multiplayer

[verbal ice]

no

That's one downside of those

You'll want your data to contain an index of sorts if you need order @pallid stone

[chrome bay]

yup no garauntee, if anything you should expect them to be in a different order

fast array is optimized towards the use of RemoveAtSwap so you don't have to constantly shuffle huge blocks of memory around

[haughty ingot]

I definitely recommend reading the behemoth of a command at the top of the fast array serializer file, I’m pretty sure it mentions specifically that order is not guaranteed, as well as pretty much explaining replication in general

[pallid stone]

It does thank you. It seems using the tfastarray is not an option as I have a 400kb initialisation for it that is larger than the max 65kb you can send as a packet and needs a custom setup via chunking to send a byte array of the data. The order setup for the replication is just a black box I am struggling to understand so changes to all the stuff I've been trying after my custom intialisation don't match up :/

[thin stratus]

You probably want to post this to #1054845249945616404 instead, even though this hardly qualifies as a tutorial and is more or less an ad for your service.

You, as a user, are basically never interacting with this community. So a blog post that largely contains lists with info from official docs, links to those official docs (of which some aren't even pointing to relevant content), and otherwise just stuff specific to the service you offer, doesn't really seem all that genuine.

I don't mind companies creating blog posts in which they, additionally, point to their services. DigitalOcean, for example, has some really great tutorials. However, their posts are hands-on tutorials, with lots of steps, examples, or even full walkthrough.

The lists at the bottom of your post seem nice as a starting point, although I do have to wonder if some of these points wouldn't benefit from a more detailed explanation or even an example.

Might be worth considering replacing that blog post with a step by step guide on how to profile a Dedicated Server in general, without your service, with images showing actual problems and how to identify them, and only posting your service related info at the end, as a bonus. At that point I wouldn't even mind pinning the post.

[lapis siren]

Deleted. Thanks for the pointers. 😄

[quartz iris]

Hi guys I have a mode where defenders must protect an artifact from being picked up by the attackers, the defenders can move the artifact around the Site but once they try to go through doorways it shows a red box visual indicating they cant go through the doorway with the artifact as they would be leaving the Site. This message and visual only shows for the server though, not the clients, however the clients. can anyone help me make it so it only effects whoever has the artifact carried variable on the player rather than just effecting the server?

This is the code which is on the doorway/sitewall

So i want it to show for whoever has the artifact and is walking into the doorway not just the server

[dark parcel]

@quartz iris thats an obvious wrong case of using Multicast.

Tell me what Multicast RPC does.

[quartz iris]

Tells the other clients

[dark parcel]

Also never cross pins like that. (The blue pin goes to other node from other events)

[dark parcel]

By whom?

[quartz iris]

I tried the on owning client

[quartz iris]

server

[dark parcel]

Exactly

So what happend when a client run multicast rpc?

[quartz iris]

nothing

[dark parcel]

It will just run the function for it self at best.

The only way for client to communicate with server is via server rpc

So if you need the client to tell server to run a function then use server rpc.

There's no instance where client speaks to other client directly.

Unreal uses server to client model.

[quartz iris]

so server rpc to owning client rpc?

from the site wall bp?

[dark parcel]

I need to read again, what r u actually trying to do?

@quartz iris who do you want to send the msg to? Both side or just the defender or just the attacker?

First of all, if that's the case. Client doesnt even need to do a single thing.

[quartz iris]

No just the player that has the "Has artifact" as enabled

[dark parcel]

You can just show the msg locally. Without a single networking if the artefact is already possessed by the time the client hit the overlap.

This will eliminate the feeling of latency as well.

Optionally you can have the server tell the client that posses the artefact.

[quartz iris]

so when they hit the collision i just do the code i already did without any rpcs?

[dark parcel]

If you are going with server telling client to display a msg.

Then do this.

Component overlap-> has authority (true) -> get overlapped pawn -> if has artifact -> run client rpc -> display msg.

[quartz iris]

because I need the visuals/msg to ONLY show for the player who hit the wall

[dark parcel]

Only the player who hit the wall or only the one that has the artefact?

[quartz iris]

Only one player can pick up the artifact so the 1 defender who has picked up the artifact

[dark parcel]

So only display msg when hitting the wall only to the player that picked up the artefact?

[quartz iris]

Yes

[dark parcel]

#multiplayer message

Try that

[quartz iris]

Ok will do, thanks!

[dark parcel]

@quartz iris btw you will need the client rpc method inside your pawn.

Create a function、call it display msg that takes text.

You can then call that function as described above.

This way the function will only run on the owning client (owner of the pawn)

[dark parcel]

@quartz iris work?

[nocturne quail]

this SoftPtr.LoadSynchronous() will work on dedicated server?

for me it gives an error target file not found

target file is the object passed using TSoftObjectPtr

loaded from hard coded FSoftObjectPath

tested with editdefaultsonly TSoftObjectPtr, giving the same error

[humble ocean]

can you please explain the context what are you trying to do?

[nocturne quail]

loading object Synchronously when its needed from a hard coded FSoftObjectPath

[latent heart]

Is the asset packaged on your server?

[humble ocean]

as far i know we do not package assets with server so that is the reason for the error it throws.

[nocturne quail]

I am sure its not excluded in packaged, so it should get packaged?

[latent heart]

If the file doesn't exist, it isn't packaged. Surely.

It's a dedicated server package? Or listen?

[nocturne quail]

Dedicated linux server, and windows client

[latent heart]

Package your server without pak files and see if it's there. Or use a pak explorer.

[nocturne quail]

Yeah its not packaged

LogStreaming: Error: Couldn't find file for package /Game/Montages/StandUnEquipMontage requested by async loading code. NameToLoad: /Game/Montages/StandUnEquipMontage
LogStreaming: Error: Found 0 dependent packages...

adding that Montages folder to cooked directory list will solve the issue?

I think since the asset path is hard coded, the cooker ignores it cause its not referenced to any bp asset?

[thin stratus]

Could be. How are you referencing it?

In theory, you don't need to manually add it to the cook list if it's used

[nocturne quail]

StandEquipMontage = FSoftObjectPath(TEXT("/Game/Montages/StandUnEquipMontage.StandUnEquipMontage"));

[thin stratus]

Yeah you should properly reference this in an Asset and not hardcode it

[nocturne quail]

TSoftObjectPtr<UAnimMontage>& TargetMontage = StandUnEquipMontage;
GetMontageSafely(TargetMontage);

UAnimMontage* UWeaponManager::GetMontageSafely(TSoftObjectPtr<UAnimMontage>& SoftPtr) const
{
return SoftPtr.LoadSynchronous();
}

[lost inlet]

And is it a UPROPERTY?

[nocturne quail]

Ok, still it is possible to add it to cook list and not refference in editor?

[nocturne quail]

No

[lost inlet]

well there you go

[nocturne quail]

TSoftObjectPtr<UAnimMontage> StandUnEquipMontage;

[nocturne quail]

Avoiding reflection for hard coded path's

[lost inlet]

why? and why are you using hard coded paths?

[nocturne quail]

If I want to make it UPROPERTY, i will just fill it in editor

[lost inlet]

the cooker won't be able to see the soft reference

[thin stratus]

Don't hardcode paths

[nocturne quail]

bp's oftenly corrupts, just to not refill it, hard code save lots of time

[thin stratus]

Like, this is wrong. Do it correctly and the cooker will find it

[thin stratus]

It's wrong though.

[lost inlet]

very wrong

[nocturne quail]

I am using lots of live coding

[lost inlet]

but I'm sure missty has a cracked reason for it

yep, there it is

[thin stratus]

Don't care. It's wrong. Do it the right way. Reference it via an Asset.

Live coding is irrelevant for this.

[nocturne quail]

Ok, and this next approach is also wrong?

UCLASS()
class ARMA_API UWeaponMontageData : public UDataAsset
{
    GENERATED_BODY()

public:
    UWeaponMontageData()
    {
        StandEquipMontagePrimary = FSoftObjectPath(TEXT("/Game/Montages/StandEquipMontagePrimary.StandEquipMontagePrimary"));
    }
    
    TSoftObjectPtr<UAnimMontage> StandEquipMontagePrimary;
};

AWeapon.h
UPROPERTY() UWeaponMontageData* WeaponMontageData;

AWeapon.cpp
AWeapon::BeginPlay()
{
    if (UWeaponMontageData* NewMontageData = NewObject<UWeaponMontageData>())
    {
        WeaponMontageData = NewMontageData;
    }
}

TSoftObjectPtr<UAnimMontage>& TargetMontage = WeaponMontageData->StandEquipMontagePrimary;
GetMontageSafely(TargetMontage);

UAnimMontage* AWeapon::GetMontageSafely(TSoftObjectPtr<UAnimMontage>& SoftPtr) const
{
    return SoftPtr.LoadSynchronous();
}

[lost inlet]

lol what is going on here

[nocturne quail]

since UWeaponMontageData is NewObjected, I think it should let the cooker know its an object and exists

[thin stratus]

Just expose the soft object pointer to the Blueprint and set the montage there.

Don't start making up a solution for a problem that doesn't exist.

[nocturne quail]

and than just TSoftObjectPtr<UAnimMontage> TargetMontage = StandUnEquipMontage; by & would not be needed?

[thin stratus]

What you afterwards do with the exposed soft object ptr is up to you.

But referencing the asset should be done via that exposed property

You can then pass it around as soft object ptr, or load it and use it, or load it and store it in a transient variable, etc.

[sick niche]

This is some serious questionable design decision

[haughty ingot]

It’s obviously for their crazy super hackers and insane performance

[quartz iris]

Hey man sorry i'm not able to update you im away for 2 days. I will be back tho to lyk !

I added you as a friend on discord 😄

[frosty harbor]

How does replication work inside a non replicated gameplay ability? I remember people saying that Epic doesn't recommend setting gameplay abilities as "replicated", if that's the case, how would you go about replicating a montage being played inside a gameplay ability so other players can see the animation being played?

I'm just running Net Exeuction Policy "Server Only" since it's for AI but I can't see the animation playing at all unless I replicate the ability

[kindred widget]

There is a play montage node specific to gameplay abilities that'll handle that for you.

[quasi tide]

I will still mark an ability as replicated just so I can do an RPC. Target data is dumb.

[frosty harbor]

oh yea forgot about it lol

ty

[meager spade]

why is targetdata bad?

[exotic wasp]

it's dogshit simple as

but also the only thing to use

[slate phoenix]

Why am I having these issues? I couldn’t really understand much from the server logs either.
Aren’t ISMs replicated?

[quasi tide]

I said it is dumb. It is annoying to setup mainly. Passing the information around is w/e.

[thin stratus]

Because your BP_BlockManager_C_1.NODE_AddHierarchicalInstancedStaticMeshComponent-1_1 isn't stably named and thus can't be resolved across the net.

(Which usually starts throwing warnings when standing on them)

[meager spade]

i wonder if this was attached]

its weird

[slate phoenix]

I use dinamic

[thin stratus]

No idea what you want to say with that.

[thin stratus]

Is that done during ConstructionScript?

I don't know if Blueprints can runtime spawn such Components outside of it, while having them properly named.

[slate phoenix]

Functions

I add new Functions

[slate phoenix]

hm

[meager spade]

and datatable eek but yeah this is problematic

[slate phoenix]

I did this intentionally.
If I limit the blocks, players won’t be able to place new ones.
I set up the ISMs to be loaded from the database.

If I don’t use a database, I’ll need to define a new block inside the Actor for every new block.
Am I thinking about this wrong, or is this correct?

[thin stratus]

Having the blocks as (H)ISM is correct. Shouldn't be a new Actor for each.

It's just that I'm not sure if Blueprints can spawn stably named components runtime.

Might be a limitation and needs C++.

[meager spade]

why does it need to be replicated for anyway?

couldnt the client just know what isms it needs to create

[slate phoenix]

i make dediced server game

If I don’t initialize the world on the server when it opens, how will other players see it?
Or would it work if I initialize the ISM builder on the client instead?

Maybe it will help you understand.

0:15 You will be able to see and visit worlds created by other players.
Each world will have its own markets, and players will be able to farm.
For players to see each other, a new server is launched on a separate port for every opened world, and that world is run by the server.
However, because ISMs are not replicated, the blocks are not visible.

If there is no one in the world, the server shuts down within 15 seconds.
This reduces RAM usage and prevents unnecessary server load.
However, when someone enters the world (either the owner or a guest), the server is started again on its own port

[thin stratus]

You'd usually replicate the data that then is used to create the ISMs instead of replicating the Components.

[opaque forge]

If I have multiple pixel streaming components in a scene, does each make an individual connection to the signaling server or do they somehow share one?

[boreal flicker]

I have a walljump system in my game. The way it works is every tick it does a sphere cast, and if it finds a wall it updates the wall variable which contains info about the wall. Then when you press jump, it checks if you’re currently on a valid wall, and if so do the walljump. I have it in tick instead of looking for a wall when you jump so that effects and stuff like that can constantly know whether you’re on a wall. The problem is the client and server disagree on if a walljump happened or not, and when the client mispredicts its pretty bad. What should i do?

[cosmic phoenix]

Hi everyone
I am having an issue with UE5 Multiplayer. When a Client approaches a 'BP_Chair' and interacts with it, certain events should trigger—for example, the character should sit on the chair. My problem is that instead of the Client who performed the interaction, the Server's character is the one sitting down, even if the Server is far away or didn't interact at all. The logic is only affecting the Server. For the system to work correctly, it must affect only the specific player who triggered the interaction, regardless of whether they are the Server or the Client. Could you please help me resolve this?

[nocturne quail]

Possible to start the game in late client mode?

[boreal flicker]

can i have more details about how it works? is the chair calling the logic on the player or the other way around? if its the chair, how does it know which player to act on?

[dark edge]

To do it the correct way you gotta get into CMC

[boreal flicker]

i am into cmc

[dark edge]

in C++

[boreal flicker]

yep

[dark edge]

is the wall check happening in the cmc tick or outside of it?

[boreal flicker]

i overrided the jump function to have walljumps and im calculating the wall in the tick function of it

[boreal flicker]

inside

is that bad

[nova wasp]

you are almost certainly accidentally using something that uses the local array of controllers/pawns etc like get player controller
https://wizardcell.com/unreal/multiplayer-tips-and-tricks/#2-beware-of-getplayerxxx0-static-functions

but I can't see what you are actually doing in the server rpc

[dark parcel]

Yeah, iirc theres a setting for late join

Haven't done one my self though

[nova wasp]

only in recent ue5 versions

[dark parcel]

@boreal flicker https://m.youtube.com/watch?v=F2IOnzzOCT0&list=PLXJlkahwiwPmeABEhjwIALvxRSZkzoQpk&index=10&pp=iAQB

[dark parcel]

Is she still trying to get the game run on toaster?

[nova wasp]

nothing wrong with making a game for lower spec systems but we will always assume people are talking about a version we all actually use

[nocturne quail]

I am on 5.4

well there is another way to achieve it, launch server from command line and join second client from command line lately

[silent valley]

Much easier to "Enable late join" in editor preferences, then just start the game in PIE as normal, there's a new button next to the Play button to load a new client

[nocturne quail]

just found Allow Late Join
now every client is a late joiner?

[silent valley]

No... the number of clients in the drop down list for Listen Server or Client mode join at the start, not late joining.

