archived-dots

[viral sonnet]

do we care about the material though?

[rotund token]

1 mesh + X chunk transforms get passed to render batch

1 mesh happens to be 1 chunk

[viral sonnet]

yeah, I see. well, I rest my case 🙂 I don't see much to improve here then

what kind of cpu timings are we talking about @rustic rain ?

[rustic rain]

Idk, I didn't profile it

I literally just managed to make it work xD

With bugs

[viral sonnet]

then don't optimize 😄

[rustic rain]

I never put it into update

Gotta try

[viral sonnet]

do you have a 2d or 3d game?

[rustic rain]

2d

But with 3d models

Can't afford sprites kek

[viral sonnet]

you probably have cell data already, right? you can hash the mouse position and build a lookup table to find the hovered cells. finding anything based on position will be useful for other things too

I'd only use this entity pick system in the editor.

[rustic rain]

No cell data

I work in coordinate space

[viral sonnet]

hm, well I don't know your game. can I imagine it like rimworld or something different?

[rustic rain]

Ever heard of space rangers?

I'm doing clone of it and ftl

[viral sonnet]

i haven't but a google search gave me an idea 🙂

so if you're doing this I assume that you not use any physics or colliders?

[rustic rain]

Yep

That's why this entity picker is god save

[viral sonnet]

yeah, I see. just don't overdo it. having no spatial awareness makes a lot of things unnecessarily tricky

[viral sonnet]

OverlapSphere doesn't look too hot for 10k 🤣

[rotund token]

is that all 1 thread?

if you only update half a frame

that's 15fps!

i do think based off what i know your requirements in general are this will probably not be fast enough 😄

[viral sonnet]

yeah, just a Run() to see how things go. I wonder how MMO servers do this stuff. they have so many entities on a server it's insane. could run on players instead of enemies but still

sad to see physics is *that *slow for that. :/

[rotund token]

they aren't triggering 10k creatures/frame

that said most mmos 95% of the creatures are asleep

usually they sleep like 50m from a player

[viral sonnet]

some form of spatial trigger has to wake them up

and yeah, I guess most of the stuff runs okay because it's just not happening that often

when the players actually do the overlap cast it'll be a lot less. still servers can have 7-10k players

I had high hopes I can just throw this feature in and be done with it ... lol fk

[rotund token]

yeah that isn't 1 server though

that's multiple machines acting like 1 single server

[viral sonnet]

yeah true, what I'm even thinking ... haha

so I guess for the demo I'd have to put the overlap cast on the enemies and not on the casters, otherwise performance is down in the dumps

[rotund token]

you could optimise this a lot

if you have a bunch of things in an area

they dont all have to cast an overlap sphere

you can cast one and use the same result for all of them

[viral sonnet]

I'll certainly keep this in mind. Don't want to cheat too much as it would be unrealistic anyway to have casters all stacked up on a few positions 🙂 but yeah, talking about unrealistic with my kind of numbers. I see the irony

physics doc doesn't even mention ITriggerEventsJob

[viral sonnet]

what's the correct way of converting? this messes up physic blobs ```BlobAssetStore blobAssetStore = new BlobAssetStore();
GameObjectConversionSettings settings = new GameObjectConversionSettings(World.DefaultGameObjectInjectionWorld, GameObjectConversionUtility.ConversionFlags.IsBuildingForPlayer , blobAssetStore);

        prefabEntity = GameObjectConversionUtility.ConvertGameObjectHierarchy(go, settings);```

I don't know where to get the BlobAssetStore

[rotund token]

subscenes? ^_^"

[viral sonnet]

for hybrid, I now know that GameObjectConversionSystem holds a BlobAssetStore but GetExistingSystem returns null.

at least from the DefaultGameObjectInjectionWorld

I'm honestly surprised that the posted method doesn't screw up the blobs from the subscene. the problem started when I put a trigger on the hybrid spellcaster

triggers run even worse (as expected) also messes up physics time step and now I have 6 updates in 1 frame ...

option A) use sphere overlap cast b) build a specialized hashMap - I have to stop at one point with this madness haha, maybe now is a good time

[rotund token]

not really sure what hybrid has to do with it

blob asset management is a bit of a pain and it's pretty complicated under the hood

if you dont care about maximum memory saving then just create, retain and disposing your own store is fine

[solemn hollow]

Is there a way to set EntityQueryOptions directly in an EntityQuery object or do i always have to work with EntityQueryDesc for that

I just want to include Disabled Entities in the query

[rustic rain]

welp, I checked

1 ms takes

for thousand of entities

to be constantly removed/added with components, kek

for sake of fail proof rendering disabling

[solemn hollow]

how is tertles suggestion not failproof though? whenever someone changes the room you add DisableRendering. whenever someone enters you remove it.
I would have thought your solution is MUCH faster actually than you profiled. there should be no memory copied around and most chunks shouldnt even change at all since most chunks are disabled anyways. the few chunks that need to be enabled again should just have their archetype adjusted and nothing more. not sure how this could take so long.

[rustic rain]

it's not failproff because there could be potential bugs due to missing such changes

meanwhile fail proof is constantly checking for "dirty" entities

[solemn hollow]

Id probably try it with a reactive system. whenever an entity changes the room you do the checks

[rustic rain]

but it's not efficient

yeah

reactive system is best scenario

but here the problem

[solemn hollow]

i guess your problem is the SCD

[rustic rain]

it's Shared Component

xD

[solemn hollow]

just have another normal ICompData holding the room ID too and always write into both. then you can filter for changes on the normal one

[rustic rain]

I am working on it

first I want to make a difference between

Player view

and player entity

[solemn hollow]

for culling?

[rustic rain]

yeah, I guess you can call it culling

just need an ability to look into other rooms

at least for debugging purpose

[rustic rain]

btw, this is not an option, since writing to shared comp can't be done through buffer

[solemn hollow]

why would you need to write it through a buffer?

[rustic rain]

so I can have actions with it in bursted jobs

rn all actions I can do: are only through filter

so far it's enough, but can't wait for unmanaged shared comp

from Unity

[solemn hollow]

I dont get the problem. when an Entity leaves a room you write to SCD and ICD at the same time. Then a System with a ChangeFilter on the ICD picks up the change and sets your DisableRenderer.

[rustic rain]

I don't think I get the idea. Could you elaborate?

[solemn hollow]

I really dont know what more to say to that.
You must have some Logic that changes the room right? Like a collision with a Stargate changes the RoomSCD. Whereever you write to RoomSCD you just also write into another IComponentData the same value. In your RoomDisableRendererSys you then Filter for changes on the IComponentData. So RoomDisableRendererSys only runs for chunks where an Entity changed the room. If nothing changes rooms it doesnt run at all.

[rustic rain]

hmm

version filter

I see

curious

Does Entity Manager support it though?

Since I don't want to run it through loop, I can literally just do it in EntityManager

[solemn hollow]

Its a Filter for a query. So if i understood your question right ofc you can use it with Entitymanger

[rustic rain]

ngl, your idea of doubling room component is great

now it also solves a question of how to acess data

for reading only

[solemn hollow]

yep

[rustic rain]

hmmm

any idea whether I can simply cast

my shared component to normal?

    public struct StarNormal : IComponentData
    {
        public int systemID;
    }

    public struct StarShared : ISharedComponentData
    {
        public int systemID;
    }

kinda like this (StarNormal) new StarShared();

doesn't really let me

[coarse turtle]

you can use an static implicit cast operator or UnsafeUtility.As with ptrs
https://docs.microsoft.com/en-us/dotnet/csharp/language-reference/operators/user-defined-conversion-operators

[viral sonnet]

what a pain, I'm not able to remove the LinkedEntityGroup from a converted GO :/

usually removing works during Convert but for some reason for this case it doesn't

it's a prefab, if I put the prefab in a subscene LEG gets removed

[solemn hollow]

Huh. I never had the case that Subscene conversion did something diffrent to single conversion

LinkedEntityGroup is pretty much the last thing that happens in conversion right? are you maybe not removing it late enough?

