archived-dots

[rustic rain]

allthough I probably should

[timber ivy]

I dont think it will be relevant for my use case my subscenes are only 100x100 meters

[rustic rain]

still better to have 1 subscene

if it's all connected world

rather than multiple

I think...

[timber ivy]

The only thing I see relating to sections in SubScene is IsSectionLoaded

am I missing something?

i guess you got to implement it yourself

[rustic rain]

yeah, kind of

just look at SceneSystem sources

it's all pretty clear what they do

to load scenes/sections

also

https://docs.unity3d.com/Packages/com.unity.entities@0.51/manual/loading_scenes.html#independently-loading-scene-sections

this

but it's all not burst compatible

hm

found this component

maybe this is how you define sections

yyep

the downside I guess is that in editor you won't have any distinction between loaded sections

[timber ivy]

so you just add that to your scene entity?

[rustic rain]

you add this to objects in subscene

and they will get assigned to selected index section

parents will be parented

I mean

parent's section will be inherited by children

But I personally avoid using parenting for perfomance reasons

[rotund token]

i think in general better to have separate subscenes for the sake of editor performance

when editing them

however sections have some nice uses though

like interiors etc

things that might not need to be loaded from a distance

[timber ivy]

oh so you can use sections like layers so to speak

have the main shit load, then as you get closer the small details load

i like that idea

[rotund token]

each subscene optionally has a collection of sections

and you can load these individually

by default there is only 1 section

and most people only use 1 section per subscene

this is first discussion i've seen where sections have actually been discussed

but knowing about sections is important for loading via scripts/jobs

because you load the sections

[rustic rain]

from what I noticed

[rotund token]

each section gets it's own world bounds

[rustic rain]

0 section is root subscene

and is loaded always

is that true?

[rotund token]

ahh

i don't think so?

not unless you actually load the subscene

[rustic rain]

I tried to load subscene

but I can't unload section 0

[timber ivy]

So is all I got to do to create a subscene entity this?

            RequestSceneLoaded componentData = CreateRequestSceneLoaded(parameters);
            Entity entity = base.EntityManager.CreateEntity();
            base.EntityManager.AddComponentData(entity, new SceneReference
            {
                SceneGUID = sceneGUID
            });
            base.EntityManager.AddComponentData(entity, componentData);
            return entity;

[rustic rain]

only unloading subscene works

[rotund token]

how did you load it in the first place?

but yeah each subsection has it's own AABB

    public struct SceneSectionData : IComponentData
    {
        public Hash128          SceneGUID;
        public int              SubSectionIndex;
        public int              FileSize;
        public int              ObjectReferenceCount;
        public MinMaxAABB       BoundingVolume;
        internal Codec          Codec;
        internal int            DecompressedFileSize;
        internal RuntimeBlobHeaderRef BlobHeader;
    }```
>        public MinMaxAABB       BoundingVolume;

[rotund token]

oh you mean by editor?

[rustic rain]

after I unloaded section 0

it says this

[rotund token]

have you tried via code though?

[rustic rain]

so I'd assume subscene is simply unloaded

and now if I try to load section 1 with unloaded subscene

[timber ivy]

Notice that the default section 0 is always there (first line) even if it is empty. The "Section: 0" part of the name is omitted, but all other sections containing at least one entity will show up with their full name.

All sections can reference both their own entities and the entities from section 0. Describing the way this reference system works is out of scope here, but an important consequence is that loading any section from a scene requires section 0 from that same scene to also be loaded. The same constraint applies for unloading: section 0 of a scene can only be unloaded when no other sections from the same scene are currently loaded.

[rustic rain]

yeah

seems like Section 0 is root of Subscene

allthough they are created as different entities

[rotund token]

i have no issue unloading section 0

when section 1 is active

[rustic rain]

hmm

[rotund token]

right is section 0, left is section 1

[rustic rain]

maybe editor limitation

[rotund token]

now you don't see it

[rotund token]

what i'm saying!

[rustic rain]

I'd worry about it though

what if in future it'll be limitation of runtime as well? xD

[rotund token]

why?

[rustic rain]

Can simply be defined as pattern though

[timber ivy]

so if i do this

        Entities.WithAll<WorldChunkComponent>().ForEach((Entity e, ref WorldChunkComponent chunk) =>
        {
            if (math.distance(pos, chunk.position) <= 400)
            {
                Entities.WithAll<SceneReference>().ForEach((Entity e, ref SceneReference scene, in WorldChunkComponent chunk) =>

It will pass chunk from the first foreach to chunk in the second?

[rotund token]

You can't nest foreach

[timber ivy]

how would I check if an entity exists in a foreach loop then?

[eager pawn]

There's not really an example of using ecs and sqlite, not sure if a db makes sense to use for game data

Also I think I would like to preload all data a scene needs when you enter the scene

The database stores static data btw. So it's stores basehp, but not hp

I think I will make an authoring component hold db data and then convert it to an entity component

[viral sonnet]

get the KernelSlot by ref via ElementAt from the nativelist.

@rotund token what did you optimize in addition to removing the memcpy?

[rotund token]

for my spatial map stuff?

All I changed today was made the quantizing write direct to key values ptrs of the hash map then changed the job to just only do the bucket calcs

[timber ivy]

        var buf = sys.CreateCommandBuffer();
         Entities.WithoutBurst().WithAll<WorldChunkComponent>().ForEach((Entity e, int entityInQueryIndex, ref WorldChunkComponent chunk) =>
         {
             var system = World.GetOrCreateSystem<SceneSystem>();
             if (EntityManager.HasComponent<RelevantComponent>(e) == true)
             {
                 if (chunk.entity == Entity.Null)
                 {

                     buf.SetComponent(e, new WorldChunkComponent()
                     {
                         entity = system.LoadSceneAsync(chunk.hash),
                         hash = chunk.hash,
                         position = chunk.position
                     });
                 }
             }
             else
             {
                 if (chunk.entity != Entity.Null)
                 {
                     buf.SetComponent(e, new WorldChunkComponent()
                     {
                         entity = Entity.Null,
                         hash = chunk.hash,
                         position = chunk.position
                     });

                 }

                 
             }
         }).Run();

buf.SetComponent is throwing this exception anyone know why? I am already using a command buffer

Structural changes are not allowed during Entities.ForEach. Please use EntityCommandBuffer instead.

[viral sonnet]

ah alright, i think i was still awake for that 😄

[rustic rain]

can't you just create your world chunks in conversion?

oh wait

you can't

you need a system that resolves those chunks once, after they are loaded

there's absolutely no purpose for you to do that expensive stuff in actual runtime loop

[timber ivy]

I have paused that for now until I ccan get something work with loadsceneasync

[rustic rain]

oh

there's easy way

Object.FindObjectOfType<SubScene>()

xD

[timber ivy]

haha

[rustic rain]

it has guid

I actually do it like this

but I store references

in an object

can be done also unmanaged

you create a system that looks for all subscene/section entities

without your own tag

or whatever component you need

so and then you simply create your own data that references those entities

store them in array

map

or whatever

or create other entities

with reference to those scene entities

[timber ivy]

Yep that is what I planned. Just having some trouble with the dots api itself now x)

are you not able to do entities.foreach twice in one system update?

[rustic rain]

you can

but be ready for Dependency troubles

hehe

seems like you are new to this

[timber ivy]

new with dots yes

Some reason even when using a command buffer I am getting Structural changes are not allowed during Entities.ForEach. Please use EntityCommandBuffer instead.

guessing this is one of the dependency issues you mentioned

[rustic rain]

In case you have troubles with Dependency
#archived-dots message

here my convo with tertle

about how Dependency works internally

[rustic rain]

no this is actually an error that says that you can't change world structure

[timber ivy]

awesome that will come in handy

[rustic rain]

during job that works with chunks

everytime you make structural change, all buffer references and chunks are basically invalidated

that's why the moment you do structural change in foreach loop you are basically done and you must return.
And that's why it's simply not allowed

[timber ivy]

So what can I do in this situation?

