archived-dots

[viral sonnet]

hm, i find dots a lot less prone for restarting. my <0.17 entities projet is something different though

just entities.foreach and componentsystems. hardly any burst 🤮

and lots of add/remove haha

Overall a good case of what you should not do

[rustic rain]

hmm

quite so annoying

can't convert primitive values to FixedString

[viral sonnet]

is burst 1.8.0-pre.2 usable?

[rustic rain]

works fine for me

[viral sonnet]

ok great. thanks 🙂

it sucks so hard that partial methods can't have return values -.-

i don't quite know how to get around that. basically i use methods that could or could not be source generated. but yeah, i need some form of placeholder. one of these methods is a get value. can't do it, it seems :/

hmmm, SO says ref works. that would be okay

[rustic rain]

can't you use out?

[viral sonnet]

no

    It doesn't have any accessibility modifiers (including the default private).
    It returns void.
    It doesn't have any out parameters.
    It doesn't have any of the following modifiers virtual, override, sealed, new, or extern.

[rustic rain]

could potentially change field

and wrap it all in some getter

get{ PartialMethod(); return _field;}

[viral sonnet]

seems dumb but partial is just a merge basically so you could have 2 methods that return. i don't see the issue though. compiler should complain. done

i don't care if enterprise coders can't get their shit together 😄

[rustic rain]

hmm

can I cast ISystem to other interface through it's pointer?

[viral sonnet]

sure, but why? just cast to your struct or do you need some generic method?

[rustic rain]

yeah, it's generic

I need to make a callback for ISystem through interface

            foreach (var system in World.Systems)
            {
                if (system is IUILoadedCallback callback)
                {
                    callback.OnUILoaded();
                }
            }

For managed it's simple

[viral sonnet]

yeah casting should work. not sure about the ptr though. never done it but in theory i see no problem

[rustic rain]

how to cast it?

[viral sonnet]

get the system struct through World.Unmanaged

[rustic rain]

this is it

not really struct

[viral sonnet]

huh? i mean get the struct and cast to your interface

[rustic rain]

that's what I do

but

I can't get actual struct

in generics

[viral sonnet]

not on my computer. look at the return value

[rustic rain]

SystemHandleUntyped

[viral sonnet]

there's definitely the system struct in there

I'll take a look when I'm re

[rustic rain]

                var sys = World.Unmanaged.GetExistingUnmanagedSystem(type);

                unsafe
                {
                    var sysPtr = World.Unmanaged.ResolveSystemState(sys);
                }

ok I got direct pointer to system state

now

bruh

cannot do direct cast either

hmm

maybe I can make a call through reflection?

[viral sonnet]

ITestInterface test = (ITestInterface)system.Struct;```

[rustic rain]

well that works if you know exact type to get

in my case

I don't

[viral sonnet]

why don't you? 😄 your code isn't a black box to you. reflection won't work i think. the isystem is all in unmanaged space.

for generics i'd stay with systembase

or you register the type somewhere to find it out through the handle

[rustic rain]

I want to have a callback in ISystem

similiar to OnStartRunning

or OnDestroy

[viral sonnet]

        {
            if (_unmanagedSlotByTypeHash.TryGetFirstValue(BurstRuntime.GetHashCode64<T>(), out ushort handle, out _))
            {
                var block = _stateMemory.GetBlock(handle, out var subIndex);
                var sysHandle = new SystemHandle<T>(handle, block->Version[subIndex], (uint) SequenceNumber);
                void* ptr = (void*)(IntPtr)block->SystemPointer[subIndex];
                return new SystemRef<T>(ptr, sysHandle);
            }
            FailBecauseSystemDoesNotExist();
            return default;
        }```

[rustic rain]

yeah, I can't use T

I only have Type

as an object

[viral sonnet]

that's how you get a block and the ptr

the untyped handle has all the info

i think you were on a good track with ResolveSystemState. you could use an empty struct that implements your interface

then call it through the struct as the interface isn't working for T

oh right, they are partials. there might be some trick here

[rustic rain]

eh

I scrapped the idea

I figured other way to have async

way to handle this

    public abstract class UIAddon
    {
        protected World World { get; private set; }

        internal void Create(World world)
        {
            World = world;
            OnCreate();
        }

        protected virtual void OnCreate() { }
    }

Kek

They are created on demand

through reflections

and then cached in some system

so later I once UI is loaded I can supply them with OnCreate virtual

sadly it's only managed

but so far, it's only managed types that need it

[rotund token]

You can have on start running in isystem

[rustic rain]

yeah, and I wanted to have similiar callback

but my own

for when UI is loaded

[elfin spire]

I'm getting loads of these when churning meshes with the Mesh API System.InvalidOperationException: Mesh data does not have {0} vertex component, any idea what that means?

that's with BURST on, with BURST off i get InvalidOperationException: Mesh data does not have Position vertex component

[rustic rain]

I would guess that your Mesh data does not have Position vertex component 😅

[elfin spire]

💪 lol

might be a mesh that is empty then

voxel engine makes loads of crap output

added a test and it's fine now, thanks haha

[drowsy pagoda]

In a situation where I need to use the managed debugger on standalone player from DOTS build using Rider. I can't seem to get it to work. Can someone help me troubleshoot this?

[rustic rain]

Can't debug bursted code with managed debugger

[robust scaffold]

Not in 2021 hahaha

[rustic rain]

Even 2022 simply disables burst

It's just that it does it in a convenient way without a need to rebuild

[robust scaffold]

But you can still debug

[rustic rain]

You debug in 2021 too

But you'll have to disable burst manually for that job

[viral sonnet]

has anyone tried source generators? I can't get any output. just testing my generator with a test project, not unity. any special sauce needed?

read this generator https://github.com/PhilSA/PolymorphicStructs/tree/master/SourceGenerators/PolymorphicStructsSourceGenerators and have set it up similar

oh wow, missed the [Generator] tag ...

[rotund token]

that sounds important

[viral sonnet]

kinda 😄 here i thought ISourceGenerator is enough

i built myself a nice rabbit hole. first i generate a cs file with data and then i need the analyzer and source generator to use some templates to write the actual code with the relevant data types, stats and methods 😄

sadly c# is not very versatile when it comes to numeric data types being generic. total pain and honestly, kind of unexpected. even c++ is more versatile

[balmy thistle]

fascinating, what in particular would be the language support you'd ideally add

[viral sonnet]

using simple data types as T for generic methods would make it a lot easier. couldn't get it to work and using the struct version of the data types is a pain and hurts performance afaik

[rotund token]

does it?

    public struct JobInt : IJobFor
    {
        [ReadOnly]
        public NativeArray<int> Input;

        public NativeArray<int> Output;

        public void Execute(int index)
        {
            this.Output[index] = this.Input[index] + 4;
        }
    }

    [BurstCompile]
    public struct JobIntOperation : IJobFor
    {
        [ReadOnly]
        public NativeArray<IntOperation> Input;

        public NativeArray<IntOperation> Output;

        public void Execute(int index)
        {
            this.Output[index] = this.Input[index].Add(4);
        }
    }```
these 2 jobs generate the exact same burst for me

        where T : unmanaged
    {
        T Add(T b);
    }

    public struct IntOperation : IOperation<int>
    {
        private int value;

    public struct IntOperation : IOperation<IntOperation>
    {
        private int value;

        public static implicit operator int(IntOperation i)
        {
            return i.value;
        }

        public static implicit operator IntOperation(int i)
        {
            return new IntOperation { value = i };
        }

        public IntOperation Add(IntOperation b)
        {
            return this.value + b.value;
        }
    }```

i'll test it as a generic in a sec

[viral sonnet]

cool thanks, but that's what i mean you have to write all these implementations

[robust scaffold]

Generic types is fine in burst. Burst just compiles all possible usages of the type

[viral sonnet]

maybe that particular path is even worth exploring. sure implementing all this is tedious but once done it works. 2nd time i want to set foot in source generators and there are other (better) solutions

but this rabbit hole goes much deeper now and i want to strip/add whole code 😄

[rotund token]

just to confirm for you

            where T : unmanaged, IOperation<T>
        {
            for (var i = 0; i < input.Length; i++)
            {
                output[i] = input[i].Add(v);
            }
        }```
generates the exact same code as just using int
the struct is just completely stripped (as i expected)

[viral sonnet]

ok sick

that's great 😄

thanks

[rotund token]

single field structs are basically just treated as the field

[viral sonnet]

ok i'm investing a bit more into this. maybe you have successfully kept me from going with source generators 😄

i prefer not using SG 😄

[devout prairie]

is it possible to profile a build?

