archived-dots

[deft pilot]

@rotund token is there some conflict between the job system 0.70.0 and 2d animation package version 9.0.0?

[deft pilot]

welp doesn't look like i can use ECS 1.0.0 with 2D right now

tilemaps and afew things are broken in 2022.2

even if i fix the package bug in 2d animations

[rustic rain]

@robust scaffold did you get tilemaps working for you in 1.0?

[deft pilot]

tilemaps just didn't work with 2022.2

so couldn't even test 1.0.0

gonna just stick with 0.51 for now

[misty wedge]

The normal unity tilemaps?

[deft pilot]

oh wow ECS uses a totally different position system

anyway to convert from world position to whatever LocalToWorld is?

[rustic rain]

ltw IS world position

[deft pilot]

isn't it a float3?

or am i crazy

[rustic rain]

ltw is 4x4

which includes position, rotation and scale

[deft pilot]

not sure what that means guess i'll just go google unless you want to explain how to set position of an entity when i instantiate it

but i only have a vector3 unfortunately not sure how to convert that

[rustic rain]

Vector3 kek = new float3(0,0,0);

just like that

[deft pilot]

yeah googled it already

thanks anyway

[robust scaffold]

Hybrid tilemaps rendering worked with the assembly ref hack. Physics did not.

[rustic rain]

@deft pilot

[deft pilot]

would need physics as well unfortunately - thanks thou

i'm okay waiting for further support

[brave field]

Do I mean GetComponent and GetBuffer API is cached by default at 1.0?

[proud jackal]

SystemAPI.GetComponent, and SystemAPI.GetBuffer indeed is cached, all of systemapi is :3

[rustic rain]

What would be a way to make system start running with a certain types of components and conitnue running with another?

[brave field]

👍 Oh nice. Is there any performance difference if I'm not using using SystemAPI to GetComponent and GetBuffer at both ISystem and SystemBase?

[rustic rain]

you should just look at codegened system

under the hood it's all just as simple as storing queries/component type handles in fields

[brave field]

Sometimes Rider can't view the codegen properly. Any setting to fix it?

[rustic rain]

idk

I just look at it

by pressing ctrl and then clicking on class name

[deft pilot]

since GetSingleton always expects an entity to exist with the query you are going for

is there something that is similar to SingleOrDefault

or do i have to do something like

            var entities = this.newEnemyPathQuery.ToEntityArray(Allocator.Temp);
            if (entities.Length == 0)
            {
                return;
            }

or should i just Entities Foreach this?

cause i'm not sure if you can nest that

Entities.Foreach((in Component1 c1) =>
{
  Entities.Foreach((in Component2 c2) =>
  {
  }
}```

[misty wedge]

You can use HasSingleton or TryGetSingleton

Also Entities.ForEach doesn't support nesting, and will be deprecated in the future.

[deft pilot]

yeah i didn't think it did

and i have read that its dead/dying in 1.0.0

[misty wedge]

Yes, it's been replaced with jobs and the query API

[deft pilot]

are you allowed to TryGetSingleton<DynamicBuffer<T>>()?

cause the standard one has a fit when you try to do that

[misty wedge]

I have a TryGetSingletonBuffer method

Not sure if that's in 0.51

[deft pilot]

nope

all good

[robust scaffold]

Plz, SystemAPI for GetComponentHandle<>().

[misty wedge]

Why not just fetch it from the state? But ideally it would be in the api. I guess it could be a pain porting the old this.GetComponentHandle from SystemBase

[robust scaffold]

Unity now throws a warning for handle creation inside OnUpdate()

[proud jackal]

There's a difference between the old this.GetComponentHandle and a potential SystemAPI version in that, the sysapi version would also cache and .update for you

[misty wedge]

Yeah, so just call Update on them?

[robust scaffold]

Yea. That would be glorious

[robust scaffold]

Yep. Except now you need to write the whole process for every handle manually. Pain.

[misty wedge]

Ah, you want to use SystemAPI.GetComponentHandle in the update method and have it codegen for you to call update there

[robust scaffold]

Yep

[misty wedge]

Yep, I agree, that would be cool 👍

[sage kestrel]

so i am following this tutorial https://www.youtube.com/watch?v=1bO1FdEThnU&t=389s ,and i am wondering,is it possible to make it so that the origin of the grid is changeable?,i currently want to generate a small grid around my player which he can walk around,any idea how?

i can also send the finished script i made if anybody needs it

[misty wedge]

Are ICleanupSharedComponentData broken somehow? It seems they can't be added by a Baker

the normal ISharedComponent appears in the subscene, but an ICleanupSharedComponentData does not

[drowsy pagoda]

Using 2022.2.0, Entities 1.0. I noticed that during play mode, if I move ECS objects in the scene view, when I exit play mode, their transform state persists. Is there a flag for disabling this feature? Or is this a bug?

[robust scaffold]

Cleanup components can not be added during baking. Only runtime

[misty wedge]

[robust scaffold]

New feature. It's something to do with the yellow dot circle thing

[misty wedge]

I feel like that should throw some kind of error

[drowsy pagoda]

Yeah, which I like that dot thing. It's awesome. But I tried moving it while putting the dot in authoring, mixed, and runtime mode. Result is the same.

[robust scaffold]

You'd think. Except systemstate was the same situation and yet, no error.

[drowsy pagoda]

Although I tried it with the dot in the Inspector. I didn't see one in the Scene window.

[robust scaffold]

No clue but it's a new setting with dots 1.0. It was mentioned in the DOTS Workflow Preview presentation and it was some sort of setting.

[drowsy pagoda]

🤔 hmm. alright I'll dig around.

[rustic rain]

they never were able to get past conversion in subscene

state components were also removed during conversion

[misty wedge]

That doesn't mean that can't throw an error

A simple check for the correct interface would suffice

[rustic rain]

why should they?

[misty wedge]

Because it's not clear?

[rustic rain]

but they literally use them too

for conversion

kek

[misty wedge]

I can do AddComponent with a cleanup component and it won't appear in the subscene

[rustic rain]

it is added to Entity though

it's just up to you how you will use it during bkaing

[brave field]

I still think Baker should support cleanup component.

[misty wedge]

^

Why doesn't it?

[rustic rain]

why should they?

they are meant for runtime only

[misty wedge]

Yes, I want them on the runtime object once the scene is loaded

[rustic rain]

then you need system which uses them

and adds them to entity

[misty wedge]

Yes, and I'd like to skip that step with baking

[rustic rain]

think of them as a flag that smth needs to be disposed, nothing else

[rustic rain]

then just use normal component?

[misty wedge]

I need it to force the destroyed entity into a specific chunk to read shared data

[brave field]

If I understand correctly, it's just like regular component that will need to move to another new chunk when adding cleanup component which is slow

[rustic rain]

cleanup components are meant for their name

to clean things up

[misty wedge]

They can just as well store data

But that explains why netcode stores all the ghost type data twice since they can't bake the cleanup component

And the data will be missing once the entity is destroyed

That's what baking would be useful for

[brave field]

And purposely need to write a system just to add cleanup component also really annoying

[misty wedge]

^

[sage kestrel]

Any help?

[brave field]

👀 Why I need to do tat if I can bake it?

[rustic rain]

think you guys using it for a wrong purpose 😅

[misty wedge]

The purpose is up to the developer

And like I mentioned, sometimes it forces you to store data twice, which is just bad

You'd need to replace the baked component with a cleanup component at runtime, which seems like an even worse workaround

[robust scaffold]

Look, we probably can but this is something ya need to learn yourself. Or wait until DOTS is out of experimental and with a lot of tutorials to learn with.

[robust scaffold]

It is pain.

[misty wedge]

Yep

[brave field]

👀 Not really. If I'm not wrong, when u destory entity, the entity will only left cleanup component for u to do final cleanup. But it doesn't stop me to add cleanup component at Baker.

[misty wedge]

Just put a flag on AddComponent in the baker if you want to keep the component or not

[rustic rain]

well, that does stop you cause they are only runtime

😅

[sage kestrel]

well the thing is,if i was using a custom grid made system,i would have probably had it done,but the thing is this is made a with NativeArray so i have no idea how to make a custom origin point for a native array

[rustic rain]

I use cleanup comp to store pointer to allocated instance of sound event in FMOD engine

so when Entity gets removed, system also disposes that sound event

[misty wedge]

Yes, that's what they are for

But sometimes you require some data that was baked to cleanup the entity

[rustic rain]

then you probably should create a destruction pipeline

[misty wedge]

Well yes, but that's more work than just being able to bake the component

[brave field]

For me I just want to avoid entity to keep moving to new chunk at runtime since currently u have been forced to add cleanup component at runtime only

[drowsy pagoda]

In 2022, did they remove Prefab editor context modes?

