bevy_replicon

[spring raptor]

You mentioned that it consists of bools

Or is there more information?

[hexed cove]

just bools

[spring raptor]

Maybe grid as components would be more useful? This way you could replicate only actually changed data.

[hexed cove]

i don't know if that's actually more performant

[spring raptor]

Only one way to find out 😅

[hexed cove]

yeahhhh

well currently i just need to update a grid of bools

as opposed to querying the coordinate, updating the entity, sending to the client the entity with the new state

i've been weighing it in my head all day and it just doesn't seem worth

working on the renet build currently

[spring raptor]

It will require additional lookup for coordinate - entity, but other then that - it should save a lot of traffic. And traffic is the most valuable resource.

[hexed cove]

grid of bools with mostly repeated 1s and 0s seems very easily compressable

[spring raptor]

Ah, true...

And you don't need to attach entity to each cell 🤔

[hexed cove]

or pretty cheap even without

it'd have to be benchmarked

but it also works well enough

i feel like all of game dev is just "works well enough" lol

[spring raptor]

Yeah, always focus on the game. You can optimize it later.

The game is the hardest stuff and the requirement of "make it fun" is super vague

So it's better to figure out what is fun and then iterate on performance

For crates it's different. You have the requirements and you need to choose the right architecture.

[hexed cove]

I'm getting this error with trying to switch to renet as the backend

[spring raptor]

Replicon versions mismatch

Make sure to use compatible versions

[hexed cove]

I moved grids to their own components and had them be handled with events and it worked perfectly, although I'm still getting the "datagram too large" issue when spawning ~20 ants. This is after setting up replicate_with for transforms

[spring raptor]

Probably the data still exceeds the maximum size somehow.
Try to switch to renet for now.

Alternatively, I can quickly patch replicon by replacing this constant with something that quinnet accepts 😅
https://github.com/projectharmonia/bevy_replicon/blob/927a8bfb3750635da037b0a260166c295ca00738/src/server/replication_messages/mutate_message.rs#L227

[hexed cove]

what was the size quinnet needs again? I can just use a fork

I tried renet but couldn't get it to work, skill issue I think

renet replicon has a lot of similar types as replicon, renet ClientId vs replicon ClientId, ServerEvent, etc

quinnet felt easier to use

[spring raptor]

I have a better idea.
Check what value is used for https://docs.rs/quinn-proto/0.11.8/quinn_proto/struct.Datagrams.html#method.max_size and I will provide a patch release of replicon with a more conservative value for now. With the next release I will fix it properly.

[spring raptor]

You don't have to touch them at all. You only need to initialize RenetServer or RenetClient. And just use everything else from Replicon. Except for disconnection, you need to call a method on RenetServer.

[hexed cove]

1162

i figured it out

[spring raptor]

Could you open a PR to adjust the constant? This way I will be able to publish a patch-release.

[hexed cove]

38 bytes of pain

yup

[spring raptor]

It's not a proper fix, but an acceptable workaround. We definitely should allow backends to set this value.

[hexed cove]

yeah i see the TODO

made pr, although might need to adjust the tests

why is this asserted as false?

[spring raptor]

Because the message is already full. We can adjust the algorithm to handle adding zeroes, but we never face it in the code, all entities have size > 0

@hexed cove could you double check that the PR solves your issue?
Just use patch because you want quinn to refer to the correct verison: https://doc.rust-lang.org/cargo/reference/overriding-dependencies.html#testing-a-bugfix

[hexed cove]

testing currently but some changes with serverevents is causing issues

apparently it was changed to triggers or something?

going from 0.29 -> 0.30 for replicon

[spring raptor]

Ah, yes. It's a recent release. Check the changelog: https://github.com/projectharmonia/bevy_replicon/blob/master/CHANGELOG.md#0300---2025-02-04

Don't forget to update the quinnet as well.

[hexed cove]

yup

[hexed cove]

getting this error when setting the visibility with the observable

removing the visibility, it does appear to be fixed

I do get this issue when summoning like 10k ants

so there is a limit lol

it keeps running, just gets laggy

[spring raptor]

Could you elaborate on this one?
The panic screenshot is too short to reason about.

[hexed cove]

these two lines on the bottom are problematic

[spring raptor]

Ah, I see what is happening here. Basically, when a new client is connected (emitted by backend) I create an entry in connected clients. But looks like your call is calling earlier.
Make sure you setup your plugins after replicon's. I think observers should be triggered in their registration order.

[hexed cove]

they should be registered after already?

[spring raptor]

Strange, I would expect this function to run first:
https://github.com/projectharmonia/bevy_replicon/blob/927a8bfb3750635da037b0a260166c295ca00738/src/server.rs#L177

[hexed cove]

hmm

i can log if the debug shows up

[spring raptor]

Well, as a temporary workaround, you can create your own event and trigger it from your own observer. This way you will be able to properly set the visibility. But it's junky 🤔

@hexed cove Better, create OnAdd observer for your teams and setup the visibility there.

But it's strange, I thought that observers fire in their registration order 🤔

[hexed cove]

they do

client connected -> setting visibility

well they do in this case, Idk if it's actually documented

[spring raptor]

On the screenshot I see logs only from Replicon, they are expected.

[hexed cove]

the bottom log "setting visibility" was placed by me

[spring raptor]

But judging by this panic, your function is running before.

[hexed cove]

on the function

[spring raptor]

Got it, my fault actually.
ConnectedClients are updated immediately, right.
But ReplicatedClients updated from this trigger: https://github.com/projectharmonia/bevy_replicon/blob/927a8bfb3750635da037b0a260166c295ca00738/src/server.rs#L187
Which triggers: https://github.com/projectharmonia/bevy_replicon/blob/927a8bfb3750635da037b0a260166c295ca00738/src/server.rs#L212

Let me put a quick fix.

[hexed cove]

and they're ordered in a way that causes a 1f delay?

[spring raptor]

They fire in the same frame, but since you observe for ClientConnected, your function was called earlier.

https://github.com/projectharmonia/bevy_replicon/pull/408

[hexed cove]

Oh I see

ClientConnected triggers both my observer and the replicated clients observer

that then generates it

[spring raptor]

Yes, and yours fire earlier due to how commands are flushed

Won't be an issue anymore :)

Once koe review both PRs, I will draft a patch release ASAP.
In the meanwhile just these apply changes locally.

[hexed cove]

Sounds good

[hexed cove]

:(

trying to replicate the grid triggers this again, so it still happens with one large entity

[spring raptor]

Interesting. Could you try to print the used value and the max size?
I would include this information in the error message. So find the string, patch the crate (maybe PR to the author, good to have) and try it.

[hexed cove]

do you know why these logs might not be showing up?

[hexed cove]

working on it

[spring raptor]

Not quite sure 🤔 I would assume that they are not called. Double check if you have only a single registration for the TeamVisibility. If you have a registration with more components and TeamVisibility, it takes priority.
For example, (Transform, TeamVisibility) will be picked over regular TeamVisibility rule.

[hexed cove]

Oh yup

lmao

[spring raptor]

😅

But could you disable compression and check the size? 🙂

[hexed cove]

oh

what if you want a bundle + replicate_with

like i dont want to replicate every transform

but I also want to use replicate_with

like replicate_group_with<(Ant, Transform)>();

[hexed cove]

also sadly doesn't work

[spring raptor]

You can do custom groups. What you pass in replicate_group_with is a bundle.

[hexed cove]

Oh I see

[spring raptor]

(Component, Component) is a bundle. Just create your own struct instead and implement ReplicationGroup

[hexed cove]

yup, missed it in the docs

[spring raptor]

I need more info on this ^

[hexed cove]

I see it now

[hexed cove]

serialization fns

[spring raptor]

Wait, but you said it worked 🤔

[hexed cove]

serializing works, this happens when attempting to deserialize

[spring raptor]

Interesting, are you sure that functions are the same on client and server?

The panic means a problem with registration

[hexed cove]

yes

these are registered in my core crate

so they run on both client/server

[spring raptor]

Hm... The error means that the received ID doesn't exist

Function ID

[hexed cove]

what function id?

[spring raptor]

When you register a replication group or component, you create its function IDs internally.
When transmitted over the network, we include this ID.
Error means that there is no such function registered for such ID.

Let me improve the error message, so we could see what is happening more clearly.

[hexed cove]

maybe my compression thing is bad

[spring raptor]

I don't see an issue with the provided code 🤔

Let's work on one thing at a time.

[hexed cove]

just gonna double check if transform works properly

[spring raptor]

Let's take a look at the datagram limit (I suspect that it's dynamic) and I improve the panic.

[hexed cove]

oh true

testing datagram logs

[spring raptor]

Opened: https://github.com/projectharmonia/bevy_replicon/pull/409
Should help us a bit.

[hexed cove]

Datagram Size 500745, Max Size Some(1414)

[spring raptor]

Interesting. So you have an entity worth of 500745 bytes

[hexed cove]

yup

[spring raptor]

And max size is different: 1414

It's probably dynamic 🤔

[hexed cove]

I think so

[spring raptor]

Okay, let's back to compression. I guess if we solve this problem, we won't need to worry about it.

[hexed cove]

50kb per team_visibility

yeah compression brings it to like

[spring raptor]

Huge 😅

[hexed cove]

50 apparentl;y

again, grid full of repeated 1s and 0s

[hexed cove]

testing this

[spring raptor]

Now I wonder if it worth even merging something like https://github.com/projectharmonia/bevy_replicon/pull/407
We don't know minimal possible size.

[hexed cove]

probably not

[spring raptor]

I bet author will appretiate if you submit the improved error message.
No in this form I mean, but put this as part of the error!.

[hexed cove]

I couldn't find a clean way to add it as part of the error

the error prints the error from quinn

[spring raptor]

Ah, I see :(

[hexed cove]

maybe a match to send a unique error for specifically TooLarge

and everything else uses the original

Ill submit a pr

[spring raptor]

Maybe just edit the display for the error 🤔

I.e. include more info in the TooLarge and adjust Display impl. But it's not as simple change as I thought. Maybe worth keeping as is.

[hexed cove]

could edit the display

but getting access to the size is a problem then

[spring raptor]

Yep, this is why I suggesting to include this data in the error to display it properly

But the author might dislike the increased error size

[hexed cove]

[spring raptor]

181 is a huge number. It means that you have 181 functions registered. Which is unlikely.
This could mean that we are trying to read some garbage.

On deserialization

Let me think

[hexed cove]

im gonna check if transform works

[spring raptor]

I think we have tests with custom serialization, but yeah, double check just in case.

Ah, I see the issue

[spring raptor]

You replacing the message here instead of appending to it

We trying to not allocate for performance reasons, this is why I pass the buffer 😅

[hexed cove]

Oh I see

I misunderstood how that works

[spring raptor]

Yeah, it's a possible pitfall. But it's very important to keep replication fast.

[hexed cove]

hmm

well

this works for a single tick

it fails to decode, but i think it successfully decodes a single time

because the correct vision shows up for a single frame

[spring raptor]

The decode most likely tried to decompress the whole buffer

Serialize the buffer size first

Then on deserialization check the size first and then pass a slice of such size to zstd

[hexed cove]

that makes sense

it works but it does eventually grow too large

compression isnt enough 😔

[spring raptor]

Damn. But at least we strees-tested Replicon 😅

You need to make it deterministic

[hexed cove]

deterministic ants?

although my end goal is multiplayer ant combat so serverside visibility is my current goal

[spring raptor]

Yeah, just don't replicate all data. Replicate only necessary data to reproduce the results on clients.

[spring raptor]

Am I understand correctly that max datagram size is per-connection?

[hexed cove]

not sure, I've been quite confused by quic

[spring raptor]

Ah, I mean in the provided API. You get the value from the connection, right?

[hexed cove]

Yup

[hexed cove]

entities will never be able to be split across packets right?

[spring raptor]

Yes, it's by design :)

Otherwise the game logic might break based on network conditions

[hexed cove]

would it be possible to pass larger entities through a different channel type?

[spring raptor]

Right now there is no such functionality :(
But even if we do, looks like single entity is already bigger then the datagram size.

[hexed cove]

well

transmitting grids through events does work

so I think there's some way to make it work through ordered or unordered channels

[spring raptor]

Replication utilizes both channels. Insertions, removals, despawns and spawns go through reliable channel. Mutation go through unreliable channels. It's because there is no sense in resending the old value, if the packet gets dropped, we send the latest value.

If you want to go this route, the events is the right tool for it.

[hexed cove]

that makes sense

max entity size probably needs some sort of documenting though

[spring raptor]

I agree. Didn't expect someone to transfer that much 😅

[spring raptor]

Switched to postcard. I described the reasons in the PR:
https://github.com/projectharmonia/bevy_replicon/pull/410

@hexed cove the release with triggers fix is on crates.io.
I will provide a fix for backend message size later. But In your case you want to use events anyway.
Highly suggest to somehow minimize the amount of transferred data and more rely on determinism.

[hexed cove]

yup

I figured out a way

[hexed cove]

it might be worth adding a fn or recipe in the docs for replicating/events with custom serialization + mapping

[raw idol]

🤔 could we have bevy_replicon no_std?

I was thinking of writing a UART or ESP32 wifi transport anyway for fun

[spring raptor]

Agree. We currently have only for just custom serialization without mapping, right?
A PR would be highly appreciated. Or do you need help to figure out how to do so?

[hexed cove]

I will try tomorrow

bc it's something I'm doing

which option do you think makes more sense here?