[viral sonnet]

yeah, must be a order problem. hm, is there a post convert? I'm not aware of any

so where it's different is. prefab in subscene -> normal. same prefab referenced in spawner subscene GO -> LEG still there

[solemn hollow]

hmm. im guessing here. the diffrence could be that prefabs don't have an associated entity when conversion starts but get them later on when they are discovered via DeclareReferencedPrefabs. Maybe you somehow manage to delete LEG before those Prefabs are discovered?

[solemn hollow]

no there isnt. i would have needed that quite often myself already 😦

[viral sonnet]

I just found this, maybe it helps here

[solemn hollow]

I think LEG is done even after late conversion group. but im not sure. feels like i looked into it once but it was quite a while ago

static void FinishConvertIncremental(World conversionWorld, Conversion conversion)
        {
            using (s_UpdateConversionSystems.Auto())
            {
                conversionWorld.GetExistingSystem<GameObjectBeforeConversionGroup>().Update();
                conversionWorld.GetExistingSystem<GameObjectConversionGroup>().Update();
                conversionWorld.GetExistingSystem<GameObjectAfterConversionGroup>().Update();
            }

            using (s_GenerateLinkedEntityGroups.Auto())
                conversion.MappingSystem.GenerateLinkedEntityGroups();

#if !UNITY_DISABLE_MANAGED_COMPONENTS
            using (s_CreateCompanionGameObjects.Auto())
                conversion.MappingSystem.CreateCompanionGameObjects();
#endif

            conversionWorld.EntityManager.DestroyEntity(conversionWorld.EntityManager.UniversalQuery);
        }

[viral sonnet]

seems so, the test with GameObjectAfterConversionGroup didn't work. I just don't get it why it's working for a subscene GO. should be the same conversion process. I mean, not completely, as it's converted into a subscene

ah thanks, that confirms it

guess I have to unlink it then. major pain ... 😄

[solemn hollow]

why are they linked in the first place 🙂

[viral sonnet]

I'm lazy and easier to setup 😄

technically they are linked. I just don't want a 144B LinkedEntityGroup in my archetype

[solemn hollow]

Well you could attach both objects to one empty GO and then in lateconversion delete that GO

then LEG wont be added ? might work

[viral sonnet]

that doesn't sound too bad actually

thanks, I'll try

the reason I'm doing this is to keep my main archetype free from any physics and collider stuff. main archetype is big enough as it is. I need some CDFE calls because of this but it's not too bad. at least, it's worth it to keep the archetype small

pointer to LTW is set once for the main entity so I even have access to translation, rotation without losing too much

[solemn hollow]

but that does mean that your entity with LTW may never change chunk?

im a bit confused

[viral sonnet]

pointers are set every frame to avoid that problem.

hm, when I destroy the entity the linked entity group entities are also destroyed, right? how would I just destroy the container?

[solemn hollow]

i hoped you can destroy the Entity in the DstWorld in conversion before LinkedEntityGroup is set up

[viral sonnet]

oh, right. well let's see

conversion mapping system errors out sadly: The entity does not exist

why are they adding something after post convert? WHO DOES THAT???

[solemn hollow]

hmm maybe the problem is that the ConversionEntity Exists and the DstEntity not

try what happens if you delete the gameobject in conversion

[viral sonnet]

it was worth a try, this is getting too hacky for my taste. I need to take a step back and reevaluate. I must be able to remove the comp at one point. and if all fails, I'll remove it in runtime for the created prefab

thanks for the help, I'm off now to a birthday party 🙂

[solemn hollow]

yes i do it in runtime too atm for the same reason. At some point in the future i hope we'll get a ExportConversionGroup like tiny had! have fun at the party!

[viral sonnet]

ah, good to know. thanks! 🙂

[hot basin]

@rotund token in .50 samples (1.0+ collections) my code with IJobParallelForDefer works

so for now I think I need to switch to simple IJob waiting for 0.51

[rotund token]

Just avoid parallel native list until you upgrade

[hot basin]

but sadly I need it more than defered job 😄

[rotund token]

Be careful because it breaks in other situations as well

[solemn hollow]

do you really need a nativelist or couldnt you work with nativestream etc?

[hot basin]

but will it work with IJobParallelForDefer?

and can I add structs to the stream?

Never used stream

sorry

[rustic rain]

ok, it turns out that EntityManager add component through query is extremely fast and I can simply solve all my renendering through FixDirty

doing it every time I want, as long as it's not every frame xD

basically every time entity gets created/changes it's room

[solemn hollow]

i never used DeferJob sorry too 🙂
i dont know what you are trying to solve. No idea if NativeStream makes sense for your problem. You need to know the size of the stream beforehand too but it wouldnt be a huge problem if you oversized it.

[solemn hollow]

yes you can

look into the Tests for NativeStream. Gives you a basic idea on how to use them.

Tertle is probably the expert on NativeStream though given he has made a whole EventSystem based on it

I used it like 3 times.

[rustic rain]

hmm, I assume GameObjectConversionSystems are just like any other SystemBase exist in world?

I just want to create native array in it

and later access from other systems

[hot basin]

which repo?

[solemn hollow]

NativeStreamTests.cs in Collections

[hot basin]

it seems that I need to specify max capacity of the stream

[solemn hollow]

yes

[hot basin]

but I don't know it

so stream is wrong type for me

[rustic rain]

@hot basin do you have entities or not?

[hot basin]

I have

[solemn hollow]

cant you even guess an upper limit? what do you want to write into it?

[hot basin]

but i'm on 0.17 as I need 2021.2+

[rustic rain]

so, why can't you know upper limit?

[hot basin]

because it's bound to setup in editor

[solemn hollow]

but then you can infer from the setup how big it might get can u not?

[solemn hollow]

the ConversionSystems live in the conversion world. you should probably use blobassets to save that nativearray and later read it in the Gameworld

[rustic rain]

nnnah

I just figured I'd add another component

kek

prob not a better idea ngl xD

[hot basin]

I can't because I don't know how many and which entities would be in the runtime

I could multiply all the maxes

but why

[rotund token]

You don't need to know capacity, just width

Native stream can resize in parallel

[hot basin]

width?

[rotund token]

Width

You don't specify capacity with native stream

You specify foreach count

[solemn hollow]

tertle to the rescue!

[hot basin]

yeah I figure that out

[rotund token]

Power went out at 7:30 last night, went to bed early to get up early. Got up at 4:50, internet went down at 4:59 with 7am+ eta

So annoying, so here I am

[solemn hollow]

oof

[rotund token]

Anyway if I was you krajca I'd just ditch parallel native list and just do ijob there and keep deferred job

Parallel native list really slow anyway and when you upgrade less code to write

[solemn hollow]

Oh man its slow? joachim said multiple times that hes impressed with nativestreams performance

[rotund token]

It's because they can't know what entities to link until all stages are done. You can create additional entities during conversion

[rotund token]

Parallel Native list is slow

Streams are fine

[solemn hollow]

ah so misstyped?

[viral sonnet]

I was talking about the LinkedEntityGroup.

[rotund token]

So am i

[rotund token]

Yeah my bad

[viral sonnet]

Eh, should just have a stage after adding the linked entity group

[rotund token]

Any stage of conversion can call create additional entity. This needs to be added to linked entity buffer

Unity does not expect you to remove this because it breaks a lot of tooling

[hot basin]

so to read from stream I need to do the whole thing?

var reader = nativeStream.AsReader();
            for (int i = 0; i < nativeStream.ForEachCount; i++)
            {
                int count = reader.BeginForEachIndex(i);
                for (int j = 0; j < count; j++)
                {
                    var axis = reader.Read<Axis>();
                    Debug.Log($"{axis.input} | {axis.scale}");
                }
                
            }

[viral sonnet]

I'm already using CreateAdditonalEntity at conversion stage and that doesn't add a LinkedEntityGroup

[rotund token]

Now I do really wish this was a capacity of 0

[rotund token]

Let me check but I'm pretty sure it should =S

[viral sonnet]

I've only seen it for GOs that already have a hierarchy

yeah, checking myself now 🙂

could be wrong

yeah, my AI entity was never linked

[rotund token]

The reason you pass in a component to create addition entity is to link it

Are you passing in null?

