archived-dots

[rustic rain]

I'd say it won't be as efficient

[whole gyro]

yeah maybe, but I'm only syncing the main character so don't really care about performance. I don't think I knew about the built-in options when I wrote it and haven't gone back to it yet.

[safe lintel]

companion transform already creates garbage per companion so id say anything else is gonna be better unless 1.0 brings improvements

[whole gyro]

IIUC, the scriptable object bridge thing is mainly there to make it easier to access input from lots of different mono behaviors. If you go the ECS route, most of the input handling will be in systems instead. It is still possible to do something similar, but the workflow will be different.

I would stick with normal game objects in this case. And just keep DOTS in your back pocket as an optimization tool.

[gloomy venture]

Thanks. It does feel like I'm setting myself up for a lot of pain otherwise. I really want to start getting a handle on DOTS, especially for my more ambitious projects, but it feels a little like coding with one hand behind my back. :p

[rustic rain]

ECS is totally other way of thinking

[whole gyro]

From my experience, ECS is great once you get the hang of it, but even then, I spend a ton of time figuring out nicer ways to handle certain problems. Input is actually one of those areas that I've been meaning to come back to. There aren't really best practices for a lot of things yet so you'll need play around with things and be willing to refactor everything.

[rustic rain]

I have found a nice approach for input for me

using components as input, and making systems run based of existance of queries

[whole gyro]

Yeah, I can't remember if I already asked you but weren't you doing the thing where you had a singleton per action?

[rustic rain]

yeah, this

tbh

this approach is not THAT good, now that I think

because you can always extend any input

by creating new input component

and extensibility was key for me

[gloomy venture]

I'm ok with this, I've been "around" Unity for a long time, but never really had time to settle into a committed period of working on it. Meaning I start getting some things figured out, then life and work happens and I step away for 6 months, and when I come back it's ground zero again. I have been wanting to jump on the DOTS bandwagon, as I think it's the future, but I also have some hard deadlines for this prototype, so will need to balance desire with reality.

[whole gyro]

I have a lot of systems that start with something like:

if (!GetSingleton<Input>().ActionIsHappening) { return; }

Using RequireSingletonForUpdate seems like a nice way to solve this.

[rustic rain]

actually I just figured

someone else's approach

by creating huge struct of all inputs

is just better

the idea is this

you create entity, give it this component data struct

and never touch it ever again

it'll be static entity, kind of like world time entity

you grab pointer reference to that struct

and viola

the fastest access

to any field of this struct you want

for now I just don't see how to ease all code boiler plate it brings

as in my approach I used reflection

[whole gyro]

I assume some system needs to update the component data and would need to be done safely. And I would still need to check for an input in each system instead of conditionally running the sytem. It seems pretty much the same as the GetSingleton<MyBigInputStruct> approach but presumably faster and without and dependency tracking

[rustic rain]

well, you're supposed to only write data through New Input System

which is done even before Initialization group

so it would be safe to use as pointer

quite literally just like Time component

[whole gyro]

Ah, didn't know when it ran. I use the C# class generation thing to receive callbacks but then write input in a system at the start of the init group

That seems like a good option for performance, but you can no longer make systems conditional on some input.

[rustic rain]

well, you can

but you shift that check

into Update method

tbh, not even sure what's faster

early return in Update, or not calling system at all through query

[whole gyro]

yeah that's basically what I'm doing now with the early return. But GetSingleton is a query so I guess its probably similar to the RequireSingleton approach. So I guess your pointer idea avoids any query.

[rustic rain]

GetSingleton is super slow

if it's only one entity

which gets this singleton

you should cache entity

and then use GetComponentData instead

[whole gyro]

Good to know, thanks. I'll can try that, although I don't look forward to having more boilerplate

[viral sonnet]

RequiresSingletonForUpdate and GetSingleton go pretty much hand in hand. does it even help to cache the entity when the system has to calculate if it can run?

[rustic rain]

GetSingleton has to find required query

while get component data looks for exactly what you want

[viral sonnet]

read again 🙂 not what i'm getting at

[rustic rain]

but I didn't mean RequiresSingletonForUpdate to be required

[viral sonnet]

when you don't need RequireSingletonForUpdate because the singleton gets created in OnCreate of the system, sure thing. with the update requirement the query would still get evaluated

[rustic rain]

still, GetSingleton is simply slower, because it has to find required query

[viral sonnet]

problem i see, most singletons are not created in the systems that also use them. otherwise, why make a singleton and not just cache the data directly in the system

i'm not saying otherwise

[pliant pike]

GetSingleton is slow

my world is destroyed, I use that all the time

[whole gyro]

Let's say I want a system to update when left mouse button is pressed. From the options we talked about:

1. Have one big input singleton. Return early if GetSingleton<Input>().LeftButtonPressed is false. Can use RequireSingletonForUpdate on the big input singleton, but let's just assume it always exists.
2. Same as above, but cache the singleton entity. Check GetComponent<Input>().LeftButtonPressed instead.
3. Have singletons per action. Use RequireSingletonForUpdate<LeftButtonPressed>.
4. Create something like the world Time component.

[gloomy venture]

I may be grabbing the cat by the tail here, but I would think the Singleton pattern won't work well with Jobs and the Burst compiler. Unless the info in the singleton is read only?

[whole gyro]

Its basically readonly, yeah. And in my case, I check inputs prior to scheduling jobs so this is on the main thread anyways.

[rustic rain]

singleton pattern doesn't usually contains data

[gloomy venture]

The new Input system is Event -driven though, so the singleton approach kinda pushes it back to the old Input. Is there no better way to take advantage of this event-driven model in combination with DOTS?

[rustic rain]

and if it does

it doesn't matter anyway

because input rarely gets to work with jobs

all of my input systems are ISystems

that run on main thread

[viral sonnet]

icomp singletons are nice for options/settings the player has. shared between many systems, mostly readonly and only some options menu can write to it

[whole gyro]

I have a system that collects these events and writes them all at once at a specific time during the frame. Maybe that is not the intended way to use the new input system, but in my game, I need a specific ordering of systems and input handling. I don't want systems getting triggered by input at arbitrary points in the frame

[whole gyro]

I'm curious, how do you handle input?

