event is called, to check if the Pawn/Character is locally controlled, or rather as an entry point for your Setup function.

@toxic cobalt ^

[mystic estuary]

I've heard some people saying to use rigs for that as it's possible to replicate less information as rig controllers will control several bones at ones. Though I haven't seen anyone doing that for anything complex like a humanoid skeleton

[nova wasp]

that's not a bad idea if you just send individual limbs

but it just seems a lot easier to just have one thing

[mystic estuary]

Jambax mentioned that they also replicate velocity... How do you get velocity of a bone?

[nova wasp]

if it's simulated it has a velocity

and if you can't get the velocity from the body instance somehow, you could easily just store the previous position and compare the distance covered over time

of course if you have to teleport this value must account for that

[mystic estuary]

I guess I should've tried searching for the keywords under the skeletal mesh component first 😅

Yeah, that works. I'm doing a little bit of ragdoll replication for my game as well, and I was only doing position and location, and it didn't look that good. Maybe replicating velocity will help it a little bit

[toxic cobalt]

I’ll try this. Again, the clients work fine, just the host cannot see the widgets properly (clients can). The host CAN see the regular actors widgets(button and doors) just not the clients.

Would adding a delay onto event begin play change anything?

[thin stratus]

Delays should never be used to fix stuff in Multiplayer.

I already told you what you can do instead of using BeginPlay.

[toxic cobalt]

Are you should that will work? Consideringt things already work for non-character widgets

[thin stratus]

Are you calling the setup code on BeginPlay?

[toxic cobalt]

Yes

It’s the same code, it’s all one system for interactions

[thin stratus]

Then the change is gonna have some effect either way, even if it doesn't fully fix your problem.

You shouldn't check IsLocallyControlled on BeginPlay.

[toxic cobalt]

Ok

[thin stratus]

If the issue or part of it still persists, then we can check further.

[toxic cobalt]

Is “oncontrolledchanged” a built in function for unreal? Or how exactly would I make them myself

[toxic cobalt]

I’ll fix it and let u know

[thin stratus]

It's built in. Pawn/Character should have it.

Not sure about the exact wording, but something with Controller Changed.

[toxic cobalt]

Ok, let me implement that

[thin stratus]

Unless you are using a really old UE version..

[toxic cobalt]

Thanks so much bud

[lament flax]

[rustic ore]

Hey all, what's the best way to spawn a non replicated actor with the actor bReplicates set to true by default?

[thin stratus]

Come again?

[thin stratus]

Maybe I deemed it unworthy? Maybe I didn't see the ping? Maybe I forgot? Who knows.

[rustic ore]

{
    bReplicates = true;
}```
spawn this ^ actor without replicating it

[lament flax]

i got my suspicions

[thin stratus]

Set the boolean to false.

That's controlling the replication after all.

[rustic ore]

I also want to replicate it in some cases.
Do I need to make a new method like:

{
  bReplicates = newReplicates
}``` 

And call it after it gets spawned?

[thin stratus]

There is SetReplicates already.

Yeah, you could do that.

If it's decided by the spawner, then you can try spawning it Deferred and then setting that based ons ome condition.

[sage lance]

maybe you could make a child class where the only difference is that it doesnt replicate. spawn that when you dont want rep

[thin stratus]

AFPSProjectile* const NewProjectile = GetWorld()->SpawnActorDeferred<AFPSProjectile>(...);
NewProjectile->SetReplicates(SomeCondition);
NewProjectile->FinishSpawning(...);

[pine cloud]

Hey, has anyone played around with Unreal's replay? The DemoNetDriver. I thought it'd replicate multicast events during a replay but they seem to not be triggering. It's for a singleplayer project though. I tried to see how they implemented replays in the Lyra project but it's too complex for me to understand.

[lost inlet]

Well it's not going to be easy since replays will only really work if your game fully supports networking. For multicasts, which actor did you choose because I vaguely remember having to override GetFunctionCallspace (or something similarly named) to get it to work

[pine cloud]

I'm calling custom events (set to multicast) in my thirdpersonchar in hopes that they'd get recorded into the replay. The movement replays well, and I managed to trigger some event using repnotify. But i hoped there was a simpler way with the multicast... Having a repnotify for every event doesn't sound like a good aproach.