[viral sonnet]

var aiEntity = conversionSystem.CreateAdditionalEntity(this); in Convert

[rotund token]

Hmm no idea just booting up to confirm

[viral sonnet]

I've done this a few times now and previously I even built the LinkedEntityGroup myself because I needed it

[hot basin]

ok I can convert it to array

[viral sonnet]

(other project)

[rotund token]

You seem to be right interesting

Now I need to figure out why this isn't working

Hmm because it's not coded to

You know what, that's annoying

[solemn hollow]

Can confirm. I dont get a LEG on createadditionalentity either

[rotund token]

Actually no looking at code it seems to imply it should

[solemn hollow]

This is super annoying together with the fact that you cant instanciate prefabs

[hot basin]

how to use deferred job with NativeStream?

[rotund token]

You don't

Defer job is only for native list

You don't need it for native stream

[hot basin]

it's heavy to convert stream to an array?

[rotund token]

Hmm a little

If you use native stream to write best to just use it to read

Ok wtf am I missing, I can't even create a linked entity group atm

[hot basin]

but I can only paralelize by width and it would not be same for every index

[rotund token]

Sorry I don't follow

[hot basin]

yeah, maybe I care too much, but ParallelFor with an array will batch work in same sizes so it will be evenly distributed

and if I will do ParallelFor with native stream I need to bach it by foreach index

[rotund token]

Generally you'd make the width chunk index or something like that

Using entityInQueryIndex is pretty poor for native stream

[solemn hollow]

what do you mean? at runtime?

[rotund token]

At conversion

[hot basin]

@rotund token is that proper reading? I didn't missed something yes?

[rotund token]

There are no linked entity groups in my subscene

[rotund token]

Ideally you read it in parallel

The main benefits of stream is lockless multi thread writing and ability to read in parallel without merging

But often you do need to read it in a single thread to use the data

[hot basin]

can I change data in native stream while reading it?

[rotund token]

What do you mean sorry? You want to write to it during reading?

[hot basin]

if I have list/array I can take element at index, change it and write it back

can I do something like that with native stream?

[rotund token]

You can only write to a native stream once

That said, if you wanted to get real dirty with pointers there nothing stopping you changing a value

[viral sonnet]

@hot basin I had very similar problems and developed this with tertle: https://github.com/enzi/Entities.Exposed

[hot basin]

how did you solve the problem?

[viral sonnet]

with a nativelist that has nearly the same speed as a nativestream for parallel writing

blocks are allocated and reused for writing until there are none left

[rotund token]

yeah krajca problem isn't speed atm

it's the fact that nativelist parallel writer is broken in 0.17

and can't be used with deferred job

hence krajca is looking at alternatives

[viral sonnet]

oh damn :/

[rotund token]

(i got internet finally woo)

[viral sonnet]

Read returns a ref, so yes

I'm not sure if that's really helpful though, do you iterate 2 or more times over the stream?

[hot basin]

in the end it will be in 100 or even 1000

[viral sonnet]

oh, it's persistent?

[hot basin]

sadly no

it's an AI for units

so I don't know how many or which type of units there will be

[rotund token]

well that's even easier then 😄

[hot basin]

so I dynamically gather all and calculate

[viral sonnet]

are you talking about elements in the stream?

[hot basin]

yes I do

[viral sonnet]

ah ok, I was asking how often you loop over the full stream because I was confused why you need to write back to the stream

[hot basin]

1 write, 1 calculation and 1 read an the end

[viral sonnet]

and the calculation needs the ref?

[hot basin]

yes or I would need to somehow connect stream element with an ID

a suppose I could make another stream

with ID + result

[viral sonnet]

if you read multiple times over the full stream I'd try to make a native array out of it. that way you can read faster and the worker threads are more balanced

[hot basin]

yeah it will be easier

[viral sonnet]

NativeStream has a quirk that sometimes worker threads do a lot more than others even when the write was totally balanced

never figured out why

[hot basin]

but I wanted to explore native stream here in the conversation as I never used them and they seems helpful

[viral sonnet]

NativeStreams are great 🙂 I like DreamingImLatios UnsafeParallelBlockList a bit more though

it's not tied to a for each count

just the 128 worker count

as with everything, both have their pros and cons

[hot basin]

how should I change native stream into an array between write job and calculation job?

[rotund token]

thats basically what my event system uses, just a fixed nativestream with 128 wide

if you are less than 128 though it does add a bit of overhead vs a nativestream

[viral sonnet]

NativeStream has a ToNativeArray method

[solemn hollow]

@hot basin what kind of AI are you going for?

[hot basin]

IAUA - infinity axis utility ai

[solemn hollow]

nice soon all of us will have their own implementation of it xD

Mabye you dont want to write back to the stream after the calculation? Pretty sure you could reject alot of scores already and only fill a small list with the highest ones

[hot basin]

I gather all the axis and then I'm doing calculations

[karmic basin]

UAI lover here ✋ Was working on mine too but halted DOTS for some time 😄

[hot basin]

all as from all ais

so I can't assume anything as I don't know relations while calculating

[solemn hollow]

Im not kidding. i bet there are atleast 10 dots implementations from ppl posting in the forums

okay i dont wanna go into details of UAI yet again but i wonder how you want to scale that up.

[hot basin]

what do you mean?

[viral sonnet]

unity physics is giving me nothing but trouble. no wonder some of you have written their own spatial query system

[hot basin]

I saw one form DreamingImLatios

[rotund token]

well dreamings custom physics comes from a different reason

of him fundamentally hating unity physics

[solemn hollow]

if i had the time i would too... physics still takes 5 ms even after reducing collisions to basically 0 (and yes safetycheck etc disabled)

[rotund token]

hvae you turned off physics integrity checks?

[solemn hollow]

yes

[rotund token]

hmm not sure

how many colliders?

[solemn hollow]

alot. but alot static

[viral sonnet]

250k! 😄 haha, then it starts the physics death spiral

[solemn hollow]

around 100 non static

[rotund token]

yeah i don't think you can have more than around 40k physics elements enzi

anyway just in case someone isn't aware what i mean by integrity checks

this nukes performance

[solemn hollow]

yes this brought physics down from 7 to 5 ms for me

[viral sonnet]

10k kinematic is 2ms for me

[solemn hollow]

wtf am i doing wrong then

well i need to get the framerate higher through other means first anyway

[rotund token]

i was simulating like 20k dynamics in less main thread time than 5ms im pretty sure

[solemn hollow]

one problem is that stepphysics runs multiple times.

atm game runs at 30 fps in editor so thats a problem...

hybrid stuff kills performance

[viral sonnet]

the dreaded physics death spiral

[solemn hollow]

yep

[viral sonnet]

really annoying, is there anything to turn off this behaviour?

[solemn hollow]

tertle suggested reducing the tickrate of physics

[rotund token]

welcome to fixed update life

you tell it to run 60 frames a second

it will run 60 frames a second 😄

[solemn hollow]

okay this is slightly off topic :
my artist was using 2D Animation Package to animate with inverse kinematics. now they run every frame and take 3.5ms... anyone dealt with this before? how to "bake" the Ik motion into the animation and not have it calculate at runtime

[rotund token]

at the very least in editor i usually drop tick rates to 30 for physics/netcode

because editor has a lot of overhead which can keep your fps low

[solemn hollow]

i feel like 0.50 introduced insane editor spikes

[rotund token]

i have noticed large chunks of editorupdates

not certain related to 0.50 or not though tbh

[solemn hollow]

well it would be about time

btw thx for the help with the EffectsSystem. It finally works after quite some trouble.

[viral sonnet]

is this normal for physics? so much idle time and only MT thread active

[solemn hollow]

looks like this for me too

i tried to schedule the heavy work before physics so it can fill in some gaps

[viral sonnet]

good idea, this gap sucks really hard

do you know what's going on there?

good thing I have system groups, now I can just move it in front of fixed update

[solemn hollow]

it seems like stepphysics is doing sth heavy on mainthread before it starts to schedule stuff. but i havent looked into it

[rotund token]

this is all editor overhead btw

[solemn hollow]

oh god

[rotund token]

                m_RemainingBlocks = 0;
                m_Safety = stream.m_Safety;
                CollectionHelper.SetStaticSafetyId(ref m_Safety, ref s_staticSafetyId.Data, "Unity.Collections.NativeStream.Reader");
#endif```

if you run a build and profile it

that fixed string stuff goes away

[viral sonnet]

oh, cool!

