ok...that was already answered, so here it goes the reasoning:
#archived-dots
night venture
If i were going to load an "external 3D model" on my ECS system, would I need to instantiate a gameobject? would I use a prefab?
zenith wyvern
Typically you would author a traditional gameobject with your mesh attached then use the conversion system to convert it to an entity at runtime
night venture
how it's supposed to load ECS "visible objects" (so far to my understanding, i added typeof(RenderMesh) to my archetype
so, I need gameobjects to load 3D meshes into DOTS !?
zenith wyvern
You don't "need" to, but the expected workflow for dots is you author in the editor with traditional gameobjects/prefabs then the conversion system converts those into entities for runtime
night venture
ok....so the expected workflow (until DOTS supports otherwise) is to have gameobjects, prefabs and such things and convert them into entities at runtime.
zenith wyvern
Doing it that way is a lot easier than trying to manually add all the necessary components at runtime as well, in my experience
night venture
yes...it might be easier, but not "pure ecs". and, tbh, sounds quite "temporary"
so far, seems the mesh is my only component left
zenith wyvern
It is "pure" ecs at runtime. The point is to leverage the existing editor tools at edit time.
At runtime you would only be dealing with entities
night venture
I'm a total newbie, but for me something like Resources.Load() or Assets.Load would have made sense instead
zenith wyvern
And no, it isn't temporaey as far as I know
It's the expected way to make dots games
night venture
It's the expected way to make dots games
sounds weird to me, tbh
night venture
@zenith wyvern seems this is what i was looking for: Resources.GetBuiltinResource<Mesh>("Cube.fbx"); Going to test...
stone osprey
Is there a way to check if an Entity is still "alive" ?
Like when you store the entity struct somewhere and you wanna check later on if you can still "use" that entity ?
north bay
EntityManager.Exists
stone osprey
Ah great, thanks ^^
night venture
ok...so my entities are being created but aren't visible.
turbid sundial
You'll want to use https://docs.unity3d.com/Packages/com.unity.rendering.hybrid@0.11/api/Unity.Rendering.RenderMeshUtility.html @night venture
This utility adds all the necessary classes to hybrid rendered entities
night venture
@turbid sundial hello STAN. let me review that. stop moving your arms. im not going to buy you anything
do you have a working example by hand?
stone osprey
When does an entity query update ? Im currently having the issue that i "1. Create an entity, 2. Search that entity ( Same frame )" and it doesnt show up...
night venture
@stone osprey a scope issue perhaps?
stone osprey
I dont think its a scope issue... because i create that entity by the entitymanager and it wont vanish once out of scope
night venture
in which method do you create the entity? is the entity object at class-level or method level?
stone osprey
EntityManager.CreateEntity();
night venture
that returns an object
stone osprey
What ? ^^
night venture
which u have to store somewhere
stone osprey
Uhmm... i dont think so, because the ecs itself stores the entities
The struct is just an identifier
Like an id
night venture
...im not sure about what you say...but my entities doesn't dissapear xD
let me explain myself:
stone osprey
It doesnt dissappear...
var entity = manager.CreateEntity();
// Run query to get all entities
// its not there ?!
Thats what is happening...
So the query doesnt find it... YET... so im not sure if i need to wait one frame
or if it normally should update instant
night venture
try Entity e = manager.CreateEntity();
and, even more...
Entity e at top, at class level
night venture
let me write what works for me so far...
stone osprey
I just have the problem that my query does not update instantly... and i just wanna know if thats normal or not
Oh i found my problem... i was still using an entity command buffer... thats why it took one frame for the query to update
night venture
then, my code is useless xD
zenith wyvern
ok...so my entities are being created but aren't visible.
light mason
odd ridge
yes
gusty comet
learning dots rn
ive faced fps drop after instantiating ton of planets
will dots help
stone osprey
BTW is it possible to render a mesh with multiple materials in the hybrid renderer ?
stone osprey
So is it possible or not ? ^^
Btw... are there any docs on how exactly the hybrid renderer works ? I would assume it offers such a good performance because it "instances" all its drawn meshes. Like opengl does with "DrawMeshInstanced". Is this right ?
karmic pilot
internally it still uses the srp batcher and URP, but as it has way better data layout. Gathering the data needed to send to the gpu each frame takes considerably less time. It is not a completely new rendering pipeline
stone osprey
@karmic pilot Thanks ! So it basically loops over all entities with a renderer each frame, gathers its data ( Transform, Rotation, Material, Shader ) and sends them to the GPU by either Opengl,Vulkan,DirectX and draws them instanced ? I just wonder how they exactly do that... because i saw examples of the hybrid renderer with 500K entities being drawn with nearly 60 fps... but in some pure c/c++ opengl examples 200K instanced objects are already lagging. So they are somehow more performant then pure opengl which makes me wonder why
karmic pilot
comparing hybrid renderer to opengl is like comparing apples to stones 🙂 the hybrid renderer cannot accelerate the underlying rendering backend (vulkan/dx/opengl), it can only improve the effort needed to prepare the data for them (render data + commands).
stone osprey
@karmic pilot Hmmm... but how its possible that its faster than a pure opengl instanced renderer ? :/ I read somewhere that it only sends the "changes" to the gpu. But as far as i know opengl/directX force you to send everything everyframe
karmic pilot
a hybrid renderer with opengl backend faster than a pure opengl optimized solution?
stone osprey
Isnt this the case ? Atleast i recently saw an dots example where someone drew about 500k entities. But the official opengl "instanced" renderer with those asteroids is only able to draw about 200-300k. Everything higher lags
karmic pilot
no idea which comparison you saw, but in general the hybrid renderer can improve performance by quite a bit compared to the old fashioned game object rendering due to more efficient queries on the data. It always depends on what your bottleneck is ... memory access, cpu computation, gpu bandwith, gpu computation... the hybrid renderer is no silver bullet
where it really fails horribly for example is using unique meshes... if you have 100k unique meshes, the hybrid renderer won't be able to save you (might even be far before that)
stone osprey
Alright, thanks ! Well thats understandable ^^ Its just possible to instance entities with the same mesh and material... and shader... Everything else is one more drawcall.
amber flicker
Has anyone here got a non-trivial dots project that when processing a low number of entities doesn't end up taking e.g. 1ms just to schedule jobs? I'm finding that I just can't get my main thread time down. The cost of scheduling (not doing work) is more than doing the same work in a traditional way on e.g. 500 gameobjects. I have to be running over more than 1000 entities before it's worthwhile. So far from performance by default. (I have many jobs and many components, this is all in a build, burst enabled even some ISystemBases where I can, reduced sync points, mostly unmanaged etc).
karmic pilot
if you have a small amount of entities, just .Run() on the mainthread
scheduling has an overhead that is only negligible if you are working on a lot of entities across multiple jobs
amber flicker
Not an option as it has to scale to a larger number of entities. I do swap to .Run() dynamically based on number of chunks where I can but it still doesn't help that much.
karmic pilot
are you jobs depending on results of previous jobs in the same frame ?
gusty comet
ehh tutorial stopped at this
Entities.ForEach()
```yup - I think the dependency a chain is a large part of my frame time. e.g. if I profile around Dependency = job.Schedule(Dependency) it seems responsible for about 30% of time across jobs
karmic pilot
does this necessarily have to happen in the same frame in order to work ?
since I don't know what you are doing, all I can do is randomly guess what might be the issue 🙂
amber flicker
yea - it's a tween library so it all has to be frame accurate sadly
karmic pilot
in general jobs for a small amount of entities has a larger scheduling overhead than the actual execution on the main thread, since as you saw you have to check for dependencies, it has to be wrapped into a job to be executed somewhere and then it has to be scheduled... this is a lot more work than a simple iteration on a handful of entities that might even fit into a single cache line alltogether
amber flicker
yes... what I didn't appreciate is how badly this scales for any non-trivial project.
karmic pilot
they are working on that 🙂
amber flicker
I also thought that burstable ISystemBases were supposed to improve things
but they don't unfortunately.. not yet anyway
do you know if that's because they haven't finished that or they're working on something else to help?
gusty comet
what if I want to add Class reference to componentData? https://i.imgur.com/qmafkvb.png
karmic pilot
I'll end by saying that the claim of performance by default in ECS / DoD only applies to many entities is not a correct assumption. DoD has a higher impact when you process more than one thing, but the expectation is that it is faster in all cases. If you have one thing, OO might be only a little bit slower while when you have many things it is a lot slower.