[viral sonnet]

which tool? it works for me

have used superluminal and vtune

you can even profile the editor

[devout prairie]

i've literally never tried just curious.. well i was thinking maybe a debug build and standalone unity profiler or something

[viral sonnet]

no idea about debugging builds, never did so. unity profiler attaches just fine

[devout prairie]

just feel all the editor stuff gets in the way when for example sorting for whats taking longest etc in hierarchy view

so i was thinking, can i avoid all that by profiling a build

[rotund token]

yes

you can attach unity debugger to development builds

[devout prairie]

as in the profiler though?

stuff like this for example

[viral sonnet]

yes i just use this setting

[devout prairie]

the EditorOnly taking up a chunk of time and skewing the sort order in the hierarchy view

[devout prairie]

i guess i'll just try it out right

well i guess

never noticed this before

[viral sonnet]

your machine will also show up here

always worked very reliable for me

[devout prairie]

nice thanks

yeah profiler launches nicely alongside Build&Run

[viral sonnet]

damn, my project is kind of broken. why is unity not compiling my asmdef that has updates to the source generator?

can i force someway to just compile the asmdef? it will compile. reimport doesn't work

this is dumb. compilation stops before getting to the asmdef. making another project then ...

[rotund token]

im not sure about 2021

but in 2020 you had to add 2 labels to your source gen asmdef for it to trigger

i believe this isn't required in 2021 though

[viral sonnet]

not using any real source gen dlls. just a simple script that evaluates a scriptable object

well, i'm generating it now in a new project. i don't get it why it just stops compiling.

maybe i can get away with just htis. that would be nice 😄

[jovial mantle]

Hi guys, anyone can point out what low hanging fruit can I benefit from using C# Jobs in a done traditionally with Monobehavior/GO ? thanks

[devout prairie]

Wondering if anybody can explain this to me because i don't fully understand it..

If i spawn a lot of physics bodies into my scene i'll see multiple build and step physics worlds inside the frame like this:

I'll get more build/steps per frame the more bodies etc i have in the scene:

[robust scaffold]

100%. That's what I'm doing.

[robust scaffold]

Fixed step death spiral.

If fixed step time becomes larger than frame time, fixed step enters a death spiral where it keeps iterating per frame

Only way to fix is disable unity's automatic fixed step ticker and implement your own:https://gafferongames.com/post/fix_your_timestep/

[devout prairie]

Yeah this is the part i'm struggling with

[robust scaffold]

The issue with implementing your own and fixing the death spiral is that it loses the "stateless" feature of unity physics

If you dont care about it being stateless, then yea. Go ahead.

[devout prairie]

I mentioned this a while back but i did notice it's possible to just null out the RateManager and it forces one build/step per frame, but i'd rather 'understand' what's going on

[devout prairie]

hmm ok, i need to get my head around this

[robust scaffold]

What is the spiral of death? It’s what happens when your physics simulation can’t keep up with the steps it’s asked to take. For example, if your simulation is told: “OK, please simulate X seconds worth of physics” and if it takes Y seconds of real time to do so where Y > X, then it doesn’t take Einstein to realize that over time your simulation falls behind. It’s called the spiral of death because being behind causes your update to simulate more steps to catch up, which causes you to fall further behind, which causes you to simulate more steps…

The fix is now "caching" the uncompleted fixed step event that took longer than FPS and not locking graphics update. Finishing the fixed step next frame and so on

Well, there is another option. Which is cutting fixed step time even more so it lands under framerate. And then interpolating the results.

That retains stateless but it may result in wonky physics with high time steps. And mandatory continuous collision detection which may be even more expensive

[devout prairie]

god damn that's a head-f*&^

but i think i understand what you're saying re caching the step and completing later ie it becomes stateful in that case

[robust scaffold]

Yep. Physics is a wild rabbit hole.

[devout prairie]

what i was thinking was:

say in some old school game, if it started to lag, the physics itself would also lag ie it would slow down in accordance with your fps slowdown

so you have say an explosion that takes 20 seconds to complete or whatever

can't the step simply be locked to display fps in that way?

[robust scaffold]

Old school may have framerate locked physics.

[devout prairie]

say if it's not networked, and you don't care about determinism or any of that

but you also don't want death spiral

[viral sonnet]

lock to just 1 timestep?

[devout prairie]

does nulling out the RateManager achieve that or is there more to it

Basically, this line of code gives me my fps back:
World.DefaultGameObjectInjectionWorld.GetOrCreateSystem<FixedStepSimulationSystemGroup>().RateManager = null;

Because it then only does a single build/step per frame

[robust scaffold]

If it takes 20 seconds to complete. You have 2 options. One: Framerate lock it. Prevent graphics update until the explosion completes. That means 20 seconds of not moving, 20 seconds of unresponsive game, and about 2 seconds before the user alt-f4s.

Alternatively, you can unlock frame update but that means the user can run around, do whatever, then 20 seconds later a very delayed explosion happens and what the user did to physics objects will not match what the computation expects

and the physics go wild, moving in unexpected ways since the user changed the state of the physics while it was computing

or the user cant move, which is just option 1.

[devout prairie]

Well what i mean by 20 seconds is, the simulation plays back in 'slow motion' because it plays back at say 2 fps or something, in an extreme lag situation

[robust scaffold]

You really cant play a physics in "slow motion" It doesnt work that way

reducing the time step doesnt reduce the computation cost

In fact, it increases it exponentially

The cost of physics is determining overlaps from axis aligned bounding boxes and then determining if, yes, this object does collide with this other object.

That is required per time step. Regardless of the length of it.

[devout prairie]

So i think my idea is basically what you were suggesting, ie frame locking it

[robust scaffold]

Then yes, setting rate manager to null will frame lock it

But the issue is that now, the physics might not simulate correctly with variable timesteps

[devout prairie]

So if it takes x ms to compute the step, it does that and does not move on to render the frame until it's done with that

[rotund token]

@devout prairie i might have missed something but is that from editor or build?

[robust scaffold]

Yep

[devout prairie]

that's editor

[rotund token]

do you have physics integrity checks on?

[devout prairie]

uhh

[rotund token]

Dots -> Physics -> Integrity checks

[devout prairie]

yeah

[rotund token]

turn that off

[devout prairie]

i do

does that become disabled in build?

[rotund token]

yes

but from your profile

to me you have like every safety feature on

which really kills physics in editor

[devout prairie]

i do yeah, in editor

[rotund token]

turn off safety, jobs debugger and integrity checks and how does it compare

if you want to run these in editor it's often worth just dropping your fixed step to 20 or 30 in editor

to avoid killing your machine

[robust scaffold]

If you can keep physics update less than framerate, then you dont need to change the default physics simulation

But lower time steps results in higher changes for object tunneling and collision misses for fast objects. Might be worth it for generally better performance.

[devout prairie]

tbh it's pretty much the same with those off:

[safe lintel]

make sure you dont have more than one PhysicsStep in your scene like I did 🤡

[rotund token]

sure you dont have full stack leak detection on?

seeing a lot of GC

[devout prairie]

no it was just set to On, trying with off

so Off is definitely a bit better:

[rotund token]

all that red stuff should not exist

how many colliders?

[devout prairie]

that's the parallel contacts job and build jacobians

[robust scaffold]

How many objects?