[nocturne quail]

hmm seems like not very usefull

[silent valley]

🤦‍♂️

[nocturne quail]

running server from editor using command line and joining it with command line early/late is more freedom

[nova wasp]

it is very useful if you want to test late joining in pie

you asked directly to be able to late join in client mode, that is an example of how to late join in client mode in play-in-editor

it is exactly as they say, it just makes the play button add a new pie client after the original number is created based on pie settings

[toxic linden]

Hi any one know is vehicle movement component with chaos wheeled supports replication

[tidal saddle]

Hello, im currently troubling with steam multiplayer connection. I have enabled both steam oss and steam sockets plugins. Steam overlay works, i can create session, but when i try to join from another pc nothing happens. Im using ue5.7.1 and yes, i have implemented accept invintation events, i attach all neccessary files. Any help appreciated 🙏

P.s Create session function on screenshot is not vanilla but my own implementation, which i attached in SGCreateSessionProxy.cpp

DefaultEngine.ini

[tardy fossil]

do you have a log for the server thats hosting?

also i've heard appid 480 can be buggy and not work if lots of peopple are using it

[tidal saddle]

yes, gimme 5 minutes i will recreate it

i use listen server

[tidal saddle]

it might be, but those features should be working as far as i know, and it worked before, back when project was on ue4.26, but then i used find session instead

[tardy fossil]

Error Unable to communicate with ANY of 30 Steam Datagram routing cluster. Possible problem with local internet connection? it cant connect to any of the steam routers for some reason.. maybe a firewall thing?

if you type ping 8.8.8.8 in command prompt, do the pings go through successfully?

[tidal saddle]

i also noticed this message and first thing i did i disabled firewall on both machines. 8.8.8.8 pings on both machines as well

Here is a repo of my project by the way, if that might help https://github.com/Kokapuk/shooter-game

[nocturne quail]

you don't need the multicast since you already have onrep
https://github.com/Kokapuk/shooter-game/blob/main/Source/ShooterGame/SGCharacter.cpp#L219

[silent valley]

Did you enable both steam sockets plugins?
You should consider using AdvancedSteamSessions plugin as it has a couple of workarounds for engine issues.
See this post where I got a minimal example working on 5.6 (I did this was just a few days before 5.7 came out).
#online-subsystems message

I also recommend enabling verbose logging on these categories, by adding this to DefaultEngine.ini

[Core.Log]
LogSteamShared=VeryVerbose
LogOnline=VeryVerbose
LogOnlineSession=VeryVerbose

[tidal saddle]

i think i still need it, since i need to broadcast death with bIsHeadshot parameter to everyone

but maybe it would make more sense to disable OnRep then

[tidal saddle]

never heard of need of the other plugin, ill try tinker around with it a bit more

[nocturne quail]

you can broadcast it in onrep with an authority check, death should only be handled on server and simulate effects on all clients

[tidal saddle]

still i need to pass bIsHeadshot somehow, which is not a case in OnRep

[nocturne quail]

void ASGCharacter::AuthDie(APawn* Killer, const bool bIsHeadshot)
{
    if (!HasAuthority()) return;

    OnDie.Broadcast(Killer->GetPlayerState<ASGPlayerState>(), GetPlayerState<ASGPlayerState>(), bIsHeadshot); // on server only
    bIsDead = true;
    OnRep_IsDead(); // simulate death effects
    GetMovementComponent()->Velocity = FVector::ZeroVector;
}

[tidal saddle]

event should fire on clients aswell so im not really following

[nocturne quail]

ok sure

[dark parcel]

@tidal saddle you can use struct to include your headshot data.

[tidal saddle]

that sounds like an idea, thanks

[dark parcel]

@tidal saddle if you need more data such as vector (e.g to dictate force for ragdoll) on top of what you have right now.

Maybe just piggy back of HitResult instead creating new struct.

At the end of the day, does the same job imo.

[rare cloud]

instead of set the velocity manually I would call
GetMovementComponent->StopMovementImmediately();

[nocturne quail]

totally agree

[nocturne quail]

just tested it and its very useful, saving lots of time

[normal crypt]

Question about RPCs: Is it possible for reliable multicast RPCs called before the client joins the game, to be received by the client once he joins minutes later? It seems to be the case in the editor, but not in packaged mode and I can't find documentation about it.

Simplification:

• Host is playing alone and calls a reliable multicast on Actor1
• Client joins the game 2 minutes later and automatically creates Actor1 locally, but isn't the authority.
• Client receives the reliable multicast.

[nocturne quail]

afaik multicast are once time run and forget, for late clients use onrep

[normal crypt]

that was my understanding too, but they seem to be queued and resent later under certain conditions in the editor.

I could be reading the data wrong though, will continue looking.

If it's the case, that's a big problem.

[dark edge]

show code

how did client make the actor but isn't the authority?

or rather, show the code that ends in calling the multicast

[normal crypt]

The actor is owned by the server and the server set it to replicate, so the client gets this actor when he joins the session

I'm taking care of the kids, will get back to this later.

[dark parcel]

No?

How are you simulating late joiners in editor?

Rpc is a one time event. Any player that is not in the game will miss it.

[normal crypt]

I set the editor to spawn 2 standalone games. Then I connect them together through a listen server.

One of the standalone is the listen server, the other connects to it later, minutes after the RPC was sent.

This isn't cut and pasted code, I edited it because it's pretty complex.
`UCLASS(config=Game)
class ASprite : public AActor
{
GENERATED_BODY()

...

void SetReplicates(bool bInReplicates);

UFUNCTION(NetMulticast, Reliable)
void ReplicatePostInitChanges(...);

}
`

`void ASprite::SetReplicates(bool bInReplicates)
{
AActor::SetReplicates(bInReplicates);

...

ReplicatePostInitChanges(...);

}

void ASprite::ReplicatePostInitChanges_Implementation(...)
{
...
}`

And I essentially call SetReplicates on my ASprite somewhere in the code.

I'm going to continue debugging my issue now, if the kids go to sleep.

[dark parcel]

are movement input replicated?

I need the movement direction for the proxies.

[nova wasp]

not directly besides the velocity afaik

[dark parcel]

I see... thanks.

[nova wasp]

you can add replicated acceleration if that helps with a setting

you can also completely replace the onreps and replicated properties of aactor (acharacter etc) with enough cheesy macros and overrides

this is only really important if you need perfect atomicity though

for a slower paced game it won't matter though

[dark parcel]

Trying to tie this to CMC predictive movement.

[nova wasp]

well, the cmc for sim proxies is pretty much just using the actor replicated movement and a few more flags/bools in ACharacter::GetLifetimeReplicatedProps

ReplicatedMovementMode, bIsCrouched and not to forget ReplicatedBasedMovement

see PostNetReceiveLocationAndRotation and UCharacterMovementComponent::SimulatedTick

you can absolutely just add a cond_simulatedonly replicated property that has more context you need but be aware unless it is in the same struct (with iris) or manually serialized all at once it will be kinda arbitrary if it shows up exactly at the same time as the velocity + transform it was intended for

[dark parcel]

If I can get away with RPC, then I can just pass the direction there, but not sure if it will break predictive movement.

[nova wasp]

rpc here doens't make much sense unless it changes extremely often and also isn't very time sensitive

for example if it's kind of okay for it to be late or too early

and if it the things it changes are not very impactful

[dark parcel]

I tried to replicate via flag,
But not sure where my PlayMontage should be.

Tried to do it on BeforeMovement and AfterMovement, but it doesn't get played on other machines.

[nova wasp]

what does PlayMontage have to do with this? you said "movement direction"

is it a root motion montage?

[dark parcel]

Oh sorry, I'm trying to do 180 turn. But putting the multiplayer aside, I noticed that it doesn't turn the way I wanted due to TurnRate, so I figured I will need the direction input for the final look at rotation which in multiplayer context need to be passed.

I'm facing too many problem right now X_X.

[dark parcel]

yeah root motion montage, but I need to tackle the first problem.

[nova wasp]

root motion completely changes the entire flow

see OnRep_RootMotion

[dark parcel]

Pretend that I haven't said anything above.

How would you handle characters playing a montage all while respecting CMC client prediction.

[nova wasp]

the regular replicated transform updates will completely get skipped during root motion

[dark parcel]

Yeah, I'm looking at this atm.

[nova wasp]

it depends on what "respecting" it means

it's literally just onrepping the transform + velocity and movement mode then simulating it locally, then smoothing the result to hide how much it jitters internally

UCharacterMovementComponent::SimulatedTick is where I would look first.. I am personally not really sure exactly how sim proxy root motion works in detail

during root motion that will kinda replace how everything else works

[dark parcel]

basically what I want to do =(. How would you tackle this?

should I still do it via compressed flag or not much sense to do it that way?

[nova wasp]

It depends... if it's just done locally via acceleration changing you could kinda just rely on that with some fuzziness I imagine

Making it a real "event" from the server will increase the chances it can miss

[dark parcel]

I tried checking speed change but other machines won't get updated fast enough resulting the condition get skipped all the time on other client.
TLDR it's been unfruitful to just apply the animation logic locally.

[nova wasp]

on simulated proxies you need to remember the mesh and actor location are entirely separate things

the true actor location is going to jitter

you can see how this actually looks if you turn on p.VisualizeMovement

the mesh location is smoothed and will interpolate based on a variety of settings

[dark parcel]

isn't that already handled for proxies

[nova wasp]

yes, but think about what that actually means

the motion you actually care about might be more the smoothed change in mesh position more so than than the raw velocity from the server

here is a cube on a sim proxy's mesh

this video looks choppy and that's not just because the recording sucks, the actual updates are very sparodic

I'm thinking I would personally try to consider the actual mesh to have its own velocity so to speak
of course you would need to consider and ignore really harsh corrections that skip smoothing it I suppose if those would cause false positives etc

honestly just trying replicated acceleration might be good enough though, I would suggest at least trying that first

the issues here imo is what do you do if it starts getting false positives that would never happen for the true local player

you could maybe have an idea of the true stage at which you did the "about-face" rotation and it was sent from the player itself from client->server as part of their saved moves somehow. You could make the sim proxies get an incrementing counter and/or a current "flip direction" as a byte or something if missing this is very bad

personally I would be okay with this kind of small animation flip being missed occasionally during extremely bad network situations... property replication is not a perfect series of events and probably should not be for sim proxies in most games unless the state is 100% required on all peers

Just remember that sim proxies do not care about saved moves, this is just a way to get that information to the server. getting that information to sim proxies has to be from something replicated on the character/a component/some rpc etc

which I think replicated acceleration will generally be pretty damn close to the intended movement direction while walking in most setups

FWIW I have not personally done that kind of turn-in-place from braking movement before so I'm sure someone else might have an evem more simple answer

[dark parcel]

All good, need all kinds of ideas atm.

[nocturne quail]

Any idea why I'm getting this crash?

[nova wasp]

show the line of YOUR CODE that it crashed on

UWeaponManager::OnRep_CachedEquippedSlot() Line 426 and the things before it

not accusing you of making a mistake here, just not possible to determine what is going on inside of a null pointer/garbage memory

also this time I am going to suggest sharing context

[nocturne quail]

void UWeaponManager::OnRep_CachedEquippedSlot()
{
    UE_LOG(LogTemp, Warning, TEXT("OnRep_CachedEquippedSlot 1"));
    if (CachedEquippedSlot != 255)
    {
        UE_LOG(LogTemp, Warning, TEXT("OnRep_CachedEquippedSlot 2"));
        if (UCharacterAnimInstanceBase* CharacterCurrentAnimInstance = Cast<UCharacterAnimInstanceBase>(OwningCharacter->GetMesh()->GetAnimInstance())) // Line 426
        {
            UE_LOG(LogTemp, Warning, TEXT("OnRep_CachedEquippedSlot 3"));
            CharacterCurrentAnimInstance->SwitchToArmedState(GetEquippedWeapon());
        }
    }
}

[nova wasp]

this has 2 ways it can crash instantly

is OwningCharacter null? is OwningCharacter->GetMesh() null?

you need to read the current state of the memory around these lines

[nocturne quail]

No, onrep is called after check ensure for owningcharacter and the mesh and animinstance on server

[nova wasp]

uh... this onrep is running from an actor channel receiving replication

why would it being valid on the server matter to the client?

[nocturne quail]

owning character is a replicated variable

[nova wasp]

which I am mostly guessing because you did not include the context of which side this crashed on but I am pretty sure this onrep is called from a regular replication onrep

[nova wasp]

so? why does that matter?

it can still be null

[nocturne quail]

server sets it

[nova wasp]

please show the state of the pointer at this line in the crash in the watch window

if you want help

[nocturne quail]

I will crash it again in a moment

[nova wasp]

the server setting a replicated property does not gaurantee the client has it ever ESPECIALLY for a random unrelated property (CachedEquippedSlot)

they are two separate properties and there is NO gaurantee they will arrive in any particular order

[kindred widget]

Even if they do arrive at same time for this particular actor, that doesn't meaning OwningCharacter has arrived and is valid yet.

[nova wasp]

to fix this just don't dereference a null pointer in here and when OwningCharacter is set it also can just do the same thing as whatever OnRep_CachedEquippedSlot does to make sure the work is done when everything is there

this is a really common problem in unreal networking

you have two things that happen at random times and they need to both be there to do X, just do X in both places for whichever one happens when both are there

[nocturne quail]

yeah owning character is null

[nova wasp]

https://www.youtube.com/watch?v=bLHL75H_VEM

so yeah, maybe have an onrep for the character pointer and have it call
CharacterCurrentAnimInstance->SwitchToArmedState(GetEquippedWeapon());
so that it will apply the right setup when it comes in

or just poll for it in the pre anim tick (sometimes dumber is easier)

[kindred widget]

It's a good habit, networking or otherwise, to always nullcheck your pointers and gracefully handle their fail states. Nothing in gameplay code should be causing an app wide shut down.

[nova wasp]

this should not even be an error log, it can be assumed it WILL be null in some cases

replication does not care

ideally the stuff that expected it to be there can be done when it shows up

[glass rune]

is there a way to debug project with rider but only for example client 1?

[exotic wasp]

dunno if there's a specific way but if they are different processes you could attach your debugger to the one you want

[glass rune]

I tried but it seems it is not working. Rider only recognize one process even tho two or more windows are opened

[twin vessel]

Where do you guys manage you ui from?

I might be super over complicating shit for myself, but basically I have a main menu, a lobby and a match level.
Both menu and lobby have access (through a ui button) to a firing range built in the level.
This means that i need to have basically the same game mode / player controller / pawn for all levels.

I guess i could put what widget to display on the HUD class and use a different one for each level. But this does not feel completely right either.

Also some ui depends on the game state, for example during the match there is a phase to select the weapons.

[bitter veldt]

hey there, im trying to make a little lobby system where players from the main menu can invite each other / create a party, and then load into the game once everyone is ready. i was wondering, what is the most ideal way for the clients to see other players names etc on UI? I was replicating a players array but noticed some issues with that

[warped berry]

Maybe try lobby data that's shared between all lobby players by using the session interface

[amber vale]

Anyone know any plugins or articles for smoothing movement on autonomous proxies? Movement corrections being performed instantly makes it impossible to have smooth player-versus-player knockback, and I'm wondering if there's an existing solution for this.

[thin stratus]

CMC corrections are smoothed fwiw

[thin stratus]

Party Systems are usually done via Beacons.

Well, the ones where you party up in the MainMenu at least

[nova wasp]

if they are a listen server they may need to enable the cvar though

but they did say autonomous proxies ah. What I said there is for sim proxies (woops)

[dark parcel]

What are beacons? R they just another backend service from steam?

[thin stratus]

No. Beacons are lightweight connections that allow replication stuff like properties and RPCs but without actually opening the server's level.

They come with a subset of classes, like BeaconHostState and ClientState. That are very similar to GameState and PlayerState.

But minus all the game related code.

So you can use them to have a party lead and party members in the main menu and replicate minimal data like names or what character is selected via them

[dark parcel]

0o? How does the connection established when players are not connected?

[thin stratus]

It's how, at least originally (not sure if still) the invite lobby of Fortnite works

[dark parcel]

Epic service?

[nova wasp]

so you can poll for information like number of users, game mode etc?

interesting

[kindred widget]

Gears of War was one of the first majors to use it. I think one of them still does.