[viral sonnet]

i have written a few ways but i settled on nothing special, not even a system. MBs Update run before entities update so I gather inputs there and done 🙂

[gloomy venture]

again, at the risk of being the noob in the room, is anyone using ScriptableObjects? it sounds like many of these singleton problems can be solved with them.

[rustic rain]

how do you imagine that?

[whole gyro]

Do you then write the input to a singleton and read that from systems?

[viral sonnet]

not quite, it's manually done and written to the correct entity. player/ai are all driven by the same input format so singleton wouldn't work. well, it would just for the player but it's not necessary, i know which entity to write to anyway

about singletons in general, use them when the data is shared between systems. otherwise, just use system data. that works also fine

(or just work with a known entity)

[rotund token]

unless your system has like 100s of queries getsingleton isn't that slow...

(i'm all for caching things though)

[viral sonnet]

agreed, it never catched my attention to anything i've done. if it makes life easier go for it. (i keep queries/systems down to a minimum - i can see a point where it could get problematic - some devs have like 300 systems or something crazy ^^)

[rotund token]

everyone enjoying their dose of dots news?!

[viral sonnet]

yeah forum is very active recently

[rotund token]

i for one love the, everyone speculates before release /s

[viral sonnet]

i'll say the same thing i say every week. next week!

[rotund token]

we'll see 🙂

[viral sonnet]

eventually i'm right. just a matter of time

i've said this a couple of times, i'm a bit dissapointed there's no mention of other things. sure, the new baker is nice and looks very streamlined but what about the other stuff. there must be more

[rotund token]

no one's asked

[viral sonnet]

lol, maybe that's it 😄

[gentle gyro]

Sorry, I meant specifically rendering ECSs using the hybrid renderer on Quest 2...

[whole gyro]

Personally, I'm most excited about the better editor integration stuff they showed off: https://www.youtube.com/watch?v=p4ct4vHWYt0

[viral sonnet]

and now i did 😄

[viral sonnet]

anything specific? journaling and the conversion switch i know of

[devout prairie]

i'm not sure if hr runs on quest 2 but just as a general comment, it does seem like quest would be a good use case for leveraging it

given it's basically a mobile platform with quite limited performance

[rotund token]

i for one am quite excited by a few features or potential api changes

(though I am still concerned about 128 entity limit, but i'll profile and bring it up once i get to test rather than jumping on the anti-hype train)

[whole gyro]

Journaling is the main thing

Also, I don't remember if I they confirmed in that video, but scene view picking of entities is something I've wanted for a while now. Based on some commits to the graphics repo a few months ago, I think it is going to be in 1.0

[rotund token]

oh i implemented this a long time ago

i can just give you this

[whole gyro]

Oh nice. I have something similar that works in the game view already. Hover an entity and press a button to pull it up in the inspector. But haven't tried to get something working in the scene view.

[rotund token]

even works with skin meshed renderers that you've attached to entities as hybrid

[whole gyro]

If you want to share it, that would be cool

[rotund token]

i think this is the primary code - it's definitely going to miss something small

i keep saying i'll release this core library but i keep getting lazy to cleanup some editor tool i need to remove first

i'll see if i can do it today because it's getting ridiculous

[gentle gyro]

Yes it would.

I'm getting shadder compiler errors (I forget what exactly since I'm not at my computer right now)

When I googled it, I get the impression it's not supported yet but I can't find a definitive answer...

[devout prairie]

afaik quest is android right which would be Vulcan ( the graphics driver )..

a quick search on here brought this comment:

in the #🥽┃virtual-reality channel

as far as i know, hybrid renderer does support vulcan so i'm guessing it's a case of getting the right combinations of unity editor and graphics driver version

in the Hybrid Renderer docs:

[gentle gyro]

I tried disabling everything but Vulkan and still have the error

Think it's an actual shader error, don't think the shader is designed for this use-case

[rotund token]

ok done, wasn't even that time consuming. going to head to gym and do a final build check when i'm back

and release when i'm back

[errant hawk]

wait what's the news?

[rotund token]

Read forums, flurry of posts

[wraith hinge]

don't worry, there's more 🙂 my top two personal favorites, if i have to pick, are the job system rewrite (purely under the hood) and bursted isystems actually working. together they represent a pretty significant speed boost

(probably biased because i work on the team for those things though)

[viral sonnet]

great to hear, i'm mostly excited about ISystem stuff. are nativeContainers also working now in ISystems or do we still have to use UnsafeContainers?

[wraith hinge]

yep nativecontainers work now

[viral sonnet]

🤤

thanks for answering! 🙂

[wraith hinge]

np! also p.s. enableable components and unmanaged shared components are pretty good too

[viral sonnet]

includes "actually working" ECB system too? that was also a roadstop for some of my systems

[wraith hinge]

EBC?

ECB! yes

[rustic rain]

you mean if I stack 10 ISystems in 1 system group they all will run with 1 burst call?

[viral sonnet]

wow you got it all covered!

[wraith hinge]

ehm. let me check.

(i would be kind of surprised if that was a big deal either way? because the burst call overhead should be small compared to a whole system update. but checking...)

[rustic rain]

I think it is a big deal, because this way it can run both, internal logic that determines whether system should run AND system's update all in one go

add to that stacked systems and it save quite some overhead

[wraith hinge]

it's 10 separate calls because it wants to have a catch around each update, and you can't catch in burst. i would be very interested in a profile where burst call overhead made a big difference here

i should say that in 1.0, systems are [alwaysupdatesystem] by default, and you have to sign up if you want to require matching queries for update. this is because everybody was always going insane because their systems weren't running and they forgot to say alwaysupdatesystem.

also, if you don't require matching queries for update, and you have a bursted isystem, you can just check if the queries match at the top of your update, and it should be just as fast

(i haven't profiled this comparison, it's just my intuition)

[viral sonnet]

interesting, that change is probably for the best.