[devout prairie]

a few 😛

[rotund token]

so 3-7?

because that's a useless response

i have absolutely no allocations when i have all safety systems off

so as far as i can tell

you have something on

[devout prairie]

Entities count on the StepPhysicsWorld system is 7204

[rotund token]

all dynamic i assume?

i can handle up to about 40k dynamics

before it breaks down

[devout prairie]

almost all using joints

[rotund token]

well that might be the problem then

[safe lintel]

thats alot of joints

[devout prairie]

it's 600 ragdolls yeah

i think 12/13 joints on each

[viral sonnet]

are they all on screen? because i see half simulation rate on far away objects a lot

[devout prairie]

not all at the same time but i'm assuming they're all being simulated either way

[viral sonnet]

what's your sim speed? i see 14.5ms for 1 physics step alone. that's almost at the 60fps limit for just a step

[devout prairie]

speed is just the default

[viral sonnet]

hm, not that the sim speed matters much here. is 1 step always this high or just when you instantiate?

the way i see it, to solve this problem is to get the step down at least to 7-8ms. 7ms or lower would be better, you want to have leeway for 1 extra step before spiraling.

[devout prairie]

instantiate is pretty quick to be honest for what it is.. this is after spawning and just letting the ragdolls settle, so it's a good average

i've early exited my other systems with 'return' just so that i can focus on what physics is doing on it's own

[viral sonnet]

some form of staggered instantiate could also work. not all at once but one by one

[devout prairie]

instantiate is ok tbh

so this is profiling the debug build::

[rotund token]

your physics is taking your entire frame budget (17ms of 16ms for 60fps)

so if you do anything else in your entire frame you will spiral down

[devout prairie]

to be clear i'm deliberately spawning a lot of these, to obviously stress test it

so my original question was why does it repeat these build/steps multiple times per frame

i think KornFlakes kindof answered it in a way that made some kind of sense

[balmy thistle]

physics often runs with a fixed small timestep

which is looped until you hit some convergence threshold

[devout prairie]

i guess the next question was, what's a good approach to combat that spiral or should i just limit how much i'm spawning basically

[balmy thistle]

because with discrete physics you're handling the collision response at the same time as your contact point generation, so you want to make sure you don't miss an intersection that might happen with a big timestep

also because it leads to more consistent convergences. You can imagine just doing one giant physics step would be much less stable than 10 little ones

[devout prairie]

so would simply locking the physics step to the frame be a bad idea?

ie the frame doesn't render until the step calculates

whether it takes the step 1ms or 50ms

[balmy thistle]

I wouldn't expect to see that in a real game

[devout prairie]

putting aside the concept of the simulation running in 'realtime' ie locked to the world clock

basically if it takes longer to sim the step, then it takes longer

is that a) possible and b) a bad idea 😛

kinda like how an oldschool game might have done it

say if you have too many explosions, everything slows down including the explosions/physics

[viral sonnet]

the obvious and unsatisfying answer to this problem is, never ever let it happen 😄

that's really the problem with fixed timestep physics. what older games did was have a variable rate of step time.

afaik they were bound to render time

[rotund token]

(this is not a new problem either btw, this is how the regular physics works in unityengine)

[viral sonnet]

yep, and it's pretty much how all modern physics engines work. i'm not aware of any other method

unless you step manually with your own timestep

the physics engine doesn't really care

[devout prairie]

i guess coming from an animation/cgi background i'm kindof used to seeing it like this:

you run an animation with dynamics simulation in the viewport, if it's not too heavy it'll playback at a nice 30-60fps

when it gets bogged down the simulation plays back slowly

[viral sonnet]

wdym slowly? how would it ever catch up? i guess they only run 1 step each frame. at least that's what would i do when realtime framerates are not the issue

[devout prairie]

that's what i mean yeah i guess it runs 1 step per frame

so if that step takes 30ms to simulate then you wait 30ms until the next frame update in the vp

[viral sonnet]

@rotund token i'm generating now a bunch of those. need to see how this solution holds. right now it looks pretty good

no need for any generics that way 🙂

[errant hawk]

should components have any sort of logic? I get that the system or jobs are supposed to handle that, but couldn't I just stuff all the relevant methods in the component and call them when necessary?

also, I noticed that a lot of components dont even have constructors

[balmy thistle]

yeah no

[viral sonnet]

hehe, i've been a pretty big opponent of using var. then i've started to use it while learning entities a lot more because types got so verbose. now i love it because it enables me to return any dataType i want without changing any code

[balmy thistle]

It helps to have an IDE that shows you the type info selectively (like Rider).

[viral sonnet]

i've turned that feature off in rider 😅 no idea why, i don't like it

[balmy thistle]

well I use emacs so don't consider this a personal endorsement

[viral sonnet]

using this struct solution, the code got cleaner than i imagined. good thing tertle brought this up. even though i'm using something completely different, it held me off enough to get knee-deep in source generators. and considering what i have now. it would have been a total waste. still code generating but no ISourceGenerator, just good old stringbuilder 😄

are you deving on linux or is emacs just a personal preference?

[balmy thistle]

personal preference, I've long since committed to emacs for everything

[errant hawk]

Has anyone done any performance benchmarks of (both using burst) IJobEntity vs a normal job? To be more specific, Im interested in raycasting performance using the Unity Physics package.

[rotund token]

IJobEntity just code gens into IJobEntityBatch

[rustic rain]

wait what

ISystems are not created during OnCreate phase?

var ccb = World.Unmanaged.GetExistingUnmanagedSystem<CollisionCommandBuffer>();

[rotund token]

what is CollisionCommandBuffer?

is that a system

[rustic rain]

yes

[rotund token]

an ISystem?

[rustic rain]

yes

[rotund token]

how is it a 'commandbuffer'

and are you getting this from a managed system?

[rustic rain]

it has a buffer

my own one

[rustic rain]

yes, I need to get it in managed system

[rotund token]

yeah

i believe all managed systems are created

then all unmanaged systems are created

[rustic rain]

welp

bruh

how do you even work with ISystem

[rotund token]

what do you mean?

you're not really meant to 'work with it'

[rustic rain]

But I need bursted update

[rotund token]

and ISystem can do that

[rustic rain]

well yeah, and I need it to work with ManagedSystem

[rotund token]

why

[rustic rain]

as command buffer

[rotund token]

nothing you have said indicates you need to access an ISystem

[rustic rain]

how am I supposed to get reference to same collection?

[rotund token]

a) i have many times very strongly expressed my disagreement with ever accessing collections in other systems
b) use a systembase

you know you can't store a native container in a ISystem anyway right?

[rustic rain]

even unsafe?

[rotund token]

sure

an unsafelist won't work though externally

only the hashmaps really will work

but again b) use a systembase

[rustic rain]

how can I burst it then?

[rotund token]

by putting it in a job

or calling a function pointer

which is all ISystem is doing

are you that pressed that you need to save 0.05ms by converting SystemBase to an ISystem?

[rustic rain]

I'm learning my tools, cause I wanted to use that approach in future

I guess what I could do instead

is create entity with Buffer

and store all data here

kek

[rustic rain]

WAIT WHAT

since when Unity Physics debug view works in game view?

pooog

brrrrruuuuh

it's "GIzmos"

I never turned it on in play mode

[hollow jolt]

why my float3x2 variable won't change after completing a IJobParallelFor ?

bounds = new float3x2( math.min( v[i], bounds.c0 ), math.max( v[i], bounds.c1 ) );```

executing the same exact line on a normal loop outside of a job would work exactly as expected

[rustic rain]

you gotta share more code than that

[hollow jolt]

    [BurstCompile] struct ComputeBounds : IJob
    {
        [ReadOnly] public NativeArray<Vector3> verts;
        public float3x2 bounds;
        public void Execute() {
            for (int i = 0; i < verts.Length; i++) {
                var v = verts[i];
                bounds = new float3x2(
                    math.min(v, bounds.c0) , 
                    math.max(v, bounds.c1) );
            }
        }
    }

    static readonly float3 MAX3 = new float3(float.MaxValue, float.MaxValue, float.MaxValue);
    static readonly float3 MIN3 = new float3(float.MinValue, float.MinValue, float.MinValue);

var job = new ComputeBounds { bounds = new float3x2(MAX3, MIN3), verts = vertices };        
job.Schedule().Complete();

that's it ...

tried both IJobParallelFor and IJob

[solemn hollow]

your job works with a copy of your bounds variable. you need to work with native containers or have the data on entities to actually get the data out of the job

[hollow jolt]

can u give me an example of a native container for a float3x2 ?

