archived-dots

[ocean tundra]

😛

again unsure how netcode and ghosts work but could you make your child abilities 'accessable'?

[fluid kiln]

wdym by that

[ocean tundra]

in my netcode that just means giving them a unique id

so when sending messages i can lookup the target entity by id

i think it netcode its the 'ghost' thing

[fluid kiln]

oh yeah, I don't think so because they aren't ghosts

Theyre ghost child

I'll try to look into what a ghost child does

Doc, is sparce

[ocean tundra]

does it not support children being ghosts?

[fluid kiln]

Not afaik might be wrong

[ocean tundra]

yea thats annoying

[fluid kiln]

Well actually

hmmm

I never thought of have other entities than the characters be ghosts

But then wouldn't that dramatically increase the number of ghosts

[ocean tundra]

i guess

not sure what the cost of ghosts in netcode is

for me 'networkable' entities are very cheap

just a ID really

and data is only ever synced when there is a change

and even then theres a ton of options to control the syncing

i cant imagine ghosts being too expensive

[fluid kiln]

Me neither but I remember reading somewhere like max 80 ghosts or smt

[ocean tundra]

max 80!!

what

no way

[fluid kiln]

Maybe i'm crazy

[north bay]

There is no hard limit that I'm aware of

[fluid kiln]

I think I jhust have to configure them correctly

Like how do I disable transform sync

Since they're abilities

[north bay]

I have used more than 80 ghosts

Latest netcode version should have overrides for that

[fluid kiln]

Any idea how I can handle my ability referencing Script?

[north bay]

Otherwise you can make a conversion component that removes TRS

What's the problem?

[fluid kiln]

A character has x abilities, that are children to it

How do I refernce these abilities from the server/client

I thought of an ID component

But now it kindof makes sense to make them ghosts

[north bay]

Hmm

[fluid kiln]

For some reason I had the preconception that ghost == player/npc

[north bay]

I wouldn't make each ability it's own ghost

[fluid kiln]

I have to, because they have important components like cooldown etc

[north bay]

How many of those will you have and how are they linked with the player?

They can still be their own entity

Ghost != entity

My bad

haha

[dense crypt]

I recently started a new project and for some reason my system is never created when I go into play mode. Any ideas??? It inherits from SystemBase and contains only a Debug.Log in OnCreate();

[fluid kiln]

A character might have anywhere between 1 and 30 abilities

So far they,re only linked through children

Hierarchy

[north bay]

You could store a lookup table on your player that maps ability type to idx in your children buffer

[fluid kiln]

Check the entity debugger if it's there

Oh okay I could do that

[north bay]

Buuuut It's rather hard to judge without knowing all your requirements

[ocean tundra]

So another fun idea 😛
A component with 30 filelds of Entities? each of them the ability entities

[dense crypt]

That's the thing, it's not 😕 Also checked in Inactive Systems

[fluid kiln]

Naah I don't like that, I want it to be dynamic ^^

[ocean tundra]

tho i have no idea how netcode handles syncing of entities references

[fluid kiln]

WHy not make them ghosts?

Since the only thing they might sync is the remaining cooldown

Do you know if children entities can use [Ghostfield] ?

[north bay]

Yea they do

[fluid kiln]

Okay so making them ghosts would be redundant

[north bay]

My setup looks like this: Character (ghost) holds a reference to AbilityCollection (ghost) which holds the abilities which aren't ghosts

So I can switch the abilities during runtime easily

[ocean tundra]

Can you explain GhostField? im interested, that just marks the field to be 'sycned'?

[fluid kiln]

Yeah basically

[north bay]

Ye

[ocean tundra]

nice

[fluid kiln]

How do you reference your abilities across the server?

[north bay]

My AbilityCollection holds a DynamicBuffer of abilities

I do not have a lot of them so I just iterate over the buffer

To find the matching one

[fluid kiln]

Okay got you

I might try to ghost them since being able to reference by entitiy would make my life so much easier

And if it's too costly, I'll find another solution

Ugh but now I have to instantiate them differently

[north bay]

Your abilities?

[fluid kiln]

Yeah

[north bay]

If they are childs of a ghost the reference is automatically resolved when instantiating

Just like the normal instantiate

[fluid kiln]

Even if they are ghosts themselves?

Worth a shot, I'll try it out

[north bay]

I haven't had a need for ghost relationships like that so far

[fluid kiln]

It's touchy

[north bay]

I like having them under one ghost so I can be sure that the reference to the children is always valid

[fluid kiln]

Yeah me too I don't want to shake that too much

[north bay]

Messing with my Character<->AbilityCollection referencing was stressful enough lol

[fluid kiln]

I'm curious how you handle casting these abilities if you'd like to enlighten me further

[north bay]

What do you mean with casting?

[fluid kiln]

For example, I like to add components so I systems can pick it up. So far my ability casting pipeline is: Player sends RPC to cast ability - Server checks if cast valid - Server Casts Ability - Server sends RPC to client to tell it the cast was succesful

But during the cast, I need to add the component OnCooldown to the ability on both the client and server

I'd ghostfield the remainingDuration

[north bay]

Little optimization, ghostfield the tick instead of the duration

[fluid kiln]

Yeah you're right that's the right way

[north bay]

That way it doesn't change every tick

[fluid kiln]

Noted thank you

[north bay]

Hmm I do not do adding/removing of components on ghosts anymore

Since that's to messy

[fluid kiln]

It's really messy and no guarantee of synchornization

But how do you handle that in systems

Yo umust have a lot of systems with if() return

[north bay]

Yes

Basically every prediction system I have

I do not have abilities in that sense btw

I call abilities/actions my character stuff like moving/shooting/sprinting

Etc

Which I inherited from the FPS sample

[fluid kiln]

Yeah no worries I can apply these concepts to my thingy

[north bay]

I don't think those rpcs work very well with prediction

[fluid kiln]

This was actually pretty enlightening, I might go the way of not adding / removing coponents too

[north bay]

You most likely want to include that into your input struct

[fluid kiln]

I tried to have "button" type fields in my inpuit

But sometimes it felt like the buttons were lost

[north bay]

They are if you sample them the wrong way, let me check if I can find the forum thread

[fluid kiln]

🤔

[north bay]

Hmm can't find it

You basically have to use "button |= Input.GetKey" instead of GetKeyDown and reset the state when you start sampling for a new tick

[fluid kiln]

Yeah I think I see where my mistake was

I'd sample more often than send the data

[north bay]

Ye the input is sampled every tick but only send once the tick has been fully predicted on the client

[fluid kiln]

Time to do some good old refactoring

So maybe I could have an int field with abilityCast or something like that

Or if I find a way to sync the ability buffers

AbilityIndex would be pretty cool to use instead

Any way to guarantee dynamic buffer order?

[north bay]

Indeed, or if you can hard code the ability type the enum to send

DynamicBuffer order should be guaranteed

[fluid kiln]

Even if I add to it during runtime

[north bay]

Well then it won't be synced and it depends on your add logic

Are you talking about adding the buffer during runtime or adding to the buffer?

[fluid kiln]

Like a character learn a new ability during ganeplay

[north bay]

Hmm

I guess you would actually need ghosts abilities for that

Since as far as I know you cannot add new components to sync to an already existing ghost

[ocean tundra]

Could you just use a rpc to sync that?

wait you cant add components to a existing ghost and get them to sync?

[north bay]

Not that I am aware, no

[ocean tundra]

hmmm actually thats exactly how my netcode currently works too 😛

[north bay]

In 0.5 idk how it is in other version netcode create a static ghost archetype (not an actual entities archetype)

That you can't modify

[fluid kiln]

Well you can add to them

[ocean tundra]

interesting

[fluid kiln]

It's just not synced

So if you send roc

[north bay]

Yea

[fluid kiln]

You need to add on client, send rpc, add on server

(it's not fun)

[ocean tundra]

yea, same for my stuff 😛 its not hard to add a new component to sync, just i havnt made it automatic yet.
I imagine its the same for Unity, they just havnt added that feature yet

[north bay]

When is your stuff going public btw?

[ocean tundra]

probably still awhile, Im almost done refactoring into a package

[fluid kiln]

:o

[ocean tundra]

but i want to build a ton more samples and docs and stuff

and discussions like this make me relise theres still a few holes

and im debating getting the modding package out first instead of netcode

[north bay]

It doesn't have to be feature complete for 0.1 haha

[ocean tundra]

even tho my netcode has more stuff, modding is simpler

haha yea, thats what i tell myself

[fluid kiln]

Conpletionism syndrome

[ocean tundra]

yea

and finally i want to figure out automated tests

this package refator broke a bit, so would be real nice to have tests to confirm bits are always working as expected

[north bay]

As a github integration?

[ocean tundra]

maybe

i actally use azure devops mainly, so would probably use their piplines to run things

but first just getting tests i can run manually

[north bay]

You gotta get that new test coverage package

[minor sapphire]

anyone know if there is any downside to using ThreadIndex for the EntityCommandBuffer.Parallel index in cases where there is no index provided (e.g. entityInQueryIndex)?

[north bay]

It's addicting seeing the coverage going up everytime you add a new unit test

[ocean tundra]

this one?
https://docs.unity3d.com/Packages/com.unity.testtools.codecoverage@0.2/manual/index.html

[north bay]

Ye there's a 1.0 version https://docs.unity3d.com/Packages/com.unity.testtools.codecoverage@1.0/manual/index.html

[ocean tundra]

😛 google must have sent me to the old one

[north bay]

Idk why unity doesn't fix their google integration regarding that

Always hit the old versions

[fluid kiln]

yeah lmao

Concerning that input sampling, how do I know when a new tick starts?