[devout prairie]

don't be a stranger! we'd welcome to have more involvement from team members around here 🙂

[viral sonnet]

{
    // This reference provides read only access to the Turret component.
    // Trying to use ValueRW (instead of ValueRO) on a read-only reference is an error.
    readonly RefRO<Turret> m_Turret;

    // Note the use of ValueRO in the following properties.
    public Entity CannonBallSpawn => m_Turret.ValueRO.CannonBallSpawn;
    public Entity CannonBallPrefab => m_Turret.ValueRO.CannonBallPrefab;
}``` so this look like unitys way of giving us reference types without unsafe code. otherwise i'm confused what the purpose of the aspect is when every value is in Turret anyway

reading through this btw: https://github.com/Unity-Technologies/DOTS-training-samples/tree/02aa55d68e8ebcd7bf2806af04fc9fa31dc52535/Ported/combat-bees

[viral sonnet]

uh [BakingType] attribute is cool. we are then able to put a dynamicbuffer or whatever on the entity which is removed after conversion

and GameObjectConversionSystem is obsolete in favor of [WorldSystemFilter(WorldSystemFilterFlags.BakingSystem)] on SystemBase/ISystem

[devout prairie]

Well that's good because i like cakes

i always felt what was missing was a good baking system

deceptively hard to get a good sponge

[viral sonnet]

yep it felt a bit tacked on and not very streamlined

[devout prairie]

yeah a bit of a mess tbh

[rotund token]

How do you currently handle transform.position in entities?

[viral sonnet]

is that question related to the aspect?

[rotund token]

Yes

Because that's what aspects can solve

[viral sonnet]

burstable transform access?

[wraith hinge]

i mean you could just directly access the underlying components of a transform. aspects are just convenient

[rotund token]

I mean the every equiv

Like translation is local

So is you want you want to set world position you need to check for parent, use local to world

Update translation to make sure you're world value isn't overridden

It's a huge pain replicating game object transform in entities atm

This is what I believe aspects can help solve

[viral sonnet]

well isn't the problem also that we can only get access to transform in IJobParallelForTransform? TransformAspect looks like a built in helper for that

[rotund token]

I'm not talking about gameobjects

I'm talking about replicating the same easy to use api that object transforms have but for pure entities

[viral sonnet]

ah i see

[rotund token]

In entities 0.51 how do you set an entity in a hierarchy to (10,0,10) world space

[viral sonnet]

but how would aspects solve that problem? the comps are local to the entity still and the same kind of data queries need to be done to determine what has to be set

[rotund token]

It wraps all the data for you and provides an api

[viral sonnet]

yeah sure, not 20 ComponentTypeHandles needed then

as you can directly query aspects the question comes up, how does it handle comps that are not added. like where we are using chunk.Has<T> now

[rotund token]

Without actually having source, hopefully transformaspect provides, position, localposition etc

[devout prairie]

Does it manage getting parent/hierarchy translation data etc in order to get set world and local values to a child?

[rotund token]

That's what I'm hoping

Otherwise I'll be implement my own anyway to do that

[devout prairie]

So I guess it could potentially be expensive if not done optimally

[viral sonnet]

if it doesn't it's kind of superfluous, just reducing boilerplate code isn't enough to get me to use aspects 😄

[rotund token]

To me aspects are the most important feature of entities in the future

[devout prairie]

That was kinda my thought tbh enzi

[rotund token]

It's what will let entities go mainstream

This reminds me of subscene hate

[devout prairie]

But yeah i think that's very true also

[rotund token]

I'll be preaching for 6 months and finally convince everyone

[viral sonnet]

it's not the same as subscene hate. it's like Entities.ForEach hate 😛

when aspects give the same control, fine. 🙂

[devout prairie]

ForEach are great tho 😛

[viral sonnet]

same goes for, can i get the pointers

and does it actually play nice with burst and simd

[rotund token]

If you want entities to be popular you need to make it as easy to use as gameobjects

And the first step is making transform usage that easy

[devout prairie]

Kinda true tbh

[rotund token]

Stop people having to write complex maths to do basic operations

[devout prairie]

The ground level stuff is currently way too involved, I do see the point

[viral sonnet]

yeah agreed - i have to say though, because i have written the current project so often in different forms. entities will never go actual mainstream. OOP is just too convenient and fast to write

i've been doing this a long time now and my productivity just can't catch up to OOP shenanigans.

so in that sense, a LOT has to happen for entities to get really mainstream

[devout prairie]

I think theres still too much obvious stuff not benefitting from dots, animation etc

[rotund token]

99% is developers don't care about this though

Most gameplay programmers will never touch a pointer in their life

[devout prairie]

One of the true benefits of unity is rapid prototyping and the ability to just throw together a game in couple of weeks.. that's really not possible with dots atm, too many hidden traps etc

Conversion is the ballache of all ball aches

[rotund token]

How so

[devout prairie]

Jeez, dont get me started, it's all over the place

Monos subscenes this works this breaks that breaks this but not that etc etc, endless headaches

[rotund token]

Works flawless for me 😅

Though I will say I put a lot of effort into getting my architecture how I want

[viral sonnet]

i guess we have to stop talking about different target audiences. i strictly talk about myself and what i want 😄 anyway, i agree with what you are saying and the steps to be more appealing for mainstream. that said, i just don't see any junior gameplay programmer doing anything in dots for a while.

[devout prairie]

Once you know what you're doing and have a workflow yeah

[whole gyro]

I feel like that's a tricky problem to solve. Are you are suggesting that Aspects should be able to read/write data from other entities? Even if that were true, I feel like there are complications. Like what if the parent's Translation was modified earlier in the frame but that change wasn't been written to the parent's LTW yet. And then you try to set the world position of the child. Is the aspect supposed to use the parent's LTW or recompute the parent's LTW from the parent's TRS components?

[rotund token]

You don't need to read other entities

You have a local to world

You just have to know if you're a child or not

[viral sonnet]

i'm definitely a child

[whole gyro]

lol

[viral sonnet]

😄

[rotund token]

All this type of stuff is secretly done under the hood for you in game object land

[whole gyro]

What about when that LTW isn't up to date? Like something else modified the Translation component of the entity earlier in the frame.

[viral sonnet]

i think the simple answer is, don't do it 😄

[rotund token]

Up to the user really

[whole gyro]

I'm curious if the transform aspect will write to LTW immediately when you modify the local position.