[rustic rain]

if it's just a tag

you can simply add component to whole query

which will from what I believe - change archetype to all existing chunks

[misty wedge]

You'd usually need some information to identify whatever you want to clean up though...

[rustic rain]

but don't quote me on that

just the fact that you need to dispose of smth that is not even created yet says you are doing smth wrong 😅

[brave field]

I can't use tat. I need to put unique cleanup component data to each entity.

[misty wedge]

Why? I want the instantiated entity from a prefab to just have a component to cleanup. The prefab doesn't do anything at runtime

[rustic rain]

well you can still do that

[misty wedge]

Adding it manually at runtime is just an unneeded extra step that requires boilerplate

[rustic rain]

just add component to query

and then query over all of them

to do your thing

[misty wedge]

You could do that in 0.17. The point is that we are requesting / asking why certain things aren't yet possible

Also I hate the fact that you sometimes need to double store data (or remove the originally baked component at runtime after instantiating the prefab entity)

[sage kestrel]

well can you give me a hint or something?

[robust scaffold]

If you're having issues with computing relative index of neighboring nodes, the CalculateIndex would instead return (x - originX) + (y - originY) * gridWidth

[rustic rain]

oof

anyone tweaked around graphics?

looks like RenderMesh is legacy

meanwhile everything is RenderMeshArray

• settings with indices

I'm guessing

RenderMesh is gone for good?

🤔

[dense storm]

Also looking for the RenderMesh replacement

[rustic rain]

RenderMeshArray

it's a bit more complex now

[dense storm]

I see :0

[rustic rain]

hmm

Have NativeContainer inside NativeContainer restriction has been lifted since they are unmanaged now?

[misty wedge]

I don't think so, there was something about that in the docs

You still need to use an UnsafeContainer wrapping the NativeContainer iirc

[rustic rain]

🤔

[misty wedge]

While you should generally prefer using the Native- collections over their Unsafe- equivalents, Native- collections cannot contain other Native- collections (owing to the implementation of their safety checks). So if, say, you want a list of lists, you can have a NativeList<UnsafeList<T>> or an UnsafeList<UnsafeList<T>>, but you cannot have a NativeList<NativeList<T>>.

[rustic rain]

but safety checks were made unmanaged now

[misty wedge]

This is from the newest documentation, 2.1.0

[austere pebble]

with which thing GenerateAuthoringComponent was replaced in 1.0.0

[misty wedge]

Try it out and see I guess 🤷

[misty wedge]

Nothing, you need to do it manually

[robust scaffold]

Huh, this works:

[misty wedge]

Or wait until they add a replacement

[austere pebble]

any example? its that bakery thing

[misty wedge]

Yes, it's the bakery

[robust scaffold]

https://github.com/Unity-Technologies/EntityComponentSystemSamples/blob/master/DOTS_Guide/cheatsheet/components_systems.md

[rustic rain]

huh

it worked

no errors

            var kek = new NativeArray<NativeArray<int>>(5, Allocator.Temp);

            for (int i = 0; i < kek.Length; i++)
            {
                kek[i] = new NativeArray<int>(5, Allocator.Temp);
            }

[misty wedge]

I think you replied to the wrong message 😅

I'm just going off the docs mind you, I haven't tried it

[robust scaffold]

NA is special. Try NL<NL<>>

[misty wedge]

Yes, NativeArray isn't even part of collections

[rustic rain]

but it works

kek

[misty wedge]

🤷

Maybe the docs are wrong, or the safety system breaks in some unforseen way

[rustic rain]

            var kek = new NativeArray<NativeArray<int>>(5, Allocator.Temp);

            for (int i = 0; i < kek.Length; i++)
            {
                kek[i] = new NativeArray<int>(5, Allocator.Temp);
                for (int j = 0; j < kek[i].Length; j++)
                {
                    var lul = kek[i];

                    lul[j] = 5;
                }
            }

for sake of more testing tried this

looks like

all guchi now

[robust scaffold]

var test = new NativeList<NativeList<int>>(WorldUpdateAllocator); Huh, no errors

[rustic rain]

yeah

[misty wedge]

Maybe the docs haven't been updated

[rustic rain]

probably

[misty wedge]

The netcode documentation is wrong like 50% of the time

[rustic rain]

oh boy oh boy

What crazy mind can think of that

[misty wedge]

I mean you could do all that with unsafe collections before, but I guess knowing unity will let you know if you forget to get rid of it is nice

[rustic rain]

unsafe sucked

so much pain handling ref

[misty wedge]

I never had any issues with it 🤷 just replacing the unsafe stuff with the normal collections now is nice though

[coarse turtle]

with Entities 1.0 - I'm taking a look at the ECBSingletons and i'm curious as to how the ECBSingleton tracks the dependencies for write operations. With EntityCommandBufferSystem - the AddJobHandleForProducer still exists and the ECBSingleton is just an Entity with a component pointing to the pendingBuffers.
Am I missing something with dependency tracking with ECBs in ISystem? 🤔

[rustic rain]

wdym?

you are acessing singletons only on main thread

so that's pretty much the only time you create ECB

[coarse turtle]

sure but i imagine you want to wait for other threads to finish adding the cmds the CommandBuffer before the EntityCommandBufferSystem performs its playback

[rustic rain]

Don't you add job handle?

[coarse turtle]

with ecbsingletons - i dont see where you could

[rustic rain]

🤔

[coarse turtle]

public unsafe struct Singleton : IComponentData, IECBSingleton
{
  internal UnsafeList<EntityCommandBuffer>* pendingBuffers;
  internal Allocator allocator;
  public EntityCommandBuffer CreateCommandBuffer(WorldUnmanaged world)
  {
    return EntityCommandBufferSystem.CreateCommandBuffer(ref *pendingBuffers, allocator, world);
  }
 
  public void SetPendingBufferList(ref UnsafeList<EntityCommandBuffer> buffers)
  {
    pendingBuffers = (UnsafeList<EntityCommandBuffer>*)UnsafeUtility.AddressOf(ref buffers);
  }
 
  public void SetAllocator(Allocator allocatorIn)
  {
    allocator = allocatorIn;
  }
}```

[rustic rain]

does ECB itself contain dependency?

[coarse turtle]

havent seen it

[misty wedge]

The ECB system has a write handle on the singleton, meaning it forces all reading systems to complete their dependency

You don't need AddJobHandleForProducer anymore

[coarse turtle]

oh - where do they store this write handle? Or is this implicit via registering the singleton?

[misty wedge]

It's when you get the singleton with the required permissions

[rustic rain]

hold up

meaning when you get ECB singleton query

it actually forces all systems with that query cached to be complete?

[misty wedge]

No, since all of your systems should only have a read permission on it

[coarse turtle]

oh right - that makes a lot of sense now - cool thnx

[misty wedge]

I assume it does force the ECB system to complete though, if it hasn't already

(this is just me guesstimating btw, I haven't looked into it extensively)

[coarse turtle]

ah lol - no worries

im trying to create a different kind of command buffer

but i also want it to be isystem friendly

[misty wedge]

You can use the normal ones in an ISystem

[coarse turtle]

oh by different - it's separate from actual entities

it's for my audio system internally

[misty wedge]

I'd take a look at how they manage the buffers internally

I think each ECB just stores a pointer to the data, and the ECB system has an unsafe list of all of those structs holding the pointers

the Singleton.CreateCommandBuffer method just registers the buffer to the system

[coarse turtle]

Yea my previous with 0.51 is done similarly

so I think I can likely just create my own component and alias a pointer to the component

[misty wedge]

What does the audio system do out of curiosity? You write to the buffer to create audio effects?

[coarse turtle]

yea pretty much - if you want to do some audio processing you just write a bitflag

i just do the playback in c++

so i process all the bitflags and then load them through miniaudio

[robust scaffold]

4 hours of work and I finally kinda recreated CGOs.

[misty wedge]

too complicated for me 😅 but sounds interesting

[coarse turtle]

lol - eventually i'll release it, but not anytime soon - i haven't really tested it in production yet 😅

[viral sonnet]

any ideas how i can get a ptr from this? ref readonly var refComp = ref lookup.GetRefRO(fixedEntity).ValueRO; due to the ref readonly no UnsafeUtility method works

just ptr casting also doesn't work :/

[coarse turtle]

I guess AddressOf is causing a potential write with ref readonly var

[robust scaffold]

What are you trying to do with it?