I saw this too: https://forums.unrealengine.com/t/multicast-not-called-in-replays/381362/2

[plucky geyser]

I'm currently creating a multiplayer project which involves frequent changes in speed (sprinting to crawling to walking etc). I was changing walk speed by setting 'max walk speed' as you would in single-player, but I noticed that when i simulate a bit of ping, you get rubber banding , because this is server-authoritative and not client-predictive.
However, by using the character movement component's crouch function, there is in-built client prediction, so the change in speed is smoothed out.
All I want to do is add some more movement modes, which function like 'crouching', and are identical to walking but have preset max speeds, and client-predictive speed transition when you swap between them.
Unfortunately, it looks like you can only add these by editing your character movement component in c++. If anyone would be willing to walk me through that process, I would really appreciate it - I feel like if you know what you're doing it should take 10 minutes, but I don't so I've been at it for days and chatgpt hit a wall!

[lost inlet]

Now I'm at my PC, this is a very old post. The problem I was thinking of if you're sending RPCs to the demo player controller - that did need some extra logic to get working

Otherwise can't say I've had an issue

And for that specifically, was overriding CallRemoteFunction

[twin vessel]

How can i lower the fps threshold for cmc to combine moves?
My laptop does not allow going over 60fps and i need to test if combining moves works fine

Edit: should be GetClientNetSendDeltaTime

[dry dove]

anyone working with custom c++ server?

[plush mauve]

Does anyone happen to now, if I replicate a large array of class references, how much data does each class reference actually take? Is it just a memory address? If so how does that work when replicating, given the server and client will have different memory addresses?

[fossil spoke]

Its an Network Guid that is actually transported

Which is mapped to Objects

These are always the same for on disk Objects

[nova wasp]

In iris it will send the full class path once and then negotiate a shared tiny ID

it's not just a memory address... it has no clue what the memory is on the other side

so it must resolve the path

at least all asset-style classes will work that way afaik

You should just use network insights to measure to see if it's even an issue

[plush mauve]

Ok so even if an actor of that class isn't spawned in the world, it still has a network guid?

[nova wasp]

If you do find that there is too much data being sent (for some reason) you could just try to bundle multiple assets together under one asset if possible

[fossil spoke]

See what Megafunk wrote in addition.

If it hasnt been mapped yet, it will send the full path

This allows the engine to load the class first

If it hasnt been used yet

[nova wasp]

It would be unlikely for this to be a big cause for bandwidth issues unless you have thousands of unique things that all must be sent at once imo

or have really restrictive bandwidth needs

[plush mauve]

OK thank you both! Much appreciated.

[nova wasp]

One thing I do wonder is if you could basically "pre-heat" the mappings

but I really doubt there's much need for that

If you wanted to you could make your own ultra simple ID to class mapping

[fossil spoke]

Precache them so they are on disk?

[nova wasp]

more so just do the path export earlier before the player spawns in

so they spread out the cost more

[fossil spoke]

I think the best you could do is just load in assets before they need to be referenced

[nova wasp]

yeah probably

[fossil spoke]

I know thats what we do

[nova wasp]

and yeah the forced "we need this loaded RIGHT NOW" is a far greater problem than bandwidth of sending the class path

[fossil spoke]

Like, we preload all Weapons, so they dont need to be fetched during flight

[nova wasp]

that's a good idea

that way a new one doesn't appear by surprise and force a hard load

[fossil spoke]

Its much easier to do that now, since RAM is pretty inexpensive.

[nova wasp]

Unfortunately I am working on a PS4 game lol

[fossil spoke]

Yeah, well what prompted us to do it was that we came across an issue where some of our Weapons would not load on Clients when they were replicated by SImualted Proxies.

We found that they were just never being loaded by the Class getting repped

So we just preloaded them all

Which has other benefits.

[nova wasp]

Iris does have a weird extra mode where you can actually still async load referenced packages, they just defer the entire packet payload (fairly complicated but it's just storing the data for later)

[fossil spoke]

Yeah righto

Im not sure how it all is in UE5

[nova wasp]

but even in that case you can still have mistakes happen

some things just force a hard load

[fossil spoke]

Id be curious to know if our issue with Client loads is fixed in UE5

[nova wasp]

GAS gameplay cues are a big source of "oh god spawn this right now"

for example

depending on how they are set up

[fossil spoke]

Yeah I hate that about GCs

[nova wasp]

yeah I'm not sure how it works exactly here... I kinda assumed it would force a reference

[fossil spoke]

I never managed to get ot the bottom of why it was happening, so cant say if it was a bug or just an order of operations issue on our end or something...

