multiplayer

[quaint roost]

What does the ping value from GetPingInMilliseconds mean? Is it affected by server performance?

As in if I get it on an empty map vs a map full of AI, will it differ?

[frail temple]

Anyone have any idea on how to replicate a split screen like It Takes Two or A Way Out? I'm currently testing a render target + 2D scene capture component but that is currently having it's own issues. Any/All split screen stuff I find seems to be directed at local play and not online play, any assistance is welcome

[dark edge]

Wait you want a split screen with with only one local player?

[frail temple]

Yup @dark edge Similar to It Takes Two & A Way Out

I've done it as a Render Scene right now, it's a squashed image. That is like the only method that I know of from previous projects. I'm not sure if there is another way to render the other persons camera view

[open wave]

is it an unreal 5.3 issue that all clients are now client 0 in prints anyone have the same issue

[late iris]

So uh, a basic Boolean isn't working for the server.
(But for the client 100 work accurately, so this on the client reads false)

In the debugging menu for both the client and the server.
The top variable reads 0, and the bottom reads 2. But specifically for the server, the Boolean reads true for some reason.

The part that stumps me is that the debugging menu reads it as those variables on both client and server perspectives, which means those variables are synced right?

[fossil spoke]

I wouldnt trust the Blueprint Debugger. You should print the values and verify they are actually what it thinks they are.

[blissful raptor]

If I send an RPC to an object on the server and MarkAsGarbage it, will it always be garbage after the RPC on the client?

at same frame

[fossil spoke]

Nothing that occurs across the network can be the same frame, you have to deal with latency.

If you send an RPC to the Server, that will take X amount of time to reach the Server.

If you then say, Destroy an Actor on the Server from that RPC

The call to Destroy the Actor takes Y time to reach the Client

X+Y

Is the time it takes to ultimately destroy the Actor on the Client

[late iris]

Interger Variables are correct, print string comes up with the variable number of 0 and 2 for the screenshot (Had it already checked with a different print string, BUT also just created two more print strings to double check.)

[fossil spoke]

Print the == as well, its likely that the debugger is giving you a false positive

[blissful raptor]

Oh, I was referring to the situation where the server is handling RPCs and MarkAsGarbage at the same time.

Both are sent from the server.

If I do an RPC(to Client) and MarkAsGarbage on the server, will it always happen in the same order on the client?

[fossil spoke]

There is no guarantee

[blissful raptor]

ok thank you

[late iris]

So, i ran the script to check, and now the debugger is giving both as 'false' and both as 'true' accurately. (it's swapped from the client being accurate to the server, which is fine, i can deal with that replication later)

But, the "less than" node, in the exact same set up.
is doing the same thing lmao, where it's being true on the client, but not on the server. (servers Print string is accurate for the variables. including the Boolean)

I'm gonna restart the engine and see if it helps lmao

[late iris]

Thanks for the debugger advice!.
(helped me find the actual issue lmao.
After a restart it helped too lmao.)

[wet nymph]

Hi, I'm facing a strange problem, when I play the project with two players the server player camera start looking to the horizon, the game has basically no logic running, It's only a pawn with a camera and two player starts on the scene

[bronze glade]

Hey everyone! I have a quick question I can't find an answer to online. When I use server travel to go to a new world with a different gamemode, my host is changing to the new playercontroller on the gamemode, but the client is keeping the playercontroller from the original gamemode. How can I fix this?

[rose pollen]

Has anyone done any performance benching marking of the built in CMC vs the GMC just for basic movement (run/jump/etc) that the CMC supports, is there a notable difference in speed/quality?

[cosmic trail]

I get this assert in replication(I think has to do with fastarrayitem) when I delete and create maybe 200 actors at once, and the deleted actors are all uproperty referenced in a fastarrayitem and the new actors are beng added to the same fastarrayitem:

DataReplication.cpp

ValidateRetirementHistory(...)
{
    ...
    checkf( Rec->OutPacketIdRange.First >= LastRange.Last, TEXT( "Invalid packet id range (First <   LastRange.Last). Object: %s" ), *GetFullNameSafe(Object) );        // Bunch merging and queuing can cause this overlap
}```

Anyone have insight into this?

[twin juniper]

What does "replicating actor" essentially mean?
Does replicating an actor essentially give it the ability to synchronize its data with the server actor?
If I set only the basic actor's replication = true, what will happen?
How does the engine use binary data (packets of bits) to change the properties of client actors?
Is this a dumb question? Should I start with basic networking theory?

[chrome bay]

Essentially, yes to all.

The networking system links the actor instances together by assigning an ID to them, so when it sends/receives data, it knows which actor instance on the local machine to apply the data to.

[twin juniper]

could you recommend books or site to me?

[empty bluff]

I have a Projectile actor which is set to replicate and during construction i set StaticMesh depending on some settings. I do this by setting a StaticMesh variable that is RepNotify.
When im on a client machine and shoot for the 1st time, I get this error. Any ideas what I can be doing wrong:

hmm. something strange is going on.. I moved the logic to a MultiCast RPC and then it was bitching over the same thing with the material

[chrome bay]

For low level stuff no not really, can't thnk of any. For general networking there is Cedric's compendium. Any resources will be pinned in this channel

[empty bluff]

Alex Forsythe did some sort of deepdive videos on the engine a year ago... could have a look on YT

[empty bluff]

then i moved the material to a Multicast too and now it doesn't get shown on the client!?!

what the heck am i up against here? the weird thing is that once it happends, the engine stalls a sec or so. Sometimes however it results in (if playing as a client), that the client is booted. Then after that, it works fine, absolutely no errors.

[empty bluff]

unless someone has experience something similar, i guess i will try and work with seperate ProjectileActors instead of one universal.

nope, that doesn't make any difference either...

[lost inlet]

small suggestion, I'm not sure why you don't replicate the least amount of data that gets you the data you want, which in this case looks like the DT row name

[empty bluff]

yeah i actually changed that after i sent the screenshot 😉

[lost inlet]

though not the biggest fan of DTs since it's not actually a tabular editor for some reason

[loud pike]

For multiplayer what is the best way to spawn a spawner ? Just place it into the world ?

[quaint knot]

how did you get that script stack what, that's cool

[empty bluff]

just checked the log

[tardy fossil]

Maybe the client receives the actor and the replicated variable at the same time making It RepNotify not called on the client

[empty bluff]

didn't do anything in particular

[tardy fossil]

Maybe on the client you need to check on beginplay too

[empty bluff]

hmm, could try

[tardy fossil]

Yeah do a print of your replicated Mesh variable on begin play, I bet it’s already replicated on spawn

[empty bluff]

you mean the static mesh variable?

[tardy fossil]

Yeah the variable that’s supposed to call that onrep function when changed

But if the actor spawns with the variable already replicated then it never “changes” from the client POV and onrep doesn’t get called

[empty bluff]

yeah, i doublechecked that too

it's set to a dummy static mesh

[scarlet nova]

why I can't send Array through Run on Server?

I thought it was possible

[empty bluff]

looks like one of your parameters are byref

RPC's are pr default one way

[scarlet nova]

hmm, do I need to copy it before sending?

or what's the potential solution

[empty bluff]

no i blieve your two parameters on that method is declared as byRef

[scarlet nova]

oh yeah, it says Pass By Reference, but can't untick it. Do you know why?

[empty bluff]

because arrays are byref always i guess hmm

[scarlet nova]

interesting, so I can't basically send arrays

aight, gonna find another way of sending them then

[empty bluff]

i feel like it's possible, but can't pin it on anything precisely

[scarlet nova]

I put the array in a struct, now it doesnt throw the error

this is the way probably

thanks for the help

[empty bluff]

ah yes that's it

thanks for reminding me

this prints 1M_Cube_Camfer which is my default/designtime box

[empty bluff]

ok, so i got rid of the errors in the projectile

actually by using two seperate Multicast calls 😦 if i try to use just one, other things fail for strange reasons

but it does not end here... when you get hit by a projectile, i transfer some debuffs to the player

and now oddly enough i get this error

[latent pelican]

Hi all im having trouble with multicast for my selected weapon ive got it to spawn properly on the client using spawn defult pawn for but that does not show up on the client any advice would be appreciated to get the weapon to show up on both

[empty bluff]

So as I read the error, the AudioThreadCheck fails deep into the engine somewhere, and it indicates that I should not call this unless im on the AudioThread?

[empty bluff]

can't read your graph, can u zoom it

when calling a multicast, you have to make sure you are on the server

either guard it with a HasAuthority check

or make a Server_DoSomething that calls Multi_DoSomething

[latent pelican]

okay so my best option would be to add has authority at the start of the weapon mesh selection

[empty bluff]

actually... you probably want to stick to RepNotifiers

if someone joins in late in game, there's no way they will know what weapon existing players are holding

[latent pelican]

thats not a problem this is going from a local character selection where everyone has to be ready first then they all spawn in at the same time at the start

im a little bit noobish when it comes to multiplayer blueprinting

[empty bluff]

better invest some time into experimenting and reading up on it... there are lots of strange pitfalls with it

[lost inlet]

it helps to install engine symbols and use the debugger

[empty bluff]

80gb though

[lost inlet]

yes, and?

[empty bluff]

Its just alot

About that error i get. Is there a way to easily execute code on thw audiothread?

[crisp shard]

is there any method to check client side if a server event fired? basically im trying to check if a client dealt the dmg that they are assuming they will hit (if hit on client), but if server event doesn't fire for some reason i'd like to have an event check on client if that dmg actually went through, i've tried a few things and im struggling to find how i can do that

i get enemy health prior to attack and then i get health after the attack but it seems like the numbers get "off" at some point if there is a missed event

enemy health is replicated (repnotify) but if that never fires, health should be the same, meaning no dmg, but client somehow doesn't get this info in time

[chrome bay]

Client sending damage to the Server is asking for trouble

The log should tell you why it wasn't sent/received

I suspect ownership

[snow trail]

hello, does anybody know how i can replicate something like this custom movement i have here?
i currently set the player actor and the mesh component to both replicate, and it works fine on server, but it definetely does not on the client
it looks like its trying to simulate locally but the server is sending it back to being still
i wonder if i should just convert my movement to the server and have it work there, or if theres a way to do it with client authority?
server sounds like the best for anticheat and stuff but then wouldnt the movement feel sloppy and delayed? (edit, tried the run on server approach and yeah movement is laggy and weird)

[lost tinsel]

how bad is it passing a UTexture2D through an RPC?

should I instead pass something else and get the reference to that or I can't worry about straight sending it

[crisp shard]

im not actually using the client dmg , just using the same dmg calulation for things that effect V/Sfx related to the enemy health. as well, im trying to just have a "miss" widget appear if the server essentially rejects the clients assumed dmg, but the server is the one doing the actual dmg calculation that would be used (if hit)

[crisp shard]

yea that might be it, im getting hte health of the enemy in a weird way at the end of the attack, essnetialy taking the local hit actor and using an interface message to get the health

the inital (which is the same function, but maybe a better way of getting the actor), is working everytime tho

[snow trail]

aaalright update on the situation
i did some more research and found out blueprints have nothing to automatically do client prediction or whatever
so running the stuff on server is really easy, but then the client feels laggy cause it has no authority and its not trying to predict the movement either

i do have smooth sync from an ancient free marketplace sale...