The current issues with low entity counts, are simply performance bugs that are in the process of being fixed.
this was at the end of 2019, no idea how much better it is since then .. haven't started using dots before mid 2020
amber flicker
yea.. I've been working on this for the past 2 1/2 years - waiting for this to be improved... as far as I've seen communicated, ISB's were supposed to be a large part of the answer
I'm at the point where I think I need to pause dev until they actually do something about it
but with no comms, I'm not sure if this is really something they think they can do more about
like you say, almost 2 years and very little improvement
karmic pilot
what if I want to add Class reference to componentData? https://i.imgur.com/qmaf...
if you want to use managed data, you need to use managed components (which are horribly slow in comparison)
amber flicker
what if I want to add Class reference to componentData? https://i.imgur.com/qmaf...
you can use a class ICD if that helps? e.g. public class myThing : IComponentData{ //managed stuff here }
karmic pilot
and Timboc, did you profile with checks/debugger on or release built player?
amber flicker
and Timboc, did you profile with checks/debugger on or release built player?
il2cpp release - though it's not much difference to in editor tbh
karmic pilot
does burst work with il2cpp? never tried
amber flicker
yup 👍
gusty comet
why am getting all these errors? :C https://i.imgur.com/fvHBDMk.png
amber flicker
why am getting all these errors? :C https://i.imgur.com/fvHBDMk.png
I think you've got a bit confused - using the ForEach parameters in the OnUpdate of a System - would recommend checking out the examples pinned in this channel
gusty comet
oh ok xD
stone osprey
@gusty comet If you have the same problem as i had ( Boxed structs... structs casted to objects ) and you wanna add/set them... i wrote an extension for that 🙂
rancid geode
if you want to use managed data, you need to use managed components (which are h...
Not horribly, just as slow as if you were using Monobehaviours
karmic pilot
thats why I said "in comparison" 🙂
gusty comet
why I can use .Schedule when ref Translation
but I have to use .WithoutBurst().Run(); when using my own componentdata?
frosty siren
why I can use .Schedule when ref Translation
but I have to use .WithoutBurst()....
AFAIK you can use ref with any IComponentData with Schedule() and with managed IComponentData, which you defining like class, you should use .WithoutBurst().Run()
gusty comet
OH so it had to be struct LMAO thanks
yea i didnt like that WithoutBurst cus i assumed it would be slower
frosty siren
yes, every .Run() will produce sync point which leads to all jobs will be forced to complete
gusty comet
oh ok
but i kinda wonder
what if your meshes are generated dynamically in monobehaviour script ?
maybe having one object with mesh generator, and he would supply the mesh dynamically to all the entities? (so mostly different meshes, eg 5 entities per mesh)
would that work
frosty siren
maybe having one object with mesh generator, and he would supply the mesh dynami...
you can make mesh generation method static and then use it in system which will set MeshRenderer component. But you must assume that MeshRenderer is ISharedComponentData and every unique value will produce extra chunk, so DOTS built-in renderer is not the best solution for mesh generation for now.
gusty comet
the generation is done with monobehaviour, i just want to "pass" generated mesh to dots entities
you mean RenderMesh component?
frosty siren
you mean RenderMesh component?
yes
gusty comet
yea but so far I know that i can access entities from entitymanager
and then i would also need to know which entity i need to access
half jay
I need to store NativeHashMap somewhere and get access in systems. What may be better create static class and store NativeHashMap in it or it can be store in component like blob asset? Also this collection should be updated and use in system with burst parallel jobs.
frosty siren
you can store it in another system
cerulean forge
How do you guys handle events with ECS? E.g. a mob has died and you need to react to that (run death animation, remove some components from the entity, drop loot, ...).
frosty siren
How do you guys handle events with ECS? E.g. a mob has died and you need to reac...
i see few variants
- Add to your entity some component and have a system that will react on it. Like DiedTag in your example. But every combination of components is one archetype and also archetype changing leads to memory copy.
- Create some event entity that will have components of "what happend" and "with what entity". And have a system that will react to this entity through link. But it's not really ecs, because you will need to access components not through query and not linearly
- Other clever approaches which i don't know about. Maybe you can search on forum. I saw some projects for dots event system.
cerulean forge
Yea that's pretty much what I've been reading on the forums
Was hoping there'd be a better way 🙂
sturdy rune
@stone osprey did you ever figure out your serialized/deserialize?
gusty comet
you can make mesh generation method static and then use it in system which will ...
But you must assume that MeshRenderer is ISharedComponentData and every unique value will produce extra chunk, so DOTS built-in renderer is not the best solution for mesh generation for now.
so you mean dots is best when values are the same (eg same mesh, same int value, same sound, etc) ?
is it useless when theyre not the same?
sturdy rune
He is talking about the shared component specifically creating a new chunk for every unique value
frosty siren
> But you must assume that MeshRenderer is ISharedComponentData and every unique...
like @sturdy rune has said it's only about ISharedComponentData. DOTS have no alternative for mesh rendering except shared MeshRenderer
stone osprey
@sturdy rune Yes kinda ^^ I actually just added a little feature to my deserializer. I call them events. And it kinda works like this, a deserializer receives a packet, deserializes it into a struct and passes it to the events. At this point we know the type and can use generics. If a event returns a certain value it wont box/return the deserialized struct. So i just hook in such a event, which basically enlists the deserialized struct into the ECB and now they arent generating anymore garbage. Which reduced the gc alloced memory from 20-40KB every few frames to about 1KB.
So there kinda like two branches for my deserializer. It either returns a list of objects... deserialized class/boxed structs... or it invokes its events to pass the deserialized structs straight to the ECB without returning and boxing them first
@iron jacinth WTF ?
gusty comet
with monobehaviors and coroutines i executed somewhat heavy code every 3 5 10 seconds, how would you achieve this in DOTS? since ONUPDATE executes every frame.
would it be OK to count time.deltatime and wait until it has passed eg 10 seconds? or thats slow
oh well theoretically I still can interact with entities via coroutine and entity manager
pliant pike
systems aren't that different to monobehaviours, they update every frame also
gusty comet
ah im still somewhat confused, but I guess its better to do 'occassional' stuff from monobehaviors, and stuff that executes every frame, from systembase OnUpdate
( theres no performance gain if something executes once per 5 seconds anyway)
pliant pike
I don't think you can use coroutines in Systems but you can do a simple timer which I doubt would be much performance difference than a coroutine
gusty comet
ah yeah i already made the timer, but the more i think of it the more i feel i need monobehaviours
i will still use dots for rendering, colliders and movement(as this is every frame), but other stuff from monobehaviours, by calling entity manager
pliant pike
yeah you can use both together just fine
gusty comet
btw, theres no fixed update for dots?
safe lintel
@gusty comet put systems into the fixedsimulationsystemgroup
night venture
not 100% ECS but here it goes:
I want my 2D board game to be static but large enough to consider using a file backend as the place to mark, for each [x,y] which tile components should I set.
