archived-dots

[left oak]

Game tiles are generally very good candidates for ECS, @tardy locust

[tardy locust]

Not doing great at it

xD

[left oak]

Yeah, I prefer square grids cause hexes are hard to grok, at first

There are definitely some clever programmatic ways to generate hex grids

[tardy locust]

I know

I wanted to try and follow this page

https://www.redblobgames.com/grids/hexagons/

They use Cube coordinates

Or prefer to, any way

[left oak]

I'll check it out, after coffee - I'd like to know how to hex, too, haha

[tardy locust]

That is a pretty complete resource on how to do hex grids

Everything from how the math is supposed to work, how to think about coordinates, etc.

If you can translate it to whatever you are using

You will have a solid grid

[left oak]

Very cool. What interests me is how isometric tiles, done a certain way, are identical to hexes

The column-row arrangement of hexes is easily translated to a NativeHashMap<int2, Entity>

The tricky part is just getting the even-odd offsets right, I guess?

[tardy locust]

That is covered too

I think there are 4 different kinds

[left oak]

What are you specifically struggling with, at the moment?

[tardy locust]

Oh right now I wasn't really trying

Just to see if I could make two rows underneath each other

But I couldn't so

I'll look into that another time

[left oak]

Haha, fair enough! I'm interested in 3d-grids, so thanks for these resources! They will certainly prove useful.

[tawdry tree]

Hex grids are just square girds where every second row (or column, depending on perspective) is offset.
And I see you've already linked to red blob, which is good.

[frosty siren]

I work exactly with hex grids and use this guide. So i can tell that it's very simple and once u implement coordinates casts (Cubic <=> Offset <=> Index) u will work with grids very comfortable

[left oak]

I'm curious if this would be a legitimate use-case of SubScene: as my content-delivery container for user-generated content. It's stored in binary, so I can easily keep in AWS simple storage, or something like that. What kind've contextual constraints come with loading a SubScene remotely into a client at runtime? SubScenes convert their objects into entities on load, right? Or can they only be prebuilt in the editor?

[left oak]

I see now that the SubScene is only converted when you rebuild the entity cache in the editor. However, each SubScene has its own EntityManager, which gives me hope that I can use it at runtime for remote content delivery.

[left oak]

All that said, Blobs will probably be much easier to work with, I imagine 💩

[dull copper]

wonder why they bumped Burst to 1.2

## [Burst 1.2.0-preview.1] - 2019-09-09

- Fix assembly caching issue, cache usage now conservative (Deals with methods that require resolving multiple assemblies prior to starting the compilation - generics).
- Fix Mac OS compatibility of burst (10.10 and up) - fixes undefined symbol _futimens```

changelog is quite tiny

but they must have their reasons

[coarse turtle]

https://forum.unity.com/threads/chunk-info-ui-explained.586729/ Oh man I wish I saw this post earlier 😅

[sacred tundra]

physics preview does not work at all for me. i am following the first part of https://docs.unity3d.com/Packages/com.unity.physics@0.2/manual/getting_started.html but the ball does not fall

regarding the convert script, using "convert and destroy" like in the link really does destroy them. they disappear. changing it to "convert and inject" does not do anything. ball remains hanging in the air

[dull copper]

@sacred tundra did you try the physics samples first?

you need hybrid renderer for the converted objects I think

[sacred tundra]

no. this guide is all i did

hybrid renderer eh? makes sense. because the simulation is running, i can see that in debug view

[dull copper]

well, you can probably get away without hybrid renderer but you can then only use like collision shapes and rigidbodies that haven't been linked to meshes

anyway, physics samples are here: https://github.com/Unity-Technologies/EntityComponentSystemSamples

I recommend looking those up as they give practical samples for a lot of these things

[sacred tundra]

hybrid renderer made it work

can i use hd render pipeline combined with this? 🤔

[safe lintel]

yes

[dull copper]

yes

not sure about built-in renderer but both LWRP and HDRP both should work

[sacred tundra]

buildtin is what i am having right now. so it works :p

[dull copper]

ah

then it's all good

you can't use HDRP's DXR with hybrid renderer atm

just putting that out there in case that's any interest to anyone

(I know I would want that)

[safe lintel]

ahh that might explain my crashing earlier

[dull copper]

it doesn't crash for me, but raytracing totally breaks around the hybrid renderer's objects

like, just get pitch black objects or some weird artifacts around the objects

[safe lintel]

oh. well at least it i know its generally incompatible

[dull copper]

Unity's RT dev said on the #archived-hdrp that hybrid renderer hasn't been tested at all with DXR

I'm sure they'll sort it out eventually

but it seems like these are two separately developed systems atm

DXR support is super early anyway

[safe lintel]

i was impressed i could run your sponza proj on a 1070

but after enabling dxr in my project, unity refuses to open without a good crash

[dull copper]

@safe lintel are you on beta 2?

at least a12 was super crash happy with DXR for me

betas are better but they've also fixed some of the crashes on SRP also

I dunno which version you downloaded, I updated the dxr sample repo a while ago for newer HDRP master

or staging, can't remember which it was (I mainly pick one that's newest from the two)

[safe lintel]

b1 was doing that for me, havent tried with b2 yet

got other dots woes just eating away at my time

[analog horizon]

Hm does anyone know a way to use reflection probes with RenderMesh's?

[mystic mountain]

I have this werid thing I'm not sure if I've managed to create myself, or something that is within ECS or the multiplayer system. I spawn a ghost entity, parent another ghost entity to it. But the Translation component of the child in the debugger seems to be showing world space instead of local space.

The child is just offset 6 units from the parent.

[mystic mountain]

nvm, I think I just parented on one side. So the values I look on is on server, so that's why it looks correct, but I'm actually displacing it in my parenting code.

[amber flicker]

I have a NativeArray bob and for each entity I iterate in a system I access the data via an entity reference (otherEntity). e.g. someData = bob[otherEntity.Index];. This random access is pretty painful performance wise. As it's very predictable - are there any cunning tricks to getting this data into cache with the other component data? Like a component that has a pointer into the NativeArray or something?? waves hands about vaguely

[vagrant surge]

@amber flicker store the index directly

caching it

[amber flicker]

not sure I understand - for clarification, it's not otherEntity.Index that's time consuming, it's bob[idx]

[vagrant surge]

not much you can do about it, really

but just accessing the stuff from linear (on the index thing) shouldnt be an issue

the cpu can prefetch those kind of accesses fairly well

[amber flicker]

hmm I need to work on a simple example but pretty sure with 100k iterations (which is a lot), that read alone adds about 1-2ms

[vagrant surge]

have you thought of having that array on a per-chunk basis?

that way instead of having a 100k long array, you have many of them, and they are chunk-local, so memory is less spread

[amber flicker]

hmm... I was wondering if there was a way to e.g. associate a buffer with a chunk (I know there's chunk data) so it would be loaded in on a single cache line (or I'd hope) but I'm not sure that's possible

[vagrant surge]

thats what stuff like chunk data is for

you can also have an array of arrays

and store the index in the chunk data

[idle steppe]

I'm still pretty new to ECS, and i'm not sure i'm doing this right.

I'm writing a nameplate system; I have 2 different Entities, one being a character entity that has a bunch character components, and the other being a nameplate entity. The nameplate entity exists as a pool of nameplates and one can be fetched and bound to a character entity.

In order to bind the two entities i'd need to have the nameplate entity reference the character entity and get it's Health Component. I'm not quite sure how to structure this relationship.

            var namePlateEntity = GetComponent<GameObjectEntity>().Entity;
            var entityManager = World.Active.EntityManager;
            var healthComponent = entityManager.GetComponentData<HealthComponent>(character);
            entityManager.AddComponentData(entity, healthComponent);

I can't tell if i'm just trying to force object oriented programming here or if there is a better way to have 1 entity reference another's component data or if I should just be adding the nameplate components to the main character entity. I worry that as I build this game the character entities are just going to have hundreds of components.

[hollow jolt]

Im checking out the first example in ECS the class RotationSpeedSystem_ForEach : ComponentSystem inside the function OnUpdate where it uses:

Entities.ForEach((ref RotationSpeed_ForEach rotationSpeed, ref Rotation rotation) =>
...

I kind of get that the first parameter is the component data, but how does it know about the second argument ? Does it automatically scan each RotationSpeedAuthoring_ForEach : MonoBehaviour, IConvertGameObjectToEntity and extracts its rotation value ?

What is the optimal way to learn ECS ? I jumped into the examples without knowing anything about it, would it be better to practice ECS without using the ConvertToEntity behaviour ?

[silver dragon]

@hollow jolt RotationSpeed_ForEach and Rotation are components. Entities.ForEach iterates over all entities with the given components, in this case over all entities with both RotationSpeed_ForEach and Rotation.

To learn ECS/DOTS: Jump in the examples, docs, and sometimes the source code 🙂 Ah, and read discord and forums!

[hollow jolt]

ah i see, how could i find out the list of all the components that are present in the Entitie by default ?

[silver dragon]

Hm, you could look at the Entity Debugger

[hollow jolt]

Just noticed the pin and found few articles on the official website will make sure to look into that

ok thanks

@silver dragon regarding performance wise is it better to reference all the components in a simple foreach loop or is it ok to separate them ? Also what is the limit of the amount of arguments i could use in foreach ?

[silver dragon]

@idle steppe You could add a NameplateRef component to the player entity which holds a reference to a nameplate entity. And a PlayerRef component to the nameplate entity which holds a the player entity.

@hollow jolt For performance you should use one IJobForeach and Burst 🙂 Limits... huh... you have to look at the overloads to find out... i think 5 or 6 components are doable. Additionally a buffer and monobehavior. Not sure...

[dull copper]

I have a question for Unity ECS experts here, what's the most perf efficient way today to sync entities with gameobjects transforms? Right now I just poll the pos and rot from entitymanager each update via component data getter using entity id but is there any more efficient way?

example snippet: cs transform.position = entityManager.GetComponentData<Translation>(EntityToTrack).Value; transform.rotation = entityManager.GetComponentData<Rotation>(EntityToTrack).Value;

hybrid renderer would be nice but it doesn't play ball with DXR atm and it has some other limitations as well

[mint iron]

Not an expert by any means but ill throw in some thoughts. There is some overhead in the checks in EntityManager.GetComponentData that adds up doing it one by one. So if you could grab all the pos/rots and targets that need updating in advance and process them together it could see improvements.

[dull copper]

yeah, when polling one by one just feels it could be faster

[mint iron]

Another idea i've been meaning to look into is using the UnityEngine.Object m_CachedPtr. Since the c# classes like for Transform are just wrappers for fixed layouts on the native side it should be possible to figure out the layout and update the pos/rot manually from a burst job with direct memory access, but obviously you've got a can of worms with that one.

[dull copper]

it could also mess up with some of the internal optimizations they do as we don't know everything that happens on their side

last time I asked the staff of being able to access transforms from native api (I can see the access points on IL2CPP code), I got warned it could screw up their GC etc

but you can also set the gameobject transforms in jobs too, so if you can somehow get these things in some list / array etc from entity side, it could be a lot faster

last time I looked into that was when I was trying to optimize bullet physics transforms for Unity GO's

[mint iron]

yeah, its an interesting area that i've seen very little around. I was thinking about it facing the lack of MaterialPropertyBlock support in the current renderer. Having to replace the material currently is a huge problem.