[dark edge]

Are you sending the texture itself or just a ref to a texture asset?

[lost tinsel]

I didn't even try tbh, I sent the player state since I needed one more information from it

[snow trail]

is this makeshift client prediction? lol

im just steering on the client and on the server too

[kindred gazelle]

Gotta be careful with stuff like this. Assuming ROS Steering sets things on the server which immediately replicates back to client, if there is too much latency you will be rubberbanding

[worldly shore]

Update on this, as I think I saved the RMS implementation:
The problem was that I was setting the dynamic RMS's force from an ability task, rather than letting the RMS itself query character rotation. The tick rate was desync'd between server and client, so the force would get updated at different times and cause divergences. Additionally, this approach was not stateful, nor did the CMC know anything about the ability task (why would it?), so net corrections would lead to using the wrong force value when replaying old moves.

Now the RMS pulls rotation data from the CMC directly in PrepareRootMotion, so on playback it's using the same state info as the CMC's physics movement. This has eliminated most of my net corrections at 200-350ms ping and 20hz server tick rate.

[lost tinsel]

well put the high latency thing

and see what happens 🙂

simulate lag for both incoming and outgoing packets and you'll see if rubberbanding is occurring or whatever you're trying to resolve

[ruby parrot]

I have a savegame object that i store everything in like health, position and name etc.
Would I,in an actor component,lets say its for health, have a "current health" variable or just point to SaveGameObject->CurrentHealth?
(tag in response please 🙂 )

[upbeat basin]

Is the way to check if we're currently in authoritive context from a UObject checking GetWorld()->GetNetMode() != NM_Standalone && GetGameInstance()->GetFirstLocalPlayerController()->HasAuthority()?

[kindred gazelle]

You probably shouldn't use the save game object for mechanics like that. Make a component instead, and when you want to save the game, have the component THEN write to the save game object.

[ruby parrot]

this way i would have the "CurrentHealth" in two separate areas,no?

[empty bluff]

anything that stands out or could explain it? I still understand it as it requires me to be on the audiothread?

hmm.. and i see there's a RPCReliableBufferOverflow for some reason at the end too?

[kindred gazelle]

I suppose, but not really. Usually you will want to have your current health be "copied" into the save game object. The save game object is basically just a perfect copy of a game, it shouldn't be used for logic outside of saving/loading

[lost inlet]

what would explain it is actually using the debugger

[empty bluff]

sorry, i've never done this before

[ruby parrot]

i guess. but i just like to keep it super clean. having the "same" variable in two places is a little meh for me 🙂 but ill keep thinking about it.

[snow trail]

it looks fine even with more than 100 simulated pktlag
only issue ive been having is when the two cars collide together they become both jittery

and also i havent managed to do the same thing with the suspension calculation
it almost seems like the suspension calculated on the server is completely different from the client, but thats just a guess

i would like to move on to a more proper system though, but we dont have 350 euro to spend for the GCM component
making my own system would probably be fine too but cant find much on the topic online

i thought about C++ however
does it handle this stuff automatically unlike blueprints? or is it the same deal?

[kindred gazelle]

I'm no cpp expert but I don't believe it does it automatically. I think the only thing that does it automatically is the movement component really

You can always build your own prediction. That's what I did. I worked from this https://ruoyusun.com/2019/09/21/game-networking-5.html

Specifically the rollback section

[snow trail]

ill look into it when im back home, many thanks!

[thin stratus]

@snow trail fwiw, you could wait for 5.4 if it release Mover 2.0

Might still require C++, but at least it's basically the CMC without the whole capsule requirement

And offical from Epic

Mover 2.0 is the "replacement" of the CharacterMovementComponent coming with 5.4

https://vorixo.github.io/devtricks/mover-show/

Because CMC will most likely continue to exist

Mover uses the Network Prediction Plugin though afaik

Which is nice for debugging

Will most likely have to work with it on the next project I join

Bit hyped to learn all the ins and outs of that system

[vague spruce]

say i have a game mode in which i have to wait until all players have connected. how do i start the server with that expected number? can i pass a parameter to the server on launch with the configuration? also how would i set this up in development? probably in here i'm guessing (SS1)?

[snow trail]

sounds interesting, do we have any info about 5.4's release date?

[grand kestrel]

Nope its not even on roadmap

[thin stratus]

You gotta count the NPP lines too :P

[thin stratus]

Guess as usual is GDC

[vital barn]

I have a bool stored in my playerstate that is IsCarryingFlag (it defaults to false)

When the player picks up an item, I set IsCarryingFlag to true (SetFlagOnServer)
When the player takes enough damage and dies, I fire from the actor (BP_FirstPersonCharacter) a function (IsCarryingFlag?) on the Player Controller (BP_FirstPersonPlayerController)
If the variable on the player state (IsCarryingFlag) == true, then it will drop the item and set IsCarryingFlag to false (SetFlagOnServer)

The playerstate in the second to last screenshot is set on begin play

For some reason, I cannot get it to read the IsCarryingFlag variable correctly, and it just defaults to false meaning nothing get dropped on the client. On the server however, it drops it correctly.
Why could this be?

[vague spruce]

it's UGameplayStatics::ParseOption

[nocturne quail]

if the base item class set to replicated true, it will apply to all of its child classes?

[dark edge]

First off, how are the cars moving and what's doing the networking for them?

[thin stratus]

Network Prediction Plugin

No

[solar stirrup]

yes

[snow trail]

the car is a pawn with basically just a static mesh and a custom component i made for the vehicle movement
the static mesh component is set to replicate
the component really just applies a bunch of forces to do everything; suspensions, acceleration, braking and steering

the way i'm handling things right now is very sketch and i definetely want to replace it (which iis where im lost), but i'm just running the simulation both locally and on the server
it does rubberband a slight bit though and i generally dont like it
i also tried implementing it for the suspension but i instead left it non replicated
i dont know why but even without running the suspension calculations on the server, they still work on the client and are replicated correctly
if i try to do otherwise, the suspension completely breaks

[dark edge]

So it's physics, and probably using the default physics replication system

What's probably happening is the sim is running on both ends, and the servers results are being replicated to the client through the default physics replication system

[snow trail]

yup, thats what's happening right now

and iit honestly mostly works

[thin stratus]

Introduce latency and you are lost

[snow trail]

i tried with 100 pktlag and iit worked

but yeah i can see why youd say that, its probably true

[thin stratus]

¯_(ツ)_/¯

[snow trail]

again, i dont like the system im using right now

i would love to move on to a more proper system, im just lost on how to actually do that

[thin stratus]

If you don't have a prediction system, then I would suggest just doing the phys locally and sending the server the result

Otherwise it's half-as'd

As soon as you involve the Server, you will get weird corrections, cause you have no proper prediction, or even reconciliation going

[snow trail]

client authority then?

wouldnt it look jittery to other clients though?

otherwise, how hard would it be to make a simple client prediction system for this kind of movement?

[dark edge]

prediction with physics is a monster of a problem

you probably have half-assed prediction right now

Basically the sim is running independently everywhere with some sort of synced inputs (your controls etc), and the built in physics replication system is keeping the client somewhat in sync with the server

That's what I use and it works great, but prediction is not really a thing like that

[snow trail]

i see

[dark edge]

My system is generally like this:

Input -> Run on Server Event

Run on Server Event -> Set replicated variables

Tick -> use replicated variables to do physics
It works pretty much perfectly, but you do have ping lag between your inputs and seeing results. It's fine for my system.

[snow trail]

hmmm

what replicated variables are we talking about? and how do you do physics with those?

is it just velocities?

i thought that was replicated already though

[vital barn]

Anyone? 😭

[open wave]

anyone else running 5.3? and successfully having the clients have their own number not just have client 0 on all clients

[sinful tree]

Can you clarify? Printing strings while running in PIE will result in whatever client is currently executing the code printing the value to display Client#: on all screens.

[open wave]

so i have started a new project i went to the map blueprint and on tick made print. this gives me client 0 on all clients

if doing this in 5.2 i get client 1 and client 0

[languid igloo]

I'm in the middle of a multicast. But I need to update the HUD, but only for the client that is possessing the character. (Instead of ALL the clients, as is currently happening. Causing a bunch of updates overwriting each other.)

Is there a check I can do before updating the HUD, that prevents it from happening on all the other clients at the same time?

[open wave]

you woudlnt do a multicast just owning client I believe

[languid igloo]

Yeah, I get that. But this function is doing a bunch of other stuff as well. Updating the GUI is just a part of it.

[sinful tree]

IsLocallyControlled into a branch

[languid igloo]

That sounds like a solution. Just gotta get a reference to the Character of which this BP is a component, and I'll give it a try.

[sinful tree]

GetOwner > Cast to Pawn > IsLocallyControlled

[languid igloo]

Really? I thought Owner was purely a replication thing?

[sinful tree]

Components are Owned by the actor they are attached to

[languid igloo]

Doesn't look like GetOwner is returning anything.

[woven bramble]

After players have been playing the game, rarely (some players) experience a bit of lag and then get kicked. How can I debug for a multiplayer game? Additionally, what do you think could be the most obvious cause of this issue?

[sinful tree]

Out of curiosity, are you getting a warning in your log that says something along the lines of "Not enough login credentails for all PIE instances"?

[sinful tree]

You said the component is attached to the character, no? The error message indicates otherwise - it looks like it's within BP_Launcher which doesn't sound like a character or a component.

[languid igloo]

It's like this. It's odd that it's not returning anything. Google says it should.

[sinful tree]

Ok, so that's a child actor component... That might not work the same way as the actor itself that is spawned by the child actor component isn't a component so calling Get Owner on that actor won't get you the reference back to the actor that holds the child actor component.

[languid igloo]

Worst case, I can just set a reference. But I'll do some googling first to see if there's a cleaner solution.

[languid igloo]

This worked. Apparently GetAttachParentActor() is the Character object that has BP_Launcher as a component. God knows what happens if you have multiple layers of components.
But the IsLocallyControlled() node worked for isolating my HUD updates to the local client only.
Thank you very much!

This works as well, so I don't need to cast. But I don't know what an instigator is, so I guess I have some learnin' to do.

[snow trail]

hello, i'm trying to increment a variable in the player state whenever the player (a car) passes through a collision box
how do i know which player state to get though?

[sinful tree]

You can't reliably use the index getters in online multiplayer. If you want the player that started the overlap, then "Other Actor" from the overlap event would likely be their controlled pawn, in which case you can cast it to pawn, and from that get the playerstate.

[snow trail]

got it, thanks!

[open wave]

where would i check this i feel like ii have seen that once

[sinful tree]

When you stop play, it should pop up in the message log.

