archived-dots

[late mural]

the fact i cant attatch it to game objects is a small annoyance, meaning i have to create a new script for each location i want it to be in, but otherwise this is real good!

[misty wedge]

Looks like an angry cloud about to shoot lightning at me

[late mural]

haha it does lol!

[rustic rain]

game objects are used for simplicity of authoring

in fact it makes it so much easier to create scenes

and lets you reuse authoring components however you want

[late mural]

yup, and i have no idea how to use gameobjects with them, so for now i wont, although tommorrow morning i will learn how to!

[rustic rain]

Any idea whether all jobs are reusable?

Can I just assign job to field

so I don't repopulate fields again?

or is there some kind of codegen that breaks this approach?

[karmic basin]

Never tried. But you can think of a similar approach where you gather your fields in a struct and keep that around

[viral sonnet]

yeah, should work. only handles need to be updated

[rustic rain]

Do jobs have any sort of code gen related to them?

except IJobEntity ofc

and also ComponentDataFromEntity does it need to be refreshed on each update?

[viral sonnet]

i think for CDFE you need another GetComponentDataFromEntity<T> call. for handles there's Update()

[rustic rain]

yeah

It says it gets invalidated

by structural change

[rustic rain]

hmm get very weird error

and it points to this line

And I don't get where it can possibly happen

Player 100% has physicsVelocity

or maybe it doesn't?

🤔

[viral sonnet]

maybe the index or version is different

i'd say the version is

[rustic rain]

brrrrruuuuuh

why they banned code blocks

[viral sonnet]

huh? they did?

wtf

this is a code block

[rustic rain]

wat

they blocked cs code blocks?

[viral sonnet]

they reduced the character limit from what I saw

[rustic rain]

xD

I didn't remove it

[viral sonnet]

spam filter?

[rustic rain]

no idea, but it's painful

[viral sonnet]

is this one of the job structs that you are reusing?

[rustic rain]

yeah

I'll try to remake it

oh wait

nvm

me dum dum

passing Entity.Null

as player entity

lol

[viral sonnet]

just keep in mind, as soon as you set a new value on the player entity the version changes so your job would have an old player entity. i dunno, reusing a job with entity fields sounds like a bad idea

[rustic rain]

it's fine, it reapplies field on every OnStartRunning

which is triggered by singleton player

kek

[viral sonnet]

there could be still a case where OnStartRunning is not triggered and there's a change on the player entity. reusing job structs is really not worth it other than opening up yourself for some weird problems.

[misty wedge]

I currently have some data in a grid that I am storing in a large dynamic buffer on an entity, and then reading in jobs. for example, the height value of a certain position. There's a few other values I plan to store here, such as the biome. However, not all jobs plan to always read all of this data.

Now I'm kind of torn between only putting single values in each buffer (e.g. 1 buffer for height, 1 buffer for biomes), and each job retrieving the buffers it requires, or just putting all of the values in a single buffer (there is some overlap where multiple values are required, e.g. many jobs require the height and biome at once).

[viral sonnet]

the design itself is bad tbh (random access) so it'll not matter if you put 1 or 5 fields in a buffer element when you profile

but if you want to continue with this, put as much data as possible in it to at least reduce the amount of random access

personally I'd redesign this to a hashmap

[misty wedge]

How does a hashmap make it not random access

Also the access inside of a job is spatially related, meaning it will (hopefully) be in the cache after the first access

I feel like a hashmap will be even slower, since I will just be mapping all grid positions non-linearly, and require an additional hash calculation for each access

[misty wedge]

Also this isn't really true, since less data in the buffer will mean more can fit into cache

[viral sonnet]

I'm not following. What's your internal buffer capacity? From what you wrote you have a large buffer so it won't be in the chunk.

[misty wedge]

Are buffers handled differently if they exceed the chunk capacity?

Basically I'm just caching some data in array form, but it's easier for me to store it on an entity to access in a job. It's also not super huge at ~1024x512

[viral sonnet]

if you don't set any capacity, the default is 8 and will allocate memory in the chunk. if that exceeds or the internalBufferCapacity is set to 0 it'll be a randomly allocated array and work just like a list

so, yes, they work very differently when exceeded

[misty wedge]

I mean if I didn't have it on an entity I would be storing it in a nativearray on a system, I don't see how that would be any better

[viral sonnet]

the bad thing about it, even if they are allocated in the chunk, their performance is underwhelming

[misty wedge]

Why?

[viral sonnet]

I don't know exactly, mostly overhead and the sum of things. are we talking about 1 buffer or several?

[misty wedge]

usually 1, max maybe 5 for a single save game

[viral sonnet]

and you have direct access because you know the index?

[misty wedge]

Access is based on world position

So e.g. for pathfinding I would access the buffer based on whatever node is currently being searched

[viral sonnet]

and what's the method of finding the correct node?

[misty wedge]

WorldPosition / ChunkSize

And then transforming that 2d position to a 1d array index

[viral sonnet]

ok, so the answer is yes, you have direct access by index. in that case, I'm sorry, it's okay what you're doing. to get back to your question, you can put as much data as you want in an element. it's just some pointer arithmetic when you have more fields in the struct. hardly any performance loss. getting to the right memory address though is pretty hefty with lots of access from jobs, so if you stay mostly in the same memory space it nets better performance than splitting up the data in several smaller arrays

and if you don't believe me (you shouldn't 🙂 ) -> profile

[misty wedge]

I don't have massive performance requirements like you, I just like talking about these things to learn and see what others would do

[viral sonnet]

you have enough performance requirements to use dots 🙂

[misty wedge]

I'm using it because because I wanted to learn something new, and I can write a chapter about it in my masters thesis since it's more interesting than "basic OOP setup #1023"

[viral sonnet]

i never had much success with splitting data in general. splitting data is really only worth it if you can vectorize code through it

[misty wedge]

Also it's much nicer for networking than something like mirror (at least I like it more)

[viral sonnet]

the tighter you pack the memory the less the cpu has to jump around in memory and if you get really low level, this is where it's at

we like to think about cache lines and all that but that's mostly machine stuff

[misty wedge]

I still don't see how this can be true since that would mean there is a significant difference between a nativearray and a dynamicbuffer

Since I'm using the same interface to access them

[viral sonnet]

in how they act, they are very different. a nativearray is always just a block of unmanaged memory. pointer and length. a dynamic buffer can be reduced to the same thing that's why AsNativeArray is a thing but a DB has the ability to either store the data in the chunk or outside of chunk memory. it has also the ability to act as a list, that's why you can add more elements which a NativeArray can't do. as long as the chunk can fit the DB data it will stay there, once it exceeds the chunkSize a new block of memory will be allocated, just like a list and copied and moved there. the data that will always be static and stay in the chunk is that a DB has a pointer, capacity and length (like a list).

to get a better feel for that. try a [InternalBufferCapacity(n)] tag on an IBufferElementData. set it to 0 and you'll see 144 allocated bytes as minimum in the archetype. pointer, length, capacity and something I forgot. set it to 4, 8 or smth, the additional sizeOf the struct will be allocated. also a reason why I don't like DBs because they take away a huge amount of data from the chunk

[misty wedge]

I guess it's a tradeoff, memory for speed. How else would you store multiples of some data that belongs to an entity?

[viral sonnet]

well that's the thing. you trade memory and speed when the data is in the chunk. you'd think that it's great design that the data of a DB stays in the chunk. sounds really fast! right? well it isn't, because it's using pointer access and that trips up any compiler. it doesn't know that the data is just around the corner. so even the whole cacheline thing won't help. also a reason why many, tertle and me included are suggesting to set the InternalBufferCapacity to 0 when the size is unclear. If it's totally fixed an IComp with pinned ptr to the first element and index access is much faster. arrays belonging to entities is the biggest problem in ecs design. the best solution I've found is to use the same entry point for the array, that's speeding things up by a lot. once the cpu knows that you're accessing the same array over and over again timings go down by a lot.

I'm mostly finding reasons for what I've measured btw. even if my reasons are wrong (I don't pretend to know every little detail) the performance characteristics are 100% true

[misty wedge]

Very interesting, thanks for the write up. I assume something like a fixed array would be much faster then

Ah you wrote that, I just missed it

the best solution I've found is to use the same entry point for the array, that's speeding things up by a lot.
By this do you mean use the same base pointer and then offset from that?

[viral sonnet]

yes

[misty wedge]

Would it be enough to get a DB, cast as native array and index it? Or is it doing something else under the hood

e.g. get a pointer to the first element and then index off that (DynamicBuffer.GetUnsafePtr)

[viral sonnet]

there's a lot of overhead to get the DB pointer which can't be optimised that well, so a plain native array would be better. once you have the ptr and length a DB and NativeArray is 1:1

also a good idea would be to cache the DB ptr/length in every way possible to reduce this lookup

[misty wedge]

Kind of an issue if the DB can change though

[viral sonnet]

right, I don't mean cache it over many frames. just in one chunk iteration for example or on job schedule

[misty wedge]

Ah, alright

I guess for a lot of these things it depends on whether or not you are actually bottlenecking on these things. I don't doubt it's a performance issue, but it feels overkill to do this for every DB

But it's nice to know if I ever run into performance issues

[viral sonnet]

I'm 100% sure, the most performance issues in ECS right now are coming from DBs. They are just so convenient to use 😄 I've replaced many different kinds of DBs now in my project and I'm 80% sure that every DB can be replaced with another NativeContainer.

