archived-dots

[errant hawk]

Granted, you dont need to control every system, just the ones that you can manage and are order-dependent

just saying a system tree that shows you the order and groups of systems would be a lot easier to visualize the system execution order

[rustic rain]

but it's already exist

literally

systems

folder

Window->DOTS-Systems

[errant hawk]

yeah but you either have to be in play mode to see your systems

or edit mode to see editor systems

I am specifically talking about just a system tree to visualize execution order outside of play mode.

[rustic rain]

eh, don't see how PlayMode is a trouble xD

especially considering there's an actual difference between play and non-play mode with systems

[drowsy pagoda]

Is there an equivalent to System.Guid.NewGuid() that is safe in background jobs?

[errant hawk]

It's already thread-safe

[drowsy pagoda]

oops, I forgot to add that it needs to be burst compatible.

[errant hawk]

I managed to get it to burst compile by including a copy of the bcrypt.dll (which is in System32) in the Assets/Resources/Scripts (or wherever your scripts are) folder

this is what burst seems to have compiled the Guid.NewGuid() to

[drowsy pagoda]

Wow really just slap new FixedString64Bytes(System.Guid.NewGuid().ToString()); and no problems? I remember using ToString method inside any burstable, schedule (single or parallel) threw fits about it.

[errant hawk]

are you doing that outside the job? It doesn't make much sense to call NewGuid and then ToString() right after in a job

Guid is a struct, so it is burst/job friendly

no need to convert it to a fixed string

[drowsy pagoda]

I have a condition inside a job that will need new guids. Currently implementing a pre-generated guid pool and using a nativearray<fixedstring> type deal.

[errant hawk]

well, you can do what I suggested as it seems to work fine

for whatever reason unity fails to load the bcrypt.dll

so just including it in your project forces unity to include it

[drowsy pagoda]

🤔 alright, I'll check it out. Thanks.

[errant hawk]

yeah np

[errant hawk]

I cant seem to find the mass field on the PhysicsMass component. Is there an extension method or something Im missing? I set the mass of an entity to 1000, but it can be adjusted and I would like to reflect the changes correctly

[rotund token]

InverseMass = math.rcp(mass),
InverseInertia = math.rcp(massProperties.MassDistribution.InertiaTensor * mass),

mass is the value from the authoring

it's stored inverted

on PhysicsMass

[opal lynx]

I have a very simple
public partial class DestroySystem : SystemBase

it's OnCreate works that's fine
I checked with
Debug.Log("DestroySystem OnCreate");

but it's OnUpdate never runs
how can I make it works??

... the entire codes are

public partial class DestroySystem : SystemBase
{

    protected override void OnCreate()
    {
        Debug.Log("DestroySystem OnCreate");
    }

    protected override void OnUpdate()
    {
        Debug.Log("DestroySystem OnUpdate");

        Entities
            .WithoutBurst()
            .WithStructuralChanges()
            .ForEach((ref Entity entity, in DestroyData destroyData) =>
        {
            EntityManager.DestroyEntity(entity);
        }).Run();

    }
}

[rotund token]

create an entity with DestroyData on it?

if a system has a query, it won't update unless 1 of the queries has entities

[opal lynx]

@rotund token Thank you for your reply

but it doesn't even update when an entity has DestroyData

and also like you said if I commentize the foreach lines
then OnUpdates works fine after hit play

[opal lynx]

so..

adding tag [AlwaysUpdateSystem] on to the system fixed it

but the EntityManager.DestroyEntity(entity);
only works after hit play
and never works even when an entity created with DestroyData component..

[rotund token]

yes because if it wasn't working before

it won't work now - it means no entity is in your query

you don't need [AlwaysUpdateSystem]

this just implies something is wrong

only works after hit play
systems don't exist outside of play mode

unless you specifically tell them to exist in editor world

[opal lynx]

sorry I said wrong

"only works after hit play"
-> only works right after hit play (maybe only 1 frame).... and doesn't work

[rotund token]

something is wrong with your world/setup then

[opal lynx]

I see...

what's the order between monobehaviour's update and ecs's onUpdate?

[rotund token]

i believe monobehaviours update happens first

i'd have to load up work project

i have no monobehaviour updates in my project to compare

[opal lynx]

@rotund token I see thanks a lot

[rustic rain]

hmm

is conditional dll is a thing?

I think so

[rotund token]

In unity? Yes

[rustic rain]

I figured I could split my publicized dll into one with runtime code and one with conditional code

[rotund token]

You probably need like 50 variations

If you want to make it universal

I can think of at least 7 conditions in entities

Which I think is actually 1000s of combinations in theory

[rustic rain]

just 2

1 for runtime, no conditionals

1 for editor

all conditionals

[rotund token]

What if you want to enable dots debugging in builds

[rustic rain]

sucks to be me

I guess

[rotund token]

You should be having different dlls for development and release as well

[viral sonnet]

@rustic rain at which point do you consider this a bad idea?

[rustic rain]

at a point it can't build

allthough maybe potentially there is a way to mark method to be stripped

before it causes compilation error

[devout prairie]

anybody explain why OverlapSphereCustom doesn't seem to work as expected with a closest hit collector like this:

if (world.OverlapSphereCustom(t.Value, radius, ref closestHit, filter))```

it seems to return a hit when they are within one unit of distance ( ignoring radius )

without specifying the fraction it returns nothing ever, but the docs mention that overlap queries don't use the fraction anyway

[rotund token]

ever solve your issue?

[devout prairie]

no i just reverted to doing a distance check on NativeList<DistanceHit>

[rotund token]

it definitely needs the fraction to not be 0

apart from that i'm not sure what's up

i don't recall having issue with overlap spheres

[devout prairie]

i think possibly a bug tbh

overlapsphere itself is fine, it works with just the NativeList

but passing the ClosestHitCollector in via OverlapSphereCustom seems to not work as expected

oddly, i tried setting maxfraction to the radius

and it seemed to work, but not 'within' the radius

[rotund token]

overlap sphere

literally just uses overlapspherecustom

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

[devout prairie]

yeah

lemme just double check actually, i kindof wondered why using ClosestHitCollector didn't require an allocator etc, maybe i was using it wrong

whereas in the code you've just posted, obviously it passes the allocated nativelist into the allocator

[rotund token]

because you used ClosestHitCollector

it only stores 1 element

it doesn't need to allocate anything

[devout prairie]

well that's what i thought yeh

it's literally just checking against the closest hit, for each hit

then returning whichever one

[rotund token]

it doesn't compare hits

at all

it relies on the physics world property of hits always returning in order

        {
            Assert.IsTrue(hit.Fraction <= MaxFraction);
            MaxFraction = hit.Fraction;
            m_ClosestHit = hit;
            NumHits = 1;
            return true;
        }```

is all ClosestHitCollector does

[devout prairie]

ah so hits are in order of closest to furthest?

[rotund token]

furthest to closest i think

but yeah check that

it's a useful property though to use

[devout prairie]

huh, i didn't realize that

[rotund token]

just break point on an overlapsphere

and look at the list

should see the order

[devout prairie]

so i don't need to distance check then in any case

[rotund token]

yeah sure

it's just a useful property

often avoids you needing to sort

[devout prairie]

so i wonder if the get first collector always returns the furthest ( or closest )

[devout prairie]

so OverlapSphere is passing the 'radius' into the MaxFraction argument of the AllHitsCollector

i assumed fraction to be a 0-1 value

as mentioned, earlier i tried it like this:

if (world.OverlapSphereCustom(t.Value, radius, ref closestHit, filter))```

which technically should do the job, based on what OverlapSphere is doing with AllHitsCollector

but passing a radius of 25 for example, i was getting hits outwith that radius

maybe i've made some mistake, but it's working ok using the same radius in just a simple OverlapSphere

i'll need to double check i think

[rotund token]

you can clearly see Fraction is only used in the assert

are you saying fraction isn't 0-1 being passed in?

because as far as i'm aware it shoul dbe

and if it wasn't you'd be getting exceptions

oh wait this method won't assert in burst

turn off burst and test

[devout prairie]

well you see:

OverlapSphere is passing radius as MaxFraction in AllHitsCollector right

[rotund token]

i don't think it should be

have you confirmed that?