[thin stratus]

That's up to you. Beacon connections are just socket connections. It's the same as normal connections, just more lightweight

You can create a special session in your backend to handle them

Steam has the PartySession name for it as opposed to GameSession

EOS, no idea

[dark parcel]

So are most of the heavy lifting already done in the steam interface for Party Session?

[thin stratus]

It's the same concept in the end. The session backend just communicates the address to connect to

[kindred widget]

It's worth noting that you don't need to use the OSS specific stuff for them.

[thin stratus]

Don't know. I only ever used normal sessions but as PartySessions for them

[kindred widget]

They're just an actor basically that can send and receive RPCs from the server without going to the map. No replication.

[thin stratus]

Yeah OSS/Backend and Beacons are two concepts

Similar to how OSS/Backend and connecting to a server are two different things

[thin stratus]

In theory, yes, Unreal Tournament used them to reserve dedicated server slots

Don't need to connect and travel to then learn that someone else connected faster

[dark parcel]

I'm struggling to see how this work. So someone still have to be the server?

In theory, how does it start when a group of friends want to form a party and chat?

[thin stratus]

I think there is a pinned post in #online-subsystems about them

There are probably plugins but natively they are c++ only iirc

[kindred widget]

One of them is the "Server". AKA Party leader.

[dark parcel]

So essentialy, it's like hosting a map for people to join without hosting the map 🤔

[thin stratus]

Correct. The PartyLeader creates a BeaconHost, which is similar to a ListenServer. Aka it has an open socket connection and waits for connecting BeaconClients.

[kindred widget]

Similar. And no replication. All RPC based.

[thin stratus]

Replication should work. Unless I misremember

Been way too long

[dark parcel]

btw I got a question about Turn Stop movement.

[kindred widget]

Maybe. I could be going off of old info. I didn't test that, just the RPC bouncing.

[dark parcel]

So I finished a mechanic to turn 180 degree for the character, but I end up using RPC and montage as opposed to including the info in SavedMove.

is this not recommended? I always thought RPC for movement related is probably a bad idea. But I see ALS community version handle turning replication with RPC too.

trying to take a video atm.

[thin stratus]

No idea tbh. Never touched ALS or had to implement a 180 degree turn

I would probably just handle that in the CMC directly. Maybe by using ControlRotation or adding something similar.

Probably needs to be predicted. If it's driven by player input then it shouldn't be too difficult to just do it in the CMC

[dark parcel]

I tried to just locally check the deceleration but can't get it to work for Sims proxies because they don't update fast enough =(.

and my attempt on predicting it also failed.

[nova wasp]

My take earlier was to use replication acceleration or just add a new sim proxy only thing for it but an rpc might be more than good enough if it ends up timed okay

[thin stratus]

If the RPC is only for the SimProxy then that's probably ok

I don't know enough about 180 turns tbh

[dark parcel]

well it goes like this.

Autonomous do 180 turn -> Tell server to 180 Turn -> Multicast

kinda eww, but coudln't get it to work other way atm.

[thin stratus]

SimProxies are a problem in all those things. Flags that can be missed usually need to be a uint8 that gets increment instead

[nova wasp]

May I ask how it actually works math-wise when simmed locally? This is partially me wanting to try it myself more than just helping I guess

I always imagined it was some kind of like acceleration threshold that eventually locks you into a little turn-around move

[thin stratus]

Yeah. You have three values to work with I guess.

InputAcceleration, so input direction times acceleration.

[dark parcel]

in my case sim just not behaving the same. For example, I calculate speed locally to do a stop animation when sprinting.
But sim doesn't update the value fast enough that it skips the check.

Ideally, I just want to rely on acceleration / speed that I can calculate locally. That way I don't need to care about networking.

[thin stratus]

Acceleration itself (current one).

Velocity, also current one.

If I hold W along the X axis it would be some (X,0,0) input acceleration

And then some resulting current accel and vel

If I then hit S suddenly, it would still be the same velocity start of that frame

But input accel now points opposite

Which is a good point to notice the 180 turn I guess

Since CMC doesn't replicate shit to SimProxies, they only get the final velocity I guess.

[kindred widget]

I'm confused. Isn't this mostly just a root motion montage?

[thin stratus]

So gotta replicate some incremented uint8 flag in the character to indicate the turn

[thin stratus]

Eh, na, this is more TurnInPlace style stuff I would say

RootMotion takes too much control away and is too stiff

[dark parcel]

initially I don't want to use root motion because I want some slide effect like ice skate. So trying to rely on acceleration / deceleration for the effect.

but I end up with root motion anyway. Skill issue.

[thin stratus]

Players with gamepads could do 170° turns

You also lose a lot of control with the RM approach

Not saying it can't be done with RM

But then that also needs to be the exact result you want

[kindred widget]

Yeah I was just thinking the quick spin around effects some games have. Probably falls apart if you do it slower than a quarter second or so.

[thin stratus]

If you notice the turn you can just play a root motion montage, sure. And you can use some motion warping to fix the cases where players trigger it from gamepad and didn't do an exact 180 turn.

[kindred widget]

And the sliding kills that idea too.

[thin stratus]

MotionMatching is a perfect system for this if it wouldn't be so damn fucking expensive

Like you don't even need to use it for the whole animation locomotion

Can just feed it some reference point calculated from the input and use that grab the best montage with the best start time of that montage to get the perfect result

[dark parcel]

tried with 400 ms, doesn't seem breaking yet. But working outside the prediction system, I am not sure if I'm going to get issues in the future.

[thin stratus]

Mid 180 turn you do another 180 turn? MotionMatching can pick the best matching montage for that too. Sadly expensive and needs tons of animations

[nocturne quail]

why is OnWeaponReplicatedNotify not always broadcasted for late clients?

void ACharacterBase::ServerPickupWeapon_Implementation(UWeaponItem* WeaponItem) 
{
    if (!WeaponItem || !WeaponItem->WeaponClass || !WeaponManager)
    {
        return;
    }

    FActorSpawnParameters SpawnParams;
    SpawnParams.bNoFail = true;
    SpawnParams.SpawnCollisionHandlingOverride = ESpawnActorCollisionHandlingMethod::AdjustIfPossibleButAlwaysSpawn;
    SpawnParams.Owner = this;
    SpawnParams.bDeferConstruction = true;

    FTransform Transform;
    if (AWeapon* Weapon = GetWorld()->SpawnActor<AWeapon>(WeaponItem->WeaponClass, Transform, SpawnParams))
    {
        const FName EquippedSocket = "Equipped_Gun";
        Weapon->FinishSpawning(Transform);

        Weapon->AttachToComponent(GetMesh(), FAttachmentTransformRules::SnapToTargetNotIncludingScale, EquippedSocket);
        WeaponManager->SetWeapon(Weapon);
    }
}

void UWeaponManager::SetWeapon(AWeapon* NewEquippedWeapon)
{
    PrimaryWeapon = NewEquippedWeapon;
    OnRep_PrimaryWeapon();
}

void UWeaponManager::OnRep_PrimaryWeapon()
{
    OnWeaponReplicatedNotify.Broadcast(PrimaryWeapon);
}

void ACharacterBase::OnWeaponReplicatedNotify(AWeapon* Weapon)
{
    UE_LOG(LogTemp, Warning, TEXT("OnWeaponReplicatedNotify broadcasted"));
}

ServerPickupWeapon_Implementation is reliable
and DOREPLIFETIME_CONDITION_NOTIFY(UWeaponManager, PrimaryWeapon, COND_None, REPNOTIFY_Always);

[tardy fossil]

probably binding the delegate after onrep is called

onreps can get called before begin play

[nocturne quail]

character binds it in beginplay

if (WeaponManager)
{
    WeaponManager->OnWeaponReplicatedNotify.AddUObject(this, &ACharacterBase::OnWeaponReplicatedNotify);
}

[nocturne quail]

tried calling onrep with 1sec delay, nothing changed

its like failing 1 out of 5

[tardy fossil]

stick a debug log print in the onrep and right after the delegate is bound and see whats going on i guess

[dark parcel]

Print on tick to check if you actually get the right value? Its probably just race condition.

[nocturne quail]

this seems to work fine, so yeah this shows that weapon onrep calls before the delegate actually binds successfully sometimes

    if (ACharacterBase* owningcharacter = Cast<ACharacterBase>(GetOwner()))
    {
        owningcharacter->OnWeaponReplicated(PrimaryWeapon);
    }

[nocturne quail]

value of weapon?

[nova wasp]

just bind to it earlier I guess

beginplay is relatively later in the lifecycle

[nocturne quail]

in ctor?

[nova wasp]

nope (there are a few other things before beginplay and not in the ctor) https://dev.epicgames.com/documentation/en-us/unreal-engine/unreal-engine-actor-lifecycle

as ultimately the onrep triggering is random... may as well just call the function you were binding for when you first bind to it if it has a meaningful value when you first see it

also as I mentioned earlier, if (WeaponManager) can silently fail if the weapon manager is null but if it's a default subobject it might never really matter

[dark parcel]

I dropped like 40 items in the map and the on Rep got called in batches

[bitter veldt]

Ahhh this is what i saw online but wasn’t sure if this still accurate tbh

[fallen fossil]

Roles question.

If LocalRole is SimualteProxy, then can we distinct if this Actor/Pawn is controlled BY server or* Other Client?
🤔
cause it looks like I have to check this other way

[dark parcel]

What do you mean by server of other client?

There is the server and there are client / clients.

Sim proxies are other pawn that you dont control.

Lets say 3 players, i am the listen server, john is 2nd player, mary is 3rd player. Each of us controll a character.

In my machine : my role is authority over all the player character.

Meanwhile in john computer : my character is simulated proxy, mary character is simulated proxy. His own character would be automous proxy because he controlled it.

In mary computer : john and my controlled character is simulated prozy. Her own character is autonomous.

[amber vale]

Simulated proxies are smoothed, but autonomous proxies aren't—at least not with a default configuration. If you're a client and the server corrects your movement, you're instantly teleported to the server's location. Looking at the CMC's code, variables like Network Smoothing Mode, Network Max Smooth Update Distance, and Network Simulated Smooth Location Time are only used when correcting simulated proxies.

[fallen fossil]

"or" , so I can see if its AI bot (server) or player :d

cause I guess ListenServer is Player

[bitter veldt]

Do you know if Beacons work fine in 5.6 with Steam? I'm searching around in the Discord and it seems like people were having issues - also saw this forum post 🙁 https://forums.unrealengine.com/t/online-beacon-compatibility-with-steamsocket-plugin/2665805

https://github.com/EpicGames/UnrealEngine/blob/684b4c133ed87e8050d1fdaa287242f0fe2c1153/Samples/Games/Lyra/Config/Custom/Steam/DefaultEngine.ini#L45 ouch haha

[exotic wasp]

you can maybe bind to it in ready for replication

I wouldn't be confident in it existing yet, though

You should probably just have a bind function that immediately broadcasts the initial value

Or something along those lines

[viscid obsidian]

will the CharacterMovementComponent automatically run PerformMovement() witht the same input vector for consecutive ticks if multiple ticks worth of time have passed on the client framerate?

Or am I supposed to implement my own tick cycle?

[nova wasp]

it works in a while loop and has settings for min/max ticks iirc

I am unsure how it handles extreme cases with keeping input or not (I would assume it keeps it)

[thin stratus]

Really? The Mesh is not smoothed on AutoProxy? Small corrections are usually pretty much invisible. It should only teleport the mesh too if the distance is too large. Maybe I misremember or it's really all just SimProxy, but I'm not convinced. Don't have access to any PC for a bit.

(not saying you are wrong, but wouldn't be the first time that someone states something that they were also not sure about in the end. Will probably check the code once more next week when I have my PC set up again)

[thin stratus]

No clue. Haven't used Beacons in a while.

[thin stratus]

Think that's only for sub stepping.

Aka to have a bit more precision for collision and such

[viscid obsidian]

Are we not supposed to change the parameters of the character movement component? I don’t see anything in the docs about the networking accounting for changes in configuration of the movement component across different ticks

[thin stratus]

Do you mean on the server?

[nova wasp]

yeah I guess it's still one tick

[thin stratus]

Changing default properties runtime is probably not really expected. Would need to be done through outside replication and potentially causes corrections.

E.g. the WalkSpeed variable is usually quite set in stone.

You can alter it via some OnRep property in the character, but that might lead to a few corrected frames

[viscid obsidian]

So what’s the intended flow for defining a bunch of custom moves that may be similar to walk or sprint or jump in nature but have slightly different mechanics?

Im not sure what the intended extension point of the system is

[thin stratus]

Inherit and extend where needed.

[viscid obsidian]

So we would inherit the component and add potentially more modes or parameters?

[thin stratus]

There are some basic examples pinned to this channel.

[thin stratus]

Yeah

Extending movement is usually a matter of inheriting and adding your own stuff in c++

[viscid obsidian]

Adding more parameters and modes doesn’t sound too bad because it should be mostly independent from the awesome framework unreal provides

I’ll be sure to take a look at the pinned examples. Thabks for the help!

[thin stratus]

This is still a bit of a vague question btw

Server doesn't tick client movement after all.

Client sends RPC with all input data and the delta time and server performs the move when receiving the RPC

Server frame rate doesn't really matter due to that

[viscid obsidian]

Doesn’t the client send on a fixed tick rate?

[thin stratus]

No

[viscid obsidian]

Oh really they send by frame? That’s odd to me

[thin stratus]

CMC uses an "independent" frame rate.

Each client uses their own frame rate.

Above 60 FPS it will try to combine moves.

That concept is also the reason that animations can look choppy on listen servers when it views other players.

Cause the pose of the skeleton is ticked via the same RPC on the Server to ensure animation and movement events are in sync.

[viscid obsidian]

Do you know how the independent frame rate is different from a fixed rate?

[thin stratus]

Question doesn't necessarily make too much sense

It's in the name what the difference is

[nova wasp]

the cmc kind of just has timestamps in each sent move

[nocturne quail]

I am now calling a dedicated function directly from the owner instead of broadcasting it, it works fine but the only thing I don't like is casting to owner to access the function

other option I can just add this function to AActor and override it in the Character class

this will be the case if I really want to avoid the cast, but I will not since this is not a rapidly call operation

[dark parcel]

Rapid call or not probably wont make difference when it come to casting

[nocturne quail]

I mean we don't want that cast on tick

[dark parcel]

Doesnt matter though?

What do you have to lose

[nocturne quail]

for me this is a bad practice to cast in tick

cast exists to access a member once and save it cache it for later use

in my case I am not saving it since character can have many weapons and players can join/leave

so the old cache can be invalidated or crash if not refreshed

[dark parcel]

Btw for your on rep condition rat race, maybe you can just use a boolean as a state to see if the required component already initialized.

Use it to determine wether to call the onRep or not.

