udon-networking

[frank fjord]

And yeah obviously latency means you can do smoothing and whatnot. It's the specifics of how this smoothing is applied, to what and where, that's eluding me.

[cold laurel]

As this can cause the cars to clip into the ground if the road suddenly has an incline.

[frank fjord]

A lot of this conceptually makes sense but I've spent a literal week trying to implement it and not getting very far so clearly I am not understanding some of the nitty gritty

[cold laurel]

What

[frank fjord]

?

[cold laurel]

Latency is in part caused by the speed of light being pretty fast but not fast enough, and the speed servers are able to relay messages.

You need to measure this.

[frank fjord]

Oh, misunderstanding

I was replying to this. I presumed you were saying that "latency" was what was "allowing" their systems to tolerate larger errors or less precise predictions

and that, by this, you meant that their systems were artificially waiting longer to collect more data and smooth it out

[cold laurel]

That would be the case of purely interpolation based systems.

[frank fjord]

I'm aware that messages take time to travel. I've been using that message travel time (from OnDeserialization result) as the measure of how far into the future to extrapolate the data received

[cold laurel]

Yes

[frank fjord]

I may not be doing that in a refined enough way, of course. Lots of things I could be doing in a naive way.

[cold laurel]

Afaik that number isn't quite right btw.

[frank fjord]

But, that doesn't change what I'm talking about regarding how 'precise' the final extrapolation estimates are, with and without acceleration/acceleration RoC as part of the calculus

[frank fjord]

I'm starting to suspect this as well, and I think this is what I'm gonna test today.

I've been trusting receiveTime - sendTime a lot

Maybe that's unwise >.<

[cold laurel]

iirc @pallid shore found a work-around for this a while ago

[frank fjord]

👀

[pallid shore]

sendTime is the time when the package get sent out, but you need the frame time because that is the time of where the position of your object is. If you use sendTime you introduce an error depending on how long the frame takes and when the networking happens in that frame. That can be a few milliseconds and that makes any interpolation or extrapolation appear jittery.

For my snapshot interpolation I calculate the delta time between the last snapshot and current and sync it with the position data. The receiving clients then calculate the delta time with sendTime and by subtracting those two values you can find the error. This works well for me and I get smooth interpolation.

 
    public override void OnPreSerialization()
    {
        double time = Time.unscaledTimeAsDouble;
        deltaTimeToLastSnapshot = (time - lastSnapshotTime);
        lastSnapshotTime = time;
        ....
    }

    private bool TryToAddSnapshot(float time) //time = sendTime
    {
        double lastSnapshotTime = snapshotTime[GetPreviousSnapshotID(nextSnapshotToFill)];

        double delta = time - lastSnapshotTime;
        double error = delta - deltaTimeToLastSnapshot;
        ....
    }

[frank fjord]

Hmm. I'm understanding some of this, but not all of this.

So the tl;dr is that SendTime (ignoring the fact that it's from the recipient's frame or reference) is referring to a time in the middle of a given frame that may not match OnPreSerialization() for the sender, which would be the part of the frame you're getting your data to send.

So in addition to the data, you're calculating some extra time value to send along with the serialization. However this time value is calculated, it needs to be... "un"-calculated, essentially, on the other end, because the recipient only has Receive Time - Send Time to go off of which, again, are from their own time frame of reference...

I'm not reeeeally understanding how you're doing that, though 🤔

[frank fjord]

I can see that on PreSerialization you're calculating how long since the last PreSerialization and storing it as a double. I presume that this delay since last PreSerialization is then synced with the rest of the payload.

I do not understand the context of this TryToAddSnapshot method, though.

Or where "snapshotTime[]" or "GetPreviousSnapshotID()" are coming from/what they accomplish

I guess... TryToAddSnapshot is being run as part of the deserialization process? You're storing an array of previous deserialization result sendTimes? Then delta is the local calculation for how much time YOU think has passed since the last serialization, and error is the difference between that local calculation, and the delta time variable that the sender synced over the network to you...?

If so, I guess I'm wondering 1) how many sendTime timestamps are being saved in that array (if I have that right?), and why you need an array instead of just a 'last value' ... and 2) what are you then using that final calculated error value for?

[cold laurel]

Quap is making use of something called snapshot interpolation
https://gafferongames.com/post/snapshot_interpolation/

[restive cliff]

I tried but to get a good result I wanted the processing power of Udon 2 first so that the client can simulate the vehicle's physics into the future for elevation changes of the road

[frank fjord]

yeah udon processing power is something I immediately start to worry about when thinking about storing a buffer of snapshots... >.<

maybe that concern is overblown though

Having said that

[restive cliff]

I don't think that will have as much as a performance impact as simulating my physics

[frank fjord]

I presume I don't NEED to do full-on start to finish snapshot interpolation, just to reap the benefits of a more accurate sub-frame-based send time

[pallid shore]

Sorry, this is just example code I copied from my system to show how I implemented it. You need to implement it in your system in a way that works for you.

TryToAddSnapshot is part of the deserialization.
The delta in TryToAddSnapshot is the same time as the delta in OnPreSerialization but because it uses sendTime it contains the error. By sending the frame based delta with the position data we can calculate the error and fix the time.

I hold around 1 - 2 snapshots in the array, but it dynamically adjusts the simulation time to achieve a smooth interpolation regardless of ping or package loss. So the array can hold more or less depending on network conditions.

Udon is plenty fast enough to do snapshot interpolation. You will run in networking limitation before you notice any performance limitations from udon. I use it to sync 16 vehicles and it works on quest with out any problems.

[frank fjord]

damn, okay, wouldn't have expected that.

What about perceived latency?

Ironically I'm doing exactly the thing that that GafferOnGames article lambasts, and trying to use extrapolation to reduce the effect of network latency to the user... :S

It's very hard to reconcile going in the opposite direction and increasing the delay between the sender and the receiver from an already pretty harsh 200-400ms to something more like 600-800ms

I guess realistically perceived latency is going to occur anyways though with any of the other systems I'm exploring using to smooth out network updates...

[cold laurel]

Well, the article says to "just sync at 10Hz lol" which is unfortunately unreasonable in the context of UdonSync.

[frank fjord]

... fair x'D

[pallid shore]

I have the system set to try to stay around 50ms + ping time in the past. With a bad ping of 200ms we simulate around 300 - 500ms in the past. The system already extrapolates quite a bit.
With better interpolation between the extrapolated snapshot and the next snapshot you could get the simulation time closer to real time.
With VRChat's networking you will never be able to achieve perfect synchronization because the most you can send is around 10hz of snapshots, for comparison Counter Strike send snapshots at 64hz.

[frank fjord]

yeah I know it can't be perfect, just want it to not be super super late...

[pallid shore]

You can see my snapshot interpolation working in the world 'Jetski Rush'. It's not perfect but it's pretty good I think.

[frank fjord]

Ah, I played Jetski Rush the other day. Nifty.

The issue is that my use case involves a singular shared vehicle that all players cooperatively pilot (or take turns piloting). So like

you know, when you see a different player on a different vehicle jitter a little, it's whatever

but when you're stuck to that vehicle too, jitter hurts a lot

[pallid shore]

You could just not allow it to jitter. If the position difference from one snapshot to another is to big you could just interpolate that smoothly.

We tested our system with two player jetskis and had no problems

[frank fjord]

Oh, to be clear, I'm not saying that as to imply that snapshot interpolation will somehow have jitter where what I'm doing won't

What I'm doing definitely has jitter

and it's stressing me out lmfao

Anyhow. I said I was going to test my basic networking again and see if I could figure out what underlying issues have to do with the serialization/deserialization step, before trying to figure out my smoothing solution's issues. It's been very hard to get my brain in the game again today, this project is bullying me so hard and I'm just tired lmfao.

[pallid shore]

Yeah, networking can be real bully! I hope you find a solution you are happy with!

[cold laurel]

Maybe you could try PID for smoothing along with snapshot interpolation where you insert a speculative extrapolated snapshot as the newest snapshot?

Then when you get the next serialization you replace the speculative snapshot with the received one. Then extrapolate the next and so forth.

[frank fjord]

One quick clarification about this. The remote client running "delta = time - lastSnapshotTime" ...

Time is deserialization sendTime. Is "lastSnapshotTime" ALSO just measured in deserialization sendTime? Or is the last snapshot time a "corrected" value that's been adjusted with the error calculation?

I'm just trying to grok the variables and for whatever reason I cannot wrap my head around what is happening specifically in this local delta calculation

[frank fjord]

This is a clever thought, I like that.

I need to figure out what it would actually mean to implement snapshot interp for my system though 🤔

[pallid shore]

This might help understand it better. lastSnapshotTime is the time with out the error.

    private bool TryToAddSnapshot(float time) 
    {
        if (!isReady) return false;


        double lastSnapshotTime = this.snapshotTime[GetPreviousSnapshotID(nextSnapshotToFill)];

        double delta = time - lastSnapshotTime;
        double error = delta - deltaTimeToLastSnapshot;


        if (error > 0.01f || error < -0.01f) error = 0;
        double newTime = time - error;

     
        snapshotPosition[nextSnapshotToFill] = positon;
        snapshotTime[nextSnapshotToFill] = newTime;
        snapshotRotation[nextSnapshotToFill] = rotation;
        snapshotVelocity[nextSnapshotToFill] = velocity;
    }

[frank fjord]

I guess the gist is that every time you get a network update, you update your list of snapshots organized by time, somehow plot a hermite spline along those snapshots, figure out how long in the past you need to 'start,' and then begin running that spline every frame to get your current state target... until you get a new network update to refresh that whole system?

[cold laurel]

This is roughly how you're calling that method right?

public override void OnDeserialization(DeserializationResult result)
{
    TryToAddSnapshot(result.sendTime);
}

[pallid shore]

Yes!

[frank fjord]

waaaait...

so based on this

lastSnapshotTime first gets set to snapshotTime[x] where x is the index of last frame's snapshotTime value.

... but then you're doing this delta calculation: sendTime - lastSnapshotTime from above

Error is the difference between that delta, and the network serialized delta from the owner.

newTime is sendTime - error

And then newTime goes into snapshotTime[]...

That means that when you populate lastSnapshotTime based on snapshotTime[x] NEXT time...

the value your getting IS a value that's been corrected for error

[frank fjord]

Maybe semantically that's what you were saying. By "without the error" you meant "corrected by subtracting out the error?"

I originally read that to mean "without having corrected for error"

[pallid shore]

Yes sorry, it's the corrected time

[frank fjord]

okay, okay

I guess, logically, that makes sense. Dunno why it was SO hard to wrap my tiny little brain around that lmfao

[frank fjord]

I presume from these snapshot arrays for position, rotation, and velocity... you then plot a series of hermite splines? With t values based on the snapshotTime array?

I'm curious though, if your local simulation is kinematic - and I figure it would have to be, wouldn't it? if you're running through the interpolation every frame - why would you need/use velocity snapshots? Since the positional changes every frame would mean you're not really... USING velocity to move the objects... right?

[cold laurel]

Because the velocity is the hermite spline's tangent.

Think of the velocity as a bezier handle in a curve

[pallid shore]

Mhm and we need the velocity for gameplay things

[frank fjord]

Oh, okay. So it's not 3 hermite splines

it's 2

one for position (which also needs velocity data)

[cold laurel]

it's just one

[frank fjord]

and one for rotation, which... I guess doesn't need angular velocity because you're doing something else aren't you--

[cold laurel]

Doing Hermite for quaternions is very hard

[pallid shore]

Rotation is just Quaternion.SlerpUnclamped

[frank fjord]

ah okay

In my case, rotation only happens on a single axis: the vessel can rotate/thrust about its yaw, and no other axis.

Still, I'll try without doing an interpolation for that, at first

it'd be nice not to have to

[cold laurel]

then you can absolutely do hermite for rotation

You just need to make sure you handle wrapping correctly

[frank fjord]

@cold laurel you mentioned that the representation of cubic bezier curves and cubic hermite splines was either the same, or very close to the same. Any chance you have pseudocode on hand for what that looks like?

[frank fjord]

yeah I've run into this, frustratingly, more times than I care to admit

[cold laurel]

It's two different representations of the same concept

[cold laurel]

I do actually!

[frank fjord]

continuity on multiple derivatives, right?

[cold laurel]

just one

[frank fjord]

or in this case, just the velocity--

right

we're just gonna cross our fingers that that's enough for human perception at this scale

[cold laurel]

You'll really only run into issues if you have vehicles that can accelerate really quickly.

[frank fjord]

... is that why they're both "representations" of the same concept? They wouldn't be, anymore, if the hermite spline was continuous on all derivatives?

The airship can't accelerate "really" quickly, but... it also doesn't have constant acceleration. Its acceleration also has acceleration.

[cold laurel]

Hermite spline is NOT continuous on all derivatives.

[frank fjord]

oh, I thought it was.

Literally thought that's what this video was about

[cold laurel]

It only has C1 continuity

[frank fjord]

clearly it's been more than a week since I watched >.<

[cold laurel]

Also called "Tangent-continuous"

Here's a cubic bezier drawn with lerps

[frank fjord]

This is a lerp of two lerps, yes?

[cold laurel]

6 lerps

[frank fjord]

oh-

okay yeah I see it

I think I have this recollection that someone's code - maybe it was Guribo's vehicle sync, or MMMaellon's light sync - using a "herminte spline" which was a lerp of two lerps.

but I guess that would not be continuous on the first derivative

[cold laurel]

https://youtu.be/jvPPXbo87ds?t=2615

This timestamp covers Bezier => Hermite conversion

[restive cliff]

-_- this discussion is putting way more thought into the networking compared to networking assets sold on the store

it is a tricky subject but definitely pays off when we can't send as many networking updates

[frank fjord]

yeah I wish that it wasn't a small (or frankly large) semester course just to do the most basic thing (send data in a way that is usable for other systems). Beyond programming all the ACTUAL functionality that's built on top of it...

but here we areeeee

[frank fjord]

Is this a representation of a spline between two data points? A start and end position and velocity? Or is this a representation of a spline between more than 2 data points?