[devout prairie]

my understanding was Fraction was a 0-1 value, representing maybe from zero to the max distance of the cast

so if the distance of the cast was 30 then fraction 0.5 would be 15 distance

[devout prairie]

it's the code you shared, what OverlapSphere is doing

[rotund token]

oh

that's all hits collider but yes that is weird

[devout prairie]

yeah

[rotund token]

        {
            var collector = new AllHitsCollector<RaycastHit>(1.0f, ref allHits);
            return target.CastRay(input, ref collector);
        }```

the raycast versions definitely use 1

[devout prairie]

and as i say, if i replicate that, but with ClosestHitCollector, and pass in my radius as the MaxFactor, i get hits outwith the radius

[rotund token]

yeah can confirm this result

so overlaps aren't passing a fraction

they pass distance

[devout prairie]

it would seem so

[rotund token]

it's definitely not a fraction

[devout prairie]

but as i say, distance isn't being adhered to from what i can tell, unless it's expecting a square or something

[rotund token]

my maxfracture value is right

translation - sphere center = fraction

it's not hit distance

it's center distance

hit point

position is at 0,0,0

overlap center at

which is 3.333787 away == maxfraction

[devout prairie]

i thought fraction would be like:

(hitposition - spherecenter) / maxdist

[rotund token]

definitely not using hit position

at the very least i think it should be distance / radius

whether it should be hit position or not i'm not sure

but the beauty of a collector

you can write your own

[devout prairie]

i'm pretty sure i was getting hits at distance 29 etc when using 25 radius

[rotund token]

because the center is outside

[devout prairie]

ahhhh wait

yeah

i think that's it

it's behaving differently from just OverlapSphere then, and not as expected

initially i thought, i'll just do a spherecast, with a distance of zero

[devout prairie]

but realised it was returning positions on the edge of the sphere i think

so i then noticed that OverlapSphere was an option, which worked as expected

[rotund token]

but yeah going to make a note for myself

about that fraction/radius

that's... annoying

[devout prairie]

however passing a ClosestHitCollector is obviously using a similar fraction approach to SphereCast

it is

[errant hawk]

anyone else notice LocalToWorld.Up is not normalized? its always maxing out at 0.1f instead of normalized 1.0f for each vector component.

[devout prairie]

is maybe your scale not 1? not sure if that affects ltw

[errant hawk]

yeah scale is 0.1 on the y-axis

hmm perhaps scale does modify it

[devout prairie]

scale must be baked into magnitude on the 4x4 matrix then

[errant hawk]

yeah makes sense

[devout prairie]

afaik it does compensate for skew etc with non-uniform scale

[errant hawk]

tbh it's a good thing because it's uniform with the underlying entity's matrix

[rotund token]

for performance all the math operations expect no scale

so if you have scale, you need to remove it from the matrix first

[errant hawk]

Is there any sources on how to do all this? I rarely work with matrix4x4 or float4x4 until I started working in ECS. The docs describe methods, but not what should be used

[rotund token]

sorry was busy

scale is just a multi on the matrix

var scale= float4x4.Scale(new float3(2,2,2);
var localToWorld = math.mul(new float4x4(rotation, translation), scale)

[errant hawk]

ah yeah makes sense

Also, I don't have to do anything special when working with similar components like LocalToWorld ? currently im doing math.transform(local.Value, point) to convert a point from local to world space.

[rotund token]

i think that should be fine

but if you do something like

        {
            return new float3(m.c3.x, m.c3.y, m.c3.z);
        }
        quaternion GetRotation(float4x4 m)
        {
            return new quaternion(m);
        }

on the matrix it will be wrong if it has scale

transform operations should be a lot easier to work with in 1.0 and aspects

[errant hawk]

yeah scale normally doesn't get changed in any game other-than for basic stuff like rocks, logs, or other static items

It will "work" but it's best to keep scale at 1.0 like you said

[rotund token]

(oh i was meaning entities 1.0 but yes in general it's advised to keep scale at 1 where possible)

[errant hawk]

How do you know it will be easier in entities 1.0?

[rotund token]

i'd say one of the major influences of aspects is making transform operations easier

[errant hawk]

Im currently making helper functions exactly for that reason. Trying to keep it in-line with Transform naming convention. For example:

public static float3 TransformPoint (this LocalToWorld local, float3 point) => math.transform(local.Value, point);

however, I think the above method is actually Transform.TransformVector since the scale is not stripped from the float4x4 matrix.

(docs say TransformVector is affected by scale, while TransformPoint is not)

[errant hawk]

Reading the docs rn and came across: ISystemStateComponentData

Is it basically a persistent component that will not be removed without explicitly removing the component during DestroyEntity ? Docs claim a common use case is for cleanup. How does it differ from just using a normal IComponentData component that is for the same use-case (cleanup) ?

[rotund token]

if you call Delete(entity)

there is no IComponentData left to help you clenaup

the entity is gone

if you call Delete(entity) on an entity with SystemStates the entity will not actually be deleted until all system states are removed

All other components are removed though

so the basically pattern is
struct Component1 : IComponentData
struct Component2 : ISystemStateComponentData

// New Entity Setup
Entities.WIthNone<Component2>().ForEach((Entity entity, in Component1 comp) => // ecb.AddComponent2
// Destroyed Entity Cleanup
Entities.WIthNone<Component1>().ForEach((Entity entity, in Component2 comp) => // ecb.RemoveComponent2

the one gotcha with system states is they can't be serialized

if you Instantiate, Copy across worlds, store in subscene etc

the component will not be copied

[errant hawk]

Ah alright I see. For example, it's best used for components that have NativeArray<T>s that need to be disposed?

[rotund token]

there's no real good way of dealing with that

because if you shut down the world they'll all leak since your job won't have a chance to run to dispose them

but yes in theory that type of thing

[errant hawk]

Really just needs to be some sort of OnDestroy for components (regardless of if a world is shutting down or not, it will always be called)

[errant hawk]

Either that's a feature that hasn't been added yet, or unity forgot to document it (like a lot of other things 😅 )

[viral sonnet]

the source code is a lot better documented. if you need info, read that instead of the docs. they are really bare bones

[little crystal]

Hi guys is there a way to query a physics shape in C#? Something like unreal engines, GetAllOverlappingActors() function.

I have this code but I can't find any functions that would let me query the overlapping gameobjects

[rotund token]

PhysicsShape is just an authoring component

it doesn't exist once you convert to entity

is this just for some editor code or something?

[little crystal]

No this is a normal C# script to spawn gameobjects. nothing to do with the editor.

I am not using entities though. just trying to use the DOTS physics. Is that not possible

[little crystal]

Is there a way to query the component, I can't find any documentation online either.

[rotund token]

it sets up PhysicsCollider

part of the physics world

are you looking to query just this collider

or just the physics world in general?

[little crystal]

I just need to query this collider

So physics world consists of all the physics colliders I set in the editor?

and to access this particular collider, I first need to query the physics world?

[rotund token]

https://docs.unity3d.com/Packages/com.unity.physics@0.51/manual/collision_queries.html#code-examples

these are cast examples

but you can do overlay spheres etc as well

[dense crypt]

How can I check if an EntityQuery is initialized/valid?

I'm making an editor tool that's only supposed to run a query when in-game and SceneView is focused, so I can't pre-initialize a query until the game is actually running and SceneView is in focus.

[late mural]

im doing some fluid physics stuff and need a computationally cheap way of rendering my particles, im wondering whether entities or unity's particle system would be better for this? (in future i may be using something like metaballs, but for now im happy to just render the fluid particles as spheres)

[rotund token]

particles are probably always going to be better

[late mural]

ok, thanks a ton!

[rotund token]

if you didn't say physics i would suggest looking at vfxgraph

side note, particle system supports getting particle data in native arrays and manipulating

[late mural]

oh ok that will be very useful then considering how im interopping with nvidia flex, which is a c++ dll

[hollow jolt]

which one would that be ?

[rustic rain]

some batched copy or memcpy

[rotund token]

my question is why convert?

why not just use the array as a NativeArray?

[rustic rain]

potentially, some unity objects don't accept NativeArray

[rotund token]

    var handle = GCHandle.Alloc(vector3Array, GCHandleType.Pinned);
    var addr = handle.AddrOfPinnedObject();

    var nativeArray = NativeArrayUnsafeUtility.ConvertExistingDataToNativeArray<float3>((void*)addr, vector3Array.Length, Allocator.None);
#if ENABLE_UNITY_COLLECTIONS_CHECKS
    NativeArrayUnsafeUtility.SetAtomicSafetyHandle(ref nativeArray, AtomicSafetyHandle.Create());
#endif

    // use nativeArray, when done call
    
    handle.Free();

[rotund token]

i'm saying use an array

and simply reinterpret it as a native array

so you can write to it directly in a job/burst/whatever

no need to double memcpy etc

[rustic rain]

wait what

didn't know this is a thing

[errant hawk]

It's effectively a block copy but for NativeArrays

[uncut rover]

I don't think there is any copy here. just pointing to the pinned memory location

[errant hawk]

block copy is also really slow IIRC... my unmanaged converter was like 100x faster or something (it's late I might be wrong 😅 )

[rotund token]

yeah there's no copy here

that's the point, you're just writing directly to the array

[errant hawk]

oh yeah true, no copy just pinning object and getting the data directly

[rustic rain]

handle is for locking managed object into memory like fixed operator?