[viral sonnet]

and cases where LTW is not up to date could mean an ECB is involved

[rotund token]

You can make your world changes after transform system group is you need up to date positions

[viral sonnet]

it probably will. too many problems otherwise i think

but we don't even know if TransformAspect covers this. it's just speculation

[rotund token]

Entities promotes flat hierarchies where possible

Which is good, it's performant and why this problem hasn't come up that much

[whole gyro]

Yeah definitely. But that requires knowing how the transform systems work. If the goal is to make it as easy as game object transforms, I think they would need to remove the delayed updates of some of these components. And maybe aspects are a way to do that.

[rotund token]

But there are still cases where you might want to move a child in world space

[wraith hinge]

aspects are just convenience; all this stuff you're talking about is to do with the new proposed transform system (not for the first experimental release), which is supposed to help with some of the harebrainedness, but still has some kinks to work out

[safe lintel]

are there any more tidbits about the new transform system you can share? will physics utilize it?

also will it(new transform system) be part of the upcoming experimental releases this year (if it is not in the initial release)?

[rotund token]

https://gitlab.com/tertle/com.bovinelabs.core
here you go, I've probably posted like 90% of the stuff in here at one point or other but at least now when I post something you can get the dependencies

it should be non-destructive in the sense that adding it to your project should change nothing by default (if you open settings though it will generate a scriptable object)

will be here once processing is done: https://openupm.com/packages/com.bovinelabs.core.html

[viral sonnet]

lots of great stuff in there. you should make a readme at some point what's all included 🙂

[rotund token]

in 0.5-2 years i'll release my game library

[rotund token]

took me 6 months to delete a folder and make it public, that might take a while

[viral sonnet]

haha, i was half joking

[rotund token]

ah bloody hell i already screwed it up didnt delete a meta updated

[viral sonnet]

so for a simple close button in UIToolkit i can write btnClose.clicked += () => { asset.ShowVisualElement(false); }; the asset (main container) is snapped there. all good, but what about when i want to move this to a method? i haven't found a way to provide the necessary paramaters to a clickEvent

[rotund token]

what do you mean?

can't rider just generate a method that matches the delegate for you

[viral sonnet]

i can make a delegate but that would also have no parameter. It's just a simple Action, no parameters. there's also RegisterCallback<ClickEvent> - it's really the typical problem that UnityUI solves with EventSystem.currentSelected or something. a method has been called but it doesn't know from where. the UIToolkit sample mostly uses methods that rely on private variables from the class. i haven't found anything else.

[rotund token]

https://forum.unity.com/threads/core.1340078/
i included a few highlights

[viral sonnet]

okay! i mean, oh K is in there.

[rotund token]

oh so you want to know what button selected?

you want to have have multiple buttons mapped to same method?

[viral sonnet]

btnClose.RegisterCallback<ClickEvent, VisualElement>(Clicked, asset); private void Clicked(ClickEvent evt, VisualElement root) { root.ShowVisualElement(false); } that's kind of sick. i'm warming up to UIToolkit 😄

[rotund token]

i find it hard to remember all the events

and what elements support what events

[viral sonnet]

hehe, i wondered the same thing with ClickEvent. how many events exist? - in my case, i can remove it. i just need the root which needs to be closed by the button

[rotund token]

https://docs.unity3d.com/Manual/UIE-Events-Reference.html

[viral sonnet]

what i find really cool is how we can provide parameters for methods. that was usually not so great

[rotund token]

i just want this, but on a single page

instead of having to click through each individually

[viral sonnet]

yep, overviews are always nice. first time i reported a problem on a help page because these basic examples are kind of meh. without a forum post i would never have guessed RegisterCallback has a custom parameter. on 2nd view, rider would've shown me but i missed it

[rotund token]

honestly i didn't know

i don't recall having a case for it, but my UI use is pretty basic atm

learning these things is why i hang out on discord!

[viral sonnet]

how do you handle close button events now?

[rotund token]

i just change UI states

        {
            var state = this.GetSingleton<ClientState>();
            state.Value = new BitArray256 { [K<ClientStates>.NameToKey("host-game")] = true };
            this.SetSingleton(state);
        }

        private void JoinButtonOnClicked()
        {
            var state = this.GetSingleton<ClientState>();
            state.Value = new BitArray256 { [K<ClientStates>.NameToKey("join-game")] = true };
            this.SetSingleton(state);
        }

        private void OptionsButtonOnClicked()
        {
            var state = this.GetSingleton<ClientState>();
            state.Value = new BitArray256 { [K<ClientStates>.NameToKey("options")] = true };
            this.SetSingleton(state);
        }```

rest is handled for me by my UI management/state systems

switching states would close current panel

(i probably need to add a shortcut for this in the base class to clean up this api a bit - been ages since i've touched it)

[viral sonnet]

oh i see 😄 you circumvented the need for it

[rotund token]

yeah for example when that clientstate for options is set

my OptionStateSystem starts running

    protected override void OnStartRunning()
    {
        this.UIStateSet(K<UIStates>.NameToKey("options"));
    }

and changes my UI state to the options window

the menu knows nothing about my actual options UI

[viral sonnet]

it's a good system. i have the use case where i build several shops that include upgrades and every shop is another panel. so i don't have that 1:1 relation but a 1:n

building a system for each would be a bit of an overkill

how to kill productivity in 2 simple steps: install entities, install uitoolkit 🤣

[rotund token]

oh i'm definitely going to have sub panels and stuff at some point

i just haven't really built anything for it yet

these states are only used for top level panels

so i'm sure i'll have more regular back buttons etc at some point

[viral sonnet]

this whole UIToolkit thing has set my knowledge about making UIs pretty much to zero. but i think it's for the best and probably a lot better once i've figured it all out. right now it's a veeeery slow process ^^

[rotund token]

entities 2.0 experience

[elfin spire]

gitlab uh, you're in Europe?

[rotund token]

no

[safe lintel]

are we still waiting on worldspace for uitoolkit?