[cold laurel]

yes

[frank fjord]

both? x'D

[cold laurel]

start => end

point A, handle A
to
point B, handle B

[frank fjord]

Okay.

So a full interpolation of a set of, say, 4 timestamps... would be 3 of these

one from 0 to 1, one from 1 to 2, one from 2 to 3, and one from 3 to 4

[cold laurel]

If you draw a spline with more points you're really just pretending it's one spline.
Each segment from point to point is a separate span.

[frank fjord]

I should know this but it's slipping my mind, what's the math to calculate the four control points (I presume p0 to p3, since those inform the LERP functions being used)?

[cold laurel]

You don't have to calculate it with 6 lerps.

It's rather inefficient to do so.

You should use the matrix representation instead.

[frank fjord]

🤔

[frank fjord]

My brain is not computing this, lol >.<

[cold laurel]

For example

Many ways to arrive at the same answer.

[frank fjord]

oh NO I don't remember any matrix math lmfao

gwuhhh

found the related timestamp for this, watching

[strange crane]

Question: Does OnPlayerLeft gets trigged for the player leaving or for the new owner that takes over?

[frank fjord]

OnPlayerLeft I believe fires on everyone, it has nothing to do with ownership

[strange crane]

Even the player leaving?

[frank fjord]

Btw, when YOU leave a world... momentarily, OnPlayerLeft fires on your client MANY times at once... for all the OTHER players

because technically, all those other players are network-disconnecting from YOUR local machine... right before you close your local copy of the world

I do not remember, but also don't believe, that it fires for you leaving ever

but also it doesn't matter because if it did, you wouldn't be able to do anything with that information before there was no world left to do it in

you don't have time to, say, send a serialization

you're already gone

[strange crane]

        public override void OnPlayerLeft(VRCPlayerApi player)
        {
            if (!player.IsOwner(gameObject)) return;

            string playerName = player.displayName;

            var keys = inWorldPatreons.GetKeys();

            if (keys.Count == 0) return;

            for (int i = 0; i < keys.Count; i++)
            {
                if (inWorldPatreons.TryGetValue(keys[i], out DataToken value))
                {
                    DataList list = value.DataList;

                    if (list.Contains(playerName))
                    {
                        list.Remove(playerName);
                        RequestSerialization();
                        break;
                    }
                }
            }

        }

This is what I'm doing and it should on trigger for the owner of the script.

[frank fjord]

So you're wondering

if the owner of this scripted object leaves, does another player assume ownership over the object before it gets called?

Or is there a problem where it runs too soon, and there is not a 'valid' owner

[strange crane]

It automatically transfers the object to the next player.

[frank fjord]

Okay.

[strange crane]

It maybe that fact is it called as soon as the player leaves.

And it most likely is getting trigged but since they aren't the owner or the fact it hasn't update the owner. Based on what you're saying.

[frank fjord]

I may be misunderstanding what your question/problem is

[strange crane]

This datalist is not getting updated.

[cold laurel]

Yeah no, this won't work.

[strange crane]

I am assuming Im going to have to call it back in like 4 or so seconds?

[cold laurel]

You're trying to run code for a player that is in the process of leaving the world.

Don't.

You should instead handle this in OnOwnershipTransfer

Let the new owner clean up.

[strange crane]

Doesn't that get called on Networking.SetOwner?

[cold laurel]

Afaik it happens any time the ownership changes.

[strange crane]

Alright.

[cold laurel]

holdon, I might be making this up actually

But it would make the most sense to me if it's this way.

@restive cliff Do you happen to know?

[frank fjord]

if OnOwnershipTransfer() runs on other players when the owner leaves?

[cold laurel]

Yeah

[frank fjord]

god I should hope it does

if it doesn't that'd be REALLY weird and unstable

[cold laurel]

Exactly

[frank fjord]

... but hahahah surely that means it must work the way we expect right

orz

[cold laurel]

Do you have time to test it rn?

[frank fjord]

uhhhh

I'm not at my desktop at this very moment so honestly it'd take me like a half hour, I'm not fast 😔

[restive cliff]

I consult the documentation every time and yes SetOwner will make everyone call OnOwnershipTransferred

[cold laurel]

Nono, we're wondering if OnOwnershipTransferred is guaranteed to fire when the owner leaves.

Like, I would hope so.

[restive cliff]

I can do a build and test to check

[restive cliff]

yes

[cold laurel]

🎉

[frank fjord]

So I'm thinking about this implementation.

On a network update, you add the snapshot with its given time, and then also update the current extrapolation whose time is the snapshot time plus how long you estimated it took to reach you. Now, at a minimum, you have two points, which means a minimum of 1 spline. You can start an interpolation from this.

As new network updates arrive, old updates get pushed back in the snapshot list but eventually just get forgotten. Since you're probably using your current time (minus some manual latency offset) to analyze where in the interpolation between snapshots you should be, this is usually fine.

Buuuut what happens if your current point in the interpolation is between your most recent received snapshot, and your current forward prediction? You're now somewhere between time = last update and time = last update + half ping

Because the point your interpolating to (the extrapolation) is guaranteed to in some way be wrong, and is being "replaced" with a real snapshot with real information, I think... you have to add your current in-between position from the interpolation, as a new snapshot, at your current real time (no added manual latency), stick the newly received update after or before that, create a new prediction, and now...

... you may have instantaneously changed the shape of the spline you're interpolating upon. Which would cause a noticeable jerk, wouldn't it?

[cold laurel]

Yeah, if you predict too far into the future and the next snapshot arrives "too early" it might "rubber band". But due to the PID it will hopefully not look too harsh.

[frank fjord]

This happens because if you have a quadratic bezier curve, and you interpolate halfway (or some other fraction) through it and then pause and add that current new point as a new control, that changes the curve on either side...

[cold laurel]

it shouldn't

[frank fjord]

Hopefully, buuuuut I did not have great luck yesterday implementing a PID controller...

It shouldn't?

[cold laurel]

Not if you do it correctly no

[frank fjord]

Hah...

Okay. Well I'm down to give it a shot, not like I have any other leads.

[cold laurel]

I'm not sure if this is necessary but a quick google brought me here: https://en.wikipedia.org/wiki/De_Casteljau's_algorithm

If your extrapolated snapshot was 100% correct and you received a snapshot "too early" it wouldn't cause any jump or stutter.

Because the data you received is along the curve you were already on. And in the perfect scenario in the exact same position at the same time as your interpolated version.

[frank fjord]

Yeah, but the extrapolation will basically guaranteed never be 100% accurate

[cold laurel]

Yes

[frank fjord]

especially since I cannot math out the drag

especially especially if I give up trying to also send acceleration and acceleration RoC data

[cold laurel]

Yeah, don't even bother with 2nd and 3d derivatives

[frank fjord]

's just a bummer, I want the math to work out >.<

[cold laurel]

https://youtu.be/jvPPXbo87ds?t=1489

About C∞ continuity

[frank fjord]

Okay, gut check me here. Is the implication here that if I have a spline from point 0 to point 1 and I just run the interpolation past T = 1, that's a valid extrapolation from the point 1 state that implicitly takes into account continuity on all derivatives from point 0 to point 1...?

[sage trail]

using UdonSharp;
using UnityEngine;
using VRC.SDKBase;
using VRC.Udon;
using UnityEngine.UI;

[UdonBehaviourSyncMode(BehaviourSyncMode.Manual)]
public class OnCollisionUpdate : UdonSharpBehaviour
{

    [UdonSynced] int playerNum = 0;
    public Text Number;

    public override void OnPlayerTriggerEnter(VRCPlayerApi player)
    {
        Debug.Log("Entering Happened");
        Networking.SetOwner(Networking.LocalPlayer, gameObject);
        playerNum++;

        Debug.Log(playerNum);
        Number.text = playerNum.ToString();
        RequestSerialization();
    }

    public override void OnPlayerTriggerExit(VRCPlayerApi player)
    {
        
        Debug.Log("Exteting Happneed");
        Networking.SetOwner(Networking.LocalPlayer, gameObject);
        if(playerNum <= 0)
        {
            playerNum = 0;
        }
        else
        {
            playerNum--;
        }
        
        Debug.Log(playerNum);
        Number.text = playerNum.ToString();
        RequestSerialization();
    }

    public override void OnDeserialization()
    {
        Debug.Log("The Deserialization");
        Number.text = playerNum.ToString();
    }

}

just a collider that should change a text UI to the current players in some collider

but there's a mistakes like higher number than the real

[cold laurel]

Ye

[frank fjord]

Interesting.

Requires at least two elements though to make the prediction, so for the first network update I need to do other stuff, like my dumb overengineered calculus :S

[cold laurel]

This shouldn't need to be synced at all

[sage trail]

if someone entered the world later he 'll have the value 0

it need to be updated from a vairiable that aleady there

[cold laurel]

Just handle it locally like this:
https://youtu.be/fjHP80CQzZ4?t=253

This video covers how to reliably count the number of players inside of a trigger collider.

Without the need for sync.

[sage trail]

aaam I really wonder how I'll watch it

but I'm really trying to understand the sync system and I don't get it

[cold laurel]

That's okay! It's a lot.
I'm just saying that your current problem doesn't require sync to solve it. And that applying sync in this case is the wrong approach.

[frank fjord]

@cold laurel btw - I'm not gonna work on it today because clearly I have plenty of other groundwork to do first, but - once I get to PID controllers, is that a subject I can pick your brain on? My attempt to implement one yesterday was disastrous. It "functioned" but no matter how I tuned it, even if it looked stable 99% of the time, it would eventually enter a death spiral and become unrecoverable.

[strange crane]

        public override void OnPlayerLeft(VRCPlayerApi player)
        {
            playersToRemove.Add(player.playerId);
        }

        public override void OnOwnershipTransferred(VRCPlayerApi _player)
        {
            if (!player.IsOwner(gameObject) || inWorldPatreons.Count == 0)
            {
                playersToRemove.Clear();
                return;
            } else if (player.IsOwner(gameObject) && inWorldPatreons.Count == 0)
            {
                playersToRemove.Clear();
            }

            var keys = inWorldPatreons.GetKeys();

            for (int i = 0; i < keys.Count; i++)
            {
                if (inWorldPatreons.TryGetValue(keys[i], out DataToken value))
                {
                    DataList members = value.DataList;

                    for (int j = 0; j < playersToRemove.Count; j++)
                    {
                        if (members.Contains(playersToRemove[j]))
                        {
                            members.Remove(playersToRemove[j]);

                            inWorldPatreons.SetValue(keys[i], value);
                            RequestSerialization();
                            UpdateInWorldPatreons();
                        }
                    }
                }
            }
        }

Since everybody calls OnPlayerLeft() I just add the player that left. Then OnOwnershipTransferred I got through the last to prevent unnecessary overhead. It seems to still not work.

[cold laurel]

Sure! I love PID controllers. Also PID tuning is an artform.

[frank fjord]

I see this now. But I did not think it would be so 'artful' as to spiral out of control at 99.5% of possible configurations. I actually figured worst-case scenario was a perfect oscillation, not a death spiral

[cold laurel]

Once you understand what each term does and in what order they should be tuned things will become way easier.

Btw. for a lot of scenarios you really don't need the integral term. A PD, or even just P controller can be fine.

In some cases you can even get away with a conditional controller

If wrong, fix it

😆

We can get away with this in games because we don't have to deal with pesky actuator latency and irl machine tolerances.

[frank fjord]

When you run your interpolation through your snapshots, how do you quickly find the indices that match your current time target? It seems to me like that would require searching snapshotTime[] every frame for the two indices that your current time target falls between, including edge cases for overshooting the last received state etc...

[pallid shore]

The snapshot arrays are ring buffers, if the time from the next snapshot is in the past of the current simulation time I just skip to the next

You only have to switch to the next snapshot when you are done with interpolating

[frank fjord]

So... if I'm thinking about this right:

If I know that my current time target falls between given snapshots 0 and 1...

My T value to interpolate on is my time target, minus the time at snapshot 1, all that divided by "time at snapshot 1" - "time at snapshot 0" (for a 0 to 1 value)

I need 4 control points:
c0 is just position at snapshot 0
c1 is position at snapshot 0 PLUS velocity at snapshot 0 ... without any scaling, right? There's no need?
c2 is position at snapshot 1, MINUS velocity at snapshot 1?
And c3 is position at snapshot 1...

And then... I can either create a new of 3-then-2-then-1 lerp functions...

... or do math that doesn't suck as much x'D

probably polynomial expansion

I do not understand matrix math

[coral ingot]

Use chat gpt 😭

[frank fjord]

lol

I'll pass but thanks.

[frank fjord]

Okay, I made a first pass at a snapshot-interpolation based sync system. It's not great.

At high - and consistent - sync rates, the rotation is actually quite good just using the quaternion SLERP. However, if sync rate changes - for example, if I release the controls, I've programmed the sync rate to go down - obvious rubber-banding occurs when acceleration or deceleration is taking place.

Position and velocity, on the other hand, are a trainwreck. Perhaps I'm implementing the cubic bezier curve incorrectly: every interpolation significantly overshoots the target position and velocity, and significantly so, resulting in a rubber-banding effect every time the pair of snapshots being interpolated between rolls over.

My control points are all based off of snapshots of received data, which I'll just call initial and next.

P0: initial position
P1: initial position + (initial velocity * X)
P2: final position - (final velocity * X)
P3: final position

For X, I've tried various multipliers in an effort to fix this rubber-banding: namely 1, 1/3, and the difference in timestamps between the initial snapshot and the final snapshot.