[rustic rain]

you can schedule structural changes through buffer

think of it this way

you make a foreach loop

and every time you trigger smth with check

you simply write it to List of tasks

and then once your loop is done

you do those tasks

EntityCommandBuffer is exactly this

but supports a large amount of actions

[winged bane]

You are looping on an entity that would already have this component. This is suspicious

[timber ivy]

im trying to set the value of the component

[winged bane]

use

chunk = new WorldChunkComponent {...} 

if a ref goes into the the entity loop, its read/writeable

no need to set component

[timber ivy]

throws the same exception

[rustic rain]

no

you just change values

of it

[timber ivy]

same exception

[rustic rain]

what exception

[timber ivy]

Structural changes are not allowed during Entities.ForEach. Please use EntityCommandBuffer instead.

so I guess

its what you said before

[timber ivy]

just can't do it ccause I got another job messing with these entities

[rustic rain]

you can't use EntityManager here

kek

you also do

LoadSceneAsync

which is also structural change

also that part is just 😅

[timber ivy]

why?

should I be using getExistingsystem instead?

and yep

that was the issue LoadSceneasync causing structural changes

time to do it with createntityinstead :/

rip

[winged bane]

ye I just saw that SceneSystem too

[rustic rain]

just cache it in OnCreate -_-

besides

you always should use GetExisting

because all of them are already created

by the time OnCreate is called

[winged bane]

@timber ivy This might be of help for you. I havent tested it but maybe it will give you some insight.

[timber ivy]

Oh awesome I will give this a look over in a bit when I am home. Thank you for the example

[latent quest]

Hi hi. I'm trying to get a grid partitioning or system into my game but I keep running into issues with nothing allowing NativeArrays, NativeLists, NativeHashSets, NativeHashMaps or seemingly anything that would make such possible. I tries this https://github.com/marijnz/NativeQuadtree but it seems to be pretty outdated and comes up with the same issues I'm running into. I can't seem to find anything online to help me figure out how to use Native-Things in anything but nothing that helps comes up. I seemingly don't understand the Native-Things at a fundamental level, so... what on earth am I doing wrong?

By "nothing allowing" I mean components and systems btw

[robust scaffold]

If you're trying to nest collections, use the Unsafe X variants.

Use UnsafeList for example instead of native arrays or native lists. They should be added to SystemStateComponents as you need to manually dispose of them

Generally, you shouldn't be nesting or adding unsafe collections to components. Kinda wrecks the cache line for burst

[viral sonnet]

hm, going back to the physics discussion yesterday. any idea how i can selectively build a BVH. i don't want to enable/disable comps. some way to get the physics BuildPhysicsWorld to run at a later point and then skip, just build entities I select

[latent quest]

How would one do quad/octrees or such stuff without nesting collections? Iirc the ecs physics system has a native octree in it though not exposed.

[robust scaffold]

Check the source (I deleted it once I finished) to see if there is a none component query to gather the initial data for the BVH construction.

[robust scaffold]

Dots physics has a BVH tree, two of them, one for static and one for dynamic objects to reduce rebuilding cost. A BVH tree is just 2 native arrays sized N bodies * 2 - 1.

[viral sonnet]

nope ... 😦

[robust scaffold]

Oof, well uh, maybe selectively remove and add physics world index?

[viral sonnet]

that could work. let me think about it. otherwise, tertle said using the BVH is pretty simple. I never looked much into it. I just need it for raycasting

[robust scaffold]

The performance of building my tree is so god damn awful. 4.5ms for 10,000 bodies. I need to reduce it to 1/10 that for viable physics.

[viral sonnet]

oh really? i have 8.7ms for 250k

[robust scaffold]

Yea. Unity's is really optimized. Mine is hand written and very bad.

8.7 for 250k is good. That's what I want. If I scale to 250k, mine takes 90ms.

[viral sonnet]

the stateless nature makes it really slow

8.7 for not doing anything is horrible in my book

[robust scaffold]

It rebuilds every frame. Such is stateless

[latent quest]

Is the n * 2 - 1 for interlacing values or am I misunderstanding? How would queries work if you can't grab the partition without looping the whole array? Sorry for all the questions ;-v-)y

[viral sonnet]

yeah but it's not colliding or smth. it's just idle.

[robust scaffold]

It needs to first figure out if something is colliding. Idle should be handled by the change filter though

But IMO, float instability means things get changed regardless even if nothing moves

[viral sonnet]

stateless or not, you can't just ignore every optimization as excuse

yep, that's a problem. a change filter doesn't care about inaccuracies though

[robust scaffold]

Just google BVH, it is originally a rendering tree for ray tracing.

[viral sonnet]

so they do check if anything has changed but then proceed to still write the same data to a persistent array ...

the change is only for BuildBroadphase

anyway, changing the worldIndex of an entity isn't that simple because it's a sharedcomp. would move to another archetype

[robust scaffold]

I just duplicated my test entity to 250k and ran a root node AABB calculation. It takes 4ms just to calculate it. Are you sure that's right?

[latent quest]

Wow, this is incredibly fascinating. Thank you a tonne for helping! I had no idea BVH even existed until now and I had been searching it with Unity while you were mentioning it at first and couldn't find much ( ;-v-). I'll be working on implementing this now :P. You've been super helpful! Thanks~!

[viral sonnet]

going down to the root of it. my ai job touches rotation every frame so there's the issue it's rebuilding 😄

not that rotation does anything for a capsule collider. meh, have to introduce a quaternion dot to not write back all the time :/

[robust scaffold]

Solid green. I disabled the BVH builder so thats why it's so fast.

[rustic rain]

how many entities?

[robust scaffold]

250,000

[rustic rain]

you mean 250k sprites?

[robust scaffold]

yea. A single sprite = 1 entity

[rustic rain]

huh

1ms of gathering sprites

pretty good

[robust scaffold]

Yea, im not worried about performance of my sprite renderer. This is without culling which I probably should implement but it's not an issue.

[viral sonnet]

a whole BVH has to be rebuilt because one entity moved? is that a normal implementation?

or is this a stateless thing again? 😩

[robust scaffold]

yep

well no, there are insertion mechanisms that allow for partial rebuild

But that requires expanding the box around an entity by a set margin

If the entity doesnt move outside the box, it doesnt need to be rebuilt

The box is 0.1 world space units around the entity, hardcoded.

[feral minnow]

What are the best tutorials on ECS on YouTube? Preferably tutorials series of building demo game 🙂

[slim reef]

that's a bit tough question with the ever changing API and outdated docs

watch all, read docs, watch them again and then have some usable knowledge

[rustic rain]

Look up moetsi

It's not YouTube

But it's the only series with full demo game in the end

I know of

[feral minnow]

can you give me a link please? 🙂

[viral sonnet]

is that implemented? that's not working for me

demo character controller touches trans/rot for change version. whole bvh is rebuilt

and something is still touching LTW. man that's annoying to find out

none of my systems are touching it. still, the check system reports it as changed. hm...

[robust scaffold]

Huh, scanning the full docs now that I finished lunch, it's not implemented. BVH from Box2D has that functionality but not Physics.

[rotund token]

If anything touches tras/rot then that'll forward onto ltw

[viral sonnet]

i check for changes on trans/rot and they are false

[rotund token]

I was writing a tool the other week to track change filters

I never finished it as I solved my issue mid way

But I intend to at some point

The annoying thing is the inspector triggering filters when an Entity is selected (might be buffer only)

[viral sonnet]

so the weird thing. it's not my player. it's happening for my many instanced ai spellcasters. and it's not happening for every chunk but only 1. i've yet to figure out what's different about those

and man, now i know exactly what you mean with naming entities. i always have to close the subscene to get rid of names and can search for entity ids

[rotund token]

I believe there's a way to search by id if you have names, something like i:

[viral sonnet]

ok it's the chunk for enemies. should have realized this sooner 😄 still no idea what's up with ltw. it's not my pathfinding system either :/

and here i thought it's "id:" thanks a lot

[robust scaffold]

One nice benefit of rolling my own transform components is that I can be 100% certain only whatever I code is what will touch the ultimate local to world matrix

[viral sonnet]

haha, yeah ... i was under the impression this will be something i easily find out and i had a very obvious flaw for my player movement. but then it still happens and i searched for every ltw and nothing. :/ when trans/rot doesn't change, ltw can't change either, right? it must be something obvious i'm missing

[robust scaffold]

Trans, Rot, and Scale = LTW. If anything even does a GetNativeArray to any component, then it gets marked as changed.

Hrmmmmmmm, looking into the source of Unity's BVH builder, I think I know why it's so fast. I'm gonna test it out on my version for a test though

[viral sonnet]

found the error. i'm a dumb dumb. parallel dot = 1 so my condition was wrong and i wrote rotation back. now, why ltw was reported and not rotation. i dunno! if it would have been rotation i would've known instantly where to look. so now i just found the right system via turning them off one by one. i have to say, pretty cool feature

[robust scaffold]

You can turn off systems?