[nova wasp]

yeah tbh I can't blame you for just pre loading them

even if it's a bit of a bodge it fixes multiple issues

[thin stratus]

I heard the current series of consoles still has limits due to their shared RAM setup. So one still has to be cautious about loading too much when wanting to release on consoles

[fossil spoke]

Yeah I made a big generalization there, completely ignoring consoles.

[thin stratus]

Understandable

I just always get reminded of trying to add a second nav mesh for The Ascent only to face the harsh truth of 300-600MB RAM being lost xD

The previous Xbox generation wasn't happy about that idea.

[spring lotus]

Anyone have any tips for me to get smoother movement in UE5.6 multiplayer? I'm finding that clients are stuttering around when their frame rate is higher than the server tick rate (which I've set to 60).

I've tried to increase NetworkMaxSmoothUpdateDistance & NetworkNoSmoothUpdateDistance to 200 and 300, but it's not really helping.

Capping the players frame rate to 60 entirely fixes the issue but I'd prefer if the player could have whatever frame rate their PC lets them run at

[thin stratus]

Custom movement code or just the unchanged CMC?

And is it the movement that stuttering or the animation?

[gritty nebula]

hey does anyone know of any resources teaching how to make stuff your pawn can enter and control like cars, tanks, jets, and mounted machine guns? Bonus points if it uses gas!

im sure i could figure out something on my own, but i want to know what the best practices are for things like these beforehand!

[spring lotus]

Custom movement where I orbit around an actor, no animation plays, just that the set actor location and rotation is super stuttering on higher frame rates

void UPlayerAimComponent::ApplyOrbitAtAngle_Common(APawn* P, const FVector& PivotLoc, float AngleDeg, ETeleportType TeleportType)
{
    if (!P) return;

    const float Rad = FMath::DegreesToRadians(AngleDeg);
    const FVector2D Offset2D(OrbitRadius * FMath::Cos(Rad), OrbitRadius * FMath::Sin(Rad));

    FVector NewLoc(PivotLoc.X + Offset2D.X, PivotLoc.Y + Offset2D.Y, P->GetActorLocation().Z);
    NewLoc.Z = FindGroundZForPawnAtXY(NewLoc, PivotLoc.Z);

    const float YawToPivot = (PivotLoc - NewLoc).ToOrientationRotator().Yaw;
    const float FinalYaw   = YawToPivot + MeshYawCorrection;

    P->SetActorLocationAndRotation(NewLoc, FRotator(0.f, FinalYaw, 0.f),
        /*bSweep=*/false, nullptr, TeleportType);

    if (ACharacter* Char = Cast<ACharacter>(P))
    {
        if (UCharacterMovementComponent* Move = Char->GetCharacterMovement())
        {
            Move->bOrientRotationToMovement     = false;
            Move->bUseControllerDesiredRotation = false;
        }
        Char->bUseControllerRotationYaw   = true;
        Char->bUseControllerRotationPitch = false;
        Char->bUseControllerRotationRoll  = false;
    }

    if (AGolfPlayerControllerBase* PC = GetOwnerPC())
    {
        PC->SetControlRotation(FRotator(0.f, FinalYaw, 0.f));
    }
}

[thin stratus]

Are you doing anything in your CMC itself with SavedMoves?

[spring lotus]

Nah, I just call this function through these local and server functions, was hoping the local would fix how it looked on the client

void UPlayerAimComponent::ApplyOrbitAtAngle_Local(float AngleDeg)
{
    if (!OrbitPivot) return;
    if (APawn* P = GetOwnerPawn())
    {
        ApplyOrbitAtAngle_Common(P, OrbitPivot->GetActorLocation(), AngleDeg, ETeleportType::None);
    }
}

void UPlayerAimComponent::ApplyOrbitAtAngle_Server(float AngleDeg)
{
    if (!OrbitPivot) return;
    if (APawn* P = GetOwnerPawn())
    {
        ApplyOrbitAtAngle_Common(P, OrbitPivot->GetActorLocation(), AngleDeg, ETeleportType::None);
    }
    if (AActor* Owner = GetOwner())
    {
        Owner->ForceNetUpdate();
    }
}

[thin stratus]

Right, well, the 60FPS limit is usually combined moves, but if you aren't doing anything in the CMC, then that's not it.
The code you shared seems to be modifying the Transform of the Character, which is not really allowed outside of the CMC:

You are affecting the movement simulation with that and that causes corrections.