CLEARLY I'm doing something wrong here... :S

@cold laurel @pallid shore

[random parrot]

oo, haven't seen this before.
this was very informative for me

[cold laurel]

I'm glad it was of use!

[random parrot]

always wanted to learn more about how this sorta networking is supposed to be done, just never really got around to it

going to go read some of the other blog posts this guy has

thanks for sharing :)

[cold laurel]

Mathf.InverseLerp();

[frank fjord]

I'm not following what this is meant to imply

[cold laurel]

If you're using a cubic Bezier and not a cubic Hermite spline you'll need to convert the velocity to Bezier control points.

[cold laurel]

Use it to find the t value from the snapshot timestamps.

[frank fjord]

oh, fair enough. I already mathed it out another way though

[frank fjord]

This I presume is what I'm fucking up

P1: initial position + (initial velocity * X)
P2: final position - (final velocity * X)
P3: final position

For X, I've tried various multipliers in an effort to fix this rubber-banding: namely 1, 1/3, and the difference in timestamps between the initial snapshot and the final snapshot.```

This was what I tried.

[cold laurel]

1/3 is the correct value

Please ensure your velocity isn't 0

And that your spline is implemented correctly

[frank fjord]

I did try that, but it continued to produce tremendously jumpy behavior..

[cold laurel]

Can you try to plot the values with gizmos?

It was very helpful for me when I was figuring this stuff out.

[frank fjord]

I implemented both the bernstein expression and the polynomial coefficient form, they behaved identically which leads me to believe I did in fact implement them... generally correctly

or I am just tremendously good at messing up in exactly the same way

It seemed like, at every pair of snapshots, the interpolation would begin slow and then accelerate into - and past - the target. As time rolled over and triggered a change in the snapshot pair, the initial position would be behind where the last interpolation ended, causing a snap-back

[cold laurel]

Can you implement the DeCasteljau form just to be sure?

[cold laurel]

I ran into this same issue too. I had to scale my velocity values by the delta time from the previous to the next span.

Iirc

[strange crane]

Does anybody know If OnPlayerLeft or OnOwnershipTransferred get called multiple times?

[frank fjord]

by this, do you mean scaling the velocity component for P1 and P2?

[cold laurel]

Yes, I think I had to multiply the hermite tangents by the time difference

[frank fjord]

I believe when I scaled the velocity component by delta time - JUST delta time, not even with the 1/3 factor - it caused the opposite behavior where instead of accelerating past the target, each interpolation would fall short of the target, causing each new pair of snapshots to force a jump forward

[cold laurel]

I can take a look at my implementation after school.

[cold laurel]

Yes, because it's the wrong value and deltaTime is generally very small.

[frank fjord]

Aye, but you suggested you had to do that yourself?

Anyhow. I'll come back to this in the morning, it's 5am and my brain is very flat from Unity bullying me all evening. Best of luck with classes. 💤

[hoary frost]

i think on ownershiptransferred might, but why onplayerleft?

[frank fjord]

what do you mean 'multiple times...?'

for a given triggering event, no

[strange crane]

        public void DoThing()
        {
            if (api.IsPatreon)
            {
                Networking.SetOwner(player, gameObject);

                playersToRemove.Add(player.playerId);
                Cleanup();
            }
        }

        public override void OnPlayerLeft(VRCPlayerApi _player)
        {
            if (patreonsInWorld == 0) return;

            playersToRemove.Add(_player.playerId);
        }

        #region Detecting Leave

        public override void OnOwnershipTransferred(VRCPlayerApi _player)
        {
            if (player.IsOwner(gameObject) && patreonsInWorld > 0)
            {
                Cleanup();
            }
        }

I add the player that left to a list and once the ownership gets transferred I check to see if they are in another datalist in cleanup. If they are I remove them. When I use DoThing() it works fine but when I use the other two methods its dosen't work. Maybe I am overlooking something.

[frank fjord]

but there are possible instances where multiple valid triggering events occur at once...

[cold laurel]

Not Time.deltaTime. The difference in time from the previous snapshot to the next one. Another word for difference is delta. Sorry for the confusion.

Sleep well btw!

[frank fjord]

Okay, we do in fact agree there: I was multiplying by the difference between the snapshot times.

💤

[cold laurel]

You're assuming that OnPlayerLeft is guaranteed to fire before the ownership transfer. This is a bit sketchy.

[strange crane]

Well there is no real good way I can go about this.Atleast that I thought of.

[cold laurel]

It looks like you're trying to make event driven UI.
Imo. It's not worth the headache.
Just rebuild the UI fully every time the data changes.

Make a single entry point, RefreshUI or something.

And then just fetch the current players in the world via VRCPlayerApi.GetPlayers();

And go from there

[strange crane]

I tried that earlier

and for some reason it will having issues.

[cold laurel]

What issues were you having?

[strange crane]

It was not removing the data from the synced DataDictionary.

[strange crane]

This is a prefab. I'm creating so I'm trying to keep done rebuidling much as possible cause it could have a bit more data then I have.

[cold laurel]

Why do you need to sync it?

Can't you just handle this locally

[strange crane]

One holds patreon data the other holds data of the current one in the world.

[cold laurel]

Nothing about this needs to be synced

[strange crane]

So using Contains() would be less overhead then just checking locally and then adding to one datadictionary?

[cold laurel]

When a player joins the world just check if the patreon list Contains their name. If it does, add them to the list of inWorldPatreons.

Do the same when a player is leaving. But remove instead.

[strange crane]

Yeah I am doing that by default.

[cold laurel]

And then remove anything sync related.

[strange crane]

The only difference is I sync it so people don't have to go through each tier. Depending on how many the player give in this prefab

[cold laurel]

You're trying to optimize by using sync?

[strange crane]

That was my thought it was better to check locally each player that joins then add them to a data dictionary for each tier.

[cold laurel]

Sync is really expensive and limited though?

Avoid it at all costs.

Also, what you're doing here isn't even close to expensive enough to warrant "optimization" via sync.

[frank fjord]

@cold laurel turns out I overlooked something.

The correct solution for the P1 and P2 points was to multiply the velocity vector component by the snapshot delta time, AND divide by 3.

From my tests, this didn't seem like it COULD possibly be the solution. Because dt is a small number, and just multiplying by dt APPEARED to cause forward-rubberbanding (undershooting the target), I couldn't imagine that further dividing that by 3 would move the system in the right direction.

Well, it did

The interpolation is now very good.

There is still a problem, which is that this implementation doesn't work great when the snapshot delta time changes.

When a player is operating a control, they send updates about the system about 6 times a second, give or take. But when the player lets go of the controls (even if the ship is still moving), I made the decision to reduce the update rate to about once a second.

The snapshot interp gets a lot rougher when in this low-refresh-rate state, sadly.

I suspect what's going on is that every time the snapshot interp goes into extrapolation (past the last known snapshot), the misses are just... very egregious, and so corrections become extremely noticeable...

[cold laurel]

Yes that is correct. Because you're using a Bezier and not a Hermite spline.
I guess I wasn't able to convey the information clearly enough. Sorry.

[frank fjord]

No, you did. I just initially discounted that solution because of what I observed experimentally

Today I just woke up frustrated and was like, fuck it, I'll try it anyways, make sure I have my bases covered.

[frank fjord]

What would be the implementation of a hermite spline?

[cold laurel]

Google :3

[frank fjord]

x'D

pain

[cold laurel]

Like with Bezier there are many ways ro arrive at the same answer.

[frank fjord]

Okay, I'll go give it a look but genuinely my attempts to even find a bezier curve implementation of what I am trying to do were fruitless

hence my endless trial and error

[cold laurel]

iirc Varneon had a really nice spline library in the works.

But I don't fully recall how they're doing the matricies.

Oh well

[frank fjord]

I feel way better about this today than I have in like a month

I could explore a PID controller on top of this as a way to maybe allow extrapolation, but even without that, the snapshot interpolation makes the replicated ship's movements very smooth. I do have to serialize the ship fairly frequently, and I don't get to take advantage of slowing down that tick when I'm not using the controls, but that's okay for the one super important vehicle.

I guess I have to seriously consider how much effort I'm willing to put into forward prediction, vs how important it is to reduce the visual delay in the ship's movements between clients...

[frank fjord]

By the way, this comment. When you say "because you're using a bezier and not a hermite spline," what are you referring to?

Is this a commentary on how the extrapolation behaves/overshoots? Just the difference in the math implementation for the control points? Or something else?

Trying to figure out if there actually IS any reason to explore implementing snapshot interp with a hermite spline

or if such an implementation would have the same properties and behaviors, ultimately

[cold laurel]

You're looking for C1 continuity. And a way to achieve that is to use a Hermite spline.
You've implemented a Cubic Bezier, and are using it to calculate a Hermite spline.

[frank fjord]

Gotcha.

[cold laurel]

To represent a Hermite spline with a Bezier you need to convert the Hermite spline's control points to Bezier control points.

Which is what you're doing when you divide by 3 and flip the last tangent

If you were to use an actual Hermite spline directly you could skip that step.

[frank fjord]

Okay. I was pretty sure that was what was going on but we keep phrasing things slightly differently going back and forth and it casts doubt that I actually know what I'm doing lol

[cold laurel]

But you would still need to scale by the delta time between snapshots.

So the scaling is mandatory

[frank fjord]

aye

[cold laurel]

but the scaling by a third is required because of the spline you chose

[frank fjord]

so no, it sounds like there would not really be any behavioral differences between this implementation, and a different implementation of a hermite spline

'cuz I AM effectively implementing a hermite spline

[cold laurel]

yeah, in a roundabout way

[frank fjord]

I'm good at roundabout solutions 🙂

(I'm actually quite bad at them)

[cold laurel]

It's fun!

A little journey if you will.

I'm glad you're happier with the sync now though!

[frank fjord]

Me too. I feel like I've been bashing my head on this forever.

LOTS more work to be done, but this is a breakthrough for sure. Thanks for your patience btw, while I'm here being an idiot lmao

[cold laurel]

Heh, no worries! I've been on the other end of this at one point too y'know :P

[frank fjord]

Yeah, but that doesn't mean it can't be a little tiring to get grilled for answers by someone that has no idea what they're doing =u=;;;

Anyhow. For now, I've got a LOT of code cleanup and testing to do.

[cold laurel]

I've genuinely enjoyed it a lot! It was a nice break from my day to day.

[frank fjord]

but based on my preliminary testing I think I'm gonna try and move on from forward extrapolation until I hit a wall that necessitates it (hopefully never)

[frank fjord]

Hah. Okay, glad to hear 🙂

Parachute emoji reminding me of how much Unity drag has traumatized me...

eughhh-

[cold laurel]

xD

[frank fjord]

I have. SO many spreadsheets

trying to model how I might be able to mathematically solve physics equations factoring in drag

it's not pretty. Or useful.

[cold laurel]

I had to do this when making the CCIP targeting for SaccFlight planes. We really wanted to be able to predict where the bombs would land after releasing them, and we wanted drag on the bombs. So down to calculus hell I went.

In the end we ended up with a pretty nice equation

(For a unity rigidbody in freefall with drag and no other forces than gravity.)

[frank fjord]

Yeah, I did eventually derive an equation for acceleration at time T with a drag force and a user force (say, gravity). However, the equation only worked if you started at 0 velocity, I couldn't figure out how to extend the equation to work for any initial velocity.

[cold laurel]

The equation we arrived at supports initial position and velocity!

[frank fjord]

👀

curious to know

[cold laurel]

I can dig it up tomorrow afternoon.

[real timber]

Had to pause the development as I have dozen other projects take priority over it, I'll just start dumping some of the files to the repo for now, I don't know how much I care about torturing myself with this before Udon 2: https://github.com/Varneon/VUdon-Splines

[cold laurel]

Thanks!

[frank fjord]

Udon 2, one day... 🌧️

[frank fjord]

Did we determine whether or not there was a consistent order of operations for OnOwnershipTransfered() and first OnDeserialization(), on a client losing ownership of something?

[frank fjord]

How are you handling snapshot storage and snapshot timestamp storage, when ownership changes? I'm noticing some NaN problems and other things hitching momentarily and I'm sure it has something to do with the fact that the snapshots in your buffer at the moment ownership changes - and ESPECIALLY the buffer of timestamps - are really off the mark when a different player takes over ownership and starts sending data.

[cold laurel]

If you were the owner and loose ownership I fill the snapshot buffer with the current local position, rotation and velocity of the object at the current timestamp.

[cold laurel]

It is now tomorrow afternoon.

[cold laurel]

private static void PredictRigidbodyPhysics(
    Vector3 startPos, Vector3 startVel, Vector3 gravity, 
    float drag, float fixedDeltaTime, int stepsToPredict, 
    out Vector3 endPos, out Vector3 endVel)
{
    float a = 1 - (drag * fixedDeltaTime);
    
    Vector3 constants = ((startVel * drag - gravity) * a + gravity * a) / drag;
    endVel = (Mathf.Pow(a, stepsToPredict - 1) * (constants * drag - gravity * a) + gravity) / drag;
    endPos = fixedDeltaTime * (Mathf.Pow(a, stepsToPredict) * (constants * drag - a * gravity) + gravity * ((a - 1) * stepsToPredict + a) - constants * drag) / ((a - 1) * drag) + startPos;
}

@frank fjord The above function is slightly wrong for some reason, it will accumulate error over time.

But it's the closest I have I think

[dense spade]

Could it be due to the accumulated error over the various math calls you're doing?

It could at least be simplified a bit

[cold laurel]

Yeah it can 100% be simplified

(and optimized)

[cold laurel]

Honestly, might be

I'm having the same issue with a newly derived equation for the future position of a rigidbody with no drag. But my iterative solution is 100% in line with Unity's internal computation.

I'm guessing the floating point issues line up with what Unity is doing when I do it iteratively, but end up different with an analytical approach.