[dull copper]

I'd really really want to use the raytraced reflections, I don't really care that much of the rest of DXR (but could also use the shadows as they are nice)

but right now it might take anything from half a year to year+ for Hybrid Renderer to get compatible with DXR, I seriously doubt it's happening with 2019.3 at least

[mint iron]

The other option is to stuff all your GameObjects into a dictionary and look them up by Id, which is what im doing for all my UI elements. So the menu panels can get switched by ECS side events. The instanceID is stored during the ConvertToEntity process when it has access to both the GameObject and Entity.

[pliant pike]

is there a reason why you cant access public const variables externally in an CompSystem?

[mint iron]

do you mean from within a job?

[pliant pike]

well its a public variable in a jobcomponentsystem

I can access an ordinary float using GetExistingSystem<system>.somevariable

but I cant access it when its defined as a public const

[tawdry tree]

Uh, constants are static meaning you acces them by the Class directly, not an instance

class SomeClass{
  public const int Constant = 1;
  public int Variable = 1;
}

//Accessed like:
SomeClass.Constant;
var instance = new SomeClass();
instance.Variable;

[pliant pike]

Oh yeah I see now, thanks Hodhandr

[idle steppe]

Is it an acceptable design pattern to have entities referencing other entities and pulling components from their references? Seems like im doing something bad.

[amber flicker]

@idle steppe very common - using ComponentDataFromEntity? Of course nice if you can design things to avoid it in an ideal world but often necessary.

[idle steppe]

Ahh interesting - that was the call I was looking for. Thanks @amber flicker . A way to fix this coupling is I could just add all these nameplate components to the character entity instead - i'm worried a few months from now i'll have hundreds or thousands of components on that entity - is it better to break things up into smallers entities when possible?

[amber flicker]

Hmm... it's always a balancing act and there aren't many general rules you can apply. If you have a character entity with 100 components, yes you likely would be better off changing some of the design I'd say. Could probably offer more specific advice for a more specific problem.

[idle steppe]

Imagine a character with buff and debuff slots like a mmo - each debuff / buff bound to the character could add multiple components to the entity on top of the already thick stack of determining factions, level, health, mana, etc. These buff/debuff slots could have no limit to how many there are.

A large chunk of systems would just be running on the same character entities but with their own unique queries. It starts to look like object oriented programming to me at that point.

[amber flicker]

I think broadly you can go one of three routes (off the top of my head): 1) (my choice I think), debuff & buff Buffers per entity - essentially an array of buffs, 2) entity per buff/debuff with a reference to your character entity, 3) a NativeList or something independent of an entity but per character storing buffs/debuffs

but if they are buffs/debuffs it's actually probably a lot more complicated as to the best architecture - I guess they have durations, perhaps exclusive conditions and so on? - might be worth seeing if anyone else here has worked on a similar system

[vagrant surge]

with unity model, its a great idea to split into multiple entities

the model generally doesnt like super-fat entities

for buff slots, multiple entities works better than you would think

[idle steppe]

How would you maintain that reference? a CharacterReferenceComponent on each of the buffer entities?

[vagrant surge]

yeah

[idle steppe]

How would you store that reference in a IComponentData? the index of the entity?

[vagrant surge]

yeah, the entity ID

[amber flicker]

{
    public Entity Character;
}```

[idle steppe]

seems weird but i'll go for it - thanks for the help!

[amber flicker]

any idea what kind of numbers you anticipate? characters & buffs?

[idle steppe]

1000s

[amber flicker]

of both?

[idle steppe]

yes

[vagrant surge]

absolutely entities as buffs then

the thing is that buffs are likel to have components like buff type, buff damage, buff time, etc. And adding/removing components is expensive

so better to leave the player entity mostly alone

and just "attach" entities to it

[amber flicker]

hmm... I'm thinking buffers would be better @vagrant surge ? I'm having to refactor a huge bunch of stuff that's per entity to buffers because at those numbers the out of cache really hurts?

[vagrant surge]

with buffers its not so flexible for the buffs themselves

you can have it 2 sided

characters have a Buff component

that holds an array of entity IDs

of buffs affecting

and buff points to character

thats somewhat how i was doing it for a MMO pirate server i was prototyping

[amber flicker]

I guess it quickly gets into the realm of what data do the buffs need to access and where does that sit etc

[idle steppe]

thats the scary thing about buffs - a designer will try and make it access everything and nothing in the same breath.

[vagrant surge]

thats why you might want them as entities

so they are configurable

can add more effects by adding more comps

[idle steppe]

I haven't been able to touch buffers yet. what's the limitations of that path? And why opt for a 2 sided approach when prompted?

[vagrant surge]

just an array as a component

the 2 sided approach is so when the character dies you delete his buffs

[idle steppe]

ahh - makes sense.

Would all of these need to be in the same chunk? the coupling of entities seems to break what unity described as this magical place where everything sits tightly in memory.
ie. Should a entity with a component referencing another entity be in the same chunk.

[vagrant surge]

nah

the buffs will be split into different chunks

by their type

so basically all "fire DOT" buffs will be on one chunk

all "ICE debuff" in another

etc

[safe lintel]

@dull copper wouldnt it make sense to use copytransformtogameobject component? any entities transforms are synced within a burst job

[idle steppe]

gotcha!

Thanks again!

[dull copper]

@safe lintel there is such component? 😄

or you mean I should make one?

[safe lintel]

there is one 🙂

both CopyTransformToGameObject and CopyTransformFromGameObject

[dull copper]

is it used by the old gameobjectentity?

[safe lintel]

no need to use gameobjectentity

[dull copper]

no I mean, is that used internally by it

[safe lintel]

no

[dull copper]

I don't see ECS sample repo using either of those things

they do the naive component data getters there

which is why I asked in the first place

[safe lintel]

i dont really understand the usage of gameobjectentity in the physics samples when they want people to use convert

i see the proxy still exists for them if you wanted to use that

[dull copper]

the reason on physics samples is partially because they want to show how people can still use that from monobehaviours I think

if you look at the physics vehicle sample, most of the vehicle code is in single monobehaviour

which is totally unoptimized as well

[safe lintel]

you can still get the entity by way of conversion though

[dull copper]

not sure where you are going with that?

[safe lintel]

you can get the entity that a gameobject gets assigned from conversion, so I dont see why they still use gameobjectentity

[dull copper]

they don't

or at least I don't remember seeing any sample that does that there

but my memory also sucks 😄

[safe lintel]

the entity to track thing does i thought

[dull copper]

nah, they store the entity id to monobehaviour and poll entity manager for component data with that cached entity id

[safe lintel]

anyway com.unity.entities@0.1.1-preview\Unity.Transforms.Hybrid\CopyTransformToGameObjectSystem.cs or CopyTransformFromGameObjectSystem.cs has the usage of those components

[dull copper]

it doesn't use gameobject entity

yeah, I'm looking at those now here, thanks for pointing out those exist 😄

quick google got me into discussion which pointed out that those CopyTransform functions only work with GameObjectEntity?

[safe lintel]

no, it just needs to be added manually(unless you intend to use the proxy equivalents)

[dull copper]

so need to feed in the GO ref manually?

well, I'll look into files, should be more clear then

docs are kinda useless on this https://docs.unity3d.com/Packages/com.unity.entities@0.0/api/Unity.Transforms.CopyTransformToGameObject.html

this is what I dislike on these new autogenerated docs

I'd prefer the old style Unity docs with proper descriptions and sample snippets

[safe lintel]

So if you have the ConvertToEntity script with ConvertAndInject selected, it should auto add the transform

yeah me too, do you know if they plan for docfx to replace the original manual at some point? i also dislike the narrower appearance

[dull copper]

no idea but since most are going for packages, it's not a bright future for us

I totally get it's easier to keep the up-to-date this way

but this is #1 reason why competitors game engine docs are so subpar on API docs

[safe lintel]

and theres like virtually zero actual documents for any of the conversion or hybrid stuff so not really sure what the long term plans for how this all works is

[dull copper]

there's some bare bones description here for transforms https://docs.unity3d.com/Packages/com.unity.entities@0.1/manual/transform_system.html#section-3-default-conversion-basic

but the rendering side isn't that great 😄 https://docs.unity3d.com/Packages/com.unity.entities@0.1/manual/gp_rendering.html

[mint iron]

@dull copper so i decided to take a break from important things and had a look at this native transform idea, and actually got it to read the TRS fine! but setting the values doesn't work 😦 bit of fun anyways. https://gist.github.com/jeffvella/6468cd84c3eccb7a57390bcd5b4ac7ac

[dull copper]

that's cool tho

[coarse turtle]

@mint iron nice

[dull copper]

wonder how likely that is to break when they change things 😄

[mint iron]

haha, yeah

[dull copper]

so this kinda works in opposite direction which I'd need myself 😄

[coarse turtle]

btw, for the CopyTransform stuff, this is really all I did to sync the translation/rotation -> the gameObject transform

        public void Convert(Entity entity, EntityManager dstManager, GameObjectConversionSystem conversionSystem) {
            if (CopyToTransform) {
                dstManager.AddComponentData<CopyTransformToGameObject>(entity, default);
            }

            if (CopyFromTransfrom) {
                dstManager.AddComponentData<CopyTransformFromGameObject>(entity, default);
            }

            if (CopyInitialFromTransform) {
                dstManager.AddComponentData<CopyInitialTransformFromGameObject>(entity, default);
            }

            if (CopyScaleToTransform) {
                dstManager.AddComponentData<CopyScaleToGameObjectTag>(entity, default);
            }
        }

for scale, I just wrote my own system to apply a Uniform/NonUniform scale back to the gameObject

[dull copper]

@coarse turtle oh, that's simpler than I thought

I don't really care about scale at all

all I need to do is to sync dots physics to GO side

[coarse turtle]

yea the scale thing was my own stuff, unity doesn't have that, just an FYI

[dull copper]

I'm kinda fighting something totally different right now, I want entities package to show up in projects solution but it just refuses to pop up there

[coarse turtle]

If you're on windows, you can try to leave it in ILSPY

the entities package

at least that's what I do so I can quickly view its references

[dull copper]

I know I can get it there if I manually move the entities package from library/packagecache -> packages but it's PITA to update when next version comes

there's this editor setting that should gen all project files automatically

but it's not working, not sure if it's 2020.1 alpha issue 😄

[coarse turtle]

ooo you're on the alpha haha

[dull copper]

yeah, for no good reason tho, I'll try again on 2019.3

[coarse turtle]

haven't looked into that yet, im still on 2019.3b lol

[dull copper]

was just testing 2020 and figured I'd check this there

[coarse turtle]

good pt

[dull copper]

there really isn't any good reason to jump to 2020.1 for me right now

will see if they add some feats soon that I can't do without

[mint iron]

if all the proj files have been generated i've just used add project to temporarily put it into the solution, it gets removed again whenever you restart unity but for that debug session it then lets you step through the code etc.

[dull copper]

the issue is that this doesn't do those csproj files for most packages

it's like it just ignores that editor preference setting fully

it's same result as without the toggle

[mint iron]

hmmm, sounds like a 2020 thing then

[coarse turtle]

@mint iron that's a good tip 👍

[dull copper]

I even wiped the library and let it process again

[mint iron]

entities is in packages for me now because i got mad at all the private/internal methods and did a find-replace to make the entire thing public.

[coarse turtle]

same goes for me with the transport layer

[dull copper]

well, it's not working on 2019.3 either for me on this project

so I guess I'll just more the package again

it was so nice when this worked tho

[safe lintel]

"all I need to do is to sync dots physics to GO side" oh you will need to make your own mini system for this 😄 the CopyTransformToGameObject system uses localtoworld, which unity physics ignores(uses translation,rotation)

[dull copper]

lol

well, thanks for the headsup

[coarse turtle]

oh, that's a good to know lol

[dull copper]

well, this starts to make more sense then

like, that must be the reason why they just poll the entitymanager for the translation and rotation data on physics samples

I'm sure they'll come up with more elegant setup out of the box at some point

those copy components, they were initially used in GameObjectEntity right?