[viral sonnet]

it's a byte array that has different data types sequentially, so i need the byte ptr and lots of casting 🙂

i see that would work ```ref readonly var refComp = ref lookup.GetRefRO(fixedEntity).ValueRO;

        fixed (byte* ptr2 = &refComp)
        {
            
        }``` i just hate fixed -.-

and i'm trying if i can ditch my UnsafeCDFE in favor of ComponentLookup

but with the special handholding. i guess not 🤷

[coarse turtle]

int i = 0;
ref readonly int t = ref i;
                           
fixed(int* ptr = &t) {
    
}

This looks kind of cursed (oh just saw you posted the same thing up there 😅 )

[viral sonnet]

it does 😄

[robust scaffold]

Yea, the fixed looks like the only option.

[viral sonnet]

    {
        readonly unsafe byte* _Data;
#if ENABLE_UNITY_COLLECTIONS_CHECKS
        readonly AtomicSafetyHandle _Safety;
#endif``` why is _Data not internal?

[robust scaffold]

because people like us are touching parts of Entities that unity doesnt want touched

[viral sonnet]

i can touch it regardless. the only question is, do they make it difficult or not 😄

[robust scaffold]

Exposed copies are pain to make though

[viral sonnet]

yeah but worth it. i'd hate to have all those dumb fixed statements around my code

[robust scaffold]

They're not that bad...

[viral sonnet]

i'm working in unmanaged code. fixing is only for the c# compiler

[spice vale]

try UnsafeUtilityExtensions.AddressOf()

[viral sonnet]

i don't like the fixed context being necessary. just take this for example ```byte* stats = null;

            if (spellData.Effect_Main.Value.snapShotStats)
                stats = statsAccessor[i].Value;
            else
            {
                ref readonly var tmp = ref StatBuffer_Final_Lookup.GetRefRO(spellOwner.source).ValueRO;
                stats = tmp.Value;
            }```

[robust scaffold]

Is the readonly required?

[viral sonnet]

if it was handled like we can handle using statements i'd be okay

yeah, ref doesn't work and otherwise it gets copied

[robust scaffold]

Oh, eww

[viral sonnet]

if i use GetRefRW i make a version bump. can't do that either

unity hasn't thought this through. very plain and simple

[robust scaffold]

If you use GetRefRW(false), it doesnt look like it'll bump version

[viral sonnet]

HOLY

i didn't even check! thought this was a given

[robust scaffold]

Yea, it's odd.

I have no clue why there's a RO version of RW while there's also a RefRO.

[viral sonnet]

haha, noice! 😄 then i just use GetRefRW with readOnly ... that's some really shitty code

no, fk that

[robust scaffold]

Maybe RefRO returns copies whereas RefRW returns pointer and the RO version of RefRW doesnt bump version.

[viral sonnet]

thanks though, i have a solution.

[spice vale]

@viral sonnet just do: UnsafeUtilityExtensions.AddressOf(in refComp)

It is a special "overload" that does not take ref, but in instead

[viral sonnet]

sorry that doesn't work

[spice vale]

UnsafeUtilityExtensions

not UnsafeUtility

[viral sonnet]

oh nice 😄 that's working

thanks man!

[spice vale]

I guess "ref readonly" is sort of same as "in" but it would be confusing to have an "in" variable as output

[robust scaffold]

Is the pointer valid?

[viral sonnet]

i'll test now. also need to see what burst does with it

[spice vale]

probably why they went with ref readonly

[viral sonnet]

i don't understand the address access much but it turned into a fine pointer

does this get some special treatment? i never found the answer why we can't write the same thing in code internal static class ILSupport { [MethodImpl(MethodImplOptions.AggressiveInlining)] public static unsafe void* AddressOf<T>(in T thing) where T : struct => (void*) ref thing;

[rotund token]

It's not valid c#

Out of interest if you tab back into unity for it compile?

i.e. It compile fine just your ide didn't like it

[viral sonnet]

unity reports the same error

[rotund token]

Ok that's interesting to me

[viral sonnet]

there's no safety with a ref anymore. i can say readOnly and still modify it ^^

but that's also not unitys job imo

i would also be okay with some unsafe/advanced methods behind a compiler directive. sometimes they used it.

when it comes to ptrs and reference values, safety gets muddy

[rotund token]

That's what is so great about these ref readonly

[viral sonnet]

true, makes for good explicit code

i have my doubts though with this line 😄 var ptr = (byte*) UnsafeUtilityExtensions.AddressOf(in lookup.GetRefRO(fixedEntity).ValueRO);

just return a void* ... jeeeesus 😄

i can get a pointer from any entities data i can think of. yet here we are with ComponentLookups

[misty wedge]

Does it make more sense to stick singleton entities from baking all on one entity or to split them up?

[rotund token]

most of my settings files are all on 1 entity, realistically though unless you have crazy amounts it won't really matter - 16KB per unique archetype

[gusty comet]

What's the difference between TranslateWorld and TranslateLocal?

[rotund token]

new translation system in entities?
haven't used it yet but I assume it's just local vs world space

so i would think transform.localposition vs transform.position

[gusty comet]

I see. Translating the localPosition would just move, say, the child GO of a parent GO, while translating the worldPosition would move the child AND parent GO? Is this reasonable behavior?

[rotund token]

no, i assume both would only move the child (again haven't looked at method)

say you have an entity with position 10,0,0
and a child with a LocalPosition of 1,0,0, it's world position is 11,0,0

if you set local position to 2,0,0, it's world position is now 12,0,0

if you set its world position to 2,0,0, it's local position will now be -8,0,0

[misty wedge]

Why can't I read and write from something on a component lookup if I use Schedule for an IJE?

[rotund token]

you can?

what issue are you having

[misty wedge]

It tells me it's declared as write only

InvalidOperationException: The Unity.Collections.NativeParallelHashMap`2[Unity.Mathematics.int2,System.Int32] has been declared as [WriteOnly] in the job, but you are reading from it.

[rotund token]

that usually means you've embedded it within another container

[misty wedge]

It's inside a component

[rotund token]

oh yeah that's another container (with the container being the whole chunk)

safety doesn't work like that

[misty wedge]

So I guess I need to make it unsafe then

[gusty comet]

Gotcha, you set one of local/world, the other one gets adjusted to compensate

[misty wedge]

Can I selectively ignore the safety restriction for just this job?

e.g. [NativeDisableParallelForRestriction]

[rotund token]

you can disable safety on the entire job - but that's about it

[misty wedge]

Nope, I guess not

How do I do that?

Also this should be safe right? Since only a single thread will be writing to this component type if I don't use ScheduleParallel

[wraith hinge]

no, because the inner container is not patched, because the job system doesn't know about the inner container. we know about it and we're exploring possible fixes

[rotund token]

i disagree with doing it, but [BurstCompile(DisableSafety=true)]

something like that

[wraith hinge]

it isn't safe because the main thread could also be writing to it

[misty wedge]

Yeah ideally I'd like to not turn off safety but I'm not sure how to go about it

[rotund token]

i'd be interesting to see if you can find a fix for it

because it would imply you could also do nested containers then

unless it was something very specific to components with like typemanager metadata

[misty wedge]

I basically have a lot of entities that "belong" to a single entity, and this entity contains some collections that need to be written to by the entities it "contains". Previously I did this by iterating over the list of containing entities (there aren't that many), but I guessed it would be better this way due to the scheduling overhead

[rotund token]

is this a singleton entity?

[misty wedge]

No, there could be 5 containing entities, with 50 entities per "container"

I still have no idea how big the scheduling overhead is so maybe the way I'm doing it atm is fine

[rotund token]

less now!

wish I could benchmark work project on 2022 somehow to actually compare

[misty wedge]

The issue is also that I need to schedule them in a chain since otherwise the safety system will also tell me there are multiple writers. The entities in the container have a sharedcomponent that puts them on the same chunk, so that could be multithreaded

I'm guessing this is also only possible by disabling safety

e.g. if I have a shared component that contains an entity value that contains additional information, I'd need a component lookup, but that isn't "safe" in the eyes of the safety system

I guess it isn't safe in actuality either with schedule parallel

[misty wedge]

How big of an issue was it previously?

[rotund token]

179/frame - Median 1.23ms, Mean 1.24ms, Min 1.02ms, Max 1.41ms

m_DependencyManager->AddDependency
171/frame - Median 0.20ms, Mean 0.24ms, Min 0.17ms, Max 0.48ms

JobHandle.ScheduleBatchedJobs
171/frame - Median 0.62ms, Mean 0.61ms, Min 0.47ms, Max 0.69ms```
were my results from work

dependency was a bigger problem for us

i believe both of these should be significantly improved in 2022 but as I have suggested against us upgrading at work i can't compare

[misty wedge]

I guess I'll just leave it like it is for now and optimize it if it becomes an issue

[rotund token]

  'Assets/Resources/SceneDependencyCache/6f5a7786b470eced84ed53b5a3e78fe1.sceneWithBuildSettings' (current owner)```

somehow managed to generate 2 subscene caches with same guid - that's a new one

[misty wedge]

What happens if you need to pass a jobhandle to an isystem? Save it on a component?