[ocean tundra]

im still not even sure whats the right way to write tests.
Those unit tests would be ok testing parts (units)
But with something like netcode i want to test it as a whole somehow

I always like hearing what stellaris does, they run a massive AI vs AI networked match each night 😛

[north bay]

That's a lot of custom stuff but that's how my input storing looks like ``` private ClientSimulationSystemGroup m_clientSimulationSystemGroup;
private EntityQuery m_commandToSampleIntoQuery;

    protected override void OnCreate( )
    {      
        m_commandToSampleIntoQuery = GetEntityQuery( typeof( CommandToSampleInto<T> ) );
        m_clientSimulationSystemGroup = World.GetExistingSystem<ClientSimulationSystemGroup>( );

        RequireForUpdate( GetEntityQuery( typeof( T ), typeof(DoPredict ) ) );
    }

    protected override void OnUpdate( )
    {
        var commandTarget = GetEntityQuery( typeof( T ), typeof(DoPredict) ).GetSingletonEntity( );
        var commandToSampleInto = m_commandToSampleIntoQuery.GetSingleton<CommandToSampleInto<T>>( );
        
        var targetTick = m_clientSimulationSystemGroup.ServerTick;

        if (commandToSampleInto.Value.Tick != targetTick)
        {
            var commands = EntityManager.GetBuffer<T>( commandTarget );
            commands.GetDataAtTick( targetTick - 1, out var lastFrameCommand );

            commandToSampleInto.Value.Reset( lastFrameCommand, targetTick );
        }
        
        m_commandToSampleIntoQuery.SetSingleton( commandToSampleInto );
    }```

space
You probably want to check out how the guys from unity's netcode did that

They also run client/server worlds in their unit tests

[fluid kiln]

tf is CommandToSampleInto

Okay I will

[north bay]

Custom component

[fluid kiln]

Thanks 😄

[north bay]

lol

[fluid kiln]

lmao for comparison this is my input system

    {
        PlayerInput input = frameInput;
        input.Tick = clientSimulationSystemGroup.ServerTick;

        Entities.WithAll<LocalPlayer>().ForEach((DynamicBuffer<PlayerInput> buffer) =>
        {
            buffer.AddCommandData(input);
        }).Schedule();
    }```

[north bay]

That code is just for preparing the command/resetting the data. Storing is also reasonably simple on my side: ``` var commandTarget = GetEntityQuery( typeof( T ), typeof(DoPredict) ).GetSingletonEntity( );

        var commands = EntityManager.GetBuffer<T>( commandTarget );
        commands.AddCommandData( m_commandToSampleIntoQuery.GetSingleton<CommandToSampleInto<T>>( ).Value );```

[ocean tundra]

What type of game are you guys building?

[fluid kiln]

Every snippet you've shared me Script have so many generics

I don't know what the hell is going on in there

hahaha

Trying my hand at a top-down rpg just for fun

[north bay]

Top-down shooter

[ocean tundra]

nice

[north bay]

INdeed, the reason for that is that it's part of my framework lol

[ocean tundra]

both will be awesome in DOTS

[fluid kiln]

Yeah dude I see so many features with DOTS that would make me cringe in OOP

It's a whole new world quite literally

[ocean tundra]

yea sooo good

[north bay]

Indeed, but it's still a very long way

[ocean tundra]

i really want to get working on my game but messing with tech/tools
Hopeing for a Voxel Space ship rts

[fluid kiln]

me waiting for navmesh, terrain, input, ui

[north bay]

haha

There's a reason while all my rendering is gameobject based

lol

[fluid kiln]

Navmesh is pretty fucking critical for me to do antyhing but I cba figuring it out until the official package is released

[ocean tundra]

@fluid kiln Have you seen Dots-Nav for navigation?

ive also converted some of A* PathfindingProject to be dots compatabile

[fluid kiln]

I don't like to go too low-level, I heard you had to still calculate the path yourself

I'd rather wait for an api that lets me GoToDestination or something like that lol

[ocean tundra]

i wouldnt say caculate... more like follow

[fluid kiln]

Care pointing me to that Dots-nav then? When I check the package doc I can't find much on how to use it

[ocean tundra]

yea i never use those APIs anyway, always wanting a bit more control about how the 'thing' moves

https://github.com/dotsnav/dotsnav

[fluid kiln]

Ooooh

I was looking at unity packages

I might try this out

[ocean tundra]

the community is starting to build up cool stuff

i dont think it will move your entities

but a follow path component is pretty easy

[fluid kiln]

Yeah no that sounds reasonable

That's actually.... pog

[ocean tundra]

pog?

[fluid kiln]

😅 like very great

[stone osprey]

How do i get an entity by its (Number:Number) name ?

For debugging purposes ?

[ocean tundra]

i dont think you can

the entity name only exsits in Editor

and only in some cases

[stone osprey]

Hmmm... what about its index ?

Is that possible ?

[ocean tundra]

ummm not really

just store the entity referece

and use a entity query to find that

whats your usecase?

[stone osprey]

WEll... its complicated. ANother question, Is a unsafe-hashmap shown in the entity inspector ?

Its items

If not... i have a damn bug somewhere

[ocean tundra]

no i dont think so

a dynamic buffer will be

but i expect that many of the native/unsafe collections dont have inspectors yet

[stone osprey]

@ocean tundra Thanks, tested it you are right ^^

Damn... i worried that it doesnt set the component right because it was just... empty

But once i loop over it, its filled so its fine

[ocean tundra]

yea inspectors is still missing some stuff

kinda annoying

i dont know if theres a way to build our own yet

I would love to get some of my bit packed components understandable UIs

so if you figure it out let me know 😛

[odd ridge]

can component buffers store pointers to regular Mono objects?

or will they only accept pure data structs?

[ocean tundra]

can a IBufferElement be a class?

[safe lintel]

my assumption would be no as you could just stuff a list in a class component

[ocean tundra]

i dont think you can have a managed buffer

yea thats probably the easyist way

[odd ridge]

can you elaborate on class component? there's nothing here xd

https://docs.unity3d.com/Packages/com.unity.dots.runtime@0.32/api/UnityEngine.Component.html

[safe lintel]

im getting antsy for the next dots release. I need more features 😩

[ocean tundra]

Wrong docs

https://docs.unity3d.com/Packages/com.unity.entities@0.17/manual/component_data.html#managed-icomponentdata

[odd ridge]

thanks

[ocean tundra]

haha yea i want more features too

anything confirmed on whats in 0.18?

working ISystemBase is the most of heard

[safe lintel]

just pure guesses, even forum posting by devs seems to have slowed but that could be my perception

[ocean tundra]

yea i think that too

or maybe they just arnt confirming stuff

[odd ridge]

can I inherit from my own IComponentData structs?

[ocean tundra]

yea

what have you been doing 😛

thats like the main thing todo with ECS

oh 😛 reread that

no i dont think you can have inheritance of structs

https://stackoverflow.com/questions/2310103/why-c-sharp-structs-cannot-be-inherited

[odd ridge]

ooooh

[untold night]

C# structs don't support inheritance

You can include a struct as part of another struct, via composition.

If you have struct type A that ONLY contains a single instance of struct type B, you can use NativeArray's Reinterpret function to view an array of A as an array of B.

You can also have interfaces that inherit from IComponentData so you can have structs with specific Methods or Properties.

ISharedComponentData, and the system state variations all do this, so additional constraints can be put on them in addition to being valid IComponentData.

[odd ridge]

thanks

[fluid kiln]

Do you where I can find the information that tells me when I begin sampling for a new tick?

Here you reset when the target tick is different than the last input's tick?

if (commandToSampleInto.Value.Tick != targetTick)

Oooh I think I get it. Something like this:

    {
        var targetTick = clientSimulationSystemGroup.ServerTick;

        //Only sample if tick has changed
        if(input.Tick != targetTick)
        {
            var inputBuffer = GetBuffer<PlayerInput>(GetSingletonEntity<LocalPlayer>());
            inputBuffer.AddCommandData(input);

            input = default;
            input.Tick = targetTick;
        }
    }```

That's brilliant

[ocean tundra]

where does input come from

and how is it set?

[fluid kiln]

My system is mapped with input events, which populate the field input

It works :p

[ocean tundra]

input events?

[fluid kiln]

Yeah I use the new input system

[ocean tundra]

the new input system?

ooo

cool

[fluid kiln]

It's just awkward cause I have no way to distinguish between UI and world when I use the mouse :/

There must be a way I don,t know about

[ocean tundra]

yea that was a issue AGES ago when i used it

had to have a game object reference to my UI MB

and that was using the new UI components which had some sort of ismouseover

but the Y was flipped from what input gave 😛 so i had to flip it first

was all really annoying

[fluid kiln]

Yeah I can imagine ^^

I like how it tho, just gotta figure that little thing out :p

[ocean tundra]

yea it was definitly better

does it give you a nice input struct you can use now? or do you map your own?

