multiplayer

[fleet raven]

could work

the worst I've seen so far was a server with 25% packet loss

surprisingly, it worked mostly fine

but slowly

[grizzled stirrup]

It all still works it's just unpleasant to see things being corrected and also knowing you could be unlucky and fire your shot when there is a moment of loss

Happens in Fortnite too though

[fleet raven]

character movement can easily recover from a little loss

it's mostly unaffected even

[rotund sapphire]

Isnt unreal networking accounds for packet loss via the Ack packets where it repeats the transmission until it completes (or something like that) ?

[fleet raven]

for reliable rpcs yes

[grizzled stirrup]

When I play Fortnite on wifi, I get a lot of rubberbanding (chests opening late, glitching to where I wasn't, shots straight up not seeming to register)

But then when I plug in my ethernet it's flawless

I can see from the indicators I'm dropping packets on wifi

But yeah not a pleasant experience

[fleet raven]

🤔

add a "packet loss defense" feature that sends every packet 3 times just in case

the server will filter out the duplicates anyway

[grizzled stirrup]

Oh no it's unavoidable, it happens in many games that I play

Just always is an unpleasant experience when I encounter it

[fleet raven]

ye but that might be a way to make it less of a problem

[grizzled stirrup]

I mean if it happens on Fortnite, it'll happen everywhere

It's a good idea though

[fleet raven]

unless the loss happens in bursts

[grizzled stirrup]

But really it's just the loss of not important packets that causes the unpleasant feeling (small skips here and there)

Light corrections

The feeling that you are shooting but you never know if you'll miss due to some loss

[fleet raven]

maybe stop using the crappy wifi 🤔

[grizzled stirrup]

Lol I pay 100 a month for it

[fleet raven]

[grizzled stirrup]

It's really fast just seems to suck when not direcly next to it

[rotund sapphire]

kylekatarn you can bring up the wifi device's detail panel and play with the settings, such as allowed power for the device. Windows on a notebook might using your wifi on a reduced rate only for preserving battery. Even if you are powering your notebook from a cord.

[grizzled stirrup]

But yeah ethernet fixes it

[fleet raven]

wait why are you paying for wifi

you don't have your own router?

[grizzled stirrup]

Yeah that's what I mean

I'm paying per month for the service

@rotund sapphire I'm pretty sure it's just proximity, it's in the next room, and my computer has a great wifi adapter

It's funnily actually faster speedwise on wifi

But less reliable

[rotund sapphire]

A magnificent wifi adapter

[grizzled stirrup]

I use it to download things at like 20+ MB/s but on ethernet I only get 4

But ethernet is fully stable in games

I don't even know 😄

It is what it is

[fleet raven]

that sounds like something is configured wrongly

[rotund sapphire]

That could be a faulty of your eth cable, since there are at least two different types. duplex and not

[grizzled stirrup]

Very true I should look in to a new cable

[rotund sapphire]

I have difficulties to recall it tho

[fleet raven]

ethernet is never slower than wifi unless you broke something

[grizzled stirrup]

It's a long cable to go to the next room

Probably it

[fleet raven]

makes no difference

[grizzled stirrup]

I also have a plug in the wall cable that has similar performance

I use that recently mostly

[fleet raven]

they can be like 100m long without problems

[rotund sapphire]

I wouldnt say that Zeblote :)

[fleet raven]

*assuming non-garbage cable

[rotund sapphire]

The amount of interference the cable picks up can be severe on that distance

[grizzled stirrup]

Actually I probably have been using the wall plug ethernet adapter when I was playing Fortnite, so it's expected to not have as much speed

But it is fully reliable

[rotund sapphire]

Yes there are better and worse cables indeed that is true

[grizzled stirrup]

Anyway point being of all this, packet loss feels bad in games and I'm sure there are many people on far worse connections playing

[rotund sapphire]

They're probably used to it already

[grizzled stirrup]

Yeah I know I was back in the day playing on xbox

It was part of the experience 😄

medium ping peer to peer lobbies

Your shot MIGHT count 😄

[fleet raven]

I have one of those crappy flat ethernet cables with no shielding going through multiple rooms to reach the router and it works without problems 🤔

[grizzled stirrup]

I think I misremembered and I had been using the shorter cable from the wall, that's most likely it

Had to tidy away the long one 😄

But yeah I'd say it's quite a luxury to have an ethernet connection for most typical gamers unless they are lucky enough to be next to the router

Most people probably don't even consider it

Especially if gaming on console etc.

[fleet raven]

using garbage wifi: 🚫
drilling holes through walls and ceilings for that low latency: ✅

[grizzled stirrup]

😄

First thing I'll do when I get a house

[fleet raven]

well even better would be to have all the rooms wired up properly inside the walls

one apartment had that and it was really nice

[rotund sapphire]

It's possible to use the power outlets to transfer net in a house, but that's where i draw the line

[fleet raven]

fancy ethernet wall socket in every room

[rotund sapphire]

No i mean the actual power outlets can be used for this purpose i see it on the internet

[fleet raven]

ye but you can also hide ethernet cables in the walls much like the power cables

[grizzled stirrup]

That'd be the dream

[rotund sapphire]

That's different, and indeed a much more favored option

[agile lotus]

How would I network this appropriately

right now hosts saves overwrite the clients

even though its just loading

all saving happens offline and in the options menu

[grizzled stirrup]

@agile lotus I'd use IsLocallyControlled to ensure that the save is happening on the machine that is controlling the player (A client or a listen server host)

[agile lotus]

only loading happens online

saving happens in the main menu only

which is offline

[grizzled stirrup]

Well if you are loading data for each client separately then you'd load it "offline"

As in the client would load it

You could then pass that info to the server if you need to

Otherwise the server will load the same file for everyone

[grizzled stirrup]

Question: when a pawn dies in ShooterGame, these two lines are called: NetUpdateFrequency = GetDefault<AShooterCharacter>()->NetUpdateFrequency; GetCharacterMovement()->ForceReplicationUpdate();

What would be the primary purpose here? To ensure all players see the pawn in the exact position he died at?

[agile lotus]

grabbing the client controller index from server worked

[twin juniper]

What is the create session thing used for?

[past rain]

Can someone explain to me what the logic behind calling "ChoosePlayerStart" before Login and PostLogin is? Because I'm dumbfounded. I have 4 teams in my game, I assign the players their teams in Postlogin, heck I even tried in Login, and then each team has a playerstart, but since chooseplayerstart is called first, no teams have been assigned and its dumb af

How would one get around this?

[grand kestrel]

Find something called before choose player start. Assign teams in choose player start. Probably designed that way so it knows where you're going before logging you in.

You can't really think it's "dumb af" to not design something around the way you do things especially since ue4 came first

Don't see why you can't override the player start at post login either

[fleet raven]

that does sound dumb though

obvious flow would be log in > chose spawn > spawn not the other way around

[grand kestrel]

Yeah

I just hope or assume there's a good reason

Maybe it is actually just done wrong

Like CMC not being modular and pawn possession system 🤭

[fleet raven]

char movement is ultimate

[wet tangle]

My multiplayer when steam is active keeps giving me the error
Invalid session info for session GameSession in GetResolvedConnectString()
when joining

[past rain]

@grand kestrel Its not really possible since the player doesnt truly exist until Login or PostLogin, so thats the earliest you can assign them a team or name or whatever

But for some reason chooseplayerstart is called before both of these functions

So I cant choose a start based on the team they are on

Im sure there is some workaround since games before have solved this

But what that workaround is... I dont know

[grave kayak]

you can probably execute your logic first before calling Super::Login @past rain , GameModeBase calls InitNewPlayer from Login and InitNewPlayer finds the player start

there is also a PreLogin you can use

[past rain]

So you're firstly saying Postlogin is way too late, right?

[grave kayak]

yes, postlogin has already executed the spawn logic

[past rain]

I cant figure out what to do with Login, since it returns a player controller

[grave kayak]

you can also not call super and do the logic in your login:

    ErrorMessage = GameSession->ApproveLogin(Options);
    if (!ErrorMessage.IsEmpty())
    {
        return nullptr;
    }

    APlayerController* const NewPlayerController = SpawnPlayerController(InRemoteRole, Options);
    if (NewPlayerController == nullptr)
    {
        // Handle spawn failure.
        UE_LOG(LogGameMode, Log, TEXT("Login: Couldn't spawn player controller of class %s"), PlayerControllerClass ? *PlayerControllerClass->GetName() : TEXT("NULL"));
        ErrorMessage = FString::Printf(TEXT("Failed to spawn player controller"));
        return nullptr;
    }

/// Assign team here for example

    // Customize incoming player based on URL options
    ErrorMessage = InitNewPlayer(NewPlayerController, UniqueId, Options, Portal);
    if (!ErrorMessage.IsEmpty())
    {
        NewPlayerController->Destroy();
        return nullptr;
    }

    return NewPlayerController;```

[past rain]

So copy paste the contents of the super, into my own

and just do my logic above it?

