archived-dots

[viral sonnet]

yeah a good stat system solves a lot of data problems

where to save it, where to read, where to manipulate

i'm re-implementing my incremental game from scratch with the new spellcasting system and the stat system is the nicest thing to work with

[rotund token]

the one thing i'm contemplating is making my stats no longer ints

but it would double my manipulators size

from short/half to just float

i really liked it as ints, but gravity is making me consider my choice

since well -9.81 standard right
now originally the idea was just representing this as -981

but then i started thinking about crit chance, +0.2% doesn't seem uncommon

and realizing maybe i'm just going to pigeonhole myself here

[viral sonnet]

how long until you want a generic stat system? 🙂

[rotund token]

do you mean different types?

[viral sonnet]

yeah

[rotund token]

i don't think ever

[rustic rain]

hm, can you store function pointer on components?

[rotund token]

yes

(im pretty sure though i have not done it myself)

i believe netcode utilizes this a lot

they use this little thing

{
    public struct PortableFunctionPointer<T> where T : Delegate
    {
        public PortableFunctionPointer(T executeDelegate)
        {
            Ptr = BurstCompiler.CompileFunctionPointer(executeDelegate);
        }

        internal readonly FunctionPointer<T> Ptr;
    }
}```

they store a lot of them

[viral sonnet]

why do they need it stored like that?

[rotund token]

it's just wrapping all the compile code

PostSerialize = new PortableFunctionPointer<GhostComponentSerializer.PostSerializeDelegate>(PostSerialize),

so all they have to do is pass a method in

[viral sonnet]

what restricts them from just calling the method?

[rotund token]

they don't know what method it is?

every ghost has a different method

[viral sonnet]

ah 😄 ok

[rotund token]

they all have their own serialize code, cleanup code etc

[viral sonnet]

that's their source gen'd stuff, right?

[rotund token]

for the most part yes

[viral sonnet]

ok makes sense then. quite cool tbh, never needed such a thing but good to be aware of

writing an older project from scratch once you've done hard stuff for months is like going back to a level 1 zone as level 80

[rotund token]

this is why i never get anything done

find a new way i like doing things

ok, 2 years of libraries need to be rewritten

[viral sonnet]

hehe. I've already written this game 2 times. i never liked it. then feature creep and ugh, still wasn't better. some games just fail at the design stage really. but you still need to make it to realize.

[rustic rain]

2 times... More

[viral sonnet]

if i would be strict i've written it 3 times lol. maybe the fourth will lead to something

1 oop, 2 ecs

the first time was in 2015 2016 ... god damn

[viral sonnet]

funny side story, this guy in another gaming discord was adamant on the point that the game uses 1024 bits to represent 1.8 e308 because 2^1024 = 1.797 e308. what i find funny about the 08/15 gaming guy is that even when told, they can't back down 😄 and he's a programmer, so he knows best anyway.

[viral sonnet]

hm, i already have a handle in place for this but what if i didn't. i have a target system which writes to TargetInfo. i have another system that needs the targetInfo, so a clear dependency. can this be designed in a way that i wouldn't require the jobHandle from the target system. would a writegroup help me somehow?

[rotund token]

isn't this just by default handled by the dependency system?

or is this like a native container

[viral sonnet]

it does? huh. i still have some huge holes in understanding the dependency system what it can and can't do.

no, entities and icomps

[rotund token]

the system that reads from it is going to get the dependency from the system writing

[viral sonnet]

ok cool, yeah. no errors. seems to work fine 😄 great then thanks

just to confirm this, the dependency is defined by the query and not the job parameters, right?

[rotund token]

it's defined by everything in the system

any GetEntityQuery, GetComponentDataFromEntity, GetComponentTypeHandle etc

this is what i mentioned the other day missing from your custom fromentity

AddReaderWriter(isReadOnly ? ComponentType.ReadOnly<T>() : ComponentType.ReadWrite<T>());

this line is what adds the dependency to the system hence it's really important!

[viral sonnet]

thanks for clarifying

[errant hawk]

the function pointers are not codegened right? You can assign at runtime like normal, right?

Right..?

[rotund token]

yes

[errant hawk]

finally a solution to my state machine conundrum

[viral sonnet]

do i need to free such calls UnsafeParallelList<T>* unsafeParallelList = allocator.Allocate(default(UnsafeParallelList<T>), 1); I've compared it with UnsafeList and there no free is done on UnsafeList<T>* listData = allocator.Allocate(default(UnsafeList<T>), 1);

which confuses me tbh 😄

[errant hawk]

yes, you need to free it

UnsafeList is what a NativeList uses internally iirc

use this to free memory: UnsafeUtility.Free(void*, Allocator)

[viral sonnet]

yeah, i know the rest. turned out i've missed another call which isn't called dispose /// <summary> /// Destroys the list. /// </summary> /// <param name="listData">The list to destroy.</param> public static void Destroy(UnsafeList<T>* listData) { CheckNull(listData); var allocator = listData->Allocator; listData->Dispose(); AllocatorManager.Free(allocator, listData); } which frees the listData

confusion solved 😄 thanks

so working with unsafeList. just disposing is not enough. it has to go through the Destroy method which is kind of unintuitive

i don't use it, but if i would, i'd have memory leaks for sure 😄

[rotund token]

i'm confused

you only need to call Dispose on unsafelist

the Create method that does

UnsafeList<T>* listData = allocator.Allocate(default(UnsafeList<T>), 1);
is Internal

and only used by NativeList

because it allocates a pointer to the unsafelist

[viral sonnet]

oh right! ...

just new UnsafeList<T> doesn't need to be freed itself

ok, then all clear. 😄 had to add the destroy pattern to my native containers because those were leaking :/ at least i catched it soon enough

the downside of the safety system. it doesn't save your ass when you are the one writing containers lol

[errant hawk]

That's why you got to add the safety system first

[viral sonnet]

i've added it. it just doesn't catch such leaks.

it's just a dumb handle. it doesn't know about the internal allocs a container makes

[rotund token]

that's what dispose sentinel is for

but yes if you don't write it properly

it doesn't help!

[errant hawk]

I thought the disposal sentinel is part of the "safety system"

[viral sonnet]

yeah, i've been calling it safety system. dispose sentinel is what i meant

[rotund token]

yeah i removed my comment didnt want to confuse

i personally consider them different things

you can add atomic safety handles etc to a container

for safety

without adding dispose sentinel

i have a bunch of containers that don't have leak safety but do have thread safety/etc

[errant hawk]

yeah for R/W access vs allocation managment

makes sense, I can see why you separate the two

[viral sonnet]

kind of excited how the new system will be with the dispose sentinel gone. will a leak be detected then even when just going through allocator.Allocate?

[rotund token]

/maybe/

[viral sonnet]

hope it does, would be cool and a benefit of just using malloc

[rotund token]

technically leaks are detected if you just use malloc atm

it just gives you no info about it

[errant hawk]

I personally just avoid using unity's built in stuff. Much easier to manage memory on your own (with the exception of their methods, I use a few sometimes)

[viral sonnet]

well that's unhelpful with no info lol

[rotund token]

and you can turn on -debugallocator

[viral sonnet]

oh, i start unity with that?

[rotund token]

yeah

what unity always tells me to do

not sure how useful

[viral sonnet]

i will try to make sure i have every leak covered now

does it report anything then with that flag?

[errant hawk]

Point Im making is it's easier to manage memory without overhead from safety system. I would honestly use the dispose sentinel over the atomic system to ensure disposal is being reported correctly

[rotund token]

the problem enzi has is the dispose sentinel doesn't actually track the memory

it tracks that you've told unity that the memory has been released

it doesn't actually verify it itself

[viral sonnet]

yeah, i could just free the dispose sentinel and do nothing else. it will leak but no one complains about it

so, tried with the flag. nothing is reported. either a good sign or it doesn't report as you said

[errant hawk]

Does PhysicsVelocity.GetLinearVelocity() slowly rotate/slide for anyone else?

having a hard time diagnosing the issue

[tropic venture]

Hello (and noob warning..)

making a mobile game and trying to push the limits of graphics; the scenes/levels are supposed to be very ambient / environmental / immersive, very scenic

basically - trees. thousands of trees and grass and vegetation as much as possibly possible within mobile (even some medium-low end devices)'s framerate capacity

they don't move around, just for pleasing the eyes

all i need them to do is: 1) collide when player tries walking through a tree 2) sway a bit in the wind

can DOTS help with the graphical rendering aspect of this? can DOTS increase performance and allow many more instances to be on scene even in mobile, if it's done with DOTS?

ok maybe i want trees to cast shadows too... and to have nice textures & shading.... and maybe i want to detect if a raycast collides with certain vegetation (bushes meant for hiding/stealth)

[rustic rain]

Yeah, but don't overestimate what mobile can do

[tropic venture]

i'm trying not to overestimate, but i am asking / pushing to discover what is the technological limit possible... whatever the limit may end up being

[rustic rain]

Some things you mentioned can't even be done on some low end desktops xD

[tropic venture]

i'm kinda hoping to theorize based on Far Cry 1 if you remember that game from 2004;
the graphics didnt really age well, by now it looks very similar to those late 90s racing/arcade/FPS games which were trying to be realistic but end up having that, well, late 90s kind of game aesthetic. However - they had low graphics settings which were able to work even on some lower end desktops

so i theorize / wonder.... can 2004 PC graphics happen on 2022 generation of high/medium/low end mobile devices?...

but yeah maybe i'm dreaming and delusional 😛

[rotund token]

i don't think you'd have any issues running a 2004 game on a high end phone

[tropic venture]

i see. but only high end then i guess?