[fluid kiln]

I have to map it cause it's not really DOTS compatible yet

[ocean tundra]

yea figured 😛

[fluid kiln]

It's kinda hard to design the input for a top down game tho

Mine looks like this so far

{
    public uint Tick { get; set; }
    public float3 mousePosition;
    public Entity hoveredEntity;
    public bool selected;
    public bool interacted;
}```

I convert the mouse position to world position on the client

The say the player right clicks to move, I send the mouse position and "interacted"

[ocean tundra]

yea i would convert it too for my older rts prototype

[fluid kiln]

But IF hoveredEntity is not null, then attack that entity instead

I find it a bit unstable? idk

[ocean tundra]

but i never synced that with server
instead, i broke it up into commands (rpcs)

[fluid kiln]

The recommended way with netcode is to send input

But I remember we talked about this before :p

[ocean tundra]

so when i clicked on something, i would take the world pos of mouse, find the entity selected, and send a command with that entity that i selected it

[fluid kiln]

Yeah I gotchu

[ocean tundra]

yea they want the raw input cause they can simulate users or something right?

[fluid kiln]

It makes it 99% server safe (?) (only breach could be corrupted input, so you gotta clean it), and also allows for ghost prediction (I'm not sure if you can do that through commands, im not too knowledgeable)

[ocean tundra]

yea

my way is all commands, client takes input, breaks it down into current command (selected actually isnt even synced) eg a move units command

and server side i have to have checks like Does unit exsit, is it my unit, is target pos valid ect

[dull copper]

oh snap... I keep forgetting 0.17 got delayed

they didn't bother updating the change log date though

https://docs.unity3d.com/Packages/com.unity.entities@0.17/changelog/CHANGELOG.html: md [0.17.0] - 2020-11-13

[visual juniper]

hi all, has anyone come across a solution for voice communication with ecs ?

[naive parrot]

can Animation Jobs be burst compiled ?

[north bay]

Since DSPGraph seems to be in a very early state I doubt it.
I had a fairly easy time integrating Dissonance

[visual juniper]

oh good to know 🙂 did you use netcode though ?

[north bay]

Using steam P2P when possible, if that fails Dissonance automatically relays the packages through the server

[hollow jolt]

tried replicating an example script but in my case it doesn't like the exact line i copied to my script

oh nvm , i have the imports all mixed up

[torn hollow]

Hi, I'm new to unity and want to make a game with lots of "boids" (objects that follow the same rules). I started optimazing it with the GPU isntancing approach (special spawner for my gameobjs). After some research I found that sth with DOTS might be even better. Is it possible to marry the GPU instancing approach and DOTS or would GPU instancing be redundant if I use DOTS?

[karmic basin]

You could use Jobs for the logic and keep GPU instancing for the meshes. Not sure if DOTS is faster right now

(especially on the rendering side)
There is an official example where they implemented Boids, maybe you could compare with your solution

If you want to release soon, I'd say keep gameobjects + GPU inst (+ jobify some parts if needed)

If you bet on the future, DOTS will be awesome in a few years

[torn hollow]

Thank you for the fast answer! Will definitely look that example up

[karmic basin]

https://github.com/Unity-Technologies/EntityComponentSystemSamples/tree/master/ECSSamples/Assets/Advanced/Boids

All the repo is a must to explore

Be aware though, that as DOTS is not ready, they sometimes cheat. For example in the Boids demo, the fish are not animated, it's a vertex displacement shader. Tricks like that

Consider only parts like ECS for the data layout, Jobs for multithreading, and such...

[torn hollow]

Okay i will google my way through it ^^' thank you so much for your help so far, i definitely have a better place to start now!

[karmic basin]

Oh if you're starting, be sure to check pinned messages inthis channel

could save you more time than google searches 🙂

and check the DOTS dedicated forum from time to time 👍

[torn hollow]

I will thank you:D

[stable sentinel]

@karmic basin didn't they say dots should be "ready" by 2022?

is CreateEntity(ComponentType[]) or CreateEntity(EntityArchetype) better to use?

[slim nebula]

the former is going to determine the archetype from your component types .the latter has the archetype already so the latter should be faster

@stable sentinel

[sage mulch]

Currently working on procedural world generation in DOTS as a performance comparison to my old implementation. Have chunks generating but the chunks have no way to communicate with each other, so I can't determine where I don't need to draw a mesh on the borders. The idea was to treat chunks as subscenes.
This leads to a lot of unnecessary drawing. Has anybody got a working implementation of this? Or any suggestions on how to solve?

From some reading I've done this seems to be a fairly common issue (looking at this repo currently but it has the same problem https://github.com/sarkahn/dots-blockworld/)

[hollow sorrel]

@sage mulch what do you mean by that? if you have the chunk index, don't you always know what chunks are bordering?

[karmic basin]

Sark actually follows this channel, you'll be able to talk with him about that :)

[sage mulch]

@hollow sorrel Thank you for the reply. Should probably make it clear that I'm somewhat new to DOTS so I may explain something poorly.
In my original implementation (non-DOTS) I store the chunks (and their data for what type of block is at each grid coordinate) in an array that can be accessed by other chunks. When I'm drawing the mesh, I am able to query neighboring blocks, and if they aren't part of the chunk we're currently working with then I can refer to this chunk array, find the desired chunk, get into the grid and determine if its transparent or not. that's the gist of it at least.
With my DOTS implementation, its basically the same but I don't have access to that chunk array because I treat each chunk as an entity, and I can't think of a way to structure the components/corresponding systems so that it knows which neighboring chunks to query for block data, without losing the advantages of DOTS in the first place. So, while I may have the chunk index, I'm not exactly sure how I can use that to retrieve the block data from a neighboring chunk entity.

[sage mulch]

Oh excellent, thanks so much. Is he just @ sark? (Don't want to ping the wrong person by accident lol

[karmic basin]

Yeah it's him. Let's not ping him. He stops by quite often anyway

[zenith wyvern]

I got pinged somehow. And I am PISSED

Just kidding

[zenith wyvern]

What I did on my last run was I created a "World" structure that I would pass into jobs https://github.com/sarkahn/dots-blockworld/blob/version2/Assets/BlockGame/Runtime/VoxelWorld/VoxelWorld.cs

It's a bit convoluted but the idea is it contains everything needed to read/write and create new chunks. So a pool of pre-made chunk entities, and a NativehashMap<int2,entity> that refers to each chunk "region", and all the BufferFromEntity/ComponentDataFromEntities along with helper functions that utilize all that for "convenient" access to any block in the world

It ended up creating a dependency nightmare which is why I ended up giving up on it. But it was working

[sage mulch]

Apologies for the ping, that was me. I thought if it didn't turn blue when I typed it it wouldn't ping you 😢

[stable sentinel]

cannot convert from 'Unity.Mathematics.float3' to 'Unity.Physics.BoxGeometry

[sage mulch]

Thanks for the reply though, I will be reading that during my break today hahah

Appreciate the help!

[stable sentinel]

entityManager.SetComponentData(myEntity, new PhysicsCollider { Value = Unity.Physics.BoxCollider.Create(float3.zero, new float3(1, 1f, 1f))});

I checked the Create function and it asks for float3

[zenith wyvern]

No worries, it's completely fine

[viral comet]

Guys, anyone using addressables with dots?

[sage mulch]

Would you say the performance improvements made the effort worthwhile?

If there were any improvements

[karmic basin]

I have only those

[stable sentinel]

yes

I was looking at the wrong API doc

[karmic basin]

BoxCollider.Create(new BoxGeometry { Orientation = quaternion.identity, Size = new float3(1.0f, 1.0f, 1.0f) }); ?

[stable sentinel]

they changed

yes that works

[karmic basin]

oh haha

[stable sentinel]

it's just that if you google boxcolliders

there is an example

pointing to an older doc

but I guess if I actually read the error I would have figured it out

[karmic basin]

at the top of the official doc there is a dropdown to select your version of the package

• a blue "latest version" button

[toxic crest]

If I have multiple NativeArray<T> which all point to the same memory and I call dispose on one, does it wipe out the memory and invalidate all that point to it or does it just remove that specific point to the memory and decrement the reference count?

[hollow sorrel]

@toxic crest yes it invalidates all the others

[toxic crest]

Do I need to do anything when I'm done with one of the copies but want to keep the others valid or do I just let it fall out of scope and Unity handles the rest?

I might be wording that oddly.

What I mean is, do I need to do anything if I'm done with one but not done with the others?

[hollow sorrel]

no you'd need to manage when to dispose yourself

unity doesn't check if multiple nativearrays are pointing to the same memory

[toxic crest]

Thats what I was hoping. Thank you

[hollow sorrel]

architecture wise, decent rule of thumb - if it instantiated the array, then it should take care of disposing the array
so if you're passing same nativearray to other classes then they shouldn't have to worry about disposing, only the one that made it does

[ocean tundra]

You cant use a SharedComponent type in a Entities Foreach With burst!!??

Or is this one of those fun ordering issues

my shared component is still a struct

[sturdy rune]

You must set the value of a shared component on the main thread using either the EntityManager or an EntityCommandBuffer. When you change a shared component value, the affected entity is moved to a different chunk. If a chunk already exists with the same values, and has enough room, the entity is moved to that chunk. Otherwise, a new chunk is allocated. Changing a shared component value is a structural change that potentially creates a sync-point in your application.

[ocean tundra]

yup

[sturdy rune]

https://docs.unity3d.com/Packages/com.unity.entities@0.17/api/Unity.Entities.ISharedComponentData.html

[ocean tundra]

not changing at all

just trying to read in a foreach

I ended up with using GetAllUniqueSharedComponentData, then sceduling the job more times (1 per unique shared component)

it will probably be faster this way 😛

cant even use a SharedComponent in a ECB 😦

[zenith wyvern]

Yes you can

AddSharedComponent

[ocean tundra]

yea from a job with burst tho?