For example: [3,0]is a sidewalk tile, while [4,0] it's a road tile...is this a BAD idea? would it be possible to write entity "components" within file to create them at runtime?
ie: .CreateArchetype(foreach_tile_types_defined_in_file)
gusty comet
what if Hybrid renderer v2 made my entities invisible?
turbid sundial
Hoping someone can give me some optimization ideas here, https://gist.github.com/jamieshepherd/d9b6a93c907e6fea5cd0744e82a7abbb
This is the biggest bottleneck in my procgen terrain at the moment, putting all of my verts etc. into the world. If the world is really small it's acceptable, but becomes noticeable at any real scale. How can I reduce this bottleneck as much as possible? It looks like the main reason I can't use ecb as parallel is because I need to pass the RenderMeshDescription a Material, can I skip this step somehow and apply this later?
https://i.imgur.com/kFg0zNZ.png
This is the current profiler snapshot, 33ms every time I need to generate new chunks
hollow sorrel
@turbid sundial you can use MeshData structs which is faster to apply mesh data than new Mesh() and can be used in jobs + burst
examples: https://github.com/Unity-Technologies/MeshApiExamples
also if you're using ecb's you don't have to use .WithStructuralChanges()
night venture
Hoping someone can give me some optimization ideas here, https://gist.github.com...
in plain human, what is your code doing?
turbid sundial
It seems like I do have to use .WithStructuralChanges because of the material passed in, at least on removal I get Entities.ForEach Lambda expression captures a non-value type 'baseMaterial'. This is only allowed with .WithoutBurst() and .Run()
I mention that in the first message ^
I did read up a little on the new mesh data api, but it seemed from the outside like it was more useful for editing existing meshes than instantiation of new ones, but I could be wrong
night venture
in plain human, what is your code doing?
hollow sorrel
it's also for new ones
i'd separate out the instantiation of new meshes into bursted jobs using mesh data and that should be decent speedup (main thread just to start with), in future you could look into separating the instantiation asynchronously, start applying meshes in 1 frame off main thread, apply them to components the next frame(s) on main thread
turbid sundial
I'll read deeper! thanks for the tip
night venture
🥺
hollow sorrel
@night venture seems it is getting pregenerated mesh data and creating new meshes with them
night venture
im not sure in interested on that approach, but FOV/LOD/distance could help a lot on this kind of things
i mean, dont render everything, but only "closer to you" things
but, i guess those are my 2 cents xD
night venture
ok...here it goes: i have a 2D array of entities. each entity has several components
how can I save it to json?
can I JsonUtility.ToJson(entityArray)?
ocean tundra
no
that wont work
you need to look at SerializationUtility
but it works from the World
night venture
...that means....?
ocean tundra
so you may need to copy/move the entities to a new world then save that
night venture
why?...
night venture
oh....so move to a blank world to only save that, right?
night venture
it might work for a test, but ill need a blank world later probably
stone osprey
Hmmm... A item is an entity in my case, i want to list them in the players inventory. So basically each item has a icon, a list item/entry and a name. Therefore i have following structs... Item{ bool stackable, int amount, FixedString32 name }, Sprite{ short spriteInDatabaseID }, Mesh{ short meshInDatabaseID } Is this too detailed ? Does it make sense to split into three components, one for sprite and one for mesh ? I thought about reusing them ^^ What do you think ? Or should i just merge them all into the item struct ?
night venture
and...2 more questions, if u dont mind
stone osprey
@night venture You can also write an extension which returns you all structs/objects of an entity. Then you an write your own serializer
night venture
<@!483700836397088768> You can also write an extension which returns you all str...
custom serializer?
i dont know what is an "extension", but probably out of my league
stone osprey
@night venture Yeah kinda ^^ Or you could use an existing json serializer... But you first need to write an extension that returns you all components as objects or whatever.
stone osprey
@night venture Look into extension, they allow you to write your own methods for existing structs and classes... without extending them
night venture
!extension xD
may I ask anyway about the other thing @ocean tundra suggested?
if using the world serialization utility
2 things left....creating/moving to a new world
it doesn't sound so complex
ocean tundra
there are methods on the EntityManager for that
night venture
and...if my entites have multiple components, after saving them...will they be stored in json, in a human readable way, with all their components and so?
ocean tundra
h sorry its binary
night venture
i need a text format 😦
the idea behind all this is to load my board/scenario from a file
stone osprey
Then a own serializer is your only way
night venture
each tile being a collection of components
@stone osprey you mean https://docs.unity3d.com/Manual/script-Serialization-Custom.html ?
stone osprey
@night venture No... i mean a custom serializer written by YOURSELF. You need to find a way to get all components of an entity as objects. Then you need to serialise them into json somehow.
night venture
<@!483700836397088768> No... i mean a custom serializer written by YOURSELF. You...
that sounds wild...
night venture
ok...i'll ask another way. reflection is there, but if i can avoid it....
stone osprey
You can avoid reflection.... but it will cost you development time
You actually need to write huge switch statements or develop some sort of plugin system for all possibly components n stuff
I would actually go with reflection if such serialisation operation do not happen during the game
ocean tundra
oh im working on that
a modding kit
will use code gen to create that switch statement
stone osprey
With json ? ^^
ocean tundra
yup
stone osprey
Oh that sounds great 😄
night venture
as you sound quite skilled, let me ask:
consider the following scenario: a board is a set of tiles.
my board is stored on a json file similar to [0,0]={Component1, Component2}, [0,1]={Component1, Component3, Component4}....
is reflection the unique way to "create the board" in unity following that?
ocean tundra
highly recommend code gen for DOTs
night venture
highly recommend code gen for DOTs
is that an answer for me?
stone osprey
If hes a beginner he should probably avoid code gen ^^
ocean tundra
so many things in DOTS needs actual types, for speed
and writing massive switch statements sucks
stone osprey
Yep
night venture
what is code gen? link, please
stone osprey
An alternative to massive switch statements is a plugin system... where you plugin your switch cases... way cleaner, but still sucks
night venture
reflection?
ocean tundra
i use https://github.com/scriban/scriban as a templating lanuage
so i write a system but with {{TEMPLATE}} stuff inside
night venture
templating
ocean tundra
then from a scriptable object i call scriban and write out a cs file
night venture
i dont think that's the way....
ocean tundra
the scriptable object is to pass in Unity values if i need it
night venture
there might be something simpler
ocean tundra
yea probably
stone osprey
Then use reflection
night venture
reflection, in fact, is simpler
stone osprey
is easy
ocean tundra
write the switch statment manually
night venture
we all know im not going to switch xD
ocean tundra
but if you have a game not a tool kit
the things your adding/remoing from save/load is quite low
night venture
i wont use it on entities, but creating entities with those components and data...why it wouldn't work?
stone osprey
Because... you need to either use Object components... or still write manual switch cases
Theres no "SetComponent(entity, object)" method yet for boxed structs
night venture
no, but i have a manager and i can manager.AddComponentData(entity, component_name_from_file), isnt it?
stone osprey
@ocean tundra Yep... trust US, because WE wrote one 😄 well i wrote it, but the others helped debugging and saved my ass
@night venture Nope
Thats not working
How should it ? It cant read your json file :p
Furthermore the generics are missing
You NEED them
night venture
i read the file, got the name, cast to class/struct using reflection, and then use it there
ocean tundra
yea sorry that wont work
stone osprey
Class is working, struct not
ocean tundra
so
you will have your struct loaded from file, thats fine
BUT you will have it BOXED into a Interface or Object
stone osprey
^ Exactly... it actually makes me "happy" that im not the only one with that problem 😄
stone osprey
But... we have a working solution ^^
ocean tundra
but having fought it as hard as you did
stone osprey
Not the best... but it works...
ocean tundra
went code gen instead
stone osprey
@night venture Im no asshole... i had the same problem. My deserialiser returned boxed structs... so i wrote this here with a LOT of help from several other people in here ( Thanks again ) !
https://pastebin.com/siRZcDBi ( You need to use ASMRefs from unity to get acess to the internals of the ECS )
The code above provides methods for EntityManager and ECB to add boxed structs to entities
ocean tundra
oh nice
ocean tundra
you lose out on all dots goodness doing that
stone osprey
^
night venture
but im not using them...i just using them to reference struct
(altough i cant probably serialize a static)
ocean tundra
so how would you add them to a entity?
night venture
Myclass.struct
stone osprey
...