[misty wedge]

I mean yeah you could stick an UnsafeList inside a component, I guess the "useful" feature is the automatic cleanup

[viral sonnet]

most recent example, every entity/npc/player needs to know with which entity it is in combat with. could be 0, 1, 250k 😄 very basic thing but I need to check with new combat events if the entity is already in combat with the other entity, otherwise I'd have lots of duplicates. my first try was with a DB and calling Contains(Entity) so a linear search. perf tanked to 8seconds a frame. then I used tertles dynamic hashmap, MUCH better, the linear search was now gone but still bad overall. I thought, well, as 250k are attacking 1 target, calling ContainsKey 250k on the boss is expensive but it turned out, the single ContainsKey on the 250k casters was the expensive part due to the random access. I then rewrote it to a hashset and hashmap, slightly more complicated and more involved but that reduced it down by a LOT. sadly I don't remember the exact numbers anymore

[misty wedge]

I don't see how a hashmap removes random access, the data is not linearly laid out (between keys)

[viral sonnet]

it's random but not totally random is my best explanation. like it's local to where the memory pointer is. the cpu knows where to start and then just access the correct index

[misty wedge]

But that should be the same for tertles dynamic hashmap. The block of data where the hashmap is stored (a dynamic buffer) is linearly laid out in memory. At the end of the day both should be storing their data in arrays

[viral sonnet]

no, my solution is one long key/value array. the db hashmap is a fragmented key/value array all over the memory.

[misty wedge]

Why is it fragmented, the dynamic buffer just knows the start of it's memory address, and the length

[viral sonnet]

for the casters, which were so expensive, there were 250k, very small key/value arrays

[misty wedge]

Ah, you meant you had many dynamic buffers before, and moved it to a single hashmap?

[viral sonnet]

which also meant,250k unique random memory access instead of 1

correct

[misty wedge]

Yeah that makes sense it would be much faster, I thought you meant you had the single hashmap stored on a dynamic buffer

[viral sonnet]

nope, that was sadly not possible with the design. the single hashmap performed, as expected really well

[misty wedge]

Your original approach would also have been slower even if it was stored as nativehashmaps, and not on a dynamic buffer. close to linear memory access is just much faster than spreading it all over the place

[viral sonnet]

right, and that's the problem when you have many entities and DBs. as long as there are not many entities, all is good but when this reaches a certain point the melting process starts

just something to keep in mind. if you know the pros and cons of DBs they work really well

[misty wedge]

I also really depends on how many entities you have 😅 I don't think very many games require 250k actors

[viral sonnet]

i mean, the whole point of ecs is to keep random memory access down. hehe, certainly not 😄 I just saw many devs considering grid tiles as entities with a DB so my alarm bells go off

you can basically write OOP code when relying on DBs too much. well, burst helps a little 😄

[gusty comet]

i am new to using DOTS but i have a quick question, is turning objects to entities improve the performance? my game is a simple lab room, and i have tones of objects in the room like beakers and lab equipments that are there for the environment style only, if i turn them all to entities, would that help ?

[oak sapphire]

If they are just static, then no, it will not change anything afaik.

There is no data being changed.

[viral sonnet]

yeah, for graphics and just a small but detailed scene you don't necessarily need entities. at least not right now because the hybrid renderer is still very alpha

[viral sonnet]

can someone eli5 write groups?

like, i don't understand their usage. to hijack some entities so another system doesn't process them?

[rustic rain]

allows you to make systems that can be disabled by components that use that write group

you have system that writes to local to world and uses write group filter

if you add any component with write group to LocalToWorld

system will filter that entity

[viral sonnet]

i see, so does this only make sense when i'm not really the programmer of a system or the system is internal and not changeable?

[rustic rain]

well, it makes sense for a lot of cases

[viral sonnet]

any examples? do you use them?

[rustic rain]

I personally haven't got a chance yet

unity itself has an example:
when player invincible he doesn't need health color to be calculated

so health calculation system uses filter

and if invincible tag is applied to entity it's filtered

in case you'll have some other component that will be supposed to disable health color system you just add to it write group filter

for example, it can be used to switch into other logic of calculating color

[viral sonnet]

ah thanks, that makes sense

[frosty siren]

should i dispose blob assets manually?

[viral sonnet]

they are disposed automatically

[rotund token]

that depends how you create them

if you simply create one yourself outside of a blob asset store, nothing is magically disposing it

var b = BlobAssetReference<int>.Create(1);

for example

a common case of this is creating your own physics colliders in a system

if you add it to a store

var store = new BlobAssetStore();
store.AddUniqueBlobAsset(ref b);
then when you dispose the store it'll dispose all the blobs

during authoring unity gives you that store for you via GameObjectConversionSystem and handles its lifecycle for you

[viral sonnet]

oh damn, that's new. is this also true when using a BlobBuilder?

[misty wedge]

Here I am creating and disposing of all of my blobs manually, didn't know blob asset store exists

[viral sonnet]

shouldn't the leak detection tell me about it? 🤔

[rotund token]

no

blobs don't have dispose sentinel

[misty wedge]

Is there any point to using a blob asset store if I currently already have a system that manages them?

[rotund token]

not really

[misty wedge]

Guess I'll leave it as is then

[rotund token]

if you read the blob builder comment

    /// 1. Declare the structure of the blob asset as a struct.
    /// 2. Create a BlobBuilder object.
    /// 3. Call the <see cref="ConstructRoot{T}"/> method, where `T` is the struct defining the asset structure.
    /// 4. Initialize primitive values defined at the root level of the asset.
    /// 5. Allocate memory for arrays, structs, and <see cref="BlobString"/> instances at the root.
    /// 6. Initialize the values of those arrays, structs, and strings.
    /// 7. Continue allocating memory and initializing values until you have fully constructed the asset.
    /// 8. Call <see cref="CreateBlobAssetReference{T}"/> to create a reference to the blob asset in memory.
    /// 9. Dispose the BlobBuilder object.
    ///
    /// Use the <see cref="BlobAssetReference{T}"/> returned by <see cref="CreateBlobAssetReference{T}"/> to reference
    /// the blob asset. You can use a <see cref="BlobAssetReference{T}"/> as a field of an <see cref="IComponentData"/>
    /// struct. More than one entity can reference the same blob asset.
    ///
    /// Call <see cref="BlobAssetReference{T}.Dispose()"/> to free the memory allocated for a blob asset.```

specifically tells you to dispose it

but yeah, doing conversion whole thing is much easier as you can just pass it to the store

and let the world handle it

[misty wedge]

Wouldn't I still need to call Dispose on the store when the world itself is disposed?

[rotund token]

the store passed during conversion is created and handled by the world

[misty wedge]

Ah I meant a custom store

[viral sonnet]

yep, I've missed that. so I just create a manual BlobAssetStore, throw the blobs in there and call dispose on the store. should be good then, right?

[rotund token]

yes you would just need to dispose the store

[viral sonnet]

thanks for bringing that up then 😄 don't want to have a leak

[rotund token]

the big benefit of the store is actually how it works

if you pass in a new blob asset that already exists in the store, it disposes what you just created and then returns the existing store

[misty wedge]

Yeah I convert a lot of scriptable object data to blob assets for use in jobs, I guess mass disposing of them when the world is disposed might be faster with a store, but probably not worth it

[rotund token]

so that you don't duplicate memory

e.g.

ok unity discord

i hate you

[misty wedge]

haha

[rotund token]

fine screenshot it is

[misty wedge]

It checks the memory contents to find equality?

[viral sonnet]

are you also having trouble with code blocks?

[misty wedge]

works for me 🤷

[viral sonnet]

there's a line limit or smth

[misty wedge]

Looks like the bot only scans new lines though

So make a small code block then edit it

[viral sonnet]

lol, that's a really dumb change for such a code centric discord

[misty wedge]

Yep

Something I was wondering about since the 0.51 migration is coming up soon for me I guess:

How exactly does the subscene workflow generate prefab entities? Are all entities that are referenced by scene objects that have an authoring component that points to the prefab created?

I still don't use any subscenes since it was broken a while ago due to some weird UIToolkit error

[viral sonnet]

can you rephrase that question?