[frank fjord]

Ah. Okay slight difference, I was looking for a function that predicted position at an arbitrary time T in the future, not a single frame

Wait hold on

I see what you're doing

[cold laurel]

This predicts an arbitrary amount of frames in the future

[frank fjord]

I see

I presume your test is comparing a local simulation with local math yeah? Not a networked position with local math?

[cold laurel]

local sim vs local prediction

for one frame

[frank fjord]

Yeah. That's weird that there is any error

[cold laurel]

funny IEEE 754 floating point go brr

[frank fjord]

I know that different clients will have different fixed update rates and that would change the outcome

Also: hate that lmao

[cold laurel]

That's an entirely different headache

[frank fjord]

Yeah. Makes me really grump my because it means even if you never code anything in fixedupdate, you still have to reconcile that physics will sim differently on different clients

Even more so than it already would (thanks non deterministic physics)

But like. Drag with no collisions

SHOULD be mathematically consistent. But it's a stepwise function that is frame dependent.

[cold laurel]

right?

[frank fjord]

Incidentally I was watching a talk by the same person who did the continuity of splines video

About writing an alternative to lerp "smoothing" (a = lerp(a, b, ratio * deltatime) )

That isn't frame rate dependent

I wonder if that same principle could be applied to the way physx drag worked...

Since they are similar in concept

not that any of us are gonna """patch""" Unity's PhysX implementation

We'd have to take over responsibility for the drag part of the physics loop ourselves

By the way. The drag equation: NewVelocity = Velocity * (1 - Drag * FixedDeltaTime)

I'm actually not sure that's really correct...!

I did some testing in Unity and discovered some interesting things about the effect of mass, force, and drag changes, on 1) how long it took something to reach top speed, and 2) what top speed actually was

but when I compared those to "rules" to just simulating out a literal frame-by-frame acceleration due to force with a drag application afterwards...

I found that my top speed actually varied from the rules I'd experimentally observed

and not just by a little

Looking back at my simulation spreadsheet I'm not sure I didn't do something wrong so take this as a hunch and not confirmed science

I will relay though what I DID observe and write down from testing forces with real rigidbodies in Unity:

For a force of 1, mass of 1, drag of 1: initial acceleration on the first frame is 1 m/s^2, and the "estimated" top speed approaches 1 m/s (but never quite reaches it because drag is applied after forces are applied, so 1 iteration of the drag equation is removed from the real observable top speed)

Doubling force doubles top speed and initial acceleration

Doubling mass halves top speed and initial acceleration

Doubling drag halves top speed but does not interact with initial acceleration

and drag is the ONLY variable that affects the time it takes to reach top speed: doubling drag halves the time it takes to reach any given top speed ratio

(these are nice numeric estimates, NOT precise mathematic calculations)

when I say "doubling" and "halving," these are ratios that are extensible to any other ratio of change between those variables

[pallid shore]

I don't need smooth transitions for owner transfers so the first snapshot the new owner sends is flagged as a teleportation snapshot which does not interpolate. The old owner adds a new snapshot with the current position, rotation, and velocity in the buffer. If you want a smooth transition, you could flag the snapshot as first from a new owner. Then you could smoothly transition from the currents snapshot to the flagged one and discard all other snapshots that are in the buffer from before the owner change.

[frank fjord]

This flagged snapshot: is the flag being sent as a synced variable? Or is this flagging process done locally? Because it seems like there may be an inconsistency in the order that OnOwnershipTransferred() and the first OnDeserialization() from a new owner run in that frame

[pallid shore]

I send one byte for extra data with every snapshot

[frank fjord]

Okay, so the flagging is done by the new owner and sent as part of the serialization

How does that work for late joiners, though?

A late joiner initially presumes they are the owner of everything until OnOwnershipTransferred runs and they get the real owner of everything

[cold laurel]

late joiners will probably "see" the vehicles interpolating from 0,0,0 to the newest snapshot

[frank fjord]

but that would imply they do not necessarily get the single snapshot that flagged the snapshot as teleport/new-owner

[cold laurel]

I'm guessing this happens so fast that no one notices

[pallid shore]

The first snapshot you receive is always handled as a teleportation snapshot

[frank fjord]

Gotcha

[frank fjord]

When you say the "old owner" adds a new snapshot with the current state, are you implying that they are adding a snapshot of the last state THEY knew of, in ADDITION to the state they receive from the new owner that's "flagged" as teleportation? Or are we just adding this newly received snapshot?

[pallid shore]

Yes, as soon as the owner transfers they add the last state they know to the buffer. The snapshot from the new owner will take time to arrive

[frank fjord]

Huh, will it? If you take ownership and request serialization in the same frame, it SEEMS experimentally like those always get sent at the same time, and get received in the same frame. Am I mistaken?

[pallid shore]

I can't say for sure, and I don't know if it's documented somewhere but as far as I know ownership transfers are prioritized so it's not guaranteed that the deserialization happens in the same frame.

[frank fjord]

ough

well, my main concern would be it happening in the opposite direction

if the expected flow is 1) detect ownership change (would be through OnOwnershipTransferred), make a snapshot as a tide-over, 2) get new data (would be through OnDeserialization running), run teleport snapshot...

if for some reason OnDeserialization runs first in the same frame - god forbid, but I am wondering if that's possible somehow - then the teleport snapshot is added BEFORE the local snapshot is added

[pallid shore]

I haven't tested it myself, but I never had problems with it. I think it would be good to test the behavior of ownership transfer and serialization/ deserialization under different network conditions.

[frank fjord]

Yeah, I should set up a testbed just for that honestly...

[frank fjord]

... o h

This... huh.

Time is unscaled-time-as-double

I'm setting a flag when either OnOwnershipTransferred runs, OR OnDeserialization runs...

and then during LateUpdate, I reset the flag if it was true. That's what "events ran this frame" is indicating

I should test this with a LOT more people than just myself, but it's pretty telling that even with 2 local clients, OnOwnershipTransferred runs at least 1 entire frame before OnDeserialization runs...

[frank fjord]

Ran with 3 clients: one just watching (log above) and 2 bouncing ownership back and forth

When a serialization is sent, I logged the time between requesting ownership (also requesting serialization) and PreSerialization running, and synced that value as the serialization

On the recipient clients, I logged when ownership was recorded to have changed, and calculated how long between then and OnDeserialization running

I also compared it, out of curiosity, to receiveTime-sendTime (which I call half-ping here)

It seems like ownership change network events arrive NOTABLY faster than actual network serializations

In my test above, all recorded ownership change events are arriving so fast the sending player couldn't even have sent their actual data serialization yet (delay between RequestSerialization() and PreSerialization())

That's. Wild. And kind of disturbing lmao

@cold laurel @pallid shore in case you're interested in network timing. This shocked me lmao

Also, looking at the last test on this log... The values here suggest that, barring some error in time measurements (which is likely, admittedly), the player taking ownership would have sent out the ownership request INSTANTANEOUSLY on the very frame they called the function...

None of the other tests reflect this margin, but it's still weirdly fast.

All the other tests reflect that either or both of the following are happening:

1. The player requesting ownership sends out the request sometime BETWEEN Networking.SetOwner() and OnPreSerialization, and/or
2. The half-ping/network delay for sending an ownership change is some amount smaller than it is for variable sync serializations.

[frank fjord]

Man. This delay between ownership change and sync variable deserialization is a huge unconsidered problem space for me. I'm 90% sure this is where basically all my current networking bugs are coming from lmfao.

[obtuse echo]

Somewhat related question:

I know you can set ownership for somebody else, but can you also set and sync data during all of that?

So instead of setting the ownership for the local player, I want to set the ownership of somebody else while syncing some new values at the same time. I assume that's a no-no because I've been having weird behaviors.

[frank fjord]

Yeah that wouldn't behave well

because of above, it's likely, if not guaranteed, that everyone will receive the ownership change before the serialization, and then reject the serialization that comes a few frames later because it's coming from someone who no longer owns the object

or the person who is sending the information may just abort the operation themselves once PreSerialization occurs

because they already know they don't own the object anymore

[obtuse echo]

Yeah, I switched to an intermediary sync to notify the player who actually will set the ownership and it works fine now. I wish it worked similar to how when you set the ownership to the local client.

[cold laurel]

Oh?

How would the ownership changing before the data arrives be problematic?

Isn't this what we should expect?

[frank fjord]

Specifically how much earlier it happens

[cold laurel]

fair

[frank fjord]

Anyhow. After significant fumbling, I've finally managed to get the snapshot interp working very smoothly and the only hitch during ownership change is a singular teleport frame from the current state to the new state. No jittering or temporary rubber banding or teleporting to the void...

I'm quite pleased. Just wish I'd figured out how to make it this consistent like a day ago

My friend is like "so how do you think you'll fix the one frame teleport?" And I'm like. Are you kidding? This is perfect compared to how this has been working lol.

[cold laurel]

💀

[frank fjord]

I wonder if there is a good place to do or solicit code reviews

Mine certainly could be better. It works though.

[coral ingot]

whats the most efficient way for me t network a players input events

[frank fjord]

What do you mean? That's pretty broad.

[coral ingot]

im making a car and am trying to network it now. I tried to put a vrc object sync but it makes it glitch around for the remote player. I was thginking asbout networking the players inputs but if theres a better way please let me know

[frank fjord]

Networking something with complicated movement is, turns out, really complicated. Feel free to read all the conversation from the last week or so to see what I mean x.x

I suspect networking a player's inputs would not be a good way of going about it. Typically that networking approach is called "deterministic lockstep" and requires a pretty precise agreement on how long things take, exactly when events happen, etc. And a deterministic physics engine... which Unity does not have.

You would inevitably get drift and error, and the local and remote clients would desynchronize.

I would explore solutions where you focus on synchronizing the state of your system - things like position and rotation, maybe velocity and angular velocity - and then finding ways to smooth that out. We had a big conversation about "snapshot interpolation" the other day, which is what I wound up implementing for my project.

... It's not easy lmao. If you really want to network something with complex movement, strap in, it's not a drag-and-drop solution 😩

https://gafferongames.com/post/snapshot_interpolation/

[coral ingot]

thank you

[timber ferry]

have you tried MMMaleon’s Smart Object Sync? that might work better

[hoary frost]

I believe smart object sync is depreciated. Use lightsync from the same author instead

[timber ferry]

ah okay, so that’s why they have light sync

well that wouldn’t work for a car then

[coral ingot]

I was able to make it work with a normal one. Thank you tho

[silk arch]

Newbie question, is there an easy way to know that a networked object is "ready" when a player has finished joining an instance?

[heavy spindle]

bool Networking.IsNetworkSettled

[wintry wharf]

i have a gameobject which get spawned by an VRCObjectPool.
This spawned gameobject have also a pool for 2 different effects how can i enable them (they get played on awake)
and how do i change the animation?

when somebody pick it up ill do this:

    public override void OnPickup()
    {
        SetNetworkOwner(GetLocalPlayer(), gameObject); // gameObject is the spawnedObject
        SetNetworkOwner(GetLocalPlayer(), pool.gameObject); 
        GetComponent<VRC_Pickup>().DisallowTheft = true;
        GetComponent<SphereCollider>().enabled = false;
        animator.Play("Launch");
        state = 1;
    }

@humble girder maybe u can help me out again 😄

[young thunder]

I'm keeping track of a list of "roles" for each user in a DataDictionary. Right now, I'm thinking of having the instance master keep the one-and-only copy of this dictionary, so that I don't have to synchronize its contents.

So, when a user needs to do something that's permissioned, they'll send a networked event to the owner of an object (which will be the instance master), and then the owner will acknowledge that request and cause something to happen in the world (like teleporting the requesting player to a new place)

When the instance master leaves, does their data get inherited by the new instance master, or am I going to lose that data?

The alternative would be to synchronize the entire dictionary over the network whenever it gets changed (which will be very infrequent!)

I suppose that's not the worst idea in the world

[young thunder]

Does RequestSerialization not work in ClientSim? I'm not getting an OnPreSerialization event.

(including with some remote players added, and when the local player is neither instance owner nor master)

[tulip sphinx]

it doesnt

[young thunder]

alright, thanks! I will test it out in-game instead

[teal pebble]

This might be a bit of a nooby question but just so i understand how Udon and sync works, VRchat hosts the room on their own servers but just appoints ownership to whatever instance a player created ,but the game itself is NOT peer to peer where someone is assigned the role of host/owner over a given cell or world?

ergo being a traditional server-client architecture for me to think about how to design my creation?

[meager meadow]

Udon works closer to peer-to-peer. VRChat hosts the files for the world of course. It takes care of synchronizing avatars. Udon synchronization works differently. Although the clients/headsets don't talk to each other directly, the VRChat servers in general just pass messages between the clients. The one case I know of where that isn't quite true is when a new user joins -- the VRChat servers send the current states of all synced variables to the new client without involving the others.

[fleet arch]

p2p via photon relay

master assigned to someone in lobby

no server side owner as such

apart from reassigning host

[obsidian hill]

so whoever the host is, holds the variable sync? is there a way to track that?

also, if I were to set a script running for someone who pressed a button to start a game, and used [UdonBehaviourSyncMode(BehaviourSyncMode.Manual)] as an attribute, would the rest of the players see the game start too?

[strange gyro]

If you set sync mode to manual it just means you have to call RequestSerialization on the script for any synced variable changes to be sent to other players

If your gameobject or component isn't enabled on the other clients they won't just enable automatically, you would have to either send a custom network event that turns on the script, or have another script that it already running on other clients that can receive synced variable changes to determine whether the script should be turned on

[obsidian hill]

alright sounds good.. Guess I'll work on it

[cold laurel]

That's not correct

[cold laurel]

The "Owner" of any particular object is the one in charge of what the value should be. But the latest synced value is cached on the server so the server can hand that to new joiners.