[solemn hollow]

hmm i really need to profile in build soon... I was changing to Subscene conversion and alot of stuff broke.. so i couldnt build for like a month now

and my artist is still stuck on some conversion topics

[viral sonnet]

every system that doesn't run with burst runs like shit ... 18 enemies, 2.59ms for the HybridAnimationSystem

[rotund token]

what are you doing in that

[viral sonnet]

eh 😄 setting animator values all the time. pretty bad actually

[rotund token]

that said, usually those get a decent speed up if you build with il2cpp

at least if you have any logic in there

be nice when we can finally get away from this editor mono

[viral sonnet]

you mean when mono gets replaced with .net core?

[rotund token]

yeah

[viral sonnet]

that'll be huge. it's taking a loooong time sadly

i'm not much of a ms fan but their work on c# is solid

[solemn hollow]

hmm i cant profile mine right now but this seems indeed slow.

nvm mine is probably equally slow. i just thought i write to 100 entities but i optimized it already

[solemn hollow]

are we talking 2022 or even longer?

[viral sonnet]

they have moved it so much. I don't know. 2022 or 2023 likely. 2023 is more likely I guess 🙂

[solemn hollow]

2025 it is

[viral sonnet]

very naive and unoptimized 😄

more like a prototype if anything

I need to figure out how to best reduce the writes

[solemn hollow]

i love the early out for dead entities comment

[viral sonnet]

haha 😄

[solemn hollow]

mine looks exactly the same. when i said optimized i meant the number of entities i write to. not the code 😄

[viral sonnet]

I have a habit of even commenting really obvious early outs

[solemn hollow]

hmm my animwriter system takes 0.04 ms

just tested with 3 entities. ill test with more

[hot basin]

@viral sonnet going back to stream questions, I can't convert stream to array if I don't .Complete() previous job

[viral sonnet]

you should be able to convert it in a job with a handle

you'd have to get the nativearray back somehow. single sized native array with a void* maybe. I think there are better solutions

[solemn hollow]

So with 18 entities i got 0.1 ms

[viral sonnet]

lol damn 😄

anyone has this problem in a system? protected override void OnCreate() { physicsWorldSystem = World.GetExistingSystem<BuildPhysicsWorld>(); mainCamera = UnityEngine.Camera.main; mainCameraTransform = mainCamera.transform; } mainCamera is null in a build

maybe it runs too soon?

[hot basin]

Do you have camera with proper tag?

[viral sonnet]

yep, runs fine in editor

[solemn hollow]

public partial class AnimationWriterSys : SystemBase
{
    
    protected override void OnUpdate()
    {
        Entities
            .WithAny<SLOD0,ZombieTag>()
            .WithNone<DestroyReq>()
            .ForEach((in HybridArtChild hybridArtChild,
                in AnimationStateOutput animationState) =>
            {
                var animator = EntityManager.GetComponentObject<HybridArtManaged>(hybridArtChild.childEntity).animator;
                if (animator == null) return;

                switch (animationState.abilityAnimationState)
                {
                    case AbilityAnimationState.frontSwing:
                        animator.SetBool("isAttacking", true);
                        break;
                    ...//some cases
                    case AbilityAnimationState.ignore:
                        animator.SetBool("isAttacking", false);
                        break;
                    default:
                        throw new ArgumentOutOfRangeException();
                }

                animator.SetBool("isEating", animationState.isEating);
                animator.SetBool("isInfecting", animationState.isInfecting);

                animator.SetInteger("abilityIndex", animationState.abilityAnimationIndex);

                if (animationState.isDying)
                {
                    animator.SetBool("isDying", true);
                }

                if (Vector3.SqrMagnitude(animationState.moveDirection) > 0.01f)
                {
                    animator.SetBool("isRunning", true);
                }
                else
                {
                    animator.SetBool("isRunning", false);
                }
            })
            .WithoutBurst().Run();
    }
}```

what is the diffrence that costs u so much performance?

[viral sonnet]

could be the triggers

[rotund token]

totally the getcomponents and maths

[viral sonnet]

also the local velocity transformation is expensive

[rotund token]

i really think in hybrid systems you should do nothing but write data

[viral sonnet]

true, the NHM lookup could also be slow without burst but I don't think it adds this much

hm, I realize now that it runs much more often but fails to get an animator

[solemn hollow]

i didnt expect the diffrence to be this grave. ❤️ burst

[rotund token]

i want to see the enzi 10x profile marker performance analysis of this job

because im curious

[solemn hollow]

😄. +1

did you guys already switch over to ISystems?

[viral sonnet]

it is indeed the NHM lookup. damn NativeContainers without burst are so slow. 0.043ms with 18 enemies. before I had 50k additional lookups

[rustic rain]

what kind of magic is this kek

my pepe algorithm is slowing game down during fps drops

            if (StartStop)
            {
                double curTime = Time.ElapsedTime;
                if (curTime - _prevTime >= _interval)
                {
                    _prevTime = curTime;
                    base.OnUpdate();
                }
            }

pepe algorithm

kek

[rotund token]

just remember though safety is extra slow in mono so you can't really tell until you run it in a build

[solemn hollow]

wasnt there a way to control updaterate of systemgroups? fixedratemanager or sth?

[rotund token]

yes

[solemn hollow]

but who can work like that 😄

[rotund token]

thats how fixed fixedupdatesystemworks

ratemanager i think it's now

[hot basin]

I end up using second stream to gather results

[viral sonnet]

my build had 2 errors. both are fixed and it's still crashing. meh

[rotund token]

🤣

this is why i try to make a build every week or so

because tracking down build crashes can be a pain if it's been a long time

[viral sonnet]

it's crashing so hard I fear it might be something unsafe related. I don't get why editor is unaffected

[rotund token]

what does the dmp show

[viral sonnet]

phew, at least log is pretty good. I know where it happens

and it's never what you think it is: EntityQueryImpl_GetSingleton_TisHandleCombatEvents_Options from all the things possible. it's a damn singleton component. I think that wants to access before the subscene is loaded. that it ends in "Attempt to access invalid address." is pretty bonkers though

[rotund token]

just a note, subscene loading is async

[viral sonnet]

yep, had my pains with it in a netcode project 😄

[rotund token]

yeah i wrote a whole management thing for it

that lets me load a bunch of subscenes into a world before the world starts running

so i don't have to a million RequireSingletonForUpdate in my jobs

[rustic rain]

what is the difference between those?

[viral sonnet]

ah cool, it's manageable with 1 subscene but when you get more than that. eh, it has to be managed

[rotund token]

from memory, FixedRateSimpleManager only updates once per frame

[rustic rain]

I'm looking for one that will smooth out ticks for target

[viral sonnet]

good doc on it: https://docs.unity.cn/Packages/com.unity.entities@0.50/api/Unity.Entities.RateUtils.VariableRateManager.-ctor.html

[rustic rain]

I was trying to find it

[misty wedge]

Is EntityCommandBuffer.SetSharedComponent not allowed in a burst function?

[rustic rain]

no

Shared comps are managed only

rn

[misty wedge]

Bummer

[viral sonnet]

@rotund token got my build up. 1.9ms-2.2ms for the hybrid animation instead of 2.5ms in editor (mostly useless NHM lookups 50k)

[rotund token]

oh you're doing 50k lookups outside of burst?

haha - gl with that

[viral sonnet]

yeah, like I said, native containers outside of burst are really slow

i'll add some tag HasAnimator to reduce this 🙂

[rotund token]

why don't you just attach the object to the entity

[viral sonnet]

too little experience with that. it doesn't change any archetype right?