[rotund token]

well i posted the giant chunk like 2 comments above

[misty wedge]

Basically, I want to know under what conditions a prefab is converted to a prefab entity

[rotund token]

then tried to post 6 lines then and discord just would not have it

[misty wedge]

I assume if a subscene component references it

[viral sonnet]

oh, yeah, right. that's so weird

i think you should test it, you can view the converted entities in the hierarchy. just make a subscene and put some gameobject in there

[misty wedge]

I guess I can test it now, I never actually attempted to use subscenes again

I just remember they broke due to UIToolkit or something

[viral sonnet]

they are just converted at editor stage and the whole scene is saved in a nice format for fast reading. the downside of this is that the entity that is created doesn't match the runtime entity index/version

so referencing in that sense won't work unless it goes through patching

most cases should work as expected though, just something to keep in mind when wanting references outside of the entity world

i.e. some monobehaviour wants the entity index/verison

[frosty siren]

@rotund token thank you for such detailed answer!😀

[misty wedge]

But it patches it for other components that reference the gameobject?

[viral sonnet]

yes

[misty wedge]

Hmm, still kind of annoying since I have some scriptable objects that reference prefabs

[viral sonnet]

yeah, those are a problem. I don't have a real solution for that.

[misty wedge]

Eh, worst case I'll just slap a component on it that maps to the assets guid and then find it that way

[viral sonnet]

I'd be really glad if the editor->runtime would match tbh. It would open up a lot of things that can be done much more elegantly

but that's probably a pipe dream and really hard to do out of the box

[misty wedge]

I don't think it would even be possible, since you can load a subscene into the world multiple times

[viral sonnet]

yeah, right. there's so much normal cases that would totally break this. maybe some additional call could be made. i just feel like there should be a solution in place for that problem. especially when this whole hybrid GO thing is here to stay for much longer

[misty wedge]

You mean the conversion workflow or actual gameobjects due to missing features (animation)?

[viral sonnet]

the actual gameobjects. if you stay within entities, references are working great

[misty wedge]

Ah okay

[viral sonnet]

it all relies on spawners now, although I'm pretty sure that's just a personal dislike and it's actually not a real problem. it's just a habit that I want to have the actual enemy in place and not a spawner. I've written a bunch of helpers to draw prefab meshes so spawners can be seen in the world so it gets better with some tools

[misty wedge]

I mean you can still just place entities right? Since a gameobject in the scene is just converted to an in-place (non-prefab) entity

[viral sonnet]

it's about the presentation of the entity, the gameobject with the audio sources, skinned mesh renderer, animations, etc...

[misty wedge]

What does DOTS convert those things to atm?

[viral sonnet]

hm, not much actually. i've not even seen one good example in their repo when it comes to that. the best thing is companionobjects that link the GO and entity but I've not got that working in the authoring stage

[misty wedge]

Hmm, ok

I just use gameobjects for all rendering / sound related stuff

[viral sonnet]

unity has us pretty much hanging in the air unless I've totally missed something. from what I know everyone is doing their own thing when it comes to Go presentation for entities

everyone does that right now 🙂

[misty wedge]

Yeah I just never checked if there was some dots native stuff out already

I saw that audio library that dreamingimlatios posted

[viral sonnet]

it was promising with the animations package and then it went dark

[misty wedge]

I'm working on a graphically very simple 2D game, so I feel like I really dodged a lot of the difficulties with more complex rendering stuff in DOTS

[viral sonnet]

hm, some what. you still need to make the link and manage, sprites, etc... so in some sense it's very similar

[misty wedge]

How do people handle something like mecanim? Have a monobehaviour that reads the component data and updates the animation controller?

Or the other way around? (system that writes values to animation controller from component)

I mean I guess it really isn't very different from settings sprites via animations...

[viral sonnet]

pretty much that, i have a animation comp that is written to via burst and then a managed system that copies the values to the animator

feels like duct tape overall 🤣

[misty wedge]

I guess for a 2d game it's at least easier to write a simple animation system that works with burst and jobs

[safe lintel]

mecanim in dots feels like hot potato coding, “no you deal with this data, no you deal with it!” Each system continuously passes the buck til it finally gets to the animator

[misty wedge]

Can you profile timings in burst jobs, or do you need to turn off burst to profile it?

[safe lintel]

i havent actually had luck profiling specific things inside of jobs but that was a while ago

[viral sonnet]

you mean like ProfilerMarker?

[misty wedge]

e.g. UnityEngine.Profiling.Profiler.BeginSample

[viral sonnet]

in system: static readonly ProfilerMarker marker1 = new ProfilerMarker("marker 1"); set this in the job then

and then you can beginSample, endSample or use auto() instead and measure the thing in brackets

works for burst too

[rotund token]

isn't it nice when you just open burst inspector, look at your job

and you have 161 lines of simd in a row first try

[late mural]

hey this is probably a stupid question, but do rigidbodies and colliders convert automatically to the entity version of those components, or do you have to replace them with the entity versions manually?

[rotund token]

the physics system includes conversion from existing physic components, so yes they convert automatically

[late mural]

oh ok thank you so much!

[rotund token]

if you use the new physics shape, physics body components though you get new features and control

[late mural]

ooh, id like to try those, how do i find them?

[rotund token]

just add physicshape/body to your gameobject

instead of boxcollider etc

[late mural]

how, i tried searching for them in the component menu, they dont appear to be there

[rotund token]

you have unity physics installed right?

[late mural]

i hope so, i just installed hybrid renderer and i thought it auto installed everything else with it, but ill go check brb

[rotund token]

it wont install physics

you need to install that yourself

[late mural]

oh ok, well then ill install it manually, thank you so much!

[rotund token]

(and body up top)

[late mural]

thanks!

[rotund token]

body is for dynamic objects

if you leave off body it'll be considered a static

[late mural]

ok that makes sense, thank you so much!

[remote crater]

day 9 and over 50 hours of not being able to compile a standalone approaching, no one answering my addressables questions. It should be easy... My game ain't making the .json file.

[rotund token]

a) probably wrong channel for #📦┃addressables
b) how are you building them anyway?

[remote crater]

I know tertle, but no one answers my question for 3 days

I'll get the info together as concise as possible. My system is building a build, takes 3 minutes a try

[rotund token]

i don't think we need that much information about building addressables (and honestly if it's more than about 2 sentences I won't read it =D)

it's usually 1 button to build them?

really just want to check if you're building them the normal way or have a custom build script doing it etc

[remote crater]

I changed build paths along the way because I thought it was putting settings.json in wrong place

Here I have no settings.json on my entire computer

Its not making the assemblies

By default, Unity should have assemblies work, I was using nothing custom at all

Error I get: System.Exception: Invalid path in TextDataProvider : 'C:/UnityProjectfiles/StarfighterGeneralOnSteam/Builds/buildme/starfighter_Data/StreamingAssets/aa/settings.json'.

Looks to be an unsolved bug from Unity from 2 years ago: https://forum.unity.com/threads/addressables-not-loading-in-build.925982/

Looks like I have to downgrade to: 1.16.15 addressables?

[rotund token]

is there addressables in

PROJECT\Library\com.unity.addressables\aa\Windows

and if not, when you build addressables what error do you get

[remote crater]

I have no aa directory

I get no error building addressables

[rotund token]

that means you've never built it in the editor

[remote crater]

I built it many times

[rotund token]

so it's never going to be included in your build

[remote crater]

Addressables group->build

Or am I doing it wrong?

[rotund token]

build, new build, default build script?

[remote crater]

yes

[rotund token]

hmm i'm looking at your addressables

is everything in 1 group?

[remote crater]

yes

I had 2 groups before

but i downgraded to 1, because Unity had problems

[rotund token]

that's going to be problematic for you

but thats your future problem

[remote crater]

I want it working now tho

[rotund token]

(it means none of your assets will ever unload from memory)

[remote crater]

If unity can't handle 2 groups, I'll do one. I'm not picky

[rotund token]

anyway click on your group

what does the inspector show

[rotund token]

you should have anywhere from 10 to 1000 groups

[remote crater]

I know, I had 2 groups to start

1 music, one my droppables

But when it didn't work, I thought Unity had an error understanding more than one group.

idea

[remote crater]

maybe I have to make the directory by hand

[rotund token]

output path seems right

can you open advanced options

[remote crater]

Unity never made those directories

Maybe it was a bug that it forgot to mkdir

I'm gonna make those directories by hand and try

[rotund token]

it should error when you build 😐

[remote crater]

my build goes through and game plays

just sends tons of errors trying to load the addreessables it can't find

[rotund token]

yeah because its clearly never been build

because its not in that folder

so it's never copied it to streaming assets when building your game

so its never been included in your build

[remote crater]

BuildPath for group 'Default Local Group' is set to the dynamic-lookup version of StreamingAssets, but LoadPath is not.

[rotund token]

if it's not in Library\com.unity.addressables\aa\

there's nothing to copy to your game

[remote crater]

I never saw the errors on a "Addressables build"

Now I see there are ones, it uses the same console unity uses for play.