I have faint memories of using those a long time ago

[coarse turtle]

yes, the proxies/componentdatawrappers

[safe lintel]

in conjunction with the proxies

yeah

[dull copper]

and GameObjectEntity is going to get removed,right?

[safe lintel]

some of the conversion stuff uses part of gameobjectentity under the hood so 🤷

[dull copper]

ah, I just noticed that the samples moved away from it, so took that as a sign

[coarse turtle]

dunno, afaik joachim said that should be going away "soon", but the conversion workflow will stay for a few more years on the forums

[dull copper]

yeah, conversion workflow will be made production ready according to him

or at least that's what I remember reading

[mint iron]

physics also completely ignores any child-parents you setup manually in code outside of a prefab Hierarchy 😦 that was a complete pain in the ass for me.

[safe lintel]

probably just get rolled into the converttoentity script but i remember trying to replicate some functionality from it a while ago and found the dependency

[dull copper]

I'm kinda sad that ecs editor took steps back

not fan of the conversion setup

I'd just want hybrid components and then editor support for the both sides 🙂

[vagrant surge]

when Entitas is a thing

the clunk and jank on conversion looks like shit

[safe lintel]

i just want more dots compatible packages, i kinda dont mind the state of the editor(except for maybe the physics joint workflow)

[dull copper]

missing proper editor is quite limiting workflow wise

especially runtime debugging is pain

it's just going back in usability, removing nice things we've used having

[vagrant surge]

its why people think ECS is hard

if it was like entitas it would get a lot more usage

[dull copper]

I'm sure they'll cover this topic at Unite to some extent

[vagrant surge]

lets see

because it seems they are going even more into conversion workflow

given that Tiny editor got abandoned

what i wonder is why is it so hard

cant they just do what Entitas does? make a component that displays the ECS components on the hybrids

[dull copper]

I don't think the tiny editor was fully dumped

like current version of it maybe

hmmmm reading the tiny thread again

well, if they can make the conversion workflow so that you still see the entities in the hierarchy, that's fine

apparently the hierarchy is kinda issue with ECS data layout tho

[vagrant surge]

yeah they want to flatten

cant they add folder type stuff ?

[vagrant surge]

a visualization beetween "gameobjets" and entitities would definitely be useful

in such a case, then the game object editor is kinda the "edit only" mode?

and then you can link the GO with entity to see the entity vals

[dull copper]

@safe lintel @coarse turtle just putting ConvertToEntity to "Convert and Inject GameObject" and adding that CopyTransformToGameObjectProxy to the same GO does update my DOTS physics rigidbodies

so, maybe they've changed that on physics package since?

I looked at the tranform copy system and it still copies the data from LocalToWorld

it also uses that TransformAccess that works on updating GO's on jobs

[safe lintel]

the translation is being updated by the gameobject?

[dull copper]

it's updated by the copy tranform system(?) not sure if you asked that tho. all I want now is this to be updated by physics sim and mirror the rigidbody transfrom to GO transform

it seems to do that out of the box

so wonder what's the deal with the physics samples not doing it this way

oh wait, when I added the copy proxy, it adds GameObjectEntity there automatically 😄

[safe lintel]

well if its working its working 🙂 my experiences were a windy road

[dull copper]

well, I get a nag about having GameObjectEntity now 😄

it basically says it will be ignored since the conversion script is there already

but the copy component requires it

which is inheriting it from ComponentDataProxyBase

[coarse turtle]

hmmm, maybe i'll just write my own version if its doing that lol 😅

[dull copper]

yeah, it should be doable

but if this works, I'm totally fine by it nagging it on the inspector 😄

they will update Entities package soon anyway and chances are that custom things will break again

[safe lintel]

well if you wanted to use the converttoentity i would add like

    public class CopyDotsTransform : MonoBehaviour, IConvertGameObjectToEntity
    {
        public void Convert(Entity entity, EntityManager dstManager, GameObjectConversionSystem conversionSystem)
        {
            dstManager.AddComponentData(entity, new CopyTransformToGameObject());
        }
    }

if you wanted to ditch the proxy stuff

[coarse turtle]

Oh yeah I did the same, I think I misunderstood previously lol

[dull copper]

that works

so considering it's that simple

why doesn't Unity use this on samples?

[coarse turtle]

tech debt 🙃

most likely

[dull copper]

yeah but they probably put more effort on the current scripts there

[coarse turtle]

yeah...I think unity is just in that phase where they might be a bit everywhere and need to show something visually that 'works'

[dull copper]

hmmmm, apparently "convert and inject to go" doesn't work with dots physics compound colliders

if I add collider to the same object with the conversion script it works

but if it's child go, it gets ignored with this setting

[frosty siren]

For what purposes we have ExclusiveEntityTransaction? There is no comments in code(

[rancid geode]

@sand cypress please use the correct channel, it is something for #💻┃code-beginner

[sand cypress]

Ok, Sorry sir.

[low tangle]

@dull copper personally I created a clone of the copy to / from systems and a new set of components to work independent

this gave me way more control and until TransfromaAccessArray goes away this wont go anywhere

then you only pay for the number of total transforms you sync per frame

I think I have no problem up till 50k transforms but I'd need to recheck as its been a looooong time

[dull copper]

Makes sense, right now there is no interpolation for physics either so could build a system that does it as well

[low tangle]

yep, once you pay for the transformaccess copy once, just do absolutely everything else in ecs and life is good

[amber flicker]

@frosty siren can't help much except to say Joachim, verse 14: "ExclusiveEntityTransaction is more for streaming large amounts of data on a seperate world."

[wary anchor]

I'm trying to perform some calculations on vertex data and I'm wondering whether it's worth bringing the vertex data into a job-friendly format to try to parallelise the work, then take it back out to the main thread at the end to piece together the final mesh.

A summary of what I'm trying to achieve:
for all atoms (tens of thousands), check every vertex and triangle for intersection with another atom.
Save vertices and triangles which are not intersecting to a surface mesh.

There's some tricksy details buried in there but that's the overall gist of it

I have been using the hybrid renderer with batched RenderMeshes which is great for performance, but doesn't allow me to do a few things I really need to do. Thinking I may have to take the final rendering step outside of ECS but use jobs to do all of the up-front calculations that need to be done

Does this sound like a sensible approach?

[amber flicker]

It sounds ok to me - what's your main concern? Is this on-load of a castep file or something?

[wary anchor]

I think the issue I'm concerned about is that I'll be completely separating out the (previously batched) meshes into what could end up being a massive single mesh

But I guess that's probably just me not wanting to give up having 33k objects drawn on screen with just 4 verts and 2 tris 😄

I would save the mesh data within a component as well as I'll be using the verts later for things like calculating possible spawn positions

Thanks for the sanity check @amber flicker going to push on to a first version and see what it's like 🙂

[amber flicker]

yea... struggling to reply with something useful. Are a bunch of instanced geo'd meshes more performant than a giant mesh? Almost certainly. Would you gain perf from jobifying & bursting vertex checking stuff? Yes - although gpu's also a consideration depending on the task I guess. As to what you're trying to achieve from a higher-level perspective - not sure 🙂

[wary anchor]

Currently my level is a collection of spheres (atoms). Previously in my OOP implementation, I generated a surface mesh for each atom with a voxel-type calculation then morphing the mesh to effectively shrink-wrap the atoms. That meant I could then select surface atoms (those with surfaceMesh.vertices.Count > 0) and find spawn points for procedural stuff that way.
I'd like very much to have a smoother surface - less obviously-just-a-bunch-of-spheres (without going into raymarching, probably!), and also have access to what is on the surface and what isn't within the job structure. My levels are protein structures from public databases - I don't have control over the structure so need to work out what the surface is just from the atomic coordinates.

Now everything is super lean in ECS I definitely don't want a bunch of gameobjects with meshes on, and now that Unity has made meshes capable of holding more than 65535 verts, my options are a little better than when I did my first OOP implementation. Assuming a single large mesh is viable, but I can play around with the geometry to make it all right I think.

[amber flicker]

😅 can't add anything useful but good luck - it sounds fun 👍

Unrelated: https://forum.unity.com/threads/can-burst-apply-function-stitching-to-boost-performance.743324/#post-4955402 🔥 😄

[wary anchor]

ouch!

[mint iron]

its probably quite stressful right now organizing the event and everything that goes on.

[low tangle]

sheeeit

@wary anchor not too familiar with how you are going about the smoothing

how big is the data set?

[wary anchor]

In other news, my gravity/collision system is finally working nicely

[low tangle]

the mesh vertexes

awesome!

I'm about to make a sphere simple gravity system soon for my game for details on characters

should be pretty fun

[wary anchor]

uncoloured version as a rough 1st go 😄

[low tangle]

oooo pretty

[wary anchor]

I can't believe how fast it runs! No colliders or physics helps

[low tangle]

no smelly slow monobehaviours as well :)

[wary anchor]

yup!

So my plan is coming down to this:

1) get pairs of intersecting atoms as collection
2) for each atom, calculate vertices by scaling then translating a template 'mesh' (vertex positions scaled to 1f around the origin, eg an icosahedron or similar)
3) check each vertex to see if it's inside any of that atom's intersection partners via collection from step 1
4) write verts to master list (need to think about whether I can do this step in parallel or combine at the end?)

The tricky bit is number 4 as I will save all triangles where at least 1 vert is not intersecting another atom, then add in the other 1 or 2 verts to keep that triangle in the final mesh. Then transform the position of those 2 would-have-been-deleted verts to a shared position with the neighbouring intersecting atom's equivalent would-have-been-deleted verts (possibly deleting one set and sharing the verts at this point).
There will be another issue when at the intersection point of 3 or more atoms, I need to think on this a bit more.

[low tangle]

makes sense

its going to be tricky with the ordering of what needs to happen before x and if you need to verify something

you might want to try doing it over a few systems and stages to make it more sane (with sync points)

then figure out how to remove the sync points later on

[wary anchor]

That's a good point, yes, syncing could be an issue

[low tangle]

the complexity is a big jump from single threaded where you can confirm that something is done before other thing, but multi its very very hard unless you do everything always even if its not needed

I would look at this problem as not exactly within ecs bounds though,

think of it more of using the ecs world / data to run a process, that churns though some large data sets, with temp memory shared all inside of standard jobs

that way you can just focus on a to b

a chain of standard stock jobs should be much easier to manage, but still off the main thread, and still able to be burst compiled

[wary anchor]

yes, I was thinking probably keep it in a single system for now: I have the atom list data from my previous system so that gives me all the coordinates and scales I need. I can ensure sync points and sharing of data like in the boids example that way

[low tangle]

use some native array temp memory inside and just work though what you need in that small tight area

[wary anchor]

sounds like a plan!

now comes the easy bit

😄

[low tangle]

haha

time for more coffee

[wary anchor]

hellz yeah. Thanks @low tangle

quick question, because I don't know up front how many verts I'll be dealing with, I think I need a resizeable collection like NativeList - is it best to use this while generating the data then for later access by other systems store it into a NativeArray of the appropriate size? Or is that microoptimisation. I've not use a NativeList yet, but I know they're slower due to the lack of up-front allocation

[amber flicker]

you can cast NativeList to NativeArray

[wary anchor]

I suppose if I brute force it, I know the max number of entries I'll need: template mesh verts * number of atoms. Might be wasteful though

[low tangle]

@wary anchor do the quick math

what size of component?

how many elements

[wary anchor]

I'm thinking... the key step is how to structure the data of triangles mapping to verts mapping to atoms

[low tangle]

well, mostly talking about the size of the data last question

doing the math on say, tons and tons of float3's

how large is the template mesh, and how many atoms do you expect to load

you mentioned it was a icohedron?