[grave kayak]

well... if you need the player controller for the team assignment for example

[past rain]

I do, I assign it in playerstate

which needs the playercontroller to get to

So I'd use NewPlayerController as the controller

[grave kayak]

yep

[past rain]

okay let me try

[grave kayak]

you can also let your custom PlayerController assign its own team in PostInitializeComponents

[winged badger]

you won't have a PS then

not unless you override half the GameMode

[grave kayak]

PlayerState is spawned in PostInitializeComponents, right?

[winged badger]

its created from PostLogin

indirectly

[grave kayak]

void APlayerController::PostInitializeComponents()
{
    Super::PostInitializeComponents();

    if ( !IsPendingKill() && (GetNetMode() != NM_Client) )
    {
        // create a new player replication info
        InitPlayerState();
    }```

[winged badger]

ah, so it is, unless its seamless travel

then it gets recreated later

[grave kayak]

yeah, seamless travel is with state migration and such

and you only have the playerstate guaranteed on the server after PostInitializeComponents, otherwise you need to wait for the replication event

[winged badger]

but to spawn a player on a startspot of your choosing, taking information not available at construction into consideration

i think you just need to override a single GM function

[grave kayak]

would be weird to override ChoosePlayerStart and do team assignment there though

[winged badger]

    /** Return true of FindPlayerStart should use the StartSpot stored on Player instead of calling ChoosePlayerStart */
    virtual bool ShouldSpawnAtStartSpot(AController* Player);

that one

just needs to return false

by the time you normally spawn pawns when the game starts that information is ready

problem is PC does FindPlayerStart before anything is ready and caches it

[grave kayak]

won't it spawn at 0, 0, 0 then though?

[winged badger]

no, ChoosePlayerStart will exec

[grave kayak]

AActor* AGameModeBase::FindPlayerStart_Implementation(AController* Player, const FString& IncomingName)
{
    UWorld* World = GetWorld();

    // If incoming start is specified, then just use it
    if (!IncomingName.IsEmpty())
    {
        const FName IncomingPlayerStartTag = FName(*IncomingName);
        for (TActorIterator<APlayerStart> It(World); It; ++It)
        {
            APlayerStart* Start = *It;
            if (Start && Start->PlayerStartTag == IncomingPlayerStartTag)
            {
                return Start;
            }
        }
    }

    // Always pick StartSpot at start of match
    if (ShouldSpawnAtStartSpot(Player))
    {
        if (AActor* PlayerStartSpot = Player->StartSpot.Get())
        {
            return PlayerStartSpot;
        }
        else
        {
            UE_LOG(LogGameMode, Error, TEXT("FindPlayerStart: ShouldSpawnAtStartSpot returned true but the Player StartSpot was null."));
        }
    }

    AActor* BestStart = ChoosePlayerStart(Player);
    if (BestStart == nullptr)
    {
        // No player start found
        UE_LOG(LogGameMode, Log, TEXT("FindPlayerStart: PATHS NOT DEFINED or NO PLAYERSTART with positive rating"));

        // This is a bit odd, but there was a complex chunk of code that in the end always resulted in this, so we may as well just 
        // short cut it down to this.  Basically we are saying spawn at 0,0,0 if we didn't find a proper player start
        BestStart = World->GetWorldSettings();
    }

    return BestStart;
}```

🤔

[winged badger]

need to take some care with ChoosePlayerStart so you don't break the PlayFromHere in editor

basically needs to return default implementation if its PIE

[past rain]

@grave kayak

[grave kayak]

but judging from that code he'll need to do team assignment in ChoosePlayerStart, which is a weird

[past rain]

That got it working

Thank you

[grave kayak]

no problem 😃

[past rain]

Just out of curioisty

is there a reason behind this madness of design?

You'd think logically you'd choose a player start after the player exists

[grave kayak]

you can also do team assignment in the PostInitializeComponents of PlayerState

[past rain]

Is postInitialize called before choose player start?

[grave kayak]

think that's the cleaner design, since then you can encapsulate the team logic in PlayerState

yes

[past rain]

am I suppose to override that function in playerstart or in game mode?

[grave kayak]

PlayerState

and only assign the team on the server and let it replicate to the clients

[past rain]

hmm

Im currently assigning players their team based on NumPlayers, which I increment at Login/PostLogin

Im gonna have to find another system then I assume?

Or just create a NumPlayers variable in playerstate

and increment it in postinit?

no that wont make sense

[grave kayak]

NumPlayers is exposed in game mode, you can get the game mode from the world

[past rain]

ah right

getauthgame mode

[grave kayak]

yup

[past rain]

Yeah that definitely seems like a clean design, Lets see if I can do it

[winged badger]

GM tracks number of players on its own

[grave kayak]

you probably want to change it based on amount of team members if you want to have fair teams, but i'll leave that to you

[past rain]

each team consists of 1 player

its a free for all

makes things easier :p

[grave kayak]

right, in that case use the already existing NumPlayerse of GameMode, it's based on the amount of non-spectating playercontrollers

[past rain]

Yup, I made the post init function, removed login

now im just using postlogin

gonna compile and see if this bad boy runs

Hmm

auto GM = Cast<ACS_GameMode>(GetWorld()->GetAuthGameMode());

It returns null in the post init component

Is it possible the game mode doesnt exist yet?

[winged badger]

no

but its possible PC doesn't know about its own world

[past rain]

Let me check

eh

it was just a dumb bug

had to restart editor

and it works now!!

Much cleaner design

[grave kayak]

make sure you put that within a HasAuthority block, game mode only exists on server, would return null on clients

[winged badger]

in PostInitializeComponents?

Authority check won't work there, no Roles are set yet

has to do a NetMode check

[past rain]

I've tested it on dedicated servers and Im not getting any errors

so it seems like the client isnt trying to get the auth game mode

or something like that

Anyways thank you for both the help, and Im sure I'll be seeing you again soon 😄

[grave kayak]

no problem

[grizzled stirrup]

In general do people use HUD classes as a nice way to store all client UI widgets and have logic to add / remove them from the screen, or do many just do that in the PlayerController?

I imagine it's preference to some degree at this point

It does seem nice that the HUD is client only so you can never accidently try to add a widget to the viewport on the server

But then again it saves a cast every time you want to call a HUD function if you just want to do it on the PC instead

Or else you could have intermediate PC functions that just call HUD functions, but I'm sure that's bad practice

[winged badger]

HUD can subscribe to delegates, you know

and its more then just a preference, managing UI from PC/Pawn is clutter

violates the single responsibility principle

[grizzled stirrup]

That principle being things shouldn't be multipurpose if possible and hold their own specific purpose instead?

I only bring it up seeing stuff like this in shootergame:

AShooterPlayerController* MyPC = Cast<AShooterPlayerController>(Controller);
    AShooterHUD* MyHUD = MyPC ? Cast<AShooterHUD>(MyPC->GetHUD()) : NULL;
    if (MyHUD)
    {
        MyHUD->NotifyWeaponHit(DamageTaken, DamageEvent, PawnInstigator);
    }

[winged badger]

makes them unmaintainable

[grizzled stirrup]

That double cast could be saved if it's on the PC but yeah I agree

[winged badger]

if you mix stuff up

[grizzled stirrup]

It'd be nice to have ALL client UI stuff in the HUD class

Frees up the PC for being a PC!

[winged badger]

once you start managing UI from different objects you'll find yourself spending an hour looking for an object that manages some piece of UI before long

[grizzled stirrup]

Well I have all client UI in the PC so I know it'll always be there

But I like the idea of it being in a HUD class too

[winged badger]

in tiny/hobby/leaning project that is not much of a problem

but when you have over 1k custom classes in the project, ouch

[grizzled stirrup]

I def won't have that many custom classes, but in my eyes it's either get controller and cast to hud and then call the function or get controller and call it directly

It's just a bit easier to get ahold of

But also easier to call it in the wrong place like on the server

[winged badger]

HUD can also hook into delegates, as an alternative

[grizzled stirrup]

Would you recommend I for example make the HUD subscribe to a delegate that gets broadcast when a client calls an OnRep_TakeDamage event?