[rotund token]

what is the final log

when you build addressables

[rotund token]

well that's a good start

[remote crater]

Yah, we're getting someplace

This is the only place I could google about it: https://forum.unity.com/threads/buildpath-for-group-is-set-to-the-dynamic-lookup-version-of-streamingassets-but-loadpath-is-not.1230894/

I'm pretty sure we're striking distance from ending this tech hurdle.

ok

my build path and load path were set to the same thing, and it can't be like that says: https://l-u-k-e.tistory.com/13

Looks like it is building now. TY tertle, you're a boss who helped me TOOOOOONS. I want to be making the hundred mil a month like other MMOs and hook you up. lol. I personally don't care about money, I'd rather just be gamin or coding.

As a UI guy, I'd recommend that if a build fails, and the console window is not visible, that it becomes rendered.

I quite literally got no feedback and assumed the build succeeded

Addressable content successfully built (duration : 0:04:00.474)

[late mural]

hey stupid question, but how do i get a reference to an entity that the script is attatched to, a bit like how you can use this.gameObject to get the game object the script is attacched to?

[remote crater]

yes

okay, simple and useful, let me get you code

You want an entity from a monobehavior script or a Systembase ?

[late mural]

monobehaviour would be nice

[remote crater]

I got you my man.

[late mural]

cause i plan on having it attatched to an entity, and it destroys the entity after existing a certain amount of time basically

[remote crater]

Its useful to use entities in some monobehaviors, especially the player controller

Wait that should be a system base

I have the code for you

[late mural]

ok

[remote crater]

if you use monobehavior for that it slows you down

[late mural]

oh ok, what should i use then?

[remote crater]

Sec, let me get u code

https://pastebin.com/F83NGB8D

If you attach a Bastian to an Entity, it sets it to destroy

Let me give you the destroy code

[late mural]

ok nice!

thanks so much

[remote crater]

https://pastebin.com/ptYSQzg5

you'll have to edit this radically to get it to work tho

Use it as notes

[late mural]

ok i shall, thank you so much!!!

[remote crater]

Basically it cycles through entities with Bastian and destroys them, and or drops loot and or destroys other entities dependent on it

Its overkill

[late mural]

ok thank you so much!

[remote crater]

I'll get you a destroy over time too

Thats more of a DESTROYNOWNOW

[late mural]

lol

[remote crater]

https://pastebin.com/7qE6H9C2

[late mural]

ok thank you so much!

[remote crater]

https://pastebin.com/vzAgWm8Z

So do you know there's two systembases

And two icomponentdatas

The systembases processes each

It might be a bit unorthodox, uneloquent and such, but this code is tapping into power and early.

But hey

let me answer your original question

how to read and write to an entity from mono

[late mural]

ok, ye ill mess around with both ways, and see what works good for me, considering how i am a bit of a newbie to dots lol (also pasting code doesnt work in this channel, just screenshot or pastebin it)

[remote crater]

Here's how to add componentdata via mono: https://pastebin.com/BETtXwzz

You want to do EVERYTHING you can in dots

[late mural]

thank you so much!

[remote crater]

The only thing you do not want to do in DOTS is your player controller, and only if you have a mature game you're expanding to have a dots level

Translation entPos =World.DefaultGameObjectInjectionWorld.EntityManager.GetComponentData<Translation>(shootingEntity);

That's how to get componentdata

[late mural]

ye luckilly im not doing much with dots, just a large amount of frictionless spheres that would otherwise grind performance to a halt without dots, but ye all you have said shall be very useful, thank you so much!

[remote crater]

Yah, I gave you some stuff to mechanically putz with

Its unedited from my current game, so you got ship wreck comments and nonsense constants you can't use

[late mural]

ill figure out some way of making work eventually though, i always do lol

[rotund token]

My character controller is written in dots >_>

[remote crater]

IT WORKED MY MAN! I should be able to make a live compile soon

[remote crater]

Thats perfectly fine. There's just a super awesome trick for people with mature Gameobject games they had on steam for years or jury rigging it. I told you before?

I'll tell it again, it's exciting.

So say you have 300 levels of games in your GameObjectgame and people loved it for like 5 years.

You want to add dots levels

But don't want to make a new game.

You take your already working Player controller.

At the very start, get position/velocity/etc of your player

[rotund token]

well, anyone who is suggesting switching an existing project to dots I have simple advice - don't ^_^'

[remote crater]

Have your player gameobject be invisible and non rendered.

Then at the end of the cycle, add the velocity, rotation

Boom!

Your player is now an entity

[rotund token]

that's just the standard hybrid workflow

[remote crater]

Super super simple

Enemies, enviro, items, etc should be all new

All new scripts for stuff except the player

Having one monobehavior script doesn't hurt dots performance

But having hundreds of mono for enemies would.

The best part is as you change your player for gameobject land, it auto changes for entity land

[rotund token]

the same controller on my characters is the same controller on my enemies

[remote crater]

You edit once, change twice

oooh,

You could... Say... Super Mario 64 sequel take over them with a hat

Or, get apower up

where every move you input

gets inputed to them

Make a dance party

Dad, tell us again how you defeated the bad guys? Oh we taught them how to dance and have fun. Turns out people having fun don't want to take over the world much.

[late mural]

another stupid thing, but couldnt you use a component system on each entity the way you would instantiate, but destroy entities instead of creating them, or is this not a good idea?

[remote crater]

when you do instantiate an entity

it has all the components the old one had

so the way I datapool

is instantiate one entity

then save it in a List

and then instantiate it by index

It generates SOOOOOO fast its basically instant

Its better than buffer datapooling in gameobjects

Thats one of many things DOTS ECS got right

ecb.RemoveComponent<Disabled>(e);
ecb.RemoveComponent<DisableRendering>(e);
ecb.RemoveComponent<DOTS_OBJECT_POOLER>(e);

I put a Disabled/DisabledRendering and DOTS_OBJECT_POOLER tag

Diabled and DIsabledRendering are UNitys

[late mural]

ooh that is clever, i should probably do that eventually, but for now ill keep using my stupid instantiation script cause it works well enough for now lol, but imma try my odd method of destruction now because even if it aint the fastes i think it may work in a stupid way

[remote crater]

DOTS_OBJECT_POOLER is mine so I don't process it in places

Sometimes you gotta get stuff working to do more

But DOTS/ECS is the way for all

Don't do MONObehavior stuff you want to keep long term

I recommend not using it all for destroy

[late mural]

i wont use monobehavior for destroy, i wanna see if i can do it in a component system in a similer way i did the instantiating, but with destruction instead

[remote crater]

Goodnight all, tertle solved all my problems

God bless, keep it real.

[late mural]

apparrently Time.time does not exist in a component system, are there any replacments for it?

wait im stupid i can just use delta time and do the equation backwards

from googling around it seems there is no way to figure out what entity the script is attatched to, atleast not in a component system, surprised there is a this.entity or something like that

guessing im missing something obvious though, so any dots magicians feel free to correct my stupidity

[drowsy pagoda]

I just updated Entities to 0.51 (from 0.50) and all of a sudden Newtonsoft disappeared from the face of the earth. I'm getting errors that Newtonsoft namespace cannot be found. I looked at my ASMDEF where they were and they are now greyed out and says None. I can't find it in the UPM either anymore. What's going on?

Just downgraded back to 0.50, and Newtonsoft.Json.dll appeared again (even in UPM). What am I missing?

[drowsy pagoda]

Hmm..updated Newtonsoft from 2.0.0 to 3.0.2, and then updated Entities to 0.51 no problem.

[rotund token]

entities removed it's dependency on newtonsoft

so unless you manually add a reference (e.g. updating it) it won't be included in your project

[drowsy pagoda]

I see. Thanks for that clarification.

But why would it disappear from UPM? I was going to add it manually but it wasn’t in the registry. I had to downgrade entities, upgrade newtonsoft and then upgrade entities lol

[late mural]

yup, the more i research the less i find, it seems instead of being able to use anything simple, ill have to ring around the posy until i find the entity that is the script is on, which seems a bit silly personally, but my research is pulling up no better solutions, odd

[rotund token]

your question doesn't really make sense to me

and i'm not sure what you're trying to do

[late mural]

so basically i have a script attatched to an entity, that destroys said entity

that is what i want

[rotund token]

so that right there

makes no sense

because you don't attach scripts to entities

you attach components

which are just data

[late mural]

oh

[drowsy pagoda]

Not quite sure what exactly you’re asking, but if you are trying to tap into the entity that is being queried inside a ForEach loop, have the first argument be Entity entity…

[rotund token]

the whole point of entities over gameobjects is to not attach logic to the entity

gameobjects have components attached but these contain data + logic
entities have components attached but these only contain data and systems provide the logic

that is like the fundamental of ECS and the difference with gameobjects

[late mural]

ok that makes sense, so rather what i should have is each entity having a float for time, that then a loop goes round for all entities and lowers teh time and if any of the entities time is 0 or lower, i destroy that entity, is that more correct then?

[rotund token]

yep

[late mural]

ok, thank you so much!