If you don't see this when hitting stop then this has nothing to do with it 😛

[open wave]

nar no errors like that

[upbeat basin]

When I try to travel to a listen server on the internet with the connection string I get from online session (steam.[SteamID]:7777), it's timing out. But it works when devices are on the same network. Is there anything I should be doing extra in order to make this work?

[rocky kestrel]

Hello! I have question about CheatEngine Speed hack. It makes deltatime faster? How I can protect my game from it?

[twin juniper]

i will try it thank you!

[glad cloak]

Hey guys, I’m making an inventory system using a custom replicated Uobject class, if I make child blueprints of that replicated UObject, will they also replicate?

[woven bramble]

Would something like this create performance issues, do you think it's appropriate?

The inside of this Door RepNotify

[vagrant grail]

Is it possible to hide the head of my imported skeleton mesh on the client without having to create 2 identical squeleton with 1 having no head for the client, while for other have the whole boddy ? So I want to have 1 single skeleton and change the visibility of the head for the client only.

[fossil spoke]

Only the Client, you would just hide the Bone below the Head on the Skeletal Mesh

This would be done in the Blueprint where the Skeletal Mesh exists

You can use the IsLocallyControlled node to check if its the Client that needs the head hidden.

Assuming its a Character

[vagrant grail]

I'm not sure to understand, why would I hide the bone below the Head (which I assume will hide the whole body) while I need to hide the head to avoid my Camera glitching with the Head mesh on the head socket ?

And what's the node to do that please ?

[fossil spoke]

You hide the one below the head because that causes it to hide all the bones above it in the bone hierarchy.

I cant remember the function name off the top of my head

Maybe do some research on how to Hide Bones

[vagrant grail]

So the bone below my head is the Torso bone, but hiding it wouldn't make that bone alos hide the torso itself ?

[fossil spoke]

You should be hiding a Neck bone of some kind

Not the torso

[vagrant grail]

The issue is I don't have the neckbone 😬

the head is directly connected to the torso

[fossil spoke]

Ok well then just hide the head bone then

[rich crag]

Does anyone possibly know how to spawn an instanced static mesh into the foliage actor for multiplayer? I can't seem to get the replication right. It either sets the right mesh but also swaps other trees to that mesh, sets the wrong mesh, or sets an invisible mesh (that is still there because I have collision logic that still detects it).

[wintry crane]

guys do you have any good source that i can learn from about rpcs and their usage

[fast mason]

Hey guys! Just if anyone knows.. I'm still debugging the issue that the DS hangs on some thread or something without log.
Just by curiosity, I checked the dedicated server running threads and found the following:

agsSetDisplayMode & png_get_io_chunk_name calls.. no idea why a dedicated server console would call those..

agsSetDisplayMode is related to HDR and D3D device.. why on earth does the DS creates a D3D device xD

I am missing something? Something that maybe I should change on config?

nvm, its start Address + offset

[woven basin]

Check pins for the Network compendium

[wintry crane]

thx

[fast mason]

any way to detect deadlocks on DS?

i dont know really what else to do.. the DS hangs, kicks all the players, if I press Ctrl+C it displays INTERRUPT CTRL+C, but never closes, it just hangs there.

[thin stratus]

Idk if someone answered you, but I just saw the message.

BeginPlay is too early to access the PlayerState. And it should also not be accessed via GetPlayerState0

[keen hound]

is this the correct way to search for dedicated servers

[nova wasp]

running a profiler on it to see which event happens and continues unabated could catch what its hung up on

Insights, superluminal etc

[upbeat basin]

If the listen server closes the server/opens a level in standalone (I don't know the terminology here), do clients need to listen for games intance's network error and manually leave from the online session?

Well, or would it be a proper solution to use AGameSession::KickPlayer for everyone just before calling DestroySession in and override APlayerController::ClientWasKicked_Implementation to leave the session?

[blissful wind]

Hello, I am trying to spawn an actor only on the client, my Blueprint class has bReplicates = false, but as soon as the actor spawns the bReplicates value gets changed to true. I have used the workaround of setting it to false right after it spawns but not knowing why this is happening is infuriating. Any ideas why this could be happening?

Code for spawning: https://ass.drison64.eu/z6Bs7NWm4LcJ
BP being spawned (TowerPreviewClass): https://ass.drison64.eu/r9CeSKLxPaat

Thank you in advance!

[chrome bay]

If you spawn the actor client-side with bReplicates it doesn't matter, it won't be replicating to anyone

All you need to do it make sure you spawn it clientside, not server side.

[blissful wind]

Yes, that works when a client spawns it, but when the host does (listen server), the actor gets replicated to the clients.

[chrome bay]

Why bReplicates would be true IDK, perhaps something in your class modifies it after it's spawned, but so long as Replicates is off, it won't spawn on clients

[blissful wind]

Should I create a forum post for this instead? Or is it okay to send few lines (38 to be exact) of code here?

oh did not know that, good to know

[chrome bay]

tbh unless anybody else is having a convo at the same time, it doesn't matter too much

But my guess atm, is somewhere in your AGameTowerPreview class (or a parent of it) is setting bReplicates/SetReplicates after spawn

[blissful wind]

GameTowerPreview.cpp

// Fill out your copyright notice in the Description page of Project Settings.
#include "GameTowerPreview.h"
#include "VRFinalProject/Player/InGamePlayerController.h"

AGameTowerPreview::AGameTowerPreview()
{
    bReplicates = false;
    bNetLoadOnClient = false;
}

bool AGameTowerPreview::IsGrabbable_Implementation()
{
    return true;
}

void AGameTowerPreview::OnRelease_Implementation(APlayerController* HoldingPlayer, ABaseCharacter* HoldingCharacter)
{
    IGrabbable::OnRelease_Implementation(HoldingPlayer, HoldingCharacter);

    FVector ReleaseLocation = GetActorLocation();

    this->Destroy();
    
    if (!HoldingCharacter || !HoldingPlayer || !HoldingPlayer->IsLocalController() || !TowerClass) return;

    AInGamePlayerController* PC = Cast<AInGamePlayerController>(HoldingPlayer);
    ABaseCharacter* BaseCharacter = Cast<ABaseCharacter>(HoldingCharacter);
    UWorld* World = GetWorld();
    if (!PC || !BaseCharacter || !World) return;
    
    FHitResult HitResult;
    World->LineTraceSingleByChannel(HitResult, ReleaseLocation + FVector(0, 0, 200), ReleaseLocation + FVector(0, 0, -1) * 2000, ECC_Visibility);

    if (!HitResult.GetActor()) return;

    PC->SpawnGridElement(TowerClass, HitResult.ImpactPoint);
}

Construction script as well the even graph are completely empty. No nodes, no nothing.

[chrome bay]

Ok. And what's the parent of that?

Unrelated but I would probably move that this->Destroy(); call to the end of that function btw.

[blissful wind]

AGameGridElement

// Fill out your copyright notice in the Description page of Project Settings.
#include "GameGridElement.h"

// Sets default values
AGameGridElement::AGameGridElement()
{
    bReplicates = true;

    SequenceManager = CreateDefaultSubobject<UTemplateSequenceManager>(TEXT("SequenceManager"));
}

bool AGameGridElement::IsSupportedForNetworking() const
{
    return true;
}

// Called when the game starts or when spawned
void AGameGridElement::BeginPlay()
{
    Super::BeginPlay();

    if (!HasAuthority()) return;
    if (!IdleAnimation) return;
    SequenceManager->Server_PlaySequence(IdleAnimation);
}

https://ass.drison64.eu/3VN3Dt8Z2ZSZ

[chrome bay]

Ok, you don't need the IsSupportedForNetworking() override, that's always true for actors

Can't see anything obvious, so I would drop a breakpoint in UWorld::AddNetworkActor and see where/why it's being called for your actor

Also, are you certain the actor that is being spawned on clients (for the listen server) is actually a replicated actor, and it's not just some error with the where/how they're being spawned?

[blissful wind]

Will try, my class is just starting, so I wont be able to respond for a while now. Is it okay if I come back in 20 ish minutes? Also could verifying the engine files help?

[chrome bay]

Verifying engine won't do anything here, this is just a code error somewhere

[blissful wind]

Okay, thank you. Gotta go now, will be back in a bit.

[blissful wind]

Thank you, this has resolved my issue.

https://ass.drison64.eu/S47WgzVG9bWJ

Thank you again!

[chrome bay]

called it 😄

[lost inlet]

weirdly inconsistent this-> usage

[unborn drift]

Hi guys

Is Epic Online Service's Relay service is free?

[lost inlet]

all their game services are free? and there is a dedicated EOS channel here

[unborn drift]

sorry, didn't notice there were dedicated channel

[thin stratus]

@grand kestrel Yo, I just noticed something on the Move Prediction Repo

void UStaminaMovement::ClientHandleMoveResponse(const FCharacterMoveResponseDataContainer& MoveResponse)
{
    if (MoveResponse.IsCorrection())
    {
        if (!MoveResponse.bRootMotionSourceCorrection && !MoveResponse.bRootMotionMontageCorrection)
        {

That stuff. We don't need to use ClientHandleMoveResponse

Which means we also don't need to check all the additional stuff below that snippet

One can override ClientAdjustPosition_Implementation (which is called if the two booleans above are false)

And just do const FStaminaMoveResponseDataContainer& StaminaMoveResponse = static_cast<const FStaminaMoveResponseDataContainer&>(GetMoveResponseDataContainer());

And handle the correction in there

Idk anymore why we override ClientHandleMoveResponse, but my original reason, when the MoveResponse stuff was just patched in, was that there was no access to the Container further down

But one can always access that (same for the ClientData one) via the getters
Similarly, the Server calls PerformMovement for 3 different Move Datas. And every time it sets the current one via SetCurrentNetworkMoveData, which can then be freely accessed in the "scope" of that PerformMovement Call.
But that one about the Server you already know.

[lost tinsel]

so I'm experiencing a problem of rubber banding and I need some hints, the problem arises when I'm lagging and I start pressing the shift key back and forth (for sprinting), when I sprint I set the stuff locally (handling animation and speeds stuff) and then I send the RPC to the server, in order to execute the stuff there aswell, but if I'm lagging and press the shift key as fast as possible I experience a lot of rubber banding ( the problem doesn't arise when I don't do that behaviour)

should I try to work around this or not let the user spam the shift key? I mean he's lagging really bad, and it's comprehensible that it's rubber banding since it's sending those RPCs back everytime, maybe I can replicate the movement only when the user didn't spam the shift key for a x amount of time? let me know your suggestions please 🙂

[grand kestrel]

I'll have a look tomorrow in the morning, almost midnight here. Feel like doing a PR by chance? 🙂

[thin stratus]

@lost tinsel Sprinting outside of CMC C++ will always rubberband

[lost tinsel]

CMC as in Character Movement Component?

[grand kestrel]

Cedric might know a repo you can use for CMC sprinting tho 😛

[thin stratus]

Might not have the time, might be easier for you to just copy paste to the other function

[thin stratus]

Yes

[grand kestrel]

All good, I'll look tomorrow then, thanks for the info

[thin stratus]

All good, have a nice sleep!