That'd then pass in the damage taken for example

That'd skip the need to cast etc.

[winged badger]

i prefer using delegates for frequent updates

[grizzled stirrup]

I thought a big point of using them would be to avoid things like binding on tick for infrequent things like updating health or ammo?

[winged badger]

UMG is exceptionally bad at... performance

function bindings are just year 2000 terrible

direct variable bindings are better, but just

just an average RPG game UI, nothing elaborate, will take 5-6ms on gamethread just to keep itself up to date when you use UMG function bindings

[grizzled stirrup]

Yeah binding on tick seems to be the worst idea ever

Do you mean function bindings as in on begin play the widget casts to the owning pawn and binds a custom event to OnHealthUpdated for example?

That'd pass in a single float only when health is updated and set the text, once

Is that still awful?

I thought that was the way to go to prevent tons of needless updating

[winged badger]

thats fine

you don't want to make a function for binding an umg variable

[grizzled stirrup]

Ah yeah I'll stay way from that

I always use SetText()

So it only sets it once, usually from a delegate broadcast

Another slight negative on using the HUD- if you only need say one or two extra widgets for different game modes, you'd either have to pack all possible widgets into a single master HUD class, or else have lots of extra HUD classes per mode that makes it a bit harder to navigate your project

Whereas there'll usually be a unique PC for each mode in my case, and in the master PC I have a create / add widget function so it's a very small bit of extra code to add

But as you said, harder to keep track and it's not fully separate

Though if the mode is TeamDeathmatch for example, I'll know any team deathmatch specific UI will be in TeamDeathmatch_PlayerController for example

[winged badger]

you can have a HUD base with most of the stuff, then just extend when you need a change

[grizzled stirrup]

Yeah I guess it really depends on the complexity of the project

If you need 10s or 100s of widgets, then HUD is the way to go to keep it separate

If you only need like 2 - 6 widgets on screen during gameplay like in my case, it's probably fine to just have a UI section on the PC

Since most modes only need a single specific widget, it wouldn't warrent creating a new HUD class for each one

There's no actual technical downside to not using the HUD class right?

Just purely a workflow decision

[winged badger]

it tends to get out of hand

[grizzled stirrup]

If I do see that happening I'll migrate to HUDs

I just hate jumping through hoops when possible

It's one less step, one less cast and one less class to worry about for the time being which makes life a bit simpler when dealing with UI

But then I can definitely see what you mean- with a large project it could get horrible

Will monitor and switch if needed

Looking through my PC, it's almost all UI anyway, quite a sparse class for what I need at the moment

[past rain]

@grizzled stirrup Was the issue from a few days ago where the client was stuttering locally only on listen-server, while it was fine on a dedicated server ever solved?

Something about having to build from source and change something in the CPC?

[grizzled stirrup]

@past rain I tried many things, some worked but all had downsides

The thing that worked the best to stop the stuttering and have frame perfect movement from the perspective of the listen server is p.NetEnableMoveCombining 0

However keep in mind this appears to send a client RPC on tick to the server

So it uses significantly more bandwidth and isn't really a great option

There are other tricks to smooth only the animation since that also only gets updated on each net update and not interpolated but it messes with the timing of animations and also doesn't look that much smoother since the location / rotation updates are the most jarring

I'm honestly not sure why it doesn't work out of the box because the server should in theory be easily able to interpolate between client updates, maybe it's something Epic didn't quite finish up as listen servers weren't their priority

[past rain]

I know plenty of games made in unreal that uses client server model

take dead by daylight for example

No clue how they get around this

[grizzled stirrup]

Keep in mind other clients see each other and the server completely smoothly

This only affects the listen server's view of clients

But yeah I'd love an interp solution that worked like it does on dedicated

[past rain]

Reading what you just said, we might be talking about different issues

[grizzled stirrup]

I'm sure it's not actually that difficult to vinterp if needed

Yeah possibly

[past rain]

Let me quickly describe what Im talking about - Client-server mode only - Server sees all clients smoothly, All clients see other clients + server smoothly

the only issue is that the client sees ITSELF as stuttering

and only itself

So only your own player is stuttering, and only from your own point of view

[grizzled stirrup]

Ah yeah I remember your post- that is definitely something to do with you setting values on the client but the server disagreeing and moving them back

It shouldn't happen for example on the default third person template right?

You had some rotation logic happening on tick right?

[past rain]

The way I have it is that, GameState holds the player speed value

Because every 30 seconds, all palyers in the game change speed

So at the begin play of the character, I get the speed from the gamestate, and set it to my player

https://gyazo.com/b702613f4d979201a27e683345153d67

This is how Im currently doing it, but I guess something Im doing here is wrong

SetSpeed is called by the server

[grizzled stirrup]

Have you logged both the client and server speed values to verify this is working?

[past rain]

I have multiple times, but I will do it agian now to check

[thin stratus]

Something like Speed should be done via an OnRep

I mean 1. you need everyone to update it 2. if you have late joiners or people getting relevant suddenly, they won't get the RPC anymore

[past rain]

No late joiners

Im doing it in beginplay and have the gamestate keep the player speedso I can set it accordingly

if they die and respawn

[thin stratus]

And they are all in relevancy range?

Just making sure you don't build something that will break

Either way you'd need to multicast the change then though

[past rain]

Im setting the speed on the server, then having an RPC on the client

[thin stratus]

Also is that SetSpeed called on BeginPlay?

[past rain]

isnt that enough?

and yes it is

[thin stratus]

No it's not enough

What's with the other players?

They should walk with the wrong speed?

[past rain]

oh

in gamestate, I get all the players

[thin stratus]

• BeginPlay is too early

[past rain]

and call set speed

https://gyazo.com/cc0465092f5e2515ed559d40dede51f2

So every 30 seconds, this function is called in game state

[thin stratus]

Then it's okay I guess

Cause BeginPlay of the Pawn itself is too early

It's not possessed at that point

• you still need the Multicast

[past rain]

So once setspeed is called from the server, it should call a multicast function that does the same thing?

[thin stratus]

Well yeah, you need to update the MaxWalkSpeed on all players

[past rain]

Isnt that what I already do? The function ClientSetSpeed_Implementation will be called on the owning client or am I wrong?

[thin stratus]

And the other players?

Or do you only have two?

[rotund sapphire]

You could just replicate this value and create an onrep call on it

[thin stratus]

Already suggested that

[grave kayak]

honestly, it's easier to just use a replicated var for this 😛

as robert mentioned

[past rain]

The logic is this - the game state called setSpeed 4 times because there are 4 players, and at the end of the setspeed function I call an RPC on the local owning client, so it should update for all clients too

I do have that sparkie, but in the end I still need to update max walk speed, since im using the cpc

[thin stratus]

If you only call it these 4 times on the server + 4 times the ClientRPC, you basically always leave out the other 2-3 players

And if you use an OnRep, as suggested, it would properly call on all Clients for all Pawns

Which is kinda like a Multicast anyway

Point is the same

[past rain]

Let me ask this then, does any of this explain why on a dedicated server it works fine

but on a listen server it doesnt

[grave kayak]

can you show where you're executing movement?

[thin stratus]

I wouldn't directly explain that. But it's still not enough to only update on the owning Client

The versions of the pawn on the other clients won't get the update

[grave kayak]

^

[thin stratus]

I can repeat this another 20 times if you want :D

[past rain]

Are you saying Im not updating on all the owning clients, but instead just 1 of them?

[thin stratus]

No

You are not understanding how a replicated Actor works apparently

If you have 4 Clients. 1 ListenServer and 3 Clients connected.

And you spawn ONE Pawn for one of the Clients.

Then this Pawn exists 1x on the Server

And 1x on EACH Client

WIth the SetSpeed function you affect the Server one

WIth the Owning Client one you affect the owner

And the other 2 are left out

[past rain]

I thought the CPC max walk speed was replicated behind the scenes

[thin stratus]

No, then you wouldn't need the ClientRPC in the first place

It's not replicated

You need to update MaxWalkSpeed by hand everywhere

That's why we suggest an OnRep float "CurrentSpeed"

That just updates the MaxWalkSpeed

[wanton pebble]

i've got a question. I'm trying to replicate a speed multiplier

{
    AddMovementInput(GetActorForwardVector() * Value / SpeedMultiplier);
}

void ASCharacter::MoveRight(float Value)
{
    AddMovementInput(GetActorRightVector() * Value / SpeedMultiplier);
}```

but if i try do DOREPLIFETIME the SpeedMultiplier, i hit a breakpoint in UnrealType.h

[thin stratus]

Let me tell you that directly: This is the wrong way to do that.