[spring raptor]

I'm curious about this as well 🤔
But I have small experience with no_std from work, we provide minimalistic hal for our cores.
But this sounds like fun to try.

From what I understand, we need to replace all std with core, switch to bevy_* (or wait for #1303128171352293410 message), add std feature and disable std by default for bytes and postcard. And the figure out what to do with the buffer I allocate to prepare messages (replace with alloc?)

[raw idol]

sounds about right to me, but I think dependencies requiring std may also be an issue. I'm not sure what replicon depends on. alloc shouldn't be a big deal, since IIRC bevy_ecs requires alloc anyway.

oh, and Instant will be a problem. this is the main issue I had with making no_std aeronet_io

[spring raptor]

Here under example I would write "Serialalizing only translation from transform:" and add one more example with something like "Serializing only entity from componentA:" and add another example.

https://docs.rs/bevy_replicon/0.30.1/bevy_replicon/core/replication/replication_rules/trait.AppRuleExt.html#tymethod.replicate_with

[raw idol]

https://github.com/bevyengine/bevy/tree/main/crates/bevy_platform_support

Looks like this handles Instant for you

[spring raptor]

We are very tiny on deps, they all support no_std, except bincode. But it's fixable with postcard.

https://github.com/projectharmonia/bevy_replicon/blob/af9129368e589d8988dbaaaebb8418c398a00cc8/Cargo.toml#L34

[spring raptor]

Ah, yes. But in Replicon we don't need Instant.

[raw idol]

that's great then, I thought you might have used it for something in replication. I use it in the IO layer.

[spring raptor]

Lackily I use it only in benchmarks 🙂

I think main problem will be Bevy. We might need to support master :(

I mean it's fine, but we won't be able to release it to public

[raw idol]

yeah, 0.15 isn't no_std right now I don't think

this is more of an experiment than anything, at least until 0.16

[spring raptor]

Yeah, I probably could create a branch 🤔

[raw idol]

I'm gonna make an aeronet no_std testing branch, and try to port aeronet_io over to no_std + add esp32 wifi support, if that's even possible

and see if I can port the transport over, which might be more challenging

are you pinning to a specific bevy rev?

[spring raptor]

How practical it is? Totally unpractical.
But how fun is it? Very fun 😅

[spring raptor]

It wold be awesome! It's cool that you have esp32. I will try to do the same with Replicon.
Will start with the current master. But I need the postcard PR to be merged. And maybe for BashRat to port bevy_internal.

So you can start, I will catch a bit up later.
Replicon should be easier to migrate since we basically just read data from Bevy and feed it back.

[hexed cove]

are you building a multiplayer project w no_std?

[dire aurora]

Did you look at how other options, like say bitcode or speedy, compare?

[spring raptor]

I trying to build a regular game: #1264625779296174110.
But Bevy community recently expanding no_std support. I wanted to join the hype, but I needed a backend that can utilize that.
@raw idol mentioned in #networking that he have esp32 and considering supporting no_std in aeronet, so how can I resist now? 😅

[dire aurora]

Bevy community
You say like it's not just @twilit islanddoing 90% of the work ... I'm sure they would appreciate seeing a no_std networking crate tho

[spring raptor]

I evaluated bitcode. It packs very efficiently, but without streaming support, all data will be byte-aligned, which defeats the purpose. But I keeping an eye on it, will be happy to migrate in the future.
I didn't try speedy, I wanted something that can be used with serde since it's what Bevy uses.

[twilit island]

I'm gambling that once 0.16 launches and people can just do:

#![no_std]
use bevy::prelude::*;

That the community will take on the challenge and my work will become competitively small

[dire aurora]

I thought they said they added streaming support in 0.6, tho I never actually checked since I already rebuild bevy_bundlication on top of replicon at that point ... And yea I guess using custom derives like speedy (or the optional, but much faster mode for bitcode) wouldn't work great, maybe one day when we get compile time reflection :')

[dire aurora]

That would be pretty neat ... I wonder if we can also no_std most of the ecosystem crates ... I feel like I could probably no_std my rollback and input queue stuff, since they largely only use the ECS and some alloc 🤔

[spring raptor]

What I like about bitcode is that it supports custom derive that we could use to more efficiently pack for replication + supports serde for components.

[twilit island]

That's exactly my thinking. Not every community crate will be simple, but Bevy already has its own ecosystem for allocation, threading, etc., so a lot of what we'd rely on std for we already just let Bevy do for us lol

[dire aurora]

I think they actually dropped support for those attributes in the custom derive 🤔
The "rewrite" PR had this comment: "doesn't require any hints (determines them at runtime)"

[dire aurora]

Yea, we'd really only need to inject a bit of std functionality where necessary, but can otehrwise interact with no_std APIs. In the case of bevy_replicon this would largely mean that you might need std for a transport (which is easy to replace with something custom when lacking no_std), but the rest could be no_std as long as all the bevy features it touches are

[dire aurora]

Looks like on the serialization benchmark (if we just look at bandwidth) the only ones that really consistently beat bincode are postcard, bitcode and whatever dlhn is 🤔

[spring raptor]

I mean they provide more efficient bare-serialization via https://docs.rs/bitcode/latest/bitcode/fn.encode.html / https://docs.rs/bitcode/latest/bitcode/fn.decode.html. Their trait implemented for built-in types and requires Encode/Decode derive for custom types. And with serde feature they provide https://docs.rs/bitcode/latest/bitcode/fn.serialize.html / https://docs.rs/bitcode/latest/bitcode/fn.deserialize.html. With streaming support we could use encode/decode for our replication messages data (ticks, manual arrays) and serde for components. So users won't need to use any custom derives, just regular Serialize/Deserialize.

[spring raptor]

Yep, bitcode is the best choice if you don't care about streaming.
I also considered dlhn, but it's quite new, I decided to wait a bit to see if it matures.

[spring raptor]

@raw idol what do you planning to use for transport? Something like https://github.com/embassy-rs/embassy ?

[raw idol]

Yeah:

• embassy-executor
• embassy-time
• esp-hal-embassy
• esp-wifi
• smoltcp
• aeronet_io + a custom layer on top
  I assume

[hexed cove]

working on the pr now, can't figure out how to map the entity without having access to the component

since we're trying to only serialize/deserialize the entity inside

[spring raptor]

Great!

You deserialize the component (MappedComponent in your case) directly first, the just call the map_entities.

Just merged migration to postcard 🎉
It's the best option for now. Looking forward to streaming support in bitcode.

[spring raptor]

@hexed cove I recently merged a change to serialization. But don't worry, I will updating the example, the changes are trivial.

Basically, the changes :

[hexed cove]

yup, just drop in from bitcode

[spring raptor]

And Cursor now just Bytes

[hexed cove]

this probably works i think

ill test it

oh wait i see an isseu

[spring raptor]

Yep, looks good.
But it's actually mirrors default ser/de functions:
https://github.com/projectharmonia/bevy_replicon/blob/postcard/src/core/replication/replication_registry/rule_fns.rs#L212
https://github.com/projectharmonia/bevy_replicon/blob/postcard/src/core/replication/replication_registry/rule_fns.rs#L195

So maybe add an extra field to the component and write a comment that we skipping the filed for replication? I.e. just to showcase a somehwat real use case scenario.

[hexed cove]

somewhat real use imo would be to just

for compression or something

(my usecase)

if you just needed to exclude fields you could just seperate the component xd

but sure

[hexed cove]

oh thats why its called serde

[spring raptor]

I agree 😅 just don't want to pull extra dependency.

[hexed cove]

yeah agree

wait

you can't just send serialize and send "entity" though

[hexed cove]

because this issue

no component to call map_entities on

unless I create a new one

[spring raptor]

You can create MappedComponent with the default value.

But not very practical 🤔

[hexed cove]

yeah

like that lol

[spring raptor]

What if we create a function compress inside the example that actually does nothing? 🤔

[hexed cove]

yeah real usecases would be like

compression or if you want to manually do bit manip to like

hyper optimize

[spring raptor]

Yeah, and in the example it's not necessary should be a real function

We can hide it by prepending #

[hexed cove]

that makes sense to me

[spring raptor]

So just copy what you have for you game and replace zstd with dummy compress😅

[hexed cove]

the compression does use cursors, but i think it can use any thing that impls Read

compression also requires you pass the size in as well

ill try writing something up

[spring raptor]

Bytes implemet Buf which is cursor-like

And you can call https://docs.rs/bytes/latest/bytes/buf/trait.Buf.html#method.reader

[hexed cove]

does this look right to you?

not sure if I did the #s correctly

[spring raptor]

Serialization looks correct. Although, I would pass Vec into compress by mutable reference, will look a bit nicer.

[spring raptor]

But in deserialization you need cursor on the latest release

And Bytes on main

[hexed cove]

I think im currently on the latest release

oh i see the issue

[hexed cove]

that would be for inplace right?

[spring raptor]

Yeah

This function doesn't do anything anyway 😅

[hexed cove]

there's a few more bugs i fixed

[spring raptor]

Thanks to postcard I switched to varint encoding for ticks: https://github.com/projectharmonia/bevy_replicon/pull/411

We use 32-bit ticks.
With bincode, ticks would start to use 5 bytes after 2^16 (which happens after ~18 minutes with 60 ticks/s).
But with postcard it happens only after 2^28 and with 60 ticks/s it starts to take 5 bytes after ~51 day.
So don't run your sessions for 51 days, otherwise messages will start to take 1 more byte 😅

[hexed cove]

https://github.com/projectharmonia/bevy_replicon/pull/412

[hexed cove]

do player/server only observables exist

[hexed cove]

workaround, use some sort of marker through resources or state and check in the observable

conditional observables should probably be added to bevy

[hexed cove]

ReplicationGroup doesn't exist :(

oh it's GroupReplication

mb

if an entity has GridCoords, VisionRange, and TeamLabel, they should all be replicated in the same entity right?

for some reason these components go missing on the client though

no error on the server

maybe they aren't added at the same time

on insert doesn't work either, maybe it's added one by one

Triggering it every time any of the 3 get added, then returning if it doesn't have all 3 is a working solution, but a little junky xd

[spring raptor]

Yeah, I also looking forward to observer conditions!

[hexed cove]

yesh

currently I can check the resources to see if it's server/client as well

there's workarounds but it's cleaner w observer conditions

[spring raptor]

Interesting 🤔
We insert components via commands. And flushing happens when we done processing entity.
So I would expect all insertions to happen first and then your observer should run.
It still create archetype moves, but I just waiting for https://github.com/bevyengine/bevy/issues/10154

[spring raptor]

Yep, that's what I do as well

[hexed cove]

but they're registered seperately

you're saying they should all be inserted even with separate registerations?

[spring raptor]

Yes. So internaly we call commands.entity(entity).insert(component), but flush when all components for entity is inserted, no matter if they grouped or not.

But maybe I misunderstanding how obsevers work 🤔

I would expect them to be called after the flush

[hexed cove]

I can double check my code tomorrow, but all three of those components should be replicated

[spring raptor]

Could you do commands.entity(entity).log_components inside your observers?

[hexed cove]

it lists the components in the error I think

GridCoords, Replicated, VisionRange, ConfirmHistory

but I can do that tomorrow

[spring raptor]

Hm... The error says that the components are present

Except TeamLabel

And that's the one you insert later with events, right?

[hexed cove]

nope, that should be replicated

replicate mapped if that matters

[spring raptor]

Weird, for some reason you have other components inserted and this one missed 🤔

Are you sure that you don't have an entity with these components and without TeamLabel?

[hexed cove]

I shouldn't, but I will double check tomorrow

need sleep, it's 6am

xd

[spring raptor]

Good night 🙂

[spring raptor]

Yeah, I also sometimes like this 😢

[hexed cove]

they should always have vision range

i only spawn vision range in 2 places and both have teamlabel

[spring raptor]

Somehow you don't have TeamLabel in the trigger, not sure why.

[hexed cove]

all components on an entity need to be sent together right?

[spring raptor]

We send all components on an entity together, yes

And insert them via commands

And then flush them.

Basically commands.entity(entity).insert(a).insert(b) and then flush.

[hexed cove]

weird

[thorn forum]

hello ! just to confirm, its not currently possible to replicate the whole Transform for insertion, and then only the translation ? im looking at .replicate_with(), but it seems to only allow the DEserialization to be different on insert and in place, but the serialization we can't choose ?
if i understood correctly then, the best way would be to replicate with Transform with only the translation, and on replicate a InitialTransform to set it initially, and after only the translation is sent.
this is for minimizing the server egress.
thanks !

[dire aurora]

I don't think this is currently possible. I'm not entirely sure if it would even be possible to add this without essentially just doing all the work for delta compression. The server can't just send the value once, since it might get lost, but then replicon also has the init messages which I think contain newly spawned components so that might be relevant 🤔

[spring raptor]

Yeah, it's not currently possible.
But you can do a workaround - create your own Rotation and Scale components that updated only once on Transform insertion and replicate only translation for Transform.

If you do this, don't forget to override the in_place function.
I recently updated doc examples to showcase it: https://github.com/projectharmonia/bevy_replicon/pull/412/files#diff-ca6867deb8d662237ab6da10d61c239e34dc35c3b2d957938a71f15bc4b6413fR98

We switched to postcard on the main, but the idea is the same.

[thorn forum]

oh yeah thats good too ill try. honestly im even wondering if i can just not even replicate my troops and projectiles positions, even with this method to only send translation. a 5min game goes from 2mb to 5mb, which mean practically 2x more expensive games to host. this amout seems normal ? here's a peek of my game, do you think its reasonable to not replicate movement ?

thanks for the answers !