[viral sonnet]

yeah

[robust scaffold]

huh

[robust scaffold]

How long did you say a full BVH rebuild by Unity Physics took for 250k entities? Can you try with only 50k (my comp chokes quite hard on 250k).

And can you post what the profiler job distribution looks like

[robust scaffold]

For 50k?

[viral sonnet]

yes

[robust scaffold]

Wow, that is 4x faster than my initial first level job.

huh

[viral sonnet]

i have a ryzen 5900x

[robust scaffold]

True, different comps. Hrm. The distribution looks the same though and how I expected looking at the source.

[viral sonnet]

so I've fixed the version bumping now . doesn't change the fact though that everytime something moves it spikes quite heavily :/

[rotund token]

you'd expect something to move every frame

[viral sonnet]

exactly!

[rotund token]

is all your stuff really close together?

[viral sonnet]

it is

[rotund token]

just seeing all the stuff basically being in a single branch for second phase

[viral sonnet]

so that's why this is only 1 worker thread?

[rotund token]

it is using multiple

its just that 1 branch is taking a lot longer than the rest

the first step breaks up N branches on ST then these branches run in parallel

[viral sonnet]

ohhhhh good to know

because that's totally unrealistic

so what's the range here? can this even be measured by range?

[robust scaffold]

You might have a very sequential distribution of nodes. Like a staircase instead of a tree.

[viral sonnet]

oh wait, we're talking about BuildBranches. hm. thought this was about the BuildFirstNLevels because that job takes quite long

[rotund token]

also just a note because i saw you talking about this before, there are 2x bvhs. 1 for dynamic 1 for static.

only the dynamic is rebuilt every frame

the static is only rebuilt if you make changes

[robust scaffold]

Static is rebuilt based off a changefilter of static transform components yea

They have not yet implemented incremental rebuild of BVH tree on the dynamic tree but there seems to be hints that they plan to. It's illogical to not.

[viral sonnet]

eh scratch that, i also have a long buildbranches. just too many worker threads that don't fit on my screen haha.

[robust scaffold]

Oh. How many threads?

[viral sonnet]

i see 23

[robust scaffold]

Pleb over here with only 10+1(main). Plz, some spare change threads.

[viral sonnet]

my old i7 also choked along with 8 😄

[robust scaffold]

Yea, I think 3 of those are virtual. Im on intel i7 as well. Pain.

[viral sonnet]

so previous one was an i7-4770k and the new one is at least 1/3 faster

easy to measure, my timings were 30ms, now 20ms

looking forward to a proper incremental rebuild on dynamics. right now all my entities are static because i saw no difference

[robust scaffold]

Alright, turning off surface area heuristic split and cutting the number of initial splits to just the root (like what BuildFirstN job does) does result in similar performance:

[viral sonnet]

hah! similar! that's faster on a slower cpu

[robust scaffold]

This job would only produce 2 parallel jobs. This will produce 8:

[drowsy pagoda]

When you "Unload" a scene in inspector. Then try to access it via script, it becomes invalid, which is understandable. However, when you "Load" it back up, it still comes back invalid unless you recompile. Is there a way to validate the scene without having to recompile?

[viral sonnet]

@rotund token Am I missing something here? For a DynamicHashMap, not multi, this could be read linear and -1 just skipped. what's the bucketNext for in a single element hashmap? i thought this is only required in multiple elements per key.

[rotund token]

more than 1 key can have the same hash

and your key/value arrays can have holes in them

the ability to remove elements from a hashmap adds all these complications

[viral sonnet]

ok i see

thanks!

[rotund token]

if you guarantee you have not removed anything from the hashmap

and added it in a single thread

then yeah you could just grab the key/value arrays

this is the reason my batch add method is a
ClearAndAddBatch

because it's only safe on an empty hash map

i wrote this i think i called it, NativeLookup

which is basically just these native hash maps stripped back to disallow remove, parallelwriter etc

so it's guaranteed to just be a linear key/value

so i could do these type of things

that said i don't use it atm because it needs a bit of love ^_^'

some day

[viral sonnet]

meh, again I'm looking at this and I just think this can't be right ... some optimized debug compilation would be great. release is just all over the place. something seems wrong with summing up these timings

[rbx+630h] must be the tick from the job parameter

nearly 30% in a long job. just no ...

[viral sonnet]

hm, so the optimal target distance system would have the source/target position on the same entity. my problem is now, when 100 casters have the same target and the target moves. 100 target distance entities would need to be updated and that update wouldn't be that fast.

[robust scaffold]

Optimal target? As in closest one?

Or one that it can see?

[viral sonnet]

nah, the caster already knows which target. just to get the distance between

[robust scaffold]

dot the positions together, dont square root it. Very cheap.

[viral sonnet]

the math is not my issue. getting the data is 🙂

[robust scaffold]

There's no other option other than random access unfortunately. You aint gonna get a nice cache line going.

[viral sonnet]

i'm using random memory access to get the position of the target

yep, that makes me sad 😄

[robust scaffold]

100 casters to 1 target, how often does the target change for each caster?

It might be good to use a shared component to fragment your chunks then per chunk, conduct a single random access to find the target for all casters under it then calculate

[viral sonnet]

i don't think whole archetype changes is a good fit for this

targets can change quite erratically

the only optimization i can think of right now is to not use LTW and translation instead.

[robust scaffold]

Depends. If it changes rarely, over the course of seconds rather than frames, I would attempt to try out shared component fragmentation. Otherwise, if this isnt multiplayer, you can instead use a fragment shader with indices to leverage the GPU to calculate it really fast.

[timber ivy]

array[i] =

[robust scaffold]

If you have no inheritance, translation should be the component you read from.

Local to world should only be touched by those actually rendering things or requires full scaling and rotation values.

[viral sonnet]

hm, frag shader. i'd still need to upload/render all my casters then, right?

[robust scaffold]

Yea, you're already requiring uploading of the caster's LtW to the GPU already. However, that's probably buried inside the rendering system / hybrid.

If you want to roll your own rendering system then yea, the values are already on the GPU ready to be used.

I wouldnt recommend it though. Anything short of 50,000 positions is cheap enough to upload. Beyond that, might consider reusing buffers.

[viral sonnet]

does dots physics alreayd has a method for Quaternion.Angle?

[rotund token]

check out Unity.Physics.Math

they implemented a bunch of the UnityEngine stuff

[robust scaffold]

Doesnt look like it

It's just two lines. Ya can do that.

[viral sonnet]

oh thanks! i expected some internal call.

gotta love some magic numbres

[rotund token]

thats 1 radian

57.296Deg × π/180 = 1Rad

[robust scaffold]

It deals with the wrap around at 2 pi.

actually, it's just 2 pi.

[viral sonnet]

man, i need to sleep ... that's the MB implementation. expected rads

[robust scaffold]

That moment when you press play and the game has 1.5 FPS because a single burst compile tag was uncommented.

[viral sonnet]

i'm starting to realize, pulling the trans/rot instead of the ltw doesn't give me any benefit. lookrotation is quite math heavy with 2 cross products

[timber ivy]

protected override void OnUpdate()
    {


        EntityCommandBufferSystem sys =
                World.GetExistingSystem<EndSimulationEntityCommandBufferSystem>();

        if (cam == null)
            cam = Camera.main.transform;

        // Create a command buffer that will be played back
        // and disposed by MyECBSystem.
        
        var ecb = sys.CreateCommandBuffer().AsParallelWriter();
        var pos = new float2((int)(cam.position.x/100), (int)(cam.position.z/100));


        Entities.WithBurst().WithAll<WorldChunkComponent>().ForEach((Entity e, int entityInQueryIndex, ref WorldChunkComponent chunk) =>
        {
            Debug.Log(math.distance(chunk.position, pos));
            if (math.distance(chunk.position, pos) <= 4)
            {
                if (HasComponent<RelevantComponent>(e) == false)
                    ecb.AddComponent<RelevantComponent>(entityInQueryIndex,  e);

            }
            else
            {
                if (HasComponent<RelevantComponent>(e) == true)
                    ecb.RemoveComponent<RelevantComponent>(entityInQueryIndex, e);
            }
        }).ScheduleParallel();
        sys.AddJobHandleForProducer(Dependency);
    }

why is this system so slow? Profiler is saying its taking most of the time on jobhandle.complete but im not even calling that

there are only 400 entities with the worldchunkcomponent

Ik you're not supposed to profile entities in editor but its making my editor run at 20 fps so I can't' really even use it

[robust scaffold]

Oh, if you need to rotate things, definitely use LTW. The matrix multiplication is definitely faster than doing cross products.