[wanton pebble]

.___.

[thin stratus]

Everything that affects the CMC

[wanton pebble]

you up to explain how do it properly?

[thin stratus]

Needs to be strictly build into it

It's a chunk more complex

I can link you a Wiki post

You need to make your own child of the CMC

[wanton pebble]

i couldn't find SetMaxSpeed in C++

[thin stratus]

And add the speed multiplier stuff properly

otherwise it desyncs

[grizzled stirrup]

https://docs.unrealengine.com/en-us/Gameplay/Networking/CharacterMovementComponent#advancedtopic:addingnewmovementabilitiestocharactermovement

That's a useful post on the topic

[thin stratus]

https://wiki.unrealengine.com/Authoritative_Networked_Character_Movement

This literally has a SpeedMultiplier example

So you should be good with that

Tbh even Sendans changing of the Speed is wrong

But well

[grizzled stirrup]

What is the cheapest way to replicate a took damage UI notification event from the serverside character after taking damage to the cilent character which can then call the necessary UI functions locally? A ClientRPC or an incrementing uint8 RepNotify property set to COND_OwnerOnly?

Thinking the RPC is easiest since no one else has to see it right?

[winged badger]

you're already replicating health

just broadcast a delegate when health changes and hook into it

[past rain]

@thin stratus How would you change the speed correctly from scratch?

[grizzled stirrup]

It can be increased though, I guess I can store the old local health before changing it? Would I have to change my CurrentHealth property to be an OnRep property or is there a way to broadcast the delegate another way?

I have already a delegate for OnHealthUpdated that the server calls when taking damage

But that's a manual specific one

[thin stratus]

@past rain Same thing that I posted for Lukas

You need to do that inside the CMC

[winged badger]

so, broadcast it from OnRep_Health

[grizzled stirrup]

so I should change my CurrentHealth property from Replicated to ReplicatedUsing = OnRep_Health ?

[winged badger]

it is a simple approach, and one that doesn't add any extra network overhead

[grizzled stirrup]

Even doing so, how can I know if I took damage or gained health from a pickup?

[winged badger]

its c++, right?

[grizzled stirrup]

Yeah, I guess I could look into the way Jamsh showed a few days ago

Passing in the old health also

[winged badger]

UFUNCTION() void OnRep_Health(float OldValue);

[grizzled stirrup]

Damn that's all that needs to be done? 😄

That's pretty sweet

[winged badger]

yes

[past rain]

@thin stratus This is pretty much the first time I dabble with multiplayer, and my game will be very simplistic and minimm multiplayer stuff, I dont really want to dabble yet in things like that

[grizzled stirrup]

Last thing- how does the OldValue get passed in?

Since the server will set CurrentHealth

Which will then trigger the on rep event

[winged badger]

the DataChannel checks how many parameters the OnRep function has when DispatchingRepNotifies

and adjusts accordingly

[grizzled stirrup]

Oh so it'll literally match and automatically pass in the old CurrentHealth before it got updated?

[winged badger]

yes

[grizzled stirrup]

That's incredible

Wow

Thank you so much

That's a much more elegant way to do it

[past rain]

So I think Ive made changes

I have no clue if its correct though

[winged badger]

its why the OnRep_Something has to be a UFUNCTION @grizzled stirrup

[grizzled stirrup]

Makes sense!

It's working great

Only slight downside is having to create local OldCurrentHealth floats before setting CurrentHealth on the listen server so he can pass it in when calling his own OnRep_CurrentHealth manually

But it's by far the best solution

Thank you very much!

[wanton pebble]

does anyone know why changing the max walk speed of the client makes it super stuttery?

like at max walk speed 600 it's fine, but as soon as i set it to 200 (crouch) or 800 (sprint)

it's stutter city

nvm, solved it

if anyone's looking, it's because the speed variables had to be set on both the server and the client

[wanton pebble]

another replication question 😅

my ADS is compromised solely of an animation

which i set using a variable, IsAiming

however, i can't get both the server and the client to show the animation themselves and see the other show it when the other is ADSing

    if (WantsToAim == true && CanAim == true)
    {
        if (Role < ROLE_Authority)
        {
            ServerSetAimingTrue();
        }
        else
        {
            IsAiming = true;
        }
    }
    else
    {
        if (Role < ROLE_Authority)
        {
            ServerSetAimingFalse();
        }
        else
        {
            IsAiming = false;
        }
    }
    //```

this is what i'm trying for now, with my implementations being just IsAiming = false/true;

alongside that i'm replicating the IsAiming variable using DOREPLIFETIME

DOREPLIFETIME(ASCharacter, IsAiming);

any idea why my server can't see my client executing the animation?

[grizzled stirrup]

@wanton pebble Server probably doesn't know about WantsToAim so it won't get set (if you are running this same function via the server)

[wanton pebble]

oh that'd make sense

yep that was it @grizzled stirrup , thanks a bunch!

[grizzled stirrup]

No probs 😃

[wanton pebble]

one final replication question before i get banned from this channel xD

i'm trying to do a dual mesh system

i've got it all set up with one exception

i'm using this system within my anim BP to make the TPS mesh bend down

but while the server sees the client executing this, the client can't see the server lean

but if i run just a singleplayer game and go into an external camera, the TPS mesh moves as it should

any idea what the issue could be?

this happens whether Aim Rotation is replicated or not

[thin stratus]

That replicated rotation is in what actor?

@wanton pebble

[wanton pebble]

It's just in the animbp for the tps mesh

[thin stratus]

AnimBPs aren't replicated

You need to replicate stuff in the Character and then pull it into the AnimBP

[wanton pebble]

So I'll need to do my function within the animbp in the character?

[thin stratus]

Anim Blueprint and Character BP are two different things

The AnimBP is not replicated.

You need to do all replication in the Character.

And then grab the Variable from the Character inside of the AnimBP

[wanton pebble]

Ok, thanks

I'll try that out tomorrow

[unreal pine]

Hello friends - Anyone know if I can override a PostNetReceiveVelocity event on a projectile in Blueprint? I'm having some projectile location jitter on clients

[twin sable]

does anyone know why when I get the resolved connection string I get the steam id but not the map url despite having loaded one on the hosting server?

I've read its supposed to look like "steam.123456789:7777//Game/Maps/MainMenu_Entry" instead

NVM its because you should have seamless travel turned off

[fossil spoke]

For initial connection it should be a hard travel yes.

[twin sable]

yeah, realised in the end

thanks though

you wouldnt happen to know why using OnlineSubsystemSteam makes the clients lag around but without its buttery smooth

they rubberband back constantly

[fossil spoke]

🤷 that could be anything

[twin sable]

agh 😛

[fossil spoke]

Whats their ping?

[twin sable]

it happens when testing in the editor

and through steam alike

the stats look ok I think

[twin sable]

looking at a fresh project I've noticed on my project the in rate never goes above 300. I'm not sure what could be causing this

[manic pine]

your outrate is 10kbs?

did you increase the default bandwidth limit

[twin sable]

I dont think so. havent messed with the networking settings

but the in rate is super low which might be whats causing it

[manic pine]

try something like this for testing

[/Script/OnlineSubsystemUtils.IpNetDriver]
MaxClientRate=100000
MaxInternetClientRate=100000
[/Script/Engine.Player]
ConfiguredInternetSpeed=100000
ConfiguredLanSpeed=100000

DefaultEngine.ini

yeah im more concerned about your out rate being 9.998KB/s, which is suspiciously close to the 10kb/s limit

[twin sable]

thats with the new settings

[manic pine]

so its using 40% more than the 10k cap

hows the stuttering now

[twin sable]

that actually fixed it

thanks

so much!

is the initial limit just low or am I doing somethign wrong do you think?