[spring raptor]

The game looks cool!

I think you can 🤔 Just send the actions, but make them do nothing on client and replicate actual game state from the server.

[thorn forum]

thanks ! ill continue to not replicate transform then. here's some test with spawning multiples buildings with multiples troops etc:
no transform: 2mb
initialtransform: 5mb
transform: 11mb
running a game without doing anything, so nothing is being replicated, no event etc, its still 0.9mb. looking with wireshark there's a lot of packets every seconds of len 68 or 72, i guess keep alive / tick, but it seems a bit excessive for a idle state or its normal ?
if you have some experience with bandwidth usage in games, do you find these numbers normal for a game of my type ?
thanks !

[spring raptor]

Yep, it's keepalive. It's a know behavior since the original renet: https://github.com/lucaspoffo/renet/pull/156
Replicon sends packets only if data actually changes. If there are no changes, we don't even send ticks.

As for traffic, I can't tell, simply don't have experience to compare. But if you are optimizing, maybe you would like to try the latest master? I did a bunch of optimizations to reduce the packet size recently.
Use patching in Cargo.toml since backends rely on release version.

@thorn forum you can also check what replicon sends by enabling tracing

RUST_LOG=bevy_replicon=trace cargo run ..

[thorn forum]

thanks for the everything!
I'll try master.
I already tried tracing and there only tick that's changing !

[thorn forum]

I don't know much about the inner protocol of renet and networking library, but I would have thought a idle game would use less bandwidth.

[spring raptor]

Then it's good, it prints the informatioon about sending bytes if there is data to send.

[spring raptor]

I agree. Ask about it in #1302745044393918646.

Let me know if you need any help with the latest master.

[echo lion]

Yes the renet ack protocol is a naive design. It really needs a rework (or at least a full analysis with improvements), but I don't have the time to work on it.

[thorn forum]

thanks, ive learned a lot i understand now how use the Fncs config. but ill just stick to not replicating positions and timer for max efficency.

[thorn forum]

so its normal, thanks for the answer. i didn't try on web with ws, but afaik renet2 still use renet in the ws connection ? i may try aeronet which use quic on native, and just ws for wasm.

[spring raptor]

Looks like this was also merged, nice!

[dire aurora]

Yep, we're just gonna need some better UX on top now, like disabling hierarchies and the like

[thorn forum]

hi again, what happens on the client when is it disconnected from the server ? like does it modify/remove some components ?
on the end screen im still showing the game in the background but stopped, and when the server quit and so the client are disconnected, multiples things are modified, including non replicated component, but these are maybe modified indirectly through my systems.

example here. the serve quit 2s after the game ended, to demonstrate the change when the client are disconnected. at the end the health bar are blank, and the animation plays again. i struggle to find the cause.

i dont have any .run_if(client_connected) systems too

it seems that some replicated component are re added to some entities ? i have a Added<Troop> for new troops, and it get triggered on disconnect

[dire aurora]

Yet another reason to do With<Troop>, Without<WhateverIsMissing> (or just using required components or hooks/observers with insert_if_new)

[thorn forum]

yeah I'm gonna use observer + required component. I was using added to do multiple things at the same time and especially being able to run it after another system that add a component, with observer I can't order them.
I didn't expect Added to be re triggered

[spring raptor]

Hm... Replicon doesn't do anything on disconnect.
It just stops sending data.

[spring raptor]

I like the art! 🙂

[spring raptor]

What conditions do you usually use? Maybe there some .if(server_or_singleplayer) that start evaluating to true when you get disconnect? 🤔

[echo lion]

Yes renet2 uses the renet protocol in websockets, but I downgrade all channels to unreliable which avoids extra overhead (beyond the baseline excessive ack packets).

Keep in mind aeronet has a custom auth protocol that is not battle-tested like netcode (I think? or is there no auth protocol at all?).

[thorn forum]

hm weird 🤔

[thorn forum]

thanks ! it's not from me tho 😆 I was wondering which art to use between this one and the last screen record I sent

[thorn forum]

oh yeah maybe, I didn't though about "server_or_singleplayer" evaluating to true on the client side lol...

[thorn forum]

hm thanks for the info !

[spring raptor]

I like the first one a bit more since I'm more into pixel art 🙂

[spring raptor]

That most likely it. If you have any idea how to properly deal with it, let me know

[thorn forum]

yeah same it has its own style

[thorn forum]

I'm may be missing a use case, but what about also including the repliconserver in the run condition?
for example, if there only a client but no server resource, it's a client even if it's disconnected. but if it has both a client a server resources, use the same logic as now ? it would be more logical imo, but I'm probably missing other things

[spring raptor]

In Replicon server and client always exist. But maybe worth reconsidering this pattern? 🤔
I have checks for existence only because cargo features might disable them.

Yeah, with resources as entities we can just insert server or client component on backend client or server 🤔

[thorn forum]

ohh yeah I got confused with the network backend resources which we add ourselves

[spring raptor]

You will still need to spawn backend entities.It's because the setup is backend-dependent.
But we will be able to attach replicon components to backends automatically which will be way nicer.

[raw idol]

to clarify, aeronet provides no auth, and leaves that as the responsibility of the user (but the IO layer is responsible for encryption, which is separate). I should probably document that

@spring raptor I was curious, how would you handle user accounts via replicon?

I'm starting to think that I need another intermediary layer somewhere above IO for Auth, but I can't quite think of a system for this yet

or rather, keep encryption as the responsibility of the IO layer, and have game-specific Auth above

[spring raptor]

I'm not sure, we just provide replication. I think that authorization and loading proper data from a database or something else is for transport/user implementation.

[raw idol]

so that would be below replicon?

[spring raptor]

I think so 🤔

Not sure what special we can provide to help

[raw idol]

well normally replication starts immediately after connection right

perhaps a "connect", "authenticate", "start replicate"

[spring raptor]

Thinking about your use case. As a temporary workaround you can have something like End state where you show the end screen. And InGame where you run your logic.

[spring raptor]

Only when the transport says that you are connected. But I think it won't say so until you finish your authorization? 🤔

[raw idol]

true, but all of the current transport implementations immediately send a replicon connect after a client connect

[spring raptor]

In renet auth is baked into connection

I.e. you receive a token

[raw idol]

renet is a bit different to what I'm thinking, since a part of the auth process happens before there's even a connection, i.e. talking to an http service to generate and request a token

[spring raptor]

So the workflow looks like this: connect to authorization server, it generates the token and then you connect to the game server.

[raw idol]

yeah

id like to do all the auth in one shot, with one server

anyway I'm gonna go to sleep and think about this a bit more

[spring raptor]

Then maybe you need something on top of Replicon, not sure

[spring raptor]

Yep, let me know how you think it could look like, I'm open to provide the necessary API.

Have a good night 🙂

[echo lion]

I disagree with this, connecting to your backend infra as a 'user' is distinct from authenticating as a 'player' in a specific game instance. Renet's connect token workflow is robust and effective.

[spring raptor]

I think there are valid use cases for it 🤔
I remember playing SAMP and you could connect to a server via password.

Or Counter-Strike

[echo lion]

There’s a good chance these cases are just logging into the backend and then the backend issuing connect tokens to specific servers.

[thorn forum]

I already have loading, countdown, running, ended, error 😆 I'll do a custom run condition to check if the logic needs to be run

[spring raptor]

Makes sense!
Yes, I would check if your game logic run in running state additionally to server_or_singleplayer.

But I planning to fix it with the upcoming rework of replicon resources as entities.

[thorn forum]

nice :)

[raw idol]

renet's way is one valid way of doing it, and I do like it. I think it works nice when you have a central authority for accounts, and it should be a supported way of doing auth. however, that's not the architecture that I would like for my game, since I want a "one-shot" auth + connect flow, where each server manages its own accounts and there is no central authority. and this is a perfectly valid way of doing auth as well.

in general, I'd like both ways to be possible, and require switching between the two to require minimal code changes (at the actual "connect" stage anyway - if you need an auth token from some other authority, then there's obviously more work you have to do)

I'm just not sure where this should live in the networking stack. right now, I'm thinking above the transport layer but below replication.

[thorn forum]

about my Added<Troop> being re triggered on the client on disconnect, i fixed it, but i would like to understand how it happened.
i used bevy track change detection to print the troop's changed_by():

2025-02-10T12:50:15.449560Z  INFO bevy_dev_tools::states: app::game::GameState transition: Some(StartCountdown) => Some(Running)
2025-02-10T12:50:41.337506Z DEBUG app::card::systems: handling Added<Troop> entity 877v6. Troop changed_by at Tick { tick: 309739 } by [...]/bevy_replicon-0.30.1/src/core/replication/replication_registry/command_fns.rs:93:42
2025-02-10T12:50:47.547765Z  INFO app::game::systems: received EndGameEvent(Some(ClientId(1739191812323)))
2025-02-10T12:50:47.564847Z  INFO bevy_dev_tools::states: app::game::GameState transition: Some(Running) => Some(Ended)
2025-02-10T12:50:48.570843Z ERROR renetcode2::client: Failed to update client: disconnected: connection terminated by server
2025-02-10T12:50:48.605955Z DEBUG app::card::systems: handling Added<Troop> entity 877v6. Troop changed_by at Tick { tick: 309739 } [...]/bevy_replicon-0.30.1/src/core/replication/replication_registry/command_fns.rs:93:42

i dont spawn or insert any Troop after the first one. so i dont know why Troop is marked as just added.
bevy's changed_by() should show from where its been added, but its showing the first add from replicon, not the second one which is added from nowhere.
could it be a bug with bevy's change detection ? because the two last_changed() are at the same tick and location, but just changed to added true (and so changed true too)

[thorn forum]

i dont insert or modify the Troop component or &mut it anywhere, only &Troop. and its still marked as added + changed

[spring raptor]

Agree, it's better to investigate! Let's take a look together.

So, backend (I assume you use renet2) systems update RepliconClient state in ClientSet::ReceivePackets.

Then we run ClientSet::Receive (the order is configured here).
Which contains the system that applies the replicated data here.

To me it's configured as expected. Client becomes disconnected first, then run receive_replication only if we are conneected 🤔
But maybe the condition under run_if is evaluated at the beginning of the schedule? That would explain the issue.

[thorn forum]

yep renet2.
from what i know and searching in the discord, run condition are run right before running the system.
but maybe some systems ordering are not clearly set when multithreading ?
i dont know where and how to debug this 🤔
my problem is fixed, so if you dont want to spend time on this no problem, ill just go next too.

we agree that its not normal for bevy's detectchanges to register the component was re-added, but it doesn't have the tick or location info ?

detect change impl: https://docs.rs/bevy_ecs/0.15.2/src/bevy_ecs/change_detection.rs.html#273-318

[spring raptor]

There is a way to check it. You can try to enable logging for replicon and see if a message is processed even after disconnect

If you want to investigate

[dire aurora]

Run conditions are ran before the first system that needs it in that schedule, then the rest usually reuse that result I think

[thorn forum]

oh yeah, ill try

[thorn forum]

interesting !

[thorn forum]

i found the problem, sorry for wasting all your time...
the system with the Added<Troop> was running on Update on the client and FixedUpdate on the server.
the fixedupdate system ran after disconnection because of the server_or_singleplayer. yesterday i tried removing it to assert it was the client update system that was running, but i must have failed to do it correctly. the second log was from the system in fixedupdate running for the first time, and so the Added was true. i learned that the same system in different schedule have different states

[spring raptor]

It's okay, I glad that we figure it out :)

I going to fix server_or_singleplayer in the upcoming release with backends integration rework. I think we can do it even without relying on Bevy main 🤔

[pale charm]

is there any example how to use replicon with avian3d? For some reason i'm having a hard time figuring it out ....

[dire aurora]

In the most basic form you'd only need to mark a couple of avian's components for replication ... Transform (or Position+Rotation), LinearVelocity, AngularVelocity, and possibly some component that lets the client know what collider and rigid body properties to set

[pale charm]

ah thanks, no idea why i havent tried that tho.... thank you very much

[spring raptor]

Why kind of game are you planning to make? Most likely you don't want to use replicon with avian without prediction and rollback 🤔

[pale charm]

just playing around. i did a small game for a gamejame with lightyear and wanted to try replicon now. But my brain seems to be afk today, so i'm struggling with pretty basic stuff

but thank you very much for your help 🙂

[spring raptor]

Ah, got it!

If you want to build a physics-based game you need a special technique called "client-side prediction". The idea is to remove input delay by applying all actions immediately locally and when server confirms the changes, correct the local state.
Under the hood it rollback your game state in time, applies server changes and re-plays local input on top of it.

And there are 2 ways of doing it:

1. Rollback pre-entity. It's cheap, perfert fit for shooters, but not so good for physics-based games or for games where you have a lot of stuff to rollback (it's cheaper to rollback the whole world instead of checking each entity individually for rollback). That's what Lightyear provides.
2. Rollback the whole world. Usually more performance-heavy, but perfect fit for physics-based games or games with a lot of rollbackable entities.

Replicon is completely abstracted from it and allows to build any of these methods on top. But the problem is that we currently don't have a crate that does it 😅
@dire aurora is working on a crate for 2, but quite busy with other stuff recently.

I planning to give a talk about replication on the next Bevy meetup, I will try to cover these topics 🙂 Maybe this will encourage someone to write a crate for 1.