If this is all ControlRotation based, then you should only call this locally and only change the ControlRotation. Then set the Character to follow the ControlRotation. The rest should work automatically. There is no need to manually update the Character Transform or to RPC this.

[spring lotus]

Ok thanks, so I guess my per frame transforms are completely bugging out when the frame rates too high. I'll try change it now

[spring lotus]

Thank you, i can already see it's much smoother using the player controller than just applying transform outside

[spring lotus]

I've just realised, it isn't just control rotation based, the character has to physically move around the ball as well, so maybe this isn't an approach i can continue with? But i do see that simply having the control rotation done through player controller is very smooth on the server

[thin stratus]

If you need to move the Character predictively and it's not possible through setting MovementInput and ControlRotation, you'll need a custom CMC most likely.

[radiant halo]

I could use some help with replication. Cause different things are occuring on the host side than the client side. If anyone have some time to spare i would appretiate it a ton. Working on a school assignment.

[mystic estuary]

Hello, how would you guys go about attaching a character to another character in a way that the two characters wouldn't move too far away from each other (a few meters), and also let character A to move character B closer when character A gets too far, but not viceversa? E.g. constrain the two characters together, and let one of them drag the other one if they walk too far away?

[radiant halo]

i would maybe use the attach actor to actor possibly

[mystic estuary]

That way one character won't be able to move at all. I need the two character be able to move, but not too far away from each other as they're bound together

I'm talking about something like Chained Together, but with one character completely dominating the movement when the chain is stretched the most

[radiant halo]

what about some sort of tether like an invisible cable component?

[mystic estuary]

I don't think that cable component affects the CMC movement in any way?

[radiant halo]

maybe not

[mystic estuary]

I could always use normal physics constraint actor, but I don't think that it's deterministic, so it will cause desync between simulations. I'm not sure, so that's why I'm asking for opinions

[sage lance]

maybe you could add a hitbox to the characterA with collisions setup just for charB
..

looking for some networking wisdom: if i put my maps in a struct will i be able to replicate them properly?

[mystic estuary]

No, use array with structs for that

[mystic estuary]

Could you elaborate?

[sage lance]

like a giant invisible box with collisions for your characters

[sage lance]

its not array oriented data

[wicked whale]

Hello everyone. I'm facing a weird issue with SetActorRotation in my multiplayer tests. I have a listen server and a connected client in a PIE session. If I call setActorRotation for the player in the listen server, it works as expected and both the server and the connected client see the correct rotation. But if that is called for the player in the connected client, the rotation looks correct in the client, but not in the server. I do see a rotation change in the server but it's completely wrong. I added a log right after the call to SetActorRotation, and in the log the rotation has the right value in both client and server, which is even weirder. Does anyone know what can cause that?

[stoic lake]

How would I go about playing a sound on an actor component that's attached to a character, only for the client that owns the character. Without having to go inside the character to call an owning client event

Im running into the component being owned by the server, regardless of what its attached to

[wicked whale]

Actually, nevermind. My issue does not seem related to SetActorRotation, but with an animation that plays right after it. It's still a bit weird since this animation plays only in the upper body and has absolutely no rotation at all, but at least that gives me something to investigate.

[dark edge]

Very difficult

Not difficult in concept, difficult to do that in the framework of the CMC and make it play nice with all the prediction and correction going on

[dark edge]

What are we supposed to be seeing here

[bronze meadow]

I’m not sure if this is multiplayer or online subsystem but I’ve heard the project cannot be online multiplayer if the projects is blueprints any validity

[lost inlet]

2nd "but I've heard" today. of course you can do multiplayer games in BP-only projects. not like I'd recommend it, and it's impossible to do any custom predicated behaviour with the CMC

[radiant halo]

Nvm, it's fixed

[dark edge]

You can do a multiplayer game in BP only but it'll be a tough go, depending on the game

BP only Advance Wars, sure

BP only Battlefield, not happeneing

[bronze meadow]

Yeah no nothing complicated more like that company game standard wise

[bronze meadow]

You know the company where players are searching facilities trying to make it out I never played it but watching it happen seems pretty cool’s but I want people to be able to play with friends and such so I think that’s a good starts
Sorry lethal company

[ember vine]

can you use an FInstancedStruct for some polymorphic rpc pipeline ? or do you always need to explicitly have the struct as an arg

[latent heart]

You can use a base struct that all your rpc structs extend from?