[lost tinsel]

I'm setting the speed on that component with

Character->GetCharacterMovement()->MaxWalkSpeed = bIsSprinting ? SprintWalkSpeed : BaseWalkSpeed;

if that's what you mean

[thin stratus]

( we are 12 hours apart, damn )

[grand kestrel]

Yeah NZST is a pain

Btw can you add me as a friend, I want to DM you a link

[thin stratus]

Check this repo: https://github.com/Vaei/PredictedMovement/tree/main

Specifically the Sprint example

There is more to it than just setting the speed

[thin stratus]

You can DM me without that iirc

[lost tinsel]

okay I'll read this, thank you

[thin stratus]

I have DMs open due to being a mod

[grand kestrel]

I actually can't

[fierce fiber]

Having a little issue. I've put this in multiplayer as I thought given I'm trying to test something multiplayer using 2 screens, It might correlate to my issue. The above video I think showcases all of whats going on, and the error at the end correlates to the image attached. Im very lost and not too sure why the splashscreen widget isn't showing for the client, given that this is just the main menu loading up, which would be entirely client side. The image is from the widget you can see working on the right screen, this image just shows what the error is potentially talking about as it doesn't provide context

If anyone has any ideas I'd be grateful!

[thin stratus]

What the f

Doesn't apply to Servers I joined in the past. How does one then apply that?

Also others can contact me just fine

I can add you, but I usually only have real-life friends on Discord

[thin stratus]

Care to explain why you start the Game as ListenServer in the MainMenu?

Mostly asking so you solve that thing yourself.

Cause it's really straight forward what you are doing wrong.

[lost inlet]

why does your game need a menu to work in a multiplayer environment at all

[thin stratus]

Ba ba ba, let them answer first :D

I'm pretty sure this is just a brainfart

[lost tinsel]

sorry I'm interested in understanding aswell since I'm new, what's the matter with this?

[lost inlet]

I'm not sure what you mean, usually the menu just opens on the client before they're connected to a server

which means it can just run in standalone

no networking required

[fierce fiber]

Oh god dont say that

[lost tinsel]

oh true

[fierce fiber]

Have I been blueprinting too long again

[lost tinsel]

I see what you mean now

[lost inlet]

it's not like you're going to connect to a game server to present a few buttons and settings

[fierce fiber]

and just cant see the issue

[lost tinsel]

yea yea you're right, thanks for explaining it anyway

[lost inlet]

the issue is that there is no issue. you wouldn't be accessing the main menu as a client connected to a server anyway

[fierce fiber]

Yeah just realised actually, so I decided to play test as a client instead

Still the same problem

[lost inlet]

that is not what is being said

[lost tinsel]

you're still connecting to it as a client, read the win names

[lost inlet]

you would use standalone with a single player

[lost tinsel]

you should run it in standalone

[fierce fiber]

Ah I see

[lost inlet]

that's how every main menu is in every game, whether it's multiplayer or not

[lost tinsel]

it's plausible, you don't want your menu to be lagging if you have bad ping

[fierce fiber]

Ok, I've started to see the error of my ways, thanks guys and sorry you had to deal with my brain hahaha

[woven bramble]

Would something like this create performance issues, do you think it's appropriate?

The inside of this Door RepNotify

[lost tinsel]

so this is executing for everytime you change the value of the door angle with the timeline? and thus sending a rep notify since it's attached to the door angle (which is replicated) ?

I'm not used to blueprinting since I'm mainly c++ so correct me if im wrong

[woven bramble]

Yes, exactly. When I open the door, the rotation value is being executed via RepNotify (naturally, via the Server). I made some changes in the Config, resolved some connection issues, but on the Client side, there are some hitches when opening the doors. It seems like they're opening with lag (although there's no lag). This might be a very detailed question, but does adding the following code to the DefaultEngine.ini help reduce the workload by limiting the appearance of certain processes on the Client side? It's like saving resources.

TotalNetBandwidth=200000
MaxDynamicBandwidth=40000
MinDynamicBandwidth=20000
ClientAuthorativePosition=false
ClientErrorUpdateRateLimit=0.015f
MAXCLIENTUPDATEINTERVAL=0.35f
MaxClientForcedUpdateDuration=0.60f
ServerForcedUpdateHitchThreshold=0.150f
ServerForcedUpdateHitchCooldown=0.100f
MaxMoveDeltaTime=0.125f
ClientNetSendMoveDeltaTime=0.05
ClientNetSendMoveDeltaTimeThrottled=0.066
ClientNetSendMoveDeltaTimeStationary=0.0833
ClientNetSendMoveThrottleAtNetSpeed=10000
ClientNetSendMoveThrottleOverPlayerCount=24
bMovementTimeDiscrepancyDetection=true
bMovementTimeDiscrepancyResolution=true
MovementTimeDiscrepancyMaxTimeMargin=0.25f
MovementTimeDiscrepancyMinTimeMargin=-0.25f
MovementTimeDiscrepancyResolutionRate=1.0f
MovementTimeDiscrepancyDriftAllowance=0.10f
bMovementTimeDiscrepancyForceCorrectionsDuringResolution=true
ClientNetCamUpdateDeltaTime=0.066
BadPingThreshold=200
SeverePingThreshold=500
BadPacketLossThreshold=0.05
SeverePacketLossThreshold=0.15

[/Script/Engine.Player]
; These numbers should match TotalNetBandwidth
ConfiguredInternetSpeed=200000
ConfiguredLanSpeed=200000

[/Script/OnlineSubsystemSteam.SteamNetDriver]
NetConnectionClassName=OnlineSubsystemSteam.SteamNetConnection
bNeverApplyNetworkEmulationSettings=true
MaxClientRate=25000
MaxInternetClientRate=25000
bClampListenServerTickRate=true
NetServerMaxTickRate=30
MaxNetTickRate=30

[ConsoleVariables]
net.MaxRPCPerNetUpdate=10
net.PingExcludeFrameTime=1
net.AllowAsyncLoading=1
net.DelayUnmappedRPCs=1```

[lost tinsel]

ideally you don't want to do that, imagine sending a RPC for each degrees of angle of the door, you're sucking up the bandwidth for that cosmetic effect?

like, is that really important that you want the server-client to be synced up on the degrees of that door?

[woven bramble]

I will do all these operations as Animation, I think this is the best.

[lost tinsel]

yes but aren't you sending a rep notify for each degree?

if you're saying it looks like it's opening with "lag" it could be some rubber banding idk

[woven bramble]

Actually, it's not really important at all. I'm just curious if the above code creates any processing efficiency. Because I had some connection problems in my game, added the above code, and it seemed to solve the issue, but I couldn't understand it well, as I need to test it with a lot of people. However, after adding the code, the doors started to open with lag. Do you think these codes have a feature to save on processing load?

So here, the door is just a test tool for connection kick-out purposes for me. I'm going to remove this system.

[woven bramble]

It seems like the only person who would understand this is @meager spade

[chrome bay]

Replicating the door angle will look bad. Replicate a timestamp of when the state changed instead.

Interpolate that on clients when they get it

AGameState::ServerWorldTimeSeconds will be plenty accurate enough for this

[lost tinsel]

that's what I'm saying, if it's a cosmetic effect why would you send so many notifies and occupying the bandwidth with useless information?

[chrome bay]

Presumably a door opening isn't just cosmetic either

[lost tinsel]

yes that's why I said, it depends on how important it is

[chrome bay]

Also, in a "real" game, you'll be lucky to get one or two updates to that variable

[lost tinsel]

but either way, even if it's important, you don't want that, you can just send a RPC to the server and get some ACK back and open the door on the client aswell

[lost inlet]

game networking is usually getting the point across in as little words as possible

[chrome bay]

You sure as hell won't get a stream of them at 60fps or whatever

[woven bramble]

Yes, it seems like the most correct and effective way is to simply replace all of this with a .fbx animation.

By the way, actually my intention for asking the question was different. Previously, some processes were being executed flawlessly (like opening doors), but there was a connection and kick-out problem. I added the above code, the issue got resolved but the doors are opening with lag. This question is completely unrelated to the door. I'll change the door, of course, there's no need to bother with cosmetics.

[woven bramble]

Do the above codes save processing load in this case?

[woven bramble]

.

[ancient root]

Hey guys, so Im working on quite large in-game collaboration based project. There is a central system within the enviroment that all players interact with. This system is quite large in scope (in terms of functionality) and is responsible for managing all of its attached children.

The problem i currently face, is that in order for the RPC to run on this system, i obv need to give it an owner. However the system needs to be able to execute its internal RPC functions when called by any player not just by the owner. What would be the best way to achieve this ? Without wrapping each function in the base player or player controller classes

[chrome bay]

Not really, but even if it did, why would you want to throttle everything

The issue with replicating a float every tick isn't going to neccesarily be the performance, it's just a poor approach and looks terrible in a "real" online game

[chrome bay]

Ultimately, you can't call RPC's on something you don't own. Route the calls through something you do own, whether that be the player, a pawn, a component on the pawn/player etc.

[woven bramble]

Thanks!

[chrome bay]

This sounds like a generic "interaction system" problem tbh, the solution is to just pass the target and the required data through a generic RPC elsewhere

[ancient root]

To do that i would have to wrap it all up, I guess i could use String and UOBjects to create generic method caller with generic params within said UOBjects..but then same issue that that needs to be owned by something...What about the game state ? Thats tech owned by the server

[chrome bay]

You need to use somethng that's ultimately owned by the controller, so the controller, or a pawn etc.

[nova wasp]

a component on it can help not have one omega object

[chrome bay]

void ServerInteract(UObject* Target, const FGameplayTag InteractionType, const FInstancedStruct& InteractionData)```

[ancient root]

Yeah but again that doesn't help me. because its a cenetral system that needs to be interact with all players. So the base playercontroller or pawn etc needs wrap all the functions of said system

[chrome bay]

Unless you can abstract the system like I showed above, then yes, you have no choice

But the above is basically exactly how my interaction system works

[ancient root]

Yeah the system is absteracted. To give more context, this is digital twin, each DT has an cenetral class and then tons of children that are either part or assembly...These classes are all in C++ and are composite so they'e made up of multiple interafaces, dynamic comp etc etc

[chrome bay]

None of that really matters, all you need is a common way to call a predefined function, optionally with some params. You must know ahead of time what those functions are and be able to ID them somehow.