[errant hawk]

it's late xD

[rotund token]

yeah

[rustic rain]

interesting

gotta make some extension method for it

[hollow jolt]

there was some unsafe utility methods i seen a year or two ago

that would convert vec3[] to NativeArray<float3>

ah found it #archived-dots message

[rustic rain]

tertle just gave you a way to do it without any copies

direct writing to same array as native

[hollow jolt]

ah damn that's some good stuff

[rustic rain]

        public static unsafe NativeArray<T> ReinterpretAsNativeArray<T>(this Array array, out GCHandle handle)
            where T : unmanaged
        {
            handle = GCHandle.Alloc(array, GCHandleType.Pinned);
            var ptr = handle.AddrOfPinnedObject();

            var nativeArray =
                NativeArrayUnsafeUtility.ConvertExistingDataToNativeArray<T>((void*)ptr,
                    array.Length,
                    Allocator.None);
            #if ENABLE_UNITY_COLLECTIONS_CHECKS
            NativeArrayUnsafeUtility.SetAtomicSafetyHandle(ref nativeArray, AtomicSafetyHandle.Create());
            #endif

            return nativeArray;
        }

poggie woggie stuff

[errant hawk]

You can use UnsafeUtility.PinGCArrayAndGetDataAddress to simplify thing a little bit

[rustic rain]

it's also a null check, huh

[hollow jolt]

no handle.Free(); ?

[rustic rain]

this is utility method

handle is in out

so user calls it himself

[hollow jolt]

do i need to dispose nativeArray or is handle.Free is all that's needed ?

[rustic rain]

nah, it doesn't work

[errant hawk]

idk if you even need the handle becuase now it is unmanaged memory. Calling .Dispose on the NA should work fine

[rustic rain]

you don't get handle

[rustic rain]

you need handle to unpin it

[devout prairie]

is there no equivalent for entityQuery.ToComponentDataArray<>() to return dynamic buffers?

[rustic rain]

huh

[devout prairie]

like i can use myQuery.ToComponentDataArray<MyDataComponent>() to return an array of those components

i take it that it's not possible to return an array of dynamic buffers in a similar way

[rustic rain]

I mean

nothing stops you from doing it manually

if you look at how Unity does it

with normal components

[devout prairie]

what this?

to be honest i was hoping someone would just say 'yeah' or 'no' haha 🙂

it's not a huge deal, was just curious

at the moment what i'm doing is on the main thread, returning a ToEntityArray from my query, then using it to iterate through a BufferFromEntity

just seemed a little contrived for what it was

probably less overhead as i'm not copying a bunch of buffers to an array in any case, but in this case there's not many

[remote crater]

I'm dumb. I forget how to disable a PhysicsBody or PhysicsShape in runtime. I want to ignore collisions.

This isn't working: ecb.RemoveComponent<physicsBody or PhysicsShape>(e); //PhysicsBody or Shape ain't considered components I guess.

I finally have a place to code at again, life is a challenge to even have a roof above your head sometimes

Question 1) How do you disable Collison

Question 2) How do you set a componentdata if it already exists and if it doesn't exist addcomponentdata?

[rustic rain]

addComponentData adds it even if it doesn't exist

set requires it to exist

basically addcomponent data can do structural change

set can't

[remote crater]

TYTY

[remote crater]

Is it possible to scale an entity yet?

[rustic rain]

Scale has been a thing for a while

[pliant pike]

I don't suppose anyone knows what to do about this error when converting a mesh into an entity Shader [Universal Render Pipeline/Lit] on [R Calf] does not support skinning. This can result in incorrect rendering. Please see documentation for Linear Blend Skinning Node and Compute Deformation Node in Shader Graph.

[rustic rain]

if you have skinned mesh

you should probably look up HR manual

[remote crater]

TY looks like it's in LocaltoWorld

[rustic rain]

it's Scale

[pliant pike]

yeah I guess I forgot it doesn't really work with skinned mesh's and that sort of thing thanks

[rustic rain]

or NonUniformScale

or

CompositeScale

[rustic rain]

I'v seen some tutorial parts

about skinned meshes

you sure?

[pliant pike]

well I guess its just finicky

I'll just stick to mesh's without bones for the time being

[remote crater]

Scale worked like a charm, now when I pick up items in my MMO, they tween up to the left in addition to a chat box message, that part seems superior to WOW already.

I think in WOW, you already had a visual of what you grabbed, but in my game with tractor beams snagging everything around you, it helps to have a visual

[rustic rain]

Interesting

            byte* ptr = (chunkComponentTypeHandle.IsReadOnly)
                ? ChunkDataUtility.GetComponentDataRO(m_Chunk, 0, typeIndexInArchetype)
                : ChunkDataUtility.GetComponentDataRW(m_Chunk, 0, typeIndexInArchetype, chunkComponentTypeHandle.GlobalSystemVersion);
            var archetype = m_Chunk->Archetype;
            var length = Count;
            var batchStartOffset = m_BatchStartEntityIndex * archetype->SizeOfs[typeIndexInArchetype];
            var result = NativeArrayUnsafeUtility.ConvertExistingDataToNativeArray<T>(ptr + batchStartOffset, length, Allocator.None);

looking at how chunks memory is done

looks like per each archetype, there's unique memory layout

very interesting

[rustic rain]

btw, this is pretty much how you can create your buffer data array manually

but that rather would be buffer accessor data array

which you probably can wrap in some nice fashion like CDFE

with [int] getter

[rustic rain]

#if !NET_STANDARD_2_0 && !NET_DOTS
#warning Project is expected to have API Compatibility Level set to .NET Standard 2.0
#endif

hmmmm

is that a thing?

I should stick to 2.0?

[rotund token]

For entities? yes

. Net4 is not supported

[rustic rain]

Oof

I'd expect smth that is so reliant on c# would support higher version

[viral sonnet]

do you need .net4 because .net2 is perfectly fine. it's not even comparable with microsofts .net2. it's weird tbh as you have a lot of new features also in .net2

[rustic rain]

Yeah, that's because of their own mono fork

[viral sonnet]

exactly, they just adhere to the standard but have added lots of features

[rustic rain]

I want c# 10.0 so hard

[rotund token]

.net standard 2.1 is mostly ahead of .net4

dont get confused by the numbers

.net 2.0 is roughly equiv to .net frame 4.6-4.8

.net 2.1 has no rough equiv in framework

unless you need some very specific, probably windows only api, then you should not be using .net framework 4 in unity

at least in 2021+

[viral sonnet]

looking forward when .net core is a thing in unity

that brought a lot of changes and improvements.

downside, some of them so big that unity can't just upgrade 😄

the biggest change is probably domain reload

also .net core is fast! behind burst but not by much

[rotund token]

yes

be nice when we're close to parity

[viral sonnet]

man i hope we get 1.0 preview in the next few weeks

can't be that far off now 😄

[rotund token]

if you say that every week

it will eventually be true

[viral sonnet]

noice! and then i'll say, CALLED IT!

no, i say that mostly because of the preparations done. sure, couldn't mean anything but i'm optimistic

if not, max 3 months

[rotund token]

fun fact

https://docs.unity3d.com/Packages/com.unity.physics@latest

points to https://docs.unity3d.com/Packages/com.unity.physics@0.60/index.html which doesn't exist

netcode etc does as well

and 0.60 is their current internal version

[robust scaffold]

All of their latest entity package links point to 0.60 versions. It's so annoying. I know to manually edit the hyperlink to say 0.51 but will the random programmer know? Probably not.

[rotund token]

random programmer probably doesn't know latest is a thing!

[errant hawk]

One of my packages upgraded to 0.60 and back to 0.51.1 once... was really weird

[rustic rain]

use DisableRendering component

[graceful herald]

god damnit everything. Why does google and unity return nothing except when specifically searching "DisableRendering". I searched for hours. Oh well. Thanks.

[rustic rain]

yeah, just assign layer to gameobject before conversion

[graceful herald]

Well yea. But I was referring to runtime.