[ember vine]

really ? i can just inherit a ustruct and theres no issues ? and just pass the base type ?

i just assumed it wouldn't be that easy

[latent heart]

I think you can, yes.

Or if you want to go a bit more ham, serialise it all to a TArray<uint8> 😛

We do that with a uint8 as a type id. Cos we just like doing that kind of thing.

[ember vine]

hmm, must be a way to do that while also using the UE NetSerialize functions u put on the struct

[mighty trout]

i am working on multiplayer game that connects player through steam and when i host the game and other player can also join the game but when i leave the game in between and host again clients can find my session but not able to join it ? I don't know why it is happening but when client restart its game it can join the game

[bronze meadow]

Using epic online or steam sessions online

[mighty trout]

steam

[bronze meadow]

I want to get a system like that started 😩

[mighty trout]

what

[bronze meadow]

So other players can join a session or create their own

[mighty trout]

any player can join or host sesssion
but the player who host the session quit the game in between and create another session then player can't join the session but they find the hosted session

[verbal ice]

Does Iris have a way of preventing a player from receiving actors at all?

I.e. I want the player to pretty much have everything irrelevant to it (except its player controller/player state, and maybe game state)

I'm doing some tomfoolery to make the engine think runtime spawned actors are statically loaded from the map (they're not) so I need to ensure the server doesn't try to send data about those actors to the player until they have the full map "generated"

I'm ensuring the names are stable for replication manually, so I just need to make it so the player doesn't get any data until they're ready

[nova wasp]

yes, filters

and net groups

as for this I am curious if you can defer them in another way

[verbal ice]

Yeah I knew about filters but it felt like the same thing as overriding IsNetRelevantFor() in all actor base classes and returning false for the player

i.e. janky, even though what i'm doing is really hacky, I feel like there's a better way

[nova wasp]

I vaguely recall the replication internals having some kind of way of deferring dependant packets but it's more for asset loading

Is the risk that these get sent too early and stuff falls through the map?

Why can't you just receive them and then suspend simulation/visuals until they are ready?

[verbal ice]

The engine yells that it can't find a statically referenced actor on the map

I'm tricking the engine into thinking dynamically spawned actors on the server & client are static/map-loaded

[nova wasp]

"I'm doing some tomfoolery to make the engine think runtime spawned actors are statically loaded from the map" - why? Is this to make them a stable name for speed reasons? to have them spawned locally?

[verbal ice]

I don't want them destroyed when irrelevant

I only want replication to stop, just like map loaded actors

[nova wasp]

hmmm... you can override actors being destroyed for network reasons to just not do that

you can tear off instead etc

I'm thinking there has to be a nicer way to do that that isn't outright fooling the engine... this is not exactly a weird thing to do to just filter

[verbal ice]

from some engine digging the only thing that tells the engine not to destroy them is if they're loaded from map/fully stable for networking

which you can manually mark an actor as, but the engine just whines with the whole "failed to serialize net actor" spam on client/server if it can't find an actor you're supposed to have loaded already

that's the part i'm trying to avoid

(+ well, I'd like if it the client didn't have other players/actors in the game running while they're spawning the map from the data they have anyway)

[nova wasp]

let me look around for a second

[verbal ice]

#multiplayer message

that's my research from last time

if it helps

Checked regular replication & Iris' way of doing it and they both have the same checks from what I gathered

[nova wasp]

Iris can deal with making an actor ending replication not be destroyed locally... but I don't know if you can easily make it restore that connection to the replicated object later

[verbal ice]

notable mention to the Epic engineer with the // Fixme: this should be a setting comment, yes I would've loved if this was just a setting on actors instead

[nova wasp]

I am really thinking filtering would just be easier here

[verbal ice]

So just filter out everything for players that are loading/generating the map

[verbal ice]

AFAIK it should just restore it later, it'll assume it's a map loaded actor, just like the engine already does without Iris

i.e. you can have a map loaded actor not be relevant but stay spawned, you just won't get networking updates for it until it becomes relevant again

95% sure at least

[nova wasp]

I'm just worried that making it think it's a stable named object will break other things

I feel like there has to be another way to do this...