[rotund token]

yes - definitely a big limit atm

[rotund token]

gitlab provides a lot more free lfs space

i don't understand why anyone uses github for personal game projects

[elfin spire]

Since I was coaxed to using gitlab by a german I'm considering moving over, it's cleaner too

[rotund token]

been using it for years

the moment i hit the 1GB cap on github was time to move

[elfin spire]

I was told gitlab is mostly used by europeans, github by US folks

habits

[rotund token]

well i am neither

[elfin spire]

where from?

leaves australia

[rotund token]

i should probably use bitbucket (actually I used to)

(but yes australia)

[coarse turtle]

I think so, but iirc wasn't there an API to translate world space to screen space 🤔

think I was thinking of this method (can't remember if they've changed it): https://docs.unity3d.com/Packages/com.unity.ui@1.0/api/UnityEngine.UIElements.RuntimePanelUtils.html

[viral sonnet]

does UIToolkit even have some kind of update event? i have a simple (cost) label that turns red when player has not enough funds. i mean, it's better when i loop over every active cost, evaluate and set - still wondering if there's any form to quickly prototype.

[rotund token]

what do you mean by update event

on value change? a tick every frame? etc

[viral sonnet]

tick every frame, something that related to a MB update

[rotund token]

no - at least i'm not aware of anything

it's reactive

[viral sonnet]

no wonder it performs better. it disallows shitty code 😄

i already stumbled over this but i forgot where. i know it's possible to query all those active cost elements

[rotund token]

do you just mean .Q<T>() ?

[viral sonnet]

yep and then followed by foreach

these selectors are pretty cool

that only works with query _rootVE.Query("shopContainer").ForEach(shop => { });

untested but Q has no ForEach

[rotund token]

oh yeah Q returns single element

[viral sonnet]

yeah, right. man, i knew that at one point - i'm getting so much information overload haha - anyway, works great to select it like that. kind of cool to have batch processing

the selector isn't a full css selector, is it? wonder if shopContainer:visible works

have to find that out tomorrow. now it's time for bed 😅

[dense storm]

Hey, just got started with DOTS.

public partial class RelativeMoveSystem : SystemBase
{
    protected override void OnUpdate()
    {
        new RelativeMoveJob
        {
            DeltaTime = Time.DeltaTime
        }.Run();
    }

    public partial struct RelativeMoveJob : IJobEntity
    {
        public float DeltaTime;
        void Execute(ref Translation translation, in Rotation rotation, in VelocityComponent velocity)
        {
            translation.Value += math.mul(rotation.Value, velocity.dir) * DeltaTime;
        }
    }
}

If I want this to run on multi-threaded jobs, I just change Run() to ScheduleParallel() and that's it?

And I'm getting this warning message, not sure how to fix.

[rustic rain]

Yes. You schedule job

[dense storm]

Is this everything I need to do to create an efficient system? What about disposing? When do I do that?

[rustic rain]

Disposing of what?

You can optimize it further by using ISystem instead of systembase

Since you don't use anything managed here

[dense storm]

Nvm I just realized only NativeArrays need to be disposed off after its done

[dense storm]

Ok I'll give it a go

[gentle gyro]

Has anyone attempted to use the DOTS hybrid renderer on the quest 2?

[rotund token]

missing BurstCompile attribute

[deft pilot]

is this crazy? can you have a base type for a component?

ie/

public abstract FooBaseComponent : IComponentData
{
  public float Value;
}

public class BarComponent : FooBaseComponent
{
  public float OtherValue;
}

[rotund token]

no

well i guess if you used a class based component?

but 99% of the time you should be using struct components

and structs can't inherit

[deft pilot]

ahh right

structs can only inherit interfaces

[karmic basin]

Love you tertle 💟 Great value for the community 👍

[deft pilot]

what makes more sense in an ECS system for types of projectiles (tower defence game):

Have different components for different types of way a tower can fire: ie/ TowerProjectileDefaultComponent, TowerProjectileInstantComponent, TowerProjectileBeamComponent

where Default is like shooting a single projectile to a certain location that has flight time, Instant has no project just hits the enemy as soon as it fires and beam is well... a beam and hits whatever comes into contact first

Or should i just have a common component called TowerProjectileComponent that has a field that contains the type and I have one system that does everything?

or do i go even deeper and have the projectile built out of components

so TowerProjectileComponent just really only has data to build the Projectile

[rustic rain]

I'd say make some component of Default tower shooter
Which will have most standard settings for basic towers.
System will use it for shooting

In case you need something special, you'll use different component and different system

meanwhile Projectile is a whole different story

[deft pilot]

so package the whole shooting into a component

hmm

the problem is

there is no default

default is do nothing (i guess)

[rustic rain]

I meant that for towers

that shoot

[deft pilot]

thats defined by the user

[rustic rain]

projectiles each should have their own logic I think

I don't think having smth generic for projectile specific logic should be the case

again, aside from basic logic

[deft pilot]

for context - you build towers with components

[rustic rain]

like for example normal movement, from tower to target (like arrow)

[deft pilot]

each component changes what its suppose to do

so there isn't really a default behaviour

so you are given a base tower that literally just does nothing

and upgrades you drag onto the tower to give it abilities

like to shoot

(and how it shoots)

only came up with this idea cause testing out ECS

[rustic rain]

ECS is composition

not inheritance like OOP

so you should probably drop any good practices while thinking in ECS world

in ECS logic will be hardcoded 99% of times

[deft pilot]

it is hard coded thou?

there is nothing "inheritance"-cy about this

[rustic rain]

then maybe I didn't quite understand you well

[deft pilot]

so consider an entity

that gets lets say afew components added to it

systems act on those components driving functionality right?

[rustic rain]

systems act based of queries of required component

they will go through all entities that match query

[deft pilot]

thats a low level overview

[rustic rain]

regardless of how many components you added

[deft pilot]

yes

so you agree right?

at a high level thats how it works

you could literally replace "driving functionality" with what you said

if you want

i just didn't think i needed to go that indepth for that

[rustic rain]

if you want "replacing" functionality of logic

Unity ECS has write groups

which can filter entities if they have components that belong to chosen write group

[deft pilot]

i'm not asking for help on how to write this on a low level

i'm trying to figure out what the best way to tackle this is

cause it depends on how i want to separate out the responsibilities depends on how i write my systems

which was basically the question

anyway dw about it i'll just think on it maybe i'm not wording the question right?

thought i was as clear as i could be in the initial question

[rustic rain]

Nah, it's just me not being smart 😅
If I wanted to make TD I'd first figure functionality categories
Tower shooting | Projectile movement | Unique projectile logic

so most towers will have most basic shooter, smth like CD based with different settings
For some unique towers, that need to shoot differently (for example based of player input) - their own component, which can also be shared among different types of towers

[deft pilot]

but you are under the assumption there are tower templates right?