[rotund token]

you sit back and ask yourself

why the hell are you doing this

[misty wedge]

I don't like it either

[rotund token]

so don't

there's pretty much no reason to do it now

[misty wedge]

I need to call multiple IJobParallelFor that depend on the contents of a native collections

[rotund token]

yeah pass the collections via singleton entities

like commandbuffersystems do

[misty wedge]

They aren't fixed

[rotund token]

or physics

[misty wedge]

There could be 5 or 10

[rotund token]

EndSimulationCommandBufferSystem holds X number of commandbuffers

[misty wedge]

So lets say I have 5 entities with components on them that contains native collections

I iterate over those entities, and then want to schedule an IJPF based on their length

How do I get the length? Since I can't fetch the component anymore

[rotund token]

so you need to sync point to schedule this job?

[misty wedge]

I need to call complete on a single jobhandle (the system writing the collections)

[rotund token]

booo

[misty wedge]

I know right

I hate it

[rotund token]

don't do it

[misty wedge]

Then how do I schedule the jobs?

[rustic rain]

With job handle?

[misty wedge]

How does that answer the question

[rustic rain]

I don't get why you can't use jobhandle as dependency

[misty wedge]

What length am I passing to the IJPF

[rotund token]

Schedule(List.Length)

how does jobhanlde help that

[rustic rain]

Hm

[rotund token]

that said you use IJobParallelForDeferred

[misty wedge]

Yeah, but then I'd need to cache either the list or a pointer to the length field

And then I still kind of have inter system dependencies, which I don't want

[rotund token]

to me everything you're doing seems wrong here

[misty wedge]

I don't mind changing it, but I don't know how

[rustic rain]

Why not do it in a single system in the first place

[rotund token]

^ was my first thought

but without knowing why you're doing something it's hard to suggest an alternative

[viral sonnet]

the singleton method that was suggested before would create a sync point. hardly a good solution

[rotund token]

?

[misty wedge]

Basically I have "world" entities, that can be a place where an actor can be, e.g. a planet, or underground, or somewhere else.

In the case of a continent, a job collects all the chunks that any actor is near to, and then a second system runs on all collected chunks per world that have been added, and creates the objects on the chunks.

The actual world entity count is low, but since a lot of chunks could be created per frame, I use IJPF to distribute that across threads

[viral sonnet]

oh unless you don't use GetSingletonRW in the update. well nvm me then 🤣

[rotund token]

In the case of a continent, a job collects all the chunks that any actor is near to, and then a second system runs on all collected chunks per world that have been added, and creates the objects on the chunks.
what's the reasoning for multiple systems when it seems like 1 responsibility?

[rustic rain]

Sounds like all should be single system

[misty wedge]

I thought about putting it in 1 system, but in theory some other systems may need to fetch the chunks that have been added this frame. For example, to generate a navmesh

I guess I could also put that in the same system, but then eventually that system will be huge

[rotund token]

ok so that's fine keep them separate

but why can't you store all chunks on a single component?

[misty wedge]

How would that work around the dependency issue?

This is what the component looks like:

[rotund token]

var list = SytemAPI.GetSingleton<YourComponent>().YourList;

[rotund token]

there are more than 1 of these right?

[misty wedge]

Yes

[rotund token]

private readonly UnsafeList<ActiveChunkTrackingCleanupComponent>* drawers;

basically just copy the SimulationCommandBufferSystemPattern

well hmm

[misty wedge]

I still don't understand how I then schedule my IJPFD

[rotund token]

at what point do you know how many activechunktrackingcleanupcomponents there are

and what component determines the length of your ijpfd

[misty wedge]

AddedChunksThisFramePerWorld determines the length of the IJPFD per world entity

[rotund token]

(also activechunktrackingcleanupcomponents is a mouthful)

[misty wedge]

So fi there are 5 worlds, 5 jobs needs to be scheduled based on their length

[misty wedge]

It very rarely changes, so I can check on the main thread as long as I don't have any jobs scheduled writing to this component

[rustic rain]

Sounds expensive

[misty wedge]

That's why I'd like to change it

[rotund token]

i'm just wondering if it'd be populated before you get to it

but you can always simply do the 'maths' approach

if you don't want to rewrite huge chunks of it

[misty wedge]

Another alternative would be to make each created chunk an entity, but that would probably not be a good idea

[rotund token]

just schedule an IJobFor with length of JobsUtility.JobWorkerCount

and pass all the data in

and each thread can work out itself what it should work on

[misty wedge]

I'd still need to cache the lists in that case though right? Since otherwise I'd need to call complete to get them

[rotund token]

im confused why you need to call complete to get them?

[misty wedge]

They're in a component

That is being written to

[rotund token]

yeah

just get the component within the job

[misty wedge]

My bad, I was still thinking about the scheduling part (for the IJPFD)

[rotund token]

it's not ideal, but it works without redoing everything

[misty wedge]

I guess that could work

I'm still not happy with the way it works though. Ideally I'd like to just schedule a single job that handles all "worlds"

[rotund token]

or just don't change anything

and just stick with SystemBase

and share them between systems

[misty wedge]

Can you not store a JobHandle in a component?

[rotund token]

you probably could

[misty wedge]

It's just a struct after all

[rotund token]

never heard it being done before though

[misty wedge]

I'll mess around a bit with it. I don't have a massive amount of entities so I don't really need to make this super optimized

Thanks for all the tips and help

I love when something works with burst off but doesn't with burst on 😬

[rotund token]

hmm what's the issue?

i recently noticed if you turn burst during play mode after making a change it uses the old version before your change

[misty wedge]

I stopped and started again

[rotund token]

and doesn't fix itself even leaving/restarting play mode

[misty wedge]

I don't think I've ever change the burst toggle during gameplay

Would clearing the burst folder fix that?

[rotund token]

how do you never forget to turn burst on!

restarting unity fixes it

[misty wedge]

I mean I never toggle it while the game is running 😅

[rotund token]

(and a new file change probably would as well)

[rotund token]

i often enter game and realize oops burst is off

let's turn it back on because i'm lagging at 5fps

[misty wedge]

regarding this, this is the behavior with burst off:

This is basically the system I was describing earlier

The squares are the chunks, and the numbers in the chunk is how many entities are "activating" that chunk

If I turn burst on it breaks, without any error messages

[rotund token]

are you relying on floats or something

for precision

[misty wedge]

the chunk positions are stored as int2

and then multiplied with chunksize (16) to get the world position

[rotund token]

hmm not sure, the only thing i know that really changes is float precision

[misty wedge]

This is with burst on:

so time to revert commits 😅

[rotund token]

the other issue is just timing

you're doing something unsafe but mono is too slow to break

and turning burst on breaks it

[misty wedge]

I seem to have broken something when moving the collections to the component, the commit before that one works 🤷

[misty wedge]

The [WithAll] etc components go on top of the struct right? I couldn't find any documentation about it

[rotund token]

yes

[misty wedge]

What do I do if burst throws an exception, but turning burst off makes the exception not get thrown? 😅

[rotund token]

attach a native debugger while burst is on

[misty wedge]

The exception is mysteriously thrown here

[rotund token]

how did you determine this

it would imply you've done something quite unsafe

and are modifying entity world while scheduling

i.e. safety missing somewhere

[misty wedge]

System.NullReferenceException: Object reference not set to an instance of an object This Exception was thrown from a job compiled with Burst, which has limited exception support. 0x00007ffacea9dbe7 (7856fbb6b4dbffaaada80d2c18c6557) [IJobChunk.cs:367]

Maybe the exception is just wrong because it's bursted? 🤷

[misty wedge]

I'm not doing anything crazy

This job broke it, seems pretty simple:

[rotund token]

not even burst compiled?

[misty wedge]

Nope, I keep forgetting 😅

[rotund token]

so how is it creating a burst compile issue

[misty wedge]

This is old Entities.ForEach code I'm converting

Which defaulted to burst

[misty wedge]

I have no idea. Maybe something broke somewhere

I already restarted unity, that didn't help

[rotund token]

are you scheduling this from a ISystem

that is bursted?

[misty wedge]

SystemBase

[rotund token]

so how do you know it's this job

[misty wedge]

I just replaced the old entities foreach with a call to this job

The execute code is identical

I'll attach a debugger and see what's happening

[rotund token]

how are you creating this job in your system

[misty wedge]

Is it normal that the generated file in the Temp folder doesn't match an exception?