[nova wasp]

(a component on something OWNED by the player) I should add

[chrome bay]

To give more context:

void ServerInteract(UObject* Target, const FGameplayTag InteractionType, const FInstancedStruct& InteractionData)
{
    IInteractInterface* IFace = Cast<IInteractInterface>(Target);
    if (IFace)
    {
        UInteractionProcessor = IFace->FindInteraction(InteractionType);
        if (UInteractionProcessor)
        {
            UInteractionProcessor->Process(this, InteractionData);
        }
    }    
}```

^ essentially the basis of any multiplayer interaction system

[ancient root]

No that doesn't help, anyways we've already got a reletionship with epic, ill pop an email to speak to a dev team...I cant expect help without giving out more detail, but too much NDA's. Thank you anyways

[chrome bay]

The rules are basically the same either way, you can't call an RPC on something you don't own, and that's not bypassable

So unless you can abstract the problem into something that can be called "generically" like the above, you'll have to provide callsites on something you do own

[nova wasp]

FInstanced structs already give an insane amount of polymorphism here and things that work like this basically just use the same pattern but with bespoke serialization

[thin stratus]

In the CMC, inside a PhysCustom function, if I do:

CurrentValue += SomeValue * DeltaTime;

Shouldn't I be able to expect CurrentValue to line up between Server and Client?
I can see that Client is combining moves, so it sends one RPC per 2 moves, but the result of the Server performing 1 Move for the 2 the Client did doesn't line up.

[nova wasp]

if they are combined isn't it... now one move?

what else is combining for?

[thin stratus]

Yeah but DeltaTime would be combined, right?

[nova wasp]

I wonder if you are supposed to change that during the combine...

unsure

[thin stratus]

If each move is 0.1 apart, the Server should move with 0.2

[5.755257] [Client - Before PhysSplineTraveling] - Distance: 740.473267
[5.755257] [Client - After PhysSplineTraveling] - Distance: 737.284319

[5.768134] [Client - Before PhysSplineTraveling] - Distance: 737.284319
[5.768134] [Client - After PhysSplineTraveling] - Distance: 730.917134

[5.768134] [Server - Before PhysSplineTraveling] - Distance: 740.473267
[5.768134] [Server - After PhysSplineTraveling] - Distance: 734.106082

First Number is the Client Timestamp

Maybe this is triggered by the initial Move being combined 🤔

Movement is so difficult that I run into these issues every time I touch the CMC.
It's not like I don't understand them, it's just so annoying to remember all the edgecases

RElatively sure MovementMode changes stop combining

Let me check

Yeah they do

So the initial one, given I change MovementMode (so the Move before I switch + the one after) should not be combined

I could disable MoveCombining while traveling along a spline, but I don't see why yet

All it does it increment a double

Well or decrement in the example above. You can travel both ways

    // Combine times for both moves
    DeltaTime += OldMove->DeltaTime;

Very very strange. I'm missing something obvious

[nova wasp]

well, does it not serialize that or something?

[grizzled stirrup]

Should you destroy sessions before calling ConsoleCommand("Quit"); or does it not matter?

[thin stratus]

Doesn't matter in theory. But doing so would ensure the Session is properly cleaned up on the Server side for a leaving client

[grizzled stirrup]

Ok will do it just in case, thanks!

[thin stratus]

It would probably time out eventually

Maybe :D

So

DeltaTime doesn't seem to be used in general. It's the Timestamp property that is send along

[nova wasp]

also if you need a "client calls RPC to server for other client" thing you can just pass over their playerstate ID as part of the RPC (on their owned actor) and have it do stuff to the other player on the server, even another RPC
if that makes more sense... It's not easy to explain without examples

the player ID would not be much different than just passing over the PlayerState actor pointer though I guess. The restrictive nature of RPCs is frustrating but easier to think of if you stop needing them to just act on the actor alone

Your system being large or unwieldly is beside the point... You can just add a component to them with the RPCs for this new thing if that makes your class look nicer

[chrome bay]

yeah pretty sure server DT is calculated as the time between packets

or between timestamps rather

[thin stratus]

How do I do this properly then

If I can't use the DeltaTime

The resulting value is used to get the Location along the Spline

If it differs it will cause a correction due to wrong location

[nova wasp]

perhaps you can work backwards, where you know the result will be fowards by another dt

you increase the spline-using value while combining as you know the DT will be some consistent amount

as for how to make that close enough, I don't really know

[thin stratus]

The thing I don't get is

If I move in straight line, just in PhysWalking, it will combine the moves.

Which means the PhysWalking stuff is called the same way my code is called, with a DeltaTime that is not the same

Which will move the Character in the end

Which would result in the same difference and cause a correction

So I'm missing something

Let me check what PhysWalking does again

sigh I need Mover 2.0 in my life

[nova wasp]

Mover 2.0 is more impactful than most 5.0 features to me lol

like that is a revelation in actually making mp games

[thin stratus]

Mover 2.0 is as awesome to me as StateTree Debugging

And that is quite sad

[grizzled stirrup]

I wonder will mover 2.0 also bring opt in features to make CMC over many characters a bit more peformant

Often the bottleneck in scenes with many chars

[thin stratus]

The CMC bottleneck iirc is not the CMC

As stupid as that sounds

The bottleneck is the recursive updating of scene components

[chrome bay]

yeah, it's just moving components most of the time

[grizzled stirrup]

Oh interesting so minimizing any additional comps on your character is actually the way to go?

[thin stratus]

So my answer would be: It's maybe more performant, but not too much

Yes

[chrome bay]

Minimising comps on literally anything that moves

[thin stratus]

Even a single comp can kill your ms xD

The Ascent had so many components that were just for "Locations"

NUKE THEM ALL

[grizzled stirrup]

Damn, I have a few situational niagara comps attached so I guess they are being taken on a useless and expensive ride!

[chrome bay]

To give some idea of how CMC just isn't the issue, we were resimulating complete moves for all 100 players

[thin stratus]

Yeah don't add FX Components if you can avoid it

[grizzled stirrup]

I'll move to a create an attach only when actually in use system

Not available always and often not used

It's handy for things like ambient FX and such where many chars have it

But if it's not on everything no point adding cost

[nova wasp]

it fights this with some scoped scene move updates at least...

but yeah it's hard to stress enough how nasty scene components are, they are just insanely insanely slow

given the actual work done

[thin stratus]

Wasn't there a better approach of just calculating the relative offset instead of literally updating the components?

[grizzled stirrup]

I guess some are kind of unavoidable like health bars but trying to think of a cleverer way to do them

[nova wasp]

thankfully turning off overlaps gets through the worst of it

[grizzled stirrup]

Yeah I've had overlaps off since day one 😄

[thin stratus]

HealthBar is avoidable unless you mean literall 3D widgets

[nova wasp]

who says you need them to transform stuff? it's like 3 lines of code to transform a local transform

[thin stratus]

We NEVER use WidgetComponents

[chrome bay]

Overlaps should be off by default change my mind

[grizzled stirrup]

I'm using widgetcomponent with HealthBarWC->SetWidgetSpace(EWidgetSpace::Screen); is there a faster way? 😮

[thin stratus]

There is a lot that should be off yeah

[thin stratus]

We usually calc that ourselves

With a Layer in the UI that Actors register Widgets to

[nova wasp]

the hard part is when you want editor support, specific rendering features, collision etc

[thin stratus]

And then we calculate the World -> Screen stuff ourselves

WidgetComponents were mostly created for VR stuff

[chrome bay]

Part of the reason scene updates are slow is the amount of time it spends chasing indirections, locking the physics thread the push updates etc.

[grizzled stirrup]

Ah that makes great sense, especially if you have lots of characters with health bars that are only visible when damaged, lots of wasted perf

[chrome bay]

Scene components with no collision data update much, much faster

[thin stratus]

It also allows you to handle all your screen widgets together

And code some "Overlapping avoidance" shit

If you want to

WidgetComponent doesn't really give you that chance

[grizzled stirrup]

Yeah that's much nicer! Are there any resources online to create a similar system to get started with?

I'm using a lot of widget components for stuff

[thin stratus]

Yeah I wonder if that will ever be improved

[chrome bay]

I've got a plugin comin' for it 😬

[thin stratus]

Not sure, google I guess.

The math is on the net

[chrome bay]

Really, truly, desperately need to get around to pushing to git

[grizzled stirrup]

Am I looking for something like "UE4 multiple screen space UI without widget component" or similar? Will get searching thanks!

[thin stratus]

Your HUD Widget would have a UserWidget with a CanvasPanel I believe, not sure anymore.
That Widget is then your "Container" for World->Screen widgets.
Then you can have an array of registered Actors. Actors can register on BeginPlay, passing in themselves and a WidgetClass.
For most of it, AActor is fine, if you need more info you can add an Interface to get info from the Actor.

Then you can calc the ScreenLoc from the WorldLoc

And position it

If you use CommonUI you could have a custom layer for it I guess

But I haven't worked much with CommonUI. Someone else fights that on our team

[grizzled stirrup]

Ok that's amazing thanks for the explanation. So basically I'd spawn a sub health bar widget (for each actor in the array) in my gameplay UI that's which then on tick updates the positions of these subwidgets from the actors, still ends up being much cheaper than them all having their own WC?

I'm not using UE5 so no commonUI

[peak lintel]

Hi. I have a question. Each time, even when I start the main menu, or when I open the map, a default camera is showing a random spot from the map and only after 1 (or half a second) the widget is triggered. This happens only if I play as a client. As a standalone play, it's working.
Does anyone knows how to solve this? I'm trying to show a loading, before possing the player's camera, but that random camera is triggered first.

[chrome bay]

It'll usually be world origin

A common solution is to have your player camera manager search for a pre-placed camera in the world if you don't have a controller/pawn etc.

Since those things will take time to replicate to you

If you're doing stuff in Pawn/Controller begin play it's too late

[thin stratus]

It will be cheaper cause they aren't affected by the whole recursive update call stuff

And you'd only spawn in your register function

[peak lintel]

Thank you! I'll give it a try!

[grizzled stirrup]

Also I assume even audio components pay the nasty CMC cost too? Will get rid of those

I assume the solution is instead of having an audio component defined in the constructor with CreateDefaultSubobject, I just define a

    UPROPERTY(Transient)
    UAudioComponent* InfrequentAudioAC;

and then in the source when needed do

if (InfrequentAudioAC == nullptr)
{
  InfrequentAudioAC = UStatics::SpawnSoundAttached
}                        

and clean up when finished with

            InfrequentAudioAC->Deactivate();
            InfrequentAudioAC->DestroyComponent();
            InfrequentAudioAC = nullptr;

Then the components are only attached for the duration that they need to be

[chrome bay]

You can also just mark them as absolute

Saves constantly adding/removing them

Just leave them in-place when they finish (or detach, etc)

[grizzled stirrup]

Ohhh that's a great idea

[chrome bay]

fairly sure cascade particle systems already did this, might be wrong tho

[nova wasp]

Audio arguably needs a real-time 3d position for some things

[grizzled stirrup]

Just by doing this when not needed anymore? Audiocomponent->SetAbsolute(true, true, true);

[chrome bay]

yee

[grizzled stirrup]

The CMC stops paying the cost then?

Thats' great

Thanks

[chrome bay]

To clarify it's not CMC - it's just moving the actor

[grizzled stirrup]

Then the inverse when needed to be updated again Audiocomponent->SetAbsolute(); (as the default is false)

Ah ok!

[upbeat basin]

Is it unsafe to loop over player array, kick them all and then change map as listen server? I'm using GEngine->Browse() to change the map and it has a function call named ShutdownWorldNetDriver(). Which makes me consider that in case of a lag or dropped packet kick RPC might not be received by clients?