[robust scaffold]

You dont need with burst, it's on by default.
Debug.log will murder your performance, remove it when profiling.

Parallel writing is not always faster than singlethreaded, try .Schedule() instead of parallel and see if it's faster.

Parallel is only really faster if you have pre-allocated and know exactly the range in which a thread will write to. This doesnt look like that case.

[timber ivy]

Schedule is slower

run is fine

[robust scaffold]

math.distance is expensive. It requires a square root calculation. Use math.distancesq and square the other side.

[timber ivy]

the delay is coming from addjobhandleforproducer ig

cause the profiler says its waiting on job.complete

[robust scaffold]

That's the ECB replaying on main thread.

Structural changes are always expensive. Try using a boolean property instead of adding and removing components if it's done often

[timber ivy]

Its not done often at all

maybe once every few frames

31 ms waiting on a semaphore yikes

[robust scaffold]

If you have a lot of entities undergoing structural changes, yep. That's expected.

[timber ivy]

that is when im standing still

and the entities arent changing at all

[robust scaffold]

Then it's probably not that job then.

Check the profiler timeline display, what jobs are being waited on

[timber ivy]

the profiler is pointing to that exact system tho

All the job is doing is looping through about 400 entities. if you have moved to a new area add a component to the entity

but im not even moving and its at max 15-30 entities that this system is even adding components to

[robust scaffold]

Thats a lot of red. That means you have a lot of memory allocation.

The debug.log has been removed right?

[timber ivy]

yes

and

where am I allocating?

i dont see anything other than that debug.log(which has been removed)

[robust scaffold]

ECB is actually, IIRC, a multi hash map just with specialized keys and values.

[timber ivy]

so am I better of just doing this in .Run() then?

[devout prairie]

definitely burst compilation enabled under menu?

[timber ivy]

Yes its enabled

[devout prairie]

i think for HasComponent, maybe pass in the ComponentFromEntity rather than use HasComponent inside the job

[robust scaffold]

If you're doing structural changes, it's typically best to do so on the main thread anyways.

[timber ivy]

LOL safety checks

was the issue

[robust scaffold]

Huh

[devout prairie]

"GetComponentDataFromEntity"

[timber ivy]

im back to 600 fps after disabling safety cchecks

[robust scaffold]

That is a wild amount of safety checks. I wonder why

I have safety checks on by default and I never see any significant amount of performance loss. Then again, I'm passing around raw pointers everywhere.

[timber ivy]

from 32 ms down to 0.032

[timber ivy]

What do you mean by this btw?

[devout prairie]

i was just about to explain..

not sure if it will make any difference tbh but the way i do that would be:

[robust scaffold]

I think the lambda code gen already optimizes it to chunk.HasComponent<>

[devout prairie]

protected override void OnUpdate()
    {


        EntityCommandBufferSystem sys =
                World.GetExistingSystem<EndSimulationEntityCommandBufferSystem>();

        if (cam == null)
            cam = Camera.main.transform;

        // Create a command buffer that will be played back
        // and disposed by MyECBSystem.
        
        var ecb = sys.CreateCommandBuffer().AsParallelWriter();
        var pos = new float2((int)(cam.position.x/100), (int)(cam.position.z/100));
        var relevantComps = GetComponentDataFromEntity<RelevantComponent>(true);

        Dependency = Entities.WithReadOnly(relevantComps).ForEach((Entity e, int entityInQueryIndex, ref WorldChunkComponent chunk) =>
        {
            Debug.Log(math.distance(chunk.position, pos));
            if (math.distance(chunk.position, pos) <= 4)
            {
                if (!relevantComps.HasComponent(e))
                    ecb.AddComponent<RelevantComponent>(entityInQueryIndex,  e);

            }
            else
            {
                if (relevantComps.HasComponent(e))
                    ecb.RemoveComponent<RelevantComponent>(entityInQueryIndex, e);
            }
        }).ScheduleParallel(Dependency );
        Dependency = sys.AddJobHandleForProducer(Dependency);
    }

[viral sonnet]

yep, safety checks are costly. that's by far the biggest increase I've seen though

[devout prairie]

ah does it

so i've just changed HasComponent to passing in GetComponentDataFromEntity

but yeah as KF says possibly it would get optimized to that anyway

[timber ivy]

ah yeah ill see if this does any better in the profiler

kinda lame the safety checks are hitting so hard to performance

[robust scaffold]

It's only 400 entities. No clue what's going on.

[timber ivy]

and i have a really good pc

[robust scaffold]

It's not a PC issue for that big of a difference

[viral sonnet]

i dunno, the timeline is all pink, not green. was that still from the debug log?

[devout prairie]

also you don't need the WithAll as it's already in the lambda

[robust scaffold]

I think Russell had customized his editor colors.

With colorblind mode or something possibly.

[timber ivy]

I have colorblind mode on

[devout prairie]

ah wait..

[viral sonnet]

oh, ok 🙂

[devout prairie]

i think you also need to chain the dependency of the ecb to the job, i've added that in the above code ^

Dependency = Entities.ForEach..

and ScheduleParallel(Dependency);

[timber ivy]

Ok I guess it was just some editor bug

restarting the editor

[robust scaffold]

Not for lambdas in fact. That's automatic.

[timber ivy]

and it isn't lagging with safety checks on now

[devout prairie]

oh really 😐

i always do it but useful to know i guess

sometimes chaining together multiple jobs in one system so i guess it's good practice either way

[robust scaffold]

For systembase lambdas and IJobEntity, dependency declaration and chaining is done at code generation so no need to do it manually

[devout prairie]

ah ok, thanks for that i didn't realise

[robust scaffold]

For system structs, yea. gotta do it by hand

[timber ivy]

System structs?

[robust scaffold]

It looks nice and slick though

[robust scaffold]

The alternative SystemBase. It's in a struct form with all the inherent properties passed into the OnUpdate in a (ref SystemState state) property.

The benefit is that you can burst compile the main thread scheduling and whatever code you have as well as the jobs.

The con is no code generation and everything needs to be written out manually

[devout prairie]

it does.. where are your nativearrays being disposed?

[robust scaffold]

Allocator automatically disposes of it.

Like magic. It's like GC

[devout prairie]

ah right

hah 🙂

is that built in to dots or your custom allocator?

[robust scaffold]

Built in. It's under World.UpdateAllocator

[devout prairie]

ohhh

[viral sonnet]

hm, can't find anything on that subject. can a quaternion be flipped like a vector by 180 degrees without multiplying?

[robust scaffold]

You can use update allocator (seen at the top of that screenshot) for all containers but it has built in special support for NativeArrays and NativeLists (not really a list though as it cant expand).

[devout prairie]

you can do math.inverse(quat) i think is equivalent?

[devout prairie]

that's useful

[viral sonnet]

not sure about that. quat inverse is used for reversing multiplications ie. to divide

[devout prairie]

in some cases i dispose of arrays between lamda jobs in the chain inside a system, so i guess it can be handy to have the option of manual or using that allocator to just handle it

[viral sonnet]

inverse has a dot product. :/ issues issues. fk it. let's go back to ltw

[robust scaffold]

https://answers.unity.com/questions/1125215/how-to-flip-a-quaternion-to-face-the-opposite-dire.html

Read Bunny's comment at the end.

quat is just weird. Im gonna stick with my 2D quats which is just sincos

I can do all sorts of wild things with 2d quats, like logical inversing

[viral sonnet]

ok, that's the long answer for what i said ^^

i'd have to multiply and yeah, i'll stop now. this is getting too math heavy for what is already precalculated in ltw

[robust scaffold]

I think inversing the LTW is just applying a negative identity matrix.

I think... i dont know

[viral sonnet]

inversing in itself isn't really useful unless you'd translate from world to local rotation or smth

[devout prairie]

yeh you're right inverse is basically just, say you have a euler of 3/3/3 it'll create -3/-3/-3

i think useful for relative rotations and as you say local to world etc

[viral sonnet]

the easiest to explain is when you have a rot and apply a 90 degree rot. inversing the 90 degree rot and multiplying by the result would give you the base rot again

[devout prairie]

yeah

[viral sonnet]

quats have no division. that's why the inverse is golden

[devout prairie]

so why do you need to rotate 180?

i think what kf's link is saying you end up with the same rot anyway right

[viral sonnet]

i'll test this quickly.