[rotund token]

phone range vastly

but think about this

there are emulators on android that play ps2, gamecube/wii games

[tropic venture]

heh true

want this on mobile

can DOTS help render some thousands of vegetation/trees on mobile... 🤔

[rustic rain]

this is already a thing for mobile even without dots kek

[tropic venture]

please do share

[rustic rain]

just gotta look around play market

Elder Srolls Blades

ark

it's not really dots that makes all of this possible

it's individual optimizations

like effective culling

also unloading unused parts of world

and etc

[full epoch]

Is good practice(performance-wise) to save entity(that contains prefab) as a system class instance field or to use GetSingleton when ever we need it(saw this on yt dots related vid)?

[rustic rain]

if your prefab is persistent

why not

saving entities as fields is a normal practice

simply because some things are just unburstable regarding getting them

and other cause sync points

[tropic venture]

though - can DOTS help at all? let's say maybe the other techniques you mention (world loading/unloading, culling etc) account for 60-70% of performance advantages, might DOTS get another 10 or 15%? or does DOTS not really solve anything for dense environment props/decoration like trees & vegetation at all

[rustic rain]

dots makes everything faster

it's great if you aim for scalable simulations

some people here process millions of entities in a matter of 2 ms

[tropic venture]

what DOTS markets on the tin is "have thousands upon thousands even millions of objects in your game fast", is dense vegetation / grass / trees not a very pertinent use case for that?

[rustic rain]

sir

Consider DOTS as another engine, inside Unity engine

that's it

[tropic venture]

interesting

[rustic rain]

it is meant to be perfomant by default, compared to classic Unity where it's simple by default

[tropic venture]

i see

so if i'm ready to bite the "not simple by default" bullet, then it's pretty much an easy - why not use DOTS, for any project, dense forests or otherwise

[rustic rain]

it's not released yet

it's more like in Alpha state rn

even though some parts of it are fully released

[tropic venture]

ah ok

not a huge deal, if it works it works i guess

[rustic rain]

hmm

public NativeParallelHashMap<FixedString128Bytes, UnsafeList<Entity>> TypeCache { get; private set; }

I have such field

var typeCache = TypeCache;
Before job I do this

then in job I get UnsafeList value, if it's not created I create it and add it to map

and fill list

but later

when I access this list

I end up with this

even though I made sure

list got filled

[rotund token]

Remember unsafelist is a struct

You have to write back to the hashmap and changes you make

[rustic rain]

huh?

I thought it's a pointer to memory

or wait

if I resize it

pointer gets invalidated?

[rotund token]

{
    var list = new UnsafeList<int>(Allocator.Persistent);
    Test1(list);
    var length = list.Length; // 0
}

public Test1(UnsafeList<int> list)
{
    list.Add(1);
}

[rotund token]

the length is not a pointer

only the buffer is

it must be passed by ref

[rustic rain]

huh

interesting

meanwhile NativeList can be used as value?

[rotund token]

Yes

Native list just has a pointer to unsafelist

[rustic rain]

ah

double wrapper

I see

hmm

is it possible to get value as ref from HashMap?

[rotund token]

sure, if you write your own lookup extension

[pliant pike]

Just another point to make for the above to why DOTS is good in my opinion its not just about performance

it helps me to write cleaner code and helps me to figure out problems a lot easier

OOP makes my brain hurt I have to figure out what is an object and is it an object of another or where is the object, some abstract concepts don't fit well into objects

with DOTS its all just data and how can you transform it in the most efficient way possible, simple

[rustic rain]

the fact that all data is public and accessible to anything is good

no need to get references

[pliant pike]

yeah that too data access is like a fundamental property, of it, OOP you've gotta figure out your own way to do it and there all more likely less efficient than in DOTS

[rustic rain]

well, not really

in OOP getting direct references is somewhat more efficient

but the fact that you have to create a whole system to get those references sometimes

is uuugh

[errant hawk]

Managed references are slower than unmanaged pointerz

[pliant pike]

really I was always struggling with getting object referencs in real time in monobehaviours

you always had to precache refs and had to know exactly what you needed before you needed to it to do it efficiently

[viral sonnet]

i got into a discussion about this. What are your arguments here?

[rustic rain]

it's not about pointer vs reference

it's about

object.field vs ECS query

[errant hawk]

Ah nvm then

[viral sonnet]

you don't have to make if object.field vs query ... the same could be done for OOP. really, the way you get references is highly variable. most of the references i need are pointers on the icomp that is being processed

[rustic rain]

well, you usually either do World.GetSystem... or query for specific entity

and most of the time you store entity reference

not direct pointer

[errant hawk]

One is managed, the other is not. Instead of the managed reference being moved/deallocated by GC, pointers are fixed to a specific address of memory. References point to an object that is fixed, which point to the actual memory iirc (kinda like pointer redirection)

[viral sonnet]

well, as i said, i have direct pointers. going through world is very OOP like. querying has a lot of overhead, the question then is, can at least the required entity be cached because a CDFE access is also very OOP like again and still really fast

[pliant pike]

I don't understand why you would use pointers in Unity to be honest

[errant hawk]

Performance

[viral sonnet]

you can do stuff you can't otherwise 🙂

[pliant pike]

I guess but they are also dangerous

[viral sonnet]

lol, not really

[errant hawk]

Type conversion, serialization, etc

[rustic rain]

wait, so if you have singleton entity and doesn't ever change archetype

does it mean you can store direct pointer to component?

[viral sonnet]

no changes in archetype, yes

[rustic rain]

huh

interesting

didn't know that

any code share how you do it btw? maybe some extension?

[viral sonnet]

i use this in quite a lot of places and only set pointers again when there are structural changes in the chunk

[errant hawk]

Not at all, unless you're careless with the bounds of the memory you're accessing

Alloc bytes > write to same bytes > share bytes > dealloc bytes > profit

[rustic rain]

I guess it's possible to make some kind of ComponentReference

[pliant pike]

well yeah that's what I'd be concerned about I might forget to release and it and the address gets assigned to something else I don't know how likely that is though

[rustic rain]

which will store pointer and chunk system state

version

[viral sonnet]

        public unsafe struct LinkStatEntityBuffers : IJobEntityBatch
        {
            ...
            public void Execute(ArchetypeChunk chunk, int batchIndex)
            {
                bool orderChanged = chunk.DidOrderChange(version);
                
                if (!orderChanged)
                    return;
                
                var references = chunk.GetNativeArray(SpellEffect_StatEntityBufferReference_WriteHandle);
                var statEntityReferences = chunk.GetNativeArray(StatEntityReference_ReadHandle).Reinterpret<Entity>();
                
                for (int i = 0; i < chunk.Count; i++)
                {
                    var reference = references[i];                    

                    var statEntity = statEntityReferences[i];

                    // todo, reduce the lookups, this could be merged into 1 lookup
                    var tmpBuffer = StatBuffer_Changes_Lookup[statEntity];
                    var tmpBuffer2 = StatBuffer_MainEffectReferences_Lookup[statEntity];
                    var tmpBuffer3 = StatBuffer_ChangesValues_Lookup[statEntity];

                    references[i] = new SpellEffect_StatEntityBufferReference()
                    {
                        chunk = StorageInfo[statEntity].Chunk,
                        bufferHeader_StatChanges = tmpBuffer.GetBufferHeader(),
                        bufferHeader_StatChangesValues = tmpBuffer3.GetBufferHeader(),
                        bufferHeader_References = tmpBuffer2.GetBufferHeader(),
                    };
                }
            }
        }``` nothing special to it. here i set the BufferHeader* so I don't have to go through additional lookups.

[rustic rain]

so every access you check for structural changes

just in case

[viral sonnet]

yes, that's crucial.

then it's really safe and fast

[errant hawk]

Once a pointer is allocated it doesnt get overwritten. You can however, read or write bytes that other pointers "own"

[rustic rain]

that's not the issue with Unity ECS chunks

they are not subjects of GC

so

they don't get moved

meanwhile data in chunks can

[pliant pike]

ok that's good

[rustic rain]

so pointer access is somewhat limited

in applications

[errant hawk]

Dont chunks only move during structural changes?

[rustic rain]

I actually meant, how you create such pointer reference

[rustic rain]

yes

[viral sonnet]

benefits of pointers. i was able to make a generic stat system with configurable data types for all stats. all is stored in a big byte array that is directly accessed via code gen'd index. it's as fast as my old system, just more flexible. that wouldn't have been possible without byte pointers that can be reinterpreted to all kind of data types

you can make magic happen with pointers. they are quite fascinating 🙂

[errant hawk]

My favorite part about pointers. Your memory can be whatever, but it will still evaluate to something regardless of the value type.