I dont think that this works
You still need to use generics in that case
Or just use the class... but then, as roycon said, you will lose the advantage of the ecs
ocean tundra
i feel like the class thing could work
night venture
i wont use the class
ocean tundra
but its not too different from the switch
you will need to pass the EntityManger into the class
and it Adds the component
night venture
but the StaticClass.struct
ocean tundra
BUT you will need 1 class per struct
ocean tundra
so proberbly even more code then the switch
night venture
no...just adding a few lines on each component...
anyway, im sure u already tested everything and tried all the paths
ocean tundra
the switch is like 3 lines per stuct
night venture
so at the end, ill have to go code gen
night venture
whatever, it will be tomorrow xD
stone osprey
@night venture Or to use the fancy extensions i posted above... wuuu
ocean tundra
if your changing the serialization loads then hyes
night venture
@stone osprey already saved to my desktop xD
ocean tundra
but honestly once your file format is set, and your components written, how often will you actually be changing it?
night venture
i need it in json at least once...
remember the objective: define the board in a file
at the end, ill define the board in the editor, using background image as reference
and save the world
but i dont want to surrender so soon!
i havent fought more than 2 hours
xD
stone osprey
Sooo... Sometimes i have the feeling that i design my components too atomic. A item is an entity in my case, i want to list them in the players inventory. So basically each item has a icon, a list item/entry and a name. Therefore i have following structs... Item{ bool stackable, int amount, FixedString32 name }, Sprite{ short spriteInDatabaseID }, Mesh{ short meshInDatabaseID } Is this too detailed ? Does it make sense to split into three components, one for sprite and one for mesh ? I thought about reusing them ^^ What do you think ? Or should i just merge them all into the item struct ? What do you guys think ? ^^
frosty siren
It's so much pain to try to use companions and SortingGroup component...
pliant pike
@stone osprey think about how you are going to use the data, like are you going to need to iterate through a single for maximum efficiency are you going to need to calculate some of it together
it doesn't really matter though you can easily change it later on as you figure things out and want to optimise
safe coral
Hey, how far from usable/stable is the new deterministic physics system? And, is there a more mature fixed point physics solution (doesn't have to be DOTS, actually probably better if it's not)?
hollow sorrel
@safe coral it's not multi-platform deterministic yet, should be same-platform deterministic, but even unity devs are giving mixed responses on the forums about that so prob need to run your own tests
unity physics devs: it is deterministic on same platform and we have tests to ensure this
unity burst devs: burst is not guaranteed deterministic even on same platform
safe coral
Okay, so completely useless to me
haha
I want cross-platform replays
Man, why are they working on a fixed point system if it's not cross-platform deterministic?
odd ridge
does DOTS have sprite sheet animations? If I have a sprite sheet animation object with a convert to entity script on it, will it run fast or still use the slow game object attached to an entity?
hollow sorrel
Man, why are they working on a fixed point system if it's not cross-platform det...
new dots physics isn't fixed point, it's all floats and they're hoping burst becomes deterministic enough
ocean tundra
@odd ridge No it dosnt
@safe coral Yea new physics its not deterministic
and buest is confirmed NOT deterministic right now
which is very confusing as it has previously been said it was
safe coral
Ah cool. Okay well that's disappointing. I was just thinking how great if would be if there was a generic number type in C# so you could parameterize the whole system (as you might in Rust for example)
safe coral
Undoubtedly
Ah well, maybe I'l investigate a third party library or just drop this idea of cross-platform replays
ocean tundra
my networking library will have cross platform replays
nice side effect
what sort of game you building tho?
safe coral
Yeah, we use Photon Quantum for a multiplayer game, and it has its own FP physics engine.
But this game is single player, it's a "turn based bullet-hell roguelike"
https://rhys_vdw.itch.io/enter-the-chronosphere
Just scoping out options for how we might expand it
ocean tundra
yea probably too "twich" based unfortuanly
safe coral
for which?
ocean tundra
sorry my networking library is aimed mostly at RTS/Management type games, if the game is too fast passed it probably wont work well
ocean tundra
theres a ton of c# 2d physics library
safe coral
I found this lib: https://github.com/christides11/2DFPhysics
ocean tundra
could you grab one of those and implement that
safe coral
Still WIP, but it might be enough.
It's just a question of whether i want to rely on something that isn't under active development for a polish feature like replays
ocean tundra
are replays so important?
safe coral
Thanks for clarification though @ocean tundra, I appreciate your time.
No, they're not, but if there was a stable physics lib that gave them for free then I'd take them.
ocean tundra
ah yea makes sense
safe coral
It actually would be very cool to have replays
Because the game turns are real time, but the gameplay input is turn based
So you could get a very cool unseen real-time replay of the action
gusty comet
can your entities contain arrays of other entities?
heady vortex
To multithread navmeshquery, I need to pass all of the data from main thread into the job, and then the job will give me a position or some other good info to apply in the main thread, right?
and is my main thread supposed to wait for the job while the job does its calculations?
zenith wyvern
can your entities contain arrays of other entities?
An entity can have a DynamicBuffer<Entity> attached to it as a component. Entities don't "contain" anything for the record
gusty comet
can I set entities to be child of other entities (and so their position and rotation would be updated)
or do I need to manually update their position and rotation, and theres no parenting?
ocean tundra
@gusty comet check the docs
There's a whole built in transform systems
And if you use the conversion flow you will get it all automatically
Definitely aim to use the conversion system, you can use prefabs and gameobjects for authoring
And unity have confirmed it's to be the recommended way to use DOTS
gusty comet
i have no need for prefabs i think
ocean tundra
You can use dots via pure code
But I don't really think it's worth it
I don't know if there's a helper method yet for code based transforms
But it's all listed in the docs
You need a specific combination of components to get it working
ocean tundra
i have no need for prefabs i think
The conversion system isn't specificly prefabs, look into subscenes. Basically pre optimized "scenes", very fast to load/unload
minor cove
Hey guys, I want to do a project and utilize a little bit of ECS (and probably dots physics).
Can I access dots in the long term support Unity client or do I sadly have to download the beta instead?
slim nebula
minor cove
@slim nebula thank you so much
yea my net is pretty slow; i was just worried i'm gonna waste time downloading a client that won't work
karmic pilot
has anyone tried to use the serialization package low level API in jobs to deserialize a lot of strings containing json ?
verbal pewter
How viable is a pure DOTS approach for procedural mesh generation right now? I remember people talking about it about a year ago and saying that there are more performant ways to go about it. Is that still the case?
I did see that the Mesh API now uses Jobs: https://docs.unity3d.com/2021.1/Documentation/ScriptReference/Mesh.MeshData.html
night venture
How viable is a pure DOTS approach for procedural mesh generation right now? I r...
perhaps ask @turbid sundial
zenith wyvern
How viable is a pure DOTS approach for procedural mesh generation right now? I r...
It works great for mesh generation. I've found uploading the data to the mesh is the bottleneck. That still has to be done on the main thread, even if you're using the job api
night venture
it makes sense to have component composition? (a component which causes entity to inherit/have 2 components)
eg: component cat, implies component animal, implies component livebeing ?
verbal pewter
It works great for mesh generation. I've found uploading the data to the mesh is...
Okay thanks, I remember you being one of the people saying that it wasn't great a while back so I'm glad to hear things have changed 🙂
stone osprey
@night venture You would normally do something like this : "Entity{ Cat, animal, livebeing}" each of them is its own component. But there many solutions... You can also have something like : "Entity{ Livebeing, Animal{ type:cat}"
It depends on what you want and how detailed ^^
Theres no right and false
As far as i know thats what archetypes are for... There no automatic including for other components yet... I think
night venture
ok. no component composition xD
btw @stone osprey do you remember i was struggling how to get the components of an entity and so? i achieved that (the easy part)
zenith wyvern
What you were describing wasn't composition, it was inheritence
night venture
@zenith wyvern not really, cause im not inheriting anything, but "adding a few subcomponents" whe adding one
anyway, ill add each one individually by now
going back to my try to serialize entities adventure...
zenith wyvern
eg: component cat, implies component animal, implies component livebeing ? You're describing an inheritance tree
Components are completely self contained from one another, their relationship is only described by the systems that operate on them
night venture
can i serialize NativeArray<ComponentType> ?
safe lintel
@night venture you might run into issues with serializing, im still waiting on this bug to be fixed and its from almost a year ago 🥲 https://issuetracker.unity3d.com/issues/trying-to-serialize-slash-deserialize-an-entities-world-a-second-time-fails-if-there-are-blob-assets
night venture
@safe lintel probably true...but i want to do it just once! xD
do you have a test working code to de/serialize NativeArray?
and...where's "Unity Bugs.zip" ?
safe lintel
i didnt try serializing native arrays, my experiment is here https://github.com/thelebaron/dots_serialization but i really dont remember what state I left it in
Roycon might know more, he gave me loads of helpful ideas on serialization 🙂
night venture
that's using SerializeUtilityHybrid, which might be similar to SerializeUtility. both save the world
furthermore, i only want to deserialize xD
turbid sundial
@zenith wyvern I saw in your blockworld repo that you end up creating gameobjects for each chunk - did you try doing this with hybrid renderer? How come you went in that direction?
safe lintel
zenith wyvern
<@!155490373123178497> I saw in your blockworld repo that you end up creating ga...