[wary anchor]

If i have an icosahedron and 33k atoms, that's going to be a total possible size of ~ 390k

[low tangle]

390k verts?

[wary anchor]

I think that's right isn't it?

12 verts on an icosahedron

[low tangle]

only 12?

yeah let me do the math then

11 megabytes

[wary anchor]

yeah that's about where I got to

[low tangle]

verts 12 float3
triangles 12 int32
normals 12 float3s

float3 = 12 bytes
int32 = 4 bytes

288 for verts and normals
48 bytes for triangle indexs

336 per ico

336 * 33k = 11,088,000

always fun to calculate data size

so yeah, brute force away, 11mb is nothing for ram

hell you can ask malloc for that in one go for no time at all

[wary anchor]

the real fun part comes later after brute forcing that in culling the unused data. Anyway I'll worry about that when I get there!

I hate keeping track of mesh data

[low tangle]

I'd say to, don't cull or chop the data

[wary anchor]

Not a strong suit for me

[low tangle]

instead figure out whats inside vrs out

by passing over the data and marking it

then pass over that and take your final set and fix the triangle indexes (which you actually dont need to bother with till the end)

heh

you basically doing 3d minesweeper

[wary anchor]

hah yeah I guess it is kinda similar

[low tangle]

if the verts were on a grid, you would just be selecting all that have at least one missing side

[wary anchor]

or rather, nuking all that have exactly 3 missing verts

[low tangle]

something to think about

[wary anchor]

It will take me a while for sure. Structuring the data as it goes into the jobs is still something that I really have to ponder 🙂 I'm getting faster at it

but still

[low tangle]

all good

by taking that step and really thinking about your data first

makes all the difference when it comes to actually solving the problem at hand

[wary anchor]

yes I realise that is absolutely the crux of working in ECS.

[low tangle]

I find I spend far more time being forced to think about my logic and data process than actually writing it

which honestly is a nice side effect

can't really hack it

[wary anchor]

yes, it makes sense when you take a step back and look at game dev in those terms! The actual implementation should be fast if everything is planned right

[low tangle]

100%

I had to finish up a chunk of my games animations and IK back in monobehaviour land

and it made me miss ecs stuff so dang much

[wary anchor]

😄

[low tangle]

but even then I was able to write my monobehavour like a ecs system with components that were nothing but data

its weird at first but strangely still fast as well?

it also made the logic super dang compact

[wary anchor]

that's how I started getting into ECS. I needed to refactor, so I decided to de-monobehvaiour everything possible and do proper unit testing

then I discovered ECS and was like... hang on

that's a much cleaner way of doing it

[low tangle]

worth it

so fucking worth it

I've been doing it for 6 months now since unity first released it

not a single regret

wait

[wary anchor]

but oh my god the mental remodelling required to get even slightly competent with ECS has been HARD!

[low tangle]

its been a year now I think

oh yeah I remember when you came in here

[wary anchor]

yeah sorry about that

[low tangle]

its fuckin hard brain remodeling

[wary anchor]

😄

[low tangle]

I went though it too

its like hold up, back to babys first programing, plus theres like a billion new things to learn about

but thankfully everything is just a simple claw hammer, does only one thing, but does it well

[wary anchor]

I quit my previous life nearly 4 years ago so have been re-training my brain solidly since then. I kinda didn't want to have to forget all that I'd learned, but it's not really like that, it's just extending my skillset and that is always always worth it

[low tangle]

humans are always works in progress, so are our skills

I never intend to stop learning, I love it too much

change is good

[wary anchor]

Yup, couldn't agree more!

Although I can't see what's past my current goal: getting players to play a game I made running around a protein whose structure I solved! That is going to be epic. When I get there 😄

[low tangle]

hell yeah 😁

[wary anchor]

combine my former life with my new one 😄

[low tangle]

I want to see you finish that game!

you got this

[wary anchor]

at least now I have a clear idea of what the game will be. but it is slow working on your own, at least I am slow working on my own

[low tangle]

it comes in bursts for me

[wary anchor]

SO MUCh ^^^

[low tangle]

slow skull dragging for awhile, then massive gratifying jumps

[wary anchor]

I've always been like that, back when doing lab work too, I'd have one epic week and get all my results, then 3 weeks of apparently nothing

Speaking of which, I have a good feeling about this mesh code. Thanks so much for the advice and the chat! Both really appreciated. I'm going to smash this today if it kills me 😄

[low tangle]

haha I was about to say the same, I'm gonna get back to it :)

good luck

[wary anchor]

and you 🙂

[mystic mountain]

Hey! So I have entities in the world (spawnPoints) marked for two teams by a Faction component. There can be multiple spawn points for each faction, and they can be occupied. (lets say marked by a occopiedTagComponent) I have an entitiy that triggers a spawn with a Faction component and what should be spawned. Is it possible to jobbify this code since I need to mark spawnPoints "taken"?

[low tangle]

yeah, but you would have to use a hashmap to keep track of who took what, or a large linear array but I'm not sure exactly how to lock / use a semaphore for the atomic flipping of the bool to mark it as used

I haven't really used any normal multi threaded stuff so far with jobs, usually trying to find a different way to do it with jobs works

usually I just use a hashmap like a locking hashset to check if its been taken at this index

[mystic mountain]

Ok ,I haven't used hashmaps in ECS yet, so I'll have to look that up : )

[low tangle]

also I finally got around to make visual studio snippets for ecs stuff

newjobsystem 
//which expands to this, placing cursor at end
class NewJobComponentSystem : JobComponentSystem
    {
        end

        protected override void OnCreate()
        {

        }

        protected override JobHandle OnUpdate(JobHandle inputDeps)
        {

            return inputDeps;
        }
    }

newsytem 

class NewComponentSystem : ComponentSystem
    {
        protected override void OnCreate()
        {

        }

        protected override void OnUpdate()
        {
            end
        }
    }


newcomponent 
     struct NewComponent : IComponentData
    {
        end
    }

ForEachEntity 

Entities.With(query).ForEach((Entity ent) =>
            {
        end
            });

Documents\Visual Studio 2019\Code Snippets\Visual C#\My Code Snippets

they go into this directory

[mystic mountain]

That's neat, I use ScriptTemplates : )

[low tangle]

I've been suuuper lazy for a long time

wrote every single lambda and system, all 120 of em

oh they add the unity usings at the top as well

Unity.Entities + jobs for the job system

[mystic mountain]

Those are neat : ) Just add to a ScriptTemplates folder * : ) I use a lot of Settings when prototyping ^^

[low tangle]

also documents?

[mystic mountain]

How do you mean?

[low tangle]

never used script templates

oooh I see

those are fuckin neato

[mystic mountain]

That's ForEach version I use as well since I prototype a lot

[low tangle]

I usually end up with like 3-4 systems in one c# file for a strain of logic so I'll probably get a lot of use out of the expand ones I made.
but really thank you for the templates as well, starting off with those is nice

this is going to save so much time I wish I did it a year ago

[mystic mountain]

Haha yep 😛

[wary anchor]

Hmm struggling a little with the best collection to use for storing intersecting pairs of atoms.

I struggled with NativeMultiHashMap because it turns out that my entityQueryCount squared was too big for the capacity (that would be the maximum possible number of intersecting atoms). Can do it with a smaller max capacity but it's being a bit slow.

conceptually I see it as a variable-length list associated with each atom, but that isn't very job friendly...

maybe I should instead just reference the atom which the vertex was inside when storing the temporary mesh data

[low tangle]

refrences are good

lots of linear arrays are good too, can you maybe do a max local sized native array/list?

[wary anchor]

I just always seem to be fighting against the possible data types!

[low tangle]

theres probally a reasonable limit for each one of how many it can be inside?

totally get that

[wary anchor]

yeah there's definitely a reasonable limit, but it irks me that I will have to hard code that

[low tangle]

nativehashmaps and running them in a bursted job is super fricken cumbersome

[wary anchor]

I tried writing every intersecting pair of atomID (int) to a NativeMultiHashMap in parallel. To be fair, that could be sped up if I reinstate my spatial hash code

[low tangle]

I feel like the atoms / intersecting verts code is more of a volume flood filling kind of problem

I wonder if there is a better way to think about the problem

[wary anchor]

then later I'm thinking it's easy to get the collection out relating to each atom of interest as it'll all be listed in that hashmap

hmm goo dpoint

[low tangle]

I wonder if you could create a SDF from the atoms (they are kinda like metaballs) and then mesh that surface instead of overlapping and overdrawing icos

[wary anchor]

it is very much a volume filling problem isn't it

I can see it now. I'm going to have to learn how to write shaders to make a raymarching shader to display them. 😄

[low tangle]

its actually super super easy

but you might be able to solidifiy those meshes

but then the meshes uvs is a intresting problem

if you want them textured

[wary anchor]

nope

[low tangle]

oh easy then

[wary anchor]

either flat shaded with a beautiful low-poly as-equilateral-as-possible mesh (LOL yeah right with no experience!) or just simple shaded. The colour will come from placeables and when I get there grass -like shaders

[mystic mountain]

Quick question, can I add a GameObject prefab to a job?

[low tangle]

not exactly no

pasting some quickly searched sdf stuff
http://jamie-wong.com/2016/07/15/ray-marching-signed-distance-functions/
https://www.iquilezles.org/www/articles/distfunctions/distfunctions.htm

[wary anchor]

kinda like this, @low tangle ?http://www.gradientspace.com/tutorials/2017/11/21/signed-distance-fields-tutorial

[low tangle]

https://www.shadertoy.com/view/Xds3zN

yep the bunnys one is good too

[wary anchor]

I've been reading up and watching videos about ray marching. It looks amazingly useful and very very appropriate, but it's not something I have any experience of in practice yet

[low tangle]

yeah its a whole different can of worms compaired to ecs and stuff

the funny thing is you could very much make these and trace them in a ecs system

I wrote a full path tracer in ecs w/ jobs

[wary anchor]

I can see it from the shader perspective (no need to modify any meshes to get nice smoothed joins)

[low tangle]

burst was stupid fast

yep

basically you cover both your high quality rendering, and the inital how do I even render this data in one go

[wary anchor]

and if it's already in the job system, it also gives me easy access to surface point data per atom which I can use for environment spawnpoints and the like

[low tangle]

hm

oh, the atom data is also a point cloud too

I've seen stuff on meshing point clouds

[wary anchor]

Hmm is exactly what I've had written on screen for a bit

[low tangle]

http://meshlabstuff.blogspot.com/2009/09/meshing-point-clouds.html

thats totally marching cubes

oooh this one looks promising

https://doc.cgal.org/latest/Poisson_surface_reconstruction_3/index.html#Section_48.1

oooh I need to save this for myself for later

[wary anchor]

so before I get overwhelmed with the magnitude and complexity of the task ahead of me...
I haven't done any C++ for a very long time. How possible is it to come up with something within C#/ECS to do this kind of job? Do I actually just need to take a couple of weeks out to learn how to write shaders first?