[rotund token]

its 4 bytes from memory

so yes archetype will change

but all the animator ones will be grouped in the same chunk

but HasAnimator will also change the archetype

[viral sonnet]

true, I'll think about it some more. I don't want to make useless optimisations that don't make sense in practice. like, how many casters will actually have no animator?

code confused me how they save MBs. somewhere in an array and the index is saved. guess this will be the 4 bytes you mentioned

[solemn hollow]

why is having an extra archetype with no animator a performance concern?

[viral sonnet]

I don't want the data in the archetype, that is pretty much all my reasoning 🙂 I just got it back to a 51 chunk capacity

[radiant berry]

What package do you need for animating a character in dots 0.5?
Or do you just need to use the standard animator still?

[viral sonnet]

standard animator

[radiant berry]

i saw a crowd demo with dots not sure if i should post link but i assume they also just used the standard animator?

[viral sonnet]

there's https://github.com/Dreaming381/Kinemation-Skinning-Prototype - haven't tried it though

nah, there was an animation package once but it got dropped

at least dropped until 1.0 has released

[radiant berry]

yeah, i think i'll wait to experiment with the idea

[viral sonnet]

FixedString32Bytes is the smallest? 😦

another useless thing in my chunk

Is data like EntityGuid, SceneSection, SceneTag, EditorRenderData also in a build?

bummer that we can't actually see what's the archetype capacity. (other than calculating it ourself)

oh, a unity dev name dropped a profiling tool I didn't know: https://superluminal.eu/

[rotund token]

i have a ministring, size 16, 15 character, but burst 1.7 kind of broke it

i havent tested burst 1.8 if they fixed it

[viral sonnet]

I hope I can find a setting so physics doesn't actually collide with anything and I can only use it for spatial querying

[rotund token]

put it in a separate physics world?!

[viral sonnet]

wait what? how would that help?

ah, you mean a world where it doesn't actually steps?

[rotund token]

PhysicsWorldIndex is what i mean component wise

someone in here managed to get this working and ticking it manually you could run it without the simulation step

[north bay]

It's really nice, using it at work all the time
Havent tried profiling a Unity app with it yet tho'

[rotund token]

or just dont put it in the simulation@viral sonnet just use the bvh yourself

use the colliders etc but leave off PhysicsWorldIndex

and then just pass it to the bvh like unity physics does

their bvh is very re-usable

[viral sonnet]

cool thanks! what do you mean "leave off index"?

[rotund token]

well if you dont add teh shared component

it wont be put in physics simulation

but you can still use the physics colliders

(purely a theoretical idea, i haven't tried to see how well it works)

[viral sonnet]

I'll play around with it. removing PhysicsWorldIndex certainly stops the physics step

and I'm a dummy. the mentioned NHM lookup is actually a Dictionary, I never figured out to use NHM to save the Animator class

saving it in the archetype sounds more attractive by the minute

hehe, the system vanished. had some free space on the physics entity. needed some data from it anyway

[misty wedge]

Any idea where this is coming from?

Too many debug lines added during the frame (max 10240 can be added from jobs). Some lines have been dropped.
To avoid this please pass "-debug-line-buffer-size <size>" on the command line or add to boot config

There is nothing being printed to console 🤷

[viral sonnet]

have you disabled one of the messages?

[misty wedge]

Nope

[viral sonnet]

hm, never had this error

somewhere it will log though, such an error doesn't come out of nowhere

I have some acceptable numbers now from physics. 14ms with 250k colliders. no dirty hacks. I set the physics step to no physics and removed the rigid body from the entities. unity says they will be now static but they can still change position and spatial querying works fine. (with AI pathing)

there's some drifting at first because for some reason all enemies spawn on the same position. - bug aside, pretty acceptable for what it is. - I tried now with physics on for the steps and it doesn't make a difference. removing the physics body is enough

[misty wedge]

I do basically the same thing, just use the physics system for querying and use character controllers for movement

Granted the game is 2d and has no complex physics anyways 🤷

[solemn hollow]

hmm interesting. id like to actually use collisions though. i would only need active colliders around the player. maybe i can disable the colliders for the rest of the simulation.

[rustic rain]

Hmm

is there an option to have random without individual components?

what can I tie it to?

I tried to tie it to LocalToWorld position value

but that ends up with 0

sometimes

leading to exceptions

[rotund token]

im not sure what you mean by this

[rustic rain]

                    float3 pos = ltw.Position;
                    var rand = new Random();
                    rand.InitState((uint)pos.x);

                    float2 deltaPos = rand.NextFloat2(MINWanderRadius, MAXWanderRadius);

this fails sometimes

[rotund token]

well anytime pos is 0

[rustic rain]

I need a reliable random

yep

[rotund token]

its also not very random

everything at the same posx

will get the same value

[rustic rain]

it's never will be in same posx tho

kek

[rotund token]

why cant it be at same pos x?

[rustic rain]

but anyway

[rotund token]

(1,1), (1, 123414)

[rustic rain]

It just won't do

I need better option

[rotund token]

i usually just generate random outside job using unitys random to give initial seed

[rustic rain]

I'v seen option with adding Random to entity as component

[rotund token]

then add the entityInQueryIndex

to get a per entity state

[rustic rain]

I kind of look for a way

rn to also make a random that will replicate itself

under same circumstances

To combat save scumming a bit xD

eh, seems like adding it as component is the only way

[minor sapphire]

A little bit extra knowledge... A few years ago I was testing the new random struct and found that skmilar (close in value) seeds produced similar results. Don't know if that has changed.

[rotund token]

think it's just an xorshift generator

fast over random

[rustic rain]

hmm

is there a way

during conversion

to query through interface?

Basically I want to be able to add to certain Components declarations my interface

and then during conversion in case any entity has component with that interface

add another component

meanwhile I don't want to add interface to authoring monos

only to entity comp

[rustic rain]

hmm

So can I somehow create IComponentData type out of ComponentType?

by myself

[rotund token]

componenttype cant be an interface

[rustic rain]

not component type

but Component itself

AIList[0].GetType().GetInterfaces()

I think I'm unto smth

kek

it's array of componenttypes

[rotund token]

you can gettype from ComponentType

[rustic rain]

yeaaah

[rotund token]

var type = TypeManager.GetType(componentType.TypeIndex);```

something like that

actually theres a shortcut componentType.GetManagedType()

i guess

which just does the same thing under the hood

[rustic rain]

            foreach (var type in AIList)
            {
                foreach (var interfaces in type.GetManagedType().GetInterfaces())
                {
                    if (interfaces == typeof(IRandomComponent))
                    {
                    }
                }
            }

so is that it?

[rotund token]

i think that'd work

[rustic rain]

can't figure out why it won't show up in inspector

hmm

oh wait

me dum d um

Is it even possible to do such thing?

[rotund token]

unity hasn't marked ComponentType has serializable

it won't show in inspector

[rustic rain]

I know

[rotund token]

it woul djust be an int though (+ access mode)

you could write your own component type with a drop down easy enough

just make a custom inspector with

TypeManager.AllTypes

THAT SAID type index is not stable

if you make a build your component type index will be different to editor

[rustic rain]

hm

[rotund token]

you can use StableTypeHash which is stable

UNLESS you ever make any change

add/remove a field

reorder fields

change namespace

change name

etc

[rustic rain]

I'd rather find another way

kek

[solemn hollow]

i went the roundabout way and wrapping the type into a scriptable object

[rustic rain]

I simply need it for a system that will automatically add Random component to AI agents that require it

[rotund token]

entity query
any {all components that need random}
none {random}

[rustic rain]

well, that's the thing

[solemn hollow]

but he wants to do it with an interface

[rustic rain]

I want to do it through interface

not hardcoding

because I want to have oepn doors for modders, that would add new components that might require it

[rotund token]

well what i say to that

stop writing oop code

😄

[solemn hollow]

but you wouldnt need a random component per component. couldnt those modders add their own RadnomAuthoringComp if they wanted to

[rotund token]

use write groups

[rustic rain]

it's not per component, it's 1

ooooh

write groups

yeah

[rotund token]

if you care about modding everything you do should be setup as a writegroup

[rustic rain]

yes

[solemn hollow]

still writegroup doesnt solve his randomComponent problem does it?