[manic pine]

great!

hmm i think the initial limit is legacy maybe... its very easy to go past it

i believe e.g. fortnite uses 40KB/s

once the client goes past it, he starts dropping packages

instead of sending them to the server

since the movement component is the one using the most bandwidth, its affected the most

[twin sable]

makes sense

bloody awesome though. now i can work on actual gameplay again

[silent sigil]

I have a question about seamless level transitions and voip. I'm running into a weird edge case

In order to debug the issue I broke down the problem into a simple sandbox project. I have a simple multiplayer game with voiptalkers on the player pawns

but when I transition between levels, if I'm talking on the mic it causes a stacktrace, that is sometimes fatal but sometimes recovered from

it usually recovers on the server and goes to the next level but the voice buffer is perpetually full

and so the voice stops working

the error is

[2019.05.13-09.42.06:750][548]LogOutputDevice: Error: Component Name: AudioComponent /Engine/Transient.VoipListenerSynthComponent_0:AudioComponent_0 World Name: World None.None Component Asset: /Engine/Transient.VoipListenerSynthComponent_0:Synth```

[silent sigil]

it looks like it has something to do with registering the voip talker to the player and the component it is being attached to being destroyed during teardown of the level

[thin stratus]

Do bots and their PlayerState survive SeamlessTravel?

I kinda doubt. In CopyProperties I'm moving my TeamState over to the next PlayerState

And that works for Clients, but bots seem to be resetted

[jolly siren]

I wouldn't think that bots would persist

[thin stratus]

Their PlayerStates are added to the List of persisting actors

:/

[jolly siren]

hmm yeah it adds the entire PlayerArray

[thin stratus]

I assume AIControllers are not kept, so the PlayerState isn't either

Yeah so "HandleSeamlessTravelPlayer" passes an AController

But handles only a PlayerController

It skips the CopyProperties stuff for everything else

Maybe it does indeed keep the AIs, but doesn't handle it properly to reinit them

Hm nvm, it does

Wtf

[jolly siren]

Does what?

[thin stratus]

WEll in the GameMode class in HandleSeamlessTravelPlayer, it filters between PC and non PC

Now I'm not sure if that is called for AIControllers

Okay it is. The TeamState is just null

Weird that it works for the Client

[distant trellis]

I have an issue where when you select number of players 2 and the server is interacting with one client my stuff seems to replicate fine but when i try dedicated server client's can't damage each other.. anyone have any ideas?

[scenic raven]

Hello everyone!

when I run mygame.exe -server

it listens on the wrong ip (a vnet I have set up)

is there a way to bind it to the right ip address?

[worthy perch]

I've never used this yet, but there's this: PORT: Tells the engine to use a specific port number.

[manic pine]

thats an OS problem

you can change priority for the windows network adapters ye

[distant trellis]

about to ditch this ability system and go back to C++ replication lol

[scenic raven]

thx

[thick shale]

In the CMC, if I wanted to override any rotation that a RootMotion montage applied, with the Player's input rotation, is there a simple way to do that?

[hushed garnet]

hey all, im having a bit of a moment here. enabled the built in steam plugin and put in the defaultengine.ini the correct variables to define my own appid, but doesnt matter what i do or set, when launched either via a built development build or standalone in the editor it always launches as the spacewar app. I have been trying various things now for a few hours

[tribal shell]

hey, I have an error everytime i Package the Project

ERROR: Unable to compile source files.

ERROR: UnrealBuildTool failed. See log for more details. (C:\Users

[twilit swift]

Hey all, does anyone know if using a RepNotify is more efficient than a Multicast? Same?

I've got a heavy multicast that's eating up network bandwidth. Looking for alternatives.

[manic pine]

its certainly not less efficient

and it allows you to specify how often to do the update(net update frequency) etc.

but if youve a multicast thats that heavy then you might have a design problem to begin with

[twilit swift]

Could be. It syncs prop transforms across all clients. Lots of props in my map and they get moved and shifted around a lot during gameplay. I'm not sure how else to sync things.

[manic pine]

whats the current RPC argument

[twilit swift]

It's a rotation and a vector_NetQuantize_10.

[manic pine]

per object right, so its in a tarray?

[twilit swift]

Yes well, there's one that syncs everything when a new client joins (that one uses an array). The one I'm currently trying to fix is called on the actual prop class and only passes with its own location/rotation.

[manic pine]

why didnt you just replicate the location and rotation directly?

or is it non replicated classes

[twilit swift]

It's a non-replicated class.

[manic pine]

mmm what you could do is looking into delta serialization

its a tarray version that only replicates changed indices

[twilit swift]

hmm... ok that sounds interesting. Thanks! 😃

[manic pine]

still the same information that needs to be sent obviously, but you wont have all the overhead related to actual rpc calls

http://www.aclockworkberry.com/custom-struct-serialization-for-networking-in-unreal-engine/#NetDeltaSerialize_and_Fast_TArray_Replication

[twilit swift]

Nice! thank you!

[tribal shell]

i have problems with the "advandes steam sessions"
I cant package project after I turn the plugin on

[versed socket]

Networking question: I remember hearing in a training livestream a few months ago that the best practice for multiplayer is not to rely on multicasted events or on replicated events in general, but rather to rely on OnRep variables and replicated variables of as small a size as you can go (e.g. Bytes, Chars, and Booleans) and then simply reacting to changes in those variables.

If that's the best practice, then doesn't that imply, then, that all of our framework classes are gonna be a mess of variables like bWantsToFaceDirection or Behavior_State? I just want to make sure that that's a correct interpretation of what they're saying.

[thin stratus]

I assume it means that if both ways are possible, that OnRep is the one to take.

If it makes 0 sense to use an OnRep variable, then don't do that.
Neither use an RPC if it makes no sense.

Technically it's no magic to find out what to use.

[thick shale]

Is it just a bad design decision to try to override Root Motion Montage rotation from the CMC?

[versed socket]

Stupid follow-up question: how do I know if it makes 0 sense to use an OnRep variable?

[fleet raven]

when you're doing a one time thing that leaves no lasting changes

and it won't matter if clients miss it

[grizzled stirrup]

Any clever ways to work out shotgun damage without sending a gigantic amount of info per shot?

Assuming we want say 20 pellets in the shotgun with each individual trace capable of damaging a different actor

No matter what way I think about it, there's going to be 20 RPCs going up at once unless I can pack them into a single RPC?

[jolly siren]

Definitely pack them

[grizzled stirrup]

Just trying to imagine how to sort them so that the server knows how much damge to deal each actor in the array of hit actors that's sent

Assuming it's fine for my client to work out how much each actor should be damaged (in my case that is fine)

My client can run a for loop of traces, populate an array of hit results, check each one if hit a damageable actor, then populate an array of actors damaged

If I also populate an array of damage floats as I'm populating the array of actors damaged, maybe they'll stay in order and I can do a for loop on the server and damage HitActor[0] by DamageAmount[0] ?

It's still potentially quite heavy sending an array of actors and an array of floats in one RPC I'd imagine

[fleet raven]

sending 10 actors (netguid) and 10 damages (float?) is 60 bytes

that's nothing

client side hit detection is very sketchy tho

[grizzled stirrup]

@fleet raven can you elaborate on netguid?

I have an array of AActor* atm

And yep float damages

Also client hit detection is fine in this case (co op game not PvP)

Against server AI

I'm having clients do as much as possible locally but this is one case where they will have to send a bunch of data per shot

I guess a realistic shotgun scenario would only be say 3 actors hit with 8 pellets each or something, so it's not as bad as it seems, just unsure of how to "link" which actors get which damage when sending to the server

[fleet raven]

a replicated actor will have a netguid which is sent as 2 or 3 bytes

[grizzled stirrup]

Oh that is super cheap!

Much less scary than I imagined

So even in the worst case scenario it'll be fine

[fleet raven]

even if it was sending 1KB per shot it wouldn't matter

unless you got some magical minigun-shotgun combo

[grizzled stirrup]

Haha nope this'll be by far the most expensive operation

Well maybe rapidly firing rifles will be just as expensive since they send many separate RPCs in a short timeframe

But I also think I'm underestimating the speed of networks and overestimating how large the data is

Any ideas on how the server can know how much damage to deal to each sent actor? Just hit me that some actors will be hit multiple times so the order of the array won't match, maybe this is where TMaps come in?

[gleaming vector]

well

replicating an actor is only cheap if that actor has already been replicated

[fleet raven]

well why not just sum up the damage to each actor and send that

[gleaming vector]

also

[grizzled stirrup]

@gleaming vector well in this case they are AI enemies that are already replicated from the server to the client, so should be cheap right?

[gleaming vector]

when doing shooting

yeah

what I mean is the first replication of the actor is the whole actor

after that point, it's done by guid

anyway

I've implemented shotguns

and other shooter mechanics

I never replicate the bullets

I replicate the shots

the server sends down information about the projectiles to the client so that it can simulate it

you can start predicting a projectile if you want, and then correct that predicted projectile when you get the authority in

[grizzled stirrup]

@fleet raven I'm not sure how to "link" them, imagine pellet 1 hits Actor A, pellet 2 hits Actor B and pellet 3 hits Actor A again, not sure how to tell it to add it to his existing damage float and also correctly tell the server- deal X damage to Actor A, B etc.