Kinda see this pattern in GAS when addressing race condition

[nocturne quail]

that's a great idea, maybe will consider it if things gets complicated again in future

[desert breach]

Hey Devs, I've a Pickable actor which I spawn on server and is replicated. It has Collision , Mesh and Audio Component, I wanted to play the sound when some player pawn touches it, So I used a client RPC to execute the call. The call is executed however the sound doesn't play.
However when I try to do it in single player it works fine .
Here is the .cpp
`void ABasePickable::ClientPlayPickupSound_Implementation(AActor* OwnerActor)
{
UE_LOG(LogGOG, Log, TEXT("Inside Client RPC call"));
// Only play if this client is the one who picked it up
APawn* LocalPawn = GetWorld()->GetFirstPlayerController()->GetPawn();
if (OwnerActor && LocalPawn && OwnerActor == LocalPawn)
{
// Play sound directly without spawning actors
if (PickUpAudioComponent && PickUpAudioComponent->Sound)
{
UGameplayStatics::PlaySound2D(this,PickUpAudioComponent->GetSound(),1,1,0,nullptr,nullptr,true);
//PlayLocalPickupSound();
UE_LOG(LogGOG, Log, TEXT("Playing Audio Cue"));
}
}
else
{
//UE_LOG(LogGOG, Log, TEXT("Owner Actor Not Same"));
}

}

void ABasePickable::OnOverlapBegin(UPrimitiveComponent* OverlappedComp, AActor* OtherActor, UPrimitiveComponent* OtherComp, int32 OtherBodyIndex, bool bFromSweep, const FHitResult& SweepResult)
{
if (CollisionActor == OtherActor )
return;

if(CurrentPowerUpActorState == EPowerUpActorState::PickedUp)
    return;

if (OtherActor->GetClass()->ImplementsInterface(UPawnInterface::StaticClass()))
{
    CollisionActor = OtherActor;

    if (GetLocalRole() == ROLE_Authority)
    {
        FString compName;
        OtherComp->GetName(compName);


        FString enumName;
        const UEnum* WeaponStateEnum = FindObject<UEnum>(this, TEXT("EPowerUpType"));
        WeaponStateEnum->GetValueAsString(InventoryInfo.Item_WeaponType, enumName);
        enumName = enumName.Mid(14);
        //UE_LOG(LogTemp, Warning, TEXT("PickedPowerUp is %s"), *enumName);

    //    UE_LOG(LogTemp, Warning, TEXT("On Server, found  %s"), *compName);
    //    UE_LOG(LogTemp, Warning, TEXT("Overlapping Pickable!"));
        if (UVehicleBlueprintFunctionLibrary::GivePowerUpToThisPlayerController(CollisionActor->GetOwner(), InventoryInfo))
        {
            //UE_LOG(LogTemp, Warning, TEXT("%s Powerup To: %s collided with %s  component"), *this->GetFName().ToString(),*CollisionActor->GetFName().ToString(),*OtherComp->GetFName().ToString());
            APawn* PlayerPawn = Cast<APawn>(CollisionActor);
            if (PlayerPawn)
            {
                AController* PlayerController = PlayerPawn->GetController();
                if (PlayerController)
                {
                    ClientPlayPickupSound(PlayerPawn);
                    UE_LOG(LogGOG, Log, TEXT("Calling Client RPC"));
                }
            }

`

here is the .h
`UPROPERTY(Category = Pickable, EditAnywhere, BlueprintReadWrite, meta = (AllowPrivateAccess = "true"))
class UAudioComponent* PickUpAudioComponent;

// Add RPC to play sound on specific client
UFUNCTION(Client, Reliable)
void ClientPlayPickupSound(AActor* OwnerActor);`

``Typically an Client RPC should be called on the client which is colliding, and then there I picked the objects audio component to play it.

Its not playing the sound, however the function executes, what is wrong I'm doing ?

[dark parcel]

@desert breach where do ClientPlayPickupSound lives?

Mind you client rpc only gets called if the actor is owned by that specific client.

So if the client rpc function lives inside the PickupActor then you are not doing it right as the server owns that replicated actor.

You can have the function in object the client own.

By defauly, its the player controller, character and player state.

I say controller make the most sense here. And if you dont want to bloat your controller, make an actor component, have the function inside the actor component then attach it to your player controller.

---

Heres what you can do.

Create a client rpc function in player controller.

Call it Client_PlaySound. Input is USound and just do PlaySound2D(InSound).

Now in your pick up actor.

1.On picked up / overlap.
2. If authority
3. Get overlapped / interacting character -> get controller -> cast to your controller and call client_playSound.

Done.

[nocturne quail]

I think player controller already have these client rpcs to play sound etc

[dark parcel]

Cool, I never explore them 😆

[nocturne quail]

Yeah, I just double checked, player controller have lots of client stuff

[dark parcel]

Nice but its unreliable rpc 👀

Guess not important enough if missed

[marsh niche]

Hi everyone, I'm running into a strange desync issue on Quest 3 (Client) vs Tablet (Server).

I am loading multiple small sub-levels using a ForEachLoop calling Load Stream Level (by Name).
To prevent freezing the HMD, I added a 0.1s Delay between each load (see screenshot).

On the Client, some Static Meshes inside these streamed levels appear displaced/offset.

• The Server sees them correctly.
• It is intermittent (happens randomly to different objects).
• Meshes are set to Static mobility.

I suspect the Delay + LoadStreamLevel is causing a Race Condition where the actor's transform replication arrives before the level is fully initialized on the Client.

Question:
Has anyone experienced this offset with streamed levels? Is FlushLevelStreaming the only way to fix this, or is there a better way to chain level streaming in Multiplayer without freezing the VR headset?

Thanks!

[lament flax]

so there isnt any proper real time replciation on the world ?

the game state server delta is very rought, when emulating bad net conditions i can get 0.5s late time

it seems it doesnt try to simulated past time but only some smooth/sync when the var is replicated, i might have to do a custom solution

[dark parcel]

@lament flax simulate past time? R u trying to get the network clock or something?

[lament flax]

i just want an elapsed server world time a minimum accurate on the clients

currently reading https://vorixo.github.io/devtricks/non-destructive-synced-net-clock/

as this probably outdated https://medium.com/@invicticide/accurately-syncing-unreals-network-clock-87a3f9262594

[dark parcel]

Yeah thats the article i would go for if you want to implement your own.

Was gonna link the firsf one.

I dont think it can get any better. It just send burst of rpc and calculate the average. You can modify the frequency of the rpc being send.

There was complain with player state server time due to player state having low net updats rate in the past but some people say it has since changed.

Ive already copy paste the code from the first articlr so i never use the one on player state.

[lament flax]

i was using time to have server and each local client simulate passed time to basically move a platform

but since any ssolution requires RPCs anyways, i might just go for the classic transform replication for my platforms

[dark parcel]

Well if all the machine can agree on the same time then the platform can be simulated locally.

Provided that the platform always move in the determistic way.

[nocturne quail]

yeah interaction sound is not important and should not be reliable I think

[hoary spear]

Interaction sound sounds like a local thing to me

[dark parcel]

I have it as client rpc atm as server is the auth for picking up item.

[hoary spear]

Ah ofc

For others to hear

[dark parcel]

I can just force play the sound locally on interact but that will be premature if there is delay since server have to give the green light if the item is picked up or not.

The moment my interact data is removed from replication. The sound also plays.

[hoary spear]

You dont predict the pickup?

[dark parcel]

If there are a lot of platforms maybe its better to have the transform not replicated. If they are determistic all you need is for everyone agree on the time.

You can translate the platform based on the time.

You can always apply correction too since tou try to sync the time 24.7.

Better than syncing all platform. And as a bonus, the platform would be on the same position even other people have significant delay. Not sure how that will play out with cmc prediction movement though.

[dark parcel]

No, do you?

What if other people picked it up? Do we have to snatch it back from the player 🙁

[hoary spear]

Yuss!

Uno reverse

Buut i suppose you could wait. I do that too atm, but its instant anyways since i just test local atm

Was thinking of dealing with it later if it gets annoying

[orchid latch]

Hi, I'm having a rather difficult time trying to solve a crash issue after updating our project from 5.6 to 5.7.
For our game we use listen servers and SteamIP net driver (with the correct location of the net driver change in 5.6: +NetDriverDefinitions=(DefName="GameNetDriver",DriverClassName="/Script/SocketSubsystemSteamIP.SteamNetDriver",DriverClassNameFallback="OnlineSubsystemUtils.IpNetDriver"))
After the update to 5.7 we could no longer play our game over the internet, LAN was fine.

I tested out SteamSockets net driver (+NetDriverDefinitions=(DefName="GameNetDriver",DriverClassName="/Script/SteamSockets.SteamSocketsNetDriver",DriverClassNameFallback="OnlineSubsystemUtils.IpNetDriver"))
It crashed due to SteamSockets seemingly not supporting non-seamless ServerTravel (https://issues.unrealengine.com/issue/UE-239936)
So I enabled seamless travel for our Lobby and Game gamemodes and added a transition map. It works, we can now play our game over the internet.
However, if we want to restart the game map, all clients crash.
Call stacks seem to indicate that some Niagara hooks into Chaos are the cause, but what I can do about, I don't know.
We have tested this with an empty map to make sure our maps aren't doing anything weird.

We are restarting with the ServerTravel ?Restart option, same as the RestartGame AGameMode function does.
The host is launching the Lobby map with the ?Listen option.

All of the testing for this is done via a packaged development build on Steam using our Steam app ID.
In PIE this all works ok (using net.AllowPIESeamlessTravel 1 to enable seamless travel in PIE)
We are using the standard distributed Unreal Engine builds via the Epic Games Launcher.

The last thing from the client before crashing (crash seems to happen after the transition map and going back to the game map):

[2026.01.12-15.30.15:452][ 53]LogWorld: SeamlessTravel to: /Game/Maps/Map_EmptyPlane
[2026.01.12-15.30.15:467][ 55]LogWorld: BeginTearingDown for /Game/Maps/Map_EmptyPlane
[2026.01.12-15.30.15:472][ 55]LogWorld: UWorld::CleanupWorld for Map_EmptyPlane, bSessionEnded=false, bCleanupResources=true
[2026.01.12-15.30.15:473][ 55]LogSlate: InvalidateAllWidgets triggered.  All widgets were invalidated
[2026.01.12-15.30.15:507][ 55]LogAudio: Display: Audio Device (ID: 1) registered with world 'Map_Transition'.
[2026.01.12-15.30.15:510][ 55]LogAudio: Display: Audio Device unregistered from world 'None'.
[2026.01.12-15.30.15:519][ 55]LogRenderer: Forcing update for all mesh draw commands: SkyLight real-time capture change
[2026.01.12-15.30.15:531][ 55]LogUObjectHash: Compacting FUObjectHashTables data took   0.64ms
[2026.01.12-15.30.15:534][ 55]LogChaosDD: Creating Chaos Debug Draw Scene for world Map_Transition
[2026.01.12-15.30.15:534][ 55]LogStreaming: Display: FlushAsyncLoading(664): 1 QueuedPackages, 0 AsyncPackages
[2026.01.12-15.30.15:535][ 55]LogStats:     SeamlessTravel FlushLevelStreaming  -  0.000 s
[2026.01.12-15.30.15:562][ 55]LogWorld: Bringing World /Game/Maps/Map_Transition.Map_Transition up for play (max tick rate 300) at 2026.01.12-15.30.15
[2026.01.12-15.30.15:562][ 55]LogWorld: Bringing up level for play took: 0.027264
[2026.01.12-15.30.15:562][ 55]LogWorld: Sending NotifyLoadedWorld for LP: LocalPlayer_2147482468 PC: BP_GamePlayerController_C_2147481819
[2026.01.12-15.30.15:562][ 55]LogNet: Verbose: NotifyLoadedWorld Begin
[2026.01.12-15.30.15:562][ 55]LogNet: VeryVerbose: GetFunctionCallspace RemoteRole Remote ServerNotifyLoadedWorld
[2026.01.12-15.30.15:562][ 55]LogNetTraffic:       Sent RPC: BP_GamePlayerController_C /Game/Maps/Map_Transition.Map_Transition:PersistentLevel.BP_GamePlayerController_C_2147481819::ServerNotifyLoadedWorld [39.4 bytes]
[2026.01.12-15.30.15:562][ 55]LogNetTraffic: VeryVerbose: Sending: FOutBunch: Channel[5] ChSequence: 324 NumBits: 315 PacketId: 0 bOpen: 0 bClose: 0 bIsReplicationPaused: 0 bReliable: 1 bPartial: 0//0//0 bHasPackageMapExports: 0 NetTokensPendingExport: 0 
[2026.01.12-15.30.15:562][ 55]LogNetTraffic: Verbose: UNetConnection::SendRawBunch. ChIndex: 5. Bits: 315. PacketId: 13134
[2026.01.12-15.30.15:562][ 55]LogNet: Verbose: NotifyLoadedWorld End
[2026.01.12-15.30.15:562][ 55]LogWorld: StartLoadingDestination to: /Game/Maps/Map_EmptyPlane

The crash call stacks aren't always the same, here's two I've seen:

Unhandled Exception: EXCEPTION_ACCESS_VIOLATION reading address 0xffffffffffffffff

ntdll
Dolly!Chaos::FEventManager::UnregisterHandler() [D:\UnrealEngine\Engine\Source\Runtime\Experimental\Chaos\Private\EventManager.cpp:42]
Dolly!UNiagaraDataInterfaceChaosDestruction::BeginDestroy() [D:\UnrealEngine\Engine\Plugins\Experimental\ChaosNiagara\Source\ChaosNiagara\Private\NiagaraDataInterfaceChaosDestruction.cpp:333]
Dolly!UObject::ConditionalBeginDestroy() [D:\UnrealEngine\Engine\Source\Runtime\CoreUObject\Private\UObject\Obj.cpp:1295]
Dolly!UnhashUnreachableObjects() [D:\UnrealEngine\Engine\Source\Runtime\CoreUObject\Private\UObject\GarbageCollection.cpp:6158]
Dolly!UE::GC::PostCollectGarbageImpl<1>() [D:\UnrealEngine\Engine\Source\Runtime\CoreUObject\Private\UObject\GarbageCollection.cpp:5773]
Dolly!UE::GC::FReachabilityAnalysisState::PerformReachabilityAnalysisAndConditionallyPurgeGarbage() [D:\UnrealEngine\Engine\Source\Runtime\CoreUObject\Private\UObject\GarbageCollection.cpp:5970]
Dolly!CollectGarbage() [D:\UnrealEngine\Engine\Source\Runtime\CoreUObject\Private\UObject\GarbageCollection.cpp:6216]
Dolly!FSeamlessTravelHandler::Tick() [D:\UnrealEngine\Engine\Source\Runtime\Engine\Private\World.cpp:8824]
Dolly!UEngine::TickWorldTravel() [D:\UnrealEngine\Engine\Source\Runtime\Engine\Private\UnrealEngine.cpp:15563]
Dolly!UGameEngine::Tick() [D:\UnrealEngine\Engine\Source\Runtime\Engine\Private\GameEngine.cpp:1873]
Dolly!FEngineLoop::Tick() [D:\UnrealEngine\Engine\Source\Runtime\Launch\Private\LaunchEngineLoop.cpp:5829]
Dolly!GuardedMain() [D:\UnrealEngine\Engine\Source\Runtime\Launch\Private\Launch.cpp:190]
Dolly!GuardedMainWrapper() [D:\UnrealEngine\Engine\Source\Runtime\Launch\Private\Windows\LaunchWindows.cpp:123]
Dolly!LaunchWindowsStartup() [D:\UnrealEngine\Engine\Source\Runtime\Launch\Private\Windows\LaunchWindows.cpp:277]
Dolly!WinMain() [D:\UnrealEngine\Engine\Source\Runtime\Launch\Private\Windows\LaunchWindows.cpp:335]
Dolly!__scrt_common_main_seh() [D:\a\_work\1\s\src\vctools\crt\vcstartup\src\startup\exe_common.inl:288]
kernel32
ntdll

and

Unhandled Exception: EXCEPTION_ACCESS_VIOLATION reading address 0xffffffffffffffff