[chrome bay]

Recommend not doing that for any primitive components though, since that would enlarge the actor bounds

[grizzled stirrup]

What counts as a prim component in general? I mainly have niagara, audio and widget comps

[chrome bay]

anything inheriting UPrimitiveComponent

aka has render/physics capability

[grizzled stirrup]

Ok thanks, in this case it would seem Niagara would fall into that category so maybe best to update them the Deactivate / Destroy way

Also I'm assuming UActorComponents don't pay any cost when moving the actor right?

Since they don't have a transform

[grizzled stirrup]

Ideally would only set aboslute when the sound has finished playing

And the component is idle, instead of paying the cost to constantly update position when not needed

[thin stratus]

I think I know my mistake

That's the second time I fell for it

I need a checklist somewhere

[nova wasp]

A sort of on-demand child component update

[thin stratus]

I remember when someone was feeding the actual location of the player to the audio WWise system

And the Camera was the one that defined the "ear" location

💀 💀 💀 those two differ slightly when walking left/right

STEREO FLIP FLOP

Yop

Fixed my issue

So, this is the second time that I forgot that I have to include data that is changed by DeltaTime in the SavedMove, even if it's not being send to the Server

Because the Client executes moves twice when combining

It performs Move X, then tries to perform Move X + 1, tries to combine, combines, and then performs it, passively performing Move X twice

Result is that the TravelDistance gets modify by DeltaTime of X and then by DeltaTIme of X and X + 1 combined

So one gotta save the "Start" value of whatever is modified (here travel distance) and then reset it inside of CombineWith

void FSavedMove_Character_JJC::CombineWith(const FSavedMove_Character* OldMove, ACharacter* InCharacter, APlayerController* PC, const FVector& OldStartLocation)
{
    Super::CombineWith(OldMove, InCharacter, PC, OldStartLocation);

    const FSavedMove_Character_JJC* JJCOldMove = static_cast<const FSavedMove_Character_JJC*>(OldMove);

    UJJCCharacterMovementComponent* MovementComponent = Cast<AJJCCharacter>(InCharacter)->GetJJCMovementComponent();
    MovementComponent->SetCurrentSplineTravelDistance(JJCOldMove->StartSplineTravelDistance);
}

That way, the combined move will work with the value that was used BEFORE Move X

Only applying Move X once

Now it is smooth as butter

I hate Multiplayer Movement

[grizzled stirrup]

😄

Well good that you found the solution in the end

[thin stratus]

The sad thing is, I'm the one who posted about it last time, that all lead to the GitHub repo to be created, which does the same thing for Stamina

Cause Stamina is, even if in concept different, the same case

But it's so difficult to keep that checklist of stuff one has to do for specific movement features in mind

[grizzled stirrup]

😄 I have come across many old posts asking for help BY ME and had forgotten the answer, I can only imagine how much it happens to you

[thin stratus]

Too often

The forums and answerhub are littered with my stupid face

[grizzled stirrup]

Would a projectile actor with a projectile movement component have the same rules here? Avoid any extra components than necessary? I think one niagara component is essential for visuals but can probably remove any others

And when looking to profile this is the best way to do stat StartFile and spawn like 200 projectiles and look for Updatecomponent or something in the profiled file?

Or is there a better way to compare before / after gains?

[thin stratus]

Any Actor

Doesn't matter which one

Testing it can be done by spawning a lot, that is correct.

[grizzled stirrup]

And best place to compare before / after would be using stat StartFile and looking for things in tick that relate to updatecomponent? Thank you so much learned a lot today 😄

[thin stratus]

Best place is to use UnrealInsights

[grizzled stirrup]

Ok thanks, not sure how fleshed out it is in 4.27 but I think they had the CPU stuff in first

It makes sense that widgetcomponents were mainly made for VR, they are usually blurry and weird in 3D space at certain scales, makes way more sense to do that stuff in 2D space also. Can't find any tutorials for similar setups (they all go for the basic add widget comp to actor) so far but I think there's enough info here to get that system away from components

Nvm Tom Looman saves the day: https://courses.tomlooman.com/courses/1320807/lectures/32515407

His method is slightly different to the one mentioned before but seems to do the same thing, he's spawning a 2D widget in each character (but not a widget component) and then that widget manually updates position on tick. Assuming this is still cheaper than having a widget component

[chrome bay]

game code managing hud sigh

[thin stratus]

o.o the HealthBar kills the AI

Ah no

Where do you see the HUD doing that

[tardy fossil]

tbh it seems like reinventing the wheel, is it really that much cheaper than a widgetcomponent set to screen space?

[grizzled stirrup]

Yes considering the hefty cost that updating any scene comp appears to add

Instead updating a widget in 2D space is much cheaper

[chrome bay]

when done properly you can separate game and UI logic, which is the real goal tbh

[grizzled stirrup]

Ah Tom is great I wouldn't say he falls into the bad rep

[chrome bay]

but yes it does improve perf, again when done nicely

[grizzled stirrup]

Comparing him tothe GameDev tv courses it's at least 100x better but yes

But he is one who can!

He's got plenty of great actual cool examples and gameplay stuff made

I hadn't seen those, surely not in any of his recent stuff?

[lost inlet]

a tick bad perpetuator, dang

[grizzled stirrup]

Right but it's rare that a comprehensive course is both cool AND generally high quality, the gamedevTV ones are them bumbling around the engine for the first time

Very hard to find a course that is generally very good

[rich crag]

many people don't realize, nor do the people making the tutorials tell them, that what they are learning are prototyping skills, not game making skills.

[grizzled stirrup]

I haven't done this latest course but the original Udemy one he did was very good, can't remember any hacks or tricks and it was all thought out and well designed but I may have been too nooby to notice anything else

But aside from him I don't think there are any other "good" courses at all

That actually go over creating something more substantial than a one off feature

[rich crag]

Speaking of multiplayer...

Can anyone tell me why I can RPC directly from my character to a foliage static mesh component and successfully run a replicated UpdateInstanceTransform, but I can't chain the RPC through another actor, I cannot have the gamemode do it because the multicast update doesnt run, I can't run the logic inside the component and have it properly replicate... I'm going nuts.

[lost inlet]

game modes aren't replicated, so.... yeah. client->server RPCs only work if the player owns that actor (or the owning actor if in a component)

[tardy fossil]

clients can only call RPC's on things they are the owner of

[lost inlet]

might help to show what you're trying to do

[rich crag]

I am trying to replicate an updateinstancetransform. Currently what i have is the painted foliage that is native to unreal, I created a static mesh component and assigned it to the trees. When my axe detects a static mesh instance, it gets the instance, the mesh and the transform, spawns an interactable tree and runs UpdateInstanceTransform to shrink the mesh instance to 0 scale. Everything works to this point. I'll add my hacky blueprints in a moment.

Ugh, blueprints are a mess from trying a million things lol

[rich crag]

"server RPCs only work if the player owns that actor" Right, so the RPC is on the character including the multicast that spawns the tree actor (on the server), then the server runs the multicast, also on the character, that runs updateinstancetransform. I want to get the logic off of the character for modularity. i would like to have a generic SetResourceTransform that takes an input of a transform, and an instance index and just updates the transform, but replicated.

[lost inlet]

wait why are you saying multicast here?

[rich crag]

the updatetransform doesnt require multicast?

[lost inlet]

it's just easier if you show what you're doing

[rich crag]

I will restore a version before the mess and give screenshots

[grizzled stirrup]

Interestingly with 200 projectiles being spawned with 1 Niagara component and 1 audio component, the audio component is only taking a tiny fraction of the cost to update transform (0.001ms for the audio components compared to 0.1ms for the empty niagara components) so there may be some audio component specific optimizations going on there. Honestly is very fun to profile this stuff

[upbeat basin]

Is there a way to check if an actor is destroyed due to relevancy? EndPlay reason only says Destroyed

[thin stratus]

iirc he made one of the earlier VR Templates for UE, or?

[true stream]

So I just found out that if I dont check reliable on my multicast the clients wont receive it, is this normal?

when I read about it I thought it was like, in case you really need it, check that.

but not that is required

[thin stratus]

I can't forget the code where the "Teleport Arc" was handled. There was a comment saying something along the lines of "Idk why I have to do Num - 2 for the end index in this loop." while doing "GetXYZ(Current)" and "GetXYZ(Current + 1)" a node later.

[tardy fossil]

you could always make a replicated TArray of FVector positions for the instanced static mesh instead of using RPC's

[thin stratus]

It's not normal, but it CAN happen, once in a while

If you don't get ANY non-reliable RPC anymore, then the chance is high that you are perform reliable RPCs on tick

[true stream]

in my case all I have is an actor spawned by the server that has a multicast event with a print, if I dont check reliable clients never get it.

tested it like.. 7 times.

[grizzled stirrup]

Also raise your bandwidth limits from the defaults if you haven't already @true stream . At least in UE4 they are set very low

[rich crag]

Can you point me toward a resource that explains this concept?

[thin stratus]

Pretty sure the Server gives 0 f*cks about the Client having spawned the Actor when the RPC is unreliable

[true stream]

for 1 message with nothing?

[grizzled stirrup]

You may be flooding your bandwidth elsewhere

[true stream]

the server spawns it.

[grizzled stirrup]

Like character movement

[thin stratus]

Yeah

[true stream]

I have no characters its an RTS.

of sorts

[thin stratus]

The RPC is not reaching the CLient cause the Client doesn't have the Actor yet

As strange as it sounds

If you need a callback for Actor being spawned, use BeginPlay

[true stream]

nani

[thin stratus]

Any replicated Data set before spawning the Actor (DeferredSpawn or ExposeOnSpawn) will be available on BeginPlay of the Client (+- if such data is another Actor that isn't on the Client yet)

[true stream]

right but my actor is already there from the start.

or Im not understanding it

[thin stratus]

You said your Server spawns it

[true stream]

yeh

[thin stratus]

Which means it's not there from the start

[true stream]

I feel like I would need an ELI5 here.

[thin stratus]

I'm relatively sure that the Server doesn't care about the fact that the Actor just got spawned when you use unreliable.
It will just not be able to route the RPC

The reliable one is probably being resent

[true stream]

you mean this? this is how I spawn the actor that calls the multicast later on

[thin stratus]

A replicated Actor, being spawned by the Server, takes time to be replicated to the Client.
The Client has several reason for not having the Actor in general, such as Relevancy.

If the Server tries to RPC in an Actor that doesn't have a replicated counterpart yet, it will just fail.

Reliable only fixes this cause it will retry

[rich crag]

OHHH Ive done the array of vector/ transform. That's not the issue. The issue is replicating the visual change of the transform. From anywhere but the character.