[devout prairie]

was interesting reading back over the past few hours discussion re bvh etc as i'm looking at exactly that at the moment

i think it makes sense in my case to just leverage physics bvh as my units are already rigidbodies

so to get for example, current list of enemies in range, i guess i have two options:

the first i was trying was have a large collider that is event only, and collect those events, but it's painful and i think someone mentioned events are slow

second is basically, do a sphere cast after the broadphase and collect the results

so if i understand correctly, doing a sphere cast will leverage the built physics bvh anyway right

[robust scaffold]

yep, and it's pretty ideal

[devout prairie]

i always think of raycasts etc as slow coming from a mono/unity background

[viral sonnet]

going back to unity5. god damn! it's fast

[devout prairie]

but i think in the case of unity.physics and dots it's pretty much just dipping straight into the broadphase bvh

[robust scaffold]

BVH is specifically built for raycasts. It's originally a raytracing tree. In fact, it's still used for raytraced lighting.

For nearest neighbor query, spatial grids are fastest but what if you wanted the nearest neighbor not behind a wall? Or the nearest neighbor in a certain direction? That's where BVH edges ahead.

[devout prairie]

there is a way to get Forward from quaternion i think

not sure if it's helpful

what exactly is happening there?

[viral sonnet]

yeah, multiplying rot with 0,0,1

the green line is the forward and the red one is the inverse of the rotation

[devout prairie]

ah!

[robust scaffold]

For 50k. not... good

[viral sonnet]

and why i need the flip: var angleFromTarget = Angle(-dirToTarget, targetForward); is to figure out if the target is in front, flank or behind of the source.

[devout prairie]

so i think convert the target to source space maybe

i really struggle with those maths though, so there maybe some dot shortcut!

[robust scaffold]

Use the vector formed from subtracting the position of the target with the source. Dot it with the forward vector of the source. If it dot is negative, it's behind. If it's positive, it's in front. Then for the flank, arccos the dot product to find the radian angle.

[devout prairie]

this guy

[viral sonnet]

that's basically the angle method ^^ private float Angle(float3 vecA, float3 vecB) { var dot = math.dot(vecA, vecB); var angleInRadians = math.acos(dot); return math.degrees(angleInRadians); }

[robust scaffold]

Yes.

The dot product is magical. You can do so much with it

You can probably figure out the flanks from a range of the dot product without needing acos if you dont need the exact angle.

[viral sonnet]

dot is really nice, yeah 😄

[robust scaffold]

dot is so good, unity named their new technology DOTS.

[viral sonnet]

lol

yeah, flank could be defined <0.7 and > 0.3 or smth

[robust scaffold]

Like if the dot product falls between 0.5 and -0.5, the angle of the target would be roughly 60 degrees to -60 off tangent.

Use cos((deg) * pi/180) to find the decimal cutoff for a degree you designate as flank. Then you dont need the acos lookup table.

Hrm

Well, I need the AABB. Thats the entire reason why I'm building this BVH

I think I understand why Unity glued together 4 AABBs together

[devout prairie]

Apparently dreamworks are making their render engine open source btw:
https://www.youtube.com/watch?v=wk-WztzAjGE

obviously it's an offline renderer, but interesting of them to do that

[robust scaffold]

Pixar and dreamworks are amazing because of the artists they have, not the quality of their engine frankly.

[devout prairie]

Hehe i always thought that multiple day per frame render times seemed excessive for what they were getting

but to be fair, they've contributed a lot.. Pixar created open subdivs which was an open source realtime subdivision algorithm

basically it's like tesselation, but for 3d content creation apps and render engines

[robust scaffold]

They were pioneers in the field dont get me wrong

[devout prairie]

and as far as i understand it, i think they also started the whole PBR rendering thing, and release the PBR algorithms to standardize that

PBR has revolutionized even game engine rendering

[viral sonnet]

weren't they also involved in blendshapes?

[devout prairie]

not sure on that one

definitely a lot of joint efforts there with like fbx evolving to support different stuff like morph targets, blend shapes, mats, skinning etc and then other formats coming along like DAE, collada etc.. all of it helping to just standardize all of this stuff across different pipelines

and now unreal with USD or whatever, universal scene description i think

actually USD was Pixar also 😛 Didn't realise

Open Subdivs is quite an interesting one to look at in terms of it's fast hierarchical data storage/access

I'm not exactly sure how it works ( been a few years since i was looking into it ) but it's a super optimized tree type structure for realtime use

[timber ivy]

Is SceneSystem.LoadParameters.Priority implemented at all?

I dont see it anywhere

[robust scaffold]

Never heard of that no.

But it would make sense.

[timber ivy]

Its there

but I dont see where its being set

def not in scenesystem

gonna dig through the code some more

ah its used by gameobjectsceneutility.LoadGameObjectSceneAsync

[robust scaffold]

There's no comments on priority though

[timber ivy]

My guess

its working like priority does on AsynOperations

and being passed to whatever they are doing to load the actual scenes owned by the subscene

[viral sonnet]

i think the most frustrating thing in todays programming is that it matters so little how many calculations you need for something. 95% is spent with waiting for data. optimizing an acos away is like, great, you saved 2ns

[eager pawn]

should i do all my data access during build time and create authoring components from the tables? i'm not writing to the db, just reading.

or should i create entities directly from db data during runtime?

[rustic rain]

Depends

You should use subscenes as much as you can

Because this is next gen loading

Can create prefabs in subscenes and then instantiate in runtime

[eager pawn]

What if it's an ItemChest containing an Item and it's possible for the player to not open it? Would the Item be created during runtime then?

[rustic rain]

There are infinite ways how this can be implemented

You can create item during conversion and add disabled tag to it

Or make a prefabs pool

And save reference to it in chest

Once it's open you randomly instantiate item

[eager pawn]

It's not random, but yeah I think I get it

[drowsy pagoda]

var sceneSectionComponents = new List<SceneSection>();
EntityManager.GetAllUniqueSharedComponentData(sceneSectionComponents);

I got the shared components. How do I iterate each chunk per shared component from here?

[rustic rain]

Get entity query, apply shared component filter

Get chunks

Repeat foreach

[drowsy pagoda]

Ah.

Thanks

[rustic rain]

I do wonder when all sections are actually loaded

I mean their entities

And I also wonder whether auto loading of it checks whether entity is already created in you do it manual t

[drowsy pagoda]

NativeArray<ArchetypeChunk> chunks = query.CreateArchetypeChunkArray(Allocator.TempJob);
If I set shared component filter, I am to assume all the chunks are of that same Shared component data right?

[rustic rain]

Yeah

Also

Use jobified chunk creationf

If you want to jobify it

Potentially can make it fully threaded too

Allthough...

I just remembered

Ah nvm, for a second I had brainfart xD

If you have lots of chunks

Jobified should be used probably

[drowsy pagoda]

I see

[timber ivy]

is it valid to cache Entity/other components in a component? or is there some special steps I must take?

[robust scaffold]

Ugh, multithreading is pain

[rustic rain]

yeah, why not

with entities there's a problem of checking tho

you'd need to check whether entity still exists

[timber ivy]

alright

just wondering in some ecs it is not valid to do that

for example in leo ecs you must use ecspackedentity

then call pack/unpack to get the actual int representation of the entity

[robust scaffold]

Entity in DOTS is just 2 uints in a single struct

[rustic rain]

yeah, basically kind of a pointer to chunk

in it's own way

[robust scaffold]

Singlethreaded, works perfectly fine and stable BVH tree. Multithreaded, all over the place. Ugh.

[timber ivy]

huh weird my entites auto generated by subscenes dont render in the game view

only in scene view

[latent quest]

Why would one be unable to use a custom struct in a job? Currently I'm using a .ForEach in my code to collect all entities that are to be processed but it's telling me The UNKNOWN_OBJECT_TYPE has been deallocated, it is not allowed to access it. Apparently this is something that happens when modifying an entities which has been invalidated but I'm not editing the entity period in the code.

prims = new NativeList<BVHPrim>();
Entities.ForEach((Entity ent, in Translation translation, in WorldObject worldObject) => {
  var pos = new float2(translation.Value.x, translation.Value.z);
  prims.Add(new BVHPrim { pos = pos, entity = ent, bounds = new float2(2, 2) });
}).WithoutBurst().Run();

[rotund token]

prims = new NativeList<BVHPrim>();

you haven't allocated your list

[latent quest]

allocated?