[hybrid kindle]

im so confused. Can someone tell me where in this script its telling the toggle to be global? I want it to be local

[cedar crescent]

I think the problem is because OnPlayerTriggerEnter event happens for any player, not just yourself. So even though the script isn't syncing anything, it's still detecting other players.

You could take the player from the OnPlayerTriggerEnter/Exit nodes and add a branch that checks if VRCPlayerApi.isLocal is true.

[hybrid kindle]

will test it later thank you

[hoary frost]

that is correct

[obsidian hill]

okay i figured so cause you gotta generate the network keys

[stone badger]

There was a discussion here about sync'ing of values earlier. I didn't quite see the problem as things work for me but I've now encountered a situation that (for me) demonstrates one limitation. If we want to assign a pooled object to a player I see 2 ways it can be done, "on join" and "on first use". I thought "first use" would be simpler but it turns out to have the same inherent issue. OnPlayerJoined is called before the sync'd values have been sync'd. That is a problem if an action is required that depends upon some shared data.

In my case I've written my own object pool code. It works perfectly fine and is understandable but I was getting tripped up by the first use logic. The first use was happening prior to OnPlayerJoined and prior to the sync'ing of objects. This meant that the player could not see the latest values in the pooled items. They saw the default values. This resulted in the wrong item being given to them. In actuality it means that every new player is assigned the second item in the pool because they can only see that the first one was handed out to the "master".

The pool items are used for in-world logging (not that it matters but it illustrates the issue). So as long as no attempt to log was made during the earliest starts up steps it works fine. I was logging an access to a file made via a URL however. That made first use occur well before the sync'ing of the pool objects which is exactly why it doesn't work.

I thought about not permitting use until after the initial syncs have occurred but there is no event we can tap to determine whether this has occurred or not. I'm obviously not going to use a timer so I'm looking for alternatives. Would the VRC Object Pool have the same issue do you think? Might there be addition code wrapped around it to handle this? I'm a little skeptical. I've looked at the code in CyanPlayerObjectPool and it may be handling it but I'm reluctant to switch systems to one I don't really understand. Particularly if it doesn't solve the issue.

Thinking about it if we had the RPC calls with parameters it should solve everything. Don't have that though.

[meager meadow]

I can't remember the exact problems I had, but I couldn't get VRCObjectPool to work reliably for giving players a unique object from the pool. It was probably related to the ownership race condition on the pool itself. That's how I ended up with CyanPlayerObjectPool. I haven't had any problems that I traced to that one yet, but I do feel like it's a bit of a crap shoot since I haven't looked closely at the code. Are the problems you are having repeatable? I am pretty familiar with CPOP at this point and if you can come up with a small test case I can try implementing it (it may take a few days before I can get around to it though). I could tell at least if it reliably fails there too.

The developer is active on github. You might try to contact him there to see if he would share insights on what he is doing internally.

[stone badger]

Hi there. I hoped you would chime in 🙂 This is my deepest dive yet into pooling. I assumed a bit too much with my implementation. And to be certain it does work in the "happy path". The problem is very much repeatable and the VRC logs explain why. It has to do with the order that world is "brought into existence" for a new player. Sync'ing of objects is done quite late, just before a VRC log entry explaining that world downloading is complete. We just don't have access to that event.

Some have suggested I assign the object in OnPlayerJoined which seemed sensible (to me as well) but as can be seen that is still too early. Logging the contents of the object pool it is plain to see that the player has not yet gotten the updated values.

So I settled on a first use scenario. Figured that if a player didn't need the object there was no real benefit to assigning it. This works fine too so long as there is attempt to use the object before all the initial syncs have completed. As I mentioned I was logging an API call and that happened quite early one. That tripped the need for the object to be assigned out of the pool and that's how everybody gets assigned the second object.

I turned off the logging in the API call and it works again. On the other hand those API calls are made at various times so it really is only this first time I can't have it log. Of course I can add some Booleans to the service to mark if the user has called it before, etc. but that's a hack. One that I may need to implement. With some thought perhaps it can be a little less hacky.

Re: CPOP. I don't doubt that it works but in my use cases it seems like overkill. I've looked the code over and it can retry, postpone and keeps track of data changes. It also seems to require double the number of objects as the player max. These are base implementation behaviors that I mostly do not need. It also syncs an array of integer values (if I recall) that corresponds to the position in the pool. I'm just sync'ing data in the objects like I would any other object. With one owner there can't really be any issues. The pool itself never syncs or needs to.

I would guess that it would fail under the same circumstances given at start up we have to rely on the VRC world startup sequence.

[stone badger]

@meager meadow just traded a few messages with Merlin about this. He suggested and I agree that it is probably best to make all the requests via the master player. It would know what the current status is. I'm going to try to hook this into what I've got.

[meager meadow]

I was about to say that just leads to the problem of now knowing who the requestor is so there is no way to know who to give the item to, but then I realized in this case since you are giving an item to everyone as they enter, that should actually work! That's the answer people give for the more general case of VRCObjectPool too, but the no-RPC-with-a-parameter limitation strikes there.

[stone badger]

Yes I looked around to find a way to get the Udon class associated with the master but there is no way simple way to get that. Thought about messaging but that only make the problem worse so... OnPlayerJoined the master can affect the assignment and everything should work.

Oh if we could only have RPCs with parameters (dream on).

[warped linden]

Is System.Collections supported by Udon?

The library.

[strange gyro]

Not yet

Only arrays and VRC's own flavor of Dictionary and List

Udon 2 should support more types

[warped linden]

Is there any workaround I can do to use it? I don’t exactly want to remake any scripts or anything.

[strange gyro]

Not to use those types directly, you'd probably want to write wrapper methods to abstract away implementation details so it's easier to switch over to a different type later

[stone badger]

A couple more false starts and I've just hit upon something that I could use. I create an isSettled state by counting the number of calls to Update(). I wouldn't tend to use it to determine where in the startup it is (it differs slightly between master and players) but in both cases if the number of calls is > 5 then all the startup stuff is complete. This means that OnPlayerJoined has been called but more importantly that the player has been sync'd. So that is the moment I can effectively set sync'd values and know they will be processed and no overwritten during the startup cycle.

It would be very useful if the various start up sequences were exposed as events (docs might have to explain what could reliably be processed at these times) but it isn't hard to create code that will "run" but not "work" if it is called too early in the cycle. I would set sync'd attributes in order to log simple "I've done this" type messages and while I could see they got set they were immediately nullified when the pending syncs were applied. In sort I could see in my log that the player set them and then that the startup sequence ignored it as the old values in the sync'd object were logged.

The good thing is (I hope) this gives me the perfect spot (and easy to monitor) to make start up calls at the end of the process.

[hearty vale]

n U#, how can I detect for a specific player ID to create buttons only my account can access?

[strange gyro]

The best you can get is checking the display name of a player with VRCPlayerApi.displayName:

if(Networking.LocalPlayer.displayName == "Chdata")
{
// Do something specific for your account
}

There's no persistent and unique IDs exposed for players

[hearty vale]

is this alone what people are building patreon and admin-command stuff off of? lol

[strange gyro]

Pretty much, you could implement some extra security if you have your own web server and send a secret code as a response or something based on the IP address of the incoming web request

If someone's determined enough it's all defeatable anyway

[hearty vale]

oh well, what I'm making does not need so much security really

I have --enable-udon-debug-logging in my vrchat launch options. But where does debug.log print to when I actually open my game not in test mode?

[strange gyro]

Right-Shift + ` (backtick) + 3

Should bring up a debug window in-game

Although you also need --enable-debug-gui to open the window

Otherwise it'll be in the logs under %APPDATA%\..\LocalLow\VRChat\VRChat

[mental furnace]

Is there a way to verify/ensure that a RequestSerialization actually made it to other players? In this test case I basically have a variable I sync with the player's ID, and I attempt to RequestSerialization until OnPostSerialization() indicates it was sent. The logs of the other test clients and their counters both show that the serialization didn't actually even make it to them.

Also confused cause when the serialization does actually make it to other clients, not all of them even receive it? The counters are all mismatched on every client.

[tulip sphinx]

how often you send stuff?

[mental furnace]

Whenever Random.value < 0.01f in Update(), but I'm also never seeing IsClogged true or IsNetworkSettled false

Again, it's a bit of a crap test case I know, but these results are bothering and I'm just hoping I'm doing something wrong

[tulip sphinx]

it still wont go above few hz afaik. so add locally timer that checks if 0.2s passed since the last Request Serialization and should be way closer

my guess at least

[stone badger]

As you might be aware it isn't a method call in the sense of it completing and having the same value available everywhere. It is part of the reason it is named "Request". First are you using the same sync'd properties for more than one player? If so there is competition for ownership.

[mental furnace]

I already set it up so that ownership is requested before it does RequestSerialization, and ownership transfers are denied if there is a pending sync until OnPostSerialization is called

[stone badger]

I can't be of much help in such a case. I know there can be issues with sync'ing in general and the more one mucks with the pattern the more likely something could go amiss. It isn't "broken" per se but rather not behaving as we suspect it should and want it to.

[stone badger]

Quick question about assigning object from a pool. I've read in a number of places that it is better to have only the master player assign objects from the pool. So is there an example of that somewhere? I can do it but it is far from straightforward as there is no way to pass a parameter in the network call. The master doesn't readily know who to assign the pool object to. Sync'd variables can't be combined with the network event. Such a solution would also tie up the object used to communicate with the master player. Is anyone actually doing this?

[strange gyro]

That's the simplest way to minimize sync state conflicts since you don't have to worry about ownership of the "manager" gameobject changing constantly

Although they did apparently change the behaviour a bit for Quest users so it can transfer master to another player when the app is suspended so it could still happen more regularly there

Still simpler to deal with than the alternatives though

https://github.com/CyanLaser/CyanPlayerObjectPool uses a master based approach and they do some extra validation when the master changes to deal with conflicts

[stone badger]

On player join, master will assign this player a free object and on player leave, master will remove assignment of this object.

That isn't hard to write. I would rather it wasn't limited to on player joined.

[strange gyro]

You could use the ownership transfer events to notify the master who is attempting to take or release ownership then have the master set a bitflag in the manager's synced behaviour to indicate that pooled object's availability status, then when a client wants to get a pooled object they check the flags in the manager first before taking ownership of a free object

[stone badger]

is anyone actually doing this 🙂

Lots of "coulds" circulating.

[strange gyro]

I don't use synced pools in my projects but a master based approach is how I'd implement one

Udon networking is pretty unreliable when you've got a bunch of players taking ownership of objects that control global state

[stone badger]

I had hoped to solicit replies and solutions from people who have done it and are using their system reliably. If it has limitations that can be noted and as a group perhaps we can a) test it, b) verify the conditions, c) provide ways around them.

[obsidian hill]

I'd join that group. Probably have little to add, but I might learn something hah

[gleaming hound]

hi guys, id like to ask if this is the right way to network triggers?

im trying my best to wrap around udon graphs and stuff so id really appreciate some help!

[tulip sphinx]

you cannot mix stuff between different flows, your api will bever be valid, its not passed

also i dont get why you trying to network it then check if player is local

also you dont check if player is local before sending network event so everyone will send one, thatd be a mess

[hardy shuttle]

how do I make a timer string to syncronize?

I already made the code but for some reason it's not syncronizing to other players

or it resets once a new player joins or it doesn't syncronize at all

[tulip sphinx]

show what you got atm

[hardy shuttle]

well, later I noticed that the problem wasn't the timer but it's the int value of another script that is not syncing when a new player joins, how do I sync int variables of the master to local players? (Udon graph)

[tulip sphinx]

either make it continuous sync or better use request serialization when owner changes the value, and apply it on ondeserialization which will happen for both current and late joiners asap

[hoary frost]

Im linking this here since its networking relevant #udon-general message

[lone zealot]

The serializer cant handle null strings. This usually occurs when using a string[] that hasnt been intialized to empty strings.

[hoary frost]

oh thats odd, im not serializing any strings on this script

only ints and bools

[hoary frost]

update: This seems to fix itself if another player takes ownership of the script. If the original owner of the script attempts to serialize anything, it wont work

odd

[hoary frost]

after building and running my entire world like 20 different iterations trying to figure out whats causing it, i think i finally found whats causing it. It was this thing here that executes on start/onplayerjoin
Might report back later once ive truly understood why

BloodClone here is a gameobject im cloning/instanciating when a condition is met, so this script here check if it exists, then syncs its drycounter to all players according to the owner

seems like despite being null, it still gets past the isvalid node. hmmmmm

[hoary frost]

Ok seems like i needed to check if the drycounter variable existed as well via Isvalid for it to properly filter it out

[coral ingot]

how do i network a vrcurl?

    public MidiWebLoader webloader;
    public VRCUrlInputField urlin;
    [UdonSynced]
    public VRCUrl URLNEXT;
    public void UpdateGlobal()
    {
        webloader.SendURL(URLNEXT);
    }

    public void URlUpdate()
    {
        Networking.SetOwner(Networking.LocalPlayer, gameObject);
        URLNEXT = urlin.GetUrl();
        SendCustomEvent(nameof(UpdateGlobal));
    }

[lusty pagoda]

what you mean network

just use ondeserialization

[coral ingot]

i want to send the vrcurl inputed by one user to all users

[lusty pagoda]

after setting it do requestserialization() ondeserialization() and continue your code there

[strange gyro]

    public void URlUpdate()
    {
        Networking.SetOwner(Networking.LocalPlayer, gameObject);
        URLNEXT = urlin.GetUrl();
        UpdateGlobal();
        RequestSerialization();
    }
    public override void OnDeserialization() => UpdateGlobal();

[coral ingot]

this worked. tysm

[stone badger]

Real quick note but as general rule I will suggest that you call RequestSerialization() first and call UpdateGlobal() second. Probably won't make a difference in many cases but consistency helps reduce hard to track bugs in your code.