[low tangle]

its pretty easy to transfer c++ logic to c#

almost trival

[wary anchor]

though it'll all be on the CPU rather than GPU of course

[low tangle]

yeah, but its mesh generation, so the final result doesn't have to change

[wary anchor]

true

[low tangle]

I can't tell you if its a good idea to go with that kind of route at all

[wary anchor]

unless it's not mesh generation and I just go all in for the raymarching for rendering as well

[low tangle]

I would personally attempt to make either marching cubes (which I have before) in ecs and try to mesh that voxel data

yeah, you pay for that in having to learn all about shaders, sdfs, raymarching, and the performance you now have as a baseline on the gpu

that second one has way better surface results than marching cubes, but its much more complex

poisson surface

[wary anchor]

I did something before in OOP which I now think may be close to marching cubes... I made a voxel grid around the protein, then marked each voxel as being inside or outside an atom, then worked out which face of that voxel was an in-out boundary, then made a mesh on that face, then distored the mesh to match the nearest atom surfaces

[low tangle]

I'm also not sure on how high res your voxel data

yeah thats pretty much a simple voxel mesher

[wary anchor]

it was the product of me working without guidance for a long time, but it seemed to work well. I'm sure it's VERY basic compared to the more sophisticated things out there. It wasn't without errors when I ramped up the resolution

although it did give me a continuous surface

[low tangle]

yeah you could basically continue that with guidance if you would want to?

[wary anchor]

Okay I need to read up on SDFs and how to generate/structure/use them. I can't put this off any longer I think. Thanks again @low tangle . Every time I come here you challenge me indirectly to learn a whole chunk more stuff than I'd originally planned to learn. It's great. 🙂

Oh I would LOVE to have guidance 😄

[low tangle]

sounds good 👍🏻

[wary anchor]

I've been working alone in my dining room on a piano stool for the last nearly 4 years

I mean, I've had toilet breaks

[low tangle]

haha

well I'm always here to help

[wary anchor]

k awesome, thanks again. I'll need to read more just now so I understand precisely what it is I'm trying to achieve.

[low tangle]

sdf / sphere ray marching to directly view the atoms on the gpu or
mesh the atom structure by treating it as voxels, or a point cloud, producing a mesh

[wary anchor]

you mentioned you wrote a path tracer in ecs with jobs, is that somewhere between the two?

I mean clearly it's not #1 tho

[low tangle]

not exactly related but related to raymarching

it replaces the unity renderer

[wary anchor]

Hmm. Yeah I'm going to need the unity renderer for most things, but for the atom structure a shader that uses raymarching would be the best possible solution. Whether I can achieve it would be another matter

but then that would beg the question, how do I actually achieve what I set out to: a set of surface data within ECS that I can use for gameplay purposes

[low tangle]

voxel mesh it imo

[wary anchor]

I can do that.

No reason why I couldn't later add in a ray marching shader and use that instead for rendering, but still use the meshed data for gameplay with ECS. Yes. Makes sense!

[low tangle]

very true

might even be able to use it as a proper mesh collider too

it would be a enclosed volume

[wary anchor]

Ooh because I would have the SDF

[frosty siren]

Ahgh, so hate memory layout changing after Create/Destroy entities. I really want to have a method that just return next Entity and schedule create/destroy.

public static void FakeMemory(Entity entity, MemoryType memoryType, int factionIndex)
{
    //Memory access
    var memoryCategoryEntity = GameInfo.entityManager.GetBuffer<MemoryCategory>(GameInfo.factionsECS[factionIndex])[(int)memoryType].entity;
    var memoryBuffer = GameInfo.entityManager.GetBuffer<Memory>(memoryCategoryEntity);
    var memoryArr = memoryBuffer.AsNativeArray();

    var index = IndexOfMemory(entity, memoryArr);
    if (index != -1)
    {
        //Instantiate entity and write returned entity to memory
        var fakeEntity = GameInfo.entityManager.Instantiate(entity);
        memoryArr[index] = memoryArr[index].Faked(fakeEntity);
    }
}

This code cause an error that said that native array is deallocated because memory layout has changed.

[mint iron]

AsNativeArray maintains the same safety as the original and points to the same data so you're not gaining anything here by using it.
EntityManager.Instantiate creates a sync point and so all jobs are completed and all arrays are invalidated. If you want to still use that array you could use ToNativeArray(Allocator) and then you're working with a copy of the data. That strategy works fine if you're only reading it but since you're updating it, you'd have to copy put it back into the DynamicBuffer.

GameInfo.entityManager.GetBuffer<Memory>(memoryCategoryEntity)[index] = memoryArr[index].Faked(fakeEntity); should be able to write with a new buffer wrapper after the sync

[mystic mountain]

Is it a valid approach to have like lookup HashMaps through singletons to find entities through IDs(integers)? My problem is regarding networked objects to parent and other things.

[mint iron]

curious how are you storing them on the singletons? in a component or buffer or something else?

[vagrant surge]

@mystic mountain very common thing to do

hashmap from YourID to EntityID

[frosty siren]

@mint iron i read somewhere that accessing through NativeArray is better for burst or something.

[mint iron]

iv'e heard that too, supposedly there are additional optimizations that are hardcoded on burst compiler if you're using a NativeArray. Worth considering when you're at the stage of needing to optimize burst jobs.

[coarse turtle]

hmm, does anyone know where Unity declares the size of the chunks?

[prisma anchor]

Is there a way to say this inside a job?

            if (EntityManager.HasComponent<FirearmData>(data.FirearmEntity)) {
                var firearmData = EntityManager.GetComponentData<FirearmData>(data.FirearmEntity);

                firearmData.IsFiring = Shoot;

                EntityManager.SetComponentData(data.FirearmEntity, firearmData);
            }

[frosty siren]

@prisma anchor u can use chunk.Has<T> inside IJobChunk. If your problem is using HasComponent<T>(Entity)

[prisma anchor]

Well yea, its using the entitymanager in a job. which is a ref type. ok ill give that a go.

[frosty siren]

List of changes for Job
EntityManager.HasComponent<T>(Entity) -> Chunk.Has<T>()
EntityManager.GetComponentData<T>(Entity), EntityManager.SetComponentData<T>(Entity) -> Chunk.GetNativeArray(ArchetypeChunkComponentType) and then work with it; ComponentDataFromEntity<T> for randomly access

[prisma anchor]

Thanks! Tony!

[frosty siren]

I didn’t fully look through your code, so I made changes to the text, sorry for that

[mint iron]

This is a pretty interesting way of packaging up more complicated functionality for jobs. https://forum.unity.com/threads/how-to-work-without-interfaces-and-other-questions.742514/#post-4959302

[pliant pike]

interesting thanks, I've been wondering about using splines with jobs for a future project

[mystic mountain]

How does blobAssetsReferences work? Could I have for e.g. lists n' reference types within one?

[coarse turtle]

Yes

I believe so

[frosty siren]

Why JobComponentSystem doesn't have PostUpdateCommands?

[coarse turtle]

struct ListRefWrapper {
  public IList<some_type> WrappedList;
}


// Create a BlobAssetReference<ListPtr> for a naive implementation
struct ListPtr {
  public BlobPtr<ListRefWrapper> Ptr;
}

@mystic mountain

@frosty siren, I think their design methodlogy is to use EntityCommandBufferSystem that belongs to your respective UpdateGroup and create a EntityCommandBuffer to playback yourself (you can always playback the command buffer on the main thread too)

[amber flicker]

to add to that - I think it's because you'd have to wait for the job to be complete and thus you'd have no idea when it would execute.. so by using a command buffer you can be explicit

[frosty siren]

yeah i know, but PostUpdateCommands is comfortable

@amber flicker ok, this is what I did not think about

EntityCommandBuffer.Concurrent always returns null Entity when call Instantiate. Am i right?

[amber flicker]

I'm not sure (I haven't checked) but I could believe it. May possibly have to fill a nativequeue instead. That or add an additional component to created entities (I use this) that a system processes.

then at end barrier remove the component (bulk change to archetype = v cheap)

[frosty siren]

yes, i'm going to set components and use reactive systems but the little problem is i need access the same data twice this way and accessing is not so fast

[amber flicker]

possibly fill a nativequeue instead then I Iguess

then you can deque, instantiate and set what you need to

[frosty siren]

Doesn't matter what to use it's means that i will need to access data twice. I store categories in dynamic buffer which is entities that represents an arrays of some data (also implemented with dynamc buffer). Accessing data looks like
DynamicBuffer<Data> dataDB = category_BFE[categoryEntity][categoryIndex];
To minimize using BufferFromEntity i map all i need to NativeMultiHashMap. When i iterate through hashmap i want sometimes create/destroy entities and always need to modify dataDB (Data contains entities). And of course i want to use it inside parallel job. So using deffered create/destroy entities leads to iterate through hashmap in mainthread.

[amber flicker]

but if you want to create/destroy entities then no matter what you do, that needs to happen on the main thread right?

all command buffer or postupdate is doing is delaying until it runs on main thread

perhaps I'm misunderstanding something

[frosty siren]

Yes, work with entitiy must be on the main thread, but i want so much mechanism that will just return the valid Entity inside job)

to avoid all this extra data flow

[amber flicker]

I'm not sure I understand... it couldn't know in the job what the entity that was going to be created is right? Also, populating a nativequeue is pretty much a similar amount of code to writing to the command buffer no?

[frosty siren]

Yes, but after entity creation i just need to write returned Entity in dynamic buffer, that's it. And only to do it i need to populate queue/schedule ecb/use reactive system. I mostly complain, hoping that I don’t know any easy way

[amber flicker]

Question about buffers - is the only way to write to a buffer to use BufferFromEntity? I was operating under the assumption that an entities buffer and it's contents will be loaded into the cache line when processing an entity but this makes me wonder - does anyone know?

[safe lintel]

there are IJobForEach with buffer variants

[amber flicker]

Re-reading the docs it does look like the memory is reserved in the chunk. I’ll take a look at ijobforeach and see what it uses (I’m using Ijobchunk)

At least the initial size anyway

[mint iron]

The default reserve amount is quite low, but if you add the attribute and set it explicitly then you can make things fit in the chunk. Otherwise, when it overflows it allocates elsewhere on the heap and the component just has a pointer out to it.

BufferFromEntity is essentially a wrapper for the typeIndex, version and safety. When you use it, it follows the pointer into EntityComponentStore. Which is basically the same as using EntityManager.

So I'd be curious to learn more about how cachelines work because that sounds to me like every access is jumping through at least a few methods/pointers, and doing a costly second type lookup for IndexInChunk (looping through every component type) then calculating the needed offsets. But it doesn't sound in the same ballpark as iterating a sequential block of data like in an array.

[amber flicker]

Hmm a lot of my refactoring is to avoid the cache miss - at least for very small sized buffers - if it fits within the chunk I hope it's avoiding some rand access?

[vagrant surge]

@amber flicker you dont have to overexcess so much that everything is perfectly cache friendly

even if stuff is a bit out of order, if your data is all in contiguous arrays and its predictable, it will be fairly fast

[amber flicker]

hmm.. I think it is actually hitting my perf quite bad. If I iterate e.g. 100k items my system takes e.g. 1ms. If I add in a line that indexes a NativeArray, that jumps to about 4ms. My assumption was that was due to the cache misses. I can't think of any other reason indexing a float in a NativeArray would add so much time?

[vagrant surge]

well man, 100k is a LOT of items

[amber flicker]

for sure

[vagrant surge]

have you thought of paralelizing it more, or just dont calculate 100k every update? like one quarter every update

but 100k indexes into a native array are slow. maybe the safety features are hitting you?