[rotund token]

you can have a LifeTime component with a value in it

and a system that iterates all these components, when it hits 0 (or when it hits X time) destroy the entity

[late mural]

ye that would make sense, i shall start figuring out how to do that!

small question though, how do you make components, is it just a regular c sharp script with the data in it that is attatched to the entity

[late mural]

also how do you loop through a list of all entities, all my googling is coming up nothing, i have a bad feeling im missing something obvious

and ye i still cant figure out how to add components, it seems that all the info on dots is really well hidden from google, and if you dare add "unity" to the end of any search it automaticallly pulls up stuff about only GameObjects and nothing about entities

[graceful herald]

@late mural https://github.com/Unity-Technologies/EntityComponentSystemSamples just poke around the samples. They have pretty much everything you could need. Either do git searches or actually look at the readme use cases for each sample

[late mural]

oh ok thank you so much!!!

also is there any good solution to not have to restart unity every time you add a component, or do anything dots related, cause i have to wait about 10 - 15 minutes each time for unity to start up again

[rotund token]

why are you restarting?

[late mural]

because for some reason i just get an error that my scripts cannot be found in the file system, despite me literally dragging my scripts from the file system onto the objects, as such i have to restart to fix this odd bug

this happens every time i create a new script, or add dots to an old script

[rotund token]

haven't seen that specific one before

that said, entities 0.51 has its problems

if you want a more stable experience 0.50 is probably where you want to stick for learning

(though you have to stick on 2020.3.3X)

[late mural]

oh that is unfortunate, guessing 0.5.0 does not work on the 2021 lts versions?

[rotund token]

*0.50.1
no 0.51.0 brought 2021 support

[late mural]

oh that is unlucky, i guess for now considering how the rest of my project is built on 2021 and i dont wanna downgrade, im just gonna keep dealing with these odd bugs and hope a patch comes out soon, and also use min max strategies to minimise the amount of restarts i do

[rotund token]

@viral sonnet finally ran into your, GetNativeArray chunk version bumping issue, ^_^
You did something similar to this right?

var original = conditionActiveTypeRead.GetUnsafeReadOnlyPtr();
var updated = this.conditionActiveBuffer.GetUnsafeReadOnlyPtr();

var hasChanged = UnsafeUtility.MemCmp(original, updated, UnsafeUtility.SizeOf<BitArray8>() * this.conditionActiveBuffer.Length) != 0;

if (hasChanged)
{
    var conditionActives = batchInChunk.GetNativeArray(this.ConditionActiveType);
    conditionActives.Reinterpret<BitArray8>().CopyFrom(this.conditionActiveBuffer.AsArray());
}```

[rustic rain]

Have you tried reimproting in editor?

[late mural]

re importing what?

[rustic rain]

Just right click in folder explorer and click reimport

[late mural]

oh ok, ill try that next time

also im trying to work out how the for each thing works, and i think i get it, but how do you get the entity, after you have gotten the component?

i have a list of all components, and all these variables, and i can change them at will, but no way of getting any entities

the samples are very useful in showing how to do all the variables in a for each loop and all that cool stuff, but they show no way of getting a list of entities

wait could i do ref Entity thing? that might work!

yes it works i think, i just dont know how to use get component on an entity, but otherwise, i have it!

i found a vague description in https://docs.unity3d.com/Packages/com.unity.entities@0.17/api/Unity.Entities.SystemBase.GetComponent.html, although it has no samples, so i have no clue what it means or how to use it, so imma trial and error until i work out what on earth this means

ok i figured it out, but now im even more confused, apparrently im not allowed to modify public floats, because they are not variables, how do i make public floats be variables?

google aint helping, as per usual, and the samples dont talk about how to make floats be variables either, so im hopeing someone might know the answer here?

[tidal gulch]

aren't you trying to access components in a Entities.ForEach?

[rustic rain]

By default for each lambda is bursted and threaded

That means you can't access fields inside

But you can store all variables in onupdate scope

[late mural]

exactly, i think

[late mural]

what does that mean exactly?

[rustic rain]

Make local variables

[tidal gulch]

here I am accessing the rotation component

[rustic rain]

Lambda can read and modify them

[late mural]

oh thank you so much!

[tidal gulch]

you can add more params depending on components you want

[late mural]

i thought you could only have 1 lol

[tidal gulch]

you can only have 8 in a foreach, but you need a custom delegate for more, and surely you will adjust your systems if you need that many

[late mural]

that will be fine, i only need 2 lol, so thank you so much!

also quick thing, just making certain you can have if statements in a for each loop right?

[tidal gulch]

yep

[late mural]

yay, thank you so much!

is this a normal warning to receive?

ok now this is weird, ive somehow made my performance dance, consistently

like it will stay at around 60 fps for about 10 seconds, then for the next 10 seconds it will drop to 10 fps, then the cycle repeats

anyone got any ideas why?

im guessing my code is quite unoptimised, because i used to get around 100fps with around 500 rigid bodies, and now that i get rid of entities that have been around for too long my game suddenly lags way more, but is more consistent

here is my code, incase im majorly messing something up

from testing, it seems to be that the line that destroys the entity is the most expensive

is there a way of getting rid of, or hiding, an entity that is more performant than destroying it?

[tidal gulch]

try to not use EntityManager, instead opt to create an EntityCommandBuffer to process entity destruction

[late mural]

oh ok, i havent heard of those before, ill do some research and try them out, thanks!

[runic spire]

I would think that would give you an error saying that you need to use .WithoutBurst()

[late mural]

i thought so aswell, but for some reason it does not

[late mural]

ok from basic research, those things look super cool, and super complex, im excited to learn them, thanks lol yet again!

[tidal gulch]

yep, whenever you use a System, be sure to use those

[runic spire]

it's pretty simple to set up, you can have it working in a few lines of code, but it does use some long class names

[late mural]

fascinating!

[rustic rain]

because structural changes already contains it

[late mural]

ok, i think im using the buffer thing correctly, just making certain though?

[runic spire]

now that you are using the command buffer, you don't need to run it on the main thread and without burst, so you can get rid of the .WithStructuralChanges()

[late mural]

oh nice!

[runic spire]

it will just queue all your structural changes to execute after you are done processing the entities

[late mural]

should i still use Run() or should i use one of the schedule things

[runic spire]

probably just Run

[late mural]

ok, ill try it out and report back on the performance!

[oak sapphire]

Why not ScheduleParallel?

[runic spire]

well its a pretty small amount of work right now

[oak sapphire]

Yeah that's fair

[runic spire]

I dont see a reason to fully multithread it, but if it's running on tons of entities and has a bit more math involved, then it might be worthwhile

[late mural]

ok, performance is up, it now averages around 20 - 40 fps, which is better, not as good as i would have liked it, but still pretty good

[runic spire]

have you fully profiled it? this is really what's causing the bottleneck?

keep in mind that the editor has burst compilation disabled by default, so it will be much slower than if it's enabled or you are in a release build

there is also stack trace stuff that gets added into the editor / development builds

[late mural]

i havent fully profiled it, i have been going off of changing stuff and seeing the result, without the entire system thing i get 300 fps for a while, until the entities get to such a great number that performance grinds to a halt at around 1 fps or less, adding the system back but not destroying the entities has the same result as the first test, destroying the entities is the bottle neck as such i would assume

[late mural]

ok that would make sense

[runic spire]

if you go here you can enable burst compilation in the editor

[late mural]

ok ill try that thanks!

enable compilation appears to have been enabled by default

[runic spire]

hmm, okay

well if it's really that many entities, you might want to extend your command buffer into a ParallelWriter and use ScheduleParallel then

[tidal gulch]

Well how many systems are there?

[runic spire]

you would just do

EntityCommandBuffer.ParallelWriter parallelWriter = ecb.AsParallelWriter();

and at the end of your OnUpdate you would do

buffer_system.AddJobHandleForProducer(Dependency);

[oak sapphire]

I am highly sceptical that destroying the entities is the bottleneck to be honest.

A simple look at the profiler should give you heaps of information

[late mural]

ok, ill mess around with that and see if i can get anything with it

[late mural]

ok ill do that

[late mural]

1 i think

[late mural]

here is my profiler if that helps

[runic spire]

you would have to click a certain point in time and look through the functions taking up the most computation time

[late mural]

how would i do that, im not very famillier with the profiler sorry lol

[oak sapphire]

We can see the scripts are wrecking you though, so that's nice 😛

[late mural]

lol

the green stuff also wrecks me, although less frequently, only about once a minute

like this

[runic spire]

wowza

something is definitely going wrong lol

[late mural]

yup lol, and i got no clue what

[oak sapphire]

Then you see at the right what % is on which script

[late mural]

oh ok, ill do that brb

[oak sapphire]

You can open those and go all the way to the methods.

Also, enable the entities stuff there

Then at the bottom you will see this

Might be you have several thousands of structural changes every frame

[late mural]

im a little confused on what you mean by any of those, but i got this if this is helpful?