[rotund token]

yes all native memory needs allocation

what you passed into the job is just a void pointer

look at the constructor for NativeList

the constructor must always be used

Might want to have a quick read of: https://docs.unity3d.com/Packages/com.unity.collections@1.4/manual/allocation.html

[latent quest]

Huh, okay then, it seems to have worked now. I always used Allocator whenever it complained and told me to. I don't really know how exactly it works in all honesty ( ; lvl)y. Thanks!

I'll read up more on Allocator for the future.

[winged bane]

it all gets easier after hitting your head against the wall several dozen times

[rustic rain]

can confirm

hmm

don't quite understand why debug collider drawing is not working

it did create those 2

but nothing is drawn

not in scene mode, not in game mode

[rustic rain]

hmm

seems like it doesn't draw if there's only 1 body

[rustic rain]

        protected override void OnCreate()
        {
            base.OnCreate();
            _centerField =
                typeof(PhysicsShapeAuthoring).GetField("m_PrimitiveCenter",
                    BindingFlags.NonPublic | BindingFlags.Instance);
        }

        protected override void OnUpdate()
        {
            Entities.ForEach((PhysicsShapeAuthoring shape) =>
            {
                var value = (float3)_centerField.GetValue(shape);
                value.z = -shape.transform.position.z;
                _centerField.SetValue(shape, value);
            });
        }

I don't get it

why it works with open subscene

but when it's closed

offset is wrong

I am trying to make literally all PhysicsColliders

at 0 on Z axis

by adding center offset

from go's position

[solemn hollow]

try OnStartRunning instead of OnCreate

[rustic rain]

this is COnversion System

[solemn hollow]

oh i see. did you refresh entity cache?

[rustic rain]

I reimported several times

restarted editor

several times

entities cache didn't help either

ok

I realised what's wron

somehow I change Z value of entity

oof, I need custom inspector for Colliders

seeing nothing is annoying

finally

@robust scaffold here's what I meant regarding 2D constraints btw

even though sprites on different levels, their colliders are fixed on same axis

so this is pretty much how you can construct 2D shapes out of 3D ones

[rustic rain]

oh wow

that's pog, thank you

[rotund token]

i havent got around to doing mesh yet but everything else is there

provides this

oops just noticed typo

should be vertex1

😄

[rustic rain]

hmm

GetHash

for type T

seems like your internal method

[rotund token]

            where T : struct
        {
            return blobAssetReference.m_data.m_Align8Union;
        }```

nothing special

you probably don't even care about seeing the hash

it just requires internal access

i just didn't want to remove something that was previously shown

[solemn hollow]

Im a bit confused atm.
Is entropy supposed to go up if i remove the most unlikely value of a list of possibilities? My limited understanding is it should go down with less possibilities.

[rotund token]

well if it's unlikely it's probably not adding much uncertainty?

[solemn hollow]

exactly. but it adds some. so if i remove it the entropy should decrease slightly right?

[rustic rain]

hmm

who do I need to give internals access to

So files in Editor folder

can access it?

Editor namespace didn't work

[rotund token]

i just have an assembly with asmref to basically everything

this is just unity.entities

[rustic rain]

hmm

I can't get it to work, weird

can't get internals

for custom assembly

in Editor folder

[devout prairie]

How comes?

[rotund token]

What does editor folder have to do with it?

[rustic rain]

idk, it just doesn't work

I declared asmref

created reference to Entities

and created [assembly; InternalsVisibleTo("nameOfasmref")]

aaaaand, still can't see internals

[rotund token]

And get hash is in nameofasmref?

[rustic rain]

yeah

both in same file in same namespace

holy

finally

made it work

now another trouble

nvm

Rider broke everything

kek

[devout prairie]

what would be an alternative approach to events?

like say i have bullets colliding ( just as an example ) and i want to spawn some effect

i think previously where i had bullets i was using raycasts anyway to detect collisions and avoid pass through

but in this case actually i'm looking at melee weapon collision, and say generating a hit effect and assigning damage etc

would i be better off doing an overlap query every frame for every weapon orrrrr

[rotund token]

really depends on frequency

we use overlap sphere for our attacks though

but we only do 1

[devout prairie]

i mean say a unit swinging a melee weapon

i could optimize by only doing the collision query during swing

[rotund token]

yeah we only do 1 overlap sphere for melee attacks

not like constantly while traveling

just has a trigger point when we fire it

(i didnt write this system just letting you know how it works)

[devout prairie]

i guess best idea would be to test against collecting events and compare, i'm leaning towards doing the overlap queries though as it's simpler

yeah that's fair

i'm using ragdolls here so i kinda want some accuracy

[timber ivy]

is it possible to write to a native array in an entities.foreach job?

cause this is throwing dependency errors

[devout prairie]

yeah just use Allocator.Temp rather than TempJob

[timber ivy]

var chunks = new NativeArray<Cell>(9, Allocator.TempJob);
        int index = 0;
        for (short x = -1; x <= 1; x++)
        {
            for (short y = -1; y <= 1; y++)
            {
                short tX = (short)(cell.x + x);
                short ty = (short)(x + cell.z + y);
                chunks[index] = new Cell(tX, ty, size);
                index++;
            }
        }
        Entities.WithAll<WorldChunkComponent>().WithBurst().ForEach((Entity e, int entityInQueryIndex, ref WorldChunkComponent chunk) =>
        {
            
           
            chunk.isRelevant = chunks.Contains(chunk.cell);

[devout prairie]

Pass the Array in with WithReadOnly

[timber ivy]

doing that throws an exception telling me to use TempJob

oh

let me try that

[devout prairie]

but if it's parallel and you're writing to it ( not just reading ) you need to treat as an unsafe parallelwriter and be very careful you're not overwriting from different threads

just reading is simple tho

Pass it in like this:
.WithReadOnly(chunks).WithNativeDisableParallelForRestriction(chunks).

Also, you don't need WithBurst()

ForEach is automagically bursted

But yeah if you're creating the array outside of the job/foreach, use TempJob like you've done ( it holds the allocation for four frames, to give the job time to play out and use the array ), and then remember to call Dispose() on it after the ForEach

[timber ivy]

 var chunks = new NativeArray<Cell>(9, Allocator.Temp);
        int index = 0;
        for (short x = -1; x <= 1; x++)
        {
            for (short y = -1; y <= 1; y++)
            {
                short tX = (short)(cell.x + x);
                short ty = (short)(x + cell.z + y);
                chunks[index] = new Cell(tX, ty, size);
                index++;
            }
        }
        Entities.WithAll<WorldChunkComponent>().WithReadOnly(chunks).WithNativeDisableParallelForRestriction(chunks)
            .ForEach((Entity e, int entityInQueryIndex, ref WorldChunkComponent chunk) =>
        {
            
           
            chunk.isRelevant = chunks.Contains(chunk.cell);
            

        }).Schedule();

        chunks.Dispose();

The system TagWorldAreasAsRelevant writes WorldChunkComponent via TagWorldAreasAsRelevant:TagWorldAreasAsRelevant_LambdaJob_0_Job 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.

still throwing the exception

[rotund token]

chunks.Dispose(Dependency);

you can't dispose that until you finish the job

[timber ivy]

ah

[devout prairie]

Dependency = Entities.ForEach and also ScheduleParallel(Dependency)

[timber ivy]

Doesn't codegen do that?

[devout prairie]

Although KornFlakes tells me this is handled automatically now

[rotund token]

yes

[timber ivy]

Yep that fixed it

tyvm!

[rustic rain]

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

        private EntityQuery _playerSingletonQuery;

huh

that works

with internal access

let's you burst getting entity

        [BurstCompile]
        public void OnUpdate(ref SystemState state)
        {
            state.EntityManager.SetComponentData(_player, new PhysicsVelocity());
        }

Anyways found a simple way to make Dynamic Bodies Kinematic with one simple constraint xD

hmmm

[rotund token]

ah

you realize there is a component you can add

to just do it right

[rustic rain]

is there a way to create children of physics bodies

[rustic rain]

which one is it?