Originally gameobject rendering was significantly faster than hybrid renderer for any kind of dynamic mesh, just due to the overhead of ECS managing the RenderMesh shared components for each unique mesh. When I profiled it recently hybrid renderer was at least on par with gameobjects for rendering large numbers of dynamic meshes. In my "version 2" branch I did try it with the hybrid renderer and it worked well. Was much simpler and didn't perform any worse than the gameobjects from what I could tell
turbid sundial
I'm hitting a similar bottleneck to one you mentioned around the actual instantiation - using hybrid renderer - mainly because of the main thread
I almost want to be able to spread instantiation over frames (e.g. like the way coroutines work) because it's not immediately necessary
gusty comet
https://answers.unity.com/questions/356638/maths-behind-transformtransformdirect...
translation.Value = new float3(
Mathf.Cos(planetData.angle) * planetData.radius,
Mathf.Sin(planetData.angle) * planetData.radius,
0f) * localToWorld.Rotation;
that kinda doesnt work
oh the rotation is quaterninon
i need vector3 hm
zenith wyvern
I'm hitting a similar bottleneck to one you mentioned around the actual instanti...
There's no getting around it - uploading to a mesh is slow. The best you can do is offload all the actual mesh generation into jobs and then decide how to handle building the meshes. You could spread it over a few frames or use some kind of algorithm to combine meshes or at least combine same-block-type-faces in a single mesh
Greedy meshing or whatever it's called
turbid sundial
Yeah I was looking into that as well
zenith wyvern
The less data you have to upload the faster it will be
turbid sundial
Can you actually spread over frames in ECS?
zenith wyvern
Not in any fancy built in way like a coroutine, but you can track when your jobs are completed (jobHandle.IsComplete) and only upload so many meshes within one frame
It's a lot of work to build a framework to track all that but it's certainly possible
turbid sundial
Got it- thanks for your thoughts 🙂
sturdy rune
wouldn't rejecting meshes not in view and etc also help wit that?
pre-culling rejection, might reduce number of meshes you actually are loading
but I am definitely a newb at this point, just learning megacity optimization
sturdy rune
pure ECS is also pretty painfult og et started
as a fellow beginner, I learned some pure ecs, took a course, and then finished by reading through ECS documentation which enforces that authoring is the direction unity will go. tiny will be pure ecs
north bay
I think a lot of people are mixing pure dots up with the authoring workflow.
Even tiny uses gameObject based authoring and conversion.
The only time I saw unity mention the term pure dots was when referring to a project that doesn't make use of any systems outside the DOTS ecosystem.
gusty comet
oh i see yeah i heard the same, that authoring is the way to go
turbid sundial
@zenith wyvern have you looked into the new mesh api? Wondering if there's any gains to be made there 🤔
zenith wyvern
<@!155490373123178497> have you looked into the new mesh api? Wondering if there...
If you mean the "MeshDataArray" api I have messed with it. It is indeed faster to upload but because you need to "ApplyAndDispose" you then need to keep an extra copy of your data around and re-apply it all any time you need to make a change
turbid sundial
That's interesting
zenith wyvern
I've found it easier to just stick with keeping your data as Native Arrays and uploading via the MeshData APIs which is an okay middle ground
sharp flax
So I was testing some performance stuff I was curious about with jobs, but found something unrelated to what I was looking, that I recall to have seen before in this channel, is it normal that the **overall ** time that a job takes to complete can be like x4 times higher when doing it in parallel than running it in a single thread?
zenith wyvern
Here's an example of how I do it in my roguelike. You can still use the Vertex/IndexBufferParams apis to upload data which seems to be a bit faster than just doing SetVertices. You can exactly describe how you want your mesh data layed out with VertexAttributeDescriptors so you know what you're looking for in your shader https://github.com/sarkahn/dots-roguelike/blob/master/Assets/Lib/Terminals/Runtime/Rendering/TerminalRenderMeshSystems.cs
turbid sundial
Cool, thanks
@sharp flax there is some overhead to jobs - so it depends on the size of the work and number of jobs you have to iterate through to decide whether or not it's worth jobifying
karmic basin
So I was testing some performance stuff I was curious about with jobs, but found...
Yeah can happen. Due to the scheduling overhead. Could mean you don't need to optimize that part (not enough entities to really bother). Could also mean you're doing it wrong 😁
But yeah also remember a forum thread where the additionnal cost of multithread was insanely unexpected
sharp flax
Not doing entities stuff, just isolated jobs testing. I just watched a talk from Scott Meyers talking about cache memory that was recommended by one of the Unity blogs somewhere and started to wonder "what would happen if...". Still don't know much about the most low lvl stuff but this has been revealing. I always knew that processing a few items wasn't worth to parallelize, but didn't expect such difference
It's not few elements either, like 250k but just adding numbers
olive kite
is anyone else getting "A Hybrid Renderer V2 batch is using a pass from the shader "HDRP/Lit", which is not SRP batcher compatible. Only SRP batcher compatible passes are supported with the Hybrid Renderer."
gusty comet
ArgumentException: GetComponent requires that the requested component 'PlanetData' derives from MonoBehaviour or Component or is an interface.
PlanetData planetData = GetComponent<PlanetData>();
how do you kinda refer to the authred IDataComponent via code
ocean tundra
@gusty comet you cant
the way its code generated makes it impossible to reference
make it manually instead
night venture
hey @ocean tundra do you have 5 minutes?
ocean tundra
probably 😛
night venture
achieved to find NativeArray<ComponentType> types = manager.GetComponentTypes(tiles[i, j], Allocator.Temp);, so I can iterate trough components each tile has
can NativeArray<T> be serialized?
ocean tundra
hmm you could always loop it and manually write it
gusty comet
make it manually instead
ok but how do you reference it? is it not possible to have both authoring component, and reference that entity from code
ocean tundra
but ive never tried any serialization librarys on native collections
gusty comet
i know i can do it in OnStart() but i wanted to do it in editor
night venture
hmm you could always loop it and manually write it
and...once is written...how i would read it back to a NativeArray<T>?
ocean tundra
@gusty comet rephrase what your trying to do and answer the following:
How are you converting your GOs? SubScenes? ConvertToEntity?
Why are you trying to access the Authoring Component?
and WHERE are you trying to do that? (System? Monobehaviour?)
@night venture Same but in reverse,
You could use the File API and manually construct the struct and then add to a native list
But that seams like a bit of a pain
I would gather your types, and convert into a managed List<T>, serialise/serialise that
gusty comet
<@456226577798135808> rephrase what your trying to do and answer the following:
...
- converttoentity
- i want to set values for many entities Data scripts in editor. to avoid loading it from monobehaviour during game Start() or from somewhere else also on start()
3.monobehaviour
ocean tundra
i see so in editor nice
gusty comet
yes but not manually(by hands) it must be via code
ocean tundra
yea so first you will need to manually make the authoring class, then you can either use a custom inspector or custom window using the normal editor tooling
you cannot access any entity stuff from that
as at that point its all still authoring
which is Monobehaviour
night venture
<@!483700836397088768> Same but in reverse,
You could use the File API and manu...
can you elaborate a bit more?...on one hand, im not an english native...on the other hand, im not a unity expert...:S
you mean to "copy" Native to List<IComponentData>, and serialize it...and reverse after reading the file?
night venture
jsonutility seem to suck (even more for 2d arrays)
gusty comet
how can custom inspector achieve it?
night venture
i have spent a few hours trying to do it
night venture
is it the same IComponent or a bunch of different ones?
bunch of them, different for each tile
ocean tundra
@gusty comet Custom inspector can be written to support multi editing, so you can have that enabled, and a button which does your code/data set
gusty comet
i need to set data from monobehaviour tho
ocean tundra
@night venture oh thats harder, i would build a "Model" class, a root class containing all your lists/components, avoiding the serialization of Interfaces
@gusty comet inspectors operate on MonoBehaviours?