[rustic rain]

in runtime you will need to change RenderMesh component

it contains layer value

[rustic rain]

I just had a genius idea for ECS

what if make codegen that will let you build state machine jobs

so you write async code in a normal async fashion with all burst/ecs constraints regarding types and etc

and codegen generates a whole job + component with all required fields for such async code running

kek

[viral sonnet]

i think that's the way for FSM. philsa did aomething similar for his polymorohic structs. source is available

[rustic rain]

I'v seen polymorphic

that's not it

I talk here about extension similiar to IJobEntity

[viral sonnet]

that's why i said similar. he wrote it for usage in FSM.

[rustic rain]

you write job, execute method with params

[viral sonnet]

i just don't see the point much to generate a whole job. way too restrictive

generating a layout for the job though to reduce the boilerplate would be cool

i dunno, maybe it has some merit. could turn out great

[rustic rain]

the reason for job is because there is no other way simply

you can only iterate it with normal system

and by having component - probably the best option

potentially maybe you can simply generate struct

with input method of DoUpdate()

yeah, that sounds like a solution

but that would mean you'll have troubles using it, since it's generated code

[viral sonnet]

you'd need to write the struct and the method, then generate code from it

[rustic rain]

yeah, I meant end result

what would it be

in my opinion best solution - generate ready system with job iteration

but if it's Unity which would do it (and I highly doubt they ever will), it can probably be generated into Job

[remote crater]

Dots physics collisions is weird... If a bunch of clumped physics collisions happens, the game freezes sometimes like a random wandering walk is happening to free em.

Is there a way to mitigate this?

[rustic rain]

build it without checks, should probably help with lags

[errant hawk]

How would I go about stripping scale from a float4x4 ?

[rotund token]

https://forum.unity.com/threads/what-do-the-float4-values-in-localtoworld-mean-and-how-can-i-use-them.724622/#post-4858469

when i have a question, i google my past self 😄

i figured it out a few years ago it seems

[errant hawk]

I actually figured it out looking at the TRS function lol

           float3x3 r = float3x3(rotation);
            return float4x4(  float4(r.c0 * scale.x, 0.0f),
                              float4(r.c1 * scale.y, 0.0f),
                              float4(r.c2 * scale.z, 0.0f),
                              float4(translation, 1.0f));

just divide, or use TRS

[rotund token]

yeah what i posted is basically how to get scale

then just invert the multiplication

oh the actual code is a few posts above

            {
                var scale = localToWorld.Value.GetScale();
                var inverted = scale.Invert();
                var value = localToWorld.Value.ScaleBy(inverted); // remove the scale
     
                return new quaternion(value);
            }```

what i linked is just the the extensions this method uses

[errant hawk]

there really needs to be a Scale property on LocalToWorld

[viral sonnet]

are buffers that are inspected usually refreshed? i have to deselect and select the entity to get the current values. i'm pretty sure the refreshing worked at some point. maybe my custom inspectors break the behaviour. any ideas?

[errant hawk]

They refresh for me. It's likely you're not updating the inspector correctly

[viral sonnet]

yeah thanks. i never implemented this in my inspectors. thought i didn't need it 😄 it seems to break refreshing of the default inspectors too so that wasn't a good idea

hm, no idea how to do it. @rotund token you talked about this at some point. how are you updating your custom inspectors?

[rotund token]

yeah buffers aren't refreshed in inspector

i believe values update but size/elements don't

[viral sonnet]

oh is that normal behaviour or something I've broken?

[rotund token]

i believe that's normal

unfortunately

[viral sonnet]

never noticed. quite lame. but thanks

[rotund token]

it may have broken at some point i'm not certain

it is annoying though

[viral sonnet]

didn't have much time for coding the past few days. I've finished the generic stat data type now. removing took me the longest time. that was tricky

[austere pebble]

Its any good template/video about netcode + ECS with unity version 2021+

[rotund token]

https://dots-tutorial.moetsi.com/dots-netcode/intro-to-dots-netcode

only tutorial i know thats half decent

[viral sonnet]

hah, wanted to post the same link. it doesn't get any better than this

[safe lintel]

I wonder just how good documentation/examples will be with 1.0

they've repeated a number of times it was one of the weak points they identified with the preview of dots so far

[austere pebble]

these examples currently its just pain

[errant hawk]

It's actually a bug. The entity count increases, but the element does not appear/refresh UNLESS you either manually refresh the inspector (selecting something else and back) or by clicking on + Add Element. Also, element value(s) update correctly.

[rotund token]

yes it's definitely a bug

i just mean the behaviour is normal, as in he hasn't broken it himself

[errant hawk]

@rotund token @viral sonnet Here's the current behaviour. Everything works other-than new entities appearing

Here's the SystemBase used:

static int c = 0;
        protected override void OnUpdate () {
            bool onAdd = Input.GetKeyDown(KeyCode.Space);
            bool onRemove = Input.GetKeyDown(KeyCode.LeftControl);
            bool onIncrement = Input.GetKeyDown(KeyCode.W);
            bool onDecrement = Input.GetKeyDown(KeyCode.S);

            Entities.ForEach((ref DynamicBuffer<IntBuffer> buffer) => {
                if (onAdd) buffer.Add(new IntBuffer { Value = c++ });
                if (onRemove && buffer.Length > 0) buffer.RemoveAt(buffer.Length - 1);

                if (buffer.Length == 0) return;

                if (onIncrement) buffer.ElementAt(0).Value++;
                if (onDecrement) buffer.ElementAt(0).Value--;

            }).Schedule();
        }

[remote crater]

What does "build it without checks" mean?

integrity checks?

I think so, thank you.

It crashed my library to uncheck that so I am rebuilding my library for the next half hour

I think it will be worth tho

This is why I have two copies of my project open always, when one crashes I work on the other, almost always one crashed, lol

If you experience this too, key is making one project light mode the other dark.

Its funny, I work in two different places on the projects like I'm two different developers collaberating

Can't work on same section since the temporary changes wack each other.

[viral sonnet]

guess i'm lucky, i don't experience any crashes ^^

[errant hawk]

That is a horrible work pattern. Use version control and create branched for big changes

[rotund token]

do you think it's crashing because you have 2 copies open 🤔

[viral sonnet]

wouldn't that be funny 😂

[errant hawk]

Probably would be a reason why lmao

[austere pebble]

seems like gamesettingscomponent got removed also with ForEach not returning void?

[rotund token]

GameSettingsComponent is not part of unity

it's your code

[austere pebble]

mhm

then I just blind and I skipped some part in tutorial

[errant hawk]

Entities.ForEach does not return void

[austere pebble]

Maybe I have other problem than this

[viral sonnet]

it does for .Run and afaik when no dependency is used

[errant hawk]

Context is based off of the error he is getting. the .Schedule() is throwing an error

but yes Run(), and Schedule() or ScheduleParallel() without a dependency return void

[errant hawk]

So I get this warning, and I'm assuming I can just ignore it right?
There is no defined ordering between fields in multiple declarations of partial struct 'WheelColliderJob'. To specify an ordering, all instance fields must be in the same declaration.' was programmatically suppressed by a DiagnosticSuppressor with suppression ID 'SPDC0282' and justification 'Some DOTS types utilize codegen requiring the type to be partial.'

[rotund token]

IJobEntity?

yeah it's annoying

you can add a structlayout tag to it if you want to hide warning

[errant hawk]

I just told VS to suppress the warning to none.. Any benefit of using the layout attribute in this case?

[rotund token]

nah

i dont really use this job type anymore so i haven't noticed the issue recently enough to find best solution

[errant hawk]

What do you use now?

[rotund token]

i pretty much only use IJobEntityBatch

[errant hawk]

What do you do differently than the source generator for IJobEntity ? To me, the generated code looks pretty decent

[rotund token]

there are a lot of optimizations you can do

store temp data per thread

change filters

memcmp on a whole chunk to detect any changes efficiently

change filters definitely the #1 though

[errant hawk]

Filters?

[rotund token]

change filters tell you if any component in the chunk may have changed value snice the last time the job run

lets you early out if there's no changes

also order filter to detect when new entity added/removed

[errant hawk]

ah alright... first time hearing about filters

[rotund token]

my favourite feature

https://docs.unity3d.com/Packages/com.unity.entities@0.51/manual/ecs_entity_query.html#change-filters

though i think you should avoid them on queries and instead do them in a job

as it causes a mini sync point

[errant hawk]

Ah ok so this video explains the same thing: https://www.youtube.com/watch?v=M3o9yOSTdrQ (1:30 for explanation). So any IJobEntity that uses ref will effectively mark that chunk as written to, wasting performance for any future jobs that need to reflect if it actually changed at all. This is regardless if the value is changed because it technically was changed since it is value-type passed as ref, and there's no performant way to determine if it was changed or not.