[zenith wyvern]

No

[ocean tundra]

giving me some weird error

yea

even tho its a struct

[zenith wyvern]

Yeah we can't u as them in jobs or burst yet

Maybe some day

[ocean tundra]

annoying

a struct based one with 0 managed references should be fine 😦

[hollow jolt]

anyone knows a position grid implementation done for dots ? im using ecs-phyics and want to apply collective forces on large amounts of bodies using a wind simulation

[ocean tundra]

its very easy to make one

[hollow jolt]

doesn't physics already uses something like this ?

i did try search for grid keyword in the source but nothing seemed to pop up

what do they use to avoid all VS all collision checking i wonder

[zenith wyvern]

Nothing that they expose to users

[ocean tundra]

i dont know if theres one in physics

[hollow jolt]

dots-physics ?

[zenith wyvern]

https://github.com/needle-mirror/com.unity.physics/blob/master/Unity.Physics/Collision/World/Broadphase.cs

[hollow jolt]

interesting choice of naming

( first time hearing this one )

thanks

[karmic basin]

They do one in the official physics sample but IIRC it's trigger based, not grid based

But yeah you can do your own spatial partitioning solution

They do one in the official physics sample
What I meant "they do" is some kind of vortex/tornado that apply forces to RB's in the zone, and IIRC they handle the all-VS-all collisions check

[acoustic spire]

Is it possible to know if code is compiled by burst or vanilla compiler?
I want to throw exceptions or mb log if I use a function not in jobs
E.g.

#if !BURST
Debug.LogError("Bad")
#endif

[winter zodiac]

I guess ... inside the burst inspector?
There you can see, which scripts are compiled by burst, afaik.

If you run the code right at the beginning of the game, you have to enable "synchronus compilation" first - because otherwise it will asynchronously compile for burst, meaning, it will run on "vanilla compiler" (I guess, C# Rosslin Compiler but IDK).

A check within the burst code would be really handy - but as this is all on the C++ side of things, or Assembly, I guess, I think the only way to view it, is in the burst inspector.

[north bay]

You can use: ``` bool isBurst = true;
IsNotBurst( ref isBurst );

    [BurstDiscard]
    private static void IsNotBurst( ref bool value )
    {
        value = false;
    }

Sorry about the formatting

lol

[acoustic spire]

Usually I define this flag manually and turn it on/off for debugging. I guess there should be used something like that internally

[karmic basin]

You can use Debug.LogError though ? Why the hassle ?

[acoustic spire]

Didn't remember if I can. But there are other use cases for this

[winter zodiac]

If it is a job, it wouldn't show the error log, no matter if compiled with burst or not (atleast, it was like this when I worked with it, that was one year ago)

[north bay]

They fixed that, you can now use the Debug methods again

[karmic basin]

https://docs.unity3d.com/Packages/com.unity.burst@1.5/manual/docs/CSharpLanguageSupport_Lang.html#partial-support-for-strings-and-debuglog

some caveats, but you could also plug a debugger

instead of printing strings :p

[winter zodiac]

Thats pretty dope and will make things much, much easier when I get back to the project in ~2 years

[gusty comet]

i hope bolt gets dots support soon
visual scripting would make jobs/ecs a lot easier to work with

[deft stump]

someone remind what's the attribute of disposing a native collection on completion?

[karmic basin]

WithDisposeOnCompletion(myNativeArray) ?

[deft stump]

normal jobs

[karmic basin]

:IJob ?

[deft stump]

yep

[karmic basin]

[DeallocateOnJobCompletion] ?

[deft stump]

that one

ty

[deft stump]

When will unity release a blog post on dots?

[hollow jolt]

where can i find this ?

[amber flicker]

I'm really struggling with reducing the main thread cost of my systems. I'm exploring caching the ComponentTypeHandle's - has anyone else tried this? Also wondering what's the best way to know when they become invalid

[amber flicker]

(caching the handles makes no discernible difference but still curious about how you know when - using the global version if the type didn't seem to work)

[karmic basin]

lemme check

https://github1s.com/Unity-Technologies/EntityComponentSystemSamples/blob/master/UnityPhysicsSamples/Assets/Demos/2. Setup/2d. Events/2d1. Triggers/Scripts/TriggerVolumeForceFieldAuthoring.cs

Used in this unity scene EntityComponentSystemSamples/UnityPhysicsSamples/Assets/Demos/2. Setup/2d. Events/2d1. Triggers/2d1c. Triggers - Force Field.unity

IIRC they implemented custom trigger events through a buffer, to keep track of entites in onEnter onStay & onExit states, as I guess it's not ready with dots. Then it's pretty much what you'd expect

[hollow jolt]

very cool site

doesn't looks like its using the broad phase volumes

[karmic basin]

nope it's not

If you find a way to use them, I'm interested 🙂

[hollow jolt]

im looking at Broadphase now and it looks like the data is stored in a tree structure Nodes

[karmic basin]

isn't it internal stuff ?

[hollow jolt]

nothing is internal in an open source project lol

[karmic basin]

oh so you wanna learn how they do it to replicate in your own code ?

[hollow jolt]

first i would have to understand what is going on there, and i would like to schedule grid based forces on the physics world without redoing / replicating too much stuff

or rather scan the tree and find the appropriate grid cell from there 🤔

[karmic basin]

You can study the Boids sample too

they did spatial partitioning

though it would be old now, and not using the Broadphase

[sturdy rune]

I am struggling to understand write groups. Just learning DOTS

When you enable write group filtering in a query, the query adds all components in a write group of a writable component to the None list of the query unless you explicitly add them to the All or Any lists. As a result, the query only selects an entity if it explicitly requires every component on that entity from a particular write group. If an entity has one or more additional components from that write group, the query rejects it.

so if you add the component b to a write group of say, component a

the query will exclude that entity with component a as if the entity had component a and b, as if it was a "with none(component b)" query ?

so any query looking for query a, will skip entity with component a unless it also has component b

[fluid kiln]

If B has write group on A

A system will only execute if the entity has A and has B

---

Does anyone know if there's a performance difference between having multiple ForEach in a system or instead multiple systems?

[ocean tundra]

I believe its faster

having many foreachs in one system

as you dont have the overhead of a whole new system

BUT you should probably manage the foreach Dependency yourself, as by default all Foreach's will be dependant on the one before

[deft stump]

why is the GC so huge?

I'm only disposing 8 items,

[ocean tundra]

@deft stump Safety checks and stuff on?

[deft stump]

now off

@ocean tundra

[ocean tundra]

@deft stump no clue sorry

What are you disposing?

[deft stump]

4 float3's and 4 floats

[ocean tundra]

Maybe they are huge things?

Oh well not huge then :p

[deft stump]

right?

AHAHAHA

[ocean tundra]

Why are you disposing them?

[deft stump]

coz.... they're in native containers?

and they need to be disposed?

[ocean tundra]

Could you reuse the container for next frame/run?

[deft stump]

I can...

so I allocate it via Allocator.Persistent?

[ocean tundra]

Yea

I normally keep all my containers around for as long as I can

Probably left over from old unity

And dispose in ondestory

[deft stump]

tried it

still spikes

[ocean tundra]

What else are you doing?

Must be something else

Hybrid objects have a leak

Somewhere

[deft stump]

@ocean tundra nope. Im not using Entities.
Just normal ijob

[mossy drum]

Hello, do you know if there is any performance benefits from using entities and the hybrid renderer only for static environment objects, in an otherwise standard Monobehaviour project ?

[frosty siren]

Should this "Auto load scene" work in build?

[karmic basin]

yes

🤔 the ConvertToEntity checkbox should be auto ticked because of the subscene though. That's strange Nevermind, that's the case for children, not the scene itself, my bad

[frosty siren]

whoooa, i tried to add convert to entity to subscene And something really strange began to happen. Tones of errors, empty new gameobjects in scene after exiting play mode.

[karmic basin]

my component displays a little more info on what scene is opened or not

& i can load entities in the editor from this component also

You're not on latest versions ?

Anyway, saving your subscene (eventually closing it?) should serialize it to disk and yeah it's supposed to autoload on a build

Just to be sure, new build workflow ? not the legacy one ?

[stone osprey]

Soooo... i have written a system which takes about 13-14 ms to execute each frame... is this considered "slow" ?

[karmic basin]

that's 70+ frames/s

but still yeah I'd consider it slow

depends on how many entities, what workload from your other systems

could probs improve perfs yeah

[stone osprey]

Thanks ! That system is a bit ugly... i basically acess mono components in there ( Animator... ) to transfer my data from the components into that animation states... well, then i need to take a look at what i could possible improve

[karmic basin]

If you did all you could do, don't get crazy and review later

If you're working with mesh data, I guess that's expected to be heavy

🤷‍♂️ so yeah depends a lot on what this system is doing

[stone osprey]

Hmm... is it possible to track the performance of single Entities.ForEach operations ?

Instead of the whole system ?

[frosty siren]

2020.3.1f1 entities 0.17

[frosty siren]

with build configuration one?

[karmic basin]

Yes with a profiler marker

[stone osprey]

@karmic basin Any tutorials on that one ? ^^ Or any examples ?

[karmic basin]

https://docs.unity3d.com/Manual/Profiler.html.