[rustic rain]

for projectile movement I'd just implement logic for each kind of movement required. Linear, drop from sky, make some zig zags and etc

[deft pilot]

would you make each type of movement a component or a value in a component?

[rustic rain]

component

obviously can store it all in one though

but that would be pretty messy

[rustic rain]

not really

I'm just imagining old TD game

[deft pilot]

fair enough

[rustic rain]

what was it called... hmm

it's from WC3

[deft pilot]

elemental td?

line tower defence?

dice td?

[rustic rain]

Element

yeah

[deft pilot]

yeah so traditional tds

have a set tower

and set upgrades

[minor sapphire]

There is no one size fits all solution of course, you just need to think about all your requirements up front. Projectiles can move , maybe they can pierce, when they hit or end what do they do... Damage, debuffs, these things can be aoe...?

Imagine now how you want to configure a projectile with these things in mind, think of how that data will drive through into systems.

You may have a debuff system here, a damage application system, the concept of damage instances, how to determine damage instances (aoe? Immunity?)

There are many ways this problem could be broken down

[deft pilot]

yeah

[minor sapphire]

Imo start with understanding your requirements, then draw up a mock config for yourself

The problems will start revealing themselves

[deft pilot]

basically yolo'd in and had to refactor so many times so its why i've stopped and considered this question

[minor sapphire]

In my company I'm the diagram guy now cause thats how I figure out how to proceed with a new feature haha

[deft pilot]

hmm so my main problem with the solution right now is

if you allow a user to "build a tower" - so lets say they add a projectile to it

and they add another projectile to it so lets say it fires with two different projectiles now

you can't actually do that

due to unique constraints on projectiles

since you can't make them a component

so i'd have to make them an entity

but then if i make them an entity how does other things the user adds to the tower effect the projectile

[minor sapphire]

You're revealing some requirements

[deft pilot]

lets say they add the on hit effect AoE as you mentioned

[rustic rain]

I rewrote one of my projects 5 times before I achieved my own zen in ECS 😅

[minor sapphire]

A turret can shoot multiple projectiles

So turrets spawn projectiles. Seems perfectly fine.

Why can't they shoot 2?

[rustic rain]

is that game rule or some technical bug you are looking to fix?

[deft pilot]

cause you can't add two components of the same type to a entity

so lets say we have TowerProjectileArrowComponent or something

you can't add that again

[minor sapphire]

Dynamic buffer

Add as many as you want

[deft pilot]

so TowerProjectileBuffer

that takes eh...

some base class of projectiles?

[minor sapphire]

I'm just saying, given you have your requirement, this is one possible solution

Some projectile config

[deft pilot]

right

projectile config

[rustic rain]

yes, you can't add same type of component, but why do you even need it?

[deft pilot]

that would make more sense

[minor sapphire]

Projectile settings

[deft pilot]

why wouldn't you if the user can author towers

[minor sapphire]

Thst can define appearance, audio, speed, aoe, piercing, etc. Oorrrr do it differently

Maybe you just give it an entity prefab

[rustic rain]

If you want tower to shoot 2 different projectiles based of some upgrades to it, you should probably make component set that allows for that

like mentiond dynamic buffer + config

[minor sapphire]

Sooo many possible ways

[deft pilot]

okay let me try this first

see how it goes

[minor sapphire]

Don't imo

[deft pilot]

i'll be back probably

[minor sapphire]

Start with thinking through other options lol

[deft pilot]

this makes sense however

[minor sapphire]

Just jump on first suggestion

[deft pilot]

it fixes most of the problems

i've had

[minor sapphire]

Ok up to you lol

[deft pilot]

i can't think of any edge cases right now

[rustic rain]

just do smth first kek

[deft pilot]

anyway thanks guys @rustic rain @minor sapphire

[minor sapphire]

Maybe you have a 'firing' config

Thst references projectile prefabs

Defines what you shoot when

The projectile entity has lots of components on it to define its behaviour

[deft pilot]

yeah that will be driven by the dynamic buffer config

rather then a prefab

[minor sapphire]

Ok enjoy 😄

[deft pilot]

config will be created by an SO i guess

[minor sapphire]

Go forth and compile

[deft pilot]

thats my train of thought so far

[muted star]

Hey everyone, where can I find the ECS pool table example scene on github? I heard that it somehow uses physics deterministically.

I would like to use deterministic physics but couldn't get it to work with my initial attempt.
The positions and rotations after the physics simulation are always slightly different between machines.

How I tried making physics deterministic:

protected override void OnUpdate()
        {
            ...
            
            fixedDeltaTime = (float)env.GetVirtualTickRate().TotalSeconds;
            physicsSimulationWorld.SetTime(new TimeData(elapsedTime, fixedDeltaTime));
            elapsedTime += fixedDeltaTime;
            physicsSimulationWorld.EntityManager.MoveEntitiesFrom(
                EntityManager,
                entitiesToTransferQuery
            );
            physicsSimulationWorld.Update();
        }

(physicsSimulationWorld is my only world where the physics systems are present)

[karmic basin]

Talking about this one I assume https://github.com/Unity-Technologies/EntityComponentSystemSamples/tree/master/PhysicsSamples/Assets/Demos/6. Use Cases/ImmediateMode

IIRC they use another world that they tick multiple times in the future to predict positions and collisions

and com.unity.physics is supposed to be deterministic between nearly every platforms yeah

[muted star]

Thanks @karmic basin I'll have a look.
As far as I remember MoveEntitiesFrom jumbles up the entity IDs non-deterministically,
so my suspicion is that this causes the physics systems to behave non-deterministically aswell.

[errant hawk]

any DOTS updates I should know about?

[robust scaffold]

1.0 has entered experimental within Unity. Not yet public yet. Probably gonna release within 2 or 3 months

[errant hawk]

Yeah, it's odd seeing trainers using 1.0 but no way to actually use them. My guess is sometime next month 1.0 will release.