[rotund token]

yes

if you want to use change filters you have to be very particular about how you code

it is a bit of a pain

and for most users not required

but if you want huge scaling performance then it's how you can do it

i.e. i was testing my effect system at 100,000,000 effects on 10,000,000 targets

and the overhead was under 2ms

optimizing this was what really enforced my belief in them

but i had to come up with a bunch of tricky (hacky) was to avoid triggering the filters

[errant hawk]

In your IJobEntityBatch how do you access the actual filter logic to determine if it changed or not?

because so far only queries show results regarding their usage

oh wait

I think I found something:

Are those and similar what you use?

Also, I think unity might expand on the functionality of filters. ie add the ability to write data but not trigger filters

[rotund token]

yep

that's the one

[rotund token]

you can already do this in IJobEntityBatch

batchInChunk.GetComponentDataPtrRO()

[errant hawk]

you said you needed to do some hacky stuff to avoid triggering them?

[rotund token]

won't trigger filter but you can write to it

so what i did was get the ro ptr

and write to it

[errant hawk]

[rotund token]

then do memcmp on the old/new data

if it's changed i manually trigger filter

(wrote an extension)

[errant hawk]

so performance for you is now a lot better because of filters?

2ms vs what?

[rotund token]

that trick optimized a job from 16ms to 2ms with 100m entities

[errant hawk]

so 60FPS to 500FPS

[rotund token]

i developed that trick for a timer component

timer.value = math.max(0, timer.value - dt)

would always trigger a change filter

but if they're all 0

nothing has changed

[errant hawk]

neat

[rotund token]

generally you'd use a component for state etc

which is fine for nearly all use cases

but i was trying to right this with completely static archetypes

as it makes saving, networking etc much simpler

[viral sonnet]

yes! preach change filters. the best feature hardly anyone knows how to use correctly

[rotund token]

my strategy for spreading the gospel is to just tell enzi then he tells everyone else

[errant hawk]

statemachines are something I really like to work with in OOP, but structs make it difficult to infer logic from types since there's no inheritance at play

[errant hawk]

lmao

once 1.0 a LOT more people will jump ship to ECS since mono/OOP kinda sucks after getting used to this design flow now

ngl a lot of shit is easier to understand using entities over MonoBehaviours

[viral sonnet]

🤣 the trickle down information exchange

[rotund token]

there's a features in 1.0 i really need, hurry up unity

[errant hawk]

Part of the reason I really like working in Burst/Jobs so much is for that extra level of "focus"

[viral sonnet]

i expect a lot of frustration and crying once 1.0 hits

[errant hawk]

meh, that will pass once some content creators release some videos/guides about ECS

[errant hawk]

Anything specific?

[viral sonnet]

most content creators will have a hard time with entities. suddenly a half assed solution won't work out. i expect a lot of abuses and event systems

teaching totally wrong stuff

[errant hawk]

yeah... as for the "right" way it wont end well for some people.

[rotund token]

well 2 things. i really want aspects though i'm scared how they'll work in ijobentitybatch

[errant hawk]

aspects.. I dont recall what those do again. I did hear about them, but it was a while ago

[rotund token]

https://portal.productboard.com/wkjyyxmtns7dipwofp19pj8r/c/1157-adaptive-game-architecture?utm_medium=social&utm_source=portal_share
small snippet here though i think they had a bit more info in a talk at gdc

but it's not a lot of info about it

but basically think, structs that encapsulate multiple components and then provide logic that uses all of them

think, a transform

(i'm hoping)

so all the parent, child, rotation, scale, position, etc info can be grouped into an aspect

then it should hopefully provide as easy to use operations as gameobject transforms

[viral sonnet]

aspects still seem so superfluous to me.

[rotund token]

try do transform operations with a hierarchy

[errant hawk]

oh shit yeah now I remember... kind of haha. But yeah, holy shit this will be the #1 go-to for a lot of reasons, especially during prototyping

[rotund token]

like, you can't use translation that's local space in a hierarchy

got to use localtoworld

[errant hawk]

ECS is the future

lmao

why this wasn't the go-to from the start kinda baffles me

like, yeah you may need to do a little extra work to get things rolling

but once the foundation is setup, things get easier

also, a single SystemBase for 1000s of entities is SO MUCH easier than having to apply changes to an array/List of MonoBehaviours

[remote crater]

Uh, parallel processing, we've had since the 90s, but the speed of computing was faster than the mind games to learn this.

[errant hawk]

I mean, the processing potential still is faster then the mind games today. What I ment was in regards to unity going with mono, or at least implement ECS sooner. My guess is Source Generators being a pretty big aspect of their design pattern

[viral sonnet]

I'd say focus just shifted. performance was never that important for unity

SoA is as old as programming. but we never had the ram nor processing power to do the crazy things we are able go now

[errant hawk]

yeah that too. It's a shame, because open-world, and large scale universes are a big market

[rotund token]

i think it's a bit more than that

the entire industry has been shifting from around 2015 towards ecs

unity just on the bandwagon

not saying companies havent been using ecs before this, just the entire industry has started shifting

[errant hawk]

yeah I noticed that too. Unity is doing pretty good as far as progress goes within the last 2 years or so regarding ECS

[viral sonnet]

first time i heard about ecs was in overwatch. before that it was just a database thing

oop just didn't cut it anymore with the amount of people working on projects and the expectations of players and massive simulated worlds

[errant hawk]

When I first heard about ECS my reaction was "why wasn't this considered sooner..."

[viral sonnet]

you could look at c# and question why some things haven't been done sooner 🙂

and then i say again, the focus just wasn't on performance but that it works and is easy

[errant hawk]

I suppose, but like what @rotund token said the entire industry is shifting, C# as well. C# devs have made a lot of good changes in the past few years

[viral sonnet]

now we have all gone pretty much back to c-style. which is funny to me 😄

glad this, make it as easy as possible but neglect the hardcore programmers era is over in c#

[errant hawk]

It honestly never should have changed. How processors work is linear... C is linear... ECS is linear... OOP is... messy

[viral sonnet]

leave it to MS to fk things up haha

i still say they did a fantastic job with c#

but our timeline could have been a lot less, lets say, convoluted

[errant hawk]

yeah C# is my go-to language... and I dont feel like learning anything else other-than C

I learned JS initially, and when I switched to C# I was like: "wow, javascript is shit"

[viral sonnet]

i used c++ in the past 5 years for ML and now i do ML exclusively in python lol

[errant hawk]

complexity != functionality

C++ IMO is over-saturated with complexity

[viral sonnet]

yep, JS is shit. most browser tech is

[errant hawk]

why there is no industry standard to just use a single language with an open source API for all browsers to use makes me sad

part of the reason why im totally on board .NET/standard getting scrapped and being just Core in the end

makes application compatibility across platforms a lot simpler

instead of one framework having specific aspects to that only work with x platform

[viral sonnet]

being able to deploy .net core on linux is 👌

it's just .net 5/6 now btw. they dropped the core part for some reason

[errant hawk]

unity?

[viral sonnet]

nah MS

[errant hawk]

oh you mean naming convention?

[viral sonnet]

i'll still call it .net core 😄

[errant hawk]

sounds cooler too

[viral sonnet]

right!

the stuff out of web tech is so cringe. like node.js ... good lord

[errant hawk]

ah ok going from .NET Framework to .NET core to just .NET

[errant hawk]

so many memes about all the garbage web devs have to go through just to display a single pixel on a web browser

anyhow, the push for unity to move everything to .NET (Core) is def going to improve the odds of ECS becoming far more mainstream. Also, that means Mono will finally die off (finally)

[viral sonnet]

yep, already said today (or yesterday?) really looking forward to .net core in unity

time for bed now. good night all o/

[errant hawk]

part of it is already here. C# 9.0 in unity 2021+... I remember being stuck on C# 7.0 for a good 2 years while C# was already C# 9.0 at the time lmao

[errant hawk]

gn

[remote crater]

The question is like saying,"Why haven't we always had cell phones. They'd have been useful as soon as we got electricity."

The reason is the increasing usefulness of ECS as multicores happen. We capped clock speed in 2003, but hardware still was able to get efficiencies + video cards.

For a while, first 2-4 cores helped run side processes in windows and such.

But as we get 8-12 cores, its on. Also in the future we'll prolly get 74/128/256 cores

The time for DOTS/ECS is now because its the right time for it.