[strange gyro]

Yep it shouldn't matter because OnPreSerialization won't happen until after everything else is done

Most important thing is making sure you have ownership before making changes

[meager quiver]

Heyo, just wanna ask about some behavior things in case they're already known. On player suspend, if you try to perform an ownership transfer from the side of the player who is suspending, will the transfer complete, or is this more reliable to be handled via other clients snatching ownership away?

[meager quiver]

Also, if a player suspends, another player leaves, and that player unsuspends, do they get old OnPlayerLeft events?

[prisma lance]

is it possible to sync positions and rotations manually instead of using an objectsync ?

[tulip sphinx]

@prisma lance ofc, thats just vector3+quaternion (or another vector3)

[prisma lance]

so there's no way around having to also update that variable every time I change the position

[tulip sphinx]

if its bind to player use plauer positions

[prisma lance]

it's 136 world objects

[tulip sphinx]

and?

[prisma lance]

it's not bound to players

I'm slightly concerned that having that many object syncs might cause issues

with multiple instances of the prefab that'd be thousands of continuously synced objects

[tulip sphinx]

i tried 150 pickups, not great but not clogged. try this https://github.com/MMMaellon/LightSync

nah, thousands of continuous sync is not possible. youd need smth else and slower.

[prisma lance]

I'll look into how lightsync works thank you

well looks like that won't work because they force a rigidbody on it

I'll just do it myself

[timber ferry]

i heard they only forced a rigidbody because it would cause a null reference error, but it should still theoretically work

so maybe you could try modifying it to not require a rigidbody

[prisma lance]

this is why you use null checks instead but I'd rather just write my own than have to constantly update the script

[timber ferry]

yeah i mean, if you think you can do it yourself go for it

i just figured i’d mention that

[obsidian hill]

if you're gonna be networking a lot of objects, I recommend CyanTrigger. it makes your life a lot easier in a lot of ways

[tulip sphinx]

[obsidian hill]

because i am lazy

not because it's more useful

[hearty vale]

If I make an udon script with an UdonSync'd variable "IsEnabled"

And then I copy that U# script onto two gameobjects

Does this create one sync'd "IsEnabled" variable or two separately sync'd "IsEnabled" copies of the variable

And can a second U# script set FirstScript.IsEnabled and it will be sync'd if I RequestSerialization() in the second script, or do I have to call requestserialization in the first script? Can I call FirstScript.RequestSerialization() within the second script?

Or do I need to make a function in FirstScript that calls its own RequestSerialization?

I'm trying to figure out how I should sync the on/off state of one game object, between two separate buttons that can toggle it...

[strange token]

Each instance of a script has its own set of variables, so two objects with their own versions of the same script would be working with their own variables, even synced versions

And for the second question yes that totally works, you can request serialization for another script exactly that way! 👌

[hearty vale]

🫰 🦿 🦐 🦿

[hoary frost]

as shrimple as that

[fringe creek]

When testing locally with two clients, does networking work as expected? I have an object with VRC Pickup and Object sync but only one of the clients can pick up and move around the objects. The other client doesn't even see the object get moved and cannot pick it up.

[fringe creek]

Ah, looks like my problem is that network instantiation isn't a thing: https://udonsharp.docs.vrchat.com/networking-tips-&-tricks/

[fringe creek]

Yep, using object pooling seems to have fixed the issue

[hoary frost]

Yep, you cant instanciate anything that would use a network id

[gloomy gyro]

heyyyyyyyyyyyyy itssssssssssss meeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeee (again)
I have once again tried to make network synced doors (or in this video just doors in general) and it once again just agressivlely spins, even the things rthat shouldnt be affekted (more mesh and the activiation cube, gave up with having the door itself as the handle). So uhm, does anyone have any specific idea to why my door is going mad?

thats the code on the cube thats also spinning, followed the tutorial i found

thats the door how its supposed to be (its supposed to well go down with the animation

[jaunty fjord]

There doesn't seem to be anything in the code that would make the door spin at all. Does it spin if you disable the Udon Script component?

[gloomy gyro]

ngl I have no idea what changed but when i press play now its just stuck in the wall

the button kindof works, it disables the object, supposed to change animation states

[jaunty fjord]

Is the picture showing the door going into the wall rather than covering the top of the hole?

[gloomy gyro]

yep

is like that now upon entering gamemode

[jaunty fjord]

Does it do that if the Udon Graph is disabled? Worth trying that first to see if it's a code issue or a problem from elsewhere

[gloomy gyro]

disabeling just pushes it into a wall diffrently

so guess its not the network

[jaunty fjord]

Are there any other things that access or modify the door? Maybe you could send a picture of what's on the door gameobject?

It's hard to say for certain what could be causing that cause I can't see what else is happening in the scene

[gloomy gyro]

thats it

[jaunty fjord]

Could be the animator, what's happening in the animator? Maybe the animation for moving it is playing repeatedly?

[gloomy gyro]

its the open and close state, one one frame where its supposed to be open, one one where its closed

thing is that has worked before even with 2 flaps in the active build and works for all the other doors

but i felt smart and went uhh sync doors and broke it for the third time

but yeah its just 2 rotation and mesh collider activiation points

thats the animator, how it has worked before, nothing changed here

sorry if i sound mean, dont mean to just want to be a bit saracastic against udon

[jaunty fjord]

It's fine, I didn't get any mean vibes from what you said, I'm just trying to think what could be causing this problem

[shrewd widget]

Hey guys. I have a script that a gameObject follows the player head but I don't know why I can't make it possible to let everyone see each other's gameObject on their head and it is always working on local even if I use the sendCustomEventNetwork

[restive cliff]

You don't have to network the position of these but the code will be similar if you have to.
For example, to get all the players you want to have a GameObject to follow, each player will get their own GameObject.
Next, loop through the players and set the position of the GameObjects. This is the non-networked method.
For the networking I highly recommend a pooled solution, there's the native vrc pool feature and there's also CyanLaser's player object pool on github. Every player will get a networked GameObject, and you can update the position here. Beware that this method can clog up the network depending on the network rate of the followers.

[fringe creek]

I want to make it so that if a player's item (owned by that player) collides with a "power up" (owned by the instance owner), the power up can notify the player item to apply that power up. My first thought (since the items are owned by different instances) would've been to use a network event, but because those don't have arguments for some reason I can't determine how to avoid applying that power up to every other player as well. Any ideas?

Ah, turns out you can send custom network events directly to the owner of an object and nobody else, so I'll just use the power up to send an event to the player's item. Shame you can't use objects tho, means I need to create like a dozen custom events, one for each power up

[vapid pagoda]

This may solve your problem (and future ones) https://github.com/Miner28/NetworkedEventCaller

[fringe creek]

That's super cool, thanks!

[hoary frost]

How is this different than regular network events? Smooth brain doesnt get it

[vapid pagoda]

you can use parameters (and also limit events to specific groups of people)

[hoary frost]

Aaah i see, thats very useful actually.
To limit them to specific people, cant you serialize the concerned player IDs/give those players tags, then send the event to everyone and check for those on arrival?

[vapid pagoda]

mhm! thats how it works

[hoary frost]

oooh is that how the package you sent works? what i was explaining is how i always did it haha

[vapid pagoda]

honestly i wish udon could let us do that built in, rather than sending data to every player and having to compare if the information is useful to us locally :c

[hoary frost]

yeahh, hopefully udon 2

[harsh pebble]

Legit, I hate working around the fact I can't send parameters with vanilla events. This might become a favorite of mine if it supports that

[strange gyro]

If you only need local events I made a small mod to the U# compiler with some helper methods to support parameters with SendCustomEvent - https://gist.github.com/NGenesis/9729ad29bfff576a9638593e852073b1

[dusky reef]

I'm having a bit of trouble with a script ive made and im unsure of how to make it sync with other people. I have the script duplicate a game object and spawn it in a position on the map. How do I make it sync properly?

[jaunty fjord]

If you're duplicating with Instantiate, the new object does not have a network ID so can't be sync'd using VRC Object sync or anything like that

[dusky reef]

damn.

[dusky reef]

is it possible when duplicating to add a network id to it?

(i dont know how the networking thing works btw)

[meager meadow]

No. The typical solution is to have a pool of disabled, networked objects and enable them as you need them. Look at VRCObjectPool.

[strange token]

The solution to support item instantiation is fairly complex and not very many devs do it, but it is somewhat possible. I gotta figure out how to benchmark when it’s worth doing, tho

Generally if you only need a few interactable items, then pooling one for each player (up to your world’s limit) means you sit at a couple hundred objects and that’s generally fine, but if you’ve got 10+ item options and plan to support 20+ players, pre-spawning that many objects might be worse for your overall world frame rate, and therefore the solution of instantiation might be worth it

The only way to get instantiation going that I’ve seen so far is:

1. Set up a disabled list of one of each of your item types
2. When a player spawns an item, network into an “inventory” array that references the item’s index in the array in step 1
3. All other players spawn a local clone of the object index from that list, then update it with networked values received on the player object

But that requires a new synced value on your player object for every possible thing a held item needs to sync

So things tend to get very hefty very quick

All of this information could definitely be compiled into a YouTube tutorial, but it should definitely include a “when to do this” explanation that needs to be based off of how many objects (even disabled) can sit in a VRC world before it starts to break down frame rate and stuff

[meager meadow]

My solution in that case is somewhat different. I have object pools (currently VRCObjectPool but I plan to change that in the future due to its limitations) that contain empty gameobjects. I set up items Zerithax described in step 1 (I call them "templates"). For step 2, I pull an empty gameobject out of the pool and sync the array index she mentions. For step 3, everyone then copies the mesh/material from the template object onto their empty gameobject. The empty gameobjects must contain all the data that might need to be synchronized by any of the template objects, but this method prevents the full array from needing to be synchronized every time someone one is modified.

[fleet arch]

Or just use cyanpool

[meager meadow]

CyanPlayerObjectPool? That gives each player a specific item. Is there a more generic one?

[fleet arch]

Ye that's the one

[meager meadow]

Cool. I use that one a lot, but they are talking about a different problem.

[finite sierra]

does it actually work? and how is the performance network wise?

[finite sierra]

Also does anyone know how many synced objects vrchat can handle?

[fleet arch]

It's more, how many can you update at the same time

[vapid pagoda]

Afaik it compacts events as much as possible using bit shifting, it adds overhead on the cpu but reduces bandwidth usage, at least thats how i remember it working i could be wrong

[meager meadow]

To follow up on what the others have said, the limits are related to network data rates rather than number of objects: https://creators.vrchat.com/worlds/udon/networking/network-details/. There was someone on this discord testing how much overhead a single sync had, but as I recall his results were that it is not straightforward. You should be able to find his results by searching here.

[dusky reef]

I have another question about the cloning object thing. Is it possible to send a custom even through server to everyones game for it to clone itself for everyone then have a script that activily updates its position and rotation for everyone else when the object owner interacts with it?

[meager meadow]

There is no built-in way to pass parameters via custom events, so unfortunately no. There is a library some people recommend that is designed to pass parameters in events, but I haven't used it and don't know anyone who has so don't know if it is reliable.

[dusky reef]

ah ok

well im looking into base unity functions

im gonna try out some stuff

[meager meadow]

Yes, that's a rude awakening when you start using UdonSharp 😦

[dusky reef]

https://docs.unity3d.com/530/Documentation/ScriptReference/Networking.NetworkIdentity.html
Not sure if this will work with udon sharp

it is calling unity for use soooo

if this works then yay?

[meager meadow]

Let us know if it does, please!

[dusky reef]

if it does ill make a package for cloning scripts

cuz this will make vending machines better

:>

[meager meadow]

I kind of doubt it will work to be honest, but would love to be proven wrong.

[dusky reef]

as long as i dont get nulled by udon im good

if you want you can read more into the page

[meager meadow]

VRChat uses Photon for communications. I don't know if that would work.

[dusky reef]

on another topic how do i report a user for illegal activity. Just wanna know

[meager meadow]

That one I don't know. Fortunately I haven't had to deal with that so far.

[dusky reef]

good

cuz uhhh im still tramatized :>

[meager meadow]

Stealing other people's work was rampant on Altspace, so I know what you mean.

[dusky reef]

oh no its not stealing its just uhhhh

videos of a type

[meager meadow]

Ah OK.

[dusky reef]

enough for prison time 💀

prob just gonna call the cops and submit their user from their

not saying vrc isnt gonna do anything just rather hand it right over

[meager meadow]

I haven't run into that, but can image it can happen if you visit a lot of worlds. There is the safety setting for URLs that should prevent you from connecting to sites that would host that.

[dusky reef]

they were discord links

[meager meadow]

My guess is VRChat would recommend just going to the authorities anyway.

[dusky reef]

yeah lol

i submitted the report to vrc yesterday though cuz thats when it happend

saw they got banned

[meager meadow]

That's good to hear.

[dusky reef]

i really hope they do launch an invenstigation on them tho cuz it was really grose

cuz uhh kids were exposed to this sutff 💀

not going into too much detail

[meager meadow]

That's plenty for me LOL. I don't need a lot of detail for that kind of thing.

[dusky reef]

besides i love playing and creating on vrchat but just like how some other people have said. "Some people would just like to see the world burn."

oh forgot to mention

im in class rn

lol

i cant really do the testing for the networking thing soooo

[meager meadow]

haha pay attention to the class!

[dusky reef]

its only spanish class

its fine

jk imma pay attention

[hoary frost]

Read that
To give an idea, lets say we have an n number of synced bools we want to serialize, how many of them would get to the maximum byte rate?

[queen coral]

hey, quick question, i semi think this works but i just want confirmation, i want to play audio for everybody when one person interacts with an object, would this do that? I'm fine with regular coding but networking i have no clue, basically i want a doorknob to open a door and play a sound for everybody, i got the opening but im not sure about the sound

[jaunty fjord]

The CustomNetworkEvent is only being sent to the Owner rather than All which would make only the world owner receive the event to play the noise

[queen coral]

oh, so if i were to change it to all would it work, or am i missing something else?