[solemn hollow]

there is none :/. AFAIK you have to use something like a NativeArray<float3x2> with a size of one

[rustic rain]

how do you read bounds afterwards?

[hollow jolt]

lol

[hollow jolt]

this.bounds = job.bounds;

after Complete()

[rustic rain]

should be it

[hollow jolt]

¯_(ツ)_/¯

[rustic rain]

can't you just share whole code block from start to end?

[hollow jolt]

ill use array[1]

var job2 = new ComputeBounds { bounds = new float3x2(MAX3, MIN3), verts = vertices };
job2.Schedule().Complete();
this.bounds = job2.bounds;

that's all there is

i'm not even scheduling it on the previous job , made sure its a separate one

[rustic rain]

verts.Length
Are you certain it's not 0?

[hollow jolt]

when should i check if not 0 , before schedule or after ?

well its not

not before , nor after

just checked

[rustic rain]

if it's 0, that would mean your bounds are unchanged

cause no work happened

try

[hollow jolt]

ill try with NativeArray[1]

[hollow jolt]

yes ?

[rustic rain]

nvm

[hollow jolt]

im speed

yep that worked

array[1] it is

[solemn hollow]

and you do only ever want to calculate bounds for one object?

[hollow jolt]

yep

i isolated it from a bigger job

will try merge those two

[solemn hollow]

i see

[hollow jolt]

its been a while since i used burst , had been using nativeArrays just now

[rustic rain]

Burst is life

[solemn hollow]

that problem you had had nothing to do with burst though

[hollow jolt]

ye i love it , gotta say the whole setup and dispose is very messy code

it nearly triples the lines of code for the same task

[rustic rain]

well, for one you don't really need Schedule, if you instantly check for result - just run

[hollow jolt]

from laggy 10fps down to 2ms 😛

ye after i merge ill just do
Schedule().Complete()

[rustic rain]

Run() is faster equivalent of that

[hollow jolt]

i don't have it

[rustic rain]

But you do

it's extension method

[hollow jolt]

oh nvm , was trying to look it up in the Handle

thanks

[solemn hollow]

why complete at all? just schedule the second job after the first with a dependency on the first

[hollow jolt]

that's what im doing rn

[rustic rain]

that's not faster

[hollow jolt]

but i want to move the code inside the first one

[rustic rain]

if you have 2 jobs that need to be run in sync

Running them is fasteest

[solemn hollow]

yeah if you want it single threaded

[hollow jolt]

what was that [Decorator] which allows writing to the same array-index in parallel ?

[rustic rain]

[DisableNativeParalellRestrictions] or smth like that

[hollow jolt]

if i would do something like single var container

[rustic rain]

use intellisense to figure actual name xD

[hollow jolt]

yes i did 🙂

was looking up Parrallel but no intelisense

btw

Run has no innerLoopBatchCount

[rustic rain]

You have IJob

in code you shared

[hollow jolt]

IJobParallelFor

in another job

[rustic rain]

in that case, yeah

you need to schedule

[hollow jolt]

what will it do if i just Run( arrayLength ) ?

[rustic rain]

no parallel

all on mainthread

[hollow jolt]

ah interesting

will keep that in mind

[rustic rain]

but it is faster than scheduling if threads not worth it

I sometimes implement double job system which checks for amount of data to process

if it's small amount, I just do Run with IJob

if it's big - parallel

[hollow jolt]

ye i figured

actually making it parallel reduced 1/2 ms

from 2 miliseconds to 1.5 miliseconds

but im doing a big innerLoopBatchCount

the test mesh im playing around with don't have too much verticies ( ~ around 4k verts )

so im guessing the parallel will help with larger meshes

finally got some bounds

now ill go read about Graphics.Draw API 😄

[radiant berry]

Does DOTS have support for pooling or like documentation on pooling with dots?
Like an example with projectiles for example?

[rustic rain]

DOTS doesn't need pooling

all memory is preallocated on chunks

if you mean ECS

[radiant berry]

yeah, any idea where i can read on the differences of memory just a very basic explanation

Just so i know why pooling is not needed in DOTS

[rustic rain]

I'd start with ECS basics
https://docs.unity3d.com/Packages/com.unity.entities@0.51/manual/ecs_core.html

[radiant berry]

I am reading through it and does it determine the size of the chunks based on the components an entity has

[rustic rain]

chunks are fixed size

chunk capacity depends on archetype

basically, how much can you fit in fixed byte array

[radiant berry]

i see, i think i understand it a little better now but not yet that much

But i am waiting for it to go to 1.0 before i will do something with it so i have time

[rustic rain]

ECS is a whole new concept that requires a lot of time to invest, before you can finally understand it

syntax is not much of an issue

it's literally "thinking" that is most of the problem, kek

[radiant berry]

yeah, i think from how it works the biggest thing is the way stuff interacts and how you should make your systems

[rustic rain]

one thing that annoys me most: OOP best practices are worst for ECS xD

[radiant berry]

yeah, that's a big adjustment i think/hope they make some good examples for how you would do something in ECS that was first done with normal gameObjects when 1.0 is out

[rustic rain]

ECS is not unique to Unity

it was made long before Unity has even started making engine

best practices of ECS are already available, basically hehe
But it's all conceptual

[cunning bloom]

I'm trying to load a subscene using LoadSceneAsync. It doesn't load in a standalone build but it does work in the editor. Is there something I've missed?

Here's the relevant loading snippet. The async loading task is started but isSceneLoaded never returns true in a standalone build.

var world = World.DefaultGameObjectInjectionWorld;
var sceneSystem = world.GetExistingSystem<SceneSystem>();

// Hard coded for testing. This is a subscene guid copied from the editor.
Unity.Entities.Hash128 subSceneGUID = new(
    x: 3353769547,
    y: 1097787817,
    z: 3813202594,
    w: 44277289);

Entity sceneEntity = sceneSystem.LoadSceneAsync(subSceneGUID);

// This will load in the editor's play mode but will never succeed in a standalone build.
while (sceneSystem.IsSceneLoaded(sceneEntity) == false)
{
    Debug.Log("waiting for prefabs scene to finish loading");
    yield return new WaitForFixedUpdate();
}

For the standalone build I'm using the platforms package. The subSceneGUID above is a subscene of SimulationEntityPrefabs in the Scenes Info list.

[rustic rain]

why not just get guid dynamically?

is subscene part of scene?

[cunning bloom]

That's what I intend to do but I wanted to make it work first.

Yes, it's part of SimulationEntityPrefabs in the Scenes Info list (Element 1).

[rustic rain]

just make serialized references in some mono to subscenes you need

and then grab all guids from it

I think this way you can even save this mono as prefab, serialized too

haven't tested it though

[cunning bloom]

Thanks, although won't that ultimately lead to the same issue? I'm eventually going to pass a guid to SceneSystem.LoadSceneAsync, so it'll essentially function like the snippet above?

[rustic rain]

in case of game object serialization, it'll be generated during build

which might be the case

why guid is different

[cunning bloom]

I replaced the hardcoded GUID with:

Unity.Entities.Hash128 subSceneGUID = sceneSystem.GetSceneGUID("Assets/Scenes/SimulationEntityPrefabs/PrefabsSubScene.unity");

According to the documentation that will look up the GUID in StreamingAssets/catalog.bin so it should be correct (and it did match the guid I hard coded).
So unfortunately the issue seems to be caused by something else.

[rustic rain]

is your subscene inside your scene?

as in game object?

[cunning bloom]

The subscene is inside the "SimulationEntityPrefabs" scene. However, that scene isn't loaded at runtime. I only attempt to load the subscene through Unity.Scenes.SceneSystem. There's no references these scenes in gameobjects.