and you can still select a specific MB to edit it
night venture
@ocean tundra look what I got so far:
Debug.Log("L1:");
List<IComponentData> l = new List<IComponentData>();
l.Add(new Tile());
l.Add(new DriveLeftToRightComponent());
foreach (IComponentData c in l) {
Debug.Log(c.ToString());
}
Debug.Log("L2: " + JsonUtility.ToJson(l));
//otuputs
L1:
Tile
DriveLeftToRightComponent
L2: { }
but I was expecting something like { "Tile", "DriveLeftToRightComponent" }
ocean tundra
your expected output wont be valid
more like
[{Tile: {TILEDATA}, {DriveLeftToRightComponent: {DATA}}]
but also unitys json sucks
so i dont think it will work with interfaces
ocean tundra
it will DEFINITLY have no clue how to deserialize types
yea
theres a package somewhere for it
and make sure you configure typed serialization
night venture
no clue what you talking about...but let me ask, before goind down the rabbit hole
night venture
does it make sense what im trying to do? each tile is a set of components, i want to edit a json with tile[x,y]={types} and to load it at runtime to create the map/board
night venture
i just want to edit a json for the map...
ocean tundra
but you could use SerializeUtility and have a map editor in game
ocean tundra
the advantage of a text format is its super easy for anyone to edit
binary format (that world serialize utilily) is much faster and probably easier to code
BUT you would need tooling in game to make the map
oh and text format you dont need to build the map tooling
night venture
can i use SerializeUtility to get a text format?
night venture
so, to sum up:
json.net to make json<->code conversion easier/better
and somehow, perhaps using json.net the serialization of nativelist would even work
night venture
well...at least native->list->json would work xD
ocean tundra
yupp
night venture
the guy responsible for json in unity...is the boss son?
night venture
i mean...unity json sucks...
ocean tundra
yea
they are migrating to json.net
hence the json package
@night venture WAIT
your going to have issues
the deserialized data will be Boxed
you wont be able to add it to a entity
ocean tundra
😛
sorry
the boxing thing is super annoying and easy to forget
you can still get it to work
just more round about
ocean tundra
some example Code:
IComponent test = new Translation(){SET POS}
so that is Boxing translation into IComponentData
when you try to use EntityManager.AddComponentData it wont let you
ocean tundra
yup
night venture
add(manager, icomponent) ?
ocean tundra
yea
make a interface all your serializable components use
that has a method like that
use that interface instead of IComponentData
sorry it will be Add(EntityManager)
night venture
you mean...struct a: Myinterface and IMyInterface : IComponentData?
night venture
i wouldnt have your IMyInterface a child of ICOmponentData
why not?....and...beside that...are struct able to "inherit/implement a method"?
ocean tundra
yea sturcts can
night venture
let me draft...
ocean tundra
just personal preference
night venture
used asset store...
Multiple precompiled assemblies with the same name Newtonsoft.Json.dll included or the current platform. Only one assembly with the same name is allowed per platform. (C:/unity/Library/PackageCache/com.unity.nuget.newtonsoft-json@2.0.0-preview/Runtime/Newtonsoft.Json.dll)
night venture
new project?
ocean tundra
it may be a dependancy of Entities somewhere
night venture
its a dependency, but i cant undo the mess
night venture
@ocean tundra fresh 2d project created xD
did u mean this?
using Unity.Entities;
[GenerateAuthoringComponent]
public struct Tile : ISerializableComponent {
public void addComponent(EntityManager manager, Entity e) {
manager.AddComponentData<IComponentData>(e, new Tile() { });
}
}
almost
manager.AddComponentData(e, this);
and entity manager might have to be passed via ref
not sure
night venture
ok...and that to make json work and store properly...right? xD
ocean tundra
maybe
test it
make a Tile and box it into the interface
then call addcomponent
it should work
😛
night venture
current json+namehandling:
{
"$type": "System.Collections.Generic.List`1[[ISerializableComponent, Assembly-CSharp]], mscorlib",
"$values": [
{
"$type": "Tile, Assembly-CSharp"
}
]
}
lets see how it handles reading it back
warm panther
Question:
What causes a converted Entity to get:
- a Scale component
- a Rotation component
< versus > - a CompositeScale component
- a CompositeRotation component?
And can I prevent them from getting either?
ocean tundra
i didnt know there was a composite rotation
night venture
there is
ocean tundra
somewhere in the transform conversion systems i guess
warm panther
Yeah and I noticed that some of my entities have one or the other.
night venture
@warm panther if you are converting gameobjects, instead of creating them
the object has a few components "by default"
warm panther
Seems like the root parent" entities have composite... SOMETIMES.
I want to know why they have one or the other.
night venture
depends of the object type
warm panther
Such as?
night venture
cube->traslation+rotation
rendermesh, renderbounds
...
so i suggest, instead of converting, knowing what you are doing "manually", to better understand
eg:
manager.AddComponentData(myEntity, new Translation {
Value = new float3(i, 0f, j)
});
manager.AddComponentData(myEntity, new Rotation {
Value = quaternion.identity
});
Material material = new Material(Shader.Find("Standard"));
material.color = Color.white;
manager.AddSharedComponentData(myEntity, new RenderMesh {
mesh = mesh,
material = material
});
But how do I do that in a IConvertToEntity?
night venture
im not converting...you can ask for each object components
warm panther
I really need visual editing so I want to base it off of a gameobject, but some of the intrinsic things it does are weird. I mean I could author as plain prefabs and then convert in my own conversion system.
but that feels like a huge drag at this stage of the project.
night venture
how does that affect you?
ocean tundra
whats the issue with the composite ** types?
night venture
you want to select/filter only some of the entities? both?
warm panther
The objects have multiple hierarchies and hierarchies of authoring components.
warm panther
whats the issue with the composite ** types?
They are treated differently in TRSLocalToWorldSystem; I am writing my own LTW replacement system for a camera centric renderer, and they constitute a number of extra cases that got me curious.
warm panther
This is my port of Unity's originalc ode.
And I want to clean it up. It could also be made branchless.
And because my system is rather specific, I would like to only deal with objects that I want to deal with, meaning - if I know what components they have, I can get rid of most of the branching even before it needs to be calculated.
night venture
sorry. out of my league :S
warm panther
(but it could also be written branchless, and I wonder why they didn't do that - they have a todo in the code waiting for Function Pointers in Burst)
I noticed that some of my objects weren't being transformed as expected, and that's because they have unexpected scale and rotation components on them. 🙂
ocean tundra
i would just leave it as unitys code for now
they will come back to it at some point
and if you need the optmization later you can either copy theirs or write your own depending on if they are done
warm panther
Yeah, right now the optimization is complexity.
night venture
@ocean tundra so...tiles was an entity[,]...i was able to iterate trough it and, for each tile get their components
IIRC, the idea was to add them to a List<Component> and then serialize as json. wasnt it?
after that, i should have a file containing aproximately
{
"$type": "System.Collections.Generic.List`1[[ISerializableComponent, Assembly-CSharp]], mscorlib",
"$values": [
{
"$type": "Tile, Assembly-CSharp"
}
]
}
yea
night venture
manager.GetComponentTypes returns NativeArray<ComponentType>...how to cast to List<ISerializableComponent> ?
ocean tundra
Component Type is not the components
is the definition of a componet
ummm think of it as reflection metadata
but not for reflection
ocean tundra
im not really sure how getComponentTypes will help with serialization
i guess you could take the compoennt type and build a entity query...
night venture
u mean to loop by myself
ocean tundra
but then you need generics to get the ToComponentDataArray....
i dont think you can...
night venture
...all that time....? xD
ocean tundra
the api is really not setup for untyped access
night venture
so...going back to basics...
json file describing tile [x,y]={list of components}
i want to read it and setup entities with those components
ocean tundra
yea all the entityquery ToComponentArray methods are generic sorry
you can just make the entity querys yourself
like for your inital file creation couldnt you just manually write it?
night venture
...no...xD
ocean tundra
i guess just manually write the json file then
night venture
but how?...i mean
ocean tundra
or you could write some 1 off code to generate one
?? just use Visual studio and edit it
night venture
...dont know what "strings" have to write in order to deserialization understands what it has to do.