@gleaming vector Oh you are miles more complex- I'm just using simple clientside traces in my case 😃

[fleet raven]

make a temp map on the client side to sort it out

[grizzled stirrup]

And the client can make the call on damage too as it's not PvP

[fleet raven]

not replicating the projectiles is interesting

[gleaming vector]

oh

[grizzled stirrup]

Ok will look into maps, thanks!!

[gleaming vector]

then replicate from client to server impacts

i kinda sorta did this on squad when I worked on that game

[fleet raven]

you mean like instead of using replicated actors, you send a multicast rpc to spawn a projectile somewhere?

[gleaming vector]

the shot was replicated to the server (with a shotid), and then the impact was replicated up (with the corrisponding shot id)

[grizzled stirrup]

I'm not even going that far, this is purely the damage amount on server- there will only be one normalized impact replicated

I just want to have each pellet count in terms of damage

[gleaming vector]

sure

[grizzled stirrup]

The cosmetic fx will be cheap and simple

[gleaming vector]

then batch your impacts

or, well, you really can't in some cases

there is some code I've seen in the engine (i forget where) that counts outgoing RPCs and if it's over a certain number, it batches a bunch of events into a single RPC

[grizzled stirrup]

Cheaper / simpler just to send one FVector_NetQuantize for the impact loc and have the clients simulate the impact normal + do a big impact FX in the center average of where the pellets hit- not as nice visually but nice and simple

But of course for more realism you'd have to send more

[gleaming vector]

oh, i see what you are trying to do

but what happens if you shoot at a corner

and half the bullets go past the corner and the other half impact it?

[grizzled stirrup]

The client will see a big ugly impact fx in the cornder

And not past it

[gleaming vector]

the bullets that keep going may impact another person a few frames later

[grizzled stirrup]

Corner*

But only remote clients

[gleaming vector]

oh

[grizzled stirrup]

The client who shoots can spawn more detailed FX if he wants

[gleaming vector]

just replicate the shot event for remotes

and then spawn projectiles that don't care about impacts

[grizzled stirrup]

Oh yeah that too, at the moment I have everyone doing the simple blocky FX

[gleaming vector]

just for effects

[grizzled stirrup]

I'll probably spawn a decal for each impact

But only one centralized particle system

[gleaming vector]

i use "spawn projectiles" here but I mean however you carry that damage event over distance (can be a line trace)

[grizzled stirrup]

From what I've tested at least, it's not usually worth spawning 20+ separate impact fx in each impact location in terms of what you get in performance

The decals are worth it

[gleaming vector]

sure

[grizzled stirrup]

But in many cases a big centralized single FX works great

[gleaming vector]

you can get away with a number of particle effects

since particle effects are actually much cheaper than decals

[grizzled stirrup]

It's a shame that decals are quite costly but they give the most visual impact imo for shotguns to see the spread like that

20+ particle systems can work great too but if they want to look decent usually it involves a separate mesh emitter for each impact hole as well as sparks/ lingering heat spot

I'd be curious on the performance differences between them, either way shotguns are just creating a lot of stuff in one frame 😄

[gleaming vector]

yeah

well

do it and profile it

then turn it off and do it another way and profile it 😛

that is, honestly, the best path forward in this situation

[grizzled stirrup]

Yep!! Will likely go cheap and cheerful (no decals, one centralized impact fx)

But first, gotta figure out sending the actual separate damage amounts / actors to the server

Thanks for the help though, got some good ideas to test for FX when I come to it later

[neon violet]

Is epics cross platform sdk3a master server/online subsystem?

And is it a better idea to not implement steamworks and wait for them to release it?

[gleaming vector]

no

and if you stick to the online subsystem abstraction, switching will probably be nothing more than an ini change

[silent sigil]

@thin stratus @jolly siren thanks for looking at it. it seems like the bug has something to do specifically with server travel while someone is talking on the mic

i'm trying to repro without seamless travel

i also can't figure out why when i have a voiptalker component on the pawn i dont need to manually call the register with player state function

it just works without

when i do call that function it crashes, but even without calling that function it still doesn't work when someone is talking

[weak fog]

hey, if anyone here happens to have a solid understanding of this diagram, pls feel free to PM me if you wouldn't mind answering a few questions

[fleet raven]

just put your question here

[weak fog]

iight so basically 2 things:

1.) What do the diamonds represent in this graph?
2.) is there a generally agreed upon "best practices" for passing data around in this system?

example for #2:
let's say I have some GameState data like the score, and some PlayerState data like health. Both of these need to be fed into the UI, preferably via event dispatchers. Should I just have dispatchers anywhere that I need to send data from, and then pull it wherever I need it, or is it better to go through some kind of hierarchy?

additionally, I could go through some kind of general event manager service, but there will presumably be a ton of different events per gamemode, and as far as I can tell, I can't put the binding of events behind an interface (I can do the calling, though)

[silent sigil]

Do I need my eyes to be checked? I don't see a single diamond

I see circles which seem to represent server/client. and i see rounded rectangles

[weak fog]

oh wow, it's transparent lol. hold up lemme find a better one

there ya go

[fleet raven]

it's probably explained near where you found this diagram

ownership/"created by" would make sense

[weak fog]

that's what I'm thinking as well, but is there no ownership of pawn ?

[fleet raven]

not really, since they can be arbitrarily spawned and you can then possess/unpossess them

can have pawns in the world that aren't related to any player

[weak fog]

i see. admittedly pawn is probably what I'm the most hazy about in terms of what type of information goes where

ahh that's a good point. a player may have a pawn or may not

in that case yeah, i think ownership is the relation

[silent sigil]

oo wow those were hard to see

[weak fog]

honestly they were driving me so crazy that when you said that, i thought maybe i imagined them

[fossil spoke]

You can think of Pawn as the visual representation of the player.

[weak fog]

right, separate from things like input control etc

[fossil spoke]

They can still implement input controls though.

Depending on your needs

[weak fog]

so an example of useful info from the pawn would be like, if you were making a minimap--relative location on the map

[fossil spoke]

Health is probably a better example.

[weak fog]

hmmm...my initial instinct was that that would go in PlayerState

[fossil spoke]

Since the Player can still exist as a Spectator when they die, you want the Pawn to have Health that then allows it to be "Killed"

PlayerState is better for information that is specific to a Player that all Clients would want to know.

Health is specific to a Pawn, not a Player, since a Player can control arbitrary Pawns with different abstractions of what Health might be.

Score is a better example of something to store on the PlayerState

Since other Clients might want to know other Players scores to display that cosmetically.

[weak fog]

okay. so i had some misconceptions but i'm understanding better now

it seems in this case like i may want to be able to call/receive events from any one of these framework classes tbh

[fossil spoke]

There is no single point of contact thats going to supply you with everything you need. All these classes work together to create the framework.

Its more about figuring out the best place for information and then how to provide it to external systems as seamlessly as possible.

[weak fog]

right. the latter part of that is somewhat eluding me

[silent sigil]

is there anyone i can have an in depth conversation about voiptalkers with?

[fossil spoke]

Its good to ask yourself what the information is doing, who might want to access it and why. This will help you decide where it should reside.

[weak fog]

basically I tried to make an event service, but as far as I can tell, you can't return a delegate (at least in blueprints). perhaps with C++ i could finagle it

but now that you're saying this, i wonder if a better way wouldn't be to use the event system to reinforce the framework

PAWN_EVENT_*, GAME_STATE_EVENT_*, etc.

and listeners accordingly

[fossil spoke]

Yes Event Driven Design is a good approach

[weak fog]

hmm. done TDD but never EDD. i like it

[fossil spoke]

It means that systems are querying data when they need them, instead of external systems strongly coupling themselves to other systems by forcefully providing that information.

[weak fog]

oh yeah, that's exactly my goal right now

you've already got the wheels turning. if i only have event types relative to the framework classes, then I can just set up my eventing based on those classes

[fossil spoke]

If your not familiar with the SOLID design principles i recommend doing a little research and at least being familiar with it.

[weak fog]

yep, i know SOLID. that's kinda how I came to these questions, through my desire to adhere to it so badly

[fossil spoke]

Your not going to write everything perfect the first time.

Iteration is your friend.

Also make sure you understand the networking context of each of the framework classes like the back of your hand.