What's currently happening is:

1. Game starts with boot scene, which is an empty scene.
2. RuntimeInitializeOnLoadMethod creates a new monobehaviour gameobject which starts a loading coroutine.
3. In the loading coroutine:
   a. Create a new empty SimulationRuntime scene (Used for gameobjects created at runtime).
   b. Unload all other scenes.
   c. Additive load Simulation scene using addressables. This is a regular scene with some gameobjects created in the editor. activateOnLoad is set to false. Wait until loaded.
   d. The snippet above is entered to load the entities subscene.
   e. Activate Simulation scene.

It was my understanding that step d would load the entities subscene into the default entities World and it's not really related to the other scenes there.

[rustic rain]

ugh

too much scenes

I got more problems than solutions from it

[cunning bloom]

Well, most of that shouldn't really matter. Loading an entities subscene should be an isolated step, right?

[viral sonnet]

are you loading a subscene or a scene dynamically? because you should load a scene with a subscene monobehaviour that references the subscene

the subscene itself does only exist as serialized data in a build

[cunning bloom]

I'm dynamically loading a subscene. Thanks, I'll look into that.

[rotund token]

netcode ghosts don't support doubles....?

[viral sonnet]

damn i got kind of addicted to generate source code. all my very generalized source that runs on spell data could be turned into specialized source code

and it's pretty much the only way how i could get the timings down even more because right now i have 3 sources of slowness. a) target stats b) source stats c) blob data. i could eliminate the blob data but i honestly doubt i would gain that much from it.

[errant hawk]

how do you access a component from a parent entity?

[rotund token]

[Get]ComponentFromEntity

or in EntitiesForEach just GetComponent<T>(entity)

[errant hawk]

Im using IJobEntity and I cant seem to find where ComponentFromEntity is. The Parent component doesn't have anything to invoke relating to getting a component.

[rotund token]

ComponentFromEntity is passed in

call GetComponentFromEntity on the system

and pass it into the job

[devout prairie]

using the mouse hover/pick code from the physics samples, i noticed that they're using shared component to store material references, and i get a non blittable error when trying to do the same thing on a standard component - i take it only shared components allow managed object types?

i did kinda wonder why they'd used a shared component for a singleton

[errant hawk]

Shared components types that implement the ISharedComponent interface can only be modified on the main thread, without burst or in jobs because they're technically managed types. IIRC they're going to eventually be replaced with normal struct types that can be accessed everywhere.

[errant hawk]

Yeah that's what it was. Also, it's GetComponentDataFromEntity<T> and ComponentDataFromEntity<T>

[rotund token]

you can have a class based IComponentData

[devout prairie]

That makes sense thanks. The samples hover/pick code does run on the main thread.

[devout prairie]

as i say i get this blittable error when i try use a UnityEngine.Material inside an IComponentData

[rotund token]

is it a class

or a struct

because you can store managed objects in a class based IComponentData

not a struct based IComponentData

[devout prairie]

ahhhhhhh

[rotund token]

public struct UnmanagedComponent : IComponentData
public class ManagedComponent : IComponentData

[devout prairie]

okay thanks

does the same apply to a buffer?

[rotund token]

i don't think so

[devout prairie]

trying to store a buffer of materials

[rotund token]

no

[devout prairie]

hmm

[rotund token]

but just store a list

on a class icomponentdata

not really sure why you'd need a class based dynamic buffer

[devout prairie]

as i say, just trying to store a list of mats

and i get the same managed error with buffer

[rotund token]

List<Material>

on a IComponentData

[devout prairie]

ahhh ok

[rotund token]

hell

[devout prairie]

so because it's a class IComponentData i can use a list ?

[rotund token]

skip the IComponentData

just attached the list to the entity

var list = new List<Material>();
EntityManager.AddComponentObject(entity, list);

[devout prairie]

ahhhhhhhhh really

i had no idea that was possible

[austere pebble]

DOTS works on 2021 unity?

[rotund token]

0.51 does

2021.3.4f1+

[austere pebble]

netcode still death?

[rotund token]

death?

[austere pebble]

I mean if it works also

[rotund token]

netcode works fine

[devout prairie]

tbh i considered just storing this stuff inside system members, rather than the singleton component approach unity have used

it's all main thread anyway

not sure if there's any benefit to having it on a singleton entity/comp

[safe lintel]

kinda breaks the paradigm of ecs if you store data inside systems

[devout prairie]

i guess so

[safe lintel]

Far easier to access managed component data using entity manager and queries than to be dipping into systems from systems too

[cunning bloom]

When I tried this I got an error saying it couldn't find the entityheader. If I look in the StreamingAssets folder in the build I see my subscene files in there but nothing related to the authoring scene. The examples I see in forum posts are passing the subscene guid to SceneSystem.LoadSceneAsync.

[devout prairie]

true in some ways i guess yeah

[rotund token]

are you building with build configurations?

[cunning bloom]

Yes, here's the config:

"SimulationEntityPrefabs" contains the entities subscene I'm trying to load.

[rotund token]

they aren't your subscenes though correct?

[cunning bloom]

Correct

[rotund token]

have you overridden automatic boot strap?

[cunning bloom]

Nope

[errant hawk]

No matter what I do I cannot get rid of this error: error CS0102: The type 'WheelColliderJob' already contains a definition for '__Rocket_Scripts_ECS_WheelCollider_AsymptotePointTypeHandle'

error is caused by the source generator

I don't even have a typeHandle anymore

[rotund token]

code?

[errant hawk]

too big to share

wait nvm there was a typo for a DynamicBuffer

[cunning bloom]

I figured out what the issue was here. In step c activateOnLoad was set to false. This caused subsequent calls to SceneSystem.IsSceneLoaded(subSceneEntity) to always return false. Activating the scene on load in step c fixes this. SceneSystem.IsSceneLoaded will now return true once the subscene is loaded.

[rustic rain]

If smth is on entity - then anything can access it

[errant hawk]

This works fine:

public partial struct TestForceJob : IJobEntity {
    public void Execute (in PhysicsMass mass, in TestForceData forceData, ref PhysicsVelocity velocity) {
        velocity.ApplyAngularImpulse(in mass, forceData.force);
    }
}

// Seperate file
[GenerateAuthoringComponent]
public partial struct TestForceData : IComponentData {
    public float3 force;
}

But this does not work:

public partial struct TestForceParentJob : IJobEntity {
    public ComponentDataFromEntity<PhysicsVelocity> velocityComp;
    public ComponentDataFromEntity<PhysicsMass> massComp;

    public void Execute (in Parent parent, in TestForceData forceData, in ApplyForceToParentTag tag) {

        if (velocityComp.TryGetComponent(parent.Value, out PhysicsVelocity velocity)) {

            if (massComp.TryGetComponent(parent.Value, out PhysicsMass mass)) {

                velocity.ApplyAngularImpulse(in mass, forceData.force);

                UnityEngine.Debug.Log("Applied force: " + forceData.force);

            } else {
                UnityEngine.Debug.Log("Failed to get PhysicsMass component.");
            }

        } else {
            UnityEngine.Debug.Log("Failed to get PhysicsVelocity component.");
        }
    }
}

// Seperate file
[GenerateAuthoringComponent]
public partial struct ApplyForceToParentTag : IComponentData { }

No exception is thrown, and it does reach the "Applied force:" log so it is working.