[viral sonnet]

quite a few ways, UnsafeUtility.AddressOf. CDFE pointer return. writing &variable

[errant hawk]

You can also use IntPtr. I don't really use them tho, so idk what you can do with them

[viral sonnet]

native containers also have getptr methods.

i have only used IntPtr to print the memory hex address 😅

debugged blob data with it to study and confirm the layout

[rustic rain]

I want to make some extension

that will get pointer reference to component

through EntityManager

meaning I could get it globally

especially useful with singletons

[scarlet robin]

Hello, Why the job system uses only the main thread? https://pastebin.com/kZ9cATh8

[rustic rain]

        JobHandle jobHandle = reallyToughParallelJobTransforms.Schedule(transformAccessArray);
        jobHandle.Complete();

because you do this

if you do Complete()

main thread has to wait until job is done

and since it's single thread job

it's faster to do it on mainthread, rather than pushing to other

[scarlet robin]

then what should the code look like to use all cores? ,what to use instead of Complete()?

[rustic rain]

if you want this exact job to use all threads

you need it to be parallel

and then ScheduleParallel

[scarlet robin]

instead of IJobParallelForTransform?

[rustic rain]

instead of Schedule

[scarlet robin]

thanks , i will try

[viral sonnet]

what @rustic rain said. don't complete any jobs that are scheduled. if you really need it to be completed just use run instead.

[rustic rain]

no

he wants parallel job to be done right now

the only way - schedule parallel and then complete

Run() makes it single threaded

[viral sonnet]

yeah, just a future reference if you ever run into this and don't care about parallel

in this case go with scheduleParallel

[scarlet robin]

solved it, because I don't use entities, I can't use ScheduleParallel() and Run() but in my case, the problem was that the objects to be moved were in another object (nested because of object pooling).

[rustic rain]

No ScheduleParallel?

huh

looks like this job extension

doesn't support parallel

no such extension even with entities

[scarlet robin]

ah

[rustic rain]

well

one thing you should probably consider

while this job doesn't support paralleling itself

it supports parallel with other jobs

so if you have multiple jobs

just schedule them all

and then complete them all at once

all of them will run in parallel

[scarlet robin]

thank you, I will most likely use it in the near future, but at the moment I'm happy that it works for me as it is (I've been dealing with this problem for 2 days already)

[solemn hollow]

it is parallel. its in its name. its always parallel and you cant Schedule it on a single thread.
its the same with IJobParallelFor. IJPF can also only be scheduled. IJPF is the old API exactly for that reason. the new way to Schedule a IJPF is to use IJobFor.ScheduleParallel() instead of IJobParallelFor.Schedule()

[rustic rain]

but judging by this guy's profiler

it was single threaded

[solemn hollow]

i see two jobs in parallel?

[rustic rain]

a bit above

[solemn hollow]

that happend cause he called complete right away

[tropic venture]

following this thought ☝️ if i'm already working with DOTS, should I aim to use it for everything in the game?
usually i'd think to use it only for objects that seem "DOTS-appropriate" such as millions of NPC vehicles in megacity, or millions of asteroids, or thousands of units in an RTS etc. and the rest of game stuff would be regular GameObjects

phrased a bit differently - when DOTS matures, will it take over as the main "engine" in unity, completely replacing the current game-object hierarchy one? (once it can fill in for all the needed features of course)

[rustic rain]

DOTS is tech stack

jobs, burst, ECS

key parts

jobs and burst are production ready and can be used anywhere

ECS is not

meanwhile ECS is also the biggest change

because it's totally different way of coding game

if you want to know more - just read manual

https://docs.unity3d.com/Packages/com.unity.entities@0.51/manual/ecs_core.html

[solemn hollow]

its a bit annoying to interop between gameobjects and Entities. For example you have two different Physics engines for each which dont interact by default. in the future i expect it to be much easier to stay in pure ecs to avoid those friction points.

but unity also said they aim to do magic ECS optimizations to gameobject code without your knowledge

[tropic venture]

ah cool

[tropic venture]

wait so - for example if i want to make tens of thousands of trees/bushes using DOTS, but i need the player gameobject to collide withe trees and enemy FOV cones to be occluded by bushes etc. - these interactions between gameobjects and DOTS entities will not be possible? Or possible, just difficult / needing some extra hoops

[solemn hollow]

extra hoops yep

[rustic rain]

ECS physics don't work together with Game Object physics

[safe lintel]

you should really pick one physics system(dots or physx) and base things from there, imo any interop between two would be a lot of pain

[tropic venture]

hm

[solemn hollow]

and those extra hoops are as convoluted as just making the player an entity too lol

[rustic rain]

yep

and make whole game entities

kek

that's the way

[tropic venture]

maybe for that reason alone i'd want to implement almost all game content in DOTS instead of gameobjects then....

[solemn hollow]

atm that causes a lot less headache than hybrid

writing back from entity code to gameobject code only in one direction is easy though

so you can still use some monobehaviours like the animator or sth

[safe lintel]

thats the trap I feel into. there's still no official equivalent to mecanim if you need elaborate animation setups though so expect to require some somewhat convoluted workarounds there

[solemn hollow]

but that ofc requires a gameobject as a representation for the entity

[tropic venture]

what about using assets tho? like the Starter Assets FPS controller, or Low Poly Shooter Pack etc. are those a no-go for putting them to use in DOTS-land?

[solemn hollow]

they are a no go anyways lol

[tropic venture]

heh

[safe lintel]

anything that is monobehaviour derrived will need to be rewritten, if you need those things you should probably just stick without ecs

[tropic venture]

hm..

so committing to ECS is committing to writing your own everything, not being able to make use of any non-ECS-based assets

[rustic rain]

yep

[tropic venture]

good to know

[solemn hollow]

you can still copy the code over to the ecs side though without much changes

[safe lintel]

@tropic venture if I were you I'd wait for 1.0 which should bring further clarity to the problem you are describing, and at least at that point there should be far more documentation and tutorials supporting it

[tropic venture]

ah, so there is a "known path forward" to move things from gameobject land to ECS land, i guess it's just tedious, needs manual work on every code file to look through and rework for ECS

[tropic venture]

is there an ETA on that?

[solemn hollow]

and if you are prototyping youd probably have way better iterationtimes with gameobjects atm anyways

[safe lintel]

supposedly preview should be any time now and release end of year or early next

[tropic venture]

nice ok

i do want to start prototyping

[solemn hollow]

yes you can more or less just split every monobehaviour into systems and Data. afterwards you care about making it burstable. then parallelizable

[solemn hollow]

your prototype most likely wont need the full ecs performance anyways

[tropic venture]

well;

[tropic venture]

...but my project is trying to be kind of a benchmark for mobile, like a "Crysis" for mobile 😛 i am aiming to push tech limits to the edge of what's possible, so i'm wondering if i'm wasting time prototyping in gameobject-land if it wont get me where i can get with ECS...

in my case i'm not really aiming for graphic realism too much, but more like graphic density, loads upon loads of scene content

detail

[rustic rain]

well, idk about graphics

but loading a lot of data is perfect with dots

subscenes allow for very fast load/unloads of parts of world

[tropic venture]

even on mobile?

[rustic rain]

any supported platforms

[safe lintel]

ecs wont speed up rendering, you could achieve similar results using DrawMeshInstancedIndirect

[tropic venture]

right, i'm eying that too

[rustic rain]

ahem

culling system

bursted one

[tropic venture]

altho i'm also hoping to support medium-lower end devices not supporting OpenGLES 3.1... where available - use instancing, where not - will need to reduce object count maybe and see what's the max i can achieve with all other optimization methods

[rustic rain]

why not just use Vulkan?

[gray tusk]

Does anyone know why this fails?

var attack = commandBuffer.Instantiate(attackSource.attackPrefab);
commandBuffer.SetComponent(attack, translation);
var attackComp = new AttackInstanceComp
{
  damage = 1,
};
commandBuffer.AddComponent<AttackInstanceComp>(attack);
// Fine until here
commandBuffer.SetComponent(attack, attackComp); // A native collection has not been disposed

[tropic venture]

so i'm worried maybe i'll prototype in gameobjects and code everything around that, maybe come up with some smart culling system in gameobject-based scripts, maybe billboarding etc. and i'll be able to achieve <some number> like i dunno, 15000 vegetation in scene even on meduim-low devices. But then needing a full rewrite for ECS - now i could achieve 25000, significant difference, prototyping in gameobjects was a waste of time...

but maybe i'm wrong about this

[tropic venture]

doesnt unity choose that automatically or something?

[safe lintel]

why not just commandBuffer.AddComponent(attack, attackComp);

[rustic rain]

you select it in player options

[gray tusk]

That fails as well

Same thing

[safe lintel]

whats the exact error message?