[thin stratus]

But in general an RPC on a freshly spawned Actor is usually redundant

[grizzled stirrup]

Have it done on a sever owned manager or something similar

[true stream]

Right, but if I understand correctly you are saying that the actor is not yet replicated, but it is, I see it on both screens. @thin stratus

[ornate crescent]

Hi guys, see that Throne and liberty are made with Unreal Engine 4.
I think is a custom dedicated server, instead of use UE dediated servers

any know if any have test with a multiplayer game with small quantity of replication to know how many play3ersw are online

[rich crag]

Exactly what conclusion I came to, but for some reason it sets everything, the clients and servers all report success, but there is no change on the client visually.

[thin stratus]

• If the RPC has no values with it, BeginPlay is enough.
• If the RPC sends values with it that are defining State, they should be done via Replicated variables, so having Replicated (OnRep fwiw) variables, setting them before Spawning (Deferred/ExposeOnSpawn) and using them on BeginPlay is adviced
• If the RPC sends values that don't define a state, I would suggest using an Actor that is already knowingly available, but at that point it usually makes sense to mark it reliable. Just shouldn't happen often that you spawn an Actor that needs non-state data reliably.

[thin stratus]

Send the RPC delayed by 1-2 seconds with a delay node

See if it works unreliable

Again if no unreliable RPCs reach Clients, then you are probably doing something reliable on tick somewhere

[tardy fossil]

tbh i dont think you even need deferred spawn for a replicated variable to show up in beginplay as long as its set right after creation.. unless you want that variable to show up in beginplay on the server side too

[true stream]

I probably am.

[thin stratus]

Could be true yeah

I do a lot of deferred spawning lately for other reasons. might misremember

[tardy fossil]

yeah id think in 99% of cases you'd want the server and client to match on beginplay

[lost tinsel]

hey @thin stratus sorry for bothering you, the topic you sent to me earlier sounded too overwhelming since I'm still a beginner, but I kept it in my fav so I can check it out when I have more experience.

For now the player is not able to spam the shift button and so experiencing the rubber banding, so there is a small timer that prevents the spamming and the result is fine, for no reason a player should be spamming the shift button each second

[rich crag]

OK, here is my trace, checks for mesh and if its a static mesh instace, runs the variable assignment and then gets the hit component on the server and spawns an interactable tree. It then multicasts to remove the instance. All works. The tree actor, runs a timer and if it hasn't been damaged in a period of time, it attempts to reverse the process with the same RPC's that the Character used, but no good. I'm aware the logic is messy, and tidying things up is what has things so weird. Several different times and several different configurations I have had the server and all clients tell me that everything works, but there is no visual reversal, so there is something wrong with my replication.

I really want to move the logic to a resource manager but, with experimentation, I cannot make the resource manager make anything happen that properly replicates. Clearly a knowledge gap

[fast mason]

its my next step, thanks!!

[rich crag]

OK, so I created a resource manager actor to respawn the static meshes. Here is the logic. The problem i have is that while it does add a replicated instance back into the place that the tree actor was, it also swaps a random number of other trees to the same mesh.... what am i doing wrong?

[crisp shard]

If I want to store a dynamic material instance to use for effects, how should I replicate this? Create on server and replicate back to clients via RPC? Or make the variable/property replicated and get it on clients that way?

[deft onyx]

Hi, does anyone know why so many co-op top down shooters, e.g. helldivers or alien swarm, limit the client players movement to the host's screen? And what kind of network method is used for consoles in this case?

[jade ibex]

alien swarm doesnt limit to the host screen

not sure how console would be any different but you just establish who the host is and focus the local player camera to their location

[jade ibex]

im not familiar with dynamic materials, but if its just an effect that does not contribute to any gameplay desync possibilities (for most cases, they are cosmetic so they should not be causing desync) you can just trigger it on client

from an event triggered by the server replicated to the clients

[crisp shard]

yea it works great on client so far, but im either getting it on client or server depending on how im setting the dynamic material instance. which right now, is on event begin play

[jade ibex]

ok... so no problems?

[sweet sage]

are Compressed Flags replicated to simulated proxies on unreal?

[jade ibex]

also does anyone know how the Allow Join In Progress bool works on a session? Like what makes a session considered to be in-progress? does the host gamestate have to run start game or something? (it does)

[crisp shard]

lol no im saying that when im setting it , i have tried setting on server on event begin play and also just locally and both results in a different outcome. the local obviously allows me to get it locally just fine but when i try to replicate that it's invalid on the server. so then i tried to put it on the server and replicate that back to the clients but the client doesn't get it then for some reason it's invlaid. i might need to create a dynamic material instance upon attack instead of event begin play? all characters will have to have a ref to the material instance

[jade ibex]

i see. why do you need to pass the material around anyway? cant it be stored on both and triggered by an event - going from server to client

[crisp shard]

that's what i figured as well, it's really just not working like other variables would from what i've tried

[jade ibex]

Im not familiar with the technicals but replication isnt meant to be passing around assets like that - it doesnt need to anyway

every game client will have it all

so you just need to trigger it

and host will assume that the material is the same on both itself and the clients

[thin stratus]

Possibly to not have them walk too far away and cause lots of stuff to load that the Host has to replicate

Not sure why though. Pretty sure we had not restrictions like that on The Ascent

@crisp shard You can't replicate a dynamic material

You replicate the values that you are setting on it

Everyone would have their own local version of it

Treat it like UI :P

[crisp shard]

that makes me feel better lol. so would need i to "create dynamic material instance" on the multicast of that effect ? instead of passing it through

[thin stratus]

Does the DynMat only exist when the multicast hits?

What is this for?

I'm sure you can create the DynMat already when whatever uses it comes into life or?

Also modifying a DynamicMaterial, in most cases, is a State change of said material

And should be done via OnRep and not RPCs

Which adds to the "Create the DynMat when the owner gets created."

[sweet sage]

@crisp shard is that what u are doing? the yellow effect?

[crisp shard]

it's for a sprite flash fx, im setting params on the sprite material when characters get hit

[crisp shard]

very simialar

[thin stratus]

Then the DynMat can already exist when the Character gets created or not?

You should be able to just create the DynMat on the begin play

And then flash it quickly via the RPC

[sweet sage]

I do create the dynamic material a single time and save as local variable on hit

[sweet sage]

It will increase draw calls a lot

[crisp shard]

yea this is exactly what im doing, but it's invalid on server

[crisp shard]

yea i thought to maybe do this

[thin stratus]

We never had any issues with that. One can optimize later.

Either way, RPC here is fine I guess, cause it's no lasting state change

[crisp shard]

do you destroy or forget that material after that or just create a new one and set it each hit, leaving the last one in there

[thin stratus]

Server not having the DynMat is a bug in your code I guess

I don't see why creating it on BeginPlay would cause that

[sweet sage]

Server don`t need that, do it?

[thin stratus]

ListenServer?

[crisp shard]

maybe i am doing something wrong somewhere, i would like it to become a rep notify but the issue is the health of the charcter is in a struct, and i dont want to set that whole struct to repnotify

[sweet sage]

this is my blueprint on-hit

ApplyDamageOverlay set this as overlay on every mesh

[thin stratus]

Can also hook it up with a Timeline fwiw

[crisp shard]

yea i was going to do this but thought it was more cleant o do on event begin play but yea

[crisp shard]

that's what i have currenlty

[thin stratus]

fwiw, I can't say creating the DynMat on BeginPlay and keeping it alive would be any worse or better

[crisp shard]

\

[thin stratus]

We shipped large titles with that, never had issues with it

[sweet sage]

It is working for me, i have no problem also

[thin stratus]

But it doesn't matter which way you create it

For it to work that is

[thin stratus]

And where are you creating the ref?

[crisp shard]

event begin play

[thin stratus]

Show it please

[open wave]

what would you use as is valid or for if a player has been here no other player can.

[thin stratus]

Sorry, can you rephrase that and explain your actual problem?

[thin stratus]

💀 Why are you getting the Controller on BeginPlay

And then not even guarded by anything

Yeah it's a bug in your code buddy

[open wave]

so if i am online i tame creature. with player 1 but if i go to that same creature i want to not be able to tame as well becasue its already owned

[thin stratus]

Please move the CreateDynMat away from any IsValid checks @crisp shard

And also move that Input code out of BeginPlay

OnControllerChangedEvent -> IsLocallyController -> INPUT STUFF

Please do that

[crisp shard]

hahah this is why i love sharing code in here

ty

[crisp shard]

will do immeditally

[thin stratus]

Stop using BeginPlay in Pawns/Characters in Multiplayer to access stuff that is only valid if the Character is actually possessed

[crisp shard]

legit had that for ages, never changed it

[thin stratus]

💀

[crisp shard]

makes perfect sense

and also makes sense why the mat wasn't working

[thin stratus]

The Subsystem is probably invalid for the Server

[open wave]

🙂 nice i will make sure i dont do that tooo ahha

[thin stratus]

Cause the Controller that is most likely valid for the Server (maybe) can't get the local subsystem of another player

So that IsValid fails

[crisp shard]

?

[thin stratus]

Assuming your Taming code is properly running on the Server, you can always save the Player who tamed the Creature

Just a simple Character Reference fwiw

If that is already valid when you want to tame it, it fails

[thin stratus]

Yes

That runs on Server and Owning Client

So make sure to guard it with IsLocallyControlled

[open wave]

well you could be assuming wrong :S hahah ill double check this

[thin stratus]

And fwiw check if NewController is valid

[crisp shard]

do i do anything w the old controller?

would that be the controller coming from my main menu level for example*?

that would make snese

[dark parcel]

U got the feet recovery working?

[sweet sage]

Yes

[dark parcel]

Does it work in multiplayer? what did u do?

I'm still trying to figure out why my stop anim don't work Client to Client

[sweet sage]

It does works

on multiplayer

It is based on "AnimNotifyState_DisableRootMotion"

But with more code

[dark parcel]

ty

wait, so u just play an anim when you an anim notify state gets called?

[sweet sage]

I disabled root motion at a certain point on ability if user requested to mvoe

[dark parcel]

So it's a montage?

[sweet sage]

It is a montage, yeah

[dark parcel]

Ahh ok, ty2

I'm trying to get it to work with just anim but I might have to try montage

[crisp shard]

yo, im attempting to change this up but not sure what exactly is happening. it gives me the old controller and new controller, would old controller be like a controller coming from a main menu level? and is new controller what i would set my inputs to?

[thin stratus]

No but if you unpossess it might be the controller that previously possessed it (:

But you can ignore that one

Check if the NewController is valid. Then check if the Character is locally controller

Then do your Input stuff

The MainMenu controller is long dead. Long live the gameplay one.

[crisp shard]

lolol love it, will update in a sec

[crisp shard]

??

ok that works great

unless something i don't know happens later haha

but i like this method much better i'd just like to understand it slightly better. is this firing the second any input comes in from the character?