[rustic rain]

it kind of does

I just need to do 1 query

question is, whether I can do it in conversion system

basically, once all other components are already added

[upper tiger]

why does this not set the localToParent?

commandBuffer.AddComponent<LocalToParent>(entityInQueryIndex, newEnt);
commandBuffer.SetComponent<LocalToParent>(entityInQueryIndex, newEnt, new LocalToParent
                        {
                            Value = localToParent
                        });

the entities in the editor havent had their LocalToParent changed at all

[rustic rain]

is it possible to get ComponentType list from write group?

hm

feels so bad ngl

I can't figure out a way to do it

[solemn hollow]

the problem you are having is that the translation component is probably set to 0,0,0

So as soon as the LocalToWorld System runs it resets your LocalToWorld to 0,0,0

if it is a static entity you can remove translation from it. If it needs translation you need to set translation instead of LocalToWorld (or set both)

ups sry i didnt see its LocalToParent. Still the same applies. Translation writes to LocalToParent

[upper tiger]

thanks

yeah seems I can set translation but not localtoparent

the localtoparent updates with translation changes automatically

[solemn hollow]

you can. its just overwritten if you dont set translation. if you need the correct position in the first frame you probably should set both

[upper tiger]

ahh gotcha

I assumed if you set one or the other, the other would be automatically calculated

but only works one way

[solemn hollow]

yes. its advised to only read from LocalToParent / LocalToWorld and write to translation/Rotation/Scale

ofc for performance reasons you could want to write to LocalToWorld in some cases but then you need to take extra care with how you use the other components

[rotund token]

an example of this is the boids sample

where they have a custom component overriding the writegroup of the localtoworld

to take ownership of writing to it

[rancid gust]

Could someone explain me why a "burst compiled" job would perform better than without? I understand that it uses LLVM under the hood, but I do not understand why unity's JIT couldn't achieve the same, and even less why burst has to be explicitly added instead of being an implementation detail

[rotund token]

because burst has restrictions

i.e. no managed data

because of this it can make assumptions about your code

[rancid gust]

And it could be handled by the compiler

[rotund token]

and perform optimizations on those assumptions

apart from that, yes llvm and the ability to vectorize parts or all your code

[rancid gust]

And unity's jit/aot technologies cannot auto vectorize?

[rotund token]

i should point out, unitys current mono jit is very old/slow and things like il2cpp or .net 6 can compile much better code getting closer to burst performance

maybe only 20-30% slower for general scalar code

[rancid gust]

I see, though it is still weird that the annotation is mandatory instead of the compiler being an implementation detail. Or at least only present for compile time errors to ensure efficient compilation

[rotund token]

the reason the attribute is required is mostly just for editor speed

its much faster to just lookup attributes than to check every piece of code it verify its burst compilable

[rancid gust]

fair enough

[rotund token]

you dont really want to tab back into editor and wait 30 seconds for burst to figure out what it can compile

[rancid gust]

And isnt there any overhead going from managed code (jit) to foreign functions? It is at least the case on the JVM, in which case I'd expect small burst jobs to perform poorly

[rotund token]

there is but its much smaller than the general speedup you get

however this is why its recommended to not have simple function pointers for example

int Add(int a, int b) => a + b```
As the overhead would be more than any speedup

[rancid gust]

Alright, thanks for the explanations :)

[rustic rain]

wait what

burstcompile works in such way?

[rustic rain]

Oh man, feels so good

2k potential targets, 1k lookers
All processed in less than 2 ms

[hot basin]

do you use any spatial sorting?

[left sparrow]

Hello
Please tell me the sources where i can read about frustum culling and occlusion culling for dots ecs?

[rustic rain]

nah I just literally collect entities in NativeMultiHashMap

and then go through it

[hot basin]

for every looker?

[rustic rain]

looking for smallest value

yep

Burst magic

🙂

[hot basin]

i wonder if sorting them into spatial buckets wouldn't be much faster

still NativeMultiHashMap but with bucket id as key etc

[rustic rain]

it was stress test tho

in real conditions

It'd be way more gentle

smth like 30 potential targets tops

[hot basin]

don't get me wrong, it's insane speed already 😄

I just wonder if such techniques have a place in DOTS

[rustic rain]

idk about sorting

but overall dots is perfect for my case

basically a huge sim

of thousands of agents

[hot basin]

can I get Entity in IJobEntityBatch?

var entities = batchInChunk.GetNativeArray(new EntityTypeHandle()); is this a proper way?

[rustic rain]

no

Get entity type handle

like component type handle

in systembase

and then assign it to field

[hot basin]

so var entities = batchInChunk.GetNativeArray(entityTypeHandle); and entityTypeHandle = GetEntityTypeHandle(), in system?

[rustic rain]

yeah

[hot basin]

thanks!

[rustic rain]

entities array is literally same as component arrays

ngl tho, so far I realised

that using jobs is meh compared to ForEach loop

xD

the only good side I noticed is that you can access shared component id

once I got rid of such need

I moved it all to foreach

meanwhile foreach loop always looks good on system debugger, while counting query count

[valid hazel]

Is the hybrid renderer supported on the switch platform?

[rustic rain]

Is dots even supported on switch?

[calm edge]

afaik hybrid rendering requires HDRP which isn't available on switch

[solemn hollow]

no URP would be fine too. https://docs.unity3d.com/Packages/com.unity.rendering.hybrid@0.50/manual/requirements-and-compatibility.html

@valid hazel its says right there that it is not validated for mobile and consoles.

It can work but you will be alone with your troubles.

[pine plaza]

does any know how to implement state machine in dots?

[solemn hollow]

@pine plaza there are multiple ways. how you wanna do it depends on the usecase u need it for. IMO there is no 1 right solution to fit all.
to name a few:

• Add and Remove StateComponents from entity. Might be good if you have very rare state changes -> structural changes
• enabled and disabled component for each state on the entity -> no structural changes needed but polling all states every frame
• enum in a component. switch in a system
• polymorphic components from PhilSA. See the example in the readme

[misty wedge]

Any debugging tips? Everything works fine except if I spawn a lot of entities, then a lot of things get spawned in an ECB, and my game hangs and eats massive amounts of memory. No idea what is going on

The last point I can see in Rider where it hangs is EntityCommandBufferSystem.FlushPendingBuffers, specifically m_ProducerHandle.Complete()

I have no idea what dependency this is waiting on though

[solemn hollow]

What do you mean by alot of things get spawned in an ECB. do you refer to the entities you instantiate in ECB?

[misty wedge]

Yes

[solemn hollow]

so how many entities are we talking about?

[misty wedge]

There are 9000 things in the list that is run through, and per spawned entity about 10 values get written to the ECB

[solemn hollow]

I dont know how Instanciating prefabs is done in ecb. maybe multiple commands are set per Instantiate to remap entities or sth? Or do you do some setup logic for each Instanciated prefab. like setting position

[misty wedge]

I use ECB.Instantiate to spawn an entity prefab, and then overwrite some component values

[solemn hollow]

well thats where those commands come from then

[misty wedge]

Yeah but I don't feel like even 90000 commands would take up 8 gigabytes of memory

When it was 2 GB before

[solemn hollow]

those commands should not be whats taking the memory. its the spawned entities

[misty wedge]

I don't think 9000 entities would take up 8 gb

[solemn hollow]

yes they wouldnt.

[pliant pike]

sometimes its best not to use the ECB to instantiate entities

[solemn hollow]

its some editor stuff i guess. i have memory problems too

did you profile it in build?

[misty wedge]

Not yet

[misty wedge]

I assume the alternative is write it to buffers and then use the batch functions?

[pliant pike]

it might be faster to just use EntityManager.Instantiate

ecb in some cases just creates more work than is needed

[misty wedge]

Would probably be tricky since I don't even know what type of entity is being spawned ahead of time

[solemn hollow]

doesnt it literally call the entitymanager functions? 0.o

[misty wedge]

ECB doesn't support batch functions

I think that's what they means

[pliant pike]

yeah but your using I presume a job to queue up commands and then play them back on the main thread

[solemn hollow]

yah batching + setup entities is a pain anyways

[misty wedge]

Yeah

[pliant pike]

that takes way time than just using the entmanager.instantiate

[misty wedge]

In batch mode

ECB just calls EntityManager functions

[pliant pike]

I can create thousands almost instantly just by using entmanager

[misty wedge]

I'm not sure how I would do that though. The archetypes aren't even the same for all spawned entities

[solemn hollow]

are you using batch commands or one by one?