[amber flicker]

It's not that it's too slow, it's that I think it could be like 3x faster

[vagrant surge]

im pretty sure ive done random access into one array from indices in other array much faster than 100k per 3 ms

[mint iron]

you could try to access the buffers from an IJobChunk

    public struct MyChunkBufferTest : IJobChunk
    {
        public ArchetypeChunkBufferType<SelectionBufferData> SelectionBufferDataType;

        public void Execute(ArchetypeChunk chunk, int chunkIndex, int firstEntityIndex)
        {
            var bufferAccesor = chunk.GetBufferAccessor(SelectionBufferDataType);
            for (int i = 0; i < bufferAccesor.Length; i++)
            {
                var buffer = bufferAccesor[i];

                for (int j = 0; j < buffer.Length; j++)
                {
                    var bufferItem = buffer[j];
                }
            }
        }
    }

[vagrant surge]

so something is amiss?

[amber flicker]

yea accessing like that works but you can't write using a buffer accessor right @mint iron ?

[mint iron]

i havent tried to write with it, but it gives you a DynamicBuffer<T> for the entity so i dont see why not

[amber flicker]

@vagrant surge take those numbers as very approximate - they're actually for looking up a nativearray<struct> where the struct has a few floats and a bool in

@mint iron the accessor seems write only to me when I try

*readonly

[vagrant surge]

even if its random access, the cost looks too big i think

what are you doing once you access it? the cpu can pipeline the memory access if its array into array

[amber flicker]

well.. I comment in/out the line and those are the differences I see in a build attached to editor

[vagrant surge]

im guessing all of the burst build options and the like are enabled?

[amber flicker]

yea

[vagrant surge]

then i dunno

[amber flicker]

I can try and make a simple repro

but I'm not that surprised tbh

a lot of speed comes from the linear memory layout and cache right? If I add a random access in every loop...

[mint iron]

it passes the readonly setting from the ArchetypeChunkBufferType, so i'm thinking u can do GetArchetypeChunkBufferType<T>(false). Guess i could test it.

[vagrant surge]

@amber flicker not really. Lots of it come from predictability and instruction caching

each CPU core is pipelined inside itself

it can overlap memory loads and other operations

[amber flicker]

hmm maybe it's not that I can't write and that I'm just not sure how to set an element? haha... I can buffer[i].Add() but I can't seem to do buffer[i][n] = ..

Cannot modify the return value of BufferAccessor<myBufferElement>.this[int] because it is not a variable

[coarse turtle]

maybe not the best idea, but maybe memset might be able to do? I imagine you can just grab the ptr address, increment it and then set it using that call

[mint iron]

yeah because it returns a copy/value you'd need to buffer[i] = new or use the ptr to make a ref

[amber flicker]

hmm but Property or indexer BufferAccessor<myBufferElement>.this[int] cannot be assigned to -- it is read only so buffer[i] = new doesn't work

bit surprised to have to use memset? Is it safe (other than the usual NativeDisableParallel and being sure to never write to the same index obviously)

[mint iron]

just a sec, ive got the example working

[amber flicker]

oh nice

[mint iron]

https://hatebin.com/aqdrgwawcg

[amber flicker]

hmmm... this is an asset to be sold.. was really hoping to avoid having to set a project to support unsafe

[mint iron]

😦

[amber flicker]

wait.. the 'Normal assignment' works for you?

[mint iron]

yep

i checked in entity debugger the count was going up

and no exceptions

[amber flicker]

I need to get my head around something... doesn't that for(int i = 0; i < bufferAccessor.Length; ++i) belong within a for (int i = 0; i < chunk.Count; ++i){} loop?

oh wait hang on let me re-read

[mint iron]

yeah its a bit confusing, i was thinking the same thing, its because its the same pattern as GetNativeArray<T>() where it grabs the whole chunk.

    [BurstCompile]
    public unsafe struct MyChunkBufferTest : IJobChunk
    {
        public ArchetypeChunkBufferType<MyData> SelectionBufferDataType;
        public ArchetypeChunkComponentType<Translation> Translations;
        public ArchetypeChunkEntityType Entities;

        public void Execute(ArchetypeChunk chunk, int chunkIndex, int firstEntityIndex)
        {
            var entities = chunk.GetNativeArray(Entities);
            var bufferAccesor = chunk.GetBufferAccessor(SelectionBufferDataType);
            var translations = chunk.GetNativeArray(Translations);

            for (int i = 0; i < entities.Length; i++)
            {
                var entity = entities[i];
                var translation = translations[i];
                var buffer = bufferAccesor[i];
                var bufferPtr = buffer.GetUnsafePtr();

                for (int j = 0; j < buffer.Length; j++)
                {
                    // do stuff.
                }
            }
        }
    }

[amber flicker]

yea - I got there 🙂 thanks!

[amber flicker]

@vagrant surge in my small repro it looks like you're right - not that big a difference... shoot... I need to investigate more.

Small aside.. how come I can write ```[BurstCompile]
struct TestJob : IJobForEachWithEntity<...>
{
public NativeArray<TestData> testData;

  public void Execute(Entity entity, int index, ref ...)
  {
           testData[entity.Index] = new TestData() { Test1 = index };
    }

}without testData requiring a [NativeDisable...] attribute? I've been using IJobChunk most places but does IJobForEach provide something extra for parallel writing? Or more likely have I misunderstood what it's actually doing? When I doreturn new TestJob()
{
testData = testData,
}.Schedule(query, inputDeps);``` that's not by copy right?

[mint iron]

it does copy something like 20-30 bytes for the ptr, length, allocator label safety etc but it doesn't copy the actual array data, that stays exactly where it was.

[amber flicker]

cool.. just checking I'm not crazy... any ideas on why it's safe to write to it in parallel? in IJobChunk I'd usually have to add the [NativeDisableParallelForRestriction] tag

[mint iron]

nah not sure 😦 but IJobForEach is effectively sugar on top of an IJobChunk so you'd think it would behave the same!

[amber flicker]

er maybe I just had the burst safety checks off and that's the difference 😅

[dire frigate]

anyone know if there's been updates to the workflow to getting the update method to run in fixedupdate?

[coarse turtle]

none so far

that I know of

[safe lintel]

hopefully unite will bring something new

[frosty siren]

Can i store CDFE/BFE inside another struct like and use attributes?

struct SomeJob : IJob
{
    public CDFE_Container;

    public void Execute() {...}
}
struct CDFE_Container
{
    [ReadOnly]
    public ComponentDataFromEntity<C> C_CDFE;
    [NativeDisableParallelForRestriction]
    public BufferFromEntity<B> B_BFE;
}

[prisma anchor]

Does it take several seconds to press play when using Unity Physics v0.2.2 for anyone else?

[merry oasis]

hi all, we're doing some collision detection and resolution a were wondering if Jobs has some sort of Multiple Producer Single Consumer process that we can easily use

the goal being that collisions will be processed by the consumer thread as the collisions are being calculated instead of just being thrown in a collection and calculated afterwards

[safe lintel]

@prisma anchor just as fast as it was in previous versions to enter play for me

[prisma anchor]

Hmm. Not doing an accurate count, its like 15 secs for me to actually get in.

I just upgraded another project's Physics from 0.2.0 -> 0.2.2. There is a noticeable slow down.

[safe lintel]

all i can suggest is try to delete the library, if that doesnt work sounds bug report worthy

[slow epoch]

Is there any site to check for ecs changes?

It's been a while since i've used it and i dunno what changed

[dull copper]

@slow epoch,
from Entities -package: https://docs.unity3d.com/Packages/com.unity.entities@0.1/changelog/CHANGELOG.html
from DOTS sample repo: https://github.com/Unity-Technologies/EntityComponentSystemSamples/blob/master/ECSSamples/ReleaseNotes.md

oh wait the samples changelog doesn't really tell much for the ECS changes anymore

it used to be super informative about changes in past

[slow epoch]

The first one works, thanks!

[dull copper]

learning from this, I added the DOTS package changelogs to the pinned message on this channel

[gusty comet]

Hello, I started today with ECS and there is something that I dont undestand. If know that I can have x entities, with theis own parameters, but in the moment of the behaviout wont they work the same? Since the system tell to all to do the same

    [SerializeField] private Material material;
    private void Start()
    {
        EntityManager entityManager = World.Active.EntityManager;

        EntityArchetype entityArchetype = entityManager.CreateArchetype(
            typeof(LevelComponent),
            typeof(Translation),
            typeof(RenderMesh),
            typeof(LocalToWorld),
            typeof(MoveSpeedComponent)
            );

        NativeArray<Entity> entityArray = new NativeArray<Entity>(5, Allocator.Temp);

        entityManager.CreateEntity(entityArchetype, entityArray);

        for (int i = 0; i < entityArray.Length; i++)
        {
            Entity entity = entityArray[i];

            entityManager.SetComponentData(entity,
                new LevelComponent { level = UnityEngine.Random.Range(10, 20) });

            entityManager.SetComponentData(entity,
                new MoveSpeedComponent{ moveSpeed = UnityEngine.Random.Range(1, 10) });

            entityManager.SetComponentData(entity,
                new Translation{ Value = new float3(UnityEngine.Random.Range(-8f, 8f), UnityEngine.Random.Range(-8f, 8f), 0) });

            entityManager.SetSharedComponentData(entity, 
                new RenderMesh {mesh = mesh, material = material });
        }
        entityArray.Dispose();
    }```

For example with that, they just move up and down

how you can tell to one entity to move to the right without doing it to all

[mint iron]

@gusty comet you would know which entity you want to change, and instead of looping through every entity, just do stuff for the ones u want to change... You could isolate an entity by making it have a component that the others don't have. Then you can use GetEntityQuery() to make a query that lets you filter just to entities matching a specific set of components.

[tawdry tree]

Does anyone know of a public game of life/cellular automata sample?
I feel like the techniques used for that would be somewhere in the intermediate range, where we lack good samples.

[mint iron]

this could be cool https://www.youtube.com/watch?v=9U0XVdvQwAI&feature=youtu.be

also, list of games on github https://gist.github.com/roachhd/d579b58148d7e36a6b72#user-content-arcade

maybe http://hextris.io/ (https://github.com/hextris/hextris/) or a dirt simple tower defense?

[tawdry tree]

I have already suggested tower defense to several people as a game/genre to try for getting into ECS, as it's mechanically simple.

But I was asking specifically about ECS game of life/cellular automata as a sample, becuase I have an idea which would use a lot of the same mechanics

[mint iron]

yeah, and u could ramp up the projectiles/effects the towers shoot and the nubmer of enemies which would be cool

https://github.com/mwharrisjr/Game-of-Life

[tawdry tree]

Still not made in Unity ECS, and that's what I'm looking for

I have implemented the game of life before

Just.... translating that to ECS...

[mint iron]

oh sorry i misunderstood, i thought u were intending on creating some examples

[tawdry tree]

Ah, no, I was asking if anyone knew of an existing one. Though if I do stop putting it off and actually sit down to make something in ECS, I'll be sure to make the repo public

[vagrant surge]

i dont think game of life makes any sense in ECS

you implement it on a bitmask usually

good use case for Burst tho

[grave turtle]

Hello, what's the proper way to associate an array of matrices with an IComponentData (DynamicBuffer? BlobArray?). The size won't change but I need to read/write from jobs

[safe lintel]

i would assume you would want dynamic buffers for most situations needing arrays

[junior moon]

I'm following some tutorials and in it you make an entity using EntityManager.CreateEntity, but mine returns an error, do we now always need Archetypes?

[pliant pike]

what's the error message?

[junior moon]

'World' does not contain a definition for 'EntityManager' and no accessible extension method

[pliant pike]

how are you using the EntityManager? what's the code?

[junior moon]

https://pastebin.com/v0bHHCjQ

it should be the same as the tutorial but its from a few months back so I guess the tut is outdated

[pliant pike]

weird it should work fine like that

[junior moon]

i even tried restart unity 😢

[safe lintel]

are you up to date for your entities package?