The generated id is different

[rotund token]

i'd clear your cache

and restart

also this is magic

[misty wedge]

What do you mean?

[rotund token]

you can break point in burst code in 2022+

Setting a breakpoint in an attached managed debugger (Rider/VS Unity Debugger...) on a method that is burst compiled, will switch off the burst code path for that method, allowing it to be debugged as normal.

[misty wedge]

Oh that's nice

It just takes my computer 900 years to switch from release to debug

[rotund token]

i find it odd you are not permanently on debug

[misty wedge]

Yeah I swapped earlier

[rotund token]

i'm still super torn how to handle menu in a multiplayer (netcode) game

i've been having it so a client always exists and the menu runs in that world

but it just seems a tad wrong to me

[misty wedge]

Why not use the default world?

[rotund token]

because then i have to duplicate random number of systems

imagine i want my menu to play a mini simulation of the game in the background for example

[misty wedge]

Ah, for UI stuff?

[rotund token]

but yeah also UI stuff

to do nice transitions between menu game

but also maybe i want to re-use options from menu in game as well

just use the same option panel

this is why i've been using client world

[misty wedge]

I think it's fine to run it in the client world then. Since all of the relevant systems won't do anything until you connect

[rotund token]

and it works even better in 1.0

because it actually ticks properly

but it's just like, gut feeling says it's off

[misty wedge]

I'm just using the default world currently, but I also don't plan to do anything fancy there

[rotund token]

1.0 kind of got rid of default world in netcode

unless you create it yourself

[misty wedge]

Yeah, I originally used it for reconnects, since I wanted to completely reset the client world. Back then you couldn't migrate the network driver

Maybe I'll also switch to using the client world eventually

[rotund token]

oh i still completely delete client world on disconnect

and creating it again basically just loads the menu by default

as that's the default state

only thing that exists outside of it then is basically a load screen to hide it quickly

[misty wedge]

I used the default world to store a component about the gameplay endpoint for the reconnect, which would then recreate the client and connect it

But it's probably overkill

[rotund token]

hmm whats your use case of reconnecting after destroying client?

is it for a game with multiple servers?

and you move between them

[misty wedge]

I think I originally had it for loading or something, but I need to rewrite my save state system so I haven't used it in a long time

The client would disconnect, the server world would be destroyed, loaded from a save file, and the client would attempt to reconnect

I also used the default world for storing the data on which save file to recreate the server world with

Swapped to debug and unity crashed on run

I switched to debug and now I don't get the exception anymore, even without changing any code

[rotund token]

fixed

[misty wedge]

maybe it was burst cache vodoo going on

I'm probably getting some new hardware soon, I really hope my iteration time goes down

[viral sonnet]

they could rename isReadOnly from GetRefRW to bumpVersion.

[rotund token]

(would it not be donotbumpversion or do you mean also invert it)

but i agree otherwise it's a bit confusing

it's a nice exposure point though

[viral sonnet]

yeah, would be inverted then. it's a nice side effect of GetRefRW

and version bumping (or not) isn't exposed a lot

makes one think about it 😄

i have rolled my own UnsafeComponentLookup (again)

[drowsy pagoda]

Woah. RenderMesh is now gone. Can someone give me some easy to understand info on how to control rendering of an entity? Say I want to disable rendering and enable rendering (instead of destroying/instantiating and entity)

[rotund token]

but can you use SystemAPI?!

if not 🙅 down with boilerplate

finally getting close to systems with nothing but jobs!

[viral sonnet]

well, it's great if your only concern is less lines of code. this exists because _Data in RefRO/W is private. nothing i can do about it

[rotund token]

i'm personally going through and removing huge amounts of my pointers

it's great

can finally have a library i'm not worried about other people breaking

[viral sonnet]

how do you expect them in breaking it? it's not that internal library jobs and data are that exposed

[rotund token]

it's not about public people

it's about colleagues

someone who has a task to fix a bug or implement a feature

someone who might not have intricate understanding of a system and what they can/can't do

i need to write code in a way that some intern in 5 years won't touch and break unexpectantly

[viral sonnet]

are you aware that IJE codegen creates ptr access instead of chunk.GetNativeArray?

[rotund token]

vaguely

i still only have 1 IJE

so i've barely touched it yet

i still have EFE to convert to it (they still work so haven't bothered with it yet while i work on other stuff)

[viral sonnet]

i mean, it says a lot about how unity sees this all under the hood.

[rotund token]

yeah but no dev is touching that

[viral sonnet]

writing code for the lowest denominator is odd

[rotund token]

it's not about lowest denominator

[viral sonnet]

what makes you think an intern doesn't change a RefRO access to RefRW because he needs to modify it NOW?

[rotund token]

you could give me your code base and i guarantee i'll miss something and cause a bug without knowing about it

because i don't have intricate knowledge of what's going on with all your timings, pointers and access

safety is extremely important, this above anything else is the most important feature of entities imo

and the 1 promise that to me is really important for studios

[viral sonnet]

i get that, yet not everything can be coded in safe space. i feel like we are talking about 2 different things

[rotund token]

it'd break because it was marked readonly

[viral sonnet]

so he removes readonly 🙂

[rotund token]

then it'd break because it needs marked to be readonly

[viral sonnet]

how can you make sure of that?

[rotund token]

because not having readonly in parallel jobs on containers

triggers safety fail

and if it's not triggering safety

then it's safe

and it doesn't matter and was totally fine to do

and if you say he puts a [Disablesafety blah blah]

then it's going to trigger analysis and be highlighted in a review

if you don't want to setup the git actions for this, an easy way to do this is simply making
[NativeDisableContainerSafetyRestriction]
a compile error and require a justification for disabling the error

[viral sonnet]

yep, that's all great. you have thought this through. i'm not sure why we are talking about it 😄

[rotund token]

my original point was just

it makes me sad you'll have to Get, Store, Update over SystemAPI 😦

you should add a codegen extension to handle it automatically!

[viral sonnet]

none of the methods i have can modify readOnly marked components

[rotund token]

actually wait

can you implicit cast the Safe to Unsafe versions?

then it'd work fine

you just want a private field right

[viral sonnet]

in essence yes, the ptr from RefRO/RW would help a lot

yeah, i guess i can make an unsafe version of them

but then they are really unsafe. unlike the UnsafeComponentLookup now

[rotund token]

can't you just do something like
public implicit operator UnsafeComponentLookup<T>(ComponentLookup<T> lookup)

or does your memory layout not match

[viral sonnet]

it does match

[rotund token]

then yeah you should be able to do that! and then just use systemapi

there'd be no difference

that'd actually be sweet then

all the benefits you want, no drawback

[viral sonnet]

yeah, that'd be cool. i'll try that. just need to find a way around to get some private fields for the implicit cast

or i try just casting 😄

[rotund token]

yeah just memory remap!

[viral sonnet]

UnsafeUtility.As?

at least it compiles public static implicit operator UnsafeComponentLookup<T>(ComponentLookup<T> lookup) { return UnsafeUtility.As<ComponentLookup<T>, UnsafeComponentLookup<T>>(ref lookup); }

[rotund token]

yeah i think that should be fine

[viral sonnet]

yeah nice idea. thanks!

[rotund token]

as long as your memory layouts match

[viral sonnet]

yep, all fine. also won't change

that's a really nice solution i must say

always good talking to you 😁

[rotund token]

well outside the times i'm being combative 😄

i'm all about the workflow though!

i'm so excited by SystemAPI removing all this boilerplate (when handle version is added)

[viral sonnet]

and even though i make it seem as i have knack for unsafe stuff, i really care about safety. there's nothing worse if things go wrong with it.

yeah SystemAPI was a really good idea

what i haven't quite figured out. systemA has a RO on a singleton. systemB has a RW and systemC has a RO again. is the dependency chain or sequence defined by the order the systems are updated?

otherwise, how would the job scheduler know that A has to run first, then wait for B and C has to wait for B

[rotund token]

i believe system order matters

there are still cases you'd want to use GetSingletonEntity<T> and access it in a job

to avoid syncs

[viral sonnet]

i'll have to test this at some point. i have an annoying dependency on a NativeList that gets shared around different systems with jobHandles. :/ pretty ugly, hope singletons can help me making this better and safer.

now i'm also missing SystemAPI.GetComponentTypeHandle

odd we get the ComponentLookup treatment first 😄

[rotund token]

makes sense

since it's used for IJE which is being pushed

with IJE you need like no boilerplate anymore

query is handled for your, SystemAPI handles lookups

you lost GetComponent/HasComponent etc from EFE so kind of needed to replace that convivence

[dense storm]

How come I can't access EntityManager from a static reference cache?

Throws me unsafe errors

[viral sonnet]

why would you need that when there's access through World?