[pliant pike]

you can just use ordinary entityquerys surely

entmanager.instantiate can accept a nativearray of entity's as an argument

[misty wedge]

Yes, that's batch mode.

[solemn hollow]

How do you set up individual entity positions then?

[pliant pike]

yeah it took me a while to find that out

I just do a simple for loop after would

on the original nativearray of entitys

[solemn hollow]

i see. but honestly why spawn 1000 entities in one frame 🙂

[misty wedge]

Yeah I think the smartest thing is just to limit it to 100 or whatever per frame

I'd still like to actually debug it and see what is going on that takes so long and uses so much memory

[pliant pike]

because its really quick and fast, I'm pretty sure I can do close to 100,000 in under a millisecond

[solemn hollow]

yes you can with batchcommands. but is that the best design you can go for?

[misty wedge]

You will still run into problems naively modifying component data, since it will cause a lot of memory copying

[solemn hollow]

sure if you have a spacebattle with 100000k spaceships fireing i see the point.

[pliant pike]

really?

[misty wedge]

I would assume so? Adding a component will change the archetype of the entity

[solemn hollow]

no you would have the components already on the prefabs and only write to them

[misty wedge]

Ah I thought you meant spawning "empty" entities

[pliant pike]

I don't generally do much adding of components in things like that, I just pre-create the archetype and use that and set its components data

[misty wedge]

Isn't there a batch function to set component data for a set of entities?

[pliant pike]

maybe I havent used it yet

[solemn hollow]

whats the point though. to create an array of data for the batching you need to loop over sth anyways. can do that once later too

[misty wedge]

I can write the data from multiple threads, then a single call to EntityManager can set it

[pliant pike]

yeah unity's method would likely be more efficient than anything we can do manually

[misty wedge]

I guess Ill try just writing the entity's to spawn to a list and then actually create them on the main thread in batches, see how that goes

[solemn hollow]

btw ecb has a batch command

just checked

[pliant pike]

yeah its still just a delayed playback system though

personally I wouldn't use ecb unless I absolutely have to

I'm not an expert though

[solemn hollow]

thats the whole point of command buffers. i dont use Entitymanager if i dont absolutely have to lol

[pliant pike]

the ecb is using the entitymanager though

[solemn hollow]

yes thats the point. its just at another point in time.

you reduce syncpoints that way

[hot basin]

and what do you use? also EntutyManager

[solemn hollow]

Sure if you manage everything correctly and manually set all your systems using entitymanager together at syncpoints then its fine. but sooooo much more work

[hot basin]

don't understand in which aspect ECB would broke it

[pliant pike]

yeah I guess if you want to schedule your sync points better then its good for that

[solemn hollow]

and u wont get much if any performance

[misty wedge]

This goes back to my point of being confused what is actually crashing the game

[solemn hollow]

its probably tooling lol

[misty wedge]

What do you mean?

[solemn hollow]

All the debug stuff for entities might have a big memory consumption. i never checked but my editor is eating tons of memory

[misty wedge]

Hmm, guess I'll test in a build

[solemn hollow]

in profiler it shows my entities is around 5mb or sth

but i dont know if that includes hybrid components not stored in chunks

[misty wedge]

Very annoying though that it seems to be impossible to figure out what is currently running

[hot basin]

welcome to multithreading 😄

[misty wedge]

It's more the issue of the native code

I would need to build and debug the C++ build

[hot basin]

but to be fair, "running currently" is what your main thread is doing

[misty wedge]

Well yeah, but that's stalling waiting for other threads

So not very useful

[hot basin]

not always

only if you .Complete() something

[misty wedge]

It is hanging on Complete

It's a dependency to the EntityCommandBufferSystem

[remote crater]

Yo, I have a gameobject with lots of child objects... How do I weld em all together so its a single entity?

[solemn hollow]

you do it in conversion. there is no magic to do that for you

for static hierarchies you might wanna use the staticoptimizeentity component

[remote crater]

Oh it welds it in conversion? Cool

[solemn hollow]

no you have to weld it in conversion.

[remote crater]

Ok, is there a proper term for welding so I can googlify it?

Thanks much

Its okay, seems like its working now

Here, let me show you, its funny!

[solemn hollow]

If you have childgameobjects which should add components to the resulting parententity you need to do this in your authoring components. id say you shouldnt build your authoring like that in the first place.

[remote crater]

Straight this is funny, I had an asset of Elon Musk's car I had in orbit of Mars way before the politics stuff happened: https://youtu.be/vytdbGtTKb4

Funny for me, but I'm weird

[solemn hollow]

not sure how this has to do with your hierarchy problem though 😄

[remote crater]

Well Elon's car was all sub childrened out and wasn't colliding

It had a complaint it wasn't childrening

Turns out I was doing the wrong inject method, which you led me to by talking about the convert process

I am reducing scope of my mmo to bare minimum

Which I thought would be SOL instead of all the star systems in the Milky Way

Now I realized a Mars base done bad would last a long time... So I am just starting with one random junk yard scrap system

Which lends itself to creativity: Just throw stuff out there, finally having fun

[white island]

VICTORY. I made the AStar system from value types only, it can cross between "worlds" (parts of the same graph that are technically in the same space but on different planes- useful for crossing through interiors vs exteriors), the whole thing can be serialized and deserialized on a "One File Per Node" basis (simplifying the serialization and allowing git collaboration without conflicts on the serialized graph), and on top of that, it can be burst compiled!

The last step is to make a KD tree for much, much faster nearest neighbor searching. Which I don't know how to make yet. But i'll figure it out.

[solemn hollow]

sounds great! wish i had bursted pathfinding :S

[white island]

this isn't the unity navmesh though, this is a hand-placed node graph for more granular pathfinding for off-screen NPC movement

[solemn hollow]

also on my wishlist xD

[rotund token]

You also need an attribute on outer class but yes, you can shortcut function pointers

[rustic rain]

huh

in 0.17

you'd need to do some weird

generic

stuff

is that in the past?

[rotund token]

For a function pointer? Well it's not related to entities it's a burst feature

But MonoPInvokeCallbackAttribute is still required for il2cpp builds

[rustic rain]

hm
That function didn't seem to be related to entities either

[rustic rain]

Did I understand correctly, that this attribute can be applied to any (burstable) method?

[rotund token]

yes (well it needs to be static)

    public static class FunctionTest
    {
        [BurstCompile]
        public static int Add(int a, int b)
        {
            return a + b;
        }
    }```

you'll want
[MonoPInvokeCallback(typeof(Func<int, int, int>))]
for the sake of il2cpp on the method

but yeah this works fine

i want to say this was allowed since burst ~1.5

it's been quite a while anyway

[rustic rain]

huh

that's very interesting

I wonder if I can try and attach burst to 2019 Unity

[rotund token]

pretty sure burst up until recently supports 2019

[white island]

that's a new one.

[remote crater]

Nonsensical!

At least the error has a whimsical sense of humor.

How is this triggering a write? I see only reads: DynamicBuffer<TriggerIgnoreBufferElement> beB = World.DefaultGameObjectInjectionWorld.EntityManager.GetBuffer<TriggerIgnoreBufferElement>(entityB);

[remote crater]

That's poor error formatting, I'll return to text only.