[verbal ice]

If it's not a stably named object it'll also try spawning the actor on the client, which is what I don't want to do

I want to network the map data to the client manually, and then have it spawn the map actors with the correct "stable" names the server would have

That way I can also cache the map to a file on the client if it's not gonna change

and just read from it the next time the client connects

TearOff in Iris would kinda work but I don't see/think you can "re-attach" after that

like in regular replication, afaik you can't undo that action

[nova wasp]

I think if you got REALLY cheesy it might be possible to fool the bridge into using an existing object

to resolve the incoming network id

[verbal ice]

I'd still have to ensure the bridge didn't try to resolve before the client finished spawning the map

right?

seems like group filtering might be the way to go

The online documentation seems outdated, Iris source itself is... confusing

[nova wasp]

yeah I would really recommend just net groups and filtering because I think that would definitely work even if it's a bit annoying to have this ack step to wait on it

[verbal ice]

Some old info I had about those seems to be wrong now?

[nova wasp]

Does the map only get generated once per level?

Iris has changed a fair bit. Your best best is to use the actual existing network filters for guidance

[verbal ice]

yeah, the server generates the map/loads it from a file after opening an empty level, players wait for it to do so, then once it has, it sends the data to the players, who then need to spawn it locally

I'm wondering if I can make a filter group that only allows connections that have finished creating the level to receive the actor

And then I just pop that on most actors outside of the game state/player controller etc, anything I didn't lie about or need during that step

What I'm confused about is inclusion/exclusion groups. If I have an inclusion group, is the actor only relevant to connections inside this group?

Is exclusion just the inverse, where the actor is relevant to all connections except those in the group?

Past documentation made it seem an inclusion group only worked to negate a dynamic filter's exclusion

[nova wasp]

https://dev.epicgames.com/documentation/en-us/unreal-engine/iris-filtering-in-unreal-engine#create-a-dynamic-filter

for example

[/Script/IrisCore.NetObjectFilterDefinitions]
+NetObjectFilterDefinitions=(FilterName=Spatial, ClassName=/Script/IrisCore.NetObjectGridWorldLocFilter, ConfigClassName=/Script/IrisCore.NetObjectGridFilterConfig)

also always double check iris docs for ini things... the only real reference is the actual live code/config

they have changed some of these ini config naming schemes in the past

[verbal ice]

I wanted to avoid a dynamic filter since you can only have one and using it for this seems like a waste

unless you can have multiple now

[nova wasp]

Ooh it's one a at a time? argh

I mean

you can just have it do double duty if it must do something else

[verbal ice]

well apparently one object can only have one dynamic filter

yeah

docs also say "don't use dynamic filters unless absolutely necessary since they're way more expensive than group filtering"

which makes sense since groups are just glorified bitmasks at the end

[nova wasp]

I think maybe a group for
Unloaded players
Loaded players

and these special actors that need to wait should only be in loaded players

then move connections that say they are done loading to Loaded players

is my guess

assuming connections can change group membership

[verbal ice]

If an object has just one inclusion group, does the engine default to exclusion for any connection not in that group?

yeah seems like they can

[nova wasp]

yeah I wish I remembered more but I am mostly familiar with the raw packet reading on the client part here more so than filtering

[verbal ice]

Hopefully we get more documentation soon since Iris is now Beta in 5.7

Surprised that's not on the roadmap

Also doesn't help that Iris looks like it's written by a mad scientist

I understand it's squeezing performance everywhere it can but it gives me whiplash every time I try to follow along

[fossil spoke]

https://tenor.com/view/guy-long-hair-mustache-magic-shia-labeouf-gif-17741293

[verbal ice]

It's funny because it's easy to understand what they're doing, but trying to follow along is hard

and then you have stuff like this

[fossil spoke]

Well AFAIK the for(;DeletedObjects;) would be equal to a while(DeletedObjects != 0)

I guess it would be hard to follow along when they do silly choices like that lol

Not sure what the implications are of using one over the other.