[junior moon]

yeah, 0.0.12?

[safe lintel]

latest is 0.1.1

[junior moon]

wait how come I can only see up to 0.0.12?

oohhh, nvm...

[wooden canopy]

@prisma anchor in the forums they explained that the slowdown after updating unity physics, is due to the burst jobs used for collider generation being compiled on start instead of on demand, this should improve when they release an updated version of burst that cache compiled jobs

https://forum.unity.com/threads/unity-physics-discussion.646486/page-17#post-4963058

[prisma anchor]

Lol. @wooden canopy That was me who asked the question. Thanks for the follow up!

[wooden canopy]

ups 🙂

[slow epoch]

Is there any way to load data to an entity like an image or sound without being a gameObject and converting it to an entity?

Or being referenced from a singleton monobehaviour

[coarse turtle]

For audio you can store clips in a blob asset, also I'd take a look at their DSP Audio, for images you can also store images too in blob assets

[slow epoch]

Is there any docs about blob assets? I haven't heard of them

[unborn bear]

Does making a job dependent on another freezes the main thread ?

[safe lintel]

@unborn bear that doesnt sound right. @slow epoch theres nothing official afaik, only stuff in the forums if you search or from going through the package code. i think physics might have some good examples, also a file called BlobificationTests.cs might have some info

[unborn bear]

weird, I made a mesh slicer where jobs share the same NativeArray, and for some reason I am having some spikes whenever I perform a cut

[coarse turtle]

@slow epoch take a look at BlobificationTests.cs in the package, it goes over the basic way on how to construct a BlobAssetReference<T> which you can store in a struct as an IComponentData

[pliant pike]

what's the best way to destroy a bunch of entity's? Destroy 1 roughly every 0.2 seconds or wait till they build up and destroy a bunch say every 5 seconds?

[safe lintel]

i think the most efficient way is to destroy using entitymanager taking an array of entities - but whether you really need to go this far for optimizing is another thing entirely

[pliant pike]

yeah I'm using an EntityManager in a component system, I'm destroying entity's when they reach a waypoint individually. So it could happen quite frequently

It doesn't seem to be impacting performance that much so I guess it shouldn't matter that much

[unborn bear]

What is the correct way to wait on a job to finish ? I am using a coroutine but it just feels wrong

[pliant pike]

JobHandle.Complete()?

[unborn bear]

that would freeze the main thread because it would force it to wait until job completion

[coarse turtle]

if you need a flag, there's an IsCompleted field

[pliant pike]

yeah if you use the Iscompleted with JobHandle.Complete I think that's probably the best way

I don't think it freezes the main thread or anything it just jumps to the main thread

[coarse turtle]

.Complete forces the thread to immediate evaluate

[unborn bear]

Yes, I know that flag. What I mean to ask is: How do you guys wait on a job to complete ? Do you guys use a coroutine OR something else ?

[pliant pike]

yeah you can use it to force the job to only run once or just make sure its completed before doing something else

[coarse turtle]

I just schedule the Job and let Unity's job system handle the completion

https://blogs.unity3d.com/2018/10/22/what-is-a-job-system/

There's a brief section abt 'Scheduling early and completing late.'

[unborn bear]

then you are basically "freezing" mainthread aren't you ? If you make a job a call complete right after it, you did not run anything a parallel

you just processed stuff with a job

[coarse turtle]

me, i very rarely call JobHandle.Complete()

I only need to call it if I absolutely need to evaluate something immediately

[unborn bear]

You using jobs with ECS ?

[coarse turtle]

Yes

[unborn bear]

Then that is probably why you have a different perspective. I am using only the job system, no ECS.

[coarse turtle]

Ah, that makes sense - I never really used the JobSystem outside of ECS yet

[unborn bear]

I see. The only way I can think of waiting on a job to complete is whether polling it on an Update function (bad in my opinion) or using a coroutine to do that (better but dirty)

I noted some weird gotchas like, making IJobs share the same NativeArray (in other words, make an IJob depend on the completion of another IJob) freezes the mainthread apparently (maybe they call '.Complete' internally ?)

[coarse turtle]

yea, since the entity manager is the one handling the jobs (I'd imagine), they just call Complete() on the job

for (int t = 0; t < m_TypeCount; t++)
    {
        ((JobHandle)(&m_ComponentSafetyHandles[t].WriteFence)).Complete();
        int readFencesCount = m_ComponentSafetyHandles[t].NumReadFences;
        JobHandle* readFences = (JobHandle*)((byte*)m_ReadJobFences + (long)(t * 17) * (long)sizeof(JobHandle));
        for (int r = 0; r != readFencesCount; r++)
        {
            ((JobHandle)((byte*)readFences + (long)r * (long)sizeof(JobHandle))).Complete();
        }
        m_ComponentSafetyHandles[t].NumReadFences = 0;
    }

[dull copper]

https://github.com/Unity-Technologies/ScriptableRenderPipeline/commit/777f7d938c121bafb9231b5dcffacaa90bb052b9

that's apparently some vertex skinning for DOTS on HDRP

[light sage]

back at it again, and straight back to ECS at that. Has been a while, did we get Boolean functionality yet or still just using byte vars?

[amber flicker]

Bool is cool now 😁

[low tangle]

@dull copper oh my god yes

might be switching to hdrp

[gusty comet]

Should we avoid using ComponentDataProxy<IComponentData> (is it going to be deprecated, or is there some negative impact on performance)?
I know there's a bunch of different ways you can create entities - I've been using IConvertGameObjectToEntity

[low tangle]

going to be deprecated,
yes
or is there some negative impact on performance
kindof, it didnt scale well in performance or development complexity

conversion workflow is the future, its also getting some love soon from what I hear

[coarse turtle]

ooh that's very nice to hear, I'm curious as to what new things they'll add lol

[low tangle]

same

I want it

please more dots stuff

cant get enough

[honest dirge]

Conversion workflow needs a lot of love before it's viable.

[vagrant surge]

but what about being able to... yknow.. click on an object in the editor view and seeing its params

[pliant pike]

or a system being able to see its variables while running would be nice

[coarse turtle]

doesn't the entity debugger already do that? if by variables you mean entities

[vagrant surge]

but you cant find an entity easily by clicking on it in viewport

[honest dirge]

You could write a system that does that if there is a graphic component

In theory

[tawdry tree]

As long as it has a position, you could make a crude ray-marching/tracing 'what did I just click at' system

Some stuff like that should be standard, I think. Maybe not part of the base package, but a package with some helpful tools for debugging and/or general development of ECS wouldn't be a bad idea.

[vagrant surge]

thats something that baffles me

there is tons of stuff they can do that would make ECS usable for development right now

but they dont

makes me think that the ECS team is just Mike Acton and 2 friends

[low tangle]

hey man I've built my whole game on ecs

its rough but doable

that debugger is more than enough info imo

but its getting worked on, just gotta wait another week for their anouncements

[tawdry tree]

What kind of game is that by the way? Like, genre and stuff. ECS fits certain types of games better than others...

[low tangle]

vr sandbox

user ran content

[safe lintel]

im surprised they dont have rough data modifying on the entity debugger

really need some enterprising individual to make an asset store ecs inspector 🙂 get in on that early user cashflow!

[coarse turtle]

lol, potentially get hired by unity to build the editor 😉

[safe lintel]

win win scenario

[low tangle]

too much work otherwise I'd be making more ecs stuff

I'm looking to just open source all my stuff in the end though

ecs is too good not to share

[coarse turtle]

same except just a module of it

been working on making a 'primitive' UI system cause uGUI is a bit jittery

[low tangle]

yeah UI is one I really want to port to ecs soon

I've gotta make it vr friendly though, in world

so far the two ecs UI's I've seen were 2d in screenspace

uGUI sucks imo

[coarse turtle]

ah, which UI's were they, if you still remember?

[low tangle]

uh I can find you one

https://forum.unity.com/threads/source-code-dotsui-open-source-ui-framework-for-dots.715880/

just had a post

didnt even have to search

aw yis

all his info was in the previous one

https://forum.unity.com/threads/showcase-pure-dots-ui-system-detailed-description-feedback.688531/

[coarse turtle]

ah very cool, I'll look into it for a bit

[low tangle]

👌🏻

[light sage]

most still doing hybrid, or is pure more feasible yet?

guessing a lot has been moved behind the scenes on a lot more of the standard tools/compilation and especially the newer packages

[wanton girder]

Hi, I have a question about what is destroyed when a gameobject is destroyed. Suppose I create an instance of a non-monodevelop class in a component (therefore create with new), then store a reference to that instance in a second component. If I destroy the first component that created that instance, would it destroy the instance as well?

My suspicion is that in this case only references to the created instance would be destroyed in the destroyed component and it would be GCed later if there were no references to it. Just need confirmation from the enlightened.

[dull copper]

I removed hybrid for now, it is always breaking with bleeding edge HDRP api changes and it doesnt work at all with DXR

[mint iron]

i can understand your frustration, im using hybrid with LWRP and its broken in a different way with pretty much every editor version since 2019.3.0a8 (like 8 versions ago). There's always packages trying to access editor API functionality that no longer exists or doesn't exist yet.

so you have to trial and error to find the magical package version soup that works, but then each package brings its own bugs that mess you up in other ways.

[dull copper]

I can fix the HDRP API changes from the hybrid package (done that many times), but fixing raytracing is like totally another level thing 😄

that could like require engine level changes

[mint iron]

@wanton girder a non-UnityEngine.Object class exists independent of where it was created, so generally it will be kept alive as long as there is a reference to it somewhere. I say generally because unity's garbage collector is different than normal .net GC. I am not certain it won't do something weird like take a while to figure out it can dispose of it.

If you have a GameObject that has 2 components, both referencing the same instance of foo; then you remove the first component. It will still exist because the second component holds a reference to it.

UnityEngine.Object derived classes act a little differently in that the class will start resolving to Null when it gets destroyed on the Native side regardless of if you have references to it.

[worn stag]

anyone has an example of conversion scriptable object to blobAsset?

[coarse turtle]

public class Bucket : ScriptableObject {
  // some data here
}

public struct BucketWrapper { public Bucket Bucket; }

public struct BucketPtr { public BlobPtr<BucketWrapper> Ptr; }

public struct BucketBlob : IComponentData { public BlobAssetReference<BucketPtr> Blob; }

// The func that will create the blob
public void ConstructBlob() {
    var builder = new BlobBuilder(Allocator.Temp);
    ref var root = ref builder.ConstructRoot<BucketPtr>();

    ref var wrappedBucket = ref builder.Allocate(ref root.Ptr);
    wrappedBucket.Bucket = some_scriptable_object_of_type_bucket;

    var blobAsset = builder.CreateBlobAssetReference<BucketPtr>(whatever_allocation_type_you_want); // Mostly you'll want a persistent one instead
    builder.Dispose()
    return blobAsset;
}

when you destroy the entity with the blob, you'll likely want to release the blobassetreference too unless you're managing the ptr yourself externally

[worn stag]

thank you

[prisma anchor]

Hi Everyone, I had a question. If I want to represent a ring buffer of IComponentData, how do I represent that? i.e

SnapShot : IComponentData {
int tick;
int[] createEntityIDs;
int[] updateEntityIDs;
int[] deleteEntityIDs;
}```

on the client & server i need to store an array of these 'Snapshots`. Would I make another component simply storing this data?

[coarse turtle]

You can likely store them in DynamicBuffers, b/c you can't store managed types in IComponentDatas

[mint iron]

@prisma anchor fholm#9083 has an unsafe RingBuffer on github that might be of interest. https://github.com/fholm/UnsafeCollections/blob/master/UnsafeCollections/Assets/UnsafeCollections/Collections/UnsafeRingBuffer.cs

[urban rivet]

Where are we with ECS these days? Is it likely to be changing much from this point?