[dense storm]

Just making things neat, but non static reference is alright

I usually initialize all manager refs in a static reference cache

Just wondering why this is not allowed

[viral sonnet]

well, it's a struct. you are asking for trouble 🙂

[dense storm]

I see 😂

[viral sonnet]

World.DefaultGameObjectInjectionWorld.EntityManager to get from anywhere. MB I guess, systems have EM anyway

being a struct is not really the reason but making a EM struct static is. caching a normal EM somewhere is fine.

also wouldn't play nice with domain reload disabled

[dense storm]

Thanks for this, got everything working well :)

[dense storm]

How to get list of systems that includes ones with [DisableAutoCreation]?
var systems = DefaultWorldInitialization.GetAllSystems(WorldSystemFilterFlags.All);
This skipped over them.

Found a solution, I just create an interface and loop through everything that implements it.

[drowsy pagoda]

Is there docs on SystemAPI? And also touching the subject of Source Gen, say if I wrote an interface that will implement source gen code, is there a way to “tag” it so source gen picks it up?

[rotund token]

The source code - it's well documented

[drowsy pagoda]

Ok thx I’ll read it over.

[proud jackal]

Not yet, but the general rule is SystemAPI is your high-level entry to the entities world. It can be thought of in four parts SystemAPI.Query which is our iteration solution. SystemAPI.QueryBuilder, which lets you easily write your query anywhere and it will be cached in OnCreate for you. SystemAPI.SomeSingletonFunction which will give you back what's essentially a cached query with .GetSingleWithoutSyncing. and lastly we have access function like SystemAPI.GetComponent, SetComponent, GetBuffer etc. which also cache their corresponding Lookups in OnCreate :3

[drowsy pagoda]

Hmm very interesting. So instead of creating hooks to trigger these actions from within a system, I can do this from MB classes as well?

[proud jackal]

SystemAPI is currently only supported in Systems and partly EntitiesForEach (ironically)

Inside a System it will latch to the world that the corresponding system is within. That's the general rule and the reason it's called SystemAPI

[drowsy pagoda]

Ah i see, hence source gen.

[proud jackal]

Indeed :3

[drowsy pagoda]

Well, more wiggle room nonetheless

[brave field]

Any ETA to make SystemAPI able to use outside System? Mostly to use it at MonoBehavior.

[proud jackal]

No, not yet, tho we do ask, why you use MonoBehavoiur in the first place? Instead of doing .AddComponentObject on an entity and then working with all your monobehaviour intercom in systems :3

[brave field]

Coz we dun have dots UI solution so still need to use MonoBehavior

[rustic rain]

ahem

UI Toolkit

it's not dots

[brave field]

I try to use system to bridge UI but ends up really suck code 🥲

[rustic rain]

but you don't have to use mono

[proud jackal]

UIToolkits, only dependency on Mono is the use of the canvas, but that can easiliy be put on an entity and accesed on systems! :3

[brave field]

The code is much clearer to just write it directly at MonoBehavior

[rustic rain]

UI toolkit let's have even cleaner solution

without monobehs

the only downside - UI won't be as clear, as uGui

UI toolkit debugger sucks

[proud jackal]

Which UI system are you using?

The old one I can imagine why you would think it. But that's where a static singleton mb's come in:

class UIHandler : Monobehaviour {
  public static UIHandler Instance => _inst;
  static UIHandler _inst;
  void Awake() => _inst = _inst == null ? this : _inst;

  public Button MyStartBtn;
}

Then systems can easily access, use and add behavoiur like:

partial struct SomeSystem : ISystem, ISystemStartStop {
  //...
  public void OnStartRunning(ref SystemState state){
    UIHandler.Instance.MyStartBtn.OnActivated += () => { Debug.Log("Button Pressed"); };
  }
}

That's a rough sketch likely some of the names are wrong 😆

[rustic rain]

there's a bit of trouble with even that due to world's getting created before scene

[proud jackal]

OnStartRunning :3

[rustic rain]

that's the worst

if (field is null) field = value;
I hate it with all my heart xD

So I implemented special callbacks on scene load for world boot

[brave field]

Currently is using Unity UI. Plan to use ui toolkit runtime in future.

👀 Wow. I thought can't put managed thingy at ISystem?

[rustic rain]

not in a field

execution itself can be managedd

[proud jackal]

True but it's been the standard for ages, cause you might end in a scene with a new one, or something just loaded weird and you ended with two, so it should handle the case 😆

[rustic rain]

but you might want it only on OnCreate or OnStartRunning

[proud jackal]

Yup, managed code is completly fine (as long as you don't have managed fields and accept that managed code in your function means that function is now no longer a burst candidate)

[rustic rain]

Here's my solution to systems creation

        private void SceneManagerOnSceneLoaded(Scene arg0, LoadSceneMode arg1)
        {
            SceneManager.sceneLoaded -= SceneManagerOnSceneLoaded;

            _bootExtensionManager.SceneLoaded();

            foreach (var system in _world.Systems)
            {
                try
                {
                    if (system is ISceneLoaded post)
                    {
                        post.OnSceneLoaded();
                    }
                }
                catch (Exception e)
                {
                    Debug.LogException(e);
                }
            }
        }

[rotund token]

i personally found this pretty gross 😄

[brave field]

At 0.51, when I use ISystem without putting [BurstCompile] tag I will just get crash at player runtime build. Not sure whether it's still the case at 1.0.0-exp.8?

[rustic rain]

it is, but what can you do

[rotund token]

just make you world created after scene load if you need something in oncreate

[rustic rain]

I tried that

it sucked

hard

[rotund token]

how so

you're basically doing it now?

[rustic rain]

I had to implement tons of async code

no

[rotund token]

unless you're monobehaviours are referencing entity world

[rustic rain]

    public class BetterWorldInit : ICustomBootstrap
    {
        private World _world;
        private BootExtensionManager _bootExtensionManager;

        public bool Initialize(string defaultWorldName)
        {
            SceneManager.sceneLoaded += SceneManagerOnSceneLoaded;
            Debug.Log("Creating world");

            _world = new World(defaultWorldName);
            World.DefaultGameObjectInjectionWorld = _world;
            _bootExtensionManager = new BootExtensionManager(_world);
            var systems = DefaultWorldInitialization.GetAllSystems(WorldSystemFilterFlags.Default);
            _bootExtensionManager.Create();
            DefaultWorldInitialization.AddSystemsToRootLevelSystemGroups(_world, systems);

            foreach (var system in _world.Systems)
            {
                try
                {
                    if (system is IPostCreate post)
                    {
                        post.OnPostCreate();
                    }
                }
                catch (Exception e)
                {
                    Debug.LogException(e);
                }
            }

            _bootExtensionManager.PostCreate();


            #if !UNITY_DOTSRUNTIME
            ScriptBehaviourUpdateOrder.AppendWorldToCurrentPlayerLoop(_world);
            #endif

            return true;
        }

        private void SceneManagerOnSceneLoaded(Scene arg0, LoadSceneMode arg1)
        {
            SceneManager.sceneLoaded -= SceneManagerOnSceneLoaded;

            _bootExtensionManager.SceneLoaded();

            foreach (var system in _world.Systems)
            {
                try
                {
                    if (system is ISceneLoaded post)
                    {
                        post.OnSceneLoaded();
                    }
                }
                catch (Exception e)
                {
                    Debug.LogException(e);
                }
            }
        }
    }

here's the full way it works now

[rotund token]

in which case you are beyond hope 😆

[rustic rain]

I kept all normal world creation the same

just added my own extensions to it

[rotund token]

and it cost you 2 weeks of work

because you couldn't use your inspector!

[brave field]

Oh. You mean write managed function is still ok. Just don't put any managed component?

[rustic rain]

it did

but it wasn't exactly it

anyway's it works now, and I'm glad I can keep working

[proud jackal]

Managed Components are managed code, managed code is fine, so by the transitive property so is Managed Comps

Just don't expect to have Burst in managed functions, a function is managed when it contains any managed code.

ISystem itself can't be managed:

partial struct MySystem : ISystem {
  string myField;
}

member functions can

[rotund token]

(which I love you created a static analyzer for - so many times i accidently made it a class wasted time trying to figure out why it wouldn't work)

[proud jackal]

Haha, yyuppp... 😆

[brave field]

👀 I see. Just can't put any managed data directly into ISystem

[proud jackal]

But you should be using Singletons for that anyways :3

[brave field]

🧐 Then actually I can migrate all my SystemBase to ISystem to get better performance even can't BurstCompile still can get quite good performance than SystemBase?