[gray tusk]

A Native Collection has not been disposed, resulting in a memory leak. Enable Full StackTraces to get more details.

But I'm not entirely certain it's related

[tropic venture]

is it literally just a drop down select and done? or needs related changes in graphics / coding / shaders / other stuff

[safe lintel]

well enable full traces and find out 🙂

[rustic rain]

it's that simple unless you use API specific code afaik

or maybe platform specific

[tropic venture]

cool then, why would anyone not use it... should be enabled by default or something

[gray tusk]

A Native Collection has not been disposed, resulting in a memory leak. Allocated from:
Unity.Entities.EntityCommandBuffer:.ctor(Allocator) (at Library\PackageCache\com.unity.entities@0.51.0-preview.32\Unity.Entities\EntityCommandBuffer.cs:1330)

😅

[safe lintel]

whats the full code for where you are using that commandbuffer? sounds like you arent disposing it

@tropic venture For Android and Linux, you should use Vulkan. However, be aware that the Vulkan drivers on many older Android devices are in a bad shape and will never be upgraded. This limits the platform coverage Hybrid Renderer can currently offer on Android devices. from the hybrid renderer docs

[gray tusk]

Looks like it

[tropic venture]

you make it sound like it's a magic Make My Game Super Fast Automagically button 😆 is it?

[rustic rain]

doubt

[gray tusk]

Disposing of it removes the exception, but it's still not spawning the entity

If I remove the SetComponent line, all's fine

[tropic venture]

is this something i can programmatically account for? or it requires a user setting/option "switch off vulkan renderer if things look horrible on your device"

[safe lintel]

@tropic venture my knowledge of mobile is basically what the docs say, so not sure

[tropic venture]

is it even possible to make choices about unity's renderer at runtime on mobile?..

[gray tusk]

I don't think so

[tropic venture]

i'd have to release a separate APK "try this version if the game looks weird on your device" or something like that

[tropic venture]

any other thoughts ☝️ re. prototyping vs. waiting for 1.0 to release?

[safe lintel]

i mean you should just do some simple prototypes regardless, maybe you will hate the workflow

[tropic venture]

hating the workflow can't possibly be a thing 😆 either you want to make a game, or you dont.... no?

[safe lintel]

some people fight conversion tooth and nail 😅

[tropic venture]

whadyou mean conversion, gameobjects to ECS?

[safe lintel]

yep

[tropic venture]

so is there any value in prototyping now? or is there a strong case to only prototype using ECS, or maybe even wait a while for 1.0 to release

[safe lintel]

by prototyping I mean start by opening the samples and make some very simple edits to familiarize yourself with the workflow, see if it is right for you. that alone should tell you if you want to attempt things further. something that is very simple where an player controlled entity moves, via the inputsystem and can play a sound from an audiosource will give you a feel for a variety of situations and how ecs works in context of other unity systems

[tropic venture]

i see

[rustic rain]

fmod is burst compatible kek

for sounds

[safe lintel]

@rustic rain personally im using myrri but split off from the latios framework

[viral sonnet]

how are you setting up the command buffer?

[solemn hollow]

did you call AddJobHandleForProducer on the ecbs?

[viral sonnet]

i'm pretty sure he doesn't use ECBs

at least it sounds that way because otherwise you don't need to dispose

so it's probably not played back

[solemn hollow]

true

[gray tusk]

https://forum.unity.com/threads/entitycommandbuffer-instantiate-fails-when-assigning-data-afterwards.1328142/

[gray tusk]

It is

[solemn hollow]

var commandBuffer = new EntityCommandBuffer(Allocator.TempJob); thats not how you create a commandbuffer

[gray tusk]

It is, according to the docs -.-

And it works just fine as long as I don't assign the other entity

[solemn hollow]

var ecb = ecbs.CreateCommandBuffer().AsParallelWriter();

[gray tusk]

https://docs.unity3d.com/Packages/com.unity.entities@0.51/manual/entity_command_buffer.html

[viral sonnet]

no, it's fine but i think you need a temp allocator. at least that's what i'm using

[gray tusk]

That's for multithreaded jobs

[viral sonnet]

not necessarily. it's when you use a entityCommandBufferSystem which does automatic playback

[gray tusk]

Well, I play it back manually. Not the issue.

[solemn hollow]

is this new? i wasnt aware i can just new() a commandbuffer

[viral sonnet]

been there for quite a while

it's for manual playback. mostly not needed but it has its use cases

[safe lintel]

@gray tusk does the prefab already have a NewComponentType on it? im not really sure why its failing on the two lines you commented out, code looks fine there

[gray tusk]

Nope

var attackComp = new AttackInstanceComp
{
    damage = 1,
    distancePercentage = 0,
    Source = entity,
    Target = _headQuarters // It works if I comment out this line
};

A bit less abstract. _headQuarters is obtained succesfully in OnStartRunning, and I can access it's component data.

[solemn hollow]

@gray tusk im out of ideas but could you try to replace otherEntity = query.GetSingletonEntity();
with otherEntity = GetSingletonEntity<OtherComponent>()

based on your forum post

[safe lintel]

yeah not sure what the outcome of a success or fail is but in oncreate you can use RequireSingletonForUpdate<T> and then use GetSingletonEntity<T> in onupdate. if you use OnStartRunning and your entity is destroyed it, your system may not know about it as it happened some time after the system started running

or use if(!HasSingletonEntity<T>()) return; if you dont want to require the singleton for update

[coarse turtle]

https://blog.unity.com/technology/pick-up-these-helpful-tips-on-advanced-profiling this feels relevant to discussion earlier with gfx backends on mobile btw 👀

[rustic rain]

(0,0): Burst error BC1091: External and internal calls are not allowed inside static constructors: Interop.BCrypt.BCryptGenRandom(System.IntPtr hAlgorithm, byte* pbBuffer, int cbBuffer, int dwFlags)
Anyone else getting this error?

in it's stack it literally mentions every single job in my project

I cleared entities cache and also fully removed Library folder

error still persist

[rotund token]

I've seen this before

Did you solve it?

[rustic rain]

nope

[rotund token]

Had this discussion with a burst dev before let me find the convo

basically GetHashCode on anything will cause that BCrypt static constructor to need to be run in Burst, which uses managed data

My cause was I used HashCode.Combine

In GetHashCode

Which wasn't supported

And this type was used as a key in a native hash map

[rustic rain]

I have non of that

weird

[rotund token]

Was a pain to track down

Had to disable my libraries 1 as a time

Then my systems

Until I tracked it down to a hashmap.add line

Thankfully as my systems are all decoupled this didn't take too long

[rotund token]

But yeah tldr, you're likely using something managed in a static constructor

[stiff mesa]

Hey everyone! Quick question: how do I handle components with BlobStrings? Can I use Allocator.Persistent on them? Do I have to track them and dispose manually?

[rotund token]

Blobs should live in blob asset store

You shouldn't have to manage this yourself they're tied to the subscene

[stiff mesa]

Thank you!

[rotund token]

If you are creating them outside of your though or managing your own blob asset store then you will have to dispose

[stiff mesa]

I'm trying to figure out how to use this little asset:
https://gitlab.com/tertle/com.bovinelabs.event
I'm doing a level manager with levels stored in Addressables and referenced through strings. I want to make an event system that triggers the level change when I send the event with the specific level name and I'm wondering if it would be ok if I attach BlobString to the event component?

[rotund token]

That's my asset 😅

should be fine

[pliant pike]

I don't anyone here would have any idea why when using an analog stick I seem to get a sort a lag where the same inputs persist over several frames

I think that is the issue its hard to debug to be honest, the inputs just don't feel as snappy as they should when moving an object directly with an analog stick, it seems fine with keyboard though

I'll post the code its really basic, but I feel like its a system or a value in a variable is not getting reset or something

[viral sonnet]

dead zone too high? acceleration too low?

wireless? it has some input lag but shouldn't be that noticable

[pliant pike]

float2 currentmovpos = new float2();
currentmovpos = playInput.DirectionInput * CursorMoveSpeed * Time.DeltaTime;
if (math.length(playInput.DirectionInput) > 0.2)        
{            
float3 tempcurr = new float3(currentmovpos.x, 0, currentmovpos.y);          
CursorSquare.Value += tempcurr;
CursorSquareQuery.SetSingleton<Translation>(CursorSquare); 
        }```

[viral sonnet]

0.2 seems quite high. you tested it with a lower value?

[pliant pike]

yeah I tested it with 0.01 I don't think that is the issue

its weird its like it occasionally goes diagonally or any direction way longer than it should sometimes

[viral sonnet]

old or new input system

[pliant pike]

new input system

[viral sonnet]

well, i think the only way you can debug this is to print out frameCount and the received value from the input. from that it should be clear how it will react

and the only possible settings are here

[pliant pike]

that's exactly what I need thanks Enzi 👍

[viral sonnet]

np, hope you find the issue 🙂

[viral sonnet]

it's somewhat mad to do it like this but also a very nice way to test my asset. the game where i'm using my asset now works very different in a lot of places. at first i thought, oh boy, this won't work out well but it turns out. DOD lends itself really nice to open custom paths. some ifs, some constants, some partials and code can be modified like crazy. quite happy how this is working out and not what i would have expected.