[safe lintel]

well its company training, not end user training unfortunately

[errant hawk]

Hmm... I guess it's still good news that they're making actual training projects with 1.0

[viral sonnet]

i think unite will be the big reveal of 1.0exp. maybe even 1.0 and exp earlier

[balmy thistle]

What makes you say that?

[muted star]

Hi everyone, I don't think I can use Unity physics for my project.
I would like to simulate physics for up to 1000 objects every 33ms.
I need a deterministic simulation. I think this should be possible with PhysX or Bullet.
(I know ECS physics can do this aswell, but not in my case)

Do you have any recommendations for a physics library which would be a good match for Unity ECS?

2D physics would also be fine

[viral sonnet]

just speculation. unite seems like a good way to get many (new) eyes on it.

[rustic rain]

There are only 2 physics engines on Unity ECS: Unity's and Havok

[balmy thistle]

In what way is Unity physics not deterministic enough?

[misty wedge]

It probably wouldn't be deterministic enough for something like lockstep networking, maybe that's what he's trying to do

[karmic basin]

@muted star There's also a framework by DreamingImLatios if you're unaware might be worth a look

[viral sonnet]

afaik dreamings is deterministic. at least as far as possible, true determinism over several cpu architectures doesn't exist yet in unity

it would require a custom datatype. (has a name, but i forgot) photon engine uses that

[karmic basin]

Ooh I thought I remembered even Unity were pretty close to a 99%+

but maybe I remember wrong

[balmy thistle]

Well, there's physics determinism, netcode determinism, and cross-platform determinism which are all kettles of fish

Unity physics AFAIK is deterministic

[rotund token]

• on same hardware

[balmy thistle]

yeah that's why I called out cross-platform determinism

[viral sonnet]

has anyone tested out the performance of uitoolkit queries?

i'm not sure if it's wise to build systems that rely on querying each frame

[viral sonnet]

hm, i really have some conceptual problem with how updating should work best with uitoolkit. if you imagine a spellbook that has like a 100 spells, not all visible and they can turn active/inactive based on if the player has the resources. previously such a spellbook button would have a simple cost and Update method that turns the button either white/red. now with uitoolkit, hm, i don't really know. should i build data in parallel to the visualElement. one system that queries visible spellbookButtons and then updates from that (would require some kind of hashmap to get the associated data) - bunch of ideas for it but nothing really great

[dense storm]

How do I transfer data between GameObject <-> Entities?

[viral sonnet]

the simplest answer is, don't. at least make it one way only. that aside, the entry point is relevant. from a gameobject you can just query entityManager and the components. from an entity, it only really works in non-burst code and having a Transform on the entity. other MonoBehaviours are supported but not all so the question is bit broad to answer specifically

[rotund token]

i'm not really sure why you need queries outside of your setup

[viral sonnet]

to determine visible spellbook buttons. otherwise i don't even know what to update - can't update all

[rotund token]

spell X unlocked, set X spell visible

what you need is to determine when a spell is activated/inactivated

i don't think you should be querying your UI to determine this

should be part of your model

also i assume you would be populating this spell book dynamically from your game data off a template

so you don't even need to query the elements, you would be creating them and already have the reference you can map somehow

[viral sonnet]

hm, i'm very familiar with the MVC pattern. big difference there, i can have code in the view, which is nice and powerful. here it's just a bunch of xml tags. and what i want to avoid is creating more code - i don't feel like that's elegant at all

[rotund token]

hmm i disagree

i think you should be writing code to build yourself custom elements to encapsulate their behaviour

i think you should have a SpellElement for example

[viral sonnet]

and that spellElement is a monobehaviour or what do you mean with that?

[rotund token]

spell element

is a VisualElement

this is the kind of thing unity does if you look through their code

they build elements from a collection of other elements and attach logic to them

etc

[viral sonnet]

ohh i see 😄 well that makes sense

where are these examples from?

[rotund token]

just first thing i found in this project

searching : VisualElement

😄

[viral sonnet]

nvm yeah found it 😄

[rotund token]

they were from the Version Control package

[viral sonnet]

collaborate namespace threw me off

at least that solves the issue of where do i even store model data 😄

thanks man 😄 these are the kind of concepts that i wouldn't have found in the docs

[urban ore]

Is there any public list of changes coming in ECS 1.0? I just ran into a really odd situation around the BlobBuilder, and was wondering if there's going to be an update to it or if this is just the way it has to be?

[rotund token]

just fragments here and there

what was the issue?

[urban ore]

Well, it's not necessarily an issue? It's just that it's a disposable struct. So if you wrap it in a using statement, you can't pass it by ref into a helper function.

So in order to get it to work, I had to unroll the using statement manually into a try/finally. Which is just really odd, to me.

[rotund token]

ah yeah, classic c# limitation

[urban ore]

Yeah. There's a lot of words online about how disposable structs aren't really a good idea. But I'd guess that this is probably a struct because it needs to be in Burst or something?

What really caught me off-guard is that I was just using Resharper's automatic extract method functionality in my code while refactoring... and it broke it because it didn't pass it by ref into a helper function.

Which it obviously couldn't do because... disposable struct with a using statement... Just odd. Was wondering if there's anything for it in the next version. If there's no clear documentation yet I'm perfectly happy to wait!

[viral sonnet]

sooo, the dragon crashers uitoolkit sample uses monobehaviours as controllers (mostly individual) and for something like a progressbar (ability cooldown in their case) a simple Update routine. really not much difference here other than everythings decoupled from UIToolkit

the big question now, why the roundabout way and not just a link from a templateContainer/visualElement to a mono script

i still don't quite get it. the nice thing about UnityUI was that with a gameobject enable/disable the UI and scripts were disabled. now i have to manage them in some way and it's quite hard to build a generic system for it. if i have any cost label that should turn red when the player doesn't have enough gold/resource i need to handle every occurrence where a cost label could show up and toggle the script based on the visibility of the uitoolkit element. this form of decoupling is total nonsense - doesn't help in the slightest

(pardon my negativity as I'm figuring this out) 🙂

[wraith hinge]

this is true for high-throughput stuff, but if you’re not transferring very much data, it’s actually totally fine and indeed necessary to control things like cameras, audio, and anything else that may not have a dots-native interface. if nothing else, you can just have a static variable on a monobehaviour that you update from a system’s onupdate or similar, as long as it’s not bursted. (If it’s bursted you can still do it but it gets complicated.)