[austere pebble]

ahh yes

[rustic rain]

leaking, hehe

[austere pebble]

Always when I tried ECS that thing happens

just spam

[rustic rain]

that means you are leaking memory

you didn't dispose of some native collection probably

[full epoch]

I had similar problem - each time i added/changed code in vs 2022 and then started my game i got the same error - i use dots physics btw - and later i found forum post that explains reasons behind the problem - and there was suggested an easy solution to turn on Vsync(game / free aspect / vsync (game view only)) and then those warnings disappeared - forum thread : https://forum.unity.com/threads/dots-multiplayer-and-deleting-an-allocation-that-is-older-than-its-permitted-lifetime-of-4-frames.760391/

[austere pebble]

vsync enabled, free aspect and same thing

[glacial hazel]

Is there a way to to cast or reinterpret a byte[] to NativeArray<int> without copying memory? Just reading the same memory location in where the byte array is, but as a NativeArray<int> ?

[full epoch]

Did you try to build the game - in one post user wrote that after he did the build warnings stopped to show?

[austere pebble]

no

[rustic rain]

there is

        public static unsafe NativeArray<T> ReinterpretAsNativeArray<T>(this Array array, out GCHandle handle)
            where T : unmanaged
        {
            handle = GCHandle.Alloc(array, GCHandleType.Pinned);
            var ptr = handle.AddrOfPinnedObject();

            var nativeArray =
                NativeArrayUnsafeUtility.ConvertExistingDataToNativeArray<T>((void*)ptr,
                    array.Length,
                    Allocator.None);
            #if ENABLE_UNITY_COLLECTIONS_CHECKS
            NativeArrayUnsafeUtility.SetAtomicSafetyHandle(ref nativeArray, AtomicSafetyHandle.Create());
            #endif

            return nativeArray;
        }

You will need to .Free() handle after you're done

[rustic rain]

turn on full stacktraces

[glacial hazel]

Wow thank you!!. I should call the .Free() after I'm done using the NativeArray or could I call it just after that method?

[rustic rain]

no you can't call it before you are done

Free lets GC to defragment this array

so the moment it does it - native array is pointing to invalid part of memory

[full epoch]

My problems with that warning spam started when by accident I pressed ctrl - b in vs 2022 and vs started to build app but tbh not sure if this makes any sense or is it somehow related lol

[glacial hazel]

Ah ok, many thanks!

Also it seems that the ConvertExistingDataToNativeArray<T>() method receives the length of the native array as input, so it would be something like :

var nativeArrayLength = array.Length/ UnsafeUtility.SizeOf<T>();
var nativeArray = NativeArrayUnsafeUtility.ConvertExistingDataToNativeArray<T>((void*)ptr, nativeArrayLength, Allocator.None);

Instead of just array.Length

(Not sure how you did it to write as code here on discord)

[rustic rain]

oh, if you are converting between 2 different sizes

then yeah

[glacial hazel]

Thanks!

[full epoch]

in case u haven't done new dots build before go to ecs manual there is explained how to do this(it's important to add scene list component) : https://docs.unity3d.com/Packages/com.unity.entities@0.51/manual/ecs_building_projects.html also moetsi tutorial : https://dots-tutorial.moetsi.com/unity-ecs/publish-builds-in-unity-ecs also u must add com.unity.platforms package

[olive kite]

What triggers the generated code to generate (temp/Generated Code)? I delete it and recompile but it doesn't get generated. Also tried toggling the "add missing partial keyword" setting in the menu.

[rustic rain]

what is your burst version?

[olive kite]

Version 1.6.6

[rustic rain]

install either 1.7.3 or new 1.8 preview

they fixed some cache problem

which triggered your issue

[olive kite]

oh ok, thanks a bunch

[olive kite]

Updated to 1.7.3 but same result, nothing generated and I end up getting the error regarding codegen nothing working properly

[rustic rain]

what IDE do you use?

[olive kite]

Exception: This method should have been replaced by codegen
LambdaForEachDescriptionConstructionMethods.ThrowCodeGenException[TDescription] () (at Library/PackageCache/com.unity.entities@0.51.1-preview.21/Unity.Entities/CodeGeneratedJobForEach/LambdaJobDescription.cs:149)

Rider

[rustic rain]

have you tried cleaning cache?

[olive kite]

not since updating to 1.7.3, i'll try now

same. made sure I was on latest versions, reset cache

What normally triggers the process? I guess I don't understand the process well enough. When a new script is created and [GenerateAuthoringComponent] is used, is it generated on compile of the script?

[rustic rain]

compilation

[viral sonnet]

anyone familiar with this? why does it run every frame?

[robust scaffold]

Live conversion, it's when you have the inspector for the GO open and it runs to convert GO data to the target entity

Turn it off under DOTS tab up near Jobs.

[viral sonnet]

i haven't selected anything . turned it off now "Live conversion enabled" - still there

[robust scaffold]

Try to close the subscene in the GO hierarchy as well.

[viral sonnet]

lot less now. still there though. eh, guess it doesn't get any better

weird i never noticed this

whole scene runs at 0.88ms so maybe that's why 0.35ms for the conversion are now sticking out 😄

as long as this is only in the editor i can live with it

would be still interested in what it's actually doing. 0.35 is a lot. but going into another rabbit hole. nah, i need to stay focused for once! haha

[forest fractal]

Starter learning dots today. Feels like a very basic question, but quoting docs "Unity ECS automatically discovers system classes in your project and instantiates them at runtime". Surely you don't want all your systems to run on all your scenes, so how do I filter which run on which scene?

Was looking at the samples and still can't figure out how they select which systems run on each sample scene

Feels like I'm missing something very obvious lol

[viral sonnet]

systems run globally, they are not scene dependent. you can manually create/destroy them and disable automatic creation with the [DisableAutoCreation] tag. though i don't recommend that as i don't think that micro management is needed. systems are very lightweight. if you let them run on the right query, they don't run at all.

[forest fractal]

Ah, got it. I see why the samples append _samplename to the end of the components then, to allow each sample's systems to only run for that sample

Thanks 👍

[viral sonnet]

can you link them, i don't remember that part

i think that's more of a naming thing. you can look at window->dots->systems to check which systems are actually loaded

[forest fractal]

https://docs.unity3d.com/Packages/com.unity.entities@0.9/manual/ecs_systems.html
Which if the first result to googling unity ecs select systems to run

[errant hawk]

@forest fractal Like enzi said. Also, if you need to run a specific system, you can attach a component to only the entities you want to modify (ie editor vs runtime/playmode.) You can of course use this method to tag all sorts of things, like: enemies, vehicles, items, players, etc.

[viral sonnet]

oh that's a very old doc btw

we are at 0.51

[forest fractal]

yeah, I noticed soon after

https://docs.unity3d.com/Packages/com.unity.entities@0.51/manual/ecs_systems.html has the same info though

Might as well add, it also says You can view the system configuration using the Entity Debugger window (menu: Window > Analysis > Entity Debugger). which seems to be deprecated

[viral sonnet]

which samples do you mean then?

[forest fractal]

https://github.com/Unity-Technologies/EntityComponentSystemSamples
And specifically was looking at https://github.com/Unity-Technologies/EntityComponentSystemSamples/tree/master/ECSSamples/Assets/HelloCube

[errant hawk]

@viral sonnet I think he means these https://github.com/Unity-Technologies/DOTS-training-samples

[forest fractal]

I see where my thought process was wrong though, thanks for the help 👍

Haven't had much exposure to ECS's before

[viral sonnet]

ok just wanted to make sure. the _sampleName is just a naming thing. there's no actual logic behind it. at least not out of the box

[errant hawk]

No one has really. ECS is still in the process of becoming mainstream.

[viral sonnet]

i think 3-4 years is a long enough exposure 😅

[forest fractal]

I've also seem some rust libraries make use of it for game development (although the gamedev scene for rust is tiny in comparison)

got me pretty interested

[viral sonnet]

you mean bevvy?

[forest fractal]

and legion before it

[errant hawk]

ECS is just now hitting 1.0 soon. 3-4 years in experimental doesn't count

[forest fractal]

from amethyst engine iirc

[viral sonnet]

there are already games released with entities. it's not that experimental for a very long time

[errant hawk]

yeah true... but in comparison there's just not even close to as many users as "normal" mono unity users

[viral sonnet]

unity is just being very careful with their wording

and there never will be as much users for it for at least 4-5 years.

i reckon at least 60% will give up and go back to MB

[errant hawk]

yeah that seems like an appropriate timeframe. I do however, expect to popularity of ECS to skyrocket jump once 1.0 is official.