[rotund token]

PhysicsMassOverride

[rustic rain]

hmmm

[rotund token]

    /// Add this component to a dynamic body if it needs to sometimes switch to being kinematic.
    /// This allows you to retain its dynamic mass properties on its <see cref="PhysicsMass"/> component, but have the physics solver temporarily treat it as if it were kinematic.
    /// Kinematic bodies will have infinite mass and inertia. They should also not be affected by gravity.
    /// Hence, if IsKinematic is non-zero the value in an associated <see cref="PhysicsGravityFactor"/> component is also ignored.
    /// If SetVelocityToZero is non-zero then the value in an associated <see cref="PhysicsVelocity"/> component is also ignored.
    /// </summary>
    public struct PhysicsMassOverride : IComponentData
    {
        public byte IsKinematic;
        public byte SetVelocityToZero;
    }```

[rustic rain]

oh

😅

[rotund token]

so just add that component and toggle it on/off as required

[rustic rain]

yeah

Sooo, any idea regarding children?

basically

I want to have 1 body

with main collider

and additional

but so that they trigger their own trigger events

I basically want to have an object with top side collider and bottom collider
If player touches top - he jumps
if player touches bottom - smth else happens

so that Unity does not unchildren it

[rotund token]

physics does not support children

it uses translation and if it's on a child it's in local space

they need to be separate and you just need to make them match positions

[rustic rain]

🤔

[rotund token]

or just use a compound collider and figure out what part it hit

[rustic rain]

even if both bodies kinematic?

[rotund token]

doesn't matter

unity transform system uses translation is local space

[rustic rain]

I assume static bodies simply won't even move their colliders with LTW

[rotund token]

only static colliders use LocalToWorld

            {
                All = new ComponentType[]
                {
                    typeof(PhysicsVelocity),
                    typeof(Translation),
                    typeof(Rotation),
                    typeof(PhysicsWorldIndex)
                }
            });```

            {
                All = new ComponentType[]
                {
                    typeof(PhysicsCollider),
                    typeof(PhysicsWorldIndex)
                },
                Any = new ComponentType[]
                {
                    typeof(LocalToWorld),
                    typeof(Translation),
                    typeof(Rotation)
                },
                None = new ComponentType[]
                {
                    typeof(PhysicsVelocity)
                }
            });```

[rustic rain]

but do static move with LTW?

[rotund token]

static colliders don't move

they're static ^_^'

[rustic rain]

huh

Debug view makes them move

[rotund token]

?

[rustic rain]

I move static bodies

[rotund token]

that's a bad idea

you have to rebuild the static world every frame then

[rustic rain]

is it much worse than rebuilding kinematic world?

[rotund token]

there's no point in making a collider static if your'e going to move it

you're slowing yourself down

[rotund token]

well instead of building just dynamic world

now you're building static + dynamic world

you have 0 optimization

[rustic rain]

well, there is... I can use parenting xD

but I don't really have dynamic world

[rotund token]

sure but you're extremely limiting number of colliders you can have now

[rustic rain]

besides 1 entity

in that case

[rotund token]

so you're not really using physics

you're just using it as a spatial map

[rustic rain]

kind of

tbh

same perfomance

all I have in this case for physics: need to get trigger events

nothing else

all bodies move

by a same value

and there's only 1 dynamic body

[rustic rain]

so, do you think it's an ok case for such mutilation of physics engine? xD

[rotund token]

as long as you're sure you never want collisions

then i guess sure

[devout prairie]

oh i hadn't seen this either

[rustic rain]

only between 2 bodies

ground and player

and only when player "dies"

basically a fancy death animation xD

[devout prairie]

i think that's kinda along the lines of what the VRising guys were doing, although i think they modified the physics engine itself

[rustic rain]

it's not really modification, it's just... using smth not the way it was meant to be used

[devout prairie]

but i think they were setting everything as static and maybe manually rebuilding when needed, or something, and moving static colliders around

yeah

[rustic rain]

and tbh

[rotund token]

there really wasn't much 'dynamic' in v-rising tbh

[rustic rain]

it can be manipulated further

with different simulations

indices

[rotund token]

the entire world was mostly static

[devout prairie]

i think something latio does or plans to do is introduce multiple ie rather than just static and dynamic there would be different dynamic for each collider group or something along those lines

[rotund token]

unity physics implemented the whole physics index in 0.50

but hasn't really used it

[devout prairie]

as at the moment you're polling the entire dynamic world for a specific set of collision pairs rather than say, each unique set of interactions having it's own bvh maybe

[rotund token]

with a little work you can build some nice query only physic simulations without needing to rebuild

[devout prairie]

i see

so how might that work?

[rotund token]

but yeah they definitely have something planned for this, not sure what yet though

[devout prairie]

i think someone was saying it looks like they might plan on making the world partially rebuild also, where there's say a bunch of colliders that don't move, rather than rebuilding the entire thing each frame

rebuild is a killer when you have a ton of bodies

even just having the sleeping thing in havok makes a big difference

[timber ivy]

what is the new version of IJobForeachWithEntity?

I need to be able to have Entity and a component as parameters for a job is that still possible?

[rustic rain]

ForEach loop or IJobEntity

both are codegen

and IJobEntityBatch in a nutshell

[rustic rain]

bruh, again troubles with filters, types and etc

static bodies don't raise trigger events?

[rustic rain]

nvm they do

[drowsy pagoda]

Alright back to moving entity into another scene at runtime. By using EntityManager.SetSharedComponentData both SceneSection and SceneTag it's not working. It's as if some other system is reset the data the moment I change it. Or at least I think I changed it. I confirmed in debug log that the data in SetSharedComponentData is indeed correct (representing the target scene I want to move them to). Someone else have any experience with moving entities to scenes at runtime, please let me know.

[viral sonnet]

why is BelongsTo in CollisionFilter important?

[rustic rain]

because if layers don't overlap no collision will happen

[viral sonnet]

isn't that all defined in CollidesWith

[rustic rain]

well, if you only had 1 layer, like classic physics

then you'd be somewhat limited with CollidesWith

meanwhile with 32 bit version you can literally have any layout you want

for example, let's say you have player

and you want some unique interactions with him

but you also want to have some interaction that doesn't know about player

or any other layers

so instead you just assign player to that layer, that some interaction works with

and voila

[thorny halo]

Hello, I'm looking to save a NativeArray to disk without copying into a managed type, however, I can't seem to find resources on how to reinterpret the array to a managed type, which is what Streamreader/File from the System library requires. Perhaps unity has some packages for saving and loading NativeArrays to/from disk?

[rustic rain]

why copying is an issue in the first place?

Aand you can't do that anyway

managed array would need new allocation

can't be done on top of native

[viral sonnet]

NativeArray has ToArray which is managed

[rustic rain]

but that's still a copy. I assume it matters to him somehow

[viral sonnet]

yeah, i can't read 😄 no idea why that matters.

still, can be done in a non-bursted job

just read the nativeArray

non-bursted because that gives you access to system.io reader/writers

[rustic rain]

I mean, simply ToArray is probably enough

I'd think it's batched so

[viral sonnet]

i need to invest into some ordered chunks. i'm getting pretty sick of references

if the order is the same i can reference correlating chunks via a chunkcomponent. then i can just access by entityIndexInChunk

this all should be deterministic, so i think it's quite possible to do so

[devout prairie]

you lost me at 'i'

[rustic rain]

yeah me too

[viral sonnet]

haha 😄

i have a spellcaster entity and a physics entity. i could merge them but physics takes up a lot of chunk space so i think it's for the best. spellcaster has no position or anything. both entities reference each other. there are multiple occasions where the spellcaster or the physics entity needs data from one another. i have to do a costly lookup to get the data then. but, every spellcaster has a physics entity and vice versa, so they should align in their respective chunk.

you getting my drift dogs?

[rustic rain]

well, I'd say

you should just make your costly calculation async

whatever entity that needs result of that calculation makes a request

systems run

it reads result of request

this way your heavy calculations can be done with chunk iteration

and you can store all data you need in request component

[viral sonnet]

making a request means a write action. that's not making it faster

the lookup isn't that costly. it's a CDFE after all. i just find it superfluous to do a lookup for something that's known and every time the same

[rustic rain]

well, it depends on how heavy your calculation is

[viral sonnet]

my current implementation does lookups and caches them in pointers. this takes 0.3ms for 250k

[devout prairie]

whats the difference between that and the CDFE version?