Dolly!UNiagaraDataInterfaceChaosDestruction::BreakingCallback() [D:\UnrealEngine\Engine\Plugins\Experimental\ChaosNiagara\Source\ChaosNiagara\Private\NiagaraDataInterfaceChaosDestruction.cpp:1922]
Dolly!UNiagaraDataInterfaceChaosDestruction::PerInstanceTick() [D:\UnrealEngine\Engine\Plugins\Experimental\ChaosNiagara\Source\ChaosNiagara\Private\NiagaraDataInterfaceChaosDestruction.cpp:2492]
Dolly!FNiagaraSystemInstance::TickDataInterfaces() [D:\UnrealEngine\Engine\Plugins\FX\Niagara\Source\Niagara\Private\NiagaraSystemInstance.cpp:1875]
Dolly!FNiagaraSystemInstance::Tick_GameThread() [D:\UnrealEngine\Engine\Plugins\FX\Niagara\Source\Niagara\Private\NiagaraSystemInstance.cpp:2605]
Dolly!FNiagaraSystemSimulation::Tick_GameThread_Internal() [D:\UnrealEngine\Engine\Plugins\FX\Niagara\Source\Niagara\Private\NiagaraSystemSimulation.cpp:1348]
Dolly!FNiagaraWorldManager::ExecuteSimulations() [D:\UnrealEngine\Engine\Plugins\FX\Niagara\Source\Niagara\Private\NiagaraWorldManager.cpp:1755]
Dolly!FNiagaraWorldManager::Tick() [D:\UnrealEngine\Engine\Plugins\FX\Niagara\Source\Niagara\Private\NiagaraWorldManager.cpp:1615]
Dolly!FTickFunctionTask::DoTask() [D:\UnrealEngine\Engine\Source\Runtime\Engine\Private\TickTaskManager.cpp:349]
Dolly!TGraphTask<FTickFunctionTask>::ExecuteTask() [D:\UnrealEngine\Engine\Source\Runtime\Core\Public\Async\TaskGraphInterfaces.h:697]
Dolly!UE::Tasks::Private::FTaskBase::TryExecuteTask() [D:\UnrealEngine\Engine\Source\Runtime\Core\Public\Tasks\TaskPrivate.h:519]
Dolly!FNamedTaskThread::ProcessTasksNamedThread() [D:\UnrealEngine\Engine\Source\Runtime\Core\Private\Async\TaskGraph.cpp:791]
Dolly!FNamedTaskThread::ProcessTasksUntilIdle() [D:\UnrealEngine\Engine\Source\Runtime\Core\Private\Async\TaskGraph.cpp:691]
Dolly!FTaskGraphCompatibilityImplementation::ProcessUntilTasksComplete() [D:\UnrealEngine\Engine\Source\Runtime\Core\Private\Async\TaskGraph.cpp:1604]
Dolly!FTickTaskSequencer::ReleaseTickGroup() [D:\UnrealEngine\Engine\Source\Runtime\Engine\Private\TickTaskManager.cpp:1035]
Dolly!FTickTaskManager::RunTickGroup() [D:\UnrealEngine\Engine\Source\Runtime\Engine\Private\TickTaskManager.cpp:2129]
Dolly!UWorld::Tick() [D:\UnrealEngine\Engine\Source\Runtime\Engine\Private\LevelTick.cpp:1743]
Dolly!UGameEngine::Tick() [D:\UnrealEngine\Engine\Source\Runtime\Engine\Private\GameEngine.cpp:1883]
Dolly!FEngineLoop::Tick() [D:\UnrealEngine\Engine\Source\Runtime\Launch\Private\LaunchEngineLoop.cpp:5829]
Dolly!GuardedMain() [D:\UnrealEngine\Engine\Source\Runtime\Launch\Private\Launch.cpp:190]
Dolly!GuardedMainWrapper() [D:\UnrealEngine\Engine\Source\Runtime\Launch\Private\Windows\LaunchWindows.cpp:123]
Dolly!LaunchWindowsStartup() [D:\UnrealEngine\Engine\Source\Runtime\Launch\Private\Windows\LaunchWindows.cpp:277]
Dolly!WinMain() [D:\UnrealEngine\Engine\Source\Runtime\Launch\Private\Windows\LaunchWindows.cpp:335]
Dolly!__scrt_common_main_seh() [D:\a\_work\1\s\src\vctools\crt\vcstartup\src\startup\exe_common.inl:288]
kernel32
ntdll

[desert breach]

Hey! This lives on the PickableActor, which is spawned on the server, its the issue of ownership, as Server owns it and the client side wants to fire audio component play, I transferred the call execution to the playercontroller client RPC, which is the owner on the client side.

[amber vale]

I'm not sure if the mesh is smoothed, but I know that the character itself isn't, which creates a jarring discrepancy from the perspective of the player, since their camera is being teleported.

[thin stratus]

I can tell you that you shouldn't put any visual stuff as a child of the capsule

It should always be parented to the Mesh

For that very reason

[twin vessel]

Do you happen to remember where the mesh location is smoothed for the locally controlled client?

I only found where it happens for simulated proxies and for clients from the listen server point of view

[viscid obsidian]

I've looked through some of the pins but I'm having trouble finding what you're talking about. Could you point me in the right direction?

This is in regards to extending the CharacterMovementComponent to support more movements

[normal crypt]

If anyone cares, no, RPCs aren't queued or cached. If a RPC is sent and a new peer joins later, he won't get that RPC.
My issue was caused by expecting I had stopped my actor from ticking, but since I wasn't the authority and the player that had authority let the actor tick, it also forced my copy of the actor to tick.

In the editor though, the authority will NOT force your copy to tick.

It's also not happening in every situation. I didn't debug this enough to figure it out.

[verbal ice]

Multicast RPCs can be queued in very specific scenarios

generally when the actor isnt available on a client yet

but is about to

afaik

that's about it

[amber vale]

I don't think it does. Mesh smoothing is done in the CMC's SmoothCorrection function, which is only called on simulated proxies or remote autonomous proxies. It even has an ensure that would prevent it from being called on local auto proxies.

[twin vessel]

Yeah, that's what i found too, but i do remember hearing about smoothing for the controlling client before

[amber vale]

If you find anything that talks about it, share it cause I'd love to see how other people do it

[twin vessel]

I guess you could interpolate the mesh component position between the pre correction position and post correction position. But this would probably interact badly with root motion (??)... not really sure.

[amber vale]

Yeah, I'm trying to get the camera component to lerp towards the new position after being corrected, since my game is first-person, and the mesh is rooted to the camera component

[eager spindle]

They will all make their own connection to the signalling server. In there, they'll each be identified by the component StreamerId

[viscid obsidian]

when extending the CharacterMovementComponent, is the idea to override Getters like GetMaxSpeed() GetMaxAcceleration() so that we can adjust simulation parameters deterministically based on boolean state flags?

[thin stratus]

#multiplayer message

[thin stratus]

No, not without having a look. Been in Mover land for too long. Don't have a PC atm (moving).

[warm willow]

guys, is it expected behaviour that notify is not firing on listen server when adding a tag to container, client is firing; But notify is firing on server/client when setting a container variable

[thin stratus]

Yes and no.

The problem is how OnReps work in Blueprints.

In C++ they don't call for Servers. The reason they call for the Server in Blueprints is that they are sort of "Property Changed" notifiers there.

But the whole thing is coupled to the Setter. The Add node doesn't trigger it.

Best thing you can do is probably set the container with itself in that case.

[warm willow]

thanks

now i see container is just array inside

[thin stratus]

Yeah, but that doesn't even matter that much. It's more the K2 node that is the Setter I believe

[warm willow]

So maybe Epic should in Add gameplay tag func make array dirty? or it would not work in BPs anyway?

thats what i was doing in c++ but I only tacked cpp replication

[thin stratus]

Idk what trigger the OnRep logic exactly

[warm willow]

push replication or whatever

[thin stratus]

Could be that it needs to do that but might be a special function call in the K2 node

Yeah it's not about push replication

You are getting the onrep on the client after all

Something inside the Setter triggers the changed notifier.

[warm willow]

Got it

[thin stratus]

Again, in C++ OnReps don't call on the server to begin with

[warm willow]

I know, you have to manually call func

[thin stratus]

You can have a Single-Player game and use the OnRep as a Property Changed notifier in Blueprints in theory. Not that you should

[warm willow]

but its magic automated in bps

[fallen fossil]

Is Struct Property replicated as whole or only its child values per change?

[chrome bay]

child values

almost all properties replicate as deltas

[modest crater]

Unless you implement a custom net serializer for your struct (just making that distinction for anyone else who might read this and not realize)

[grizzled garnet]

UE documentation mentions a Remote RPC alongside server/client/multicast. I've never seen this before, is this something new and from what version of UE?

[nova wasp]

iirc it's contextual based on who calls it to be either way

https://unreal-garden.com/docs/ufunction/#remote

It's new in 5.5. I have not tried using it yet personally

[fossil spoke]

From what I remember its possibly related to the experimental server meshing (cant remember its proper name).

[rare cloud]

MultiServerReplication ?

[fossil spoke]

That could be it yeah.

I feel like it was related to that

I could be wrong

[rare cloud]

from my understanding it's like you have an identical Server and Client RPC, you can use Remote and it will be call on the other side

but the use case seem niche

[fossil spoke]

It could very well be just a convenience like that

Might not be tied to anything in particular

Though they wouldnt have added it unless they had a reason to

In the MultiServer paradigm I wouldnt be surprised if it has something to do with like "local" authority for a particular server instance in relation to another instance next to it.

🤷

Server instances need a mechanism to talk between each other without a Player involved.

Might be related tothat

"If I am a Slave instance, send this to the Authority, if I am the Authority, send it to Slaves"

🤷

Tracking down the commit that it was introduced in might give clues

[nova wasp]

remote object is kind of the term here

basically every single TObjectPtr has code now that considers remove objects

[dark parcel]

How would you check if a controlled actor is controlled by the listen server?

GetNetMode() it is.

[verbal ice]

IsLocallyControlled()

with IsNetMode(NM_ListenServer) if need be, or just an authority check

[dark parcel]

IsLocallyControlled just check if the actor is controlled by local controller.

I need to check if the character is owned by the listen server.

all characters will return authority on server machine.

To put it into context, I am dealing with interaction on server side.
With lag in mind, I will tolerate extra interaction radius when interacting character is owned by a client.
But for server, I want the radius to be checked as is because server have no delay.

if (InteractingCharacter->GetNetMode() == ENetMode::NM_ListenServer)
{
                
}

Not sure if this is even right 🤔

Perhaps I don't think clearly...

i can probably do the opposite. Check if the controller is owned by a client, otherwise it's the server.

Get Remote role or something.

if (InteractingActor->GetRemoteRole() == ENetRole::ROLE_AutonomousProxy)
{
                
}

maybe this?

[verbal ice]

Why even bother

I'd use the same leniency on the server, it just doesn't matter

IsLocallyControlled just check if the actor is controlled by local controller.
So what do you think that does when the local controller also has a net mode of ListenServer

:p

[dark parcel]

right right

I'm getting hit too much at the gym lately.

[verbal ice]

haha

but yeah I wouldn't even bother

[dark parcel]

my test still early and I never get into the stage for a playable demonstration but you don't come across any issue when the character is at the edge of the interaction distance?

I don't want to have a scenario where the player is at the edge of the interaction radius, have the interaction data prompted as it was about to leave / enter the interaction radius and server go sike, you are off by a few units so you don't get the item.

[verbal ice]

No I meant I wouldn't bother removing the leniency for the listen server

Might as well keep it the same for everyone

@dark parcel

[dark parcel]

// Called on server.
    if (ACharacter* InteractingCharacter = Cast<ACharacter>(InteractingActor))
    {
        if (!bInteracted)
        {
            const UAGInteractionComponent* InteractionComponent = Cast<UAGInteractionComponent>(InteractingCharacter->GetComponentByClass(UAGInteractionComponent::StaticClass()));
            float _distanceToInteractingActor = (GetActorLocation() - InteractingActor->GetActorLocation()).Size();
            float _interactDistance = InteractionComponent->InteractionRadius;
            if (!InteractingCharacter->IsLocallyControlled())
            {
                // If interacting character is owned by client, allow tolerance for the interaction radius.
                _interactDistance += InteractionComponent->ClientInteractionRadiusTolerance;
            }

            // Check distance between the fragment and interacting character.
            if ((GetActorLocation() - InteractingActor->GetActorLocation()).Size() <= _interactDistance)
            {
                
            }
        }
    }

Thank you for the help btw, as this function is called on server, I can just check with IsLocallyControlled as you say.

[nova trout]

Is there a proper way to use ans in mp? I PIE with 1 instance in Client Net Mode, I expect ANS end to be called 2 times but its being called 11 times instead. I have no idea why and how to solve this.

[dark parcel]

I have no issue with ANS in multiplayer. Perhaps show what you have.

[nova trout]

UAnimInstance::Montage_Play called 2 times but ans end called 11 times

[dark parcel]

Can we get the ANS logic?

[dark parcel]

Also if you want to call this once per notify why are u using ANS instead of Anim Notify

Oh you want begin and end, nvm.

So the ANS pnly calls set state. I cant see any code that link it with your combat component.

If the reload weapon is the issue, show the ANS for that.

[nova trout]

reload weapon probably has the same behavior, I tested this with another ans end and it has the same issue.
When I put a debug point on notify begin or notify end they being called more than they need to be called.

[dark parcel]

Go to your montage and open the detail panel.

Screen shoot it

Theres montage tick type or something like that.

Or something that is called que.

https://dev.epicgames.com/documentation/en-us/unreal-engine/python-api/class/MontageNotifyTickType?application_version=5.5

Try swapping what ever you have right now, see if it helps.

I gtg sleep, if that dont work. Hopefully someone else spot the issue.

[nova trout]

neither tick type or que is in asset details of montage.

[nova trout]

good night

[dark parcel]

Im sure its somewhere there, if not check the doc and figure out where to change it.

Also just because u running 1 instance of client, there is still the server.

[nova trout]

yeah ik

thats why I expect 2 calls

[dark parcel]

Yeah try changing the montage tick type. I know it calls more than required.

Iirc shouod be branching.

But swap what ever you have right now.

https://share.google/kdtUBxcDzIhRpjAJv

[viscid obsidian]

This is the best resource I've come across for extending the CharacterMovementComponent: https://youtu.be/urkLwpnAjO0?si=Cpfou99a1LZs50Zm

the framework unreal gives out of the box is great. in Unity it took me like 3 months to make the same framework from scratch

That said, I understand why Unity doesn't

[dark parcel]

Cmc is a dinosaur, not great for games with a lot of characters.

Does unity even have multiplayer out of the box? Thought you have to download plugin or roll your own.

[viscid obsidian]

You're right. Its not out of the box, but they have proprietary solutions that you use to roll your own. More involved than unreal but much less than doing multiplayer from low level libraries.

NetCode for gameobjects is probably what you're thinking of

[nova trout]

found montage tick type defaulted as queued, setting it to branching point fixed multiple runs.

Thank you a lot ❤️

[fallen fossil]

so is

UGameplayStatics::GetPlayerController(GetWorld(), 0)

reliable to get host PC in ListenServer case?

not sure if I tagged, did not mean to do it

[warm basin]

What is best way to create loading screen with readyup/wait for players. It means if all players accept after map load, screen is removed and game starts

[silent valley]

Look into AGameMode, as it supports the concept of Match state.

[dark parcel]

Loading screen is just a widget you slap to the viewport.

You can choose to remove it when ever you want. Look up Lyra common loading screen.

[quasi tide]

Loading screen is just a widget you slap to the viewport.
It's not

Loading screens in Unreal are deceptively more complicated than what they friggin' need to be.

Sure - you can probably throw a widget on the screen and add it to the root and all that. But that has issues (that elude me right now, busy at work). The most common approach, that has been pretty successful, is hooking into UE's movie player. I know - doesn't make a whole lot of sense. But one of the big reasons is because when it gets created and destroyed and when it gets called. Also it is on a different thread if I recall.

[dark parcel]

Widget by default get destroyed on level change.

By making the UUserWidget a shared ptr, it stops them from being destroyed and GC. I havent copy anything else from lyra and havent play with the movie player.

But my loading screen can be removed and added at anytime. Atm i wait till my map finished load and for my cosmetic data to load.

I may be wrong but I think exi says, even lyra loading screen isnt truly async?

I probably just havent encounter future issues yet.