[runic spire]

you can physically move your mouse and click on a time in this graph

[late mural]

how do i open them, i dont get that, i tried clicking on it, and takes me all over the place

im not very clever sorry lol

[runic spire]

it will show you a snapshot of all the collected profile info at that point in time

[late mural]

oh ok, let me try that

ok i did that, it did nothing

[oak sapphire]

You can click the arrows by the way.

[late mural]

issue, i dont have the timeline button, it isnt there on mine, and pressing the arrows just shows their children, none of which seem helpful

[oak sapphire]

Like that

[late mural]

yes, it looks like that

[oak sapphire]

Show me a screenshot of your whole profiler

[late mural]

ok!

[oak sapphire]

It's probably hidden at the bottom

[late mural]

oh ok

so after clicking on that, what do i do next?

[oak sapphire]

Set to timeline

[late mural]

woah that is cool!

[oak sapphire]

Hide the black box 😛

We can't see the scripts

[late mural]

how?

[oak sapphire]

Don't hover over the function

[late mural]

im not

i never was

[runic spire]

click something else then

[late mural]

it opened automaticall

[late mural]

that then creates another box

[runic spire]

lol

[late mural]

lol

[runic spire]

press copy then and maybe it will go away haha

[oak sapphire]

Just click on the timeline again

Select a different frame

[late mural]

tried that, did not work

[late mural]

tried that, did not work

im stuck with this box one way or another

[oak sapphire]

Works for me

[late mural]

aha, by selecting all on the heirarchy, i have managed to remove it somehow!

[oak sapphire]

Lots of Job stuff

Can you open that one too?

[late mural]

open what one, how can you open stuff here?

[late mural]

oh, i didnt see those...

[oak sapphire]

Profiling is an art 😛

[oak sapphire]

I have never seen anything like that 🙂

[late mural]

neither have i, although i suspect for different reasons, cause ive never used the profiler before

[oak sapphire]

Isnt it the Run thats wrecking him?

[late mural]

it is the green stuff and the blue stuff, that is all i know

[oak sapphire]

Does it run all the entities in small jobs now?

[late mural]

i have no clue

[oak sapphire]

I hope Danon5 does

[late mural]

same lol

[oak sapphire]

I always ScheduleParallel my stuff so 😛

[late mural]

ill try that, just to see how that affects performance

[runic spire]

I don't think I have looked at the dots job profile section before, so I'm not really sure what it should look like

[late mural]

lol

[tidal gulch]

Run should not ruin performance like that

It's something else

[runic spire]

but 12 milliseconds on basically the most simple task ever seems strange

[oak sapphire]

For me it's huge blocks of parallel work

On all 12 worker threads

[rotund token]

a) are you creating/destroying entities every frame
b) you are in a fixed update death spiral

[late mural]

yup, very weird, should i go over to the frames when the green stuff takes over and see how the profiler looks then?

[late mural]

i am destroying entities every frame

[rotund token]

you can't destroy something that doesn't exist

so are you creating entities every frame as well?

[tidal gulch]

Can you show how you are creating entities?

[late mural]

and creating them, although creating them quite efficiently, i get around 100 fps without destruction

[late mural]

sure!

[rotund token]

i assume your entities have physics on them

[late mural]

yup

[rotund token]

i think you're triggering a full physics rebuild every frame

[late mural]

oh, that aint good, how would i avoid that

[rotund token]

can you
a) turn jobs debugger off

b) make sure leak detection is off

[late mural]

ok ill try that!

[rotund token]

c) turn off physics integrity checks

then profile it

[late mural]

do each one individually or all at once?

[oak sapphire]

All at once

[late mural]

ok i shall!

[rotund token]

actually for b) On is fine, just make sure it's not full stack traces (Expensive)

but yeah make sure jobs debugger is off (unlike my screenshot) and integrity checks

[late mural]

ok, it crashed my unity trying to turn the integrity checks off, so is it ok if i dont do that one?

[rotund token]

thats the most important one

did you turn it off while in play mode

[late mural]

oh

[late mural]

nope

[rotund token]

odd

all it changes is a precompiler flag

[late mural]

well frozen, not crashed

[rotund token]

and triggers a recompile

[late mural]

must be a long recompile then

been a minute and a half since it started

should i force close unity, or wait for something to happen?

it loaded yay!

[runic spire]

lol

[late mural]

ok now i can continue with the steps

a nice drop into 4 fps land

here is the green domination part, ill do the blue one next

here is the blue

[late mural]

now that ive done all this, any ideas?

[oak sapphire]

The jobs are still scuffed.

[late mural]

ye, it is weird

also the green domination that randomly occurs is odd

oh well, it seems to have had interestin affects on fps

randomly dropping to around 5 fps every 20 seconds or so

sometimes getting up to 100 fps, but usually staying at around 30 fps

anyone got any ideas what on earth is going on?

[oak sapphire]

If the other people don't have suggestions you can try making a development build and profile that.

[rotund token]

how many entities do you have?

[oak sapphire]

That way it is way better in the Jobs and not a lot of overhead

[late mural]

i will try that if there are no other options, but that is cool cause i did not know that you could profile dev builds

[rotund token]

to me it looks like you have a lot of physic colliders really close

alternatively can you save your profile and upload it?

[late mural]

a lot, a couple hundred, the destroy script is meant to keep the numbers lower for hopefully better performance

[late mural]

oh, how would i do that?

[rotund token]

are you creating them all on top of each other

[late mural]

yes, randomly positioned within a 1 ^ 3 space

[late mural]

ok ill do that!

[rotund token]

yeah i think you're literally just killing physics by having too many collisions in a small area

[rotund token]

as for this something in your editor is killing you

but that wont happen in build

[oak sapphire]

Would that mess up the Jobs so bad like that?

[rotund token]

its creating a lot of collision event pairs

or whatever they call it now

[late mural]

ok that makes sense, but it doesnt match reality, when i dont have the destroy script, i get insanely high fps until the number of entities gets too high

[oak sapphire]

The more you know 😛

[rotund token]

yeah but it probably only calculates it once

[late mural]

calculates what once?

[rotund token]

this is what happens when you spawn 1000 physics objects in same spot

[late mural]

oh i see

[rotund token]

next frame

4 seconds vs 2ms

[late mural]

as a solution, should i rather space out the spawning place for my instantiated entities, like have them randomly spawn in a 10 by 10 instead of a 1 by 1?

[rustic rain]

What system is that that calculates everything?

Transform?

[late mural]

yup it is definitely the collisions, just from a little bit of a space out, it is doing much better

getting around 100 fps now

[oak sapphire]

Didn't know the Physics ones also showed up in Jobs, learned something. 👍
Problem is fixed then?

[late mural]

yup, although gonna make my random selection position code be a bit better to help fix the issue even more!

also a bit of a math check up, if i use the random.value, and i times it by a number, call it x, it will give me a number between 0 and x right?

[rotund token]

there's a math.Random function

for inside jobs

but yes your logic is fine

[late mural]

it isnt inside a job luckilly, cause im doing the instantiating very stupidly dont worry

[rotund token]

(though I'd always just use Range() anyway)

[late mural]

oh ye, i forgot that exist, i should probably use that

[oak sapphire]

Yeah, but remember it does not include the last number. So Range(-3, 3) = -3, -2, -1, 0, 1 and 2

[late mural]

ok, is there a range but for floats?

[oak sapphire]

Range(0, 1) also is one of those things you think should work but doesn't

Yeah there is, google it

[late mural]

guessing it just gives 0 all the time, or perhaps an error lol?

[oak sapphire]

Just 0

[rotund token]

only for int

[late mural]

could be useful for claiming a gambling app has a random function, while secretly is still rigged

[runic spire]

yeah, it's just to make it easier to generate a random index

[oak sapphire]

Oddly specific 🤔

But its possible yeah

[late mural]

lol

wondering how parenting works with dots, let us say i have 1 parent object with a rigidbody, and then the colliders on seperate children gameobjects, where should i place the conver to entity component, and should i put it on all of them or only 1?

[rustic rain]

it works disgusting xD

children have Parent component

[late mural]

ah, that helps me a little

[rustic rain]

parents have Child buffer

[late mural]

oh

ok

how do i do any of that?

[rustic rain]

just add Parent component to child

Transform system will resolve it for you

[late mural]

ok, so i dont need the child buffer on teh parent then?

and i have convert to dots on everyone then?

[rustic rain]

sometimes you do need it, but hopefully you won't face that trouble

[late mural]

i hope so aswell

[rustic rain]

I personally had to write an algorithm that restores child buffers through painful checks

[late mural]

yikes

[late mural]

where do you find the parent component?

[rustic rain]

I think it's Transform namespace

not sure

Rider solves them all for me xD

[late mural]

what is rider?