[frosty siren]

@karmic basin i don't use new build workflow. Today i have tried to find some docs, but only found few forum messages

[karmic basin]

and https://docs.unity3d.com/ScriptReference/Unity.Profiling.ProfilerMarker.html

[karmic basin]

With DOTS you HAVE TO

[stone osprey]

Thanks a lot ^^

[karmic basin]

don't build from the build menu, build from the config file

or asset thingy

[frosty siren]

where can i read about it?

[karmic basin]

Like first page of the manual

[frosty siren]

O_o

[karmic basin]

https://docs.unity3d.com/Packages/com.unity.entities@0.17/manual/install_setup.html#standalone-builds

Well second one 🙂

[stone osprey]

Uhmm... another question... im using those "Object Components" as some bridge between gameobjects and ecs. Is there already some way to let a Entities.ForEach with them run scheduled ? Normally the acess to Object Components was main thread only

[karmic basin]

Not sure. I would capture relevant data on main thread and pass it to jobs in allowed collections

[frosty siren]

i see. I was skipping this every time because whole page named ProjectSetup

thank you👌

[karmic basin]

^^ + they only detailed this on 0.17

Took me a few days to understand it was a problem, because building from the menu looks like it kinda works

[frosty siren]

yes!

[karmic basin]

So at first I duidn't understand subscenes workflow at all, when the basics are simple

[hollow sorrel]

@stone osprey what k said but also 13ms is very slow for 300 matches even main thread, would prob figure out the root cause before hiding the cost with other threads

[stone osprey]

@hollow sorrel Thanks, im gonna look into it, its once of my only systems that heavily interact with unitys Animator and AnimationClips. Probably mess up somewhere

I do a lot of lookups in there each iteration for example to override animation controllers

            // Override animations
            Entities.ForEach((ref Animation anim, in AnimatorOverrideController animator) => {

                var animationState = anim.state.ToString();
                
                // Get animation from database
                var contentStorage = database.GetContentStorage(anim.clipID);
                var replacementClip = contentStorage.Get<AnimationContent>();
                
                // Receive a list of all overidden animations
                var overrides = new List<KeyValuePair<AnimationClip, AnimationClip>>();
                animator.GetOverrides(overrides);
                
                // Check if replacement already happened for prevent it from doing this every frame
                var clipToReplace = animator.animationClips.FirstOrDefault(clip => clip.name.ToLower().Contains(animationState));
                if (overrides.Exists(pair => pair.Key.Equals(clipToReplace) && pair.Value != null && pair.Value.name == replacementClip.ClipName)) return;
                
                // Replace the animation of the certain state with out new animation we wanna play
                if (clipToReplace != null) {
                    var animatorr = animator;
                    replacementClip.Clip.LoadAssetAsyncIfValid<AnimationClip>(handle => animatorr[clipToReplace.name] = handle.Result);
                }

            }).WithoutBurst().Run();

For example this here

[frosty siren]

arrrgh, every time feel pain installing preview package

[stone osprey]

Right top you can see the alloc of the system... i bet this is too much garbage created each frame, right ?

[hollow sorrel]

thats a lotta allocs

the overrides list you can cache outside the foreach, and instead of using linq firstordefault a regular for loop will be faster

then you still have a whole bunch of string allocs but not sure how you can get rid of that, i dont use unity animator

if you can use id's instead of strings that'd be better

[stone osprey]

Thanks im gonna try that ^^ Yeah... im using FixedString32 but well... the unity mono stuff uses regular strings, so i need to convert the fixedString32 to its string form which also causes massive gc allocs. The problem here is, neither string nor FixedString32 provides any method to compare each other without any allocation happening :/

[hollow sorrel]

yea fixedstring prob not gonna help

but clips should have an unique id i think, if you could use those for comparison instead of name thatd be nice but not sure if possible here

[stone osprey]

@hollow sorrel I really should do that :/ Its just so much nicer to use strings because its more readable xD

Besides that... Any idea how i could implement a FixedString32 equals method which compares itself to a string ?

Without converting the fixedstring32 of course 😉

I assume that causes like 90% of my allocs right there

[safe lintel]

Pretty sure theres a clipname to hash function for animation

[stone osprey]

@safe lintel I need to look into this ^^ Nevertheless a better equals for the FixedStrings would be great... something that uses pointers without allocating a new string every time

[stone osprey]

What the hell... my optimization approaches are not working at all 😮

[karmic pilot]

well you could create a FixedString32 from the given string once and use that one for comparisons. Comparing FixedStrings is cheap and fast

[stone osprey]

Thats probably a smart idea... does this allocate anything ?

[karmic pilot]

In general you could try to keep the usage of strings within ECS to a minimum and only use them when interacting with your "traditional" OOP layer for stuff that is not yet ECS native

it only allocates the length of the string on the stack in form of a struct

[stone osprey]

On the stack only ? That sounds fine ^^

That shouldnt count as a gc alloc call in the profiler, right ?

[karmic pilot]

you can always try and measure 🙂 especially with performance related metrics, measuring is most of the time faster than guessing

[stone osprey]

Im still a noob in terms of profiling... For example this here :

            // Play animations
            Entities.ForEach((ref Animation anim, in Animator animator) => {
                
                var clipName = animator.GetCurrentAnimatorClipInfo(0)[0].clip.name;

                // Set state and override animation to the one from our animation component, for triggering transistions
                foreach (var trigger in anim.triggers) 
                    animator.SetFloat(trigger.Key.ToString(), trigger.Value);
                
                anim.clipName = clipName;
            }).WithoutBurst().Run();

A little system i wrote... it basically sets a variable inside the animator to play an animation

And the profiler tells me that it causes garbage

About 400Bytes

For one single entity 😮

And i have no clue why

Theres no allocation going on... atleast i see none

[karmic pilot]

toString() allocates

[stone osprey]

Oh...

That explains the allocation

[karmic pilot]

no idea what SetFloat does internally, haven't used that for some time, but I guess some string comparisons and allocations as well

[stone osprey]

Hmmm... is there any way to convert a FixedString32 to an string without allocation ? I assume not, right ?

Thats a dumb question, of course theres always an allocation

I mean if theres a smarter way

Or any to prevent this

[hollow sorrel]

there is not

[karmic pilot]

as far as I know you have to allocate... I do not think there is a way to reuse the memory of fixedstrings for string even if the memory is on the heap

[stone osprey]

Oh thats great :/ Damn... that makes it awfull for many entities

[hollow sorrel]

comparing fixedstring to a string should be doable tho, dunno why thats not built in already

[karmic pilot]

it is, but it uses ToString().Equals(other)

[stone osprey]

Yep it is... i already wrote a extension method which does not allocate ^^

Simply loops over its chars/bytes and compares them

[hollow sorrel]

:+1:

[stone osprey]

Probably i should use a class instead to store strings. This way i wouldnt be forced to conver them each frame

But then this cant make use of burst...

But that doesnt matter because its running on the main anyways

[karmic pilot]

is this system only about driving animation triggers ?

[stone osprey]

Yes ^^

Actually yes

[karmic pilot]

does this need to happen every single frame for every single animation ?

[stone osprey]

Cant explain everything but it works hand in hand with the networking... i receive packets from the server which manages the animation states and i write it into the ecs for triggering the translations

hmmm

actually not

[karmic pilot]

can you separate the part of making the decision of which animation on which entity to progress and the actual setting of the value? the first part can probably be done completely without any OOP part, only when actually updating the animation is it important to interact with the real animation system. At least that's my guess from your code sample

[stone osprey]

Yes, that should be possible ^^ i store the "state" inside the Animation struct. I only use that to tell the animator it should which conditions are true. But Those do actually rarely change. So i could mark them somehow to trigger those iterations when they are really needed

[stone osprey]

This alloc thing is weird

I call this method in one of my iterations :

With those comments in there... guess how much it allocates per frame ?

240 Bytes

And it doesnt make sense at all, why should this here allocate bytes ?

[karmic pilot]

what is the allocation per call ?

[stone osprey]

The profiler wont tell me that... atleast i havent found any way to acess those details

[karmic pilot]

try enabling deep profile

[stone osprey]

But well, i have one entity which matches the loop. And in that entities.foreach iteration i call that method once. So i assume that its 240Bytes per call. Which kinda fits the 0.9MB i get for hundreds of entities

Im gonna do that ^^ one moment

[karmic pilot]

or you could add profiler markers into the function

[stone osprey]

The lambda job 3 is my iteration above which calls that GetClipByName method

Without GetClipByName its allocation is... zero

But well, my method is pretty empty besides that loop :p im also gonna add some profiler markers in there for the next debug

[karmic pilot]

how does the bracket operator look like for the AnimatorOverrideController ?

[stone osprey]

I thought the same... but im not using the bracket operator for AnimatorOverrideController, im using it on .animationClips which is a "AnimationClip[]" array

Heres the profiling for the GetClipByName method

And well... 240 Bytes allocated within one call

Oh wait

Is this the problem here ?

[karmic pilot]

in your first screen the get animation clips access on the animator seems to be the offender

[stone osprey]

I see, i see... the ".get_animationClips()", right ?

[karmic pilot]

yeah

it seems like an extension method, can you post this one ?