[jaunty fjord]

I think that is all you would need to change

You might also need to make the 'person who interacts' the owner of the game object but I can't remember off the top of my head if that is needed to send a network event

I would change the target to All then build your world with two clients to test it

[queen coral]

if i moved the block node after the setowner node here would that be the fix if it didnt work

[tulip sphinx]

block is a lie

[queen coral]

is the block node wrong? i didnt know any other way to make 2 seperate things happen when interacting with 1 object

[tulip sphinx]

havent read through previous stuf but block is just a decoration

you just connect the output to the next input?

[queen coral]

oh

[tulip sphinx]

and you do a looot of cudtom events overall, since graph doesnt like merging etc. in u# its kinda more natural to do

so dont be shy on customevents

[jaunty fjord]

Is the script sync type set to Continuous or Manual?

Just out of curiosity

[queen coral]

manual i believe

[jaunty fjord]

Will the door play a different sound if it is opening or closing?

[tulip sphinx]

k checked your events while playing jingle should be fine, overall the main way to handle networked stuff is serialization

[queen coral]

nah, i kinda only want it to play a sound once, my world is like 1 player walkthrough world but with the ability for others to spectate, which is why im using networking in the first place, but the door should only have to open once

[jaunty fjord]

Ok that's fair then

I was going to suggest playing the sound in the OnDeserialization event but a CustomNetworkedEvent works perfectly fine for that use case

[queen coral]

alright, also, mostly unrelated note, i just tried to build and test on windows multiple times but its giving me an application not found error, what does that mean

[tulip sphinx]

so if soujd doeskt play disconnected from other stuff, weird. if its dependent on more constant stuff, then its not

[queen coral]

njevermind i think i found it

for some reason it wasnt pointing towards where vrc was installed, it was in some random default folder

[tulip sphinx]

set vrc location in 4th tab of sdk, or judt use 0 and then launcher in vcc, way better than default stuff

[queen coral]

whats the launcer in vcc?

[tulip sphinx]

hard to miss, the only tool they have

[queen coral]

oh okay, i see, but i cant use that anyways cause i havent published to pc yet

[tulip sphinx]

huh

it haz local build option

[queen coral]

oh okay

another question, how would i make an interact go away after it being used the first time?

would/could i disable the collider, or is there an easier way

[dusky reef]

Currently doing heavy testing

Not sure if this will work but its worth the test

Assets/clone script testing 2/Cloning.cs(35,35): Method is not exposed to Udon: 'Networking.Instantiate(VRC_EventHandler.VrcBroadcastType.AlwaysUnbuffered,CloningPrefab.name,CloningTarget.position,CloningTarget.rotation)'

💀

Sadly using networking doesnt work for cloning

[jaunty fjord]

It generally doesn't as a rule of thumb, it's likely possible to achieve some level of sync but you will have to create your own system rather than using the built in Networking tools.

[jaunty fjord]

If you want to turn off the Interact prompt and outline you can set the DisableInteractive property to true.

[finite sierra]

has anyone used https://github.com/Miner28/NetworkedEventCaller?tab=readme-ov-file and not being able to send a gameobject over with parameters?. i get Caller not assigned unable to send method - {0}, Invalid target unable to send method - {0}

[finite sierra]

@obsidian cedar is it something you have knowledge of?

[obsidian cedar]

GameObject ? You can't send scene references over network

[finite sierra]

It's not a scene reference. It's an object. Created runtime wise

[obsidian cedar]

public enum Types
{
    Boolean,
    Byte,
    SByte,
    Int16,
    UInt16,
    Int32,
    UInt32,
    Int64,
    UInt64,
    Single,
    Double,
    String,
    Decimal,
    VRCPlayerApi,
    Color,
    Color32,
    Vector2,
    Vector2Int,
    Vector3,
    Vector3Int,
    Vector4,
    Quaternion,
    DateTime,

    //Arrays    
    BooleanA,
    ByteA,
    SByteA,
    Int16A,
    UInt16A,
    Int32A,
    UInt32A,
    Int64A,
    UInt64A,
    SingleA,
    DoubleA,
    DecimalA,
    StringA,
    VRCPlayerApiA,
    ColorA,
    Color32A,
    Vector2A,
    Vector2IntA,
    Vector3A,
    Vector3IntA,
    Vector4A,
    QuaternionA,
    Null,

    //Variable Size
    Int16V = 60,
    Int16VN = 70,
    Int32V = 80,
    Int32VN = 90,
    Int64V = 100,
    Int64VN = 110,
    UInt16V = 120,
    UInt32V = 130,
    UInt64V = 140,
}

These are all the supported types

[obsidian cedar]

What kind of an object ? A GameObject ? GameObject is a scene reference

GameObjects, Transforms etc is not possible to sync over network

There's nothing to sync actually

[finite sierra]

Ahh so that's why it would result in null.

[obsidian cedar]

Yep

[fringe creek]

I was wary to use this initially due to the risk of having UdonSharp/VRC update and break this (requiring me to come back and fix the world in the future), but honestly it feels like it's worth the risk. I can't believe that event parameters aren't a built in feature already, though maybe games aren't really the focus of VR chat?

[fringe creek]

@obsidian cedar thank you for the wonderful extension! Was the possibility of skipping events ever addressed or was it a dead end?

[tulip sphinx]

so, i had a problem with teleporting people around, i heard there was some need to offset it in time but im not sure if its appliable. its not like im syncing the destination, they get one beforehand, it just does some weird stuff when they claim to be at new destination ie seeing stuff there, but you can see them at the og point floating in space, wha, how?

[obtuse echo]

You have to be careful when teleporting inside OnDeserialization. If that's what you're doing, then just use SendCustomEventFrames to call a method where you teleport the player.

I got a question about the VRCObjectSync component:

I feel like I must be doing something wrong, but it seems like it's jittering when owned by a remote player. It appears to apply gravity as it tries to fall down a little, but then immediately jumps back up. Is that supposed to happen?

Specifically, the jitter/bounce happens when the object sync is further away. I understand that the rate at which data is sent is being reduced based on the distance or something. But is it really supposed to bounce? Is there a way to remove that or something?

[strange token]

I kept running into issues with the teleportation basically ignoring the setting for "lerp teleport" or whatever until I delayed it by a frame

So players would just get stuck against the wall when trying to move between rooms

One frame delay on teleports (or on the method that includes the teleport logic) works to resolve it tho

[obtuse echo]

Oh and this seems to only affect quest. Could this be a bug or something?

Another thing with the object sync (less important compared to the bouncing issue):
When you disable and then enable the object, it seems like data is being received on a much late frame. It makes it difficult knowing when it's actually ready to be repositioned. Am I doing something wrong here again? Or are you not supposed to disable them in the first place?

[obtuse echo]

Okay, so I'm manually counteracting the downward force in the scenario where it's the most egregious. Helps at least mitigate some of the problem.

[obtuse echo]

Okay, so I found some relevant tickets about VRCObjectSync.

1.) SetKinematic/SetGravity is inconsistent. This happened to me too and I had to set them in multiple places as failsafes just in case.
https://feedback.vrchat.com/udon/p/vrcobjectsync-doesnt-sync-gravitykinematic-flags-reliably

2.) Disable VRCObjectSync and then enabling it again seems to cause trouble in general. When I enable the object and I want it to move to a certain position, it seems like values are being received and overwrite whatever changes you did while enabling it. It would be cool if the values set after we take ownership would take priority over what we receive later.
https://feedback.vrchat.com/udon/p/object-sync-problems

3.) This one I created for my issue because it seems different. I feel like all 3 are kinda connected though. Seems like the state handling on VRCObjectSync is off when you disable the object and then try to enable it again.
https://feedback.vrchat.com/udon/p/vrcobjectsync-bouncing-on-android-devices-when-further-away-gravity-not-disabled

[timber parcel]

God I always forget that I can't send network events with parameters.

It's really makes things a lot more difficult. I can work around it, but it's going to slow me down a bunch and might lead to a sub optimal solution with more bugs.

It'd be really good to get a better version of SendCustomNetworkEvent at some stage. I can imagine this is slowing down a LOT of people.

[timber parcel]

hmm. Can I sync a prefab refrence?

To be instatiated by clients later

[strange gyro]

Nope, only primitive types like numerics and strings, bool and some unity types like vector/quaternion and VRCUrl

[timber parcel]

riight

so I'll neet to sync a index and have a array somewhere of all possible prefabs?

[strange gyro]

Yeah

[timber parcel]

is start called every time a object pooled item is spawned or just at the start of the server?

[strange gyro]

It gets called the first time that object instance is made active in the scene, and if the pool reclaims the object and reuses it it wont be called again as far as I'm aware

[timber parcel]

nice

[finite sierra]

@obsidian cedar btw with your Netcaller thing. you state the amount should be MaxInstanceSize * 2 +2. so to be sure here that means if there capacity is 60 it should be 122?. and is there a specific reason for the double of netcallers required?

[obsidian cedar]

That's old information for when there only used to be 1 number you could set. But these days set it to whatever the Max is +2

So eg. if your max instance is 60, set it to 62

It's because there's 2 reserved slots that can go over the cap

[finite sierra]

Ahh

[unique elk]

I want to assign each player a box and later move these boxes around (possibly based on which player owns that box).

Should I use cyanplayerobjectpool and then loop through the items in the poolassigner?

Or is there a way to store the pool items in an array once they are enabled and deal with them there? (This seems like it would good if there is more than one item for each player)

[hoary frost]

Ive been trying to find a definite solution to this as well.
Cyanplayerpool could prolly be used, but you could also code it yourself

[finite sierra]

its extreamly easy. what i have done is to litterly just create a object in unity editor. and have a custom class on it that inherits from udonbehaviour. then when someone joins a world every person indivually adds this person to their own collection of available players. to get the actual players in the world all you do is to use the VRCPlayerApi.GetPlayers() that each person also needs to have

each person then locally assign a given object, id, name, etc to that player we get from the VrcPlayerAPI. and tell it to update each person cube based on their current position. this is done all locally and syncs quite well

[unique elk]

Thank you for your suggestion, I’ll try to test if this is viable for me

[hoary frost]

how would you assign the given object an id, name, ect?
It feels simple af but i cant compute it for some reason

[finite sierra]

you would just get the id,name and so on from the person who joined.

[hoary frost]

Well yeah but how do you assign it to the player in a way thats referencable later?

Having an array of gameobjects is cool and all, but yeah

[finite sierra]

thats how u assign it. and whenever u want to reference a player u just find them in your array by ID.

locally that is

since every person will have the exact same set of gameobjects

[hoary frost]

im still confused, could you show me a short example?

[obtuse echo]

Just FYI but DataDictionaries are very convenient for cases like these. Just map the playerId to your object.

[hoary frost]

that indeed seems convenient

[obtuse echo]

Yep, I'm as much of a fan of fiddling with arrays as the next person, but dictionaries are just more intuitive.

[finite sierra]

like any normal way when you want to assign any variable.

[fringe creek]

Why is my OnDeserialize function being called on an object with no synced variables quite a few times when the second player first loads? I would've thought that since none of the variables are synced, there'd be no reason to deserialize anything

[finite sierra]

i am pretty sure any Serialize or DeSerialize has to run on each person in a world regardless as soon as you override it or add it to your object.

[timber parcel]

Grabbing a object out of a vrcobjectpool and requesting serialization for it on the same frame is having inconsistent results for me. Sometimes the second player gets the data... Sometimes they don't.

Anyone got experience with this?

[timber parcel]

In the end I simply delayed the Serialization by ~0.25 seconds and that solved the issue. However it would still be good to know what's up with the object pool that sometimes the objects need a moment to initilize.

[meager quiver]

This isn't an issue with the pool specifically, vrchat in general has consistency issues when dealing with sync with disabled gameobjects. Given that's what a pool does, funk is going to happen.

You can cause similar issues messing with enable/disable state with late joiners - if a user joins late with an object disabled, vrcobjectsync won't put stuff in the right position when enabled unless it's been touched at some other point between them joining and enabling the object (though even then it's kinda hit and miss)

AFAIK the data gets written, the event just doesn't get fired correctly (but it's been a while since I've checked, this specific behavior detail may have changed).

[meager quiver]

If any component on the object has sync (including VRCObjectSync), the entire object will get the event, as serialization is done on a per-object basis rather than a per-script/per-component basis. This also means one script can cause another script to serialize without requesting, if it's on the same object.
To avoid sync here, you are best setting the sync mode to none, and making sure it isn't on an object that gets synced elsewhere.

You also won't get the event if the object hasn't serialized any data since the instance was started (i.e. there's no late joiner state to sync).

This is also why you can't mix sync modes, as the sync mode applies to the object, not the component.

[timber parcel]

Ahh, so the more reliable (but less performant) way to do a networked objectpool would be to hide the unused objects off in a corner rather than disable them?

[meager quiver]

Generally I'd say the delay works fine, a pool does work well. This kind of issue though is why you sometimes see people making their own pool implementations, so sync and such can be micromanaged around such edge cases.

Making a pool script yourself isn't all that hard either - array of booleans (though an array of bytes used as a series of bitflags would be more efficent network-side, or just a u64 to get 64 slots without the array serialization overhead, but you lose the O(1) lookup for slots via an IndexOf no matter what) and on deserialization, check if any flags have changed, and if so toggle their enable state.

The built in pool does this + on disable it sets its position to the location it was seen for the first frame the pool itself was enabled, the latter of which doesn't sound like something you need.

If you wanna shove objects into the corner, you can, but enabling/disabling also means you don't have to worry about update overhead if you use an update loop and aren't running a custom loop for it.

If you wanted to shove into a corner and keep enabled, a decent way to optimize this is to turn off the renderer component when it's unused. You wanna keep udon enabled, same with any sync, but renderer off means the gpu can skip it.