[olive wagon]

an IDE

[late mural]

ok

[olive wagon]

like notepad but better

[late mural]

it isnt, i cant find it there

[late mural]

ok lol

[rustic rain]

Use manual's search to figure out namespace

[late mural]

why do i need the namespace

sorry im quite dumb lol

[rustic rain]

well, you ask about where do you get it

I assumed your IDE didn't resolve it for you?

[late mural]

you need a name space to get it?

[rustic rain]

new Parent()

[late mural]

oh i meant in unity

[rustic rain]

If you convert from GO

they are added automatically

[late mural]

oh good

[rustic rain]

through normal transform hierarchy

if you want to create children manually through code

you add Parent

[late mural]

ok thank you so much

[glacial hazel]

(No intention to interrupt) Hello there! Quick question, is it possible to iterate in parallel (in a job) through the different values per key of a NativeMultiHashMap?
I saw that before there were IJob's dedicated to that kind of collection, but they were removed

For example, before existed IJobNativeMultiHashMapVisitKeyValue

[late mural]

im a newbie to dots, but i swear ive seen the exact thing your talking about somewhere in a unity fluid simulation tutorial

[olive kite]

I'm getting a few JobsUtilities errors when upgrading to v.51 using 2021.3.2f1:
'JobsUtility' does not contain a definition for 'ClearSystemIds'
Any idea how this might be fixed?

[glacial hazel]

Hmm, do you have the proper name of the tutorial or indications of how to find it? xD

[late mural]

ill send a link quickly brb

[late mural]

think it was this one https://medium.com/@leomontes_60748/how-to-implement-a-fluid-simulation-on-the-cpu-with-unity-ecs-job-system-bf90a0f2724f

it uses hashmaps for something i remember, hope it helps!

[glacial hazel]

Hmm yeah, it just goes through all the values, instead of the keys, with an IJobParallelFor

I guess I could do that instead too

[late mural]

welp i wish you luck in finding a solution!

[glacial hazel]

Thank you @late mural !

[karmic basin]

Try to upgrade unity to 2021.3.4 at least

Yeah manual says so

[late mural]

wondering how do you add rigidbody constraints, is it possible to do in the editor, or does it have to be done via code

[runic spire]

https://github.com/Unity-Technologies/EntityComponentSystemSamples/tree/master/PhysicsSamples/Assets/Demos/4. Joints/Scripts Unity's physics sample project has code for physics joint authoring components, and one of those is a limiter that acts as a rigidbody constraint (I think it's the LimitDOFJoint)

[late mural]

oh thank you so much!

woah that stuff is cool looking! I'll try it out soon, thanks a ton!

[gusty comet]

When I first tried DOTS way back when it was just called ECS, the docs were a bit weak and I had to rely on youtube to fill in the gaps. Coming back, it looks like the docs are a lot better, but are there still parts of the workflow that I'll need to learn from third-party? And if so, has enough changed with the 0.50 update to invalidate older youtube content?

[rustic rain]

it didn't really invalidate

a lot of things are obsolete now

but overall ideas are still legit

mostly it's due to ComponentSystem -> SystemBase

switch

also

ECS is part of DOTS

[gusty comet]

That whole thing confused me to begin with so it's probably for the best that I'll need to relearn it

[rustic rain]

DOTS is a lot more then just ECS

and each big part can be used independently

[gusty comet]

Yeah I'm aware, it just used to be called ECS before the rest of the packages were developed

[rustic rain]

it's still called ECS

but it's not DOTS

it's ECS package

DOTS contains: Jobs, Burst, ECS

all are independent

[gusty comet]

I know... I was referring to the period of time where all of the new systems, including things which do not fall under the ECS label, were just called "Unity ECS" as an umbrella term.

[rustic rain]

ppl tend to call ECS DOTS, because ECS is using all power of DOTS as dependency

[misty wedge]

How fast is allocating a large amount of temp memory? Say I want to speed up my pathfinding by moving from a nativehashmap to a nativearray, I would need to request the entire grid (~5MB) at the start of the job. Also, how big is the temp memory pool?

[rotund token]

from memory there's only 256KB per block of Temp memory

at 5MB (per thread) you'll probably be using the fallback allocator

it won't be that bad though if you can do it per thread (entity chunk), not per entity

you can do something like this to re-use allocations per chunk that ends up on the same per thread (literally wrote this today)

private NativeList<BitArray8> conditionActiveBuffer;

public void Execute(ArchetypeChunk batchInChunk, int batchIndex)
{
    var conditionTypeBuffers = batchInChunk.GetBufferAccessor(this.ConditionType);
    if (!this.conditionActiveBuffer.IsCreated)
    {
        this.conditionActiveBuffer = new NativeList<BitArray8>(conditionTypeBuffers.Length, Allocator.Temp);
    }

    this.conditionActiveBuffer.ResizeUninitialized(this.conditionActiveBuffer.Length);```

[viral sonnet]

@tertle hehe, I knew you would run into it sooner or later. good solution to do a memcmp. i did it on an individual level because I had something like 5+ handles that could or could not change. your solution seems better

[misty wedge]

So running it with batchesPerChunk set to 1 should be fine?

Also I'm guessing the fallback allocator for temp is just tempjob?

[viral sonnet]

@rotund token oh, what are you using for the var updated (conditionActiveBuffer)? is this a temp array? just wondering how you keep the original and updated around without too much copying

[rotund token]

funny enough i just posted this like the comment before you replied

[rotund token]

^

per thread nativelist

[misty wedge]

Why does this work? Does each thread reuse its job struct for all chunks it handles?

[viral sonnet]

oh missed that! quite involved. i just use ref and write to the readonly ptr. then bump the version manually

in my case I know when changes happen. something you don't seem to have the luxury of

[rotund token]

yes, stripping out a lot of code it basically simplifies down to this

{
    // If we have no range to steal, exit the loop.
    if (!JobsUtility.GetWorkStealingRange(ref ranges, jobIndex, out beginBatchIndex, out endBatchIndex))
        break;

    for (int batchIndex = beginBatchIndex; batchIndex < endBatchIndex; ++batchIndex)
    {
        jobWrapper.JobData.Execute(batch, batchIndex);
    }
}```

[misty wedge]

Huh, interesting. That would also mean if you write any "state" data to the struct from execute (which you shouldn't anyways I guess), that would not "reset" on the next batch

[rotund token]

so each thread gets their own jobdata struct and just how structs work this means state is preserved yes

[misty wedge]

Cool trick, thanks

Another thing, if I request "too much" memory using the temp allocator and it falls back, I'm guessing I still don't need to dispose it manually

[rotund token]

nah you dont need to worry about that

[misty wedge]

Yeah that would be awfully error prone 😅

Is this also true for unsafe collections allocated with temp allocators inside a running job? (im guessing they are also automatically disposed for me)

[rustic rain]

hmm

rn I started to wonder whether systems check whether they should run every frame

or only after structural changes

Judging by source code - every frame

but this is so inneffective

[misty wedge]

iirc it's every frame

[rustic rain]

literally comparing all queries every frame

[rotund token]

well it compares queries you tell it to compare

[rustic rain]

meanwhile actual state only changes during structural changes

feels bad

[rotund token]

by default that's any query in the system

but you can restrict this with RequireForUpdate

[rustic rain]

yeah, but I mean

it can be done only once

on structural change

kind of like: make all systems dirty after structural change

[rotund token]

a) how does it know there's a structural change
b) checking a structural change requires a sync point

because you can implement b) if you want

simply to [UpdateAlways] so it doesnt check any query

[rustic rain]

Structural changes run on main thread anyway

meaning you can just add additional code to it

[rotund token]

then do a if(query.IsEmpty) return;

[rotund token]

oh sorry you meant structural changes

not change filters

ah most systems don't care about structural changes for updates

[rustic rain]

ahem

aren't most systems run based on queries?

[rotund token]

yes

[rustic rain]

well

then I think that checking whether queries exist every frame is a waste

quite large one

[rotund token]

oh you mean if the query is valid?

ah you realize unity already caches this right?

public bool IsEmptyIgnoreFilter => _QueryData->GetMatchingChunkCache().Length == 0;

[rustic rain]

it's more about

hmm

It's checking all queries

array of them

meanwhile everything can be simplified to just boolean

[misty wedge]

Have you profiled if it's actually slow?

[rustic rain]

and be 100 times faster

[rotund token]

it's literally just doing int == 0

sure it will be faster

but i doubt you'll even profile a difference

using chunk caches was one of the big improvements of entities 0.50

but even with that, did you notice a performance difference? because it is something like 20x faster than 0.17

if profiled independently

but when we profiled our entire project, we couldn't notice more than 0.5-1% in real world, well within any error margin

[misty wedge]

I imagine it would be pretty fast to check if any archetype chunk exists even without caching it

[rustic rain]

oh

I see

but you know

it does such checks for all queries

[rotund token]

it doesn't check each query

it checks until 1 returns true

[rustic rain]

welp, not sure I understand, but hopefully it's all cached

[rotund token]

it only requires 1 query to return true for a system to run

[misty wedge]

You only need a single query that contains an entity to run the system

I learned this the hard way when I made a really dumb mistake initially:
I create a LogColor singleton so that each network world (client and server) would print logs in different colors.
I then had a system that inherited from SystemBase that my systems would implement. The color was fetched with EntityQuery.GetSingleton.
This meant that since all of my systems inheriting this class had an entityquery that was always true (the log color singleton always existed), all of them ran each frame. So maybe don't do that...