[viral sonnet]

i don't want to be debbie downer here, but it won't. as long as audio, animations, particles, navmesh isn't rock solid implemented

[errant hawk]

and rendering. AFAIK there is no ECS camera

[viral sonnet]

and we have zero clue about the timeframe of audio and animations which i consider most crucial.

i don't think there ever will be an ecs camera. it can be converted via hybrid

[errant hawk]

well, considering all their resources are poured into ECS and not exclusive modules suggests within a years time after 1.0 for 1 or 2 modules to be production ready

[rotund token]

i dont find audio to be that critical

[viral sonnet]

because we all use fmod? 🙂

[rotund token]

pretty much

[viral sonnet]

my wrong spawn offsets made it look weird ... 😄

[viral sonnet]

hm, i need an inspector for a dynamicbuffer and not its elements

would Inspector<DynamicBuffer<T>> work?

[rotund token]

you're shit out of luck

if you find a way to do this without editing package let me know

but as far as i can tell not possible

i wanted this to write a custom inspector for my DynamicHashMap

[viral sonnet]

lol, damn. i'm in a similar spot again with my generic byte buffer

just tried, doesn't do anything

hm, i can hack it for my case at least as i can move the ptr around from the current byte i'm at. the UX is shit but at least i can see some values

ok, a debug IComp that has the pointer to the buffer header could be a solution

[rotund token]

this is my pain

            {
                IComponentProperty CreateInstance(Type propertyType)
                    => (IComponentProperty)Activator.CreateInstance(propertyType.MakeGenericType(typeof(TComponent)), TypeIndex, IsReadOnly);

                var type = typeof(TComponent);
                if (typeof(IComponentData).IsAssignableFrom(type))
                {
                    if (TypeManager.IsChunkComponent(TypeIndex))
#if !UNITY_DISABLE_MANAGED_COMPONENTS
                        Property = CreateInstance(type.IsValueType
                        ? typeof(StructChunkComponentProperty<>)
                        : typeof(ClassChunkComponentProperty<>));
#else
                        Property = CreateInstance(typeof(StructChunkComponentProperty<>));
#endif
#if !UNITY_DISABLE_MANAGED_COMPONENTS
                    else if (TypeManager.IsManagedComponent(TypeIndex))
                        Property = CreateInstance(typeof(ClassComponentProperty<>));
#endif
                    else
                        Property = CreateInstance(typeof(StructComponentProperty<>));
                }
                else if (typeof(ISharedComponentData).IsAssignableFrom(type))
                    Property = CreateInstance(typeof(SharedComponentProperty<>));
                else if (typeof(IBufferElementData).IsAssignableFrom(type))
                    Property = CreateInstance(typeof(DynamicBufferProperty<>));
                else if (typeof(UnityEngine.Object).IsAssignableFrom(type))
                    Property = CreateInstance(typeof(ManagedComponentProperty<>));
                else
                    throw new InvalidOperationException();
            }
        }```

its basically this code here

unfortunately i see no easy way to insert my own type in here

[viral sonnet]

well, i got it working. i see no way of doing a "proper" solution unless unit enables us to add our own types

[rotund token]

i just thought of a hack

[viral sonnet]

simple checkbox on authoring that adds debugging capabilities. i think that's nice enough

[rotund token]

make it inherit IComponentData as well

😄

[viral sonnet]

on the IBufferElementData? i don't think this would even work.

[rotund token]

sadface

[viral sonnet]

what's so gross about a debug comp? it's not that you need to see the values all the time

[rotund token]

if you name it similar enough that when you search it appears it's fine

the problem is it could break subscenes

if it's an editor only component

so you'll have to add it at runtime

[viral sonnet]

yeah that's why i have an authoring flag

[rotund token]

what do you mean by that?

[rotund token]

I actually dont know if this is still an issue i haven't tested in ages but way long ago changes to components in build vs editor broke subscenes

as they built with editor data

i suspect this might be fine now

[viral sonnet]

what was your use case there? i never did that. i mean, this is really only an editor thing as i need to see the damn values and not convert bytes by hand lol

[rotund token]

i simply had a component that was only existed in editor for debugging

[viral sonnet]

and just that broke the build? wow. i can test this later down the line what an editor directive would do

[rotund token]

well if the component is included in the subscene

but doesn't exist in a build

and you try to load the subscene

bad things going to happen

[viral sonnet]

yeah makes sense. i never investigated at which point subscenes are serialized. i expected them to be serialized at build stage

[rotund token]

yeah i believe you could build it fine if properly stripped

[viral sonnet]

found an interesting edge case. the setup: a subscene GO that has a manual prefab comp and has an authoring element that creates an additional entity. the additional entity has a missing prefab comp then. when you instantiate the prefab the link to the additional entity doesn't get rewritten. it doesn't even get created

also interesting, when the additional entity has a prefab comp there doesn't need to be a LinkedEntityGroup for it to work

[rustic rain]

hmmm

is it possible to make DynamicBuffer parallel safe?

[rotund token]

if you're not resizing it probably

just same as nativelist parallel writer

you'd have no safety to ensure this though

[rustic rain]

ah, that's the thing

[errant hawk]

A dynamic buffer is just a BufferHeader pointer. Same restrictions apply like using a NativeList. Unfortunately there is no parallel writer yet.

[rustic rain]

yeah, I get it

It's just that I want to try and expand idea about using DynamicBuffer on entity as EntityCommandBuffer

so not only buffer is accessible out of anywhere

but also ECB is fully burstable

[errant hawk]

from an ECB you want the ability to resize the buffer during playback?

[rustic rain]

I want a parallel safe native collection that is resizable, like ECB

because you don't know how many commands you'll write to buffer

could be 0, could be thousands

[errant hawk]

Yeah same. Also, looking at the source code it totally looks like it would be possible with DynamicBuffer or future similar types.

[rustic rain]

I assume

dynamic buffer without capacity limit

is just UnsafeList

but I'm not sure whether it supports parallel writing

[remote crater]

I can do this, but I want to inquire: What is the most efficient way to determine if you're within "distance radius" of another object?

Doesn't have to be a perfect radius, I don't need square math, I can do x+y+z

[errant hawk]

In the mean time, you can create a mock ECB that uses interlocking to declare when a resize is happening, freeze when a thread is resizing the array/list, then append data.

[rustic rain]

wouldn't that be?
(posX - posY).magnitude < radius

[remote crater]

Yah, I was just overthinking I spose. That was what I was gonna do.

[remote crater]

Goodmorning

[remote crater]

I was just making sure ECS/DOTS didn't have anything built in

[rustic rain]

math is the only thing that DOTS has built in xD

btw

it has

distance

math.distance

[remote crater]

Cool, but I want it fast, so even the inbuilt math isn't as fast as addition. Great!

Thats kinda what I wanted.

[rustic rain]

for fastest

use distance squared

[errant hawk]

yeah use math.distance for absolute distance, or math.distancesq for faster results against a squared radius

[rustic rain]

square roots are expensive

[remote crater]

Yup! Double checkin. Good morning all. Happy devvin.

[rustic rain]

not sure

whether comparing it with squared radius is correct though

[remote crater]

Good to know the exact commands, and both of those probably are more expensive than addition.

This isn't even something I need to maximize around, I become a perfectionist dork at a certain point.,

[errant hawk]

math.distancesq is just (y - x) * (y - x) where yx are your types (float)

[remote crater]

Cost of multiplication is 4x that of addition/sub: https://stackoverflow.com/questions/39719364/what-is-the-computational-cost-of-floating-point-multiplication-vs-addition

[errant hawk]

wait no

that's just for floats

[remote crater]

You guys gave me waaaay more info that I needed, and I'm just being a computational dork in something that doesn't matter much.

God bless, have an awesome dev day!

[errant hawk]

Ah yeah, it's lengthsq(y - x) which is math.distancesq

god it's too late for this lmao

[remote crater]

You're putting way to much effort into thinking, lolz. Just a fanciful q.

[errant hawk]

I seriously would not worry about the performance difference of float arithmetics. If performance is a concern, I would look at managing entities in a spatial tree that you can compute distance from.

That way you're not checking against hundreds of thousands of entities every update

[remote crater]

You're 100% correct, I assumed that too. But did you ever start out on a whole new architecture and doubt yourself, but still want to dive into coding. You guys assured me I was taking the right decision... Morning stuff.

I'm adding a component in my game when you get close to space stations, you get a menuing window [Dock] [Hail] [Buy] [Sell] [Forge] [Rob] etc

A zone might have 1000-5000 stations tops, so I am not sure I even need a spacial tree

The spacial tree is MOST def the efficiency grabber tho, too early to remember that

[errant hawk]

is this single or multiplayer?