[viral sonnet]

at least that way i only need to do 1 per frame

hm, not that much on the surface. it's more for advanced usage then. i can memcpy whole comps for example. i can't do that with CDFE

like, copy every position from the physics LTW to the spellcaster translation

[devout prairie]

i often wonder what the impact would be of say.. splitting up components into just single value components, and only passing in the comps/'values' needed.. ie rather than having say a health component that has health/maxhealth/whatever, split those up into individual components.. would there be a benefit to that.. or for example if ECS only pulled in the values used from components inside jobs

[viral sonnet]

there's quite the benefit of single value comps with simd

[devout prairie]

yeah i do wonder about that.. so thinking in terms of values rather than components that are groups of values

[viral sonnet]

and with my proposed method i'd be able to get a whole nativearray of the referenced data. 1 lookup per chunk

i see it this way, we are game devs. we always deal with references in some sense. the naive ways suck pretty hard. :/

something i'd really like to figure out how to do better

and my first idea of aligned chunks goes way back ^^

[devout prairie]

it's funny i'm looking at a similar problem myself at the moment and i'm thinking, rather than worry about efficiency i'll maybe for now just lump everything into a single component, just to get things up and running, and then split it up later.. in one case i'm doing similar, a CDFE for get 'parent' values while iterating over 'children' and i'm thinking is there a better way.. in other case i'm splitting my values into two different components but actually i'm pretty much always using those two components together anyway

i'd be curious how your proposed idea would work, tbh i know very little about the under the hood technicalities of chunks etc and work pretty high level with dots

at some point i'll look deeper into all that, utilization etc, but for now, the task at hand is complex enough without micro optimization 😛

[viral sonnet]

i made a test quite recently. the job used all the data i was merging. and i merged all into 1 exact 64byte struct. to my surprise the job was significantly slower

[devout prairie]

same data but merged into a single comp and it was slower?

[viral sonnet]

yep

all data was touched

[devout prairie]

i wonder if it gets optimized out

ahhhh

interesting

how is that possible i wonder

[viral sonnet]

i didn't really figure it out why. i was using refs too for writing.

[devout prairie]

hmm maybe it's a layout thing in some way, not sure

[viral sonnet]

anyway, 1 value comps are king. tbh ecs and simd expects that

[devout prairie]

not sure the cache implications of either approach, ie large comp vs split into smaller comps

right

[viral sonnet]

that's why i made it 64 bytes. to get all in 1 cacheline

[devout prairie]

would be very interesting to test, split comps up into a bunch of single values

under different circumstances

i do remember reading it's better to split up heavy calculations etc into multiple smaller jobs than do it all in one, if possible

i think even if it's working on the same data.. but again, would be interesting to test that out

see the thing is with Unity, a little bit of community engagement and they could have people doing all this work for them

running tests, benchmarking different approaches, etc

a lot of the things we chat about on here are problems that should have been solved already

just go the Blender model and open source Unity 2.0 ECS ecosystem and let us build it together

they can still monetize the ad platforms etc etc all they want

[rustic rain]

ahem

[viral sonnet]

yep, they are really hands-off when it comes to community. wasn't always the case. before going dark they were pretty open and interested

[rustic rain]

I am sceptical on this one

[devout prairie]

fair enough to be sceptical, i do look at the success of blender though and think, why not.. blender is sitting right on the bleeding edge of tech, is rapidly evolving, and the model seems to work 🤷‍♂️

it's beating the hell out of the competition hands down, with the exception of maybe houdini

[viral sonnet]

i think he was talking about this. and yeah, that's a myth.

reading data is costly too, so you want to do it only once and keep it in stack.

my first iteration of the project was based on this. no idea who started that claim ^^

[devout prairie]

yeah i'm pretty sure i read this in the docs recently, in passing

[viral sonnet]

ok, yeah I'm pretty sure they expect to work on seperate data. tertle did mention the same thing, that he started merging jobs and it helped a lot

[devout prairie]

i do think at times, should i just merge all this logic into one job or one system rather than splitting it up just for the sake of being more organized and modular

obviously doing that might mean also merging some components into bigger comps, or even the opposite actually try splitting data into multiple smaller or even single value comps, why not

[rustic rain]

I finally came to a point I need it aaaand... It uses your private UnsafeListPtr<T>

[rustic rain]

                    var e = buffer.Instantiate(prefab);
                    spawner.lastSpawned = e;
                    var prefabDepth = ltws[prefab].Position.z;
                    buffer.SetComponent(e,
                        new Translation
                        {
                            Value = new float3(spawnerPos.x,
                                spawnerPos.y + spawner.random.NextFloat(spawner.maxHeight),
                                prefabDepth)
                        });

hmmmm

is there a way I can do smth like that?

[viral sonnet]

i don't see why not. the only problem i see is spawner.lastSpawned. does it get patched?

[rustic rain]

this is runtime

and it doesn't work

kek

I get tons of this

[viral sonnet]

spawner would need to go through buffer.SetComp to be ptached

[rustic rain]

probably because field is not remapped

uh huh

I guess this is the moment when logic separation is required

hehe

[viral sonnet]

where is the error coming from? from the lastSpawned?

the rest looks fine

[rustic rain]

it comes from accessing this entity

so yeah

I guess it needs remapping

and logic separation

[viral sonnet]

just do buffer.SetComponent(spawnerEntity, spawner)

it'll get patched at the end of frame then

[rustic rain]

well, that's a lot of data

to pass

I'll move that field to separate component

[viral sonnet]

makes sense

[rustic rain]

oh god, smth went really wrong xD

[viral sonnet]

lol, i'm always amazed how easy it is to rewrite stuff in ecs 😄

run this query on other entities, add some data here, copy it there. and no care in the world that it would break something else. what a relief

in bigger OOP projects I nearly went mad what every change implicated

guess it's easier when your mental image of a project is just a fraction. a blissful state, but when not, man, pretty exhausting

[devout prairie]

useful

[worldly isle]

How do I do a simple trigger with unity Dots? I have been reading through the physics sample and all I can see is raycasting or collider casting. I just want a simple box trigger zone and I want to apply a tag to anything that is inside that box trigger

I can't seem to find anything about this in the documentation nor the source examples

[viral sonnet]

this example should get you started: https://github.com/Unity-Technologies/EntityComponentSystemSamples/blob/master/PhysicsSamples/Assets/Demos/2. Setup/2d. Events/2d1. Triggers/Scripts/TriggerVolumeChangeMaterialAuthoring.cs

also this: https://github.com/Unity-Technologies/EntityComponentSystemSamples/blob/master/PhysicsSamples/Assets/Demos/2. Setup/2d. Events/Scripts/Stateful/StatefulTriggerEvent.cs

i just linked the important scripts here. it's best to load up and look at these samples

[drowsy pagoda]

Is there a dots equivalent to UnityEngine.Bounds.Encapsulate() but with AABB ?

[rotund token]

MinMaxAABB has the method

AABB stored as Center/Extents
MinMaxAABB stored as Min/Max

they can be implicitly converted between depending on your required operation
there is the cost of calculation for converting hence they have different methods and the one you use should be dependent on the operations you do most frequently

e.g.
a lot of encapsulations? keep it in MinMaxAABB
a lot of contains? keep it in AABB

[drowsy pagoda]

got it, thanks

[viral sonnet]

hey tertle, have you read what i wrote about aligned chunks?

and i can't remember, did you ever successfully use an unsafelist in a chunkComponent?

[rotund token]

i only have 1 chunk component and it uses an unsafe list

as for aligned chunks dont recall reading anything

[viral sonnet]

started throwing the idea around here: #archived-dots message

and this is a chunk that always exists, right?

[rotund token]

oh wait sorry

i replaced my unsafelist

with an UnsafeParallelHashSet

for some optimizations

[viral sonnet]

i got the conceptual problem with this idea how to dispose the unsafeList

[rotund token]

i just dispose in OnDestroy

private NativeQueue<UnsafeParallelHashSet<ArchetypeChunk>> chunkLists;

        {
            while (this.chunkLists.TryDequeue(out var list))
            {
                list.Dispose();
            }

            this.chunkLists.Dispose();```

[viral sonnet]

other than aligned chunks, i could also call them parallel chunks. all i need would be a list of chunk that are parallel because the chunk capacity could differ and a startIndex of the entityIndexInChunk

the goal is to reduce reference lookups to 1 per chunk when entities are linked or have linked data

[rotund token]

how do you intend to force all entities in 1 chunk reference all entities in another chunk?

this was one of my bottlenecks i had to solve when testing 100m

i had 650 entities/chunk

but each chunk referenced ~145 other chunks on average

[viral sonnet]

i haven't thought this through for more than 1 reference tbh. starting small 😄

would require a list of 2 levels. 1 for the reference type, 1 for the chunks involved

[rotund token]

the only solution I thought of to improve this was shared components

but this doesn't really work until unmanaged ones

doing 1:1 shared components is not memory viable

[viral sonnet]

yeah