[weak fog]

yeah. the trickiest one was PlayerController imo. still gotta make sure i'm super comfortable with them all tho

[fossil spoke]

Which are the columns of your image you posted

Yes it being unique as it is.

[weak fog]

yeah, i spent a while just console logging everything and seeing if it printed client or server, or both, and what could replicate and on which policy

you've really helped a lot. this discord is so nice

luv u all

[silent sigil]

I'm just gonna dump some stuff here

so, voiptalkers

I have a fps character. i have a voiptalker attached

[fleet raven]

theres one pretty important gotcha with this gameplay framework: replication sux and a client can have a pawn before he has a player or game state

[silent sigil]

yeah, i have a wait macro

to fix that very thing

it just hangs till the player state is valid

https://blueprintue.com/blueprint/31shxb1m/

[winged badger]

one can force its hand by changing around some NetPriorities on those actors @fleet raven

[silent sigil]

https://blueprintue.com/blueprint/5cjrb6no/

[winged badger]

Pawn is default 3, actor 1

[silent sigil]

so i have two blueprints there

[winged badger]

and GS/PS don't have the defaults altered

[silent sigil]

one registers the voiptalker to the player state

the other does not

[fleet raven]

that's not 100% reliable though, if packets get dropped they can still replicate in different orders

I would like a feature that simply guarantees, 100%, for certain things to replicate before others

[silent sigil]

what exactly is register with player state doing?

i copied the bp wrong so it doesn't show the voip talker as connected to that block as the target but it is

i poured through the code last night

and i see the map of talkers linked to player ids

what confuses me is that the talker works fine without this mapping

i wasn't sure if there was some magic happening with the component setup

[winged badger]

that is a tall order Zeblote

[silent sigil]

but i couldnt find any in the engine source

[fleet raven]

is it really tho

I wouldn't have a problem if it results in delaying a join by a few times latency by simply pausing the replicating until the client has confirmed receiving the game state

wait - can I not already do this by using the replication graph?

[plush wave]

How does one limit voice chat to certain players and not all players?

Or is that something that would require Vivox, for example?

[silent sigil]

what do you mean by limit?

like limit who hears who

[thin stratus]

@weak fog As stated in the Document, this is taken from an outside source (recreated).
The Compendium is meant as a collection of information that is spread around the internet, so while a lot of text and images in there are mine, I found these diagramms to be so good that I used them (see Source 2).

[silent sigil]

So I did solve my problem

I still don't understand the register with player state, since not using that seems to solve the problem.

but it was 2 problems

1 was, i had to not use that, and 2 was i was calling start networked voice in "begin play", which doesn't get called after map travel. so that's an embarrassing mistake but all working now.

[thin stratus]

When using Seamless Travel, is the GameMode the exact same one in the next level (if same class) or does it also pass over data from old to new?

Cause I'm kinda confused as to how UT manages their TeamInfo class.
They just copy it over from Old PlayerState to new PlayerState and never add it to the ActorList that should survive.
Now if I understand that correctly, it should survive anyway because it's referenced in the PlayerState?

Is there an easy check to see if an Actor is from a previous level?

Other thing is, that they don't even check if the TeamStates array is empty or not and always recreate it.

What is the OnCopyProperties call for then >.>

I'm currently checking if the array is empty or not and post seamless travel it seems to be.
So either my GetSeamlessTravelActorList doesn't work (which technicallly it should do already cause the states are referenced in PlayerState, GameState and GameMode) or this is a new GameMode.

Gnah, it should be possible to make the TeamStates survive a travel so that everyone just keeps their Team.

[silent sigil]

@thin stratus being referenced by player state doesn't necessarily guarantee it survives afaik

[thin stratus]

I'm adding it to the list of actors that should survive anyway

[silent sigil]

is teaminfo an actor?

i tried searching for it

and couldnt find it

online

[thin stratus]

Yeah

But not part of UE4 of course

It's custom

[silent sigil]

oh if it is a custom actor

it must be added to the seamless travel actor list

[thin stratus]

Literally what I wrote not even a minute ago :D

[silent sigil]

lol yup

im confirming that, the reason is the reference gets saved but not the actor being referenced.

[thin stratus]

I think the way I set everything up is already wrong. I need a proper way of creating the Teams and then reusing them later.
And that has to work for Players and Bots.

[silent sigil]

hmm

and you want the bots to persist too i assume

[thin stratus]

Currently bots only join a team if they are created, sadly they also survive (which is technically okay) the travel, which never adds them to a team again.

[silent sigil]

oh i see

[thin stratus]

Which technically is also okay, if the freaking TeamStates would survive properly

[silent sigil]

why don't you have the team state be an enum var on the various controllers

[thin stratus]

Cause it does a lot more than just keeping score.

[silent sigil]

i see

i mean

it sounds like you want to have a score state, that is for the game as a whole

[thin stratus]

Keeps track of leaders, top scorer, party members etc.

[silent sigil]

that is one piece of information

and the individual team information for the players

team state sounds like it is doing too much

[thin stratus]

Na, it's the same thing that Unreal Tournament does

[silent sigil]

oh ok

[thin stratus]

It's a solid system if it would survive the travel

[silent sigil]

gotcha

[thin stratus]

I'm just again a bit too lost with all the gotchas of travelin seamless

[silent sigil]

i def was dealing with that

are you in c++

[thin stratus]

Yeah

[silent sigil]

hrm, yeah i exposed the actor list in the game mode to blueprints to debug what i was working on

was a dead end but i did get the pawns surviving the travel

as a thought

there are some functions for when an actor finishes travelling

add some debugs there to the team stats class

[thin stratus]

Right now it works like this:

1. GameMode::InitGame creates a TeamState per needed Team.
2. GameMode::InitGameState passes the TeamState Array to the GameState so everyone can access it.
3. When the Player leaves the Spectator State (In PlayerController) it asks for a Team to join.
4. TeamState array is added to list of surviving actors.
5. PlayerState is passing TeamState over in CopyProperties.

[silent sigil]

and see if maybe the actors are surviving but the references need to be reset on the players

[thin stratus]

Yeah that would be helpful

Have to check AActor api to see what exists for that

Just for debugging

[silent sigil]

let me pull up my project, i am doing a much simpler thing, but i know the game mode had a function you could override

[thin stratus]

Seems like actors have a "bActorSeamlessTraveled" boolean

[silent sigil]

i wasnt sure if it was just for controllers though

well that'll work for you

[thin stratus]

At least for debugging yeah, but still not sure about the whole process of when to create the states and when not, what function is best for that etc.

[silent sigil]

i suggest putting something debug logic wise in the actor that it'll self fire.

are you sure game mode init game wont be called post seamless travel?

[thin stratus]

e.g. InitGame calls even after the travel

Yeah it does, which is one of the points why I think UT doesn't really persist the Teams anyway

Even though they pass the TeamState via PlayerState

¯_(ツ)_/¯

[silent sigil]

i see

but you are checking if a new one is needed

and not generating a new one if not

i assume

[thin stratus]

Yeah i check if the Array is already setup

Which it isn't

Which makes me believe the TeamStates array isn't surviving or at least not as I want it to

[silent sigil]

for the game mode there is some copy logic

you might want to override

[thin stratus]

Idk if the GameMode has a function to move over stuff

[silent sigil]

to make the old array copy into the new one

it does

i believe

[thin stratus]

I only know about the PC and PS stuff

So "bActorSeamlessTraveled" is false for GameModes post SeamlessTravel

.>

[silent sigil]

yeah

[thin stratus]

Only thing I found

[silent sigil]

yeah im looking through the github code for gamemodebase

it looks like maybe you can use the game state class

but im not seeing how tbh

[thin stratus]

Can't see any function here either though

[silent sigil]

SeamlessTravelTransitionCheckpoint was my best guess

but by the time you hit the transition map

the game mode will have changed

[thin stratus]

There is also no way to access old and new GS or GM

[silent sigil]

it's weird cause the docs def say the game mode actor should persist

on the server anyway

are you only looking at the server

or the client gamemode as well

[thin stratus]

Client has no gameMode

[silent sigil]

sorry, yeah, i was confused

i thought it was reped for some reason

try setting a bool in the game mode on begin play, seemless travel and see if the bool is still set

im just grasping

[thin stratus]

I wonder if I can grab the CurrentWorld in the new GameMode

But that also sounds so hacky

    /** the world we are travelling from */
    UWorld* CurrentWorld;
    /** set to the loaded world object inside that package. This is added to the root set (so that if a GC gets in between it won't break loading) */
    UWorld* LoadedWorld;

LoadedWorld is the new one that the new GM is in

if I can grab the CurrentWorld (and it's still valid) and get the GM of it, I could access it

But that sounds wrong too

I feel more like I need to ge the same path PlayerState is going somehow

So instead of keeping the TeamState, I need to copy over the data at some point

But where the fuck is the entry point for something like that

[silent sigil]

I feel like you should be able to keep the team state via GetSeamlessTravelActorList, and then just find all actors of team state post seamless travel

and rebuild your array

[thin stratus]

Let me try to use PostSeamlessTravel to just print the amount of TeamStates in the World

[silent sigil]

sounds like a good start

[thin stratus]

Cause technically, it should be 4 instead of 2

[silent sigil]

gl man, i gotta get some sleep. hope you figure it out

[thin stratus]

Yeah sure, thanks for your time!