[pale charm]

thank you for your explanation. i'm looking forward to your talk 🙂

i really love how easy it was to setup replicon and how helpful the docs where. damn good job

[dire aurora]

Would also be nice if we got something for lockstep ... Lockstep + some input delay is a fairly common approach too for games that can more easily place limits on how much latency you'll have (like games that have lots of regions the way LoL does it, player-hosted games, or projects that are designed to be more local)

[dire aurora]

@spring raptor How are we actually supposed to use the example backend? It just throws this panic:

bevy_replicon_example_backend::server::set_running could not access system parameter ResMut<RepliconServer>
```Which happens because the example backend uses the the one from the git dependency, while your code uses the versioned replicon 🤔

[spring raptor]

Damn, I forgot that third-party crate won't be able to use it this way...
Should I publish on it crates.io then?

I think it's quite explicit that it's just for example 🤔

[spring raptor]

@dire aurora will it work if I set the version of replicon in Cargo.toml?

[scarlet imp]

Hello! I'm updating some of my game's server events to use the new server triggers, but when I run it as a listen server only the non-server clients are receiving the triggers, not the one running the listen server. It runs as expected in single player mode, and it worked previously as a server event. The server_trigger documentation mentions triggering locally if ClientId::SERVER is a recipient of the event, but I can't find how to make it a recipient. Using SendMode::Direct(ClientID::SERVER) as the mode in the trigger also didn't appear to work. I'm was hoping to get some help with that. I've got the following code snippets:

#[derive(Event, Serialize, Deserialize)]
pub struct SetTurnBasedState {
    pub state: TurnBasedState,
}```
```rust
app.add_server_trigger::<SetTurnBasedState>(ChannelKind::Ordered);
app.add_observer(receive_turn_based_state);```
```rust
commands.server_trigger(ToClients {
    mode: SendMode::Broadcast,
    event: SetTurnBasedState {
        state: TurnBasedState::RoundBeginning,
    },
});```
```rust
fn receive_turn_based_state(
    trigger: Trigger<SetTurnBasedState>,

) {
    warn!("Received new turn based state: {:?}", trigger.state);
}```
Thanks for the help!

[spring raptor]

Could be a bug... Going to sleep right now, but I'll take a look tomorrow.

[scarlet imp]

Thanks! I managed to build a workaround, but I'd be happy to provide any additional context tomorrow if it would help. 🙂

[dire aurora]

Maybe? That's how you do it for published vs unpublished crates, just put both a version and path, not sure what that does for git version tho

[spring raptor]

Just checked, it's only for releases on crates.io

I'll just publish it. For convenience the version number will match the replicon version.

[spring raptor]

@dire aurora but I want to hear your opinion as a user on this ^ first before I publish 🙂

[thorn forum]

i was also wondering how to trigger an event to both the clients and the server at the same :)

[spring raptor]

It's a bug, I will push a fix after work later :)

[thorn forum]

oh ok thanks :)

[spring raptor]

It should work exactly as remote events.

[dire aurora]

Should be fine, as long as crates.io also has the warning the readme in that subfolder has

It takes some real mental gymnastics to use a crate with such a warning and example in all the names as a real backend 😂

[spring raptor]

Do you think using the same version is fine? Just thinking, there might be some cases where we want to update the example backend separately 🤔

[dire aurora]

Could use same major but different patch releases

I don't think pushing features to just the example backend outside of a replicon update would be particularly worth it at least

[spring raptor]

@dire aurora Published for 0.30:
https://crates.io/crates/bevy_replicon_example_backend
As discussed, I'll maintain the same minor and major versions, but different patch versions.

I will adjust the readme on crates.io in the next release. It contains the warning, but lacks some context 😅

[dire aurora]

Quite a surprising error to get ... Did StartReplication become a trigger? 🤔

Requested resource bevy_ecs::event::collections::Events<bevy_replicon::server::StartReplication> does not exist in the `World`

[spring raptor]

Yes!

I mentioned it in the changelog

[dire aurora]

I guess this just shows how painful the Trigger vs EventReader thing is with Event merged as one concept

[spring raptor]

Yes, it makes the migration awful :(

And you can't know if the type is a trigger or an event by trait impls, you need to rely on authors to explicitly state it in the docs.

[dire aurora]

Having to integrate both loading of inputs while resimulating and listen servers really makes me need to rethink some of the design of my input queue and rollback stuff ...
Questions like:

• How do I avoid generating new inputs while resimulating
• How do I make sure the listen server's inputs have been written before sending
• How can I make the input queue load inputs instead of expecting the generation when it doesn't know rollback exists

[spring raptor]

Could you describe these issues in more details? 🤔

[dire aurora]

So the first one is pretty obvious, as a part of FixedMain you need to generate inputs, but only if it's a real frame, and not a resimulation. I could do some run condition or system set dance but that seems very error prone

The second is a bit more specific to how I've set things up. I receive inputs from clients, use them to update the input queues, then send the inputs. I do this in PreUpdate, however if you are a listen server you also generate inputs, but in FixedMain, and thus miss them. If you send inputs after FixedMain (for example in PostUpdate), you can miss updates because they got consumed from the InputQueue while the listen server was running

For the last one I need resimulations to load inputs from the input history for all players, including the one with authority, instead of only trying to load it if there is no authority (they are unlikely to be there, but I check it just in case anyway, since it can avoid some mispredictions)

[spring raptor]

Ah, interesting!
But isn't server is the source of thruth?
I thought that you don't need to re-simulate inputs when you are a listen server 🤔

[dire aurora]

On the lister server we don't re-simulate, I just need the server to send an accurate list of inputs before it starts consuming them

So clients have inputs for resimulating

[spring raptor]

Ah, got it now!
I have an issue similar to 2 in my input crate. I need to read inputs in PreUpdate but in FixedMain I need to output it at specific point in time.
LWIM does it, but I still didn't get to it.

Not the same as your problem, but related

[dire aurora]

Yea ... They sort of come from the same source, which is the fact that we can't really determine during PreUpdate where the inputs should go

If we could, I could easily populate my input history in PreUpdate and immediately send the inputs (now including the host's inputs)

I guess I'll just ignore this specific problem for now and send the inputs at the end of the frame instead, it contains 3 of the past inputs too, so the server needs quite a bit of lag for it to matter 🤔

Still leaves the other two issues ... I guess the simple solution for the first one is to just generate inputs in FixedPreUpdate, I only resimulate FixedUpdate by default, and if you change things you can just use a custom schedule ... The last one ... I guess I should just make it a bit smarter with when it loads, inputs aren't stored until FixedPostUpdate, so if I only load something if there is history I think the problem should be non-existent, but ofc the responsibility to clear the previous frame's input is now in the developer's hand ... Which I guess is fine, it's their type anyway 🤔

Actually I guess the last issue would also not exist if inputs could be generated ahead of time, then you could just always load from the history

Seems to work, but then I suddenly hit this:

bevy_replicon-0.30.1/src/core/replicon_client.rs:209:32:
client should have a channel with id 98
Encountered a panic in system `bevy_replicon_example_backend::client::receive_packets`!

Also seems to happen the other way around 😂

bevy_replicon-0.30.1/src/core/replicon_server.rs:175:32:
server should have a receive channel with id 10
Encountered a panic in system `bevy_replicon_example_backend::server::receive_packets`!

[spring raptor]

Looks like a deserialziation issue. Or do you have 98 client events?

[dire aurora]

Pretty sure I don't 😂

The only events I register:

.add_client_event::<Connect>(RepliconChannel::from(ChannelKind::Unordered))
.add_server_event::<CurrentTick>(RepliconChannel::from(ChannelKind::Unreliable))
.add_client_event::<LocalEntities>(RepliconChannel::from(ChannelKind::Unordered))

[spring raptor]

Looks like an issue on my side related to the example backend.
I have tests, but maybe it triggers only with multiple channels.

Here is how I read an write messages, quite simple: https://github.com/projectharmonia/bevy_replicon/blob/master/bevy_replicon_example_backend/src/tcp.rs

Just tried with multiple channels, still can't reproduce

Could you show how the events look like?

[dire aurora]

Oh right and these too ofc:

app.add_mapped_client_event::<InputHistory<T>>(ChannelKind::Unreliable)
    .add_mapped_server_event::<HistoryFor<T>>(ChannelKind::Unreliable);