ocean tundra
so easist way is to write some code to manully save
night venture
json file describing tile [x,y]={list of components}
i want to read it and setup entities with those components
ocean tundra
which is one off
yea so you want a List of a model class (Tile)
X;
Y;
List<serilzedComponent> components
}```then use 2 for loops for X and for Y to create your inital grid (5 by 5?)
and add them all into the list
then serialze that
night venture
i cant loop through entity[,] ?
ocean tundra
boom you will have a nice 'template' json
if you serialzie a 2d array you will have to also serilze the width and height
im honestly not sure how json handles 2d arrays
night venture
?...sorry...im exhausted...
look, this is what i got when serializing 1 entity:
{
"$type": "Unity.Entities.Entity, Unity.Entities",
"Index": 42,
"Version": 1
}
in other words, nothing
ocean tundra
yea no you cant serialzie a entity
sorry i think im missing what you want
end goal is a josn file defining your 'grid' of tiles?
night venture
yes, each tile being an entity
ocean tundra
ok and you dont want to manually write that file
night venture
entity being a set of components
i can write it manually, since the moment i have a pattern
but, as im using entities, i dont have one
using oop (not ECS) that could be fine
but i want to save "an entity"
(and in json format, not binary) 😛
ocean tundra
yea ignore dots/oop/data orientated. you just need to get saving/loading working
thats just adding complexity
night venture
ok...how i save an entity? xD
ocean tundra
ok so you have a inital grid?
night venture
atm, made by hand, set of entity[10,10]
ocean tundra
ok
so make a new system
and have some trigger (like input.getkeyup...
on that trigger
ocean tundra
you want to have a entity query with all your CURRENT tile components in it
night venture
already
ocean tundra
then for every tile component you want to call ToComponentArray
you now have many arrays of your components
night venture
ok...hold a sec.
ocean tundra
next you need to loop x and y and construst a list of SerializedComponent
and grabbing out your tile compoents from those arrays and putting it into that new list
now your in memory data is close to your target json
and i 'think' you should be able to just feed that list into json.net and have your output
night venture
would Entities.WithAll<Tile> do the job?
ocean tundra
personally i would resturcture it so your not serializing a 2d array, instead a single long list of tile (which has X Y and component)
no
thats for Entities.Foreach
you need to be manually creating entity queries for this
GetEntityQuery()
but a single query with just Tile will probably work
night venture
but why foreach wont do the job?
ocean tundra
you could probably just use EntityManager.Has/Get component for all your other components
i guess you could... you will have to use run and without burst
its just it generates a ton of 'magic' behind the scenes, for this you dont really need any of it
yea sorry, this is one off code, we are not caring about performance
you dont need entity query or for each
just For X and Y of your entities[x,y] array
and use EntityManager.GetComponent or HasComponent
and built the 'per tile' output lists of serialziedcomponents
night venture
you mean manager.GetComponentData?
ocean tundra
sorry my default for dots is to try follow best pratcices and write for performance 😛
yea
and HasComponent if you have optional components (components only on some entities)
so sudo code:
night venture
you are running too fast for my asleep mind.
ocean tundra
for y //Loop over your entitys
entity = array[x,y]
maintilecomponent = EntityManager.GetComponentData<Tile>(entity)
AdditionalComponet1 = EntityManager.GetComponentData<AdditionalComponent1>(entity)
serizlaedtypesList = getListFor(x,y) //get the list for this entity/tile position
serizlaedtypesList.Add(maintilecomponent)
serizlaedtypesList.Add(AdditionalComponet1)
//a optional component
if (entityManager.HasCompoinent<Optional1>(entity)
serizlaedtypesList.Add( EntityManager.GetComponentData<Optional1>(entity))
sorry does that make more sense?
night venture
manager.GetComponentData is typed, hence it will return only one component. inst it?
ocean tundra
yea sorry, missed some generices
night venture
so the only way of getting all components would be manager.GetComponentTypes ?
ocean tundra
honestly forget about that
you dont need it
how many components do you currently have?
5? 10? 20?
night venture
some tiles 4, others 8
night venture
we have been dealing with this for a while...
ocean tundra
your only using it to generate your inital json file
night venture
oooh, ok
ocean tundra
after that you edit/copy your json file for everything
night venture
hold a sec
ocean tundra
your requirement was for deserialization (loading) more then serialalization (saving)
this will give you the inital template and patterns and then you use that instead of this
it may help to generate a few different sizes initially
minor sapphire
quick interjection: what is the appropriate job type for looping over elements in a list, e.g. NativeArray or hopefully NativeHashSet?
thought IJobForEach but it's obsolete now?
ocean tundra
i think that was for entities...
minor sapphire
Oh yes indeed
ocean tundra
so??
night venture
Get component data can not be called with a zero sized component.
night venture
and just add a new Tile?
ocean tundra
for y //Loop over your entitys
entity = array[x,y]
maintilecomponent = EntityManager.GetComponentData<Tile>(entity)
AdditionalComponet1 = EntityManager.GetComponentData<AdditionalComponent1>(entity)
serizlaedtypesList = getListFor(x,y) //get the list for this entity/tile position
serizlaedtypesList.Add(maintilecomponent)
serizlaedtypesList.Add(AdditionalComponet1)
//a optional component
if (entityManager.HasCompoinent<Optional1>(entity)
serizlaedtypesList.Add( EntityManager.GetComponentData<Optional1>(entity))
//for tags
if (entityManager.HasCompoinent<Tag1>(entity)
serizlaedtypesList.Add( new Tag1())
heres some updated sudo code
😛
you should probably write at least 1 component that contains data
so you have it in your template
ocean tundra
all manual serialzation is terriable atm
unity are very strict about what they open on the entity manager, they DO NOT want us to have any methods that are 'slow'
night venture
ok...first thing i notice is it seem its not properly serialized
ocean tundra
lets see some code and json output?
night venture
List<ISerializableComponent> l = new List<ISerializableComponent>();
for (int i = 0; i < tiles.GetLength(0); i++) {
for (int j = 0; j < tiles.GetLength(1); j++) {
if (manager.HasComponent<Tile>(tiles[i, j])) {
l.Add(new Tile());
}
if (manager.HasComponent<RoadComponent>(tiles[i, j])){
l.Add(new RoadComponent());
}
}
}
Debug.Log(JsonConvert.SerializeObject(l, Formatting.Indented, new JsonSerializerSettings { TypeNameHandling = TypeNameHandling.All }));
results in:
{
"$type": "System.Collections.Generic.List`1[[ISerializableComponent, Assembly-CSharp]], mscorlib",
"$values": [
{
"$type": "Tile, Assembly-CSharp"
},
{
"$type": "Tile, Assembly-CSharp"
},
{
"$type": "Tile, Assembly-CSharp"
},
...
]
}
yea ok
so yor first line List<ISerializableComponent> l = new List<ISerializableComponent>();
you need to make a new one of these PER tile
so its actually
List<ISerializableComponent>[,]
other then that looks right
you could probably play with TypeNameHandling setting
you may be able to remove the highest level one (the list) but keep all the others
another tip i would recommend is to have a containing model class instead of a direct list
thats mainly a webdev json rule
night venture
too many suggestions per paragraph xD
ocean tundra
but its helpfull if you want to add a 'top' level setting of some sort
night venture
ok, so looking at the json, seems its working
ocean tundra
yea you just need to break apart the serilzedtype list to be per tile instead of a single massive one
night venture
simplified using handling=objects:
[
[
[
{
"$type": "Tile, Assembly-CSharp"
}
],
[
{
"$type": "Tile, Assembly-CSharp"
},
{
"$type": "RoadComponent, Assembly-CSharp"
}
],
...
nice
ocean tundra
😛
you should just be able to do the reverse
json.net wise i mean
then you will need to loop x and y
and do the following steps:
1 EntityManager.CreateEntity()
2. Loop over every component
2a. call component.AddComponent(Entitymanager, newEntity)
night venture
hmmm...like this? :
string txt = JsonConvert.SerializeObject(l, Formatting.Indented, new JsonSerializerSettings { TypeNameHandling = TypeNameHandling.Objects });
List<ISerializableComponent>[,] l2 =JsonConvert.DeserializeObject<List<ISerializableComponent>[,]>(txt, new JsonSerializerSettings { TypeNameHandling = TypeNameHandling.Objects });
for (int i = 0; i < l2.GetLength(0); i++) {
for (int j = 0; j < l2.GetLength(1); j++) {
tiles[i, j] = manager.CreateEntity(archetype);
foreach (ISerializableComponent c in l2[i, j]) {
c.addComponent(manager, tiles[i, j]);
}
}
}
yea that looks right
night venture
today u deserved a beer, coke or whatever
ocean tundra
it works?