[stone osprey]

I try... its not mine ^^ Its from unity itself

[karmic pilot]

k then I can find it myself 🙂

[stone osprey]

This makes it even worse

Actually i only acess the array... why should it even call those weird ".get_animationClips()" methods ? Where they ? .o

[karmic pilot]

looks like a copy of the internal list

so every time you access animator.animationClips, it returns a copy of the internal list as an array

[stone osprey]

Well, that explains the massive gc alloc going on there

I actually wonder why they implemented it like this

[karmic pilot]

🤷‍♂️ who knows, I haven't worked with animations for years

[stone osprey]

Thanks for helping me ^^ i think im gonna try to cache this somehow to prevent that copy every time. I hate that animation system so much :p

[karmic pilot]

probably to make sure that there is no way you can edit the clips without going through some sort of API on the controller

if the data never changes, create a lookup for it once on creation of the system (or wherever needed)

so unless you add/remove clips to animators, this data should be constant

[frosty siren]

can it be that order of systems are different in editor and build? (Those systems that have no order attributes)

[ocean tundra]

maybe a little bit

if its causing issues then you should add order attributes

[frosty siren]

but there can be an issues that i can't understand just by playing build

[ocean tundra]

i also wouldnt be supprised if il2cpp had a different order too

[frosty siren]

and while i don't know exact build system order i can't assume what cause issue

[ocean tundra]

if your code has any dependancies between systems you should add order attributes

but otherwise order shouldnt effect your code

maybe youve missed a dependancy?

[frosty siren]

yes, but i want to minimize order usage

and there can be dependencies you are not thinking about

until it thrown an error

[ocean tundra]

sounds weird to me

if the order matters then you need the order attributes

but ideally your code should work reguardless of order

and the worst that happens is the processing is broken up over frames instead all the same frame

[frosty siren]

Yes, but we use editor mostly to see how things works without do full build every time. So if things go different in editor and in build then editor idea is little broken. Maybe i'm not understand something

[ocean tundra]

😛

i would look elsewhere for the cause of the error, rather then thinking it might be ordering

unless you know 100% that system X will throw if before system Y, and in that case you should have a order attribute

[frosty siren]

I already have fixed my issue and it's gone after i write order atttributes) In editor no changes happend, but build now works correctly

[ocean tundra]

nice

[frosty siren]

i'm afraid about my future 🙂

maybe there are more hidden order bugs in my project

[ocean tundra]

i dont think ordering has much performance cost btw, its something done on start so not something to worry about

[frosty siren]

there is no performance cost, it's just system instantiation ordering

[karmic pilot]

if you are afraid about ordering bugs, you might need more world tests in built players :>

in general all dependencies are logical consequence of some read/write collision or explicit defines, so there shouldn't be some hidden dependencies you do not know about

[stone osprey]

So i heavily optimized my gc alloc code... Basically via caching ( Did you know that even Object.name allocs ? Who the hell designed this ? ). One issue left... FixedString32.toString(); is there no way to get a temporary string from it ? Is it probably possible by writing an own extension ?

Wow... that cache system is the ugliest one i ever wrote... there multiple dictionary lookups in each entities.foreach iteration just to prevent the convertation from FixedString32 to string... or to prevent .name lookups... so damn ugly

[ocean tundra]

why so many strings?

can you replace them with a enum

or numbers?

or even use the string hash code?

[hollow sorrel]

if you find a way to lemme know, i've tried also and it seems a .net string will always be an allocation no matter what because it's managed memory
pretty much impossible to have an unmanaged string because it'd no longer be a string (not including different types)

[coarse turtle]

hmm well if the string names are unique what about using a checksum like hash 🤔 it doesn't exactly solve the problem of char* to string allocation in C#. But it's pretty much something I ended up doing for "nameable" ids

[minor sapphire]

has anyone experienced an issue with hybrid renderer v2 where the screen goes black with certain camera positions?

[ocean tundra]

never seen that

what rendering setup?

HDRP/URP, V2 define added?

[minor sapphire]

hdrp v2 define added

it's been working most of the time, very odd, particular camera angles

hmm in frame debugger I can see the step

post processing hmm

[ocean tundra]

your using post processing too?

[minor sapphire]

yeah I guess I need to try turning all the effects off

oof turned them all off and no change... perhaps a default one is causing problems 😐

[ocean tundra]

turn off post processing completly?

[minor sapphire]

yeah I just did

black and white scene now but at least I can see something 😄

the culprit is somewhere in here 😄

[ocean tundra]

i had soo many issues with hybrid renderer

was using it with URP + shader graph shader

was awerful

cant wait till the update it so it works

😛

[minor sapphire]

oh man it's still pretty buggy

but it's getting there 😄

the culprit was bloom

I don't get it though, there is bloom in the default settings, and then I have a global volume override, but they don't seem like the same thing...

turn off override, it calls back to default, fine

turn off default, scene changes, even though I have the override enabled

hmmmmm

I must be misunderstanding how this works

I don't know how bloom makes the entire screen turn black though 😐

stop NaN's on camera fixed it

[stone osprey]

Hmmm... Lets say i have an iteration which looks kinda like this...

Entities.ForEach(...) => {

    .
    .
    .
    if(someCondition) return;
    SomeClass.InvokeCallback(() => {}); 
});

Would this execute gc alloc each iteration ? As far as i know lambdas require a gc alloc... I would also like to know if it would alloc the lambda each iteration, or just when the condition is met... any ideas ?

[slim nebula]

The lambda has no capture variables so the new action will be cached statically and reused each time.

[coarse turtle]

that's cool, had to look at SharpLab to see what the compiler would do lol

[remote crater]

I'm confused. Why can't I set an Entity variable to null?

What should I use in its stead?

[north bay]

Because entity is a value type and not managed (or a ptr most likely in your case). Use Entity.Null instead

[remote crater]

oh, ty

I was already starting on making 2ndary vars bool myEntity exists;

lol

typo psudocode

Another dumb question, what is the standard way of linking entities in my scene to be used in scripts?

I think I could make a script to assign this to a singleton's public static entity SpecificEntity;

But every single Entity would require a custom script

I'm not going to endlessly ask questions.

These two would be good for an hour

Wow, funnily enough, this is the standard way of doing stuff!

https://forum.unity.com/threads/reference-to-another-entity.847954/

using Unity.Entities;

[GenerateAuthoringComponent]
public struct Enemy : IComponentData
{
public Entity currentTarget;
}

So like then you assign it to the singleton. I suppose what I will do is have an Entity variable and an index variable to know what Entity it is to feed into the right singleton value. It is crazy you can't just assign it into a Monobehavior.

They way they're doing it seems pretty dumb: They should probably fix that so you can drag and drop ENTITY stuff into monobehavior that declare public ENTITY myEntity;

As it is now, when I declare a: public Entity myEntity in a monobehavior, UNITY doesn't let me drag and drop anything into that slot. I add public int i; and it lets me drag, so its definitely not liking autopopulating Entity on monobehavior.

[zenith wyvern]

Entity has no edit time representation

[remote crater]

Well it could

Simply link scene index to what the translation becomes

Super simple

[zenith wyvern]

You're meant to build traditional gameobjects/prefabs in the editor then use converison to change them into entities

[remote crater]

Yes, I know, but if the gameobject to change into entity is in the scene, it could have an index

Which translates to the entity

Its not even tough to think about

Welp, I'll just accept the DOTS programmers made a UI error. I've worked with bad UIs in the past like Starcraft2 mapeditor

Maybe they can fix it later.

[zenith wyvern]

Interesting way of looking at it, good luck brother

[remote crater]

Thanks bro. I probably should be working at Unity, lol

Ok, seeing how basic functionality is missing, I have to ask: Have they added a Debug.Log facsimile in DOTS, or is this article still relevant: https://forum.unity.com/threads/as-debug-log-is-a-problem-in-job-i-made-this-simple-trick.860464/

[zenith wyvern]

You've been able to use Debug.Log in bursted jobs for a few versions now

[ocean tundra]

Look at burst function pointers too
https://docs.unity3d.com/Packages/com.unity.burst@1.5/manual/docs/AdvancedUsages.html#function-pointers

[stone osprey]

@slim nebula Thanks ! but what if i use variables inside ? Then it should use gc allocs right ? I

@ocean tundra Thanks im gonna look into this ^^

[slim nebula]

yup. then it reallocates every time

[stone osprey]

@slim nebula Thanks... that explains my damn gc alloc calls... the profiler couldnt find the exact position so i needed to guess. Guess i need to find a cheaper way

[slim nebula]

the capture variables compile as member variables on a class. If you're using the lambda class on multiple threads, you need multiple instances of the captured variables, hence multiple versions of the lambda class

since the language doesn't know if you're using multithreading it just always does it

[stone osprey]

In my case i dont use multithreading ^^ it actually runs on the main, the example above was just simplified 😄

[slim nebula]

then just use a static function in your system, and set member variables to replicate the captures

[stone osprey]

I think a good alternative is to dispatch commands into a queue for executing those callbacks else where

@slim nebula Member variables ? You mean params ?