#[serde(bound(deserialize = "T: for<'de2> serde::Deserialize<'de2>"))]
pub struct InputHistory<T: InputTrait> {
    list: VecDeque<T>,
    updated_at: RepliconTick,
}
``````rust
#[serde(bound(deserialize = "T: for<'de2> serde::Deserialize<'de2>"))]
struct HistoryFor<T: InputTrait> {
    entity: Entity,
    tick: RepliconTick,
    past: ArrayVec<(u8, T), 3>,
    future: ArrayVec<(u8, T), 7>,
}
```The T in question:
```rust
struct GameInput {
    direction: Option<Dir2>,
}

struct Connect;
struct CurrentTick(GameTick);
struct LocalEntities {
    car: Entity,
    ball: Entity,
}

It seems to happen after some period of time, never right away 🤔

[spring raptor]

Strange, maybe there is some weird TCP thing I didn't take into account 🤔 I will try to reproduce.

But let's start with a few simple things

[spring raptor]

Fixed, thanks: https://github.com/projectharmonia/bevy_replicon/pull/414

[scarlet imp]

Thank you! I'll give it a shot a bit later today.

[spring raptor]

If you want to apply the fix, don't forget to use patch section in Crates.io (to avoid the issue with backends depending on a release version)
I planning to draft a new release soon.

@dire aurora could you try with trace logging enabled? Just to see that the amount of sending bytes matches the amount of received.

I.e. RUST_LOG=bevy_replicon=trace cargo run ..

[dire aurora]

It doesn't actually show the number of received bytes anywhere 🤔

[spring raptor]

Ah, it only shows the number of messages...

[spring raptor]

Could you try this branch?
https://github.com/projectharmonia/bevy_replicon/pull/416
This will require you to switch to the latest master which contains the postcard migration. We have a changelog, but maybe seeing this PR will also help (just look at the docs changes): https://github.com/projectharmonia/bevy_replicon/pull/410

[dire aurora]

The world's most painful migration for sure:

-    cursor: &mut Cursor<&[u8]>,
-) -> bevy_replicon::bincode::Result<()> {
+    cursor: &mut Bytes,
+) -> bevy_replicon::postcard::Result<()> {

And changing the example backend version to 0.1 for unknown reasons 😂

client:

2025-02-12T17:31:57.629624Z TRACE bevy_replicon::core::replicon_client: sending 13 bytes over channel 2
2025-02-12T17:31:57.645861Z TRACE bevy_replicon::core::replicon_client: sending 2 bytes over channel 0
2025-02-12T17:31:57.646642Z TRACE bevy_replicon::core::replicon_client: sending 13 bytes over channel 2
2025-02-12T17:31:57.662504Z TRACE bevy_replicon::core::replicon_client: sending 2 bytes over channel 0
2025-02-12T17:31:57.663042Z TRACE bevy_replicon::core::replicon_client: sending 13 bytes over channel 2

server:

2025-02-12T17:31:57.646761Z TRACE bevy_replicon::core::replicon_server: received 1 message(s) totaling 2 bytes from channel 0
2025-02-12T17:31:57.646768Z TRACE bevy_replicon::core::replication::replicated_clients: ClientId(56474) acknowledged mutate message with Tick { tick: 441520 }
2025-02-12T17:31:57.646780Z TRACE bevy_replicon::core::replicon_server: received 0 message(s) totaling 0 bytes from channel 2
2025-02-12T17:31:57.646788Z TRACE bevy_replicon::core::replicon_server: received 0 message(s) totaling 0 bytes from channel 3
2025-02-12T17:31:57.646792Z TRACE bevy_replicon::core::replicon_server: received 0 message(s) totaling 0 bytes from channel 4
bevy_replicon-6598e87287870c1c/13d2c93/src/core/replicon_server.rs:179:32:
server should have a receive channel with id 10
Encountered a panic in system `bevy_replicon_example_backend::server::receive_packets`!
2025-02-12T17:31:57.663598Z TRACE bevy_replicon::core::replicon_server: received 0 message(s) totaling 0 bytes from channel 0
2025-02-12T17:31:57.663613Z TRACE bevy_replicon::core::replicon_server: received 0 message(s) totaling 0 bytes from channel 2
2025-02-12T17:31:57.663617Z TRACE bevy_replicon::core::replicon_server: received 0 message(s) totaling 0 bytes from channel 3
2025-02-12T17:31:57.663621Z TRACE bevy_replicon::core::replicon_server: received 0 message(s) totaling 0 bytes from channel 4

Before that the server still seemed to receive the constant 13 bytes over channel 2 and 2 over channel 0 fine 🤔

[spring raptor]

Do you have a lot of custom serialization?

[spring raptor]

Yeah, it's because on the latest master it's still 0.1, but I already have a PR opened for it.

[spring raptor]

I think I have an idea, let me quickly try one thing

[dire aurora]

I have none in this example, this one place is just the authoritative write function :')

[spring raptor]

My theory is that's because I write into the socket in consequential writes. But I need to write it in one go.

[dire aurora]

Yea that was also my theory, but I don't see why that would be the case ... Maybe if it's non blocking and it picks up a partial message first, then sets it up for failure later? 🤔

[spring raptor]

I suspect that it might picks only channel or size, then hits WouldBlock, exits, but the stream have these bytes read now.

I pushed a small change to tests this theory. Could you try it again?

[spring raptor]

Pushed a less ugly fix to avoid extra allocations.

@dire aurora sorry, pushed to a wrong branch 😅 Now should be available at https://github.com/projectharmonia/bevy_replicon/pull/416

[dire aurora]

Not sure if it fixes it (since it's a pretty rare bug) but at the very least it's harder to hit now 🤔

I guess to really make it reliable the reading side should also ensure it either reads everything or nothing 🤔

Otherwise it could still go wrong on ticks with a lot of data that might cause things to be split up over different packets

[spring raptor]

Let me know if it triggers again. But I think that the change is good anyway, it's a single syscal now.

[spring raptor]

I considered this, but I don't know the size ahead of time :(

[dire aurora]

The common approach would be to peek the first 3 bytes, then if that fails you continue, but if it succeeds you read the whole message

Does require a buffered reader tho I think

[spring raptor]

Ah, I didn't know that TcpStream provides peek. Let me adjust the receive impl as well

[dire aurora]

Oh it does have a peek ... Nice so it's basically free to fix

[spring raptor]

Pushed

[spring raptor]

I managed to replace &World in the replication sending system with FilteredEntityRef. It allows other systems that write to non-replicated components to run in parallel. But I started all this to to make it work nicely with server as a component 😅

The logic is almost the same, but I access the pointers via FilteredEntityRef.
The downside is that it's 2x slower (!) due to the extra lookups.

However, I think I can create a custom system parameter that provides access to the replicated components, just like before, but with granular access similar to FilteredEntityRef 🤔

https://github.com/projectharmonia/bevy_replicon/pull/417

[dire aurora]

Which lookups exactly are causing the slowdown?

[spring raptor]

Access check validation and component storage type fetch. I think I can get rid of all of this by writing my own system param.

[dire aurora]

You mean the switch from the unsafe get_component_unchecked to get_by_id(id).unwrap_unchecked)?

[spring raptor]

Yep, basically this:
https://docs.rs/bevy_ecs/0.15.1/src/bevy_ecs/world/entity_ref.rs.html#2412-2418
that calls this:
https://docs.rs/bevy_ecs/0.15.1/src/bevy_ecs/world/unsafe_world_cell.rs.html#950-962
Is slower then my old approach where storage type were cached with no access checks.

[dire aurora]

Interesting ... I'm surprised it makes that much of a difference ... I guess this just shows how efficient the architecture is otherwise ... Iterating over archetypes and using an archetype cache is pretty cheap after all 😅

[spring raptor]

Yeah, I also didn't expect it 🙂
I will try wrapping UnsafeWorldCell into a SystemParam with granular access. Will work just like before, but Bevy should be able to parallelize it nicely. And without that ugly ParamSet.

[dire aurora]

Ugh, it seems like just resimulating FixedUpdate won't work ... I wonder if I should just recommend the approach I did for my game (a custom PreSimulation, Simulation, and PostSimulation, in a SimulationMain, which is run in FixedUpdate and by rollback)

In case anyone is wondering what the issue is: There are certain things that should always happen before and after the simulation, and you really don't want to schedule every system around them ... Things like loading inputs and storing the new prediction

[spring raptor]

Makes sense to me. Unless until Bevy provides a nicer solution

[spring raptor]

Done: https://github.com/projectharmonia/bevy_replicon/pull/418
The performance is the same for single thread and Bevy can parallelize it better.

[spring raptor]

But my main goal is to make components as entities to work nicely.
Now I need something similar for the receive side. But the problem is that user can write data into non-replicated components 🤔

[spring raptor]

For this I'll wait for actual "components as entities". Because unlike with components, you cannot define access as "all resources except...".

So RepliconServer and RepliconClient will be resources for now.
But we can do some other cool API improvements already 🙂

[dire aurora]

I think main might actually have that capability because of a somewhat niche feature rendering people wanted 🤔

[spring raptor]

Fixed the UB I caused 😅
https://github.com/projectharmonia/bevy_replicon/pull/419
Now it's implemented properly.

[spring raptor]

@dire aurora quick question: do you need full world access in write functions or only ComponentId?

[dire aurora]

I need read access to some specific resoureces too, we could make a way to whitelist those tho 🤔

[spring raptor]

Yeah, this could work.
And the same for components.

[dire aurora]

You mean like non-replicated components you can modify?

That would be necessary too I guess, my write functions are for any component, but they always write to the AuthoritativeHistory component

[spring raptor]

Yep, additional non-replicated components.

I also have another idea to try: provide a special world that blocks access to resources used by replication receiving system.

[neon canopy]

So folks, [I'm looking into still setting up a basic synced app. ](#networking message)

And I know @spring raptor has helped me out but I'm still struggling to set up a good way to set up synced camera and a synced up mesh, which I plan to be changing in the future. Basically 2 seperate apps should have the transform components of their cameras synced up as well as the Mesh3d shape and it's texture as well. I haven't even gotten to syncing textures yet and I'm not even sure if swapping the path on the material will even be the way to do it. But for now just having a camera perma synced would be massive help.

I tried doing the replicated components path and a Camera marker component and it replicates, but it's not too fab as you get a No Camera warning a few times until replication starts. So maybe there's a thing there of creating a window on the client only when the replication starts.

And the second thing is that in the required component path I need a nice way to access my global settings resource as the z value of the camera position is set in the settings, which I'm struggling to grok how to do.

Maybe I should start a thread in the help forum if this would be more helpful.

[spring raptor]

but it's not too fab as you get a No Camera warning a few times until replication starts
That's expected. In actual game you most likely need a camera to draw menu stuff and when you receive the replication, just despawn the original camera via a trigger.

I need a nice way to access my global settings resource
For that you want to mix it with the blueprint pattern: create a trigger that listens for in-game camera insertion and just update the value from settings.

Maybe I should start a thread in the help forum
This thread is a good place to ask questions like this.

[thorn forum]

on wasm im getting a panic

replication_registry.rs:141:32:
replication `FnsId(63)` should be registered first

im not using any fns or custom impl, only some replicate mapped
thanks :)

[spring raptor]

Interesting, it means that on deserialization we read number 63 in a place we shouldn't 🤔
When you do replicate_mapped, it registers a default FnsId. But I don't think you have 63 replicated components.

I remember that you using this branch https://github.com/projectharmonia/bevy_replicon/pull/405 to fix another problem that happens only in wasm. Am I right?

Does it happen constantly?

[thorn forum]

it happen everytime when connecting, but only on wasm with websocket renet2 (native socket no problems).
im using the branch trigger-listen-server. since i didn't encounter the panic after switching wifi, i stopped using the branch disabling it.
no i have two replicate_mapped component, and none of them are spawned when connecting

i triend branch main and same thing

[spring raptor]

Interesting, might be a transport issue then.
Ping @echo lion, maybe you have an idea how to debug it?

[echo lion]

The transport should not be mangling messages

[spring raptor]

You better use this branch instead, I recently introduced an UB in the latest master 😅

[echo lion]

So I imagine it’s a replicon issue, possibly UB

[spring raptor]

But it happens only on transport change 🤔

[echo lion]

It happens in WASM, so perhaps the UB has different effects

[thorn forum]

wasn't hapenning before i was using unsecure connection, and now it happen since using serverconnecttoken, but it can just be a change i made in my code ordering

[echo lion]

mirsella can you test webtransport? Should work on chrome

[spring raptor]

Okay, let's try to debug it. @thorn forum could you enable logging to see how much bytes we receiving and sending first. Just in case.

[thorn forum]

thanks, ill just use latest main commit then

[spring raptor]

No, no, not the lastest master, use the mentioned branch that fixes the small problem :)

[thorn forum]

oh its not yet merged mb

[spring raptor]

Yep, I pushed it yesterday, but forgot to request a review. @echo lion will merge it once he have time.
It's not related to your problem, just a fix for the previous commit in master.

[thorn forum]

with bevy_replicon=trace and with the fix-replication-read-ub branch

[spring raptor]

I see, a lot of bytes 🤔

Now could you show logging from the server as well?

RepliconTick(839), we definitely receive something weird

[thorn forum]

hm yeah seems a lot, dunno if thats just my usage, im replicating a few things on startup

[spring raptor]

Yeah, it's quite suspicious. The idea is to compare how much replicon sends and receives.
These are bytes without any transport data, just pure replication and events.

[echo lion]

Does that client id match your expectations? 552481

[thorn forum]

on client the "applying message for replicontick" is 51 like on the server

2025-02-17T16:51:06.967353Z TRACE bevy_replicon::server: incremented ResMut(ServerTick(RepliconTick(50)))
2025-02-17T16:51:07.016956Z DEBUG bevy_replicon::server: `ClientId(811578)` connected
2025-02-17T16:51:07.016986Z DEBUG bevy_replicon::core::connected_clients: adding connected `ClientId(811578)`
2025-02-17T16:51:07.017001Z DEBUG bevy_replicon::core::replication::replicated_clients: starting replication for `ClientId(811578)`
2025-02-17T16:51:07.017075Z  INFO app::game::replicon::systems: client ClientId(811578) connected
2025-02-17T16:51:07.017089Z  INFO app::game::replicon::systems: spawning castle for new ClientId(811578) at Tile(3, 4)
2025-02-17T16:51:07.017625Z TRACE bevy_replicon::core::replicon_server: received 0 message(s) totaling 0 bytes from channel 2
2025-02-17T16:51:07.017636Z TRACE bevy_replicon::core::replicon_server: received 0 message(s) totaling 0 bytes from channel 3
2025-02-17T16:51:07.017642Z TRACE bevy_replicon::core::replicon_server: received 0 message(s) totaling 0 bytes from channel 4
2025-02-17T16:51:07.017657Z TRACE bevy_replicon::core::replicon_server: received 0 message(s) totaling 0 bytes from channel 0
2025-02-17T16:51:07.018208Z TRACE bevy_replicon::server: incremented ResMut(ServerTick(RepliconTick(51)))
2025-02-17T16:51:07.018468Z TRACE bevy_replicon::server: sending update message to ClientId(811578)
2025-02-17T16:51:07.018478Z TRACE bevy_replicon::core::replicon_server: sending 8851 bytes over channel 0

had to truncate the logs for discord, but nothing before or after

nothing in the sense no changes or messages are sent

[thorn forum]

oh comparing how many are sent and recv, ok will do that

[thorn forum]

yeah im using current time as micros seconds for now since its anonymous

should just use ms since epoch instead

[spring raptor]

So you send 8851 bytes. That's quite a lot.
But could you run everything together again just to ensure that sizes match?

[thorn forum]

yep the sent bytes matches, both 8972 this run

[spring raptor]

That's a lot of data to send. I wonder what takes so much. Let me improve the logging.

[thorn forum]

i just tested its also this much data when connecting with native. but it doesn't crash on native. could it be something on wasm being buggy and that much bytes corrupt something ?

[spring raptor]

I suspect it might be caused by packet constant limit 🤔
Could you quickly try this branch? https://github.com/projectharmonia/bevy_replicon/pull/407

[thorn forum]

i set

[patch.crates-io]
bevy_replicon = { git = "https://github.com/newclarityex/bevy_replicon" }

but i get

warning: Patch `bevy_replicon v0.30.0 (https://github.com/newclarityex/bevy_replicon#673caad1)` was not used in the crate graph.
Check that the patched package version and available features are compatible
with the dependency requirements. If the patch has a different version from
what is locked in the Cargo.lock file, run `cargo update` to use the new
version. This may also occur with an optional dependency that is not enabled.

i ran cargo update and deleted cargo.lock to try but same thing

[spring raptor]

Strange. But it does work like this on the latest master?
If yes, clone the repo, and adjust the constant to some conservative value, like 1000 and use path = "<path to the folder>".

[echo lion]

It shouldn’t be a packet size limit, because the message arrives with the same size on the client

[thorn forum]

doesn't work i get serde function IDs should be obtained from the same instance on wasm and on native it works like always. i think its not really using the newclarityex:master branch

do i still try locally changing the limit ?

yes server and client native and client wasm where recompiled

[spring raptor]

If you get the warning above, then yes

[thorn forum]

panic

[spring raptor]

I meant this warning ^

[spring raptor]

I would try locally, yes

[thorn forum]

oh yeah

[spring raptor]

Maybe packets ordering is swapped 🤔

[thorn forum]

ill try locally.
btw i think the patched version doesn't work because its version 30.0 while your versin is 30.1, even if its patch version change maybe its incompatible since its a lower patch version

[echo lion]

mirsella do you use renet2_setup on both server and client?

[thorn forum]

yes

[echo lion]

Ok. I will try to test extra-big messages in an hour or so.

[thorn forum]

@spring raptor ive sent max_pack_size to 1000, but same thing. anyway even 1000 is still bigger than what im sending ~8900 bytes so it doesn't matter ?

[thorn forum]

only on wasm 😁

[echo lion]

Right I will test all the transports

[thorn forum]

do you still want me to try webtransport ?

[echo lion]

Sure that would help

[spring raptor]

Got it. I suggested a smaller value due to possibly incorrect splitting. We can't fit ~8900 in a single packet. So we split data across multiple packets.

Okay, one more thing to try :)

Could you add byte printing to the debug message locally?

[thorn forum]

oh okay, so no need to put it even lower

[spring raptor]

No, why?
Single packet size is limited. And replicon tries to split data up to maximum size. But it depends on the transport. It usually bigger then 1000. We currently have a hardcoded constant, but I planning to fix it in the next release :)
If you have the same error with 1000, then it's not what causes the issue.

[spring raptor]

Let's try this. Just print the whole message before sending and receiving. Just search for "sending {} bytes" and adjust the trace message

The idea is to compare what we send on client and server.

[thorn forum]

looking at my code, there shouldn't be even close to 8900 bytes sent, in not replicating a lot

ill try updating the trace

i just print the raw bytes u8 ?

[spring raptor]

Yes

[spring raptor]

I will add more detailed logging soon

[thorn forum]

btw i tried not spawning my tiles, and so not even sending start game event, and it didn't crash

https://pastebin.com/x3PjSJuQ

[spring raptor]

Says not available

[echo lion]

How many tiles do you replicate? That could add up to a lot.

[thorn forum]

oh sorry i thought "unlisted" would work instead of public. https://pastebin.com/0ZLwHC1F

[echo lion]

Also not available

[thorn forum]

144. whats replicated on a tile, is a transform, a marker component, and a u32

wtf

they're marked as public but doesn't work

[echo lion]

Transform is fairly big, looks like each tile is probably 62 bytes.

Although that does seem a little too big, not sure

[thorn forum]

https://textbin.net/cucbciqf9f

[spring raptor]

That's one. But I need both, server and client to compare

[thorn forum]

oh ok sorry

[thorn forum]

oh there also Name replicated on debug mode, and when the client connect, a new building is spawned for it, which is max 200 bytes

[echo lion]

Hmm well transform is ~40bytes afaict. Name is probably 10bytes or so.

[thorn forum]

do you want me to just check if the message on the server is the same as on the client ? simpler than sending two pastebin

[spring raptor]

This also works :)

[thorn forum]

what the hell i worked

ill re try it

the only two things that changed is the message length is smaller, and especially running the server in release

the web wasm app was always compiled in release, while the server was in debug. the serialization are different i guess

[echo lion]

Interesting, strange

[thorn forum]

its working everytime now

[echo lion]

Maybe this is the problem. If Name isn’t registered in release mode, the fnsid won’t be known by the client. But idk why the fnsid would be 63.

[thorn forum]

thanks as always for all your helps :)))))

i tested more, and it seems to be this, Name being replicate on server, and not on client

i tried debug server, release client, and it work as long as both have the same thing for replicate Name

but the amount of data sent, and the number of FnsId is still weird 🤔

[spring raptor]

Probably just reads some weird value due to how data serialized

Name is quite huge, so I don't see anything suspicious

[thorn forum]

I didn't see the problem being debug vs release, because usually when I mismatch compilation mode I get "serde function IDs should be obtained from the same instance" but not this time

[spring raptor]

Not sure if get you. Could you elaborate?

[thorn forum]

when I run the client in debug and server in release, the app panic with this message "serde function IDs should be obtained from the same instance"

but with release client and debug server it didn't panic, so I didn't see the two had different compilation mode

[spring raptor]

Ah, I see, interesting 🤔

Maybe we should send a hash of the defined protocol? Just to avoid footguns like this.

[echo lion]

Yeah good idea. Can send it in the first init message to clients. Might need to use one of the message bits for it.

The error message can include a list of possible reasons for mismatch: rule registration order not the same, replicon channels not the same, different features affecting replicon rules enabled on server/client, etc.

[spring raptor]

We won't be able to detect this from the hash. Do you want to send more detailed information?

[echo lion]

No I’m saying the error message has a list of possible reasons it can not match

When you check the hash and it doesn’t match

[spring raptor]

Ah, agree!
Opened: https://github.com/projectharmonia/bevy_replicon/issues/420
I will try to take a look into it soon. Want to rework of sending and receiving first.

Working on receiving right now

[spring raptor]

I had previously finished sending, but today I opened a PR with a small fix.

[spring raptor]

Won't work.
For receive we expose Commands that can remove resources.
Unless we provide custom commands that refuse to remove some of the resources 😅

[spring raptor]

And the same problem with whitelist

Our current approach with removing all required resources isn't bad.
I just can't do the same with components.

Ah, we have https://docs.rs/bevy/latest/bevy/ecs/prelude/struct.EntityWorldMut.html#method.take
Never mind 🙂

[dire aurora]

Wouldn't this cause an archetype move? 🤔

[spring raptor]

Yes, but it's only for a single RepliconClient component on a single entity.

[dire aurora]

Oh, that shouldn't be a major issue then ... Not really sure why any other code would even try to touch it tho 🤔

[brittle mulch]

Might I ask how one implemente i nterest management? With replicon

[dire aurora]

Depends on what you mean by interest management ... It could be as simple as using per-entity visibility or as complex as needing to add features to bevy_replicon 🤔

[spring raptor]

Yeah, I don't think anyone would touch it. But if someone did, it would lead to UB without the removal.

[spring raptor]

Yeah, it's a bit of a broad term.
We provide a per-entity toggle: https://docs.rs/bevy_replicon/latest/bevy_replicon/core/replication/replicated_clients/client_visibility/struct.ClientVisibility.html
With a policy on the server plugin: https://docs.rs/bevy_replicon/latest/bevy_replicon/core/replication/replicated_clients/enum.VisibilityPolicy.html
It's fast and opt-in.
Third-party plugins can built on top of it things like this: https://github.com/UkoeHB/bevy_replicon_attributes
It would be nice to have a plugin that automatically toggles visibility based on distance or visibility 🤔 Shouldn't be hard to implement.

[dire aurora]

I mean fwiw it's a broad term because you kind of need all the approaches to end up with usable interest management ... Just per-entity visibility isn't enough, you want component visibility, priority + packet limits, send rates, etc

[spring raptor]

Yep, component visibility, priority, and other features are planned. I don't think it can be implemented as a third-party crate. I just haven't gotten to it yet 🙂

[pulsar garnet]

Heya! I'm going through the quickstart guide, everything is working so far except the ClientConnected trigger.

I followed exactly as it appears here https://github.com/projectharmonia/bevy_replicon/blob/master/bevy_replicon_example_backend/examples/simple_box.rs#L40 but the trigger is never called, even though I can see replicated stuff showing up in the client! I'm using the bevy_replicon_renet backend, is there something I need to do with it to get the connected trigger going?

[pulsar garnet]

Ahh, I was using MinimalPlugins on the server, switching it to DefaultPlugins made it work. Wonder which bevy plugin was missing

I'm getting these logs even using DefaultPlugins, are there specific bevy features that need to be enabled for replicon to function properly? I have default-features = false in cargo.toml.

2025-02-20T00:37:38.947731Z  WARN bevy_ecs::system::system: System 'bevy_replicon::server::reset' has not run for 3258167296 ticks. Changes older than 3258167295 ticks will not be detected.
2025-02-20T00:37:38.947749Z  WARN bevy_ecs::system::system: System 'bevy_replicon_renet::server::RepliconRenetServerPlugin::set_stopped' has not run for 3258167296 ticks. Changes older than 3258167295 ticks will not be detected.

[pulsar garnet]

The quickstart uses MinimalPlugins which is a bit confusing though, if ClientConnected isn't expected to work with it

[spring raptor]

It should just work because we rely on the connection event in Replicon itself: https://github.com/projectharmonia/bevy_replicon/blob/af9129368e589d8988dbaaaebb8418c398a00cc8/src/server.rs#L177
If you see replication happening, then the event was triggered 🤔

[spring raptor]

Didn't know about this warning!
But it's expected, reset and set_stopped run on server shutdown. This is why it says "has not run for N ticks".

In the next release these systems will be triggers, so you won't see the warning.

[pulsar garnet]

strange.. I might try figure it out later but for now happy enough to use default plugins.