night venture
haven't tried yet, but looks promising
(and furthermore when undesrtanding what it does) 😛
ocean tundra
yup
night venture
ArgumentException: Unknown Type:Unity.Entities.IComponentData All ComponentType must be known at compile time. For generic components, each concrete type must be registered with [RegisterGenericComponentType].
ocean tundra
implement IComponentData directly on your structs?
night venture
probably at c.addComponent(manager, tiles2[i, j]);
ocean tundra
not via a parent interface
night venture
but then, how it know it has to implement addComponent?
night venture
ok, i forgot xD
ocean tundra
and can you show me the code for your addComponent
night venture
using Unity.Entities;
[GenerateAuthoringComponent]
public struct Tile : ISerializableComponent, IComponentData {
public void addComponent(EntityManager manager, Entity e) {
manager.AddComponentData<IComponentData>(e, this);
}
}
yea thats wrong
night venture
xD
ocean tundra
manager.AddComponentData(e, this);
should work
but also this is a tag component...
you may have to use manager.AddComponent<Tile>(e) instead
but for non tags i think its right
night venture
....so....? xD
ocean tundra
so
try this first manager.AddComponentData(e, this);
if that dosnt work
then try this manager.AddComponent<Tile>(e);
night venture
on my way
ocean tundra
if that still dosnt work then no clue 😛
night venture
should ISerializableComponent implement IComponentData ?
night venture
seems it works
ocean tundra
nice
safe lintel
im hoping it brings a whole slew of updates to every other package too.
also want an easier to use dots audio package
and navigation
minor sapphire
I have been keeping away from dots audio so far
is it usable?
or should I keep away a while longer 😄
velvet oxide
is there an alternative to using .Equals on structs? I keep getting a Burst error saying my IBufferElementData does not contain IEquatable<object> even though I do have it implemented.
Burst error BC1002: Unable to find interface method `System.IEquatable`1<object>.Equals(!0)` from type `DotsPhysics.PhysicsHitBuffer`
can you show the component code?
velvet oxide
public struct PhysicsHitBuffer : IBufferElementData, IEquatable<object>
{
public FixedString64 name;
public PhysicsPositionData positionData;
public PhysicsHitData hitData;
public PhysicsDebugData debugData;
}
i wonder if .equals is doing some sort of boxing
oh of course
its to object
also shouldnt there be a equals method?
velvet oxide
I tried doing my own Equals as well and still got the burst error
ocean tundra
IEquatable<object> change object to the actual type
objects = class which wont work in burst
and im pretty sure you need to implement the interface
velvet oxide
lol thanks no idea why its workign now but it is. I swear i tried this before
thanks again @ocean tundra
ocean tundra
all good
velvet oxide
I must of had some other burst error junk before
minor sapphire
If I do
Dependency = Job1.Schecule(Dependency);
Dependency = Job2.Schecule(Dependency);
Will job2 always run after job1, or only if there is actually something in job1 that job2 reads/writes?
ocean tundra
yea job 2 will be after job1
minor sapphire
I'm trying to schedule these two to run at the same time
ocean tundra
yea ok
minor sapphire
but it keeps hating me when I try lol
ocean tundra
var initialDependency = Dependency;
var job1D = Job1.Schedule(initialDependency );
var job2D = Job2.Schedule(initialDependency );
Dependency = JobHandel.Combine(initialDependency , job1D , job2D );
JobHandle handle1 = new Job1().Schedule(count1, Dependency);
JobHandle handle2 = new Job2().Schedule(count2, Dependency);
JobHandle.CombineDependencies(Dependency, handle1);
JobHandle.CombineDependencies(Dependency, handle2);
Ok so this produces error:
InvalidOperationException: The system System1 reads DynamicBuffer1 via System1.Job1 but that type was not assigned to the Dependency property. To ensure correct behavior of other systems, the job or a dependency must be assigned to the Dependency property before returning from the OnUpdate method.
so I thought, I must do
Dependency = Job1.Schedule...
ocean tundra
no at the end
where you combine
you need to assign back to Dependency
**Dependency = **JobHandle.CombineDependencies(Dependency, handle2);
minor sapphire
sigh lol
I'm still not used to dependency management really
been doing it for ages but it still catches me
ocean tundra
yea me too
minor sapphire
ctrl f on CombineDependencies to check all usage lol
ok good I did get it right in other places haha
frigid quarry
Hi guys, is there a way we can manual control culling in ecs/ hybrid rederer? I have some object have shader that changes position, find the current culling system doesn't work on that
ocean tundra
@frigid quarry im not sure which system(s) control it
but you should be able to find them. copy them into your project
and disable the original systems
sullen mason
is this tutorial still relevant?
https://www.youtube.com/watch?v=BNMrevfB6Q0
having some trouble replicating the behavior shown 🤔
more specifically, it seems like their bullets are converted to entities, but still use the update function... am I missing something here, or is this tutorial just skipping crucial parts of information?
ocean tundra
havnt seen it
but at a guess no
oh i was at that session
😛
definitely not relevant
the ideas and concepts will still apply
but any code will be long obsolete
sullen mason
got it. got any recommended tutorials to follow? been pulling my hair out over this for the past few days...
ocean tundra
umm just the offical docs really
i dont know any up to date tutorials
i think unity learn has something
but not sure its for the latest
the problem is its still very work in process
so not much offical unity focus on docs
and anything the community makes gets out of date fast
sullen mason
right, definitely seems that way after the tonne of tutorials ive been trying to follow 😂
ocean tundra
the core concepts should all be the same
and then look up the same 'thing' in the entities doc
and see the current way of doing it
EG any tutorial that has you writing a job manually to loop over your entities
you should use Entities.Foreach for that
and theres a massive doc page on that
and feel free to ask here, most people are happy to help
sullen mason
any chance you could link me to said entities docs? 🤔
sullen mason
ah, sweet. Thanks 🙂
sullen mason
WARNING
Do not use the Build and Run menu "File > Build and Run" to build DOTS projects. It might work in some cases but this approach is not supported. You must build your project using the Build or Build and Run buttons at the top of the Build Configuration Asset Inspector window.In particular, entity subscenes are not included in builds made via the Build and Run menu and will fail to load with the following error message showing up in a development build: "Could not open <path>/<guid>.entityheader for read".)
what exactly does this mean? i cant find any "Build Configuration" inspector window...
ocean tundra
its another package
sullen mason
Standalone Builds
Making standalone builds of DOTS projects requires installing the corresponding platform package for each of your target platforms:com.unity.platforms.android
com.unity.platforms.ios
com.unity.platforms.linux
com.unity.platforms.macos
com.unity.platforms.web
com.unity.platforms.windows
After installing the platform packages you need, create a "Classic Build Configuration" asset for each platform (via the "Assets > Create > Build" menu). The properties of that asset will contain a "Scene List", which is the only way of adding subscenes to a standalone project. Make sure you add at least one scene or that the "Build Current Scene" checkbox is toggled on.
speaking for this?
ocean tundra
yea those
sullen mason
aight, ive got an error when installing those packages... will reinstall and post here
Library\PackageCache\com.unity.entities@0.16.0-preview.21\Unity.Entities.Editor\LiveLink\StartLiveLinkWindow.cs(370,39): error CS0104: 'Platform' is an ambiguous reference between 'Unity.Build.Platform' and 'Bee.Core.Platform'
sullen mason
so i should rename the "Platform" that these errors are referencing
to "Bee.Core.Platform"?
ocean tundra
what no
'Bee.Core.Platform' is the wrong one
never heard of that
name that something else
ocean tundra
it must be something you added in your project
sullen mason
the only packages ive installed, with ... 5-6 small scripts?
none of which have the namespace "Bee"
ocean tundra
whats your code namespace?
sullen mason
DotsTest
ocean tundra
weird