[rustic rain]

Currently my systems run based of RequireSingletonForUpdate

mostly these are state entities

hmm, I do wonder though

if I add requirment for several singletons

they are will be combined requirment, right?

and probably same for queries

[misty wedge]

RequireSingletonForUpdate just gets a query for that type and then calls RequireForUpdate

[rustic rain]

yeah, but is it using OR or AND logic when you add it this way?

[misty wedge]

You can add multiple required queries to a system; all of them must match at least one entity for the system to run.

[rustic rain]

ah, yes
I didn't notice

[misty wedge]

More explicitly this is the code that actually checks the required entities

[misty wedge]

@rotund token sorry to bother you with this, but is this true? I'm not sure how to check since I have no idea where temp memory is freed, and after the job runs the unsafe container's IsCreated is obviously still true since it wouldn't know about it. Calling dispose on it after the job runs doesn't throw an exception though (due to the missing dispose sentinel)

[olive kite]

What about the components "Occluder" and "Occludee" for occlusion culling, are these replaced or just not needed anymore

[viral sonnet]

temp allocated memory in jobs doesn't need to be disposed

[misty wedge]

Same for unsafe containers?

Since safe containers throw an error it's easy to check for those

[viral sonnet]

hm, good question. I'd say no because it doesn't implement anything for the job to know

I'd stick to native containers for any temp allocations

they are optimized really well especially in burst jobs

[misty wedge]

What if I need to stick a container in another container inside a job

[viral sonnet]

I'd say rethink your design approach 🙂

[misty wedge]

But what if I wanna do it

[viral sonnet]

I'd still say the same thing ^^ I mean, it's possible for sure and I'd handle disposing just to be sure

[misty wedge]

TBH I think they are disposed, otherwise it would be impossible to dispose any unsafe containers that use temp memory, since UnsafeUtility.Free is a no-op for temp memory, meaning you could never dispose an unsafe container if it wasn't already being auto-disposed. I don't think the temp dispose mechanism has anything to do with jobs

[viral sonnet]

could be true, there's a mechanism in place that prevents of doing temp allocs outside of jobs

[misty wedge]

That's not actually true, you can alloc temp memory outside of a job iirc

You just can't pass it to a job

Not 100% sure though

This is all it says in the docs:
Allocator.Temp has the fastest allocation. Use it for allocations with a lifespan of one frame or fewer. However, you can’t use Temp to pass NativeContainer allocations to jobs.

[viral sonnet]

ah ok, so yeah, it seems then that temp is just freed every frame

and every thread has its own temp memory

[misty wedge]

Yeah, I was just wondering about unsafe collections, but I guess they are automatically freed to, since again, you can't actually dispose them manually since UnsafeUtilities.Free doesn't do anything for temp memory

[oak sapphire]

Hey all, I'm trying to get all the positions of all my FireballTag'ed Entities in a MonoBehaviour .
Currently I have this:

    EntityManager manager;
    private void Awake()
    {
        manager = World.DefaultGameObjectInjectionWorld.EntityManager;
    }
    void FixedUpdate()
    {
        EntityQuery eq = manager.CreateEntityQuery(typeof(FireballTag));
        manager.GetComponentData() <<< No idea what to do here, I want a list or an array of all the fireball positions, or should I use a Job for this and prepare a list in that?
    }

What I am trying to do is put the positions of all the fireballs in a pixel of a Texture2D that is fed to a VFX Graph so I can batch tens of thousands VFX spells in a single graph.
Any tips/links/suggestions would be appreciated. I want this piece of code to be as performant as possible, because I would like hundreds of thousands of fireballs.

[viral sonnet]

good writeup: https://www.jacksondunstan.com/articles/5446

[misty wedge]

You can create an entityquery and then call ToComponentDataArray on it to get all components of a specific type from it

[viral sonnet]

Allocator.Temp appears to be backed by a "bump allocator" that is cleared between every frame. Whether you call Dispose or not, you should only read and write to memory allocated by it during the same frame that you allocated it. Starting with the next frame, you'll either get an exception or data corruption.

[misty wedge]

I tried reading it a frame later and it still worked, but it's definitely not going to always work

the safe container threw an exception

[viral sonnet]

sometimes it lingers, guess when the temp memory space isn't exceeded or smth. hard to say what the logic is as we don't have the source code for it

or rather, the data lingers because it's not overwritten

[misty wedge]

I mean why wouldn't it linger, it's in the virtual address space of the program that allocated it. As long as nothing wrote over it it won't change

[misty wedge]

Is there a reason unity doesn't publish the source for stuff like the alloc, free, and scheduling stuff? Because it's part of the engine source code?

[oak sapphire]

Awesome, thanks! 👍
Are you aware of a faster way then this?

This is working now, but just checking if there is more to be gained.

[misty wedge]

Depends a lot on how you have your project set up. Ideally you aren't using any monobehaviours and are running all your code in jobs with burst enabled

[frosty siren]

how is it fast to read from blob reference? Can i use it like normal component data, or should i cache data from blob to read it faster?

[oak sapphire]

Well currently I'm doing it this way because as far as I know, jobs can't change values in my VFX Graph, and can't change the Texture2D I am using for positions. But maybe I am wrong about that. And the VFX Graph has so much overhead that giving all the Fireballs their own Graph is way less performant then doing it this way.

https://gdl.space/uzemebicik.cs
That is currently the only script in my scene.

[olive kite]

I do the same thing to instance my vfx (from positions stored in texture) 🙂 works well

[oak sapphire]

You do it in a MonoBehaviour as well? Or do you do it in a Job? Is that even possible?

[olive kite]

@oak sapphire If I can remember correctly, I do all the calculations in burst but had to set pixels in MonoBehaviour let me look

I am using a property binder to the vfx graph to assign the values

is that what you are doing, i know this is not what I was answering, pulling it up now

[oak sapphire]

Yeah, I have exactly that. All my entities behaviour via the SystemBase ScheduleParallel method. And the Mono changes the Texture2D and the bindings with the VFX graph.

My question was about what would be the fastest way to do this.

But if you are doing the exact same thing, it will probably be good 🙂

[misty wedge]

A blob reference is just a pointer to the memory where the data is stored

[olive kite]

Yeah i remember testing it and it was really efficient , but I'm definitely not an expert on micro optimizing in jobs and burst

[misty wedge]

I would profile it first and see if it's even a performance issue

[olive kite]

Yeah you should be good, looks like your texture isn't too large

[misty wedge]

If you do need more performance, there is a much faster way to write pixel data to textures

[olive kite]

please share

[misty wedge]

You can write to the memory backing the texture by getting a native array to it's color data using Texture2D.GetRawTextureData

e.g. Texture2D.GetRawTextureData<Color32>()

You can then write the data in a burstable job if you like

[olive kite]

oh duh, yeah that is much better than what I am doing

[misty wedge]

Just note that you aren't supposed to store the native array in any way, the intent is to use it immediately and not dispose it iirc

[oak sapphire]

Depends on what you think a performance issue is.
Currently I have DOTSNET networking working with 1 million entities all updating the rotation and position, while being server and client.
I have an animation system working with 78.804 animated characters at 70 FPS.
And now I want to make those characters shoot VFX entities, and would like to get the most performance possible here too.
Then I will combine them and make a small game, that's my plan.

[misty wedge]

Yeah I just mean it really depends on the project. imo it's a performance issue if you don't hit the required metrics, e.g. 60 FPS or whatever your target is on the minimum hardware

I just mean check if it's actually an issue before trying to optimize it

[olive kite]

yeah, of course, if trying to get product out in the most efficient/effective way, no point in fixing what isn't a problem. Sometimes the use case is "I wonder the absolute most efficient way to do x" which somehow makes it into my project pipeline when it shouldn't or isn't an issue. But still fun sometimes

any of you know what might be causing the componentdata to be unknown here? you all know why I might be getting "InvalidOperationException: The previously scheduled job Jobs:CreateRigidBodies reads from the UNKNOWN_OBJECT_TYPE CreateRigidBodies.JobData.PositionType. You are trying to schedule a new job CalculateGroundHeightSystem:Calc_Ground_Height_Job, which writes to the same UNKNOWN_OBJECT_TYPE (via Calc_Ground_Height_Job.JobData.__translationTypeHandle). To guarantee safety, you must include Jobs:CreateRigidBodies as a dependency of the newly scheduled job.

[misty wedge]

The discord recently started blocking large code blocks for some reason