Shoving things into a corner this way will require you to make a pool of some form anyway if you want to use a similar kind of API (get me the next unused one pls) so you might as well go full ham on it

But if not, the delay works fine. A 500ms delay works for most possible situations of high latency - and that's a fairly extreme case (like US to australia).

[timber parcel]

That's the exact situation I've been testing in, but any level of delay seems to be fine, as long as it's long enough to be sent in different network packets.

[meager quiver]

Then that should work!

Also just a general sanity check - make sure you're actually taking ownership here. Pools don't do that for you. I assume you've checked, but always good to flag up regardless :P

[timber parcel]

Even 2000 ping should be fine if there is a delay long enough that the 2000 ping player recives the network update AFTER reciving the pool enable update

[timber parcel]

Oh yeah of course

Perhaps too often though

[meager quiver]

I mean if you try to take ownership when you're already owner, nothing happens, so ay

[timber parcel]

Is it okay to take ownership the same frame as requestingserialisation()

[meager quiver]

Yep, and that's my usual pattern.

Bonus thing I learned recently too - if you request serialization when you are not owner, the moment you take ownership that pending request will be executed (well, same serialization schedule tick) - without requesting again

The state that is serialized will be whatever is there at the time of it actually going through, as usual

[timber parcel]

... that... is dangerous

[meager quiver]

Only for things that depend on the act of serialization to trigger events, really. Like RPC systems.

In general vrc behaves pretty well when doing that - it schedules it, it doesn't serialize anything ahead of time.

[timber parcel]

More that it means that I need to check ownership in a couple of cases where the owner was syncing up the position now and then be everyone was running the script

As I was just telling everyone on to requestserialization with the intent that the other players would never actually send it

[meager quiver]

Yeahhhhhh that may be some of the source of your issues then lmao

[timber parcel]

That hasn't cause any of my issues yet... but I can see how it would

[meager quiver]

In my case it's handy - I've used it to act as a middleware layer for VRSL pixel triggers, to allow long (20 minutes plus) animations and audio to late joiner sync
I rely on this so if the master disconnects as it triggers, things will still work as is, as all players only store when they saw it change and send it on serialization - they only actually trigger stuff when they get notice from the network.

(there's also a bunch of extra redundancy to deal with questies suspending their HMD as this transition occurs)

[finite sierra]

its eaiser to make a pool yourself and use VrcPlayers. to get a person when they join

which is always in sync. also remember requesting a serialization triggers for everyone.

[fringe creek]

Oh that’s interesting, maybe that’s the case, thank you!

[fringe creek]

Interestingly enough, I've determined that just having a synced variable (even a new one that isn't set) is enough to cause OnDeserialization to be called twice a second. Remove the synced variables so that nothing is synced and it's fine, but add a single [UdonSynced] var even if it isn't ever set and the syncing will always occur on non-owner clients for whatever reason

[meager quiver]

This usually only happens if you have sync mode set to continuous (which is also enforced if you have the object sync component attached) - do you have any scripts with that on? With continuous set, sync will happen on its own.

You can also use the onpre/onpost serialization to monitor when this happens - this is called even with continuous, and even on scripts that lack sync, if the gameobject itself is flagged to sync

[fringe creek]

Of course, that is it. I had figured that manual means that it needs a manual sync request and that continuous just meant continuous watching of the variable for changes, not constant ser/de for the object no matter what

[meager quiver]

The exact rules of when continuous syncs or not aren't completely clear to me either, but it does try to minimize how often this happens at least.

[fringe creek]

Thank you for the help, I'm finding the odd design choices of udon networking to be a real pain

[meager quiver]

Worth noting too, the on var changed only triggers if the heap value changed - even if it got data (at least in the case of sync specifically). This can also lead to an unintuitive understanding of how variables are synced, if you base your knowledge around that.

I'm not 100% sure if it also skips the event here in the case of one script setting it on another.

[fringe creek]

No kidding? I haven't implemented any variable-specific listeners yet but that is very good to know

Okay yeah I've definitely been using the wrong type of variable syncing for my stuff, should've read the docs for that more clearly first

[finite sierra]

anyone noticed that the amount of data out from a world increases when vrc peak happens? and also desync is very common when vrc peak happens

also anyone noticed the inconsistency with delays? and sometimes worlds wont load properly and break?

[heavy spindle]

This might be coincidental since the networking in the instance is done Peer 2 Peer with steam relays as the backend. I'd say it more has to deal with steam's traffic, but at their scale, I'd assume their relays are chilling

During VRC's peak, the only things that would be affected are avatar, world, instance, group etc info fetching speeds. I believe their avatar and world downloads are behind a cdn separate from them which I believe is google cloud but I may be mistaken on the google cloud part

[fleet arch]

networking is done via photon server, not peer to peer

[heavy spindle]

I was under the assumption photon was only used for the ephemeral instance itself which was like the info and a shared known host for udp hole punching

[fleet arch]

nope :3

[heavy spindle]

Welp. Always happy to know how it all works anyways

[finite sierra]

is there a way to put a inherient delay into a world to prevent multiple people clicking a button immediately?

[meager meadow]

Begin with the button disabled, in Start() set up SendCustomEventDelayedSeconds(), then in the callback for that enable the button. But if the problem is multiple peolpe pressing the button simultaneously, rather than just too early, that won't be a solution since people could just click the button at the same time as soon as it is enabled. There really isn't a way to prevent that; you just need to account for it in your code.

[frozen apex]

you can also keep track of the time since the last button click, and only fire the code if the time exceeds X seconds or whatnot

[meager meadow]

That one would work for one user spamming the button, but not for multiple users simultaneously clicking the button. Network delays between users create a period of time where a button can be clicked by one user and no one else knows it happened.

[violet mist]

if a synced object is destroyed locally by one player alone in an instance, then another player joins that same instance (while the other player is still there) and attempts to call SetOwner() on that object, will VRChat recognize that the object has been destroyed, and set the new owner properly?

[frozen igloo]

If that object has an OnOwnershipRequest function in an udonbehaviour on that object then destroying it would prevent the owner from returning true, so it's possible that could prevent it from transferring. But if not, then it should transfer with no problems

[violet mist]

okay awesome, so as long as I don't override that method in the object's behaviour, i'm fine

[frozen igloo]

Yep

[violet mist]

okay cool, tysm !

[finite sierra]

sadly i dont think Vrchat has a internal time for that.

[tulip sphinx]

as it was said, you cannot stop users from spaming a button simultaniously, you should just make what happens next to be unbreakable.

you can use smrh like GetServerTimeInMilliseconds to determine whos input should go first, tho thatd require some queue

tho its probably an overkill unless youre remaking family feud

[finite sierra]

yea i know. atleast i should be able to prevent to many clicks in a rapid session.

[finite sierra]

what events run when soemone joins a world? on vrc it says it runs OnDeserialization when that happens. but it clearly does not.

[tulip sphinx]

it does, for joiner. not instantly tho, onenable/start happen earlier. everyone runs onplayerjoined

[stone badger]

A slight variation might be more workable. I use this on my click buttons to make them "blink" and even more so on those that process API calls. Immediately upon Interacting you disable the button and on the sync callback re-enable it. In my API calls that can mean waiting 5 seconds but the time can be set on the button. It also means the network sync has already taken place so it is ready to be used again.

[lone zealot]

OnPlayerJoined

However be aware that for the player who just joined the event is run for every player already currently in the instance and themselves as well

Also Start runs for the local joiner

[finite sierra]

uhm. it doesn't seem to run for every person currently in the world. OnDeserialization that is

[lone zealot]

Thatss why the first line of my message says "OnPlayerJoined"

OnDeserialization should also run if you have anything that has previously been serialized though

[finite sierra]

hmm. right but shouldn't it then show me in the log that OnDeserialization runs? when i have a debug in it?

[lone zealot]

Yeah...dont remember the specifics on that though. Its been changed a few times

[finite sierra]

sooo. i found several wierd things going on.

when a person joins a world and we sync our data up etc. everything works as intended. However as soon as one person leaves and rejoin everything breaks. they can still see the data i have. but i cannot see theirs and nothing in the world works for that person when they rejoin. nor does OnDeserialization fire for anyone else in the world.

However. When they then rejoin things work again for them but it then breaks for everyone else in the world. causing us to need to resync twice both of us before its working fully again.

[frozen igloo]

deserialization only fires on join if that player has persistent data on that object to load. But onplayerrestored will always fire even if there is no data to be restored

Oh sorry I thought this was #world-persistence

[finite sierra]

i wish heh.

[frozen igloo]

Are you just talking about standard late join sync? It's not super clear what situation you're referring to

Or are you talking about players already in the instance observing someone join?

[finite sierra]

late join sync or if anything happend in the world before someone joins

so yea late join

[tulip sphinx]

not in the world but with the exact object+variables that are set up for manual sync - ondeserialization fires 100% for late joiners and this is the way late join consistency is realized overall.

[finite sierra]

i can tell you it does not atleast i cant seem to get any debug for it.

everything i do is manually synced

[tulip sphinx]

it does, so if you cannot observe it, time to show some code

[finite sierra]

can't do. and even if i did it wouldn't show you anything. since i am just looking into why it doesn't call the onDeserialize and also does that not get called the first time an instance is created and someone goes into it anyway?

[tulip sphinx]

why would it get called the first time. its called on objects other people owned, changed and requested serialization already.

[frozen igloo]

Are the objects with ondeserialization perhaps disabled on load?

Disabling objects will prevent a lot of networking from working

[finite sierra]

they are yea. since each person a object that follows them around. and if that object is not occupied its not enabled until well a person joins. could that be the reason to it not working?

all through if i am reading it correctly on vrc docs. their object own pool which i dont use. does also allow for enable and disable of the gameobject. and it does not say that it could prevent it from working correctly.

[finite sierra]

after further testing. even if i locally override the state when that person joins it still behaves as if its disabled

hmm after more testing and setting the owner of the button to the person clicking it. it does seem to work. however now what happens. if the master leaves and rejoin and resync. the button breaks for the new master

[frozen igloo]

That's the problem, it needs to be enabled on load not just on start

[frozen igloo]

Object pool handles it in a special way that is able to work before udon start, which you can't do

[frozen igloo]

Enabling it with udon later isn't guaranteed to fix it. You need it to be enabled from the very beginning, when you upload the world

[finite sierra]

@frozen igloo hmm that doesn't solve the wierd issue of when master is being transfered. that it completely desyncs everything

[frozen igloo]

There's lots of ways that could break, you're gonna have to share your code or describe it

[finite sierra]

the way i have tested it so far is to have two people join the world. then both of them syncs up to each other. by clicking a sync button. then when they have confirmed they are synced i then make the master or rather the current owner of all objects to rejoin. when this happens the sync button turns to the oppesite state of being active so inactive. and the new master cannot interact with said button.

[twilit pewter]

I made this and it works. It generates a number of unique functions that I can use with SendCustomNetworkEvent(). Is there literally any alternative to this that scales in Editor mode for runtime?

This wouldn't be necessary if there was a way to pass any kind of argument, and I absolutely must keep this automated in Editor to maintain scalability for anyone to comprehensively use. I'm sure it's understandable that this isn't something that I want to condon giving to end users given the nature of generating code that needs to compile with compiled code. This is awful if there's literally no alternative. I really hope there's a better solution and I'm just ignorant of what it is.

try
{
    AssetDatabase.Refresh();
    Debug.LogWarning(AssetDatabase.FindAssets("VRCNetworkArgumentHandler.cs"));

    Debug.LogWarning("Writing " + System.IO.Path.GetExtension(path) + " file to " + path);

    //Prefix goes here.
    string fullFile = "using UdonSharp;\n" +
        "using UnityEngine;\n" +
        "\n" +
        "namespace VRCNetworkArgumentHandler\n" +
        "{\n" +
        "\t[UdonBehaviourSyncMode(BehaviourSyncMode.Manual)]\n" +
        "\tpublic class " + scriptName + " : UdonSharpBehaviour\n" +
        "\t{\n" +
        "\t\t[SerializeField] private EnemyLogic _enemyLogic = null;\n" +
        "\t\t\n" +
        "\t\tpublic void _cannonHit(int cannonID, int damageLevel)\n" +
        "\t\t{\n" +
        "\t\t\tSendCustomNetworkEvent(VRC.Udon.Common.Interfaces.NetworkEventTarget.All, \"Cannon_\" + cannonID + \"_\" + damageLevel);\n" +
        "\t\t}\n\n";

    //Function Generation goes here.
    if (cannons.arraySize > 0 && damageLevels.intValue > 0)
    {
        for (int cannonID = 1; cannonID < cannons.arraySize; cannonID++)
        {
            for (int damageLevel = 0; damageLevel < damageLevels.intValue; damageLevel++)
            {
                fullFile += "\t\tpublic void Cannon_" + cannonID + "_" + damageLevel + "() { _enemyLogic._hit(" + cannonID + "," + damageLevel + "); }\n";
            }
        }
    }

    //Suffix goes here.
    fullFile += "\t}\n" +
        "}";

    System.IO.File.WriteAllText(path, fullFile);
}
catch
{
    Debug.LogError("There is no file at path or unable to access it: " + path + ". Please create the folders and the U# file there first.");
}```

[finite sierra]

what exactly is this? are you trying to create a method that does something? cause what your doing seems to be it just creates a new script. with some new value that wouldn't ever work in any application even outside vrc.

[twilit pewter]

I'm generating a new script because I don't want this cursed method of generation to exist outside an easily deleted bubble since it's possible a mistake can prevent compiling.

[finite sierra]

huh.

[twilit pewter]

The scenario is I have here is a number of unique cannons need to convey their targets with how much damage they're doing to them.

[finite sierra]

it generates a new script? for what purpose. u never need to create a new script. for every difference in value of in this case towers etc