[rustic rain]

yeah, they are lightweight

it's like comparing GameObject and Entity

kek

            foreach (var (rma, mmiVARIABLE) in SystemAPI.Query<RenderMeshArray, RefRO<MaterialMeshInfo>>())
            {
                
            }

Is that correct way to query over shared comp?

[misty wedge]

I had the same issue, I have no idea. I instead just used and entityquery 🤷

[rotund token]

you know what hasn't had enough credit yet? unmanaged shared components

[misty wedge]

Yeah I love it

I can finally burst my spawn functions

[rotund token]

i used to have like duplicate components

[misty wedge]

Yeah I still kind of have one kind of duplicate component, but it has a use to force entities into the same chunk for some stuff

But if I can I plan to remove it

[rustic rain]

oh god

[rotund token]

yes? 😈

[rustic rain]

tertle, remember that selection tool?

it's broken

[rotund token]

oh is it? i had some huge issues in editor for a while

until i realized it was conflicting with unitys own solution so i removed it

i haven't tried at runtime though

[rustic rain]

I'm trying to figure how to bring back mesh chunk gathering

[rotund token]

oh because rendermesh doesn't exist?

[rustic rain]

yeah

[rotund token]

ah yeah that's a bit of a pain to redo

[misty wedge]

I'm still not sure how to deal with collections not being writable and readable inside a component from a lookup. I guess unsafe for now and see if unity can add the safety system for that in the future? 🤷

[rustic rain]

wdym not being able?

[rotund token]

components are effectively containers (chunk)
and safety doesn't work nested containers

[rustic rain]

ah

[misty wedge]

It's what I mentioned yesterday, the safety system will complain telling you that it's write only (I wasn't being entirely clear in the first message)

[rotund token]

so you can't inject safety handles while scheduling a job so it always errors with this is write only

[misty wedge]

I think you can write to it but you can't also read from it

[rotund token]

native containers on components only usable at system level

good for sharing between systems though

[misty wedge]

So I guess make them unsafe for now? It would be cool if that last bit could be added to the safety system, it sounds tricky though

[rotund token]

and you can pass them in from the system

[misty wedge]

If it was just lists I would replace them with DBs, but I have some other stuff in there too

[rotund token]

and it'll work fine, it just breaks if you access the component itself in the job

[misty wedge]

Yeah if it was a single one I'd do that, but I have a few of them. If the container is unsafe this is completely ignored right?

[rotund token]

yeah there's no safety on them

[misty wedge]

And I'd still have dependencies at least from the read / write permissions on the component

[rotund token]

yes

[misty wedge]

So just be careful when defining those

[rotund token]

^

as long as you define your read/writes correctly it should be fine

[deft pilot]

anyway to specify do something every second in a system?

like i could store the last delta total and action it when its over 1f

but thats prone to error

or should i just work off framerates?

ie/every 60/120 frames do x

[rustic rain]

well, that's how it works though

[deft pilot]

i guess deterministic would require me to do it frame by frame

i'll just do that

[rustic rain]

there's built in RateManager btw

FixedStep group uses it

[deft pilot]

i'll take a google of that thanks

[deft pilot]

is a buffer FIFO?

aka if i add 100 elements the first element is always going to be index 0 when i get the buffer assuming no changes

and 100th element is always gonna be 99th index

[spice vale]

yes, it works like a list

[deft pilot]

alright making sure there isn't any weirdness under the hood

[deft pilot]

just realized a problem

with using tilemaps and ECS

since i have no way to call TileMap's methods

how am i suppose to get the world center of a tile?

i'm guessing the answer is something like

just store every tile in a buffer along with the position

[pliant pike]

I mean you could just calculate it, if you know the size of the grid and the size of each tile?

[pliant pike]

is there no HasBuffer() method I thought there was 😕

[proud jackal]

SystemAPI.HasBuffer 😄

[pliant pike]

I'm on 0.51 so maybe its different

Don't tell me I'm going to have to update 😟

[half jay]

is it possible to use Hybrid approach now? In some cases I have GO which doesn't destroy and then im control it from ECS. Also Im using AR module which stroing in shared component and manage it in system. So i did it through convertation. If i now move all AR monobehaviours from main secne to subscene will it break AR module? Will the monobehaviours logic work in Subscene?

[pliant pike]

There's lots of ways of using GameObjects not just with subscenes

I highly doubt the monobehaviours logic will work, its only the data that is converted, and its only the data that Unity knows about and or is explicitly created

[gusty comet]

Any ETA on being able to add an initially disabled component in a Baker?

Right now as we know, it's forced to be enabled by default

Actually, would a temporary workaround be implementing a BakingSystem that disables the desired components? This would require that BakingSystem only runs once for each component during conversion/baking

[coarse turtle]

Why not disable the gameObject in the subscene? It adds a disable tag.

[gusty comet]

I only want to disable certain components

[coarse turtle]

oh sorry I misread

[gusty comet]

youre good

[brave field]

@rotund token Have u get dots netcode working at hello world project? Seems like it's broken that until now I can't get a working player runtime build yet

[viral sonnet]

there are a bunch of build problems currently. even without netcode. what error do you have? cant compile or crash?

[whole gyro]

I was just testing this and I don't think it is possible with bakers or baking systems. Setting the enabled bit in a baking system does not seem to carryover to runtime.

@viral sonnet I believe you mentioned that you were unable to set enable bits on prefabs using an ECB. I just tried it seems to work fine, despite what the docs say. ComponentLookup.SetComponentEnabled also works for disabling compoennts. And entities instantiated from the prefab have the same enabled bit as whatever prefab they came from.

But this is all at runtime. Couldn't get anything working in a baking system, just like you were saying.

[viral sonnet]

yeah runtime works fine

so have you managed to get it to work for baking systems?

[whole gyro]

No, tried a bunch of things and no luck so far.

[viral sonnet]

but you can change a prefab enabled state and then the instantiated are also correct?

[whole gyro]

Yeah exactly

Luckily, all my objects that I was planning to use disabled components on are instantiated from prefabs, so I'm just going to disable the components on the prefabs in the first frame

As a workaround for now

[viral sonnet]

ok, i'm gonna wait this out and just set the flag on in the instantiation

[gusty comet]

That's good to know. I'll just ignore component enabling/disabling until I really need it

Also, does Baking only work in Subscenes for now?

[whole gyro]

Yes

[gusty comet]

Are Subscenes the way to go from now on?

[whole gyro]

Yeah, they removed runtime conversion/baking so subscenes are the only way to bake entity data.

But there are some cases where you need to create entities from a mono behaviour for some hybrid setup. For that, you need instantiate the entity and add components manually. This is similar to how runtime conversion worked, but all manual. You can't just add the ConvertToEntity component to your game object any more.

[gusty comet]

Thanks!

[drowsy pagoda]

Am I understanding this correctly? If I use SystemAPI.QueryBuilder within the OnUpdate scope in order to pass it into a job.Schedule(query, deps), the Source Gen will move that query builder to OnCreate and cache it there, so even by calling it every frame, I'm actually always receiving the cached version right? It's not building it from scratch each frame?

[rustic rain]

🤔

[rustic rain]

you can quickly check source tbh

just compile it

and then take a look at source of codegen

[rotund token]

pretty much

the description are pretty good for this

Calling this method should allow for construction of EntityQueries with almost no run-time performance overhead."```

[drowsy pagoda]

yeah that's how I understood it from description, just wanted to make sure I understood it correctly cause that's an awesome feature.

[rotund token]

that's basically SystemAPI in a nutshell

[rustic rain]

hmm

if it does caching under the hood

then you can save so much space

in files

pog

[rotund token]

yes

the only thing systemapi is missing is Handle versions

and you can basically remove OnCreate from most systems

and if you already use IJE you already can

because the GetComponentLookup etc are all cached/auto updated for you

[rustic rain]

oh man

I wish to fix this stuped entity selection faster

so I can start reworking systems

I even consider moving rendering to single thread due to complexity of lookups

I mean

to non-instanced version

[patent kelp]

You guys have been so great and the tutorial that you gave me 5 days ago was a dream... this one... is there a good tutorial for DOTS 1.0 collision? (link that you gave me:https://github.com/Unity-Technologies/EntityComponentSystemSamples/blob/master/DOTS_Guide/ecs_tutorial/README.md )

[rotund token]

downloading the actual physics samples
https://github.com/Unity-Technologies/EntityComponentSystemSamples/tree/master/PhysicsSamples

is probably the best

you can see on that page how many different samples they have

[patent kelp]

Do those still work in 1.0?

[rotund token]

yes

they were updated

[patent kelp]

Thank you Turtle

[rotund token]

only thing not updated yet is netcode

which is coming soon~

[spice vale]

Is it always better to use WorldUpdateAllocator rather than Allocation.Temp whenever possible?