[thin stratus]

No, it fires when the Controller Variable changes

And you need a valid controller to check if the character is locally controlled

And you want to check that cause only the local player of a given pawn (the owner that controls it) should set up input

Otherwise every player tries to do this for every pawn/character

A freshly spawned character is unpossessed. And even if it instantly gets possessed it will replicate to the client and call BeginPlay before the possess stuff happens

So high chance the code would simply never work

[crisp shard]

the old code that was there or this new one

[dark parcel]

Would appreciate any input before I have to fall back to other methods

[sweet sage]

Use distance matching

[dark parcel]

I will read about that, ty

[rich crag]

Apparently i just don't understand the way the instanced foliage actors and mesh components work. If i try to remove a single instance of a mesh and respawn it with RemoveInstance / AddInstance, the remove works as expected, but the add instance adds an instance and at the same time swaps all the alike meshes, as seen here:
https://www.youtube.com/watch?v=iGqlQHz5Huk&ab_channel=randopixel

[solar stirrup]

Can confirm that UObject inventories that contains a fast array of items w/ FInstancedStruct

Has to be the hottest way of doing an inventory and item system in UE5

It just works, UObject inventories to move inventories around, item instances that hold a map of FInstancedStruct so that all your items use the same basic struct for instances, but can hold different data

I use a TMap<FName, FInstancedStruct>: wanna know if your item is stackable? Check if its data map contains your stackable FName. No looking up data assets, 100% data oriented solution

Lightweight as fuck too

[thin stratus]

From what I heard FInstancedStruct is not the best thing to replicate though. At least for inventories

[solar stirrup]

How so

[thin stratus]

@chrome bay would need to explain that

Also, if you already have UObjects why not just use those for the items

Gives you inheritance etc too

[solar stirrup]

Lots and lots of item instances at a time, went from structs for that reason

UObject for inventories = no copying of item arrays when transferring an inventory between actors

Inheritance is done through item data assets

item data asset has a CreateItemInstance() function, each deriving item asset can add more data to the created item instance struct :3

[thin stratus]

I mean stick with your setup. That's all fine. I just got told FInstancedStruct for this is not a good idea

But I can't explain why

I didn't code that type

[solar stirrup]

Yeah I'd love to know more

If it's fixable I'd rather do that sooner than later

[sweet sage]

are lots of UObjects slow for a mmorpg backend? where each item is a UObject?

[thin stratus]

Highly doubt. UObjects are pretty lightweight

[sweet sage]

what about garbage collection?

[solar stirrup]

Mostly a pain to deal with outers etc for replicated subobjects

[twin pendant]

Were you ever able to fix that issue? Or anyone else? 😄

[solar stirrup]

whereas here the only outer i have to handle is the inventory's

[thin stratus]

Replicated Subobjects are pretty straight forward though

[dark parcel]

I applied the bandaid reddit suggest

[thin stratus]

Not sure what you struggled with

[sweet sage]

I`m using UObject because some items can have OnHit callbacks, OnConsume callbacks

[dark parcel]

But i have to disable the bandaid when playing montage or I get all kinds of error

[solar stirrup]

Nah replicated subobjects are fine - until you play with outer changes

[dark parcel]

So my montage still look somewhat jittering for the listen server but the rest of anim are fine

[thin stratus]

#multiplayer message

[solar stirrup]

since outers don't replicate heh

[twin pendant]

eh which one, would you have a link for me eventually?

[twin pendant]

damn, why is something that important so hard to fix

[solar stirrup]

My item instances all replicate atomically anyway by design

I'm not worrying about it

they're pretty lightweight

[dark parcel]

Easiest fix is to go dedicated imo

[thin stratus]

I mean he added those FInstancedStruct. I trust him when he says they are shite for Inventories. ¯_(ツ)_/¯

[solar stirrup]

oh yeah you definitely have to be careful

[thin stratus]

Also I still don't see the need. For me a UObject is all I need

[solar stirrup]

but I didn't mind about the atomicity because I didn't want desync between props

[thin stratus]

struct FInventorySlot : public FFastArraySerializerItem
{
    UDataAsset* ItemDefinition;
    uint32 StackCount;
    UObject* ItemInstance;
};

^ that's James' InventorySlot

And that's how I would probably also always design it

But that's a choice of course

[dark parcel]

@twin pendant https://www.google.com/url?sa=t&source=web&rct=j&opi=89978449&url=https://www.reddit.com/r/unrealengine/comments/a0n455/listen_server_host_sees_clients_jittering_fix/&ved=2ahUKEwjCl-KOxpyEAxWRbWwGHY4JCHgQjjh6BAgiEAE&usg=AOvVaw0k9gnyV5YseHlQgTqj_R3C

[thin stratus]

+- some changes based on requirements of course

[solar stirrup]

yeeee

for me not all items are stackable

so i save on bandwidth by having stack data be part of the instanced structs map

[twin pendant]

ty! reading through the post right now. what solution have you applied there?

[solar stirrup]

• makes it easy to check if an item is stackable (stack data in data map? stackable, else nope)

don't have to lookup the data asset or anything else, it can all be inferred from just the struct data

[thin stratus]

But you still replicate that data?

Stack Info would be in the ItemDefinition

Which is mostly a Guid

[solar stirrup]

It doesn't exist if the item isn't stackable

so there's less data to replicate

[thin stratus]

You mean the StackCount?

[solar stirrup]

yep

that's just for stacks but it applies to anything

[thin stratus]

Just.. don't serialize it if the ItemDefinition says it's not stackable?

Maybe I don't see the problem

[solar stirrup]

the item definition asset has the function to create an item instance

if the item definition asset reads that the item doesn't stack, it just wont add stack data to the data map

if it does, it will

[dark parcel]

@twin pendant w.e independent bee said is what I applied

[thin stratus]

Well, that's also all besides the point. I was mainly trying to ensure yo uare aware that the FInstancedStruct migth be a bad idea

[solar stirrup]

I will defo keep it in mind

So far no issues

I'm a bandwidth fanatic, I already replicate as little as I can and infer what I can

[thin stratus]

How many players will the game have?

In one Session of course

Given that Iris and RepGraph are somewhat locked to ~100

I assume something similar?

[solar stirrup]

Yeah 100 max for my project

Maybe one day I'll switch to Iris

if it's stable

it'll probs help

[thin stratus]

Right, yeah then it's probably a good idea to keep teh footprint small

If you would have said 5-10 I would have probably said it's barely worth it

[solar stirrup]

yeeee

also my item instances are separate from the FAS struct

[thin stratus]

The amount of sh*t we replicated in The Ascent for 4 players didn't even remotely reach any bandwidth limits

And that's a looter shooter

With hords of enemies

[solar stirrup]

so like the inventory fast array serializer item holds item instance data and whatever is relevant to inventories like position

[thin stratus]

It's quite surprising how much one can get away with now-a-days

[solar stirrup]

yeah, not like before where you'd crash your player's router if you were not careful

[thin stratus]

I'm more worried about ms

Even RAM is mostly a joke with newer consoles

Wasn't fun work for the last gen of consoles though..

[solar stirrup]

Haven't worked on consoles yet

Work hasn't given me a devkit for Path of Titans cause i don't handle console stuff on the project

good cause i barely have space in my room

I'm happy staying on sane platforms

[twin pendant]

Thanks a lot!

[solar stirrup]

Additional optimisations I was thinking about for my project's items are to strip the FPrimaryAssetId and change it for a simple FName

No need to replicate the primary asset type it never changes

and maybe switch the TMap<FName, FInstancedStruct> for TMap<uint8, FInstancedStruct>

[thin stratus]

Yeah the Type is probably redundant. Only the unique per asset part would make sense.

I usually create GameplayTags for those that I pull the FName from. Not sure that's good for highly optimized replication of them though

Does however make it very organized 😅

[solar stirrup]

I wish Epic added a way for us to replicate FNames as their index like engine ones

kinda sucks there's no way for us to add to theirs

I guess I could modify the engine but ehhh

[thin stratus]

I think they are sent as FStrings right?

Quite yucky

Would mean my GameplayTag doesn't hurt though

At least not more

[nova wasp]

gameplay tags

[thin stratus]

Are they not just handled as FNames and thus share the same ?

[solar stirrup]

yeah but it adds more work

now i gotta create a data asset + a new gameplay tag, eh

[crisp shard]

i've got everything working great now, just setup a new variable that essetnailly just equals the player health and on rep'd that and created a dynamic material instance off of all the sprites

[nova wasp]

if one macro is more work I guess

[solar stirrup]

also prone to gameplay tag load order issues

if i wanna support modding in the future

[nova wasp]

true enough

[solar stirrup]

that's why i skipped gameplay tags

[crisp shard]

for the record, the creating of the dynamic material instance each hit seemed to be the thing that solved it

[thin stratus]

So, FGameplayTag replicates like a normal FName by default or?

As far as I can see, the boolean controlling that is false by default

[thin stratus]

It would have solved it if you create the Dynmat properly on BeginPlay too

Instead of behind some random IsValid check :D

[crisp shard]

lmfao yea that def was a solid find in the midst of this

[thin stratus]

UGameplayTagsManager& TagManager = UGameplayTagsManager::Get();

if (TagManager.ShouldUseFastReplication())

[crisp shard]

but i did do that and it was successfully being created on event begin play and i still could not get the material to pass throug

not sure why

but the sprite passes through fine

[thin stratus]

Pass through?

[crisp shard]

then i make it from that

[thin stratus]

You can't replicate the DynMat

In case you tried that yet again

[crisp shard]

how would i have gotten that fx to show for other clients observing? they weren't getting the instance i had to create it to fix ti

even with it being set on begin play

[thin stratus]

Eh

If you create it on BeginPlay

Everyone would have that locally

There is no need to send it via an RPC

It's completely redundant and wouldn't work

The only thing you have to replicate is the value that you change

Or fwiw the event that triggers the change

[crisp shard]

yea i mean i really feel that's what i was doing, just getting the material that i had stored from event begin play, but on the other clients it was always invalid

this was on a multicast event, and was not firing for the owning client , that was being done seperately but i changed it all to happen on a repnotify

but when i first did the repnotify, i still did not see the flash fx , and was getting invalid on the material

but when sending just the sprite, it worked

same logic/code

then i just created the instance off of that sprite instead of getting it from the begin play variable

[solar stirrup]

i think so yeah

but ticking that bool is free bandwidth if you don't do weird load order things

[thin stratus]

Well, I'm sane enough to not develop Multiplayer Games for myself

I'm not gonna have to tick that box

Maybe on client projects

[solar stirrup]

guess im insane

[thin stratus]

Most certainly

[sweet sage]

I`m developing a mmorpg for myself

Easy task

[thin stratus]

At least not sane enough

[solar stirrup]

yeah i am

cant wait for Iris

[thin stratus]

can't wait for Mover 2.0

[solar stirrup]

can't believe there's still no 5.4 roadmap

;-;

[thin stratus]

can't wait for all the half broken things that take 8 versions to fix

[solar stirrup]

fix?

[sweet sage]

Is iris broken? is it not on fortnite?

is fortnite broken?

[solar stirrup]

Not out yet

Don't think Fortnite uses it yet

[thin stratus]

Pretty sure you can use Iris already

[solar stirrup]

I played the current season for a while and they are extremely aggressive in the replication

[thin stratus]

Just not really documented

[solar stirrup]

like, anyone not in your city has low update rates and they get culled pretty close