Here's what the system looks like:

protected override void OnUpdate () {

    //TestForceJob job = new TestForceJob();
    TestForceParentJob job = new TestForceParentJob {
        massComp = GetComponentDataFromEntity<PhysicsMass>(),
        velocityComp = GetComponentDataFromEntity<PhysicsVelocity>()
    };

    job.Schedule().Complete();
}```

[rotund token]

velocity.ApplyAngularImpulse(in mass, forceData.force);

you haven't written the velocity back

you are just modifying the local copy of the struct

        velocity.ApplyAngularImpulse(in mass, forceData.force);
    }```
this case is passing it in by ref

you need to do
velocityComp[parent.Value] = velocity;

after you apply changes

[errant hawk]

what a weird way to write the velocity back

[rustic rain]

I wonder if it's possible to get a ref value with component data from entity

Just like with entity manager

[errant hawk]

Yep, works now thanks

[rotund token]

yeah unfortunately you have to write your own copy as the fields are private

can't write an extension with asmref

[rustic rain]

nothing is private with hacks 😅

Publicizers rule

[rotund token]

yes thats what i mean

you have to write something

it's much clearer to just have an extension method to add support

which you can do for most things in entities as for the most part everything is marked as internal

so it only takes a few seconds to add

[rustic rain]

        public T this[Entity entity]
        {
            get
            {
#if ENABLE_UNITY_COLLECTIONS_CHECKS
                AtomicSafetyHandle.CheckReadAndThrow(m_Safety);
#endif
                var ecs = m_Access->EntityComponentStore;
                ecs->AssertEntityHasComponent(entity, m_TypeIndex);

                CheckComponentIsZeroSized();

                void* ptr = ecs->GetComponentDataWithTypeRO(entity, m_TypeIndex, ref m_Cache);
                UnsafeUtility.CopyPtrToStructure(ptr, out T data);

                return data;
            }

well, tbh all of that is available easily through other means

most complex I guess is EntityComponentStore

because you don't want EntityManager in jobs

yep, feels bad it's private

well, one solution I see

use publicizer, build assembly

load as dll

this way it will bypass accessability I think

since dlls are not recompiled

[errant hawk]

Im really hoping 1.0 will fix all this API garbage that doesn't make much sense. It feels "messy" and overly complicated when it comes down to doing something simple. I can understand the requirements to access the components, that im fine with, but it's everything else that is annoying to work with at times.

[rustic rain]

but, it's not API garbage

it's literally language limitations

which are quite same in c++

what I simply tried to do here: make a hack with pointer

[errant hawk]

Doesn't sound like a language limitation if you can bypass it with a hack

plus, that was kinda my point

the API in some areas is sloppy, I'm absolutely certain that 1.0 will bring a lot of needed changes to the API the will mentally make more sense. I really dont care if I have to do 3 extra steps to make something work so long as it makes sense doing it

Naming conventions is a major aspect of any API. Personally, since a lot of people are going to be familiar with MonoBehaviour, having a similar naming convention for obviously similar components, behaviours, and systems would be best.

[rustic rain]

but it is xD

things that are passed by value are passed by value

pointers are not meant to be standard C# workflow

they require unsafe context

even Unity tried to avoid user from unsafe context

by wrapping it all in safe methods

with tons of checks

[errant hawk]

yes, as does 90% of DOTS logic, which suggests that using unsafe is a standard practice for unity's ECS

[errant hawk]

"safety and performance first" but they still provide pointers to be used, and even promote it in some scenarios

[rustic rain]

it is if you aim for maximum perfomance and you are advanced in C#

but I don't find it as a requirment

[errant hawk]

For DOTS/ECS it absolutely is a requirement

[rustic rain]

meh

[errant hawk]

maybe not for the developer making a game, but for the library it is

I personally prefer using pointers over objects since the logic flow is far more in tune with the logic of how a processor works (linearly)

[rustic rain]

managed logic is part of Unity ECS too

[errant hawk]

yes, and Im perfectly fine with that. Some managed memory on a core aspect of a library is perfectly acceptable

[viral sonnet]

https://github.com/enzi/Entities.Exposed i've posted this quite a few times since you are here. 🙂 it has a lot of these extensions to get ref or pointers

[rustic rain]

what

did you make your own ComponentDataFromEntity? xD

[viral sonnet]

yeah long time ago. can't live without it 😄

i was begging on the unity forum since i dunno, before 0.17 or smth and they did discuss it internally but then never did bother adding it

[errant hawk]

Lets hope 1.0 they do, because now it's going to start getting much easier to add features after 1.0 is released

[viral sonnet]

yeah they should. i consider the default CDFE implementation pretty trash because it returns a struct copy

using the ptr or ref is a pretty substantial increase in performance when you use it for the whole project. also, no copying back needed which results in much nicer code

[rustic rain]

hmmm

does your unsafe CDFE assign dependency to system?

probably not

[viral sonnet]

it goes through the same checks as CDFE but the EntityManager.GetUnsafeCDFE doesn't add a reader/writer because i didn't bother getting the systemState. the ISystem/Systembase version does though 😄

i suppose, don't use the entityManager version. i just needed it to write tests

[rustic rain]

no, it's about adding dependency

internal

for systemstate

        internal void AddReaderWriter(ComponentType componentType)
        {
            if (CalculateReaderWriterDependency.Add(componentType, ref m_JobDependencyForReadingSystems, ref m_JobDependencyForWritingSystems))
            {
                CompleteDependencyInternal();
            }
        }

this one

[viral sonnet]

i literally just said it did so above and you have the code 😄

[rustic rain]

ah

I didn't know how it was called until I found it

kek

[viral sonnet]

yeah it's a weird method name 😄

[rustic rain]

hmmm

I wonder what would be better solution

own CDFE

or hacky extension

[viral sonnet]

i have no idea what you consider a hacky extension. i think the important part it that it'll still be usable after entities updates, easy to integrate and doesn't require any rewrites or local versions of entities

[rustic rain]

hacky extension is compilation with publicized values

so you just access private fields without any regards for it's accessability

[viral sonnet]

i think the answer is pretty obvious if you use words like hacky 🙂

what's the upside of the "hacky extension"?

[rustic rain]

no need to have your own implementation

just use native one

[viral sonnet]

you won't get very far with an extension as the fields are private and not internal

[rustic rain]

that's why it's hacky

publicizer makes everything public

[viral sonnet]

publicizer?

[rustic rain]

hacks

😅

https://github.com/krafs/Publicizer

[viral sonnet]

i'm baffled you are even thinking about it ...

[rustic rain]

well

normally you'd use reflection

but publicizing is fast

and it's native

[viral sonnet]

feel free to do it. idc 😄

[gusty comet]

How could I instantiate a specific entity to a location that a raycast hits? I'm pretty new to DOTS & ECS.

[rustic rain]

instantiate it and then set component of your choice

[whole gyro]

if (collisionWorld.CastRay(input, out RaycastHit closestHit)) {
    Entity entity = EntityManager.Instantiate(prefab);
    EntityManager.SetComponent(entity, new Translation { Value = closestHit.Position });
}

[gusty comet]

Thanks a lot 🙂 But I guess I gotta learn some of the actual meanings of the stuff used by reading the docs before I start playing with the code.

[balmy thistle]

rarely hurts

[errant hawk]

I honestly dont know what Im doing wrong...

just, trying to keep it hovering has been an absolute pain

How Im applying force:

velocity.ApplyImpulse(
    massParent[parent.Value],
    translationParent[parent.Value],
    rotationParent[parent.Value],
    local.Up * force.strength,
    local.Position
);

What I do not understand is why it's bouncing when the rays are not even making contact with anything. It just bounces without input from the rays

console will log when a ray makes contact... multiple times I see it make contact and bounce without the log increasing at all

Here's what it looks like hovering (Hasn't made contact with the collider yet, which IMO is causing the bouncing)

[solemn hollow]

that behaviour looks correct to me based on the fact that you always apply the same force per frame if a ray hits the ground. you will need to do something like vary the strength based on distance to ground and current downwards velocity. You could look at PID controllers

[errant hawk]

The behaviour in the first video is the issue. The second one is expected.

[rustic rain]

if you don't apply force

is bouncing present?

[errant hawk]

Yes, I completely disabled my system and it still bounces like crazy

[rustic rain]

welp

what are you material properties?

for both: ground and hover

[errant hawk]

Physics materials?

I dont have any templates set atm

[rustic rain]

they are set inlined if you don't give it asset

[errant hawk]

It does it with a normal cube wtf. Something is wrong with the physics engine

Here's the components the cube has:

[rustic rain]

what about ground

[errant hawk]

Just a physics shape:

[rustic rain]

and what's inside template?

[errant hawk]

Default:

[rustic rain]

do you have any special logic for filters?