[verbal ice]

From what I see in code comments, inclusion groups only override dynamic filters?

 * Inclusion groups are used to allow overriding the effect of dynamic filtering which can be useful to always allow replication of team specific objects for example.

     * Add a group to the filtering system. This group is used only for filtering out objects. Exclusion groups are processed before dynamic filters, those implemented by UNetObjectFilter. 
     * By default an exclusion group disallows replication for all objects in it. Use SetGroupFilterStatus to change the behavior.

this is for exclusion groups

so like, using SetGroupFilterStatus(), I can "invert" what they do?

this is so confusing

[nova wasp]

Iris code is pretty tough to follow because it tends to prefer doing things with raw integer ids and c arrays

the internals are about as tryhard as you can get

they pretty much use 0 unreal containers

[verbal ice]

yeah

[fossil spoke]

Pedal to the metal

[nova wasp]

which sometimes makes sense but sometimes I question if they just didn't know about an existing unreal type...

[verbal ice]

I'm all for it, just wish the docs were a little less confusing

[nova wasp]

which in fairness it's not like I could rattle off every bitset template lol

[verbal ice]

 * By default an exclusion group disallows replication for all objects in it. Use SetGroupFilterStatus to change the behavior.

like what does that do? if I change the filter status, what's the new behaviour?

Epic please my sanity

from what I see in code, if I change the filter status to Allow it just ignores the filter?

super useful /s

like yeah sure maybe if I wanna disable a filter temporarily but rip

Wish I could just poke someone who works on Iris heh

[halcyon ore]

Not sure how to word this question.
But, is there anyway to disable whatever C++ auto fixes replicated actor refs?
Actor 1, with a ref to actor 2.
If you are near 1, but no 2, the var is invalid.
But, when you get near 2, it suddenly becomes valid, without any extra var replication from my understanding?

Or, is this some like innate "feature" of how object references work deep in C++?

[verbal ice]

The engine updates mapped references when they can be resolved on a client

What exactly are you trying to do?

[halcyon ore]

I'm not exactly sure what I'm doing yet.
Its for future experiments/ doing shit.
Just specific vars, not every var, since I know that'd break a ton of stuff.

[verbal ice]

I don't think you can disable that behaviour without some engine modifications

Unless there's a macro definition/cvar you can use for that, but none come to mind

[verbal ice]

After taking a break for dinner, I think I found out that it can be done simpler.

You can have one exclusion group, let's say PlayerLoadStatus, which excludes all connections by default.
Then, when a player has the map fully loaded, you can do ReplicationSystem->SetGroupFilterStatus(GroupHandle, PlayerConnectionId, UE::Net::ENetFilterStatus::Allow);

So yeah I was just being dumb

Just need to find out if inclusion groups override that

As long as they don't, then whatever

[verbal ice]

Does Iris consider all objects as always relevant by default?

[median shell]

I figured out that without a source build, you can still run the custom config in Shipping using the same -customconfig=Steam trick that you use with the editor.

I've been trying to find a way to make it work without the CLI arguments OR a source build though...

Seems like if you're building for multiple Targets, Epic just expects you to be building from source at that point.

[dark edge]

Hell yeah.

They do it for performance.
I do it because I learned a bit of C in 1997 and it stuck.
We are not the same.

[quasi tide]

"How did you make it fast?"
"We decided to not use any of Unreal"

[verbal ice]

TI was right!!!

[exotic wasp]

generic container good enough for most uses but certain applications benefit from a custom made container

SHOCKER

[verbal ice]

wee little bump

[quasi tide]

Search for siliex talking to Cedric about Iris. siliex talks about stuff like this I believe. It was a thread when Cedric was trying to figure out relevancy.

[verbal ice]

found it yeah

kinda helps, kinda doesn't? still confused somehow heh

i'll try it out

[keen turret]

Hello, I imagine this is a fairly common problem, but I'm still a beginner and have been trying to replicate my physics on an authoritative server for several days now. I'm using the Smooth Sync plugin, which works relatively well for all my mechanics, but when my clients' first-person characters try to push my physical objects, they experience constant rollbacks. When the plugin is disabled, my clients can push the objects without any problems, but their movements are jittery. I've tried changing the plugin settings, but it doesn't make much difference. Has anyone else encountered this problem? (i don't know if i should ask this in physics or multiplayer channel)

[verbal ice]

<@&213101288538374145>

[obsidian ermine]

Have you tried only allowing the servers proxy to move it?