[solemn hollow]

hmm i guess i missunderstood reinterpret :S

var packedEnablerCounts = cellEnablerCounts.AsNativeArray().Reinterpret<int4>(sizeof(int));
                                bool allValid = math.all(packedEnablerCounts[compatiblePattern] > new int4(0, 0, 0, 0));

                                if (cellEnablerCounts[compatiblePattern].Value > 0 &&
                                    cellEnablerCounts[compatiblePattern + 1].Value > 0 &&
                                    cellEnablerCounts[compatiblePattern + 2].Value > 0 &&
                                    cellEnablerCounts[compatiblePattern + 3].Value > 0)
                                {
                                    if (!allValid)
                                    {
                                        Debug.LogError("allvalid is wrong");
                                    }
                                }```

this always throws the error.

i thought the evaluation with math.all does exactly the same as checking each entry individually. what am i doing wrong here?

[rotund token]

hmm

your index is wrong

you're using the same index for the reinterpreted and original arrays

if compatiblePattern == 1
then for
packedEnablerCounts[compatiblePattern]
you're looking at elements
4,5,6,7
for cellEnablerCounts[compatiblePattern]
you're looking at elements
1,2,3,4

length of packedEnablerCounts is /4 of the size of cellEnablerCounts

[solemn hollow]

ah damn. ofc. okay then everything works as expected. thats bad news cause that means my bug is somewhere else

[viral sonnet]

Reinterpret<int4>(sizeof(int)) i don't get it. why int4 and then int?

never noticed before there's a parameter for Reinterpret 😄

in case the base type is int. okay. is it actually needed or does it help in any way?

[solemn hollow]

its needed. getting runtime error if i leave it out

System.InvalidOperationException: Types {0} and {1} are different sizes - direct reinterpretation is not possible. If this is what you intended, use Reinterpret(<type size>)

[viral sonnet]

ah i see. private static void CheckReinterpretSize<U>() where U : struct { if (UnsafeUtility.SizeOf<T>() != UnsafeUtility.SizeOf<U>()) throw new InvalidOperationException(string.Format("Types {0} and {1} are different sizes - direct reinterpretation is not possible. If this is what you intended, use Reinterpret(<type size>)", (object) typeof (T), (object) typeof (U))); } never ran into it because my types always have the same size

[rotund token]

^ you figured it out but yeah it's for different size reinterprets

[rustic rain]

Any idea whether chunks are fully persistent?

I want to save pointer to chunk

[rotund token]

i believe so

they can change though

[rustic rain]

that's what I'm trying to check

    public unsafe struct ComponentReference<T> where T : struct, IComponentData
    {
        internal IntPtr ptr;
        internal uint orderVersion;

        public T Read
        {
            get
            {
                UnsafeUtility.CopyPtrToStructure((void*)ptr, out T output);
                return output;
            }
        }

Kek

[rotund token]

if you have a chunk with XYZ components and you do a query A AddComponent, it will change the chunk to be XYZA

[rustic rain]

I don't want to enforce EntityManager reference

but I want to keep checks for order version

[rotund token]

oh i store chunks

for order changes

[rustic rain]

chunks?

what is size of chunk type?

[rotund token]

hmm?

i just store the actual ArchetypeChunk

[rustic rain]

        internal ArchetypeChunk(Chunk* chunk, EntityComponentStore* entityComponentStore)
        {
            m_Chunk = chunk;
            m_EntityComponentStore = entityComponentStore;
            m_BatchEntityCount = kUseChunkCount;
            m_BatchStartEntityIndex = 0;
        }

I see

[rotund token]

why bother with ptrs

[rustic rain]

well, I kind of want to store it on entities at some point

[rotund token]

    {
        internal UnsafeParallelHashSet<ArchetypeChunk> Targets;
    }```

i do that

well not on individual entities

i store it as a chunk component

[errant hawk]

How can I create an entity from a GameObject that will be used as a reference to a child entity? The conversion workflow docs only mentioned prefabs, nothing about getting an existing GameObject.

Example:

public void Convert (Entity entity, EntityManager dstManager, GameObjectConversionSystem conversionSystem) {
    DynamicBuffer<WheelBase> buffer = dstManager.AddBuffer<WheelBase>(entity);

        buffer.ResizeUninitialized(wheels.Length);

    for (int i = 0; i < wheels.Length; i++) {
        //WheelBase wheel = new WheelBase(conversionSystem.???);

        //buffer.Add(wheel);
    }
}

[rotund token]

GetExistingEntity(object)

[errant hawk]

wow that... is a badly named method

should be something like InterpretObjectAsEntity or something

regardless, thanks

[rustic rain]

hmm

How to get global system version?

Is that kind of data that can only be obtained from EntityManager?

[rotund token]

why would you want it

i guess is my first question

[rustic rain]

to SetChangeVersion

                var indexInTypeArray = ChunkDataUtility.GetIndexInTypeArray(chunk->Archetype, typeIndex);
                chunk->SetChangeVersion(indexInTypeArray, );

    public unsafe struct ComponentReference<T> where T : struct, IComponentData
    {
        internal IntPtr ptr;
        internal uint orderVersion;
        internal Chunk* chunk;

        internal int typeIndex;

        public T Read
        {
            get
            {
                if (chunk->GetOrderVersion() != orderVersion)
                {
                    Debug.LogError("Order changed, component is invalid");
                }

                UnsafeUtility.CopyPtrToStructure((void*)ptr, out T output);
                return output;
            }
            set
            {
                var indexInTypeArray = ChunkDataUtility.GetIndexInTypeArray(chunk->Archetype, typeIndex);
                chunk->SetChangeVersion(indexInTypeArray, );

               
            }
        }

Just figuring out stuff hehe

[rotund token]

            where T : unmanaged, IComponentData
        {
            SetChangeFilterCheckWriteAndThrow(handle);

            var typeIndexInArchetype = ChunkDataUtility.GetIndexInTypeArray(chunk.m_Chunk->Archetype, handle.m_TypeIndex);
            if (typeIndexInArchetype == -1)
            {
                return;
            }

            // This should (=S) be thread safe int writes are atomic in c#
            chunk.m_Chunk->SetChangeVersion(typeIndexInArchetype, handle.GlobalSystemVersion);
        }```

ComponentTypeHandle, ComponentFromEntity etc

have this stored

[rustic rain]

uh huh

so global version must be obtained beforehand

welp, that sucks

[errant hawk]

Im not seeing any method called GetExistingEntity from the GameObjectConversionSystem or even the EntityManager

[rustic rain]

but it's fine

[rotund token]

sorry, GetPrimaryEntity

[errant hawk]

kek that is what I was using, but I was getting nullref so I thought I was using the wrong one

[rotund token]

every system update adds +1 to the globalsystemversion

[errant hawk]

guess I missed something else where

[rotund token]

you have to get it before you return from a system

you can't get it in a job, it's too late

(at issue)

[rustic rain]

well I know

but I assume

getting this value

doesn't cause any sync points

so it's fine

[rotund token]

nope

systems just have a field called this.GlobalSystemVersion

[rustic rain]

my whole idea here is just

[errant hawk]

@rustic rain what is it you're tying to do anyways

[rustic rain]

to have a way to set data

to static entities

also read

without EntityManager

or sync points

allthough writing seems very unsafe

[rotund token]

it only causes a sync point if it needs to

[rustic rain]

but still requires EntityManager

[rotund token]

so?

[rustic rain]

no EM in jobs

[rotund token]

you know you have componentfromentity

right

[rustic rain]

It's good normally

but when you have a lot of entities with only 1 component

that never change

and you know you only read them and sooooometimes write to them

CDFE feels like a huge overkill

[rotund token]

why

cdfe is super fast

[rustic rain]

this is even faster, kek

[rotund token]

sure

until entities moves your entity to a different chunk

[rustic rain]

yeah, and that is not meant to happen

[rotund token]

why?

[rustic rain]

that's why I put order version check

[rotund token]

ok so now you're checking

[rustic rain]

because this is meant for static entities

[rotund token]

i don't think it's faster anymore

[rustic rain]

similiar to Time entity

[rotund token]

you're converting 50ns into 20ns

this like, next level over optimization here

[rustic rain]

no, I'm actually reducing total amount of code

because now I don't need to get cdfe

store field for it

[errant hawk]

but that's over 50% !

[rustic rain]

also I can store it on entity

I think...