[errant hawk]

no custom system is running atm. it's pure ECS logic if there is, other-than some gizmo stuff

[rotund token]

my cubes do not bounce

not like tha tnayway

[errant hawk]

Im about ready to give up on ECS and go back to GOs

this shit dont work

Making a new project with 2021.3.8f to see if there's a difference, otherwise, I will be postponing the usage of ECS until at least 1.0

[solemn hollow]

shot in the dark: but what about that physics step component? maybe you get penetration over multiple physics frames and it adds forces to push out the collider multiple times

first time i see that component

[errant hawk]

physics step afaik just syncs it to the fixed update

[solemn hollow]

its in fixed update by default

remove the component and see how it behaves

[errant hawk]

the same

[solemn hollow]

try enabling continuos collision detection

[errant hawk]

@solemn hollow what editor version are you using?

[solemn hollow]

2021.3.6f1

[errant hawk]

same as me atm

[errant hawk]

I dont see an option for CCD in either of the components

[solemn hollow]

i think dynamic + interpolation is CCD

[rustic rain]

smoothing

property

can you show cube settings?

[errant hawk]

cube is in its own subscene

still happens with CCD

also tried using a Rigidbody and Box Collider components but same effect

[rustic rain]

have you removed physics step component alltogether?

[errant hawk]

yes

[solemn hollow]

but its not the cube doing the bouncy collision right? its the round shapes

[errant hawk]

all colliders do it

[rotund token]

What's the ground mesh collider

Plane collider?

[errant hawk]

yes

ill try mesh

[rustic rain]

try box

[errant hawk]

YOOO

something's fucked with plane colliders

zero issue with convex

for the ground plane*

[solemn hollow]

i just tried to replicate your scene. no issues at all

plane collider or box collider for ground both works

[errant hawk]

let me try something

[errant hawk]

can you set the scale of the plane to 100x100x100 and see if it bounces?

[rustic rain]

using scale 🤔

[errant hawk]

yeah that is the issue kek

[rustic rain]

and what if you don't scale Z axis

or whatever axis that is meant to be 0

[solemn hollow]

no bounce

all colliders i use have a nonuniform scale too

[errant hawk]

can you also set Size under the physics Shape to match the size of the plane? I think the issue has to do with the Size property because setting it to 1, 0, 1 works as expected, while having it match the scale causes the bouncing

for example, I set the plane to 100x100x100, and the plane Size property to 100x0x100

[solemn hollow]

you mean with the button fit to rendermesh? works for me

[errant hawk]

Ok so

why tf did I have Size set to something so large in the first place

[errant hawk]

yeah works for me as well

[solemn hollow]

i cant reproduce your bounce at all :S

[errant hawk]

set your plane to this:

[solemn hollow]

no problems

[errant hawk]

huh weird

im going to move the ground outside of the subscene

either way it turns out to be the Size property causing the issues. anything greater-than 10 causes the bouncing

[solemn hollow]

my entities are not in a subscene

[solemn hollow]

i had the size 10 times larger even and nothing happend

[errant hawk]

weird

so this might be a isolated issue

[solemn hollow]

@rotund token is burst 1.8.0 pre2 good to use? im still on 1.7.2.

[rotund token]

seems fine

[solemn hollow]

"Setting a breakpoint in an attached managed debugger (Rider/VS Unity Debugger...) on a method that is burst compiled, will switch off the burst code path for that method, allowing it to be debugged as normal." this seems great

[errant hawk]

I've been using 1.8.0 and it seems to work fine

[rotund token]

only works in 2022.2+

[solemn hollow]

T_T

[errant hawk]

is the hybrid renderer required?

[rustic rain]

not unless you have your own rendering solution

[errant hawk]

My entities are invisible without the hybrid renderer, so im going to take that as a yes

[errant hawk]

Did you let the cube plop straight down or did you bounce it off of a corner/edge?

because in a new project it does the same bouncing

[solemn hollow]

straight down. but i rotated it so it would land flat

[errant hawk]

needs to land on an edge

[solemn hollow]

yes it did land on an edge

[errant hawk]

that's so weird

that im the only one with this issue so far

brand-new project and editor install as well (2021.3.8f1)

[solemn hollow]

might be time to post in the physics forum.

try what happens if you switch to havok

[errant hawk]

good idea, ill try that

[errant hawk]

I'll make a post and submit a bug report to see if someone can reproduce it

my guess is when the cube penetrates the plane, the Size property somehow alters the rate of de-penetration... going to have a look at the sourcecode

[jovial mantle]

hi all, probably been asked before, can you (schedule) instantiate/destroy with C# Jobs ?

[solemn hollow]

do you work with entities or just gameobjects+jobs?

with entities you can use commandbuffers to schedule instantiation to a syncpoint in the future. with gameobjects youd need to create a list of all the objects you want to instantiate inside the job and then later outside of the job run through the list and instantiate them

[jovial mantle]

GO + Jobs, no Entities

is it possible to use Jobs, (to use another thread) for only this purpose ?

[solemn hollow]

the instantiation of gameobjects needs to happen on the mainthread

whats your usecase? something that an objectpool cant solve?

[jovial mantle]

right, good point. Honestly, was tinkering wiht the idea of using different threads for different tasks

[solemn hollow]

you shouldnt need to care about what thread does which tasks if you use the job system. it just uses as many threads in parrallel as possbile. what you need to care about is how many syncpoints you introduce where jobs need to be finished.

[rustic rain]

you can schedule it

make a list

add it to job

and if you need to destroy smth

add it to list

after job is completed

do whatever you want with that list in managed code

[hollow jolt]

is there a quick way to convert Vector3[] to NativeArray<float3> and back ?

[rustic rain]

no quick

I mean, you'd need some unsafe extension for that

but it is possible to do very fast conversion between two

lol

Publicizer crashed PC when I tried to compile

attempt #2

eh, it won't compile

oh wait

I know

how to make it actually convinient

Instead of creating publicized dll extension with useful stuff

I can do it with only field that are required for useful stuff

basically, just make a extension for every private field that needs to be publicized

lemme test whether publicized dll works in editor

yep, it works

no crashes

[viral sonnet]

that seems so much worse than the asmref trick ...

you are not using the lookupcache btw.

has to be resolved every time

[rustic rain]

this is literally copy of [int] getter

with different return

besides

I decided to do it differently

I will make a library with methods like this

        public static void* m_access<T>(this ComponentDataFromEntity<T> cdfe)
            where T : struct, IComponentData
        {
            return cdfe.m_Access;
        }

providing access to private fields in different objects

through extension methods

this way it'll be so much easier to make good extensions

without a need to recompile anything

[viral sonnet]

if you go for that, why not publicize the entityComponentStore? just access in cdfe will not be interesting to any other struct

[rustic rain]

I can publicize anything this way

but I want to do it through extensions for each field

the only problem - I'm struggling to get good reference to Unity API

[viral sonnet]

what happens if you share your library? it won't compile, right?

[rustic rain]

wdym?

it can only be compiled with nugget

I tried to drop sources to project

but Unity does not recognize anything regarding publicized fields

I have no idea what's up

it won't build anymore xD

it did build hour ago

[viral sonnet]

i don't get you, why you are going through this. there's a perfectly valid solution in front of you that has no drawbacks or eleborate setup and is shareable. yet you are wasting time on this. 🤔

[rustic rain]

wdym solution?

you can't asmref smth that is private and not partial

[viral sonnet]

most private variables are gettable in some other way. i never found a roadblock yet and i've gone balls deep into extensions. yes sometimes you just have to copy parts of unitys code but you need to do that anyway if you are extending.

[rustic rain]

well I want ComponentDataFromEntity as ref

[viral sonnet]

most important stuff is internal

[rustic rain]

I don't want to reimplement my own cdfe

[viral sonnet]

and the solution is to make something more complicated? the cdfe code is like what, 200 lines?

[rustic rain]

but it's not more complicated

it's actually so simple

and in case unity will make them internal instead

It'll be as easy as just remove old library

and fix naming for properties/fields

[viral sonnet]

yet you are not finished yet. i had the cdfe ref done in like half an hour because i was just copying unitys cdfe and adding my methods to it.

[rustic rain]

well, I work kind of slow after actual work xD

[viral sonnet]

and other people can just install the package in their project and done. no nuget or any other dependencies

[rustic rain]

well

that's my goal

I just build dll

and people can just install it in their project

and done

they can make their own extensions with it

using private fields

if I would want to add new private fields to it - this is when you need to build with nugget, yes

[viral sonnet]

and that dll has the publicizer in it or how does that work?