[humble idol]

For hard travels, I don't think you can use widgets, you have to run a "movie" or image on a separate designated thread.

[dark parcel]

Depends on what you mean by cant. If you mean the widget get destroyed then no, you can prevent that.

Infact i am hard traveling soo far. Havent got the need to server travel yet.

[humble idol]

Aren't widgets still rooted in the "world?" I thought that everything basically gets destroyed during hard travels, causing a synchronous load on the game thread.

[dark parcel]

If you make it a shared ptr it will prevent the widget from being destroyed and gced.

If its a raw ptr then yeah they get destroyed on world change.

[humble idol]

Is the widget still displayed and running during the synchronous load?

[dark parcel]

Yes

[humble idol]

Hmmm, that's interesting. I'll have to try it.

[dark parcel]

Survive both level. I spawned mine in subsytem and make it shared ptr.

Async load level then open level (hard travel)

Only fade it out after conditions are met in the new world.

[grizzled garnet]

When a client joins a multiplayer match, does the player's state get spawned before or after the level's actors have spawned?

[fossil spoke]

On the Server side, the PlayerState is created as soon as the PlayerController is created.

On Clients, its arbitrary because its created through replication.

[grizzled garnet]

ok tldr I need some way to defer spawning the player server side until functions client-side complete, like loading a sublevel. An actor is handling the sublevel loading manually, and not all levels require a sublevel and the actor.

I'm suspecting replication should handle the sublevel to the client, so I suspect currently with the sublevel not yet loaded then the server is spawning players before it's loaded, which I could handle via the gamemode searching if the actor exists and waiting until they've completed their task before spawning players

[latent heart]

Or spawn players in a temporary "spectator" pawn until things are ready for everyone.

[grizzled garnet]

I could do that, I just need some way where actors spawned in the level can tell the gamemode that things are ready

I have an actor which does that (gametype, which not to be confused with gamemode, handles how gameplay occurs including spawn rules) but in my current code its not finding the actor which it needs to wait for

oh, the issue is because the gametype actor is in a different always loaded level, so the ugameplaystatics::GetAllActorsOfClass isn't finding the actor

[latent heart]

BeginPlay() -> Server RPC

[nocturne quail]

what should I choose for a stuff like shrinkable bluezone that covers the whole map and resize over time
Subsystem or Actor with few components ?

Responsibilities of the bluezone system:

1: Dealing damage to players
2: Bombing a random point in available playzone
3: Generating a random airdrop point
4: Spawn airdrop plane that move towards airdrop point, when it is reached drop an airdrop package

[latent heart]

If it needs a visual representation, an actor. If it's just a map thing, a subsystem is probably fine. You could do both. Have the subsystem spawn the actor.

Map = minimap / map overview.

[nocturne quail]

My current approach

void AArmaGameMode::BeginPlay()
{
    Super::BeginPlay();
    
    BlueZoneManager = GetGameInstance()->GetSubsystem<UBlueZoneManager>();
    BlueZoneManager->Initialize(this);
    
    ZoneVisualizer = GetWorld()->SpawnActor<ABlueZoneVisualizer>();
    BlueZoneManager->SetVisualizer(ZoneVisualizer);
}

[latent heart]

Seems fine.

Though I'd probably have just thr server spawn it and have it replicate.

In case you want to transfer any data about it.

[nocturne quail]

ZoneVisualizer is replicated just for visual representation

[latent heart]

Yes, but that subsystem is going to spawn one on both server and clients. Unless the subsystem is only created on servers...

[nocturne quail]

Yeah subsystem is only on server

it does all other stuff like

-- Global state management (current zone phase, timers, damage)
-- Damage calculation and application to all players
-- Zone shrinking logic and timing
-- Random point generation for bombing/airstrikes
-- Airdrop scheduling and coordination

[fallen fossil]

I need to validate one thing.

Does Multicast runs locally if actor does not own connection ? 🤔

[latent heart]

Correct

[torpid storm]

Anybody here who has got far enough to be hosting dedicated servers - are you opting for the -nothreading option?
I think traditionally it has been used to allow for more servers to share a single machine, but it implies that the server would need to run at a lower tick rate, plus I'm unsure how well things like HTTP background tasks work in that scenario.

[fallen fossil]

you are smarter than GPT 😆

[exotic wasp]

depends on the game. if you heavily use threading you might have to keep it on

many games are made to run on a single core or even be on a shared core. all depends on how the game was made and how difficult it is to run

even with threading on there are ways to force a program (like your server) to run only on assigned cores

[torpid storm]

I'm not really asking for advice, just to get an idea of what people are doing. By default UE spreads things over a small number of threads, and obviously on a server the render thread can be ditched which reduces the count further, but it's not 1. So I'm curious how people are approaching this.

[silent valley]

I profiled my server on 5.3, and it was pretty much single threaded anyway. The other threads had v little activity.

[lament flax]

Usually a MT server costs to much so games works with single threaded servers

[copper raft]

heyy guys, do i need to set input mapping context to work on multicast if i want to switch them when enter in a vehicle? i tried to ask google gemini but he said input mapping needs to run only on owning client, but when i'm trying to do how he said nothing works, with multicast works fine but when my car starts to go, my second player from passengers seat starts to changing his location when vehicle is moving

i tried to make this event, but i don't understand why it shows Enhanced Input Local Player Subsystem is not valid

[tardy fossil]

players possess the vehicle right? i'd switch the input context in the pawns SetupPlayerInputComponent function which gets automatically called on the client

[copper raft]

yes, player successfully possess the vehicle, this event is inside my car blueprint and works fine

if i call this event after possess node, all works fine, but i don't understand why gemini said to use run on owning client because is better

[copper raft]

you mean to switch them inside player controller blueprint?

[tardy fossil]

no it'd be inside the vehicle class.. in c++ i'd use SetupPlayerInputComponent() but im not sure about blueprints.. maybe client restart? i just know you dont need to use an rpc

[dark parcel]

Especially a multicast to setup an input of a client. That make 0 sense.

[copper raft]

so input switch needs to run only on owning client, without server?

[tardy fossil]

yeah input context is completely local so it doesn't exist on the server... server only cares about the data sent from whatever those inputs are doing locally

[copper raft]

ok i will try again thanks

[copper raft]

omg, i just changed the possess node after call Client_SetupVehicleInput node and now works fine😅 , and i spent 3 days on this thing😵‍💫

[gritty mason]

Hi guys! I have a strange problem. I develop a listen server multiplayer game in 5.4 and when the players arrived in the Lobby everything is fine ,they can control the character but when we go to the game map, the client can not control the character, the character controller is None, the client doesn't see the host character when the host moving around it. I use seamlesstravel and bdelayedstart. What could be the problem? I've tried almost everything. GameMode and GameState is the same, in the character Blueprint the auto possess player is Disabled.

[nocturne quail]

if you run a timeline on both server and client to set transform of a UStaticMeshComponent, why both desynced?
generally both should have the same transform for the mesh! but they don't 🤔

[nocturne quail]

the issue was server first set transform and than client is trying to set it also, in result client get stuck cause server is again applying new target scale...
solution was to set the UStaticMeshComponent replicated and let the server set its transforms

[nova wasp]

what replicates? what changes? what times the timeline? what does the timeline actually do? does anything else change the transform?

and yeah of course if they are both fighting over setting it the last one will apply... if you need a separate local only value make a separate thing

replication is not going to send at a perfect framerate either so you can safely assume it will jitter without local interp

[nocturne quail]

this was started by both server/clients

void ABlueZone::TimelineProgress(float Value)
{
    NewScale = FMath::Lerp(StartScale, TargetScale, Value);
    ZoneSphereMesh->SetWorldScale3D(NewScale);

    NewPosition = FMath::Lerp(StartPosition, TargetPosition, Value);
    ZoneSphereMesh->SetWorldLocation(NewPosition);
}

[nova wasp]

edited some spelling mistakes, very late for me so I gotta go

"server first set transform and than client is trying to set "

they are separate machines... what else does the server do to send the result to the client? what is actually "synced" at all besides the timeline starting?

[nocturne quail]

yeah it was started in beginplay

yeah this is what log says at least, first server started than client started

[nova wasp]

I would suggest using a visualization like vislogger

instead of logging transforms... that is very hard to read

at least for me personally

[nocturne quail]

actor was replicated, so any change made in the actor server replicates

[nova wasp]

that is the context you left out that makes what is happening here a lot easier to figure out

remember: we can only see what you type into the messsage

or is that what you changed after the original message? either way I can't assume the transform is replcated based on the first question but oh well

with this stuff It's really dependant on what the goal is imo... For example arguably you don't need to know about the true end position if the start position + start time is known and decently synchronized on both sides (accurate server time etc)

[nocturne quail]

Yeh I logged, server : 500, 500, 1000 client : 500, 500, 1000 in the same frame and then server sets new value in next frame and client skip the next frame change, in result we have difference

[nocturne quail]

its important for both to have same transform values, cause if client is outside from zone on server it recieve damage and on client side it still not outside of zone cause the client has different transform values

[nova wasp]

I don't know what "the client" is in relation to a random static mesh

[nocturne quail]

static mesh is visually represent as a wall for client

[nova wasp]

why is the wall moving? is it the zone edge?

[nocturne quail]

yes

[nova wasp]

that would have helped to know earlier because I think that kind of thing is going to change very slowly over time and in a consistent way

you only really need to send when it starts and the client just needs to represent where it is going visually imo but if it changes speed/direction often it might be worth keeping realtime

[nocturne quail]

it moves to a random point and scaled down
red circle is server transforms, which don't match to the client transforms

[nova wasp]

all they need to do is share the same known min/max size and their time along it

[nocturne quail]

for late clients it also needs to be synced

[nova wasp]

I don't recall saying this is an rpc (in the sense it's only sent once ever)

clients also can receive world time from the gamestate (via requesting it or replicating it down etc)

https://vorixo.github.io/devtricks/non-destructive-synced-net-clock/

this is not needed though since there will only really be 1 of these and it being ~100ms off is not the end of the world

[nocturne quail]

🤣

[nocturne quail]

my current approach to keep it synced while minimal rep load

I am sure it can be further improved, but it works perfectly at least

[nocturne quail]

I think I have to move this new system to the gamemode sybsystem, and use instance AInstanceStaticMeshActor for visual effects

AInstanceStaticMeshActor->UInstancedStaticMeshComponent supports runtime scaling/repositioning?

[nocturne quail]

one more optimization, these don't needs to be replicated, they will be the same as targetposition and targetscale in the end

    FVector NewPosition;
    FVector NewScale;

[dark edge]

replicated initial state and synced network clock

CenterPosition and StartTime is enough

[nocturne quail]

I'm using ServerSnapshot as the initial state for late clients

USTRUCT()
struct FBlueZoneSnapshot
{
    GENERATED_BODY()
    FVector Position;
    FVector Scale;
    int32 PhaseIndex;
    float Timestamp;
    float ShrinkProgress;
    FVector BomberTargetPosition;
    bool bIsRedZoneActive;
    bool NetSerialize(FArchive& Ar, class UPackageMap* Map, bool& bOutSuccess);
};

I would like to further optimize those replicated vectors, is it possible to apply bit packing on them? its a crazy idea but if it is possible and will work it will be a game changer!

these

    UPROPERTY(ReplicatedUsing = OnRep_TransformState)
    FVector TargetScale;

    UPROPERTY(ReplicatedUsing = OnRep_TransformState)
    FVector StartScale;
    
    UPROPERTY(ReplicatedUsing = OnRep_TransformState)
    FVector TargetPosition;

    UPROPERTY(ReplicatedUsing = OnRep_TransformState)
    FVector StartPosition;

[dark edge]

Why do you need to replicate progress and scale, that's deterministically derivable from StartTime and the synced clock

[nocturne quail]

I tried to make client locally progress but it was problematic, a small ping drop will desync the progress with the server progress

and about synced clock I read the article https://vorixo.github.io/devtricks/non-destructive-synced-net-clock/
looks like the system will need big changes to apply this

USTRUCT()
struct FPackedVector
{
    GENERATED_BODY()
    
    uint32 PackedX : 24;  // 0-16,777,215 cm (167km) map support
    uint32 PackedY : 24;
    uint32 PackedZ : 16;  // Usually 0 for blue zone
    
    // Unpack to FVector
    FVector Unpack() const
    {
        return FVector(
            PackedX / 100.0f,  // Convert cm to meters
            PackedY / 100.0f,
            PackedZ / 100.0f
        );
    }

introducing custom vector for this specific task

    void Pack(const FVector& Vec)
    {
        PackedX = FMath::Clamp(FMath::RoundToInt(Vec.X * 100.0f), 0, 0xFFFFFF);
        PackedY = FMath::Clamp(FMath::RoundToInt(Vec.Y * 100.0f), 0, 0xFFFFFF);
        PackedZ = FMath::Clamp(FMath::RoundToInt(Vec.Z * 100.0f), -32768, 32767) & 0xFFFF;
    }
    
    bool NetSerialize(FArchive& Ar, class UPackageMap* Map, bool& bOutSuccess)
    {
        bOutSuccess = true;
        Ar << PackedX;
        Ar << PackedY;
        Ar << PackedZ;
        return true;
    }

and finally

USTRUCT()
struct FOptimizedZoneUpdate
{
    GENERATED_BODY()
    
    uint8 Flags;

    // Only send what changed
    UPROPERTY()
    FPackedVector Position;
    
    UPROPERTY()
    FPackedVector TargetPosition;
    
    UPROPERTY()
    FPackedScale Scale;
    
    UPROPERTY()
    FPackedScale TargetScale;
    
    // Phase index (0-7 fits in 3 bits, but using 4 for safety)
    uint8 PhaseIndex : 4;
    
    // Shrink progress (0-100%, 7 bits = 128 values)
    uint8 ShrinkProgress : 7;
    
    bool HasPositionUpdate() const { return (Flags & 0x01) != 0; }
    bool HasScaleUpdate() const { return (Flags & 0x02) != 0; }
    bool HasTargetUpdate() const { return (Flags & 0x04) != 0; }
    bool IsShrinking() const { return (Flags & 0x08) != 0; }
    bool IsRedZoneActive() const { return (Flags & 0x10) != 0; }
    
    void SetHasPositionUpdate(bool bValue) { Flags = bValue ? (Flags | 0x01) : (Flags & ~0x01); }
    void SetHasScaleUpdate(bool bValue) { Flags = bValue ? (Flags | 0x02) : (Flags & ~0x02); }
    void SetHasTargetUpdate(bool bValue) { Flags = bValue ? (Flags | 0x04) : (Flags & ~0x04); }
    void SetIsShrinking(bool bValue) { Flags = bValue ? (Flags | 0x08) : (Flags & ~0x08); }
    void SetIsRedZoneActive(bool bValue) { Flags = bValue ? (Flags | 0x10) : (Flags & ~0x10); }
    
    bool NetSerialize(FArchive& Ar, class UPackageMap* Map, bool& bOutSuccess)
    {
        bOutSuccess = true;
        
        // Always serialize flags
        Ar << Flags;
        Ar << PhaseIndex;
        Ar << ShrinkProgress;
        
        // Only serialize what changed based on flags
        if (HasPositionUpdate())
        {
            Ar << Position;
        }
        
        if (HasScaleUpdate())
        {
            Ar << Scale;
        }
        
        if (HasTargetUpdate())
        {
            Ar << TargetPosition;
            Ar << TargetScale;
        }
        
        return true;
    }
};

[rare cloud]

free code, let's go

[nocturne quail]

uint32 ABlueZone::PackPosition(const FVector& Pos) const
{
    // Pack position with 1cm precision
    // Map is 800000cm (8km), so we need 20 bits per axis (0-1,048,575)
    // We'll use 21 bits to be safe (0-2,097,151)

    int32 X = FMath::Clamp(FMath::RoundToInt(Pos.X * 100.0f), 0, 0x1FFFFF); // 21 bits
    int32 Y = FMath::Clamp(FMath::RoundToInt(Pos.Y * 100.0f), 0, 0x1FFFFF); // 21 bits
    int32 Z = FMath::Clamp(FMath::RoundToInt(Pos.Z * 100.0f), -1024, 1023) & 0x7FF; // 11 bits