I'm curious if there's a way to have a "replicate only once/on insert" kind of behaviour. I have logic components that are replicated that are basically what certain units are doing, like "move to X". With those replicated the same behaviour will play out on each client. The only issue is when a new player connects they don't get the other units current Transform because I don't have it replicating, so they see the other unit walking from Transform::default().

My best idea is to query for Query<&Transform, With<Replicated>> on the server when a player joins and send them all over with an event to patch the initial positions.

[spring raptor]

It's a planned feature: https://github.com/projectharmonia/bevy_replicon/issues/331

As a temporary workaround you can create a custom InitialTransform component that on insertion updates regular Transform.

[pulsar garnet]

Ah yea, makes sense. I initially expected Replicated to be an enum with that kind of choice, for what it's worth 🙂

I don't know if that would help, I'm trying to avoid syncing positions constantly, if I replicated InitialTransform wouldn't that leave me in the same boat? Sorry if I didn't make the criteria I was going for earlier more clear :). I'm going with click on ground to move, so replicating a component that says "moving to x,y" is the only thing that needs to be transmitted ideally (assuming the unit starts in roughly the correct place 😛 )

[spring raptor]

I, right, my suggestion won't work...
I guess yes, you will need to send positions manually for now.
Damn, I need to get to this feature asap 😢

[pulsar garnet]

no rush ^^

[spring raptor]

We also have writing function that provides world access.

https://docs.rs/bevy_replicon/latest/bevy_replicon/core/replication/command_markers/trait.AppMarkerExt.html#tymethod.set_marker_fns

It's executed after receiving on client. This way you can override writing into Transform, for example, into something else.

Serialization happens as usual.
What is your use case?

Have you considered using reflection?

It's a bit more expensive, but if you need to send dynamic data, it might work

That'll work, yeah

But you will have to track changes on your own

Why not use reflection?

@vocal violet this example is for event, but you can do the exact thing for components:
https://docs.rs/bevy_replicon/latest/bevy_replicon/core/event/client_event/trait.ClientEventAppExt.html#tymethod.add_client_event_with

We provide access to the type registry

[spring raptor]

@dire aurora do you need access to client's ticks on server? 🤔

[dire aurora]

In what context? 🤔

[spring raptor]

I working on a component-based API. The idea is to replace these resources:
https://docs.rs/bevy_replicon/latest/bevy_replicon/core/replication/replicated_clients/struct.ReplicatedClients.html
https://docs.rs/bevy_replicon/latest/bevy_replicon/core/connected_clients/struct.ConnectedClients.html
https://docs.rs/bevy_replicon/latest/bevy_replicon/server/client_entity_map/struct.ClientEntityMap.html
that map ClientId into data with entities with data as components. It makes it way more ergonomic. And I additionally split ReplicatedClient into ClientVisibility and ClientTicks. And I wonder if this ClientTicks even needs to be exposed. It stores methods like mutation_tick, update_tick, etc (like ones on the current ReplicatedClient).

[dire aurora]

There are context where the acked ticks are useful, but you'd probably want to explicitly receive it then anyway

Like delta compression

Otherwise I don't really use it

[spring raptor]

It's for listen server and singleplayer case. Useful if your game needs to work in both modes.

No, singleplayer is when your server is not running.

No, listen server is when server is also a client. All events just getting re-emitted locally. And other players can connect. No packets actually sent locally for efficiency, it's just work.
Singleplayer mode is when you don't use server at all.

Of course you can implement singleplayer by running server locally, but with replicon it's not needed.

In the quick start guide I tried to explain the trick we use to support all modes. Just navigate to the events section because for listen server you don't run replication locally.

[spring raptor]

You don't send packets. You send events. And events are re-triggered locally. See the events section in the quick start guide.

You can also try running any example in singeplayer mode

@vocal violet
To simplify:

1. When you emit ToClients<T> on server and singleplayer, it's also emitted locally as T.
2. When you emit T it's also emited as FromClient<T> on server and in singeplayer mode.
   In singleplayer server won't be running, but events will still be re-triggered.
   Let me know if I can somehow improve the quick start guide to make it more clear.

[spring raptor]

Don't afraid to ask more questions if it's still unclear.

[spring raptor]

Why?
You are a server (or singleplayer) and emit T - you get FromClient<T> locally.
You are a client and emit T - server receives FromClient<T>.

Not at all, I really need to put a table or something like this to make it more clear 🤔

If you have any ideas how to improve the docs - let me know.

I think it's there, but I'll double check.

Good idea. If you have a vision of it, a PR would be appreciated.
But I planning to take a look into it myself before the next release.

[spring raptor]

@dire aurora do you still calculate statistic on your own? 🤔

I.e. stuff like packet loss, rtt

[dire aurora]

Yes, but only rtt, and only because backends can't report accurate latency ... And there is that feedback loop that adjusts the timing if inputs ever get dropped

[spring raptor]

Do you think it's useful to provide statistic in backend-independent way?

[dire aurora]

Would be helpful for anyone that uses statistics from a backend if they ever want to switch ... Could also maybe simplify clock sync in cases where a more accurate out-of-band measurement isn't practical (yet) 🤔

[spring raptor]

Why rtt provided ba backends aren't accurate? I.e. why do you calculate it manually?

[dire aurora]

The backends run as part of the bevy update loop, which adds latency on each side based on how the packets align with the frames/updates

In theory backends could have an async part to them to avoid that ofc, but not many of them do

[spring raptor]

Maybe we should contact backend developers 🤔

But why? 🤔

[spring raptor]

Yes, I can't provide world access there, unfortunately.

[spring raptor]

The example runs on CI

Make sure you have all necessary imports

use serde::de::DeserializeSeed;, for example

[brittle mulch]

@spring raptor Mr shatur are u experienced with ai development?

[spring raptor]

Unfortunately, no

Ah, I think it's only for events 🤔
Do you have a component with reflection inside?

I can provide it like I did for events. Could you open an issue for it? I'll fix it soon, just need to finish some other work.

I can't provide all resources or world access nicely. But having TypeRegistry is reasonable.

[spring raptor]

Just opened a big PR with API rework:
https://github.com/projectharmonia/bevy_replicon/pull/423
You don't have to look at the code, I provided a high-level explanation about the changes in the description.

[echo lion]

What problem does this solve?

[spring raptor]

I dislike our current ergonomics.
Data related to connected clients is located in separate resources, and accessing it is inconvenient. For example, to configure visibility, I have to get ReplicatedClient from ReplicatedClients using client ID (O(n)), and then call visibility_mut. Fetching it from a query is more direct and O(1).

Since we now have identifiers, I removed ClientId because conversions between them were quite confusing for users.
This change leads to using OnAdd / OnRemove triggers to listen for connect/disconnect events. As a result, I removed the special triggers along with the disconnect reason (which we also can't represent nicely).

Not entirety sure about the API, but I think that the direction is promising. It feels more Bevy-way. I'm open to suggestions.

[raw idol]

I like it but I'm clearly biased because I already use entities for everything

it's the same conclusion I eventually reached, that as much as possible should just be in the ECS since it makes access easier and more ergonomic (unless there's a very good reason to store it in my own data structure in a resource)

[swift hamlet]

Does any of the messaging backends for replicon implement nat traversal?

[spring raptor]

Not aware of any backend that has this 😢

[dire aurora]

In what context are we doing NAT traversal here?

[swift hamlet]

If you do hosting mode, you can run your game at a much lower cost if you don't need to pay a full relay.

[dire aurora]

So you want a player host to be reachable by other players while everyone is behind NAT?

[swift hamlet]

So when the connection gets established, there would be a similar implementation of what Matchbox is doing, most clients would be able to direct message each other instead of going through a relay. Which is both lower cost and faster.

[swift hamlet]

correct.

[dire aurora]

Yea, you'd probably need a custom transport in a backend for that ... I do something similar for my game (except I only have clients behind NAT but the server is a dedicated one without NAT issues), I handle an initial handshake trough a web API and that allows me to immediately connect afterwards

[flat holly]

Is there a preferred way with replicon how to despawn entities with an effect so that they are not removed instantly? (e. g. showing different animations if an entity dies, teleports away or going out of range etc.)

just an animation should be fine, can you explain the difference between fully use state and dead state? my first simple search would be something like, fully use state -> client can handle the identification itself, dead state, the server sends every state and client have to follow?

Ah I see, thats somewhat not yet defined since I would adapt it based on libs like replicon which works best

would you create a new entity on its own with the state on the server and sync or send the dead animation via channels

but I think thats actually it with the new entity then there are no conflicts of the order of changes

ok thanks I will give that a try

[dire aurora]

Despawning + spawning a new entity should work fine if I understand the way replicon handles despawns and spawns correctly

[spring raptor]

Games usually do like this ^

Replicating death component or some component removal is inefficient.
It's better to just despawn the entity on server, client replicates it and on despawn spawns a special decoy entity.

[dire aurora]

Yea the exception would be if the thing that died can come back to life, in which case you need to retain everything, and have some way to prevent them from moving

(That's the reason why death is a status effect in my game, tho I currently still just despawn enemies immediately when their HP hits 0)

[thorn forum]

hello :) on server commands.server_trigger(ToClients { mode: SendMode::Broadcast, event: () }); doesn't trigger the event locally if the feature client is not active, only the clients are triggered

[swift hamlet]

And then the server isn’t needed anymore? (Meaning the server only handles handshake?)

[dire aurora]

Yea, once the handshake is done that web API I used to do the handshake is no longer relevant

[swift hamlet]

an it wokrs with the exiting transports? which one are you using?

[dire aurora]

It's a custom transport, it's pretty simple to make one when you just need to establish a connection based on some known details and use encryption with a key that you also shared trough that handshake

[spring raptor]

@echo lion Quick remainder about the review 🙂
After that I'll open a few PRs with fixes and cut a new release.

[echo lion]

Ok I’ll try to get to it this weekend

[thorn forum]

up :) i think its not the expected behavior ? thanks

[dire aurora]

The good news: The rollback example seems to function decently without nearly as many hacks as my own game had
The bad news: I suck at this example game

[brittle mulch]

For some reason that made me notious very trippy

lol

[dire aurora]

You mean nauseous? Understandable with my poor driving skills + the awful controls + the complete lack of any references to comprehand what is happening 😂

[brittle mulch]

yes nauseous

But it is nice too se a working physical sample in replicon

Congrats on your achievement

[dire aurora]

Now we can brag about how the worst rocket league clone in the history of rocket league is written with bevy_replicon

[spring raptor]

Sorry, I forgot to reply. I think I fixed it in the latest master. Could you try it?

[spring raptor]

Awesome! Maybe apply a grid texture to the ground? And maybe adjust the camera a little 🤔 Just interpolate it and increase arm distance when the box speedups

Maybe one of these textures: https://www.kenney.nl/assets/prototype-textures

[thorn forum]

im already using this commit https://github.com/projectharmonia/bevy_replicon/commit/545d0bbe097bcfba26d8af5209cea0b7ee3290b7 if thats the one you're talking about

[spring raptor]

Ah, I think it's expected. When the server is not a client, it's not a listen server. All local re-triggering happens only if client feature is enabled.

[thorn forum]

alright then, but maybe we can improve the documentation to say the server_trigger are not received if the client feature is not activated

[spring raptor]

Will adjust docs soon, need to merge one big PR first to avoid conflicts :)

[swift hamlet]

Is that available somewhere?

[dire aurora]

Not yet, but it's in a state where I could probably publish it soon, just need to make the two smaller crates compile again

And then solve the biggest problem: I don't have a name, so it's currently called some_networking_thing_idk 😂

The repo has some stuff for entity disabling (might be able to remove that if I account for changes on main), entity management (mostly reusing existing entities if the spawn conditions are identical), input queue stuff, and of course rollback

[brittle mulch]

Mind describing it I like to think I can help

[dire aurora]

It's basically a set of tools for networking more complex things on top of replicon, so the naming will probably be x_rollback, x_input_queue, etc ... See my previous message for the list of what I currently have in there ... Still need to add some clock sync stuff tho

[swift hamlet]

RepliconRewind? Or are you look for something more abstract?

RepliconImpulse?

[spring raptor]

Naming is so hard for me as well. I still don't have a good name for my life simulation game...

[dire aurora]

Isn't it called "Project Harmonia" for now? Better than my game which still goes by "Unnamed ORPG"

I do have a better name for it, I just haven't put in the tiny amount of effort to do the rename 😂

[spring raptor]

It's a placeholder name; I just didn't want to call it "Unnamed Life Simulator" 😅

[brittle mulch]

How about Mr Shatur, game of life

[spring raptor]

You mean like Sid Meier's Civilization, but Mr Shatur's game of life? 😅

Game of life is also quite recognizable, it's an old school classic.
And a game named just "Harmonia" also exists 😢

[brittle mulch]

exactly

[spring raptor]

I wonder if I violate EA's IP by calling my game "Shatur's The Sims" 🤔

[brittle mulch]

Knowing their policies

Probably yes

[dire aurora]

"Shatur's Harmonia" 🧠

[swift hamlet]

The sims had other names before (during development):

• Dollhouse
• Home Tactics
• People Simulator

Bur probably none of them have that certain feeling to it.

[spring raptor]

Interesting, I didn't know that!
Yeah, The Sims is very catchy 🤔 The fact that they call characters "sims" and money "simoleons" adds to its charm.

There's another game in this genre that's currently in development: Paralives. They call their characters "parafolks" 🙂

[thorn forum]

code name ShH 🤫

[spring raptor]

Do you have any up-to-date footage? I thinking about mentioning your project in my upcoming Bevy talk

[dire aurora]

Haven't touched it since then

[spring raptor]

It's okay, I'll do a text mention 🙂

The video is alright, just needs grid textures. With solid colors it's hard to see the movement.

[dire aurora]

I could push the repo if you want to fix it, but I probably won't have much time to fix it for a while cause I've been focusing more on getting the CNC stuff done

[spring raptor]

The talk will be in 2 hours, I don't think I'll be able to make it in time.
But it's okay even without a video.

Take your time and enjoy your hobby 🙂

[dire aurora]

Ah yea, 2 hours is definitely too short to fix this example 😂

It's in a pretty rough shape

[dire aurora]

I guess with how long your presentation took it's good we didn't have fancy demos 😂

[spring raptor]

True 😅
I should have cut half of the slides and include "part 1" in the title.

It also was my first time going public like this. In the beginning of the presentation I was very nervous 🙂
I'm an introvert person, but like to try new things.

So it was a stressful but interesting experience for me 😅

[dire aurora]

I think you did pretty well overall ... And unlike Jondolf you actually had a functioning microphone 😅

[brittle mulch]

Wait

I lost the presentation

[dire aurora]

It was streamed on youtube so you can just watch it back

I have to say the most disappointing part was when the 0.16 features were brought up and no one mentioned entity disabling

[spring raptor]

Damn, I totally forgot about it 😢
Yeah, it's huge for rollback!

I definitely needed you there as a 3rd talker, I would be able to cut 4 slides 🙂

[spring raptor]

Here is the link to the recording: https://www.youtube.com/watch?v=fNuolgA_88k

[dire aurora]

But then you might've had less time to talk and those 4 slides took the least time 😂

[spring raptor]

True, I really just rushed through them 😅

[dire aurora]

I do agree that relations are big tho, for pretty much the reason you described ... If we find a way to network these nicely in replicon then I can throw away all the most cursed parts of my game in one go :')

[waxen barn]

Yes me too, I'll have to watch the full thing later but you're not coming off as nervous just skimming thru the video

great job

And honestly I find networking really hard to vocalize

[spring raptor]

Thanks!
You probably know all of this stuff, it's more for beginners 🙂

[dire aurora]

Just wait until we get to part 3 and even Joy learns new stuff 😂

[waxen barn]

hahaa

quantum entangled messaging system, true zero lag

[spring raptor]

Ah, yeah, I think the way we replicate in ECS might be new 🙂

I also planned to make a small call for action in the end: encourage people to build their crates. It would be really nice to see a shooter-style rollback crate.
And things like the mentioned asset streameing could also be a neat crate that utilizes replicon.

But I forgot about that 😅

[swift hamlet]

Still hoping @dire aurora will release their crate this week 🤗. At least that’s already one 😉

[swift hamlet]

Is there a way in Replicon to add artificial lag, jitter and package loss?

[spring raptor]

There are tools for that.
On Linux check out Comcast, on Windows - clumsy.

[swift hamlet]

As far as I understood those work on the network interface and won’t work if you are testing with localhost? Or do they also allow you to test network conditions locally?

[spring raptor]

Should work on localhost as well. Clumsy even showcases it in their gif: https://jagt.github.io/clumsy/index.html

[dire aurora]

Yea I usually do this on localhost with some tc commands ... I have observed it to cause some undesirable side effects however, like adding latency to communication to the session API from both the client and the server ... As long as you're aware of what side effects it could cause it shouldn't be a major problem tho

[dire aurora]

I could try to fix all the naming, push it to github and claim some names on docs.rs at least ... Probably makes more sense to wait for 0.16 before I do a real release tho (cause rollback without entity disabling really sucks)

[swift hamlet]

For a crate release makes a lot of sense.

On GitHub for just testing and trying it out would be really nice before that.

Just curious to see how you implemented it and how it works

[spring raptor]

Did you come up with a crate name? 🙂

[dire aurora]

Also wish 0.16 included Resources as Entities ... The part where I do rollback on resources is easily the worst

I think I'll steal that "rewind" name, tho I still have to figure out how I'm going to make the subcrates sensibly structured 🤔

[spring raptor]

Same, I really hoped for that.
Maybe use singletone entities in the meanwhile? Or do you have some third-party you want to rollback?

[dire aurora]

The only resource I use rollback on is Collisions from avian

Outside of that resources just aren't part of my game's simulation

[spring raptor]

Ah, from Avian, right...

[spring raptor]

bevy_rewind? Sounds nice.

But if we get a shooter-style rollback, it might be a bit confusing 🤔

Maybe better go with bevy_replicon_rewind to clarify that it's for networking

[dire aurora]

Yea I'm planning to do bevy_replicon_rewind, tho I'd probably also claim bevy_rewind in case I ever make it generic across replication layers, or something like replicon gets upstreamed but rollback doesn't

Just gonna need to figure out how to name all the subcrates, there's the rollback, input queue, entity management, and disabling (still have to experiment on bevy main if this is still necessary tho)

[spring raptor]

Yeah, naming them as bevy_replicon_rewind_something would be weird

I remember you mentioned that they are fairly standalone? Maybe just call them bevy_replicon_something?

[dire aurora]

The input queue is standalone at least, would have to check for the other two

They aren't entirely separate tho, like you could use the input queue outside of rollback, but making your own input queue that plays well with this rollback would be annoying

[spring raptor]

Understandable.
I wish we could have scopes in crates.io

I would just put them under bevy_replicon_something even if some of them rely on rewind

[dire aurora]

Hmmm, entity management is coupled to rollback (which makes sense I guess, otherwise it ends up doing extra work for entities that aren't even predicted)
The other one I'm not super sure I can remove yet ... Upstream entity disabling has no counter in Disabled, nor does it trigger anything when you disable/enable entities ... Those triggers aren't super necessary but OnAdd/OnRemove observers would need filters (or we'd need query observers) to easily get the same behavior 🤔

[spring raptor]

Filtered observers would be really nice. I currently just do checks with queries in my game. Or is it expensive for what you are doing?

[dire aurora]

I think it would be viable in my game, I don't have all that many predicted entities, and the majority of them are the ones I'd need the observers for

I think it mainly depends on the timeline of this feature

[spring raptor]

Ah, so the upstreamed solution is better because it doesn't trigger anything?
And current solution is a temporary replacement to make it work with 0.15?

[dire aurora]

Idk if I would necessarily say better, it lacks features because they would've been controversial, but this makes utilizing it much harder

[spring raptor]

Ah, I see 🤔

[dire aurora]

I knew there was something I still had to do for the entity management crate ...

has_authority: (), // TODO

It doesn't support cases where server vs client isn't statically compiled

[dire aurora]

Ah yes, of course I can't get this working on 0.15 ... Would be difficult to get a working example on bevy main tho, considering they don't track main 🤔

[spring raptor]

Ah, you rely on Disabled, makes sense.
I can port Replicon to main, but we will also need to port a backend and avian...

[spring raptor]

BTW, wasn't this under a feature?

[dire aurora]

The backend is the example backend so that shouldn't be too hard

[spring raptor]

Ah, right!

[dire aurora]

A separate crate, but it's the last part I need to get right before I can publish something

[spring raptor]

So do you need main? I can port replicon to main this evening, just currently at work

[dire aurora]

Would probably be ideal ... Especially if we also want to mess around with the relations networking before the release ... Would be amazing if we the replicon release for 0.16 would include something for that 🤔

Iirc the state of networking related entities is that the rework to make it possible was done, but the feature isn't there, right?

[spring raptor]

Okay!

About relations, In 0.16 we'll have one to many. I don't think we'll need anything special 🤔 If I understand it correctly, you can replicate Parent and Children will be created automatically.

[dire aurora]

Yea the only thing we'd need is the ability to group them into the same packet ... And I guess a way to specify which relations should have that behavior (only ones where desynchronized state can cause problems would want this I guess?)

[spring raptor]

Ah, yes 🤔
Right it won't cause any issues because insertions and spawns are a single reliable message and updates, even if split, won't be applied unless the minimum required update is received. So you will never receive a component that points to an invalid entity.
But it's a bit conservative. With relations we could make it much more performant.

On second thought, if there weren't any insertions, spawns, or removals, but some entities changed their parents and the changes get split, it might be weird to see only half of the changes applied.

So you are right, we might better group related entities smartly

[dire aurora]

Yea, there's quite a few cases where them updating separately results in something that is wrong in some way

Which is why I currently still use that map for status effects (even if it is wildly inefficient because it now sends all of them when the map changes, rather than just the status that changed)

[spring raptor]

Definitely will try to look into it before the release

[dire aurora]

But yea it's definitely gonna be important to have some control over this ... Not every relation will need this, and even when we need it, it might not be needed for every instance, though I'm not sure if replicating huge hierarchies of replicated entities is currently a thing people do 🤔

[spring raptor]

It's similar to what Lightear call "entity grouping".
But I deliberately waited for relations for it.

[dire aurora]

Yea without relations these usecases aren't easy anyway

The replication/rollback side of my status effects is a mess, but so is keeping the stupid map up-to-date 😂

[spring raptor]

I wonder if we could always use unreliable channel and use relations to ensure partial world updates aren't a mess 🤔
We will of course need 1:1 and many:many

[spring raptor]

With the last serialization rework it should be doable, yes.

We also now have connected clients as entities as a PR. Might be useful for relations as well.
But koe doesn't seem to like this change 🤔

[dire aurora]

koe's main concern seems to be the removal of client ids

Which I guess makes some sense ... With things like he indexing feature on main (at least if that got merged) or a map from client id to entity we could have both pretty easily however ... The ergonomics of working with entities instead of maps with client id -> client data is definitely better tho

[spring raptor]

Ah, I totally forgot about the indexing feature! Hope it gets merged.

Yes, after some thinking I also more inclined to keep ClientId and just provide a map (until we don't have the indexing feature). The map will be useful for some third-party plugins, while users can continue to use Entity for better ergonomics.

[sand radish]

I'm using bevy_replicon (with renet) and replication works fine, but when trying to get events to work, I get a panic with Called 'send_message' with invalid channel 2 (same but with receive on server side). I just use the default channel init, so I'm not sure if I'm missing something obvious here, anyone knows where to look further?

[spring raptor]

It means that the channels weren't setup properly 🤔
Double check how you setup channels, you should pass them from Replicon: https://github.com/projectharmonia/bevy_replicon_renet/blob/aa3443181aadacd33b893f554c2e40a1cea0d624/examples/simple_box.rs#L108

[sand radish]

Yes that's what I did. Client side setup looks like this: https://gist.github.com/almindor/030386a484e82ca514fcca875d05269c

[spring raptor]

Looks correct to me. Are you sure that you register events before setting up the client and server?

Try running with RUST_LOG=bevy_replicon=debug cargo run ..., it should show us what you registering

[sand radish]

you mean add_client_event should be called before the insert_resource of client/transport?

doesn't seem to change anything if I change that order