[rustic rain]

that dll is already compiled publicized

access restrictions exist only in IDE

in compiler

after it's compiled - it'll work just like that

[viral sonnet]

the publicizer modifies the entities dll, yes?

[rustic rain]

no

well, kind of

it simply makes everything public

all fields, methods, properties

so during compilation of dll, you can use it all

but it does not modify any actual dll

it's only for building

[viral sonnet]

so if i want to use your dll i have to use a modified entities dll?

[rustic rain]

in runtime there is no such thing as private

[rustic rain]

no

you simply use my dll

that's it

any idea where can I get a good reference of ECS packages?

for Class Library project

I don't want to make it inside Unity project

[rustic rain]

so, that's it

    public static unsafe ref T AsRef<T>(this ComponentDataFromEntity<T> cdfe, Entity entity)
        where T : struct, IComponentData
    {
        #if ENABLE_UNITY_COLLECTIONS_CHECKS
        AtomicSafetyHandle.CheckReadAndThrow(cdfe.m_Safety());
        #endif
        var ecs = ((EntityDataAccess*)cdfe.m_Access())->EntityComponentStore;
        ecs->AssertEntityHasComponent(entity, cdfe.m_TypeIndex());

        CheckComponentIsZeroSized(cdfe);

        void* ptr = ecs->GetComponentDataWithTypeRW(entity,
            cdfe.m_TypeIndex(),
            cdfe.m_GlobalSystemVersion(),
            ref cdfe.m_Cache());


        return ref UnsafeUtility.AsRef<T>(ptr);
    }

And now I built my own extension method

[rustic rain]

If anyone wants to try xD
https://github.com/bustedbunny/UnityECSPublicized/releases/tag/0.1

[rotund token]

Doesn't this just break all conditional code

[rustic rain]

tbh, idk

I'll try to build

ah yes

it is indeed breaks

hmm

how can I deal with it

I guess I can't

sucks

[rotund token]

I wasn't even implying builds that's a whole other can of worms

I was thinking more things like turning off automatic bootstrap etc

[rustic rain]

I tried build because most of those conditionals don't exist in build

feels bad

[misty wedge]

Anyone here have issues with unity using more RAM each time it recompiles? Just trying to rule out experimental burst / dots stuff as the cause... 😦

[rustic rain]

hmmm
fixed is only usable on managed objects or?

btw, are NativeArrays or other collections pinned in memory or they can be subject of reallocation in heap during defragmentation?

[errant hawk]

How can I update a SystemBase in the equivalent FixedUpdate loop? adding the attribute [UpdateInGroup(typeof(FixedStepSimulationSystemGroup))] throws job dependency errors (trying to write to the same native array)

[rustic rain]

what kind of dependency errors?

[errant hawk]

InvalidOperationException: The previously scheduled job Solver:ParallelBuildJacobiansJob reads from the Unity.Collections.NativeArray`1[Unity.Physics.RigidBody] ParallelBuildJacobiansJob.World.CollisionWorld.m_Bodies. You are trying to schedule a new job WheelColliderJob, which writes to the same Unity.Collections.NativeArray`1[Unity.Physics.RigidBody] (via WheelColliderJob.JobData.collisionWorld.m_Bodies). To guarantee safety, you must include Solver:ParallelBuildJacobiansJob as a dependency of the newly scheduled job.

[rustic rain]

looks like you mess up dependency in your OnUpdate

mind sharing code?

[errant hawk]

there isn't any dependencies. Just scheduling a IJobEntity in OnUpdate

[rustic rain]

well IJobEntity is subject for dependencies

[errant hawk]

        static RayGizmo[] rays;

        protected override void OnUpdate () {
            NativeList<RayGizmo> rayList = new NativeList<RayGizmo>(4, Allocator.TempJob);

            WheelColliderJob job = new WheelColliderJob {
                velocityParent = GetComponentDataFromEntity<PhysicsVelocity>(),
                translationParent = GetComponentDataFromEntity<Translation>(),
                rotationParent = GetComponentDataFromEntity<Rotation>(),
                massParent = GetComponentDataFromEntity<PhysicsMass>(),

                collisionWorld = World.GetExistingSystem<BuildPhysicsWorld>().PhysicsWorld.CollisionWorld,
                deltaTime = Time.DeltaTime,

                rayDebugger = rayList.AsParallelWriter()
            };

            job.Schedule().Complete();
            rays = rayList.ToArray();
            rayList.Dispose();

            Debug.Log(Time.DeltaTime);
        }

[rustic rain]

every time you access any entity and their components you create a dependency

job.Schedule().Complete();

doing this in OnUpdate

[errant hawk]

Not supposed to?

[rustic rain]

it is a sync point, at very least

that results in more main thread idling

but you also

when you Schedule

you don't write Dependency

to it

so what happens

is that it simply runs

[viral sonnet]

try this protected override void OnStartRunning() { this.RegisterPhysicsRuntimeSystemReadOnly(); }

[rustic rain]

and Unity notices it and shouts at you for that

meanwhile during that moment, some other jobs can access component data you requested

Schedule(Dependency)

would probably solve it

[viral sonnet]

yeah, and additonaly schedule with Dependency

[errant hawk]

yeah two jobs try to write to the same memory, which is obviously not allowed or safe. That's what job dependencies are for, but I didn't know how to fix the error

[viral sonnet]

but first you have to register the dependency with the code i posted

[errant hawk]

yeah let me try it

[rustic rain]

World.GetExistingSystem<BuildPhysicsWorld>()
🤔

[viral sonnet]

there's also readWrite to adjust accordingly

[viral sonnet]

yeah, that should be cached in create

[errant hawk]

yeah I havent touched the system since last night. Been trying to figure out other things

[errant hawk]

Alright, so the system is correctly running on the fixed loop, but I get this error: InvalidOperationException: Adding/removing components or changing position/rotation/velocity/collider ECS data on dynamic entities during physics step

[rustic rain]

add true to all CDFEs

if you only read them

and assign readonly attributes to fields in job

[errant hawk]

Im writing to the PhysicsVelocity component on the parent entity, which is causing the error still.

I've set everything else to readonly

idk how else I'm supposed to apply a force to a rigidbody

[rustic rain]

so

how your attributes look?

for system

[errant hawk]

Just this: [UpdateInGroup(typeof(FixedStepSimulationSystemGroup))]

[rustic rain]

🤔

[errant hawk]

should I specify an after as well?

[rustic rain]

try to add

hmm

what is it called

in world system

AddInputDependency

or smt hlike that

oh wait

you still do Complete

right?

[errant hawk]

yeah

[rustic rain]

add

UpdateAfter( ExportPhysics...

just make sure

your system runs

after physics step is fully done

near FixedStep ECB

[errant hawk]

Yep that's what it was

so many weird requirements to get something to just update in fixed lol

Like, I understand what they're doing

[worn walrus]

do u know how to make an ai to follow you

i am dumb

[rustic rain]

update order is very important in ECS

you need to know where your systems need to be called

[errant hawk]

Yeah, but having to use attributes makes it very messy sometimes. It would be nice if there was a system execution order window in the actual editor (like script execution order)

[rustic rain]

it relies on constraints

not on some list with priorities

[errant hawk]

you can still have constraints if it was done through the editor

instead of specifying the attribute, you just add it to a group and set its order accordingly in the editor

[rustic rain]

eh

I don't favour such approach, because it won't support modding very well

[errant hawk]

you can still have attributes

it's just to visualize it and manage it the editor approach is better

[rustic rain]

but it's already visualized

when you create new system you simply just figure where it needs to be

kek

assuming you assign order as soon as you create system

[errant hawk]

That's the problem. Maybe you dont know the order it should actually be in

there's nothing to reference it after

[rustic rain]

most of the time

you don't care about the order

so you just let Unity do it

and when you do care, you probably know where it needs to be

[errant hawk]

well see, when working on prediction and networking it is absolutely critical you have full control over the order of execution.