[slow epoch]

You should wait a bit for the copenhagen

It's the next week

And i'm sure they'll announce new features

[trail burrow]

@urban rivet you going to unite? 🙂

[trail burrow]

@dull copper i see ur going to unite

how did you get ur fancy tag?

[dull copper]

look at the #📢┃announcements

it has instructions

are you going too?

[trail burrow]

i am

so now i have to spend 4 days trying to avoid you... ish 😄

❤

[dull copper]

yeah, we've talked online for like 6-7 years without any IRL contact, why ruin it now 🤔

[trail burrow]

@dull copper exactly

there we go

someone gets it

😄

[dull copper]

to be serious tho, I'm kinda worried about not being the anonymous guy, I like keeping some distance to my online "personality"

[trail burrow]

@dull copper yes same

i prefer to stay in the shadows 😄

[low tangle]

its weird at first but its nice after the inital wears away

at least you already have something in common :^)

[urban rivet]

@trail burrow couldn't make it this year. Unity was really kind to extend an invite too

[trail burrow]

ah 😦

sadface

[urban rivet]

I probably won't come until I've got some headway done on my game out of sheer procrastinatic embarassment tbh

haha

can't turn up red faced without a decent demo!

[low tangle]

wheres that game ;)

[tawdry tree]

Clearly not here, since hippo's nowhere to be found 😛

[low tangle]

🚶🏻 💨

[pliant pike]

I dont suppose anyone knows why if I use the below code then the whole system just dissapears or destroys itself

public class CountRobotsatWaypoint : ComponentSystem
{
    public Entity countent;

    protected override void OnCreate()
    {
        if (!EntityManager.Exists(GetSingletonEntity<CountRobots>()))
        {
            countent = EntityManager.CreateEntity(typeof(CountRobots));
            EntityManager.SetName(countent, "CountRobots");
        }
        else
        {

        }
    }

[tawdry tree]

Systems will disable when they have no more 'targets', if that's the entire system then they'd detect 'oh well, nothing more for me to do here' and disable it.
I think it was [AlwaysUpdateSystem] which stops that behavior. The ondisable or whatever method can be useful for figuring out when it happens.

[pliant pike]

that's not all the code I have a foreach loop as well, which seems to activate after but then the whole thing just disappears

its not even disabled its like the whole system is destroyed, its really weird behaviour

and using AlwaysUpdateSystem doesn't work

[safe lintel]

is there anything in the update loop for the system to actually work on?

[pliant pike]

yep, I'll post the whole code maybe that will help

[safe lintel]

also if you stick something in like Debug.Log("test"); before the update foreach does it force it to run?

[pliant pike]

[AlwaysUpdateSystem]
public class CountRobotsatWaypoint : ComponentSystem
{
    public Entity countent;

    protected override void OnCreate()
    {
        //if (!EntityManager.Exists(GetSingletonEntity<CountRobots>()))
        //{
            countent = EntityManager.CreateEntity(typeof(CountRobots));
            EntityManager.SetName(countent, "CountRobots");
        //}
        //else
        //{

        //}
    }

    protected override void OnStartRunning()
    {       
        //if (!EntityManager.Exists(GetSingletonEntity<CountRobots>()))
        //{
        //    countent = EntityManager.CreateEntity(typeof(CountRobots));
        //    EntityManager.SetName(countent, "CountRobots");
        //}
       
    }
    
    protected override void OnUpdate()
    {

        Entities.WithAllReadOnly<ToCurrentWaypoint>().ForEach((ref ToCurrentWaypoint dude) =>
        {
            if (dude.value == 1)
            {
                if (HasSingleton<CountRobots>())
                {
                    var countem = GetSingleton<CountRobots>();
                    countem.value += 1;
                    Debug.Log("Why is this not working");
                }

            }

        });
    }    
}```

[tawdry tree]

Does any entities with ToCurrentWaypoint exist?

[pliant pike]

yep

[tawdry tree]

Hm, weird

And the force update...

Sounds like you might have struck a bug, or at least a edgecase that should have a warning or documentation

[pliant pike]

I have to have some way of checking that CountRobots exists because if I did it in Onrunning then it created 2

[safe lintel]

i mean stick the debug before the ForEach and if it shows in the debugger as running, double check that the system does or doesnt have entities to process

[tawdry tree]

Theoretically those entites could lose their components or be destroyed, though, are they still in the entity debug menu thingy?

[amber flicker]

just to say, the way it looks to me, if (HasSingleton<CountRobots>()) should be outside the Entities.WithAll anyway no? You wouldn't want to check it for each entity

unless you create it in the loop or something I guess

[tawdry tree]

That's a good point too, a singleton check should definitely be outside of any loops (unless, as you said, the singleton might be created/destroyed in the meanwhile)

[pliant pike]

yeah good point

and the DebugLog works, it slows it down like hell

[tawdry tree]

Then it sounds like aggressive optimization from the system-system (whatever keeps serial killing innocent systems)

[amber flicker]

It might just be a lambda thing they don't yet support?

[pliant pike]

yeah I'll post it as a bug for the Unity devs see what they say

[coarse turtle]

You can also do RequireSingletonForUpdate<T> in OnCreate

[pliant pike]

It seems like I've found another bug, for some reason OnStartRunning is run twice

its effecting another system, making it run way to long and not in the way I want

I don't know if I'll use a singleton you have to use SetSingleton to set it and I'm not sure if that works for what I want

[mint iron]

you have to be careful with OnStartRunning because every time a system is not run because of RequireForUpdate or from the queries used it will cycle both OnStartRunning OnStopRunning again. So you really can't use it for initialization. I've been bitten by that numerous times.

[pliant pike]

the trouble is I cant use OnCreate either as the Gameobjects don't exist or the other system doesn't exist

[hollow sorrel]

it's weird how oncreate runs before monobehaviours, but onupdate runs after monobehaviours

or other way around

i forget

[mint iron]

i know what you mean, its messed me up where singletons from ConvertToEntity i needed hadn't been created yet in OnCreate 😦 gets frustrating.

[pliant pike]

yeah I can kind of get around it but the Systems really are confusing how they are run

protected override void OnStartRunning()
    {
        if (MoveWaypointslist.Length != WaypointGameobjects.Length)
        {
            WaypointGameobjects = GameObject.FindGameObjectsWithTag("Waypoints");
            for (int i = 0; i < WaypointGameobjects.Length; i++)
            {
                var tempwaypoints = WaypointGameobjects[i].GetComponent<Transform>();
                var someotheval = (float3)tempwaypoints.position;
                MoveWaypointslist.Add(someotheval);
            }
        }

    }```

that seems to fix it for now

[coarse turtle]

hmm just my curiousity

is it possible to just have a gameobject with a list of points, convert it and add the list of points to a buffer on an entity

is it possible to just have a gameobject with a list of points, convert it and add the list of points to a buffer on an entity

[mint iron]

yep

[pliant pike]

yeah of course

[coarse turtle]

I think the caveat is you'd have to write some scene view tools if you wanna edit them in editor time

[mint iron]

yeah it woudl be more complicated if you wanted to keep them in sync or edit the entities directly from handles or something

[pliant pike]

that's what I'm basically doing except they are separate gameobjects and I'm adding them to a list instead

[coarse turtle]

yea, jw

[pliant pike]

Does anyone else here use Rider? Does anyone know if it's worth using?

[coarse turtle]

I tried it for a bit, it's certainly nice with a ton of features

local history view seems to be really nice for a lot of ppl

but the problem I had was that it slowed down a lot for me trying to query a lot of things i just ended up sticking with omnisharp + vim

[pliant pike]

I guess its worth trying, a guy is giving out free promo codes for six months

[coarse turtle]

haha yea, i'd certainly give it a shot, it takes some time getting used to the hotkeys it has

[pliant pike]

https://www.youtube.com/watch?v=luibp9pmyk8

if anyone else is interested

[amber flicker]

@pliant pike just in case you didn't know, in OnCreate you can use GetOrCreateSystem<> and if the system doesn't exist at that point, it will create it then and there. If you require that system to have run, then possibly you want your second system to require a Created component or something to be present on the entities you want to change. There are also [UpdateInGroup] attributes to specify order of systems where needed.

[pliant pike]

Oh ok, thanks @amber flicker

[gusty comet]

Edit: Figured it out, problem with other system, not with input system

[coarse turtle]

Might be this issue here

PlayerInput input = inputs[j]; <- this might be a copy of inputs[j] which means that you're not modifying what's in the chunk

[gusty comet]

yea but in the end I'm writing to inputs[j]

[coarse turtle]

oh totally didn't see that

[gusty comet]

so basically that's the only line I'm confused about, since the debug.log is firing correctly when there is input

[coarse turtle]

hmm, it's the only system modifying the input right? you dont have a separate system zeroing it out immediately after I assume

[gusty comet]

I have another system which is ReadOnly accessing it

This is not using a job, so I'm less familiar with how to handle it

[coarse turtle]

same here 🤔

[gusty comet]

maybe there's some sort of timing issue going on, since the other system is using a job

🤷

[coarse turtle]

GetArchetypeChunkComponentType<PlayerInput>(false); <- false means you can read/write right?

[gusty comet]

yea

[coarse turtle]

hmm, 🤔 yea i wouldn't know till I try it out, srry

[honest dirge]

Does anyone have any thoughts why modifying Translation component before CopyToGameobject runs would cause so much jitter?

[prisma anchor]

@coarse turtle & @mint iron thanks for the info

[gusty comet]

@coarse turtle I figured out what was happening.
The Input system was doing it correctly and the chunk was updating.
In my movement system, I was switching from Translation + Rotation to LocalToWorld. However, I had left the old Translation and Rotation archetypes in the code (I had added LocalToWorld to the query as well), and there's a catch
https://docs.unity3d.com/Packages/com.unity.entities@0.0/manual/transform_system.html
User code may write directly to LocalToWorld to define the transform for an instance, if no other transform components are associated with the same entity.So after I completely removed the Translation and Rotation archetypes from the query and job, everything started working (it's confusing, since you can still see those types on the entity in the entity debugger....)

I guess the documentation means you can write to LocalToWorld as long as no other transform component is part of your query/job

[coarse turtle]

Ah makes sense

[slow epoch]

I have a design question, if i'm working in a Console system with commands and stuff. I suppose the Console itself is a system and the commands are IComponentData, but where is the logic shared with a given command?

Can a component store logic and that logic be executed by a system?

[mint iron]

a component could contain logic, but its sort of against the concept of having a clear data/logic separation.

[slow epoch]

What about having a delegate in the Component and executing that delegate inside a system?

[mint iron]

Sounds dubious, can you tell us more about what you're trying to accomplish?

[slow epoch]

A Console system like CSGO where you write commands and does stuff

And i still get problems designing for ecs

I don't know if it's about not having enough tools still for ecs or that i don't know how to approach it

[mint iron]

I assume you'll have a lot of different commands, so having a different system for each command would not be practical. Leaving you with switching somehow within a single system.

You could do a different component per command, or have a type within a command component.

The logic is still in the system, the component contains whatever info is needed to complete the command.

[slow epoch]

Yeah, the game is gonna be based on the console and i can end up having hundreds of commands

So a system per command is not an option

[mint iron]

A simple approach would be to have a different component for each command, and different IJobForEach<MyCommand> for each one. Then the job becomes the container of your command specific logic.

But im not sure how that will scale if you're having hundreds of them.