[rotund token]

they are kind of do different things

i think for the most part you should replace TempJob with WorldUpdateAllocator

[spice vale]

WorldUpdateAllocator can be rewinded, right? but not Allocator.Temp?

[rotund token]

temp is rewound every frame

kind of

it's a fixed pre-allocated chunk of memory per thread

[spice vale]

so what is WorldUpdateAllocator then?

[rotund token]

there are 2 WorldUpdateAllocator that swap every frame

but there is only 1 set of of them on main thread, none in worker threads

i.e. it cant be used in jobs

it's resizable but reclaims it's memory back if it hasn't been used recently

[spice vale]

right

[rotund token]

so doing a 1 off giant allocation won't bust it

because it will release it soon

so if you have a pretty consistent per frame memory usage for it, it won't actually need to allocate/deallocate much

it's rewound every 2nd frame (because there are 2 of them they alternate usages) allowing jobs to not be force synced end of frame and can flow through into next

sorry i shouldn't say every 2nd frame

i should say, every 2nd update

thats one of the big difference to tempjob which was tied to frame rate, every 4 frames

rewind allocator you use them yourself (for example ECBS have their own) rewind on demand

so you might have a very slow ticking once a second system that can allocate out memory to other systems

with its own rewind allocator

and you only rewind it every second or so, very useful

(though this is off topic and not specifically WorldUpdateAllocator but this is what it uses)

[spice vale]

WorldUpdateAllocator rewind functionality is hidden for us then I suspect?

[rotund token]

it's also moved from 0.51 where the rewinding happens

but you can see how it works

RewindableAllocator

in Unity.Collections

and if you want to add your own EntityCommandBufferSystem has a good example of using one

[spice vale]

OK, thanks for sharing such useful knowledge, this should be pinned.

So as far as I can see, there is no benefit of using Allocator.TempJob, when we have WorldUpdateAllocator, correct?

[rotund token]

i'm unsure of a reason to use it

but i don't want to blatantly say it should never be used

at the very least, unsafeutility.malloc does not seem to support custom allocators

[spice vale]

Oh I see, we can access the actual allocator via the World, and rewind from there, if we want to do it manually.

Thought we just had access to the label for a sec

[rotund token]

i don't think you should ever be rewinding the world allocator yourself

that is going to cause some serious issues

good chance a job is currently using memory from it

[spice vale]

You think so? It looks like it keeps track of used/unused memory blocks. Still I probably won't do it since it automatically happens every 2nd update like you said.

nvm, I think you are right

[rustic rain]

I don't get, what's the use case for that world allocator?

[spice vale]

Afaik it seems like better than Allocator.Temp/TempJob when allocating on main thread.

I know Temp can get full, but I don't think WorldUpdateAllocator can.

[viral sonnet]

i'm guilty of not having switched to world allocator. I'm still using Allocator.Temp/Job mostly. Rewindable World allocator works like a double buffered memory. the memory isn't allocated from thin air but already allocated and swapped each update. at least, that's as far as I understood it

[rustic rain]

🤔

[viral sonnet]

so i see the usage, for memory that you need per frame which is pretty big. i don't have any idea what the limits are. like, can it reach 2gbs or something

[rustic rain]

So it's just faster allocator than jobtemp?

As For use case

[viral sonnet]

here they use it quite a lot: https://github.com/Unity-Technologies/EntityComponentSystemSamples/blob/master/ECSSamples/Assets/Advanced/Boids/Scripts/BoidSystem.cs

line 266

huh, that warning only comes up in ISystems? [Conditional("ENABLE_UNITY_COLLECTIONS_CHECKS"), Conditional("UNITY_DOTS_DEBUG")] void CheckOnUpdate_Handle() { #if ENABLE_UNITY_COLLECTIONS_CHECKS || UNITY_DOTS_DEBUG if (Burst.CompilerServices.Hint.Unlikely((m_Flags & (kIsExecutingISystemOnUpdate | kDidWarnIsExecutingISystemOnUpdate)) == kIsExecutingISystemOnUpdate)) { Unity.Debug.LogWarning($"'{new FixedString512Bytes(m_DebugName)}' creates a type handle during OnUpdate. Please create the type handle in OnCreate instead and use type `_MyHandle.Update(ref systemState);` in OnUpdate to keep it up to date instead. This is significantly faster."); m_Flags |= kDidWarnIsExecutingISystemOnUpdate; } #endif }

[rotund token]

it has no limit

it'll resize to whatever you want

it also has the beautiful benefit of not having to dispose it

so you don't need to schedule another job after your job to dispose your tempjob memory

[viral sonnet]

so am i right in the assumption that if you allocate 100Mb per frame, with the rewindable memory it'll actually be double, 200Mb?

[rotund token]

if you do it every frame

it wont have to allocate anything

after the first allocation it'll just re-use it

but yes it would cause 200MB of memory

[viral sonnet]

ok

huh, this seems to mostly work for fixed allocations then. can a NativeList also be allocated with rewindable memory?

[rotund token]

yes

i believe all the native containers can be now

[viral sonnet]

do you know how this will react when the memory grows? will it take rewindable memory also? i can't see this working out of the box honestly

but if it does, i would actually not need persistent lists for performance reasons

[rotund token]

why not?

[viral sonnet]

but the cap would start at the init cap. 0/1 or whatever

when the list grows during the frame it'll also resize several times (possibly)

[rotund token]

yes?

just like any allocation

[viral sonnet]

i guess i'm mixing 2 things 😄

but still, the rewindable memory must be tracked and resized also then

pretty cool then

i'm mixing up a persistent native list that doesn't need to resize anymore.

[rotund token]

        public struct AllocatorHandle : IAllocator
        {
            internal ref TableEntry TableEntry => ref SharedStatics.TableEntry.Ref.Data.ElementAt(Index);
            internal bool IsInstalled => ((SharedStatics.IsInstalled.Ref.Data.ElementAt(Index>>6) >> (Index&63)) & 1) != 0;```

this is what is passed now and stored in the container

and there's AllocatorManager

which does most of the lifting

[rotund token]

i don't really see the benefit of replacing a persistent container though

as long as it doesn't have hugely spikes of memory

in which case rewind has the benefit of cleaning it up and not sticking around

[gusty comet]

Can a Baker remove components? I'd like to create a Baker that completely removes all Transform-related components in order to reduce chunk usage per entity

[viral sonnet]

true dat. i'll evaluate those persistent lists, some Capacity can be calculated so that would make it possible for rewindable memory

[viral sonnet]

no, only a bakingsystem that uses ECB

[gusty comet]

Gotcha. Guess I'll rely on StaticOptimizeEntity then

[gusty comet]

Oh wait, sure! I'll try that

[viral sonnet]

if it's just transform there are solutions

GetEntity(authoring, TransformUsageFlags.ManualOverride);

[gusty comet]

Does that remove the components?

[viral sonnet]

it doesn't even add them

i use this to get rid of them. works fine

or var additionalEntity = baker.CreateAdditionalEntity(TransformUsageFlags.None);

if you create child entities

[viral sonnet]

the sticking around and process of resizing is bit annoying to handle. good benefit then

[gusty comet]

Thank you for the info. I should have specified though that I want the current entity that the Baker is processing to lose all of its Transform components

Kind of like this:

public sealed class Positionless : MonoBehaviour
    {
        
    }

    public sealed class PositionlessBaker : Baker<Positionless>
    {
        public override void Bake(Positionless authoring)
        {

        }
    }

[rotund token]

if you want no transform components just do
GetEntity(authoring, TransformUsageFlags.ManualOverride);
as enzi said

[gusty comet]

Gotcha. Thanks!

[viral sonnet]

GetEntityTypeHandle got the .Update treatment. nice 👍

now, should i rewrite all this or wait for SystemAPI? 😅

i'm really not in the mood to rewrite all this 🤣

[rotund token]

i made the mistake of redoing all this before i realized the power of systemapi

that said, really wasn't that time consuming

[viral sonnet]

do you know how netcode handles tick? is it in a singleton?

[rotund token]

oh it's super complicated

the first question you have to ask is

what tick

[viral sonnet]

oh damn 😄 server tick

[rotund token]

clients can tick multiple partial frames

[viral sonnet]

it's super simple for me. maybe i just need to use SharedStatic. I had a static uint in a SystemBase. Now ISystem and Burst doesn't like static of course.

[rotund token]

in 1.0
NetcodeServerRateManager

they used to have custom system groups and stuff but got rid of them in 1.0

and switched to ratemanagers which they just add to the default SimulationSystemGroups etc

imo best approach - what they're designed for

[viral sonnet]

NetcodeServerRateManager is a systemGroup, right?

[rotund token]

no