[solemn hollow]

thats a write right there

[white island]

ok so I have no idea what this neabs

I'm just adding a struct to a list

here's the stack trace and relevant code

 [System.Serializable]
    public struct KDPoint
    {
        public byte d;
        public float3 point; //theoretically this is redundant
        //-1 is no child
        public int leftNode; //left node index
        public int rightNode; //right node index
    }

 public struct KDTree //specifically for 3 dimensions
    {

public NativeList<KDPoint> nodes;

        public KDTree(int defaultLength, Allocator allocator = Allocator.Persistent)
        {
            nodes = new NativeList<KDPoint>(defaultLength, allocator);
            Debug.Log("KD tree allocated");
        }

public void addPoint(float3 p)
        {
KDPoint newPoint = new KDPoint
                    {
                        point = p,
                        d = newD,
                        leftNode = -1,
                        rightNode = -1
                    };

                    nodes.Add(newPoint); //HERE

[rotund token]

the error basically means you're trying to resize to a negative value

the question is what in your code is doing this

my best guess is you're calling addPoint on a KidTree that hasn't been instantiated via your constructor

so it's invalid memory

[white island]

it's been constructed, since the debug.log message appears, the stack trace follows a constructed instance, and unity panics that I haven't deallocated the list, since I don;t get the chance to before it goes boom

[rotund token]

well your log only implies that a KdTree has been instantiated, not necessarily the one you're referencing
without seeing the kdtree file we can't really see what is setup wrong

[white island]

as the stack trace implies, I am creating it through the factory at the end of the file there

[rotund token]

hmmmmmmmmmmmmm

[white island]

and the debug points . length shows me 2, as expected

so im not instantiating it with 0 length or something

[white island]

the error says it's on line 72 by the way

[rotund token]

yeah i cant see anything obviously wrong

i'm going to give it a run at some point when i have free time

and see if i can repo it

[white island]

huh. changing Add to AddNoResize (because I already instantiated it with the correct size) causes this error.

what does the requested length mean?

[rotund token]

what is going on 😫

but anyway

what htis error is saying

you set capacity to 2

but it's trying to add a 3rd element

so it's definitely not been setup to the correct size for the logic

[white island]

yeah hm.

also it shouldn't be adding a 3rd element. i need to do some testing

Okay, it's instantiated now for capacity 4, adds two elements, and then goes pppbtht

[misty wedge]

What does this option do exactly?

Reuse worker threads?

[rotund token]

if you untick this

you will only use mainthread

[misty wedge]

It will run all jobs on the mainthread?

[rotund token]

all jobs will run on mainthread, there will be no parallel work

[misty wedge]

Awesome, that's exactly what I need

[rotund token]

useful for debugging and potentially dedicated servers

[misty wedge]

Also why dedicated servers? For determinism?

[white island]

and I initialize my native list with a capacity of defaultLength + 1, which is 3, but then it says my capacity is 4? somthing is fishy...

[rotund token]

there's overhead for running code in parallel

if you're running a bunch of instances on the same machine it can be more efficient to just run each instance on a single thread

rather than running each instance over multiple threads interfering with each other

[misty wedge]

👍

Also any other tips on debugging? Everything on my project works fine and then it will randomly stall and commit 30+gb of memory, then crash

It's locking on a dependency, but if I interrupt with rider I can't find what the issue is

I assume it's an infinite loop somewhere, I don't think anything else could take that much memory

[white island]

and instead using a nativearray creates an infinite loop. What.

[rotund token]

yeah see i thought an infinite loop could be the problem

but i couldn't figure out how it would go to a negative length unless you added 2.1billion elements

which i thought would cause issues earlier

[white island]

i uh hope im not adding 2.1 billion elements

also I have no idea how there could be an infinite loop

add is only called twice

[white island]

The stack trace says it's in the factory function, but for the record I am calling it like this:

print("Building KD tree");
            IOD.NPCGranularNavigationSystem.KDTree.KDTree kd = IOD.NPCGranularNavigationSystem.KDTree.KDTreeFactory.buildTree(
                (from p in nodeList.ToArray() where p.worldIndex == 0 select p.position).ToArray() //??? error here???
            );

            print(JsonUtility.ToJson(kd, true));
            kd.Dispose();

the LINQ statement produces a length of 2 which is expected

this is in a different file

[rotund token]

@white island

you have a loop

for (int i = 0; i < nodes.Length; i++)

in this loop you add a point

nodes.Add(newPoint);

increasing nodes.length

[white island]

... youre right.

as it keeps expanding

it keeps going

got it

it should be capacity instead

and i gotta be sure to actually return from the function, whoop[s

ok. that did it. Thanks. Lmfao

sometimes I'm blind to my own code

[viral sonnet]

does collisionWorld.OverlapSphere actually have a cap? like, does the NativeList keep growing over the start capacity?

[rotund token]

dont think it's capped

you can cap it if you want

or custom filter it

and all sorts of things

(physics is very flexible for this)

[viral sonnet]

I hate code that ends in interfaces ... lol - yeah, how can I cap it?

[rotund token]

OverlapSphereCustom

and implement your own ICollector<DistanceHit>

the default overlap sphere just creates an all hits collector from your list

            where T : struct, ICollidable
        {
            var collector = new AllHitsCollector<DistanceHit>(radius, ref outHits);
            return target.OverlapSphereCustom(position, radius, ref collector, filter, queryInteraction);
        }

physics provides a few defaults for you

but you can implement any with custom rules you want

[viral sonnet]

ah, so I early out in the AddHit when the cap is reached?

[rotund token]

pretty much

[viral sonnet]

ok cool, thanks a lot

weird they have so many but none for capped

[rotund token]

🤷‍♂️

it's a pretty obscure use case

i've never considered a need for something like that

[viral sonnet]

pretty normal for RPG skills. hit N closest or smth

[rotund token]

yeah that seems like a reasonable use case

i believe from memory it does add them in distance order

[viral sonnet]

do you know what the return value of public bool AddHit(T hit) is?

            {
                return false;
            }``` only found this false in QueryInteractionCollector

[rotund token]

based off that code alone seems like whether it should fire physics events 🤔

but honestly i dont know havent looked

[viral sonnet]

well, at least I know where to start. it's late and I'm not getting some of these implementations. like ClosestHitCollector is just overwriting m_ClosestHit.

[viral sonnet]

ah, comment above // Called when the query hits something // Return true to accept the hit, or false to ignore it

                    {
                        hadHit |= leafProcessor.ColliderCastLeaf(input, hitData[i], ref collector);
                        if (collector.EarlyOutOnFirstHit && hadHit)
                        {
                            return true;
                        }
                    }``` I found some more answers but the ClosestHitCollector is still a mystery to me how it actually works with the code and no early out

good thing is that I can control breaking the loop after the cap has been hit with setting EarlyOutOnFirstHit

[rotund token]

oh actually i think i was backwards

closest hit is last

which is what the ClosestHitCollector relies on

[viral sonnet]

now it makes sense 😄

and my cap collector wouldn't work. eh, I'll think of something else. no biggie

i need to debug this to confirm. so the list I get back has the farthest first

[rotund token]

i believe so

i think i've used this before

so if you only care about x closest hits

just take it from the end

[viral sonnet]

yep, easy enough

[rotund token]

but yeah you'll want to check this

and maybe add in an assert just to confirm in case unity ever changes this behaviour

[viral sonnet]

initially I wanted to cap the memory allocations. but with the nearest last, this doesn't really work out

[rotund token]

i haven't used it before

but you could try maybe UnsafeRingQueue

but i have a feeling this doesn't actually loop 😦

which i thought it did

oh well

[viral sonnet]

interesting concept though

[rotund token]

but yeah basically just use a native array and index % length

you also get the count from memory on return

yeah NumHits

would always replace the furthest away

[viral sonnet]

what do you mean with this? NumHits is only set internally in the collector. I don't get the hitCount from the for loop

eh, something for tomorrow 🙂 Have a good night all o/

[rotund token]

isn't it written to the collector which you pass in

and it's passed by ref so you should be able to read it after?

[half jay]

Hello, what can cause such a load in this job?

[hot basin]

I'm guessing high number of children in your hierarchy

[half jay]

every entity has no more than 6 - 7 children

[hot basin]

and as I can see by colours, you have burst turned off?

[half jay]

oops

[ionic sierra]

Hello. Anyone know what kind of performance difference to expect out of a 2D game in Dots. I remember the early demos, zombie games, and Dots cameras, boids, etc, but since they've changed to a more hybrid approach recently , what are the current performance gains and are there any current examples?

[rustic rain]

Depends.
Biggest con of ECS is that it takes a lot of perfomance price for scheduling stuff
But with that price you get an open gate to jobified and threaded jobs, which show themselves on scale