[errant hawk]

CDFE's are a bit cumbersome to use

[rotund token]

and when entities adds their chunk garbage collection in 1.0

and things start moving

and you have to rewrite it all again

[rustic rain]

yes😅

[errant hawk]

That's going to be interesting... do you think it preform a structural change every time it runs

[rotund token]

i can not confirm this exists

but it was mentioned a while ago on forums

(but so were a lot of features we want that probably aren't coming =D)

[errant hawk]

hmm... interesting ntl

Im sure a lot of well-needed additions are tucked away from the userbase that will only be disclosed on 1.0 release day

[rotund token]

There is nothing you can do about it right now, but fragmentation & tiny / mega chunk support is definately something we will look at in the coming months.
2018 😄

i have heard defragmentation a few times

[errant hawk]

I mean, ECS as of right now is still pretty good at the core. A few things could change yes, but it is functional and doesn't seem to have any blindly obvious major issues as of right now

[rustic rain]

there is one

some people noticed here

that if you create millions entities

whole system slows down

[errant hawk]

I would like to see more control over how the memory is managed by ECS

[rustic rain]

noticable

[rotund token]

not really sure what you expect ^_^'

[errant hawk]

I mean, that is to be "expected" with million of entities

[rotund token]

create millions of gameobjects and see how well you go

[rustic rain]

no, I mean

other systems

that don't process those entities

they slow down too

[rotund token]

yes

it still has to look through every chunk

to fill a query

systems don't magically know what chunks it uses

while there is some caching now

[errant hawk]

memory lookup times gets slower as more entities/chunks are added

[rotund token]

soon as you start spreading your memory around things are going to slow down

^

[errant hawk]

It's likely why they will add a GC for ECS so they can pack similar entities/chunks together for improved memory lookup times

[rotund token]

fragmentation has been a long term concern

used to cause huge issues on our console release

[errant hawk]

yeah

[rotund token]

1 hour later fps would be down like 25%

due to significant chunk increases

[rustic rain]

that's why you switched to static archetypes?

[errant hawk]

oof that is pretty bad pref drop just because of fragmentation

[rotund token]

oh we don't have static archetypes at work

but we did significantly reduce it

i just like the concept of static archetypes

makes everything easier

network serialization, saving

and things just work faster

i don't need to worry about component checking if something has been removed

[errant hawk]

static as in it's assigned to a static field or what?

[rotund token]

as in i don't do any add/remove components

[rustic rain]

unchangable archetype

[errant hawk]

ah yeah

[rotund token]

i also just like the challenge and i think it can make some beautiful crazy scaling code

i do not think most devs should focus on this

it makes a lot of things more complicated

[errant hawk]

entity per function of a complex system, instead of a component

Idk, I would have to figure out some other use cases

[rotund token]

i should say i don't have completely static archetypes

and i do occasionally use tag components for long state changes

but i require the archetype of an entity to be rebuildable from the saved original archetype

i.e.
component with value = 1
would addtag3

therefore i load an entity with value = 1

it would automatically readd the tag

[errant hawk]

no need to add a tag then, right?

[rotund token]

tag removes it from a system

or adds it to a system

as mentioned above, long state changes

minimum seconds, ideally minutes+

[errant hawk]

oh yeah I see, you mean long

this would be used for timers or other similar events?

[rotund token]

only timers i have atm are actually static

[rustic rain]

if I do GetEntityQuery in several systems, will they all share same query?

[errant hawk]

for (int i = 0; i < wheels.Length; i++) {
    if (wheels[i] == null) continue;

    Entity e = conversionSystem.TryGetPrimaryEntity(wheels[i]);

    if (e == Entity.Null) continue;

    WheelBase wheel = new WheelBase(e);

    buffer.Add(wheel);
}```

I keep getting the error: An Entity index is larger than the capacity of the EntityManager. This means the entity was created by a different world or the entity.Index got corrupted or incorrectly assigned and it may not be used on this EntityManager.

Do I have to register it with the EntityManager somehow?

[rustic rain]

I don't quite get what you are doing

[errant hawk]

adding an Entity to a buffer when converting from a GO

it's a child entity

[rustic rain]

why are you making a Entity.Null comparison

instead of just using return value

of TryGet

[errant hawk]

idk I figured it may have been part of the issue

it doesn't even work, just covering all bounds atm

[rustic rain]

What is buffer?

[errant hawk]

DynamicBuffer

#archived-dots message

[rustic rain]

weird

are you converting in subscene?

[errant hawk]

yeah

wait what

It works only when it's outside of a subscene

[rustic rain]

when it's outside of subscene

you create it directly in target world

so entities don't get remapped

[errant hawk]

So it won't work in subscenes then since the world is different?

[rustic rain]

it will

if you use supported container for entity

and buffer is

[errant hawk]

which is what im using

and it throws that error

[rustic rain]

        protected override void OnUpdate()
        {
            Entities.ForEach((DefConverter converter) =>
            {
                var converterEntity = CreateAdditionalEntity(converter);
                DstEntityManager.AddComponentData(converterEntity, new DefType { value = converter.defType });
                var buf = DstEntityManager.AddBuffer<RegisterDefReference>(converterEntity);
                foreach (var converterPrefab in converter.prefabs)
                {
                    var def = converterPrefab.GetComponent<DefAuthoring>();
                    Convert(buf, def);
                }
            });
        }

        private void Convert(DynamicBuffer<RegisterDefReference> buf, DefAuthoring def)
        {
            var e = GetPrimaryEntity(def);
            buf.Add(new RegisterDefReference { value = e });
        }

here's what I do

quite same thing

add buffer, get primary entity, add

wait a second

you resize

uninitialized

thus why value might be corrupt

try not resizing

[errant hawk]

I only did that because the sourcegen did it

lmao ok yeah that was the issue

It didn't like the invalid buffer elements ig

[rustic rain]

I assume it tried to remap entity with weird numbers

and failed

[errant hawk]

yeah

anyhow, thanks

[rustic rain]

Do you guys have any better ideas?
I have this unique system that makes async read of current cursor mouseover

and in order to trigger it from different implementations

I made this

    public interface ISelectionRequest
    {
        public ComponentType[] Desc();
    }

so any system that wants to read selection result from that system

just implements this

and then I create update requirment for that selection system from that interface

[errant hawk]

why not use events and pass in cursor data any time there's a selection?

[rustic rain]

burst

my current async implementation allows for burst

in initialization system group new input system and ui elements create components of input

right after simulation ECB, I run selection system

which writes result to singleton component

and then all other systems can read it in bursted code

[errant hawk]

ok but your ISelectionRequest contains a managed array

[rustic rain]

it's only read on OnCreate

        public static EntityQueryDesc[] GetAllRequests()
        {
            var list = new List<EntityQueryDesc>();
            foreach (var request in TypeUtility.InstantiateAllOfType<ISelectionRequest>())
            {
                list.Add(new EntityQueryDesc() { All = request.Desc() });
            }

            return list.ToArray();
        }

RequireForUpdate(GetEntityQuery(SelectionUtils.GetAllRequests()));

like this

And then in managed system, I make a selection, and set component data to selection entity

[devout prairie]

@rustic rain you recently mentioned a component to disable render..

RenderDisable or something?

[rustic rain]

DisableRendering

[devout prairie]

Aha! Thanks

was just looking for a quick shortcut to on/off some elements

[rustic rain]

state.RequiredEntityQueries
hmm, if I get this list

maybe that's the way

I want to try and make HashSet out of all queries in all systems that request selection

and then add them to SelectionSystem

for update requirments

[devout prairie]

any idea if RenderBounds or WorldRenderBounds are axis aligned to local orientation or of they are both AABB to the world axis?

just trying to figure out a way to get a mesh's approximate bounds

[rustic rain]

I use WorldRenderBounds to get bounds

I use it to dynamically scale this selection overlay

tbh, you could just select entity in inspector to manually see what they represent kek

[devout prairie]

hehe yeah the objects are currently placed in default prefab orientation so world orientated

i'll just rotate them and check, thought i may as well ask first 😛

[devout prairie]

so yeah i think RenderBounds represents the locally aligned bounding box, whereas WorldRenderBounds is the world axis aligned AABB

[pliant pike]

I have a weird System issue I think, it seems like a method call from PlayerInputActions.IPlayerControlsActions is only being called every few frames

the System is update every frame but this method for example is not

public void OnDirectionalControls(InputAction.CallbackContext context)
{
        inputdirection = context.ReadValue<Vector2>();
        //Debug.Log("The actual input is " + inputdirection);    
}```

maybe that's normal behaviour although I'm sure it is receiving input every frame so I feel like perhaps that could be the cause of my problem

[rustic rain]

hmmm

Is there a way to get list of all managed components of type?

ok, it is possible

but it's super slow 🤔

[viral sonnet]

seems like doing thispublic partial class HandleCombatEventsSystem { private partial void CustomHandleCombatEventsUpdate() { breaks codegen with error CS0579: Duplicate 'global::System.Runtime.CompilerServices.CompilerGenerated' attribute

well there's more to it than the code i posted but calling a partial update method from the main update doesn't seem possible

[rustic rain]

if I dispose native array of unsafe lists

unsafe lists are untouched, right?

[rotund token]

yes

[rustic rain]

hmm

so anyone figured ways to make animations?

not skinned mesh ones

just general, for translations/rotations and etc?