[slim nebula]

if the thing you're passing the lambda has X params and you need to capure more variables (Y) to use inside the function body, then you can create static member variables for Y, and the parameters (X) still get passed in the function prototype normally.

but the function prototype matches the caller requirements, if you need more variables you use a class variable

w/e there's many ways to work around the issue. whatever works for ya hehe

[stone osprey]

I Thanks, im gonna try a bit around ^^

[turbid sundial]

I've been reading up on Utility AI/IAUS and thought that from the outside - DOTS would make a good choice for a prototype. Many NPC entities need to consider and score many things in parallel - but when it comes to implementation I have some questions with the ECS style model. If for example, one of my AI considerations is - "distance to nearest tree is less than 50 units" - how might I go about calculating this or rolling this up into component data? Should every NPC have a list of all the trees within a radius of them that is constantly updating? Is that too much data to store? If there are trees instead as entities and I just want to do some basic distance math - can I query for tree entities while I'm working on a different entity? (e.g. in an NPC entity foreach, can I then query all of the tree entities to get an updated list of nearby objects?) Kinda just thinking out loud here but if anyone has any thoughts I'd appreciate it 🙂

[slim nebula]

at a high level, yes you can do all those things

more specifically ECS is a way to store data and a framework that compiles into native code that operates on that data. you still need to structure and organize your data (index/hashmap/whatever) depending on your requirements. I think there are many ways to approach what you're planning

[zenith wyvern]

Re "In an entity ForEach I can query for x" You cant run a query/ForEach inside another

[ocean tundra]

@turbid sundial Have you taken a look at unitys AI Planner

[turbid sundial]

That's good to know - what might be a good way to determine "nearest tree", if that needs to be constantly updated

[ocean tundra]

its on hold now but was looking good

[turbid sundial]

I have yeah, I'm just experimenting at the moment - trying lots of things

[ocean tundra]

could dig into its code to see how it did things

[turbid sundial]

Tangentially related - if I were to build a procedural world with trees - but I want to cut down a tree, travel a thousand miles - then come back and make sure that tree is still cut. Initially my mind goes to instead instantiating the trees procedurally and having a list of "cut" trees, and preventing them from being instantiated if they were in that list - but 1) that doesnt seem like it would scale well into late-game and 2) it doesn't take into account trees that a player might plant themselves

[stone osprey]

Arent you actually storing each chunk ? Well actually only those chunks which were modified ? Atleast thats how minecraft does it... player travels trough the world, chunks wont get saved... player placed a block or modified the chunk, modified blocks are getting saved into its own textfile.

Those are based on seeds, that means when the player comes back some day, the chunk generates the same chunk based on the seed. Then it removes/adds blocks based on the added/removed log from the chunk file

[turbid sundial]

Yeah chunking them like that would definitely be more scaleable - I guess ultimately there will always be a limit though

that's a good point though, thanks

[remote crater]

Confused on DOTS.forEach. What is the syntax to read three variables? Entity myEntity; int index1; int index2;

[ocean tundra]

😛

all in the docs

@remote crater https://docs.unity3d.com/Packages/com.unity.entities@0.17/manual/ecs_entities_foreach.html

basicly you cant attach raw types (int) to a entity

you need to make a struct containing your data, attach that, then you query for that

using ref (for read write) or in (for read only)

there are also some special values you can get which are indexes

maybe you were trying to get those?

https://docs.unity3d.com/Packages/com.unity.entities@0.17/manual/ecs_entities_foreach.html#special-named-parameters

[remote crater]

Of course I had a struct but it wouldn't let me post it here. I let me see if I can link the problem: https://old.reddit.com/r/Unity3D/comments/mfxqf6/confused_on_dotsforeach_what_is_the_syntax_to/

[ocean tundra]

so replace your ???? with
in DEntity data

[north bay]

You access (ask for) the struct not the individual fields @remote crater

[ocean tundra]

and if you want THIS entity then your first paramater will be Entity

[remote crater]

So: .ForEach((ref Entity entity, in DEntity data) =>

[ocean tundra]

almost

dont ref the entity

or in

its just (Entity entity, in DEntity data) =>

[remote crater]

One final question and I assure you I'll stop asking questions for many hours: How do I access index1 to print? UnityEngine.Debug.Log(data.index1.ToString() ); //gives an error on run: Burst error BC1016: The managed function int.ToString(int* this) is not supported

[ocean tundra]

dont tostring?

[remote crater]

Dif error

[ocean tundra]

or use string interploating or whatever its called

the $"{variable} text {nextvariable}" syntax

[remote crater]

But the variable I am accessing is correct: data.index1 ?

That's my main question

[ocean tundra]

yup

[remote crater]

Shoot I said I wouldn't ask another question, but I guess its clarification on the same!

[ocean tundra]

😛

[remote crater]

Thanks everyone, you rule.

[ocean tundra]

all good

we are here to help

kinda

and ask a ton of our own questions

[remote crater]

Hey if I make a bunch of money on my game, and you remember me and you see me or email me, I'll send a brown manilla envelope in the mail.

[ocean tundra]

haha

[remote crater]

🙂

[viral comet]

Hi! I understand correctly, if value type parameters are captured, then there will be no allocations though, right?

[slim nebula]

iirc value types get boxed for lambda captures, which makes sense since the lambda updates the values if read outside the lambda afterwards.

so no, still would allocate a new lambda object each time it's used

https://docs.microsoft.com/en-us/dotnet/csharp/programming-guide/types/boxing-and-unboxing

[viral comet]

sad 😦 Any chance to pass an Entity then?

avoiding allocs

[slim nebula]

If you just use a normal function instead of a lambda, and store what was previously a capture variable instead in a class member variable.

the problem stems from multithreading. if these callbacks are all running at the same time with different captured values, then you need to cache those values somewhere for your function to use.

you could create an array of classes in a static list so it's not allocating 100 each frame, but 100 up front and reusing them or something like that

I dunno there's many ways around it

a lambda function is just a new class with a function matching the lambda prototype, and boxed member variables for each capture. you could code this out the long way and adjust from there to reduce allocations

¯_(ツ)_/¯

I dunno if I"m helping

I think it depends on the specifics of the problem

[viral comet]

Thanks for the explanation, helped a lot 🙂

[turbid sundial]

https://i.imgur.com/8bCq3yR.png

MSPaint graphics style. This is a procedural streaming system. Blue circle is where the player currently is, as he travels left - the chunks obviously change (standard procgen stuff). In my DOTS design, when the player moves a set amount - I create a new array with all the positions, create chunks that don't exist yet, and keep chunks that already exist and should still exist after the update. However I'm not entirely sure what to do with the cleanup chunks (now in red). One thought I had was, before doing any updates - apply a DeleteTag component to all the chunks that exist - then as we go through and find ones that exist and should still exist, take that component off. I worry that this touches too much though? Any other good ideas?

[ocean tundra]

probably better to keep lists of your chunk entities

and not use tags

as each tag add/remove is a structural change

but then it kinda depends on how big each chunk is as the bigger they are the less they move in/out of range so less changes

[minor sapphire]

could go with dynamic buffer of neightbours? when all neighbours are remvoed you can delete it?

[ocean tundra]

i was meaning seprate native lists within a system

actually

this is your v0.1 right? as in just starting?

i would stick with tags, and add/remove whenever needed

when your further along and needing to optmize theres a change that Unity will have optmized tags by then, OR you could do any of 100 other things 😛

so im also planning some chunk stuff, my plan was to have a single component storing every directions neighbour

and use Entity.Null to tell if they exist or not

[minor sapphire]

suddenly the DisposeSentile.Create()s in unity physics are taking 25ms 😐 wtf did I change 😐

[ocean tundra]

😛 physics is a pain

so make sure you have burst on

and see if that helps

[minor sapphire]

yeah burst is on lol

I was running at 60fps no problem yesterday, now 20

it's all physics dispose sentinels

but I don't know why now lol

[ocean tundra]

turn off safty checks?

[minor sapphire]

I did that with no effect which is odd cause DisposeSentinels are safety system. But more to the point, I haven't had this issue with safety checks on before

[ocean tundra]

close reopen?

[minor sapphire]

I need to go through source control and look at every change in yesterday's check in lol

I restarted PC 😄

[ocean tundra]

clean librarys folder?

[minor sapphire]

hmm interesting idea

I'll try that after source control check

[turbid sundial]

@ocean tundra lists of the chunk entities is what I have - but the entities that are already instantiated but no longer in range need to be cleaned up - which is why I was thinking add the "Deletable" tag to all of them, and remove the tag if it is still needed

Maybe that's a weird way to think about it though

Ultimately the entities that are no longer in range need to be removed somehow

[ocean tundra]

na it will work quite well

you could use the EntityManager EntityQuery opperations

they are the fastest

[sage mulch]

Hi, playing around with some code rn. How would an Entity field in a struct work? Is that a reference to an Entity? And would having that reference be a bad thing in ECS?

I found this on the docs but was curious if anyone could tell me if it was bad to do or not:

Also curious which of these types it would fall under:

[minor sapphire]

It is the last dot point, a struct containing unmanaged blittable fields

and yes you can store references to entities in your components

it is like a reference in the sense that you can use it to lookup an entity, but it is not a reference type

it is stable in the sense that an entity will never change it's Id or Version, and it is unique.

@sage mulch

[sage mulch]

Thanks!

[sage mulch]

Ah this was where I was confused

[stone osprey]

Sooo... Im trying to write a deserializer for my packets to convert the content into structs for attaching those to entities.

IList<object> FromPacket(Packet pk);

As you can see im running into a problem... Boxing, but the returned list needs to store objects because the packet can contain multiple different structs. Any smart idea how i could return a list of different structs without boxing them ?

[karmic basin]

IList<struct> ? 😛

[north bay]

Allocate ptrs with your data so you return the list of ptrs instead of the boxed references?

[stone osprey]

That would be pretty smart, totally forgot that c# has pointers ^^

[north bay]

You could also do that with one ptr tho idk what's easier to do but one should be more performant

TO be honest dots feels more like C++ than C#

[stone osprey]

:D im gonna try around a bit till i have something working... Yeah, atleast when you go into detail ^^

[stone osprey]

Besides that... lets say i have a List<IComponentData> and i insert a Player : IComponentData{ FixedString32 username }... does this box ? The struct implements the interface so i would assume not... or am i wrong ?