    // Pack: 21 bits X | 21 bits Y | 11 bits Z = 53 bits total
    // We'll split across two 32-bit values
    uint32 Packed = (X & 0x1FFFFF) |
        ((Y & 0x1FFFFF) << 21) |
        ((Z & 0x7FF) << 42);

    return Packed;
}

error C4293: '<<': shift count negative or too big, undefined behavior

🤔

[rare cloud]

this is because you manipulate int32 and put value into an uint32

oh and also << 42 can't be possible for a 32-bit integer

even the comment talk about two 32-bit "We'll split across two 32-bit values"

[nocturne quail]

yeah I want int64 , thanks

uint32 ABlueZone::PackPositionX(const FVector& Pos) const{
    // Pack X and part of Y into first 32 bits
    // X: 21 bits (0-2,097,151)
    // Y_low: 11 bits (lower 11 bits of Y)
    
    int32 X = FMath::Clamp(FMath::RoundToInt(Pos.X * 100.0f), 0, 0x1FFFFF);
    int32 Y = FMath::Clamp(FMath::RoundToInt(Pos.Y * 100.0f), 0, 0x1FFFFF);
    uint32 Packed = (X & 0x1FFFFF) | ((Y & 0x7FF) << 21);
    return Packed;
}

uint32 ABlueZone::PackPositionY(const FVector& Pos) const{
    // Pack remaining Y and Z into second 32 bits
    // Y_high: 10 bits (upper 10 bits of Y)
    // Z: 11 bits (signed -1024 to 1023)
    // Remaining 11 bits unused
    
    int32 Y = FMath::Clamp(FMath::RoundToInt(Pos.Y * 100.0f), 0, 0x1FFFFF);
    int32 Z = FMath::Clamp(FMath::RoundToInt(Pos.Z * 100.0f), -1024, 1023);
    uint32 ZPacked = (Z < 0) ? ((~(-Z) + 1) & 0x7FF) : (Z & 0x7FF);
    
    uint32 Packed = ((Y >> 11) & 0x3FF) | (ZPacked << 10);
    return Packed;
}

[nova wasp]

this kind of packing into 1 32 bit doesn't make as much sense to be as two 16 bits

depending on the ranges

most integer types in unreal are serialized around the idea of them being a lower range of values

from what I can tell from the source

but at this scale it won't matter much to shave off 4 bits

[nocturne quail]

yeah I switched back to snapshot packing which is more simple and only takes one packed struct with all data needed for replication

[nova wasp]

that's good because it will all arrive at once

[nocturne quail]

and gives super smooth update

[vocal current]

If you're using NetSerialize then there are already some custom ways to serialize vectors, if I remember right

As opposed to the default Ar << MyVector

[meager spade]

you can searchfor NetSerialize

we normally do like FVector_NetQuantize& ReplicatedLocation = ReplicatedLocations[i]; ReplicatedLocation.NetSerialize(Ar, Map, bOutSuccess); if we dont need precision

theres FVector_NetQuantize, FVector_NetQuantize10 and FVector_NetQuantize100

 *
 *    2 decimal place of precision.
 *    Up to 30 bits per component.
 *    Valid range: 2^30 / 100 = +/- 10,737,418.24```

for example

[nocturne quail]

I will do it next time when I will need it for sure

[meager spade]

these can reduce rep sizes

we have done in the past, where we exposed normal FVector for gameplay

[nocturne quail]

// server sets it and calls onrep.
FZoneSyncData NewSyncData;
NewSyncData.PhaseIndex = NextPhaseIndex;
SyncData = NewSyncData;
OnRep_SyncData();
    
void ABlueZone::OnRep_SyncData()
{
    UE_LOG(LogTemp, Warning, TEXT("OnRep_SyncData called - Phase: %d, HasAuthority: %s"),
        SyncData.PhaseIndex, HasAuthority() ? TEXT("Yes") : TEXT("No"));

}

it only logs called by server, no idea why

[meager spade]

but made it not replicated, then handle the rep via a quantized and set back to the normal property

this way to the end user, its just an FVector

you really should avoid server calling OnRep_ like that

you should make intermediate functions

that OnRep and server called, like a shared path

i really hate people doing OnRep_Blah on the server side path

its ambiguous

[nocturne quail]

for dedicated server setup it also matter?

[meager spade]

yes

cause the issue is, OnRep == Client callback when property changes

if server ends up calling it, its really hard to debug

it becomes a nightmare

i can look at my callstack and be like "Thats server side"

but if i see OnRep_ to me its client side

but if server is calling OnRep then thats not client side..

[nocturne quail]

Adding an intermediate now, if it fixed, it will make my day 😄

[meager spade]

whats the issue anyway

i didnt read

also this is wrong

[nocturne quail]

the onrep only calls by server

[meager spade]

    // Pack X and part of Y into first 32 bits
    // X: 21 bits (0-2,097,151)
    // Y_low: 11 bits (lower 11 bits of Y)
    
    int32 X = FMath::Clamp(FMath::RoundToInt(Pos.X * 100.0f), 0, 0x1FFFFF);
    int32 Y = FMath::Clamp(FMath::RoundToInt(Pos.Y * 100.0f), 0, 0x1FFFFF);
    uint32 Packed = (X & 0x1FFFFF) | ((Y & 0x7FF) << 21);
    return Packed;
}

uint32 ABlueZone::PackPositionY(const FVector& Pos) const{
    // Pack remaining Y and Z into second 32 bits
    // Y_high: 10 bits (upper 10 bits of Y)
    // Z: 11 bits (signed -1024 to 1023)
    // Remaining 11 bits unused
    
    int32 Y = FMath::Clamp(FMath::RoundToInt(Pos.Y * 100.0f), 0, 0x1FFFFF);
    int32 Z = FMath::Clamp(FMath::RoundToInt(Pos.Z * 100.0f), -1024, 1023);
    uint32 ZPacked = (Z < 0) ? ((~(-Z) + 1) & 0x7FF) : (Z & 0x7FF);
    
    uint32 Packed = ((Y >> 11) & 0x3FF) | (ZPacked << 10);
    return Packed;
}```

you dont want to be backing your vectors like this

[nocturne quail]

since it should called by clients/server both when a server changes the property

this only logs has auth yes

void ABlueZone::OnRep_SyncData()
{
    UE_LOG(LogTemp, Warning, TEXT("OnRep_SyncData called - Phase: %d, HasAuthority: %s"),
        SyncData.PhaseIndex, HasAuthority() ? TEXT("Yes") : TEXT("No"));

}

[meager spade]

but why this weird packing

[nocturne quail]

UPROPERTY(ReplicatedUsing = OnRep_SyncData)
FZoneSyncData SyncData;

[nocturne quail]

I don't have it anymore

[meager spade]

what you using instead?

[nocturne quail]

clients are doing this locally now

[meager spade]

doing what locally

show the code

cause im lost

[nocturne quail]

I was doing it for replication purpose to make the packets smaller, but now clients are simulating the timeline locally

[nocturne quail]

I was doing all that custom serialization to minimize this load

void ABlueZone::GetLifetimeReplicatedProps(TArray<FLifetimeProperty>& OutLifetimeProps) const
{
Super::GetLifetimeReplicatedProps(OutLifetimeProps);

// OPTIMIZATION: Only replicate to simulated proxies (clients)
DOREPLIFETIME_CONDITION(ABlueZone, TargetScale, COND_SimulatedOnly);
DOREPLIFETIME_CONDITION(ABlueZone, StartScale, COND_SimulatedOnly);
DOREPLIFETIME_CONDITION(ABlueZone, TargetPosition, COND_SimulatedOnly);
DOREPLIFETIME_CONDITION(ABlueZone, StartPosition, COND_SimulatedOnly);
DOREPLIFETIME_CONDITION(ABlueZone, NewScale, COND_SimulatedOnly);
DOREPLIFETIME_CONDITION(ABlueZone, NewPosition, COND_SimulatedOnly);
DOREPLIFETIME_CONDITION(ABlueZone, bTransformRep, COND_SimulatedOnly);
DOREPLIFETIME_CONDITION(ABlueZone, BomberTargetPosition, COND_SimulatedOnly);
DOREPLIFETIME_CONDITION(ABlueZone, ServerSnapshot, COND_InitialOnly); // Only for late join

}

now using this synced with the game time

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

USTRUCT()
struct FZoneSyncData
{
    GENERATED_BODY()

    // Phase timing
    UPROPERTY()
    int32 PhaseIndex = -1;
    
    UPROPERTY()
    float ServerStartTime = -1.0f;
    
    UPROPERTY()
    float Duration = 0.0f;
    
    // Geometry
    UPROPERTY()
    FVector StartPosition = FVector::ZeroVector;
    
    UPROPERTY()
    FVector TargetPosition = FVector::ZeroVector;
    
    UPROPERTY()
    float StartScale = 0.0f;
    
    UPROPERTY()
    float TargetScale = 0.0f;
};

[halcyon ore]

Hey all.
I have a weapon system.
Its a weapon actor attached to the player.

I'm having an issue, where the server tries to replicate the weapon actor, before the player/ owner.
(even with the use owner relevancy or whatever)
This appears to be, because the server thinks the client knows about the owner, but just doesn't. (since at the end of the day, its a distance check)

Is there anyway to change this, so the server/ client both know there owner is valid?
Or, do I just need a begin play loop, waiting for a valid owner?

(Also, specifically just the clients, server always has a valid owner, not like the owner var just isn't set.)

[split drum]

Hello, I'm new to Unreal and Multiplayer and have been having an issue with my RPCs in my Listen Server game.
I have a function that spawns a debug sphere in front of the player.

UFUNCTION(Server, Reliable, BlueprintCallable)
void Server_SpawnDebugBall();

UFUNCTION(NetMulticast, Reliable)
void NetMulticast_SpawnDebugBall();

UFUNCTION(BlueprintCallable)
void SpawnDebugBall();

void AFourManScrambleCharacter::NetMulticast_SpawnDebugBall_Implementation()
{
    SpawnDebugBall();
}

void AFourManScrambleCharacter::Server_SpawnDebugBall_Implementation()
{
    NetMulticast_SpawnDebugBall_Implementation();
}

void AFourManScrambleCharacter::SpawnDebugBall()
{
    FActorSpawnParameters SpawnParams;
    SpawnParams.Owner = this;
    FVector SpawnLocation = ItemSpawnPoint->GetComponentLocation();
    FRotator SpawnRotation = ItemSpawnPoint->GetComponentRotation();

    AActor* SpawnItem = GetWorld()->SpawnActor<AActor>(SummonItem, SpawnLocation, SpawnRotation, SpawnParams);
}

I have a key-input that runs Server RPC in my C++ code. While I'm playing as the server, it will spawn the ball for both the server and the client. But when I press the button as the client it only spawns on the client

As well I tried hooking it up through blueprints instead. Of the three methods on the right, the top one is the former I had just mentioned. The middle one, (which you can see implemented in the bottom left), will spawn two balls in front of the calling player, and the bottom one, which just directly calls the Server RPC function, will work correctly, spawning a ball in front of the appropriate player on both the client and server.
Can anyone help me understand how these variations are happening?

[crisp shard]

when it comes to equipment / itmes that you can equip, would it make sense to "pool" these actors? seems like it would make sense in the way i understand the benefits of pooling. i suppose same would go for item drops in general, like if you were to "drop" an item, this is current an actor, but not sure if there's a way to make it not a full blown actor , as in most cases, it's just pointing to a single int32 value which is just an index to the actual item to pickup. if i made them non-actors somehow (and stil could interact the same way), then i'd think pooling would make sense

i've also considered the issues of equipables being pooled as they would be replicated actors, and having those just sitting not used also seems like it could be unnecessary, but yea curious on what somet akes on pooling things like item drops or equipment like weapons that would be effectively spawned and despawned potentially frequently (equip / dequip, etc, or dropp item -> pick up item)

[halcyon ore]

To me personally.
Pooling doesn't make sense for stuff like that.
It also very much depends on the game.
You say, your items are an index to the actual item.
But, it may not be like that for some games.

and, then like you say for replication.
That just an extra actor that exists for no reason.

Pooling I feel like only makes sense, where an inventory is sorta locked.
Think looter shooters.
You don't have like 30 guns.
You have like 5 static ones.
Pistol, machine gun, melee, grenade, idk. 😛

You could also conceal the pooling/ loading in the unequip.
Sure, that still the "issue" of entire actor creation and shit.
But, thats 1 gun, compared to like 100 bullets, from a machine gun, or something.

[crisp shard]

that makes sense, will most likelly avoid the pooling for equipables. for item drops tho, im thinking of instead of spawning the item, have a single actor item drop manager or something w an item struct that can replicate the Item id and it's location... which in my head i think would work, but not sure about if multiple were spawned at once or something how it would handle that but maybe i can do soemthing like that for items. but could prob just keep them actors too (if single items). loot chest/storage boxes would remain actors as they aren't something quickly destroyed

[halcyon ore]

The way I kinda see it.
In more dynamic games, a weapon/ tool, could be 50 different things, or just do nothing at all.
While most shooter games (arma, looter shooter), you kinda only can shoot a million bullets.
But, a survival game, could have shovel, pickaxe, sword, full auto gun, primitive crossbow, bow and arrow, nothing.
Theres so many possibilities, you'd be pooling like 50 extra weapons/ actors.

I do agree, that some sort of item drop manager could work better, especially with such basic item drops like you said.
Could also instanced static mesh it, and lookup the drop, based on the instanced index.

[crisp shard]

yea 100p when you describe it like that, it would be a nightmare pooling all the different possiblities lmao. i also like the ISM approach or even a simple niagara to represent it via sprite, could put that on the itme manager, then just do a "distance check" to the location to pickup the item, as i most likely will not be using collision on these, that would prob do more harm than good. and distance check could just pick up items by closest / most relevant

[halcyon ore]

Yeah, it very much depends.
I could never use just a basic int32 var, for my items. lol

[crisp shard]

lmao, well it's just an item id which points to the actual struct w the info needed, but yea in my case, majority of items are sprites so easy to represent them in world w just an index

[halcyon ore]

Oh, I see. 😛
I thought your items were legit just a int32, for an index on array, or some DataTable. lmao

Why have a int point to the item, and not just the item in the dropped item.
Out of curiosity?
Is that just how you designed your game, or did you have some performance, or like networking in mind?

[crisp shard]

tbh i just thought it would be easier to replicate and send around a int32 and then do the heavier strcut work later when needed rather than it's full item strcut, but it may very well be unnecessary but all my items have an ID which points to the struct in a datatable. in fact, i most likely convert it in the pick up logic of the actor version rn, immediatley , so it is probably unecessary lmao, but again, that variable is just replicated so i thought it would be a bit more efficent, and i wouldn't need all the extra info other than it's visual representation

im def always thinkinga bout performance, but given that im using zero new tech (no nanite, lumen) on forward rendering, and have essentially zero shadows/lighting, i prob have plenty of room lol

and all my charcters/items are sprites, so dont use skeletal meshes at all

biggest bottleneck is probably just the size of the world

and im not currently using world partition (at least i don't htink i am)

[halcyon ore]

I tried like 5 different methods for my complex item replication.
None of them worked.
So, it was either low level, custom replication channel.
or, RPC spam.
I chose RPC spam. 😛

[crisp shard]

loll whatewver will get the job done and keep perfomrance maintainable is all we need, but ciurous what you mean by RPC spam

i have some functions that i set to reliable after realizing GAS abilites are reliable by defualt (at least as far as i understand them)

i dont use GAS tho ( i did for a month, never again... unless...)