[pliant pike]

I have an idea for how I might implement animations if I get around to it 🤔

I'm not sure if it will work that well but I will try a stop motion type animation system

where I swap mesh's out in sequences

I already have a script that can extract single mesh's from animations in Unity

[rustic rain]

tbh, I figured my game doesn't need animations xD

but not certain

currently I have 2 things in game that potentially require animations:
space ship making jump to other system
space ship landing on planet

first thing is quite solvable without animations

but second...

[pliant pike]

yeah depending on the game you could get away with minimal animations

spaceship landing should be pretty easy without too

[rustic rain]

Can't figure what would be a better option

1 constraint - while landing ship must be vulnerable for combat

[pliant pike]

tweens could work well for that

[rustic rain]

yeah, hardcoded animation

[pliant pike]

Ok, I've definitely found an issue with the way I was receiving inputs

[devout prairie]

i'm baking clips into a blob asset ie the pos/rot values

[pliant pike]

notice how the square flies off in random direction

[devout prairie]

then i have a system which increments the current frame on each animated thing, and applies the pos/rot

[devout prairie]

i think you could do this with code..

[pliant pike]

I don't know if its a bug or what but it seems using this way public partial class PlayerInputStuffSystem : SystemBase, PlayerInputActions.IPlayerControlsActions { to collect actions from the input system is not a good way

and now it means I'm going to have to write a new system to process inputs

but at least I solved it

[rustic rain]

input can be solved with ISystem

actually

[pliant pike]

really? I havent used ISystems yet I kind of want to wait until they are fully implemented

[rustic rain]

well, the implementation is simple. you use ISystem to reset entity that stores all input tags xD

 var inputActionTypes = TypeUtility.GetAllAssignableFrom(typeof(IEventAction)).ToList();

            foreach (var assetActionMap in controls.asset.actionMaps)
            {
                foreach (var inputAction in assetActionMap.actions)
                {
                    var name = inputAction.name.ToLower();
                    foreach (var type in inputActionTypes)
                    {
                        if (!type.Name.ToLower().Equals(name)) continue;
                        if (allStarted.TryGetValue(name, out var started))
                        {
                            inputAction.started += _ => em.AddComponent(inputEventEntity, started);
                        }

                        if (allPerformed.TryGetValue(name, out var performed))
                        {
                            inputAction.performed += _ => em.AddComponent(inputEventEntity, performed);
                        }

                        if (allCanceled.TryGetValue(name, out var canceled))
                        {
                            inputAction.canceled += _ => em.AddComponent(inputEventEntity, canceled);
                        }

                        if (allPressed.TryGetValue(name, out var pressed))
                        {
                            inputAction.started += _ => em.AddComponent(pressedEventEntity, pressed);
                            inputAction.canceled += _ => em.RemoveComponent(pressedEventEntity, pressed);
                        }

                        inputActionTypes.Remove(type);
                        break;
                    }
                }
            }

I do it this way: store all interfaced components

get their generics

with

Started<Event>
Performed<Event>
Canceled<Event>

[pliant pike]

that seems a bit complicated

[rustic rain]

sometimes I add
Pressed<Event>

[rustic rain]

yeah, but then it's so liberating

cause now you have global input values

as components

and you can create trigger systems

that react on certain queries

and all of that is burstable

[pliant pike]

I mean I was just setting values in a componentdata and that worked fine for me

[rustic rain]

well, that works too

but I enjoy having systems with reaction through queries

[pliant pike]

bools for buttons and float2s for mouse/gamepad etc

[rustic rain]

        public void OnCreate(ref SystemState state)
        {
            var em = state.EntityManager;

            _cameraEntity = em.GetOrCreateSingletonEntity<SpaceCameraTag>();
            state.RequireSingletonForUpdate<Pressed<FocusCameraOnPlayer>>();

            _playerQuery = state.GetSingletonEntityQueryInternal(typeof(PlayerTag));
        }

        [BurstCompile]
        public void OnStartRunning(ref SystemState state) { _player = _playerQuery.GetSingletonEntity(); }

        [BurstCompile]
        public void OnUpdate(ref SystemState state)
        {
            var em = state.EntityManager;

            var pos = em.GetComponentData<LocalToWorld>(_player).Position;
            var sysID = em.GetComponentData<StarNormal>(_player).systemID;

            em.SetCameraPosition(_cameraEntity, pos);

So then whole system that makes pressing button follow player looks like this

[pliant pike]

that's interesting 🤔

[rustic rain]

it's probably less fast then simply checking component

but it feels good to use xD

also it supports modding really well

cause modders can add their own input components and any way they can be triggered

[pliant pike]

yeah exactly, and either way I bet its faster than a monobehaviour

[rustic rain]

I might actually make it a library

[pliant pike]

DOTS just gives way more flexibility

[rustic rain]

hmmmm

what if I instead make those chunk components

I wodner if queries will still work

the reason - I want to limit amount of chunks for input entity to 1

[pliant pike]

I have no idea, I haven't messed with chunks much yet, haven't seen any need to

[rustic rain]

if they can be used the same way

by other systems

I might actually make it way faster

allthough

it's probably

0.01ms vs 0.008ms xD

welp

turns out it's very simple

https://docs.unity3d.com/Packages/com.unity.entities@0.51/manual/ecs_chunk_component.html

and looks like this is what I'm looking for

[pliant pike]

interesting

[errant hawk]

Burst compiled function pointers are slower than normal method calls if it is not a complex and expensive process. I messed around with them in the past, doing something simple, and performance was always 10x slower for such a small workload. Things may have changed tho, so you can only truly know if you benchmark it.

[rustic rain]

yeah, should probably benchmark it

huh, looks like Burst already supports struct in function pointers

[pliant pike]

what do you guys think would be the best place to post about that problem I have above, the DOTS forum or the Input Forum it feels like its an issue with both

[rotund token]

supports struct if passed by pointer

also, i find your input solution fascinating

i can't tell if i love it or hate it

[pliant pike]

yeah 😆

[rustic rain]

I have a feeling input forum is way more dead than dots

[rotund token]

from a purely data/reactive design it's quite nice

what is your issue calabi?

i feel like the type of issue would determine where it's posted

[rotund token]

this is actually how the Unity physics guys made their character controller

[pliant pike]

I get a sort of lag issue where inputs are sustained if you check the gif I posted above using a certain method to get inputs

like if use the methods directly in the classes from public partial class PlayerInputStuffSystem : SystemBase, PlayerInputActions.IPlayerControlsActions { that

then it doesn't update every frame

I presume its using fixed update, but I can't figure out how to change that, and I have changed the input settings in the project setting to dynamic update

[rotund token]

https://github.com/Unity-Technologies/EntityComponentSystemSamples/blob/master/PhysicsSamples/Assets/Common/Scripts/DemoInputGatheringSystem.cs

that's how they do it here

i think we need a bit more code than your interface

to see what's going on

they handle it similar to me, store the value in the system

then write to a singleton in update

i don't use the interfaces though, i prefer not hard coding my link to input

and pass in inputaction instead via configs

but effectively the same thing

[pliant pike]

yeah that is close to how I'm using it

I'll change my code to that example anyway, I can't remember where I got the example for using it the way I did

[rotund token]

[AlwaysUpdateSystem]
[UpdateInGroup(typeof(InitializationSystemGroup))]

are you doing always update?

[pliant pike]

yep

the update method is called every frame its just a method like the below is not called every frame

public void OnCursorPosition(InputAction.CallbackContext context)
    {
        inputMove = context.ReadValue<Vector2>();```

I'm not sure if its a bug or just my implementation

[viral sonnet]

hm, yesterday i wrote that this is not working so i have reduced it down and this version is indeed working. 🙂

one method to either support mods or support modules for asset development

still interested if there's a better way of doing this

most of the code is not important. what this does is calling another implementation of an update method

where i still struggle a bit is how to set this up with asmdefs. main reason, the implementation of a partial method has to be in the same asmdef. so an empty implementation has to be provided or code file that is meant to be modified. but moving this file out of the asmdef directories ends in the implementation not being there and compile errors. then i tried setting up a custom asmdef but then you have circular references.

found out that calling GetSingleton leads to these weird codegen errors

their codegen is too dumb 😄 Unity.Entities.EntityQuery __query_0; protected override void OnCreateForCompiler() { base.OnCreateForCompiler(); __query_0 = GetEntityQuery(new Unity.Entities.EntityQueryDesc{All = new Unity.Entities.ComponentType[]{ComponentType.ReadOnly<NZSpellCasting.MonsterLineQueue>()}, Any = new Unity.Entities.ComponentType[]{}, None = new Unity.Entities.ComponentType[]{}, Options = Unity.Entities.EntityQueryOptions.Default}); } just creates a __query_index without checking if that's already been created

[stiff mesa]

Hey everyone! I need an advice. I'm making a level manager for a modular game system. Level loading logic should work like this:

Level Load Called | For Each Module Call PreUnload Function | Unload Level | Load New Level | For Each Module Call PostLoad Function | Anounce Level Loaded

I've prototyped this using Coroutines and it's pretty simple there. Each PreUnload and PostLoad function is just another coroutine and they are ran one after another.