Have a good night!

[silent sigil]

np!

[thin stratus]

Welp, it's indeed 4 teamstates

[silent sigil]

cool, so def being added to the list

i was just thinking you maybe added them wrong

[thin stratus]

Welp, now I have to find out how to properly get a grab on them.
It's easy to just find and reinit them, but I have to stop the other 2 from spawning again.
This freaking boolean should be true for a seamless travel gamemode >.>

[silent sigil]

no arguments here

[thin stratus]

Wait what

// scope because after GC the kept pointers will be bad
AGameModeBase* KeptGameMode = nullptr;
AGameStateBase* KeptGameState = nullptr;

// Second pass to rename and move actors that need to transition into the new world
// This is done after cleaning up actors that aren't transitioning in case those actors depend on these
// actors being in the same world.
for (AActor* const TheActor : ActuallyKeptActors)
{
    KeepAnnotation.Clear(TheActor);
    TheActor->Rename(nullptr, LoadedWorld->PersistentLevel);
    // if it's a Controller or a Pawn, add it to the appropriate list in the new world's WorldSettings
    if (TheActor->IsA<AController>())
    {
        LoadedWorld->AddController(static_cast<AController*>(TheActor));
    }
    else if (TheActor->IsA<APawn>())
    {
        LoadedWorld->AddPawn(static_cast<APawn*>(TheActor));
    }
    else if (TheActor->IsA<AGameModeBase>())
    {
        KeptGameMode = static_cast<AGameModeBase*>(TheActor);
    }
    else if (TheActor->IsA<AGameStateBase>())
    {
        KeptGameState = static_cast<AGameStateBase*>(TheActor);
    }

    TheActor->bActorSeamlessTraveled = true;
}```

So the old gameMode is marked as Traveled

Are you shitting me

Weirdly enough, the GameMode is copied over to the new world usually

if (KeptGameMode)
{
    LoadedWorld->CopyGameState(KeptGameMode, KeptGameState);
    bCreateNewGameMode = false;
}

[silent sigil]

so whats yours doing?

[thin stratus]

We will soon find out :P just breakpointing it

[silent sigil]

gotcha

[thin stratus]

Just in case someone is/was interested in what shit is being moved over

So it does indeed completely survive the travel

Not even recreated, it's the same actor

Now the question is ,why is the obvious bActorSeamlessTraveld false in InitGame post travel

Man I need a new coffee

I hope all of you just reading this are actually learning something

[silent sigil]

half awake but excited to learn how it ends

[thin stratus]

So, before I go further through the breakpoints: Yes it is keeping it and the TeamStates are still in the array at that point.
Now the question is: Why is the Number of TeamStates etcl not work anymore post seamlesstravel

Cause I'm doing this in the InitGame function: if (TeamStates.Num() < NumTeams && !bActorSeamlessTraveled)

And right now, if I open the inspector, TeamStates is of num 2 and NumTeams is 2

Even bActorSeamlessTraveled is true

// Init the game mode.
if (AuthorityGameMode && !AuthorityGameMode->IsActorInitialized())
{
    AuthorityGameMode->InitGame( FPaths::GetBaseFilename(InURL.Map), Options, Error );
}

This should never call as the Actor is initialized at this point

And it indeed doesn't go in there. Now I'm confused. How am I able to break point that function if it's skipped

Magic?

[pallid mesa]

wait, hold on, weren't controllers and ULocalPlayers the only ones meant to survive to seamless travels o_O

[thin stratus]

By the looks of it, the LocalPlayer isn't even teared down

So it doesn't have to persist ?

PlayerState def persist

[pallid mesa]

that... should be new, I don't recall this behaviour lol

[thin stratus]

Well I only add the TeamStates by hand

The list above from Visual Studio is otherwise untouched

[pallid mesa]

lmc

[thin stratus]

Oha now it makes more sense

This SeamlessTravel stuff calls twice

Once from A to Transition and then from Transition to B

And A to Transition the GameMode survives, Transition to B it doesn't.

Now is that my bad for not setting something up properly is the question

[silent sigil]

do both game modes exist in the transition map?

[thin stratus]

Both?

[silent sigil]

the game mode for the end map

[thin stratus]

I never touched the Transistion map

[silent sigil]

when does the final game mode come online

[thin stratus]

Thought that just uses the same GameMode and that's it

[silent sigil]

during transition

or on final map load

if it is during transition you can copy everything over

if both actors exist in that map

i dont know if they would

[thin stratus]

As far as I understand, it should keep the GameMode from T to B too

[silent sigil]

i agree

[thin stratus]

Cause from A to T it's one and the same GameMode

[silent sigil]

maybe the transition map doesnt have a game mode assigned

[pallid mesa]

and T to B GameModeBase?

[silent sigil]

is it an empty map?

[thin stratus]

Well I sadly skipped the loop

Gotta have to run a travel again to see what actors survive from T to B

I tell you, develop a Splitscreen, LAN, Online, heavily UT inspired game nearly alone

And you will sooner or later know the freaking engine

[pallid mesa]

you got splitscreen? nice

[thin stratus]

Yeah....

Would rather not have it, cause UI is a mess with this

And they want Online Splitscreen at some point

I will die

[pallid mesa]

welp it looks pretty nice

but yeah, I feel you

[thin stratus]

List from T to B

Missing GameMode, GameState and GameSession

Super not cool

[pallid mesa]

Okay, I'm going to debug a side project to see if I get 1:1

I don't do gamemode discovery in this one, but not sure if because my url option impl I got the right behav

or just because the engine handles it for me

[thin stratus]

All I do in this one is ServerTravel with the ?game=TDM url iirc

So it keeps the GameMode

[pallid mesa]

yeah, that seems right, iirc

try to read the built in options string

to see if it has your settings, while this thing boot up

[thin stratus]

Yeah also checking the point where they try to keep all actors

[pallid mesa]

single process, right?

[thin stratus]

Launch Game via uproject file

[pallid mesa]

k, dc?

or just listen?

[thin stratus]

Actually just offline splitscreen atm

Doesn't really matter if you listen to something or not

[pallid mesa]

well yeah, there are a couple shenanigans implied on empty servers and dc that can give you headaches, but gona try purely SP

[thin stratus]

I already found the problem

void AGameModeBase::GetSeamlessTravelActorList(bool bToTransition, TArray<AActor*>& ActorList)
{
    // Get allocations for the elements we're going to add handled in one go
    const int32 ActorsToAddCount = GameState->PlayerArray.Num() + (bToTransition ? 3 : 0);
    ActorList.Reserve(ActorsToAddCount);

    // Always keep PlayerStates, so that after we restart we can keep players on the same team, etc
    ActorList.Append(GameState->PlayerArray);

    if (bToTransition)
    {
        // Keep ourselves until we transition to the final destination
        ActorList.Add(this);
        // Keep general game state until we transition to the final destination
        ActorList.Add(GameState);
        // Keep the game session state until we transition to the final destination
        ActorList.Add(GameSession);

        // If adding in this section best to increase the literal above for the ActorsToAddCount
    }
}

It doesn't keep itself when traveling from transition

Only if toTransition

Now please someone tell me why

I can easily keep these three in my own gamemode with !bTransition or just overriding everything

[pallid mesa]

🤔

[thin stratus]

But not sure if that's a good idea

Do I want to keep the GameMode?
It's most likely in a late MatchState

Doubt they reset it already

[pallid mesa]

If anything, debug T->B and look for the initialization code

[thin stratus]

Well you can clearly see that they don't keep the GameMode from T to B

[pallid mesa]

yeah no they don't but the URL I guess stands for something