[keen turret]

Yes that's the case in the video i sent

[obsidian ermine]

Vid shows your overlap moving it locally, then the server correcting, then the server overlap moving it.

[keen turret]

If i turn off the collisions locally, my client character will move trough the physics cube due to smooth sync interpolating it

[obsidian ermine]

don't disable collision, still needs to block all pawns.

Set your servers tick rate

[keen turret]

Okay done, so to allow only the server proxy to move my physics objects what am i supposed to do ?

[obsidian ermine]

Test it with the new server tick

might not need to adjust anything else

[keen turret]

The problem seems to persist :/

[dark edge]

because on your screen the box is in the way until the server moves it

What's the box's replication setup, just replicate movement and that's it?

[keen turret]

No the Smooth Sync plugin overides the Replicate movement

The smooth sync parameters

If i disable the Smooth Sync plugin and activate native movement replication, i don't have rollbacks anymore but the movement of my physics props avec very jittery

[keen turret]

Smooth sync interpolate and extrapolate my physics objects to have a smooth movement over the network, im guessing i have server corrections because my character is overlapping the physics props server side, but i don't know how to fix this, sorry if i'm being slow, i am new to physics replication and i am having trouble finding documentation

[dark edge]

You can tune physics replication to be less strict

Do you understand the problem though? On your computer, your character is in the future relative to the server's view of the world, but server authoritative movement of the box is in the past.

[keen turret]

Yes i understand that, but the thing is whenever there is even a small amount of interpolation, my client will have rollbacks. Should i do a client side prediction ?

[dark edge]

I would suggest you first have a loose enough sync on the box that the client can locally push it away from where the server is saying it is

doesn't have to be much

right now on the client your character is running into a box it can't move because it's position is constantly being set from the server.

On the server the character is pushing the box just fine

but on the client, the character is predictively moving and running into the lagging box

Say your ping is 1 time unit

at T = 3, you are moving your character. At that moment, the server sees you at T = 2, and you see the box at T = 1

you are bumping into the box

networking is hard as shit, especially when you mix prediction with non-predicted stuff

[keen turret]

Okey i tried to loose the sync on the box but i still had rollbacks, it seems that the plugin is working differently from the network physics settings of unreal,for i don't fully grasp the working of all this networking stuff i might turn off the plugin for now and just use the native tools, it looks like it works better for character collisions. Thank you for your patience. If you have by any chance any good docs to learn games networking i'll gladly take it !

[plush mauve]

When testing multiplayer in PIE, is there any way to force a delay between when the server is started and when the client is started and connects?

[thin stratus]

Not for the first client I believe. But you can manually late join the second

It's a setting you can enable that then adds a button next to the start stop buttons for adding another client

[dark edge]

For testing I really like just making a couple bat files to launch server and client

it's a nice middle ground between PIE and shipping

#multiplayer message

[plush mauve]

That would probably do what I need. Do you know where I can find that setting?

[thin stratus]

Project Settings or Editor Preferences

Late Join or something like that

Just search for it

[ashen plume]

A few cmc questions, recommended way(s) to handle cmc ground friction changes? Saw that i could add surface types to a project wide array somewhere as well i think? Lastly is there a combined friction calc type setting or type?

[pliant cypress]

i'm using this code for FABRIK, TP mesh. should i do the same but with IsLocallyControlled() for FP mesh??

if (bWeaponEquipped && EquippedWeapon && EquippedWeapon->GetWeaponMesh() && SoldierCharacter->GetMesh())
{
    LeftHandTransform = EquippedWeapon->GetWeaponMesh()->GetSocketTransform(FName("LeftHandSocket"), ERelativeTransformSpace::RTS_World);
    FVector OutPosition;
    FRotator OutRotation;
    SoldierCharacter->GetMesh()->TransformToBoneSpace(FName("hand_r"), LeftHandTransform.GetLocation(), FRotator::ZeroRotator, OutPosition, OutRotation);
    LeftHandTransform.SetLocation(OutPosition);
    LeftHandTransform.SetRotation(FQuat(OutRotation));
}