But with DOTS it is not as simple. Logic tells me that PreUnload and PostLoad functions should be systems and should position their updates using UpdateBefore and UpdateAfter attributes. But the problem is every step of the process should start only when the last one is finished. So, for instance if you have 5 modules, Unload Level step should start ONLY when all 5 PreUnload functions have finished their execution. How would you approach this?

[viral sonnet]

this doesn't need to be 6 different systems. make 1 system and 6 different jobs. chain their dependency. done 🙂

you are still able to make 6 systems. the jobhandle has to be shared though. it works but isn't as elegant

[stiff mesa]

Yeah, sounds reasonable. So, I should just inherit jobs from IPreUnload and IPostLoad, then get them through reflection and stack them one after another in a single system, right?

[viral sonnet]

why would you need the abstraction of the interface and reflection? just write an Entities.ForEach or job struct

unless you have different PreUnload jobs

[stiff mesa]

Well, each module will have its own logic for those things, so the code for each module's PreUnload will be different.

That's why I thought about multiple systems.

[viral sonnet]

hm, i see. then your initial approach sounds reasonable. here's what you can do. you can define systems what they require to run. one of those is RequireSingletonForUpdate<T>. the "For Each Module Call PreUnload Function" systems can require a LevelLoadCalled comp (it's just a single entity with this comp) which triggers the PreUnload stage systems. after you are done, destroy all singleton entities you created. this way you can chain the systems. also set the order with UpdateBefore/After as you mentioned. either this way or with shared jobHandles.

[stiff mesa]

I'll try this. Thank you!

[viral sonnet]

you're welcome

[rotund token]

so all that partial class, partial job stuff

worked fine in 0.17

we had to rewrite a bunch of partial systems in 0.50+

was a painn

I reported these errors in my mega thread way back when 0.50 came out

[viral sonnet]

oh, wasn't on my radar when you posted this 😄 did you have more problems? i'm just having trouble with GetSingleton

[rustic rain]

What is the benefit of partial classes?

Why not just have static helper class?

[viral sonnet]

i'm using it to provide easy modification of base behaviour. could also help with modding support

[rotund token]

you can't split jobs across files

[viral sonnet]

huh, i'm doing that. splitting how? anything special? i've a lot of methods all over several files

[rustic rain]

Only if it's library

Modders won't be able to have asm ref

[rotund token]

are they code genned though?

[viral sonnet]

ah no they aren't

all job structs

[rotund token]

yeah which is fine

[rustic rain]

Unless it's one mod per game with such changes

Which is destructive pattern

Imo

[viral sonnet]

true dat, personally i just need it for my asset

modding support is such a pandoras box 😄 setup has to be much much more mindful

[rustic rain]

I figured best solution for mods I think

I create interfaces, base classes for many many things

Boot with different hooks for all stages

Ui

Even input

So my code can obtain it all through reflection

What isn't satisfying for modders can be improved with harmony

Meanwhile unmanaged game logic pretty much does not need much support in this regard

Cause it's all data anyway

And if someone wants to modify it he just does it

Also a key is that I base my own logic on these tools

[viral sonnet]

great, my spell loader isn't loading with subscene closed (not unloaded) all entities are there. the fk ...

oh the loading order changes. that's annoying. one would expect that the loading systems are running in the same frame and before my systems. seems to not be the case

[errant hawk]

FloatMode = FloatMode.Deterministic I found this new float mode that is a placeholder. Will this be cross-platform determinism, or something else?

[viral sonnet]

that's the plan but it's been on ice for a loooong time

and at that point, questionable if it will ever see true cross platform

i don't know the details but it strikes me still as weird that there float isn't deterministic on all platforms 😄

i mean, you'd think engineers would be thinking about this when the first network cable was plugged in. but apparently, everyone does its own thing despite norms

[rotund token]

that's the cpus fault

and float specifications

[viral sonnet]

that's what i wonder. why are they so different.

[errant hawk]

There's no standardized method of float math

That's why

which REALLY sucks

you have to either have a compiler do the work converting all floats to be deterministic, or use integer math (it's likely they both do that, just the compiler does it for you) which is slower

[rotund token]

from our testing, all modern desktop cpus (not including apple m1 etc) produced deterministic results

but as soon as last gen consoles (ps4 etc) enter mix

results vary

i.e. intel/amd meeting x64 specification are fine which is not surprising

that said our testing was just just for debug tools, we don't require determinism so don't have thorough testing for this

[dense crypt]

How do you guys handle helper functions for components? Say I have a component that represents currency and this currency is represented by an int, but can be expressed as Gold/Silver/Bronze (think World of Warcraft).
Typically I'd just slap a function on the component that converts the int to the gold representation but components should only contain data, so this approach isn't recommended.

I find that I end up with so many random helper classes for my components that it's starting to get confusing, how do you typically organize your files/components/helpers?

[errant hawk]

I just use extension methods and nest the file in the same directory and namespace as the component.

For pre-existing components, like PhysicsMass I put them in a special folder, but put them under the same namespace as the component.

[minor sapphire]

I would just have a property in the component itself to set/get the int with gold/silver/bronze. Recommendations by unity in certain contexts are not rules 😄

If it makes sense to do it, then it's fine, imo. Your example is literally just a reinterpretation of data fully contained in the component. Seems pretty harmless to me.

Break free from the shackles of vague recommendations!

[rotund token]

my question is, outside of representing this to the user visually, does it ever need to exist outside of its pure int form?

i don't think there's much harm in throwing a few properties on a component to present the data in different ways

[errant hawk]

Unity even has properties for some components

Is there any way to transfer unity terrain over to ECS?

[dense crypt]

For this example it's probably practical to just use the integer, but it was just an example. But yeah, I suppose having a few properties here and there would be fine.

[viral sonnet]

unitys recommendation is concerning logic. just data transformation is fine

[errant hawk]

Do lights not work? I cant seem to get entities to project any kind of light

Lights also dont work in subscenes

is there a setting I need to enable/disable?

[honest plinth]

Lights (and cameras and many other things) are still game objects. Only certain things is converted to entities

[pliant pike]

well I changed it to follow that Unity GatherInputSystem sample, tertle posted and I get pretty much the same laggy result in fact it feels like the cursor goes more haywire

[rustic rain]

hmm

when I use query in EntityManager.AddComponent

it modifies chunks instead of moving entities?

[rustic rain]

hmm

so here's what I'm trying to achieve

I have a singleton entity which gets tons of tags added

which results in dozens of chunks created for no reason

what I want instead: change it's chunk

switch it's archetype

while adding component to chunk batched seems doable with query
Removing components is not

Removing dozens of components from query seems just impractical and very costly

[viral sonnet]

seems to me like you want to fix bad design. is it that important that the singleton is used for queries. i mean it's 1 entity. put some flags in a comp

[rustic rain]

it's dynamically created

[viral sonnet]

but you must have a finite amount of tags, right?

[rustic rain]

no

for modding support must allow for any amount

[viral sonnet]

can you create an archetype for it?

[rustic rain]

?

[viral sonnet]

creating an archetype with all the tags would at least prevent the dozens of archetypes that are created via addComp

otherwise i don't see much you can do, it's just how entities work under the hood

[viral sonnet]

hm, another dependecy issue. how is this not working? ```var updateIntervalsHandle = new UpdateSpellEffect_Intervals
{
...
}.ScheduleParallel(mainEffectsQuery_Intervals, Dependency);

var updateDurationHandle = new UpdateSpellEffect_Duration
{
SpellEffect_DurationEndTick_ReadHandle = GetComponentTypeHandle<SpellEffect_DurationEndTick>(true)
...
}.ScheduleParallel(mainEffectsQuery_Duration, Dependency);

var removalHandle = new RemoveSpellEffects
{
...
}.ScheduleParallel(mainEffectsQuery_Removals, updateDurationHandle);

var removalHandle2 = new RemoveSpellEffects_WithStats()
{
...
}.ScheduleParallel(mainEffectsQuery_RemovalsWithStats, updateDurationHandle);

Dependency = JobHandle.CombineDependencies(removalHandle, removalHandle2, updateIntervalsHandle);leads toThe system NZSpellCasting.CombatEffectSystemServer reads NZSpellCasting.SpellEffect_DurationEndTick via UpdateSpellEffect_Duration but that type was not assigned to the Dependency property. To ensure correct behavior of other systems, the job or a dependency must be assigned to the Dependency property before returning from the OnUpdate method.```

[devout prairie]

Could you have mods effectively produce an entity per modification and then gather them into a single archetype at runtime?

[rustic rain]

are you using your unsafe CDFE?

[viral sonnet]

nope

straight GetComponentTypeHandle and GetBufferTypeHandle

my unsafeCDFE has AddReaderWriter calls btw when used in SystemBase or ISystem

[rotund token]

Is there another exception after this

[viral sonnet]

that's the only one thrown

chaining looks okay, right? maybe the problem is elsewhere

[viral sonnet]

odd thing is that everything works fine. just the error gets thrown every frame while an effect is running 😄