[halcyon ore]

RPC spam.
RPC for item add, item remove, item move, item drop, item changes.
Recently did custom serialize shit, so I can pass var changes around as bytes, rather then 1 RPC per possible var.

Its the only way I could really find to do it.
Got some smart people here, but all the methods they suggested would not have worked.
(This is only like 50% of my RPC's, for my item system.)

[crisp shard]

ohhhh yea, that's exactly what im doing, easily the worst part of the user experience for my game is moving items around, but i just coudlnt think of any easy way to do some sort of client preiction or making things feel smoother

all reliable lmao i love it

i never used reliable casue i thouht it was "bad" then like i said , once i realized how GAS worked, i was kinda pissed casue i could've had so much easier logic by setting some things to reliable. of course, they are "gameplay critical" things (that's the term i've heard)

[halcyon ore]

Yeah.
If you need it to happen.
Such as item moves, then reliable makes sense.

For instance, player movement isn't reliable.
Cuz, while it could be seen as "critical", it also interpolates, so whatever var it recieves, will account for the lack of other RPC's it sends.

RPC's also help, for the 4 different ways my items can replicate.

[halcyon ore]

Your calling the _Implementation
Don't do that

[split drum]

Oh! So I just call the "NetMulticast_SpawnDebugBall"?

[halcyon ore]

yeah

Implemetation is like the I recieved the replication sorta stuff.

[split drum]

Ah gotcha gotcha!

Thank you 🙏 It worked

[fiery wadi]

Hi, can someone explain why the 1st controller Controller_0 doesnt spawn a character but the 2 conneccted clients do please? Even with the breakpoint Controller_0 never even triggers the Set Actor Location, I did consider destroying the Pawn and then trying to spawn a new one and assigning the Owner of the new pawn to the ConnectedController but feels a bit odd to destroy a character just to create another.

[fiery wadi]

Nvm I forgot I overrode CHoosePlayerStart which was returning nothing....

[halcyon ore]

Reposting this, cuz now its hidden behind all these other messages. 😛

I have a weapon system.
Its a weapon actor attached to the player.

I'm having an issue, where the server tries to replicate the weapon actor, before the player/ owner.
(even with the use owner relevancy or whatever)
This appears to be, because the server thinks the client knows about the owner, but just doesn't. (since at the end of the day, its a distance check)

Is there anyway to change this, so the server/ client both know there owner is valid?
Or, do I just need a begin play loop, waiting for a valid owner?

(Also, specifically just the clients, server always has a valid owner, not like the owner var just isn't set.)
Thankfully its not exactly a critical issue right now, so i'm not in a rush to care about/ fix it. 😛

[viscid obsidian]

for extending the CharacterMovementComponent: when networking one off events like a dash, should we extend the SavedMove struct to auto replicate, resulting in larger bandwidth per frame, or should we just send a one time ServerRPC to replicate?

[nova wasp]

serializing something like a dash can be a single bit that says if it's in the current structure or not... this is not going to be a bandwidth concern if you put any thought into how it is sent

[viscid obsidian]

right that's what I was thinking, for some reason the tutorial i was following used an rpc. maybe he just wanted to show how you could do it for some reason you needed to

[nova wasp]

using an rpc will basically mean doing it outside of the cmc by turning off corrections during the ability etc (if it ends up causing corrections)

but you can try that if it seems easier

[viscid obsidian]

for the serialization, are you saying you could have a bool like bWantsToDash that indicates whether or not any additional dash variables are stored in the networked struct?

[nova wasp]

no need for a loop of any kind here really... you can just have both beginplays check for the presence of the other and maybe a local trackimg thing to see if they initialized with X thing

[nova wasp]

that or basically having a single bool in the serialized thing representing if the dash has data in it etc... replicating a boolean is absurdly cheap. The cost to avoid here is sending new floating point values if you don't need to (or at least can avoid it in the common case)

[viscid obsidian]

thank you

[nova wasp]

but honestly you might be able to spam extra fvectors in there and never really care if bandwidth is never under pressure so profile it I guess

[halcyon ore]

I thought of that.
I guess I wanted it to be 1 sided coding.
The weapon manages itself, and the player just spawns it.

If that’s the only logical method, then that’s what I’ll do.

[nova wasp]

or just change it to be that the weapon replicates as a subobject (component/or new uobject) of the player pawn
and then have the actor created locally when it arrives to just serve as visuals/collision etc
(this is of course a big change to make and not always nicer but to me it would reduce confusion)

[nocturne quail]

what happens if an actor in world call ForceNetUpdate(); when it is loaded for late join client?

[nova wasp]

I assume it would replicate stuff regardless of net update rate/prio? Are you not seeing it arrive on the client?

[halcyon ore]

Can you not do that with just direct actors?
I barely touched replicated sub objects. 😛

[nova wasp]

I'm honestly not sure if forcing an update matters for late joiners

[nocturne quail]

yeah, it forced only defaults to be replicated, any custom replicated property is ignored

seems like it needs to serialized forcefully

[nocturne quail]

yeah it not matters

I was doing it cause to test it to see if the actor will sync the late joiner in the current state of the actor

[nova wasp]

well... I guess it depends on what is interesting about the weapon to send down

[halcyon ore]

Not sure I follow?
It’s just easier to put it all in an actor for my brain.
Rather then some comp on the player, and some fake actor weapon.

[nova wasp]

oh absolutely lol, I'm definitely going to agree it is easier to understand on a human level to have it just be a thing you can click on in the editor that says it has x and a mesh

I am not sure actors can "require" other actors to show up first in normal replication

I'm kind of surprised them being attached doesn't influence it or something (assuming they are)

So if anyone knows more than me free free to chime in but I don't really use this approach so I am unfamliar

[halcyon ore]

The issue I guess is.
The server doesn’t check if the client knows of the actor.
So, at the end of the day, even if there attached, it still does a distance check, for replication.

Cuz, it influences it by using the owners relevant function.
But, of course if the client doesn’t have a valid owner, it doesn’t care for that.

[nova wasp]

apologies if I misunderstand... is the distance check happening too early and making it appear when it should not or not appear at all?

so the relevancy is getting stuck?

I wonder if you can make it use the attach parent for relevancy

[halcyon ore]

The issue is.
It does use the attach parent.
But then just becomes a distance check.
And is purely a server only thing, so the server doesn’t know if the client knows of said actor, just that it’s within the rep distance.

Hence the issue.
The weapon replication is received/ calculated first, before the player, cuz the server doesn’t actually know/ check that the client knows of the player

[nova wasp]

So what happens? does the weapon appear at world origin for a split second?

Could you not just have it hide itself until the parent shows up if you know it has a parent? Not sure

ofc if you intend to ALSO have weapons that don't actually have any parent on the server (dropped on the floor) you would need to consider that state too I guess

[halcyon ore]

I haven’t checked that honestly.
But, not sure.

The issue isn’t showing, it’s that’s the owner ref is invalid, so begin play fails.

Hence my potential thought of looping for valid.

[nova wasp]

you can keep the logic in the weapon

just have the character appearing find any weapons it can see attached to itself and call the weapons init thing

[halcyon ore]

Right, yeah.
That very well may be what I need to do.
I was just hoping to keep it a bit fully self contained.
I just like that.
Even for the super basic “player finds weapons”

Unless someone else sees this.
I’ll just end up doing that. lol

[nova wasp]

it is not self contained unfortunately

if you want you could make a simple interface or something to hide the weapon from the character

fwiw there might be a nicer setup here, I'm mostly just guessing on this working and you might find inside of the AActor beginplay it might not see attachments yet etc

a delay one frame in a loop is very sad but sometimes can avoid headaches with stuff you can't listen for

[halcyon ore]

Oh, no an interface, or reference isn’t the issue.
Just more personal design method of self contained stuff.
I don’t like having required secondary code, or multiple copies of code.

So far UE replication has been fucking my personal design method. lol

[nova wasp]

Maybe the weapon actor can override it changing what it is attached to?

to react to the owning character appearing etc

[halcyon ore]

Oh, that’s valid.
I didn’t think of that.

[nova wasp]

not to skip calling super:: here, but to just know when it happens

[halcyon ore]

I’ll need to check for a func when I get home.

Yeah.
Thankfully I don’t plan on doing other shit with the weapons.
It’ll always be equipped.

[nocturne quail]

that trick actually works to make late clients sync with the current replicated state by using the server world time

 FBlueZoneSerializedSyncData NewSyncData;
 NewSyncData.CurrentPosition = LevelCenterPoint;
 NewSyncData.CurrentScale = FVector(Scale, Scale, Scale);
 NewSyncData.TargetPosition = LevelCenterPoint;
 NewSyncData.TargetScale = FVector(Scale, Scale, Scale);
 NewSyncData.PhaseIndex = -1;
 NewSyncData.ServerStartTime = DoTimeTravelInServerWorld(); //replicated
 NewSyncData.Duration = 0.0f;
 NewSyncData.ServerShrinkProgress = 0.0f;

 ZoneSyncData = NewSyncData;
 void OnRep_ZoneSyncData();

now if the bluezone in any phase, late clients will join the current phase, duration, scale, location synced

and the best thing its now one time notify, and client simulate their state locally based on the synced data they recieve

[nova wasp]

The hard part is making sure the time is closely synched but for something that moves slowly it being slightly off might be perfectly fine. The cool thing here is the amount of data is just the start and stop

https://www.forrestthewoods.com/blog/tech_of_planetary_annihilation_chrono_cam/ Planetary Annihilation is an RTS game that takes this concept to the extreme where even unit positions and things are sent as a "curve" with a timestamp

[nocturne quail]

yeah all of 7 phases are slow , the final 3 phases are fast enough but I think it will not be an issue

if it will be, I will slow them down also 😄

[nocturne quail]

LogTemp: Warning: [CLIENT] Received zone sync data - Phase: -1
LogTemp: Warning: [CLIENT] Setup - Progress: 0.00, Duration: 0.0
LogTemp: Warning: [CLIENT] In shrink delay - starts in 0.0s
LogTemp: Warning: [CLIENT] Starting shrink
LogTemp: Warning: [CLIENT] Starting timeline - Progress: 0.00, Remaining: 0.0s
LogTemp: Warning: [SERVER] === Phase 0 Announced ===
LogTemp: Warning: [SERVER] Starts at: 11.1 (current: 6.1, delay: 5.0)
LogTemp: Warning: [CLIENT] Received zone sync data - Phase: 0
LogTemp: Warning: [CLIENT] Setup - Progress: 0.00, Duration: 120.0
LogTemp: Warning: [CLIENT] In shrink delay - starts in 5.0s
LogTemp: Warning: [SERVER] Timeline started
LogTemp: Warning: [CLIENT] Starting shrink
LogTemp: Warning: [CLIENT] Starting timeline - Progress: 0.00, Remaining: 120.0s

late client 1
LogTemp: Warning: [SERVER] Timeline Progress 0.15, Duration: 120.0
LogTemp: Warning: [CLIENT] Received zone data on join
LogTemp: Warning: [CLIENT] Setup - Progress: 0.16, Duration: 120.0
LogTemp: Warning: [SERVER] Normalized Artifect - Progress: 0.17, Duration: 120.0
LogTemp: Warning: [CLIENT] Starting timeline - Progress: 0.17, Remaining: 101.5s

one more thing I think I need is to make server just lerp from start to target without timeline

[viscid obsidian]

what do i do if i want to see logs from both the client and server? right now im only seeing logs from the client and I'm running with 2 players in viewport while playing as Listen Server.

[nova wasp]

where are you looking at the logs? is this a single process PIE?

[viscid obsidian]

nevermind the issue is with my code not unreal

[nocturne quail]

if you are asking about text formatting this how you can do

FString RoleString = HasAuthority() ? "server" : "client";
UE_LOG(LogTemp, Warning, TEXT("Fire executed on: %s"), *RoleString);

[viscid obsidian]

what settings do you use to test multiplayer code? do you run width a dedicated server?

[nocturne quail]

when you run as client it automatically creates a dedi server for you and client connects to it

than you can click late client to add more clients

[dark parcel]

which ever environment you want to support.
If you are going to use dedicated server then play as client,
for listen server, play as listen server.

If you want to support both, then you use both.

[nova wasp]

use GetDebugStringForWorld instead

that is local role only

just because you have authority does not mean it is the server (locally spawned things would have authority etc)

it is still useful to see local role but it would make more sense to consider that together with the local netmode from GetDebugStringForWorld

[nocturne quail]

yeah GetDebugStringForWorld is a perfect thing

[nova wasp]

the one downside is in some cases it might be too early for it to see the debug string but that is quite rare

[nocturne quail]

that's a good note

[viscid obsidian]

is world time used as a synced clock between client and server

even though they are not at the same time at the same point

[nova wasp]

https://vorixo.github.io/devtricks/non-destructive-synced-net-clock/

basically it is not quite close enough for some things so the alternative is to replace it or have a simple rpc back and forth to constantly update it with simple math (running ping average that deltas the local one)

it depends on what the timer is used for though because incoming replication data is always in the past (well, from it)

[halcyon ore]

BTW, there is one. 🙂
The OnRep for the attachmentreplication, given I only have issues for the client side.

[nova wasp]

neat, I hope that works

[halcyon ore]

It does.
Already tested

[boreal flicker]

I’m trying to make a sprint mechanic in my game. Sprinting costs stamina, and the way I have it right now is that it consumes stamina in the sprint phys function, and if there’s not enough stamina it exits sprint to another movement mode. However, I noticed that clients consume stamina about twice as fast as the server, and I think it's because of replaying moves. I wanna just not consume stamina if we’re replaying moves, but I don't know how to know if we’re replaying moves from the phys function. Should I not be changing variables in phys functions at all? Should I be doing that in a different function? If so which one?

[dark parcel]

@boreal flicker #multiplayer message
Maybe this will help you

[silent valley]

GetServerWorldTimeSeconds is the built in function you want. That is synced across all instances. No need for custom solution for most cases.

[nova wasp]

it won't be close enough in some case unfortunately but it will work for simple stuff that doesn't need accuracy

definitely easier to try that first and then change it if you find jittering stuff etc

[viscid obsidian]

sorry to bother yall again, I'm misunderstanding something about the CMC. I thought that on autonomous proxies, we run movement logic then build a saved move, which we then compress and serialize into NetworkMoveData, which gets sent to the server and rerun on the same movement pipeline. My confusion stems from how the server takes the NetworkMoveData and reruns the code. Is it true that when the server recieves the NetworkMoveData, it doesn't rebuild a saved move and use PrepMoveFor() to update the CMC state, but instead relies on manual setting of CMC state from NetworkMoveData?

[coarse patrol]

were u able to fix this?

[copper raft]

how to fix bugs with other player location when a car is moving? movement is set to None, collisions are disabled, attach actor to component are all snap to target, weld simulated bodies tried to enable tried to disable, nothing changed, and why works fine on driver screen? passengers screen only have this problem

[nocturne quail]

driver is server?

[copper raft]

yes, passengers too, i just divided them with a bool, if is driver seat -> posses node, if not -> switch camera input only

all this is inside run on server event

[nocturne quail]

are you attaching to socket?

[copper raft]

this inside OnRep_IsInVehicle, inside player character blueprint

[copper raft]

yes, sockets inside car skeletal mesh, i tried to attach them with a scene component location, but same problem

[nocturne quail]

try ignoring the vehicle collision channel when attach to vehicle

and disable physics simulation

[copper raft]

where to put this node? inside EnterVehicle which is running on server?

[nocturne quail]

before attaching to vehicle

[copper raft]

disabling physics of the character?

[nocturne quail]

first try on server if not work do it on multicast

[nocturne quail]

yes

[copper raft]

im not using multicast at all

[nocturne quail]

for disabling physics/collisions try with a multicast