```2025-03-07T19:25:42.448582Z DEBUG bevy_replicon_renet: creating channel config ChannelConfig { channel_id: 0, max_memory_usage_bytes: 5242880, send_type: ReliableOrdered { resend_time: 0ns } }
2025-03-07T19:25:42.448615Z DEBUG bevy_replicon_renet: creating channel config ChannelConfig { channel_id: 1, max_memory_usage_bytes: 5242880, send_type: Unreliable }
2025-03-07T19:25:42.448620Z DEBUG bevy_replicon_renet: creating channel config ChannelConfig { channel_id: 0, max_memory_usage_bytes: 5242880, send_type: ReliableOrdered { resend_time: 0ns } }
2025-03-07T19:25:42.448624Z DEBUG bevy_replicon_renet: creating channel config ChannelConfig { channel_id: 1, max_memory_usage_bytes: 5242880, send_type: Unreliable }
2025-03-07T19:25:42.448774Z DEBUG bevy_replicon::core::event::client_event: registering event gizmo::entities::Controller
2025-03-07T19:25:42.448784Z DEBUG bevy_replicon::core::channels: creating a client channel with ID 2
2025-03-07T19:25:42.577590Z DEBUG bevy_replicon::core::replicon_server: changing RepliconServer running status to true
2025-03-07T19:25:43.295525Z DEBUG bevy_replicon_renet::server: forwarding event ClientConnected { client_id: 0 }
2025-03-07T19:25:43.295622Z DEBUG bevy_replicon::server: ClientId(0) connected
2025-03-07T19:25:43.295637Z DEBUG bevy_replicon::core::connected_clients: adding connected ClientId(0)
2025-03-07T19:25:43.295645Z DEBUG bevy_replicon::core::replication::replicated_clients: starting replication for ClientId(0)