[remote crater]

MMO actually I have research level code from 2003 that worked when I had an inhouse Tekkenlike mmo

This is a client side check so no big

[errant hawk]

yeah you're going to want a spatial tree of some kind. Not just for managing stations, but other ships in range, projectile targeting, etc. It's best to always have some performant tree of some kind to rely on when accessing a small section of data from a large dataset

[remote crater]

Now we're talking!

[remote crater]

Is there a built in DOTS.ECS spacial tree? I figure nothing outside of like 500/500/500 units could interact with each other

I am doing the frame shift every 5000, where the world moves around you due to 32 bit limitations

but an actual zone could be 500,000 to 5,000,000 or more I think... Haven't got there quite yet

[remote crater]

You're going insanely above and beyond my original question.

[errant hawk]

The best method imo is some kind of node tree that can query chunks deterministically. I made a fully burstable octree that has on average 0.2ms execution time. Something similar can be used, because my octree is balanced for voxel generation reasons.

[remote crater]

Code monkey has something: https://www.youtube.com/watch?v=hP4Vu6JbzSo

[errant hawk]

it's better than you coming back here pulling you hair out a month later rewriting everything

[remote crater]

I can figure the query, but what about .foreaches? Maybe have a component that designates each sector?

[remote crater]

EXACTLY!

[remote crater]

At least I can mitigate the refactor knowing about this.

it'd be pretty ghetto code to have a componentdata type for every single sector of about 20,000 sectors

[errant hawk]

yeah pretty much that, but more of a tree-type for even better performance. Research octrees for 3D or quadtrees for 2D

[remote crater]

And I wouldn't trust the inner workings of dots to deal with 20,000 component datas. I'll find some way of tricking the .foreach

I've used spacial isolation since early 90s I believe, gotta isolate only stuff that matters.

[errant hawk]

you want dynamic/lazy loading of sectors/chunks. Minecraft can have near infinite worlds, but doesnt load everything.

[remote crater]

In space, you might not be able to interact with something a few million km away, but you can see it

I want the entity in memory and rendering, but I update the network less often when it's far away.

Now I'm thinking when checking laser hits and such, that stuff will be crowding my .foreach

Oh I know

I'll just add one component

OutOfRelevantRange

and I can add/remove it

This is huuugely good stuff, I knew there was a reason I asked a dumb question earlier

I hate being satisfied with an answer, because now I feel like I accomplished something and want to slack off.

hha

[errant hawk]

just a quick tip: dont use Entities.ForEach for performant code. Use either IJobEntity or IJobEntityBatch (non-batch is easier as it uses codegen to help you out like ForEach)

[remote crater]

Ah, my entities system is an amalgomation system that is ugly but works without conflict or crashes

Dude, this was awesome. Okay, take care, get some rest. Anyone who helps me, if I start making bank of millions/month like mmos do, find me and ask for some change.

lataz

[rotund token]

Ecb is already fully bursted

[rustic rain]

they used function pointer?

[rotund token]

Yes

[rustic rain]

huh, all right

[rotund token]

playback has been bursted for years as long as you don't use a managed operation (shared component etc)

Ecb is so much faster than it was 3 years ago

Magnitudes

[errant hawk]

Last thing: https://www.youtube.com/watch?v=xqtwXD0lFTY Things that are far away, don't actually need to be loaded 😉

[errant hawk]

damn, that sucks

[rotund token]

Phone auto correct

[fallow thicket]

Do you think this game is made by dots? https://play.google.com/store/apps/details?id=com.dxx.firenow

[rotund token]

Why do you think it is?

[gusty comet]

There’s a very big amount of Bots and all that stuff

[errant hawk]

I think it's not because ECS can handle way more than just a couple hundred AI. The AI also looks pretty simple too.

[rotund token]

I'm not even sure it's made in unity

[viral sonnet]

TF relased on Aug 9, 2022 and already over 1 million downloads. that must be some sick ass marketing. like, posting in different discord channels and asking useless automated questions. - who asks if it's made BY dots.

[rustic rain]

Is there a way to register component type manually?

through reflection

instead of RegisterGenericComponent

by the looks of it, no

unless I use hacks

like harmony

[pliant pike]

your probably right it could be a sneaky marketing tactic I was almost tempted to try it

[rustic rain]

what is TF?

bruuuuh, I am so tempted to harmony patch TypeManager

xD

an ability to register generics through reflection

is a must

[rustic rain]

huh

modifying assembly

through reflection

bruh

assembly is abstract

and it has no fields -_-

[devout prairie]

any way to select entitys in either scene or game view and have them be selected in the DOTS Hierarchy?

i do have a selection system for picking in game, maybe there's a way to send an editor event to the hierarchy window?

[rustic rain]

you'll need a whole system for that

[devout prairie]

is it possible to send events to the dots hierarchy window?

[rustic rain]

                #if UNITY_EDITOR
                Unity.Entities.Editor.EntitySelectionProxy.SelectEntity(state.World, selection.mainEntity);
                #endif

[devout prairie]

ahhhhh

how'd you find that little gem

ah! it was right there

works beautifully, thanks for that @rustic rain

[rustic rain]

either tertle, or some repo on github enzi sent

[devout prairie]

slow motion high five dots crew

[glacial hazel]

I need to allocate a huge amount of memory (1 GB), for a super big native array. If I just go with the standard new NativeArray<T>(length, Allocator); it hangs the app for more than a second, because it takes too long to allocate that much memory in one frame.

So my question is, is there a better way of allocating a lot of memory, not necessarily faster, but asynchronously or spread into more frames I guess

[rustic rain]

well, you can quite literally make an async method

oh wait

nvm

I don't think that would work

[glacial hazel]

I think I tried that but it doesn't let me create native arrays in other threads

Yeah

[rustic rain]

I wanted to suggest icnreasing capacity with async method

hm

doesn't it have jobified allocation?

[glacial hazel]

Currently it's persistent allocation

Since things take so long

[rustic rain]

wait

try

[glacial hazel]

Is that what you mean?

[rustic rain]

try creating it without clearing memory

that might be way faster

[glacial hazel]

How is that?

xD

[rustic rain]

it won't clean memory to zeros

while allocating

[glacial hazel]

How do you create it without clearing memory?

[rustic rain]

it's parameter

[glacial hazel]

Ah I see, let me check

UninitializedMemory!

I'll try that

Wow yeah, it's extremely fast

It took just 1.4 ms

Hmm, and if I wanted to create a byte[], I guess with that I'm fecked

[rustic rain]

why not just use native collection in the first place?

[glacial hazel]

Yeah my context is that I'm doing massive hacks, just to try to serialize native arrays

[rustic rain]

serialize huh

[glacial hazel]

I mean they seem hack to me, just because I'm not super experienced with managing memory

So what I'm doing is creating a byte array (which is serializable)

Then reinterpreting that byte array as NativeArray, thanks to the method that you sent me the other day

And then serializing in a C# job system what I want to serialize to that NativeArray

And everything is working great, except one step now, which is when I create that byte array

And my goal is just not to stall or hang the app in the process

Hold on, I'll send a block of code to explain better

[rustic rain]

I get what you mean

but I don't think you can construct managed array without initialization

from what I found this feature was added only in .net 5

[glacial hazel]

Hmm I see

I mean I don't need that necessarily

I just need to not to stall the app or main thread

Well anyways, I'll keep thinking about it, thank you very much for your help @rustic rain, you're the best!

[viral sonnet]

what kind of data are you talking about? and where do you need it to go? i see much better ways than just allocating a big block

[rotund token]

Project day 🥳

What problem to tackle today

[viral sonnet]

is that celebration worthy?

[rustic rain]

huh

interesting stuff is in manual

very interesting

[rotund token]

hell yeah

i get a project day once every 2 weeks

best day of the fortnight

[viral sonnet]

so a day where you can do whatever your want at work?

[rotund token]

i get every second friday off

which is project day!

[viral sonnet]

ah 😄

so what's on the menu?

[rotund token]

character controller i think

i've been really interested in motion matching recently

and want to have a look at that

but i need a controller first

[viral sonnet]

hm, CC is a big one 😄

[rotund token]

i'm not going to make anything crazy

probably just base it off the physics sample one

i was having a quick look before work yesterday morning, setting up my data

got to have your classic gravity fields, etc

and then i was like, wait

make everything a stat!

i have a stat system, whole point is to use this for everything

so now gravity is a stat

which i think is awesome

i can just have an item now that changes your gravity

or create a trigger in world that changes it

no extra work

i'm loving the idea of just making everything a stat

or a skill could inverse a monsters gravity for a short while

so they float away