[karmic pilot]

if you access structs through interfaces, they are always references and therefore boxing occurs. If you do not want them to be boxed, use the actual types directly

[stone osprey]

Damn... Ah fuck it, ill just go with pointers. Who doesnt love pointers ?

[karmic pilot]

if you don't like the boxing, maybe your approach is flawed and you need to look for another approach, depending on the reason for why you do not want boxing to occur

[stone osprey]

Well it causes gc allocs during the frame which is kinda bad for the performance.

[karmic pilot]

then why do you need the interfaces to access the data ?

[stone osprey]

I dont ^^ But as i said before... i deserialize a packet here and it needs to returns a list somehow to attach the deserialized structs to the entity

I just thought a List<Interface> would avoid boxing

[karmic pilot]

using a list in general will allocate for whatever you throw in there. If you know your types and you only want to pass them to the caller, use the low lvl variations for your structs like Unity.Collections.NativeList or something similar

[stone osprey]

And i dont know the types... its some sort of generic packet. Some kind of build instruction for entities on the client :/ So i cant use a specialised list like List<Player>

[karmic pilot]

well, then you might be stuck with pointers or try to disassemble the packet reading into something less "generic", as you always have a finite set of types that can actually appear and for which it might be possible to create specialized structures/caches

unless ofc you have no control over what the other side can potentially send you

[stone osprey]

@karmic pilot Alright, thanks ^^

Besides that... is it normal that an ECB allocates memory ?

Well the creation of the command buffer

Is there any way to cache this ?

Or is there any way to check if a command buffer is still valid ? I remember that i often received errors that told me i should create each frame a new command buffer

[pulsar jay]

Did you profile this in a build? Or at least with all safety checks disabled? Because DOTS does a lot of things differntly when enabling safety checks. For example using a standard List instead of NativeList to store command buffers

Does anybody know a good trick to copy an entity into an archetype? I want to create a copy of an entity which only contains the components in a specific archetype. I could use a ForEach with all components as in parameters and then attach them to a new entity via ECB. The problem is this reaches the component limit quickly and does not handle optional components

[stone osprey]

@pulsar jay Havent profiled it yet.. but i think i should, thanks ^^

Does anyone know if i really need to call .CreateCommandBuffer each frame ? Is there some way to cache this ?

[pulsar jay]

If you follow the CreateCommandBuffer call you can see what happens

It will just get added to the pendingBuffers list

Which gets executed (flushed) and then cleared in OnUpdate:

protected override void OnUpdate()
{
    FlushPendingBuffers(true);
    m_PendingBuffers.Clear();
}```

[stone osprey]

Hmmm... so that means we could actually reuse one command buffer ? We would just need to add it to pendingBuffers manually ?

[pulsar jay]

Thats not what I wnated to say here 😬

It might probably work but it is definitely not intended

It is intended to just be called each frame

As the command buffer is a struct it should not allocate memory on the heap (at least without safety stuff enabled)

[stone osprey]

Well... the problem is that im playing around with reflection because i cant acess the internal methods ( yes i tried using those fancy asmrefs )... and creating a new ECB each frame kinda breaks my reflection calls. I convert the methodinfo to a delegate to prevent unnesecary gc allocs. And when the command buffer changes each frame, i need to convert it each frame... so i actually want that i can use the same damn CommandBuffer :/

[pulsar jay]

Ok I see the problem now

But the methodInfo is created from the type (which does not change) so you could cache the method info and then just invoke it on the newly created command buffer each frame?

[stone osprey]

Yeah i also thought like this... and it works, but... it causes a massive amount of gc allocs. Because we can only pass a new[]{ entity, cmp } into it. This creates allocs and .Invoke(); also causes gc allocs. Thats why i went with the delegates instead :/

So i either accept the massive gc allocs... or i find a way to reuse the same damn ECB

Damn im gonna start a petition to force unity to just make all their internal methods public. This would make development much easier

[deft stump]

Or

make their code avaible to be read-only on github.

[pulsar jay]

You can already ready all the DOTS code because its in packages but it would at least allow forking

[pulsar jay]

I would second this. Just recently had the same problem with running into a lot of dead ends with addressables as everything is internal 😫

[north bay]

Oh btw @stone osprey maybe you can use NativeStream instead of a List of ptrs

[stone osprey]

@pulsar jay Oh damn.. im also using adressables but i didnt had any issues yet... but im using a hybrid system where i just use the ecs for the data... go's are used for rendering in my case ^^ @north bay Hmmm... thanks, im gonna look into them !

I think im gonna look for a way to reuse command buffers... They even have a dispose method, so i bet theres someway to do that

Is boxing/unboxing shown in the profiler ? Or is it just "known" as gc alloc in there ?

[pulsar jay]

How do you handle referenced entities that might have been destroyed? I have a lot of stuff like this: var targetLTW = GetComponent<LocalToWorld>(targets[0]); where it will throw an error if the target has been destroyed. Do I have to check if the entity is still valid in any system that accesses the target buffer or is there a better way?

[stone osprey]

I run a HasComponent() first. That should prevent this if im right

[north bay]

Can you clean the targets array up in some way?

For example I have a DestroySystem which runs after the EndSimulationSystemCommandBuffer but in the SimulationSystemGroup to destroy all entities flagged with a DestroyComponent.

This makes it so that all entity references are valid when CommandBuffers play the changes back (that was a huge pain for me before introducing that system).

But this also allows me to run a system after that DestroySystem which would clean references up and set them to Entity.Null.

[pulsar jay]

I tried this some time ago and it didnt work as expected (the cleaned up targets where still found by later systems). I did not really understand the dependency management back then though. I guess I would have to make the other systems depend (call complete) on the clean up system (or run the clean up system before a sync point)

[stone osprey]

Quick question... does unboxing also create gc alloc calls and result in garbage ?

[karmic pilot]

gc allocation usually happens when you convert the value type to a reference as you then need the indirection. unboxing is retrieving the value and should not allocate additionally (ofc there is now a copy of the unboxed value on the stack/register/wherever the value is used)

In general to better understand boxing/unboxing in C#, especially when it occurs, microsoft has a really good article on it in their docs on .NET

[ocean tundra]

@stone osprey As nasty as it is, codegenerating a huge amount of code to advoid boxing is pretty easy 😛

hmm but i see you want it into a list, It it a list of ANY components, or 1 component?

my networking syncs all the 'same' data together in the same packet (meaning i probably have too many packets right now :P)

but allows me to simplfy some code around handeling the packet as i know all data will be the same type/sizes

one of my earlyier tests had me generating a struct that contained a native list for EVERY networkable component type (so there were like 20 lists in that one struct)

[gusty comet]

Any thoughts on how to get a singleton entity inside a monobehavior? It seems GetEntityQuery is not available inside monobehaviors so

[ocean tundra]

World.DefaultInit.......EntityManger.Get(Or Create??)EntityQeuery

depend on if your using many worlds

or just the default

i use that to Create Entities from my Networking lobby

BUT my networking lobby script is added to a gameobject with convert and inject

so when reading/updating (using querys to change menu state) i have that in a system instead

[gusty comet]

I'm just not at all happy with the character controller in the physics samples so I'm trying to have a lot of data for characters in ecs but the movement needs to be on a gameobject

[ocean tundra]

if your doing alot of hybrid you should probably have all the stuff in systems/ecs and use gameobjects as 'Views'

as otherwise what is the 'truth' for your world 😛
like is ECS data right
Is MB?
Or is it a mix, for players the truth is MB, for AI its ECS

ect

[gusty comet]

I mostly need to access the character controller's .move function

[ocean tundra]

that can get pretty confusing

[gusty comet]

Because with ECS there is actually not a simple way to move an object with collision without rigidbodies

[ocean tundra]

ah i guess not if your mixing physics

[gusty comet]

You would actually have to write all the collision code yourself - I don't even have unity physics (dots) installed right now

[ocean tundra]

well you could call CC.Move from a system

using without burst and run()

but you would also need to find a way to send all the collision events though into ECS data

not too sure on the best way to mix legacy MB/Physics and ECS

[gusty comet]

I'm not sure if the character controller converts to an entity well. They have a really, really crappy ECS character controller in use cases examples but it was clearly just written to show off that it works it's impossible to understand or work with

[ocean tundra]

im guessing the DOTS shooter CC is even more complex 😛

[gusty comet]

I mean unity devs are still wrongfully obsessed with rigidbodies. I have no idea why they haven't figured out yet that most games dont use rigidbodies for their main characters

[ocean tundra]

honestly no clue, I dont usualy make character type games

normaly focused on RTS type ones

and definitly dont have RB's on them

well untill they die 😛 then it can be fun watching them all fall

[gusty comet]

Anyways this works - thanks for the advise

    SysInputGathering inputSystem = World.DefaultGameObjectInjectionWorld.GetOrCreateSystem<SysInputGathering>();
    Entity ent = inputSystem.m_HumanoidInputQuery.GetSingletonEntity();

[ocean tundra]

if you do World.DefaultGameObjectInjectionWorld.EntityMangaer there should be some sort of get entity query method

so you dont need to get a system first

but depends on if you want that system for other things

probably a good idea to have a 'bridge' system that most of your MB stuff uses to communicate into ECS