thread 'Compute Task Pool (6)' panicked at /home/ales/.cargo/registry/src/index.crates.io-1949cf8c6b5b557f/renet-1.0.0/src/remote_connection.rs:349:13:
Called 'receive_message' with invalid channel 2```

that's server

[spring raptor]

From this log I can see that you register events after you create channels

[spring raptor]

Before you read channels from RepliconChannels

[sand radish]

aaah

that could use a warning section in the quickstart 😄

[spring raptor]

Yep, I definitely need to rework my quickstart guide. My todo before the next release 🙂

[spring raptor]

Done: https://github.com/projectharmonia/bevy_replicon/pull/428
It's a quick and dirty migration, but it compiles, tests pass and examples run.
I need to update docs and properly implement mappers on contexts, but I'll do it later. Don't want to get into large conflicts with other PRs. But this should unblock you 🙂

[sand radish]

what's the expectaion of mapping "this entity is what the player on a client controls" ? I have things set up such that server is standalone and creates the whole world using replication, including the "player controlled" entity. How should the flagging be typically handled for the client? e.g. how can I inform a client "this newly replicated entity? that's your thing for controlling"

[spring raptor]

2 options:

• Spawn entity on client and send an event to server that says "Hey, I control this entity", server inserts the entity into https://docs.rs/bevy_replicon/latest/bevy_replicon/server/client_entity_map/struct.ClientEntityMap.html and it just works
• Server spawns an entity and says "Hey, you control this entity".

[sand radish]

i mean I guess I could send the entity id over as an event to specific clients?

right 🙂

thanks

[sand radish]

is it safe to use an Event as both an event and as a Component ?

e.g. I have a component on an entity, that's also an event I actually send over for event handling

[spring raptor]

Yes

[sand radish]

how does ParentSync work exactly? Do I assign it to the children entities, the parents, both? Do I need to register it for replication?

[spring raptor]

Only to children 🙂
See this example: https://docs.rs/bevy_replicon/latest/bevy_replicon/parent_sync/struct.ParentSync.html

[sand radish]

and do I need to add it to app.replicate ?

ah no, works without that 😄

[spring raptor]

Won't be needed for Bevy 0.16. In 0.16 you will be able to just register Parent for replication.

Because of relations

In 0.15 this helper is still needed.

[sand radish]

how would you go about "hybrid" replication for physics simulated entities you don't want to fully replicate movement of (e.g. fast moving simulated projectiles)? I was thinking if I could somehow "init replicate" them with physical properties, but then let them simulate fully on the client that'd be best (of course "resolution" events would be server authoritative but that's out of the scope of this)

[spring raptor]

With the described approach you would rely on determinism. And it's really hard to guarantee.
I would recommend taking the Rocket League approach. In short: you simulate locally, the server sends you the state, and you correct it by rolling back the entire world to the last confirmed tick and re-simulating until your current timeline.
@dire aurora is working on a crate that implements this on top of Replicon 🙂

[sand radish]

What I'm thinking is that client side simulation would only be "path simulation" (e.g. gravity effects etc. yes, but no colission detection or impact on objects). This way server would remain fully authoritative on the "fast/many" elements and their lifetime and impact, but their "traversal" would not need to be replicated. If there are discrepancies, that's ok it should be fairly minor

so e.g. a shot on the client will "fly through" and not impact a target unless the server replicates the despawn and impact info

I'd need to implement a sort of "init only" approach to replication on these tho, e.g. "create a shot with initial transform and velocity" through replication but then stop replicating transform

[spring raptor]

It depends on the game. If weird glitches are acceptable, then it could work.

[sand radish]

it's a top down (2d plane) space shooter kind of game, using capital ships with mounted/aiming weapons

I'm hoping to come up with a replication scheme that will make it viable for up to 10 players

there will be hitscan but most things will be projectiles, physics plays a role since everything is essentially simulated and gravity is a gameplay mechanic (e.g. black hole, that sort of thing)

[spring raptor]

Then I would go with Rocket League approach

Here is a good video: https://www.youtube.com/watch?v=ueEmiDM94IE

Is starts with physics, so skip to 23:30 for the networking part

[dire aurora]

Definitely sounds like a good usecase for the rocket league approach yes ... Things like projectiles are already kind of a pain with the shooter approach, but it gets so much worse when physics are involved

[sand radish]

I'll at least try and see how much discrepancy I get with the "init and forget" approach. My hope is if I rollback and re-apply the shot physics (e.g. move the shot physics simulation forward to "server time" on client side replication moment) it should minimize visible discrepancies. But I need to test this. Right now I'm just happy to have a jerky basic game going. Thanks for the library btw! it's great. I also love how ECS starts to click on all these things so nicely

[sand radish]

Am I correct in assuming that replication and events are not ordered in any order? e.g. if I spawn something replicating on the server and then send it's mapped entity over in an event, the client could get that event before the replication spawn on its end?

[spring raptor]

No, we guarantee that events will never reference an invalid entity

[sand radish]

Hmm, I'm seeing a very odd issue, not sure if it's bevy or replicon/bevy level. If I run my server and client (both render, but slightly in different ways) in one workspace (e.g. 2 windows side by side) things seem to work fine. If I put one on another workspace things get extremely choppy to the point of beeing frozen on both sides. And I don't mean just input event level things, even basic simulation on the server stops. Interestingly tho, debug log shows things such as client sending input events on mouse moves normally

even the initial connection sometimes takes like 10s when done like this, they both sit there, server reports connected, client reports connecting, but nothing happens for a while and when it does it's like 10s lag

I'm guessing it's a window focus issue

[sand radish]

nah, not focus, just "visible at all" issue. even unfocused, with continuous winit setting it works, but as soon as the server window is obscured completely it stops processing

works if I go headless

[spring raptor]

Merged connected clients as entities with all concerns addressed 🎉
Planning to submit a few more minor fixes and draft a new release.

[spring raptor]

Fix for max message size is up: https://github.com/projectharmonia/bevy_replicon/pull/429

[dire aurora]

Hey Shatur, what are your opinions on https://github.com/bevyengine/bevy/pull/17687?

[spring raptor]

I like the new ergonomics, but the requirement for EntityMapper::set_mapped is a bit odd...
In the 0.16-dev branch I created, I just stubbed it since I can't implement it.

[dire aurora]

I like the general approach, but with the avian porting work I'm doing I'm just a bit confused about it

There's 3 traits for this now and they aren't all compatible

If they were compatible this could really cut down on a lot of multiplayer boilerplate tho

Could even get rid of mapped vs non-mapped if the trait can always provide the mapping and that gets optimized out when unnecessary

[spring raptor]

VisitEntities, VisitEntitiesMut and Component?

[dire aurora]

Yes and MapEntities also still exists

[spring raptor]

I agree, I don't quite understand why VisitEntities, VisitEntitiesMut are still needed, Component provide these methods

[dire aurora]

You also get some fun issues like how EntityHashMap can't be marked with #[entities]

Because it doesn't have VisitEntitiesMut

[spring raptor]

Ah, that's why we need separate trait for non-components 🤔