[rotund token]

sigh been putting this off for a couple of months as i knew it was going to depress me

but i finally tested entities 1.0 128 entity per chunk cap

and my effect system update time goes from 2.03ms +-0.1ms to 4.74ms +0.2ms

on 10mill idle effects

[rustic rain]

1.0 is here?

[rotund token]

no

soon™

[minor sapphire]

Ouch

[muted star]

Yea but that doesn't mean that we can't use different libraries when necessary

[muted star]

Moving entities between worlds seems to mess up the determinism of unity physics. Probably because the order of the entities keeps changing non-deterministically.

[muted star]

Theoretically it should be deterministic enough for lockstep networking as long as all machines are using the same CPU architecture. So Intel or AMD is okay but they all need to be either x64 or x32.

[muted star]

There is a library I was using up until recently called soft floats. It does what you're talking about.

[muted star]

I've already taken a look at his code and it his psyshok physics implementation seems geared towards his demo and not a universal physics solution 🤔

[rotund token]

hmm i think this should be deterministic (though i haven't tested enough to be certain)

moving entities between worlds though would definitely require the exact same tick on physics though

moving then would have to be done in fixed update

[muted star]

Do you mean fixed update from a monobehaviour?
This is how I'm doing it currently.

        protected override void OnUpdate()
        {
            ...
            
            Debug.Log("Moving Entities from Simulation World to physics World");
            fixedDeltaTime = (float)env.GetVirtualTickRate().TotalSeconds;
            physicsSimulationWorld.SetTime(new TimeData(elapsedTime, fixedDeltaTime));
            elapsedTime += fixedDeltaTime;
            physicsSimulationWorld.EntityManager.MoveEntitiesFrom(EntityManager, entitiesToTransferQuery);
            physicsSimulationWorld.Update();
        }```

[rotund token]

no fixed update system

what you're doing there doesn't look deterministic at all to me though

you aren't updating a fixed update potentially multiple times in a single frame

it's update could drift quite far away from what it needs to

hmm actually wait no

maybe that's fine

i can't tell if your fixedDeltaTime

is actually fixed?

you seem to set it every frame

implying it's not the same

but if you're using fixed step simulation group inside your world this doesn't matter

[muted star]

fixedDeltaTIme is always the same, I could have set it in OnCreate

[rotund token]

so you're passing a fixed time to a world

in a regular update?

aren't you just either ticking your world faster or slower than it should be

or did you remove FixedStepSimulationSystemGroup

[muted star]

I removed all physics systems from all my worlds except for the physics world

[rotund token]

yes but do those physics systems still update in FixedStepSimulationSystemGroup?

and if so why are you passing in a fixed step to the world from a non fixed update

[muted star]

I was under the impression they did when I called physicsSimulationWorld.Update() but I'm not 100% certain they do this every time.
Should I transfer the entities from a system in FixedStepSimulationSystemGroup?

[rotund token]

FixedStepSimulationSystemGroup manages the fix step for you

converting non-fixed time to fixed time

hence i'm unsure why you're managing this yourself

and it's doing my head in if it'll actually be a problem

i think it definitely could be if you haven't set the timesteps up to match

but i'm pretty sure it is. if you're updating at 100 fps (10ms tick) and fixedDeltaTime is 50fps (20ms tick) then you're passing 20ms into the physics update and updating every frame when it should only be updating at 10ms and skipping every 2nd frames

yeah the more i write this out the more wrong it seems. you're basically updating your physics world in Update with the wrong delta time

you can quickly test if i'm right/wrong
change your (float)env.GetVirtualTickRate().TotalSeconds to be x10 larger

if the simulation is the same i'm wrong

if it speeds up 10x i'm right

save me thinking about it

[muted star]

I can try to explain how my code works overall perhaps it will be more clear then:

1. Simulation world systems run.
   The last system in the simulation world transfers all simulation entities to the physics simulation world and manually updates it (once all long running jobs were completed).

2. The last system to run in the physics simulation world immediately transfers all simulation entities to the presentation world.

3. The presentation world copies all the data from the simulation entities to presentation entities.
   The last system in the presentation world immediately moves all simulation entities to the transfer world.

4. The transfer world's only job is to wait for the next message from the host.
   The host sends a message to all clients every 33ms in production. (This is the GetVirtualTickRate().TotalSeconds seen in the code above)
   Once the wait is over, the transfer world moves all the entities to the simulation world and the process starts from step 1 again.

TLDR: It's fixed lockstep

[rotund token]

hmm so it's lockstep
have you set your FixedStepSimulationSystemGroup to use the same FixedStep?

that said it shouldnt matter

it might just tick twice per frame

[muted star]

The physics simulation world is the only world with systems in the FixedStepSimulationSystemGroup

[muted star]

I use SetTime on the physics simulation world before I update it. Could it still update twice in a single frame?

[rotund token]

yes

FixedStepSimulationSystemGroup sets the fixed update to 1/60

so your physcis world is ticking at 16.67ms

if you're updating it at 33.33ms

it's going to tick twice per update you call

maybe once, maybe 3 times depending on float rounding

[muted star]

Could it be a solution to move all the physics systems outside of the FixedStepSimulationSystemGroup then?

(if thats even possible)

[rotund token]

easier to just set FixedStepSimulationSystemGroup.RateManager = null

so it stops running it's own fixed update

and will use whatever you pass in

[muted star]

okay I'll try that 🙂

[errant hawk]

is there a parallel way to do GetComponentDataFromEntity ? I need to access some nested entity components and errors get thrown when I don't use [ReadOnly] but I need write access

disabled parallel restriction, seems to work fine

[rotund token]

thats the way

[errant hawk]

I guess CDFEs are atomic internally when it comes to R/W?

[rotund token]

no

you are responsible for ensuring what you are doing is thread safe

when you turn off safety

you are saying that no other thread will be reading this element

[errant hawk]

when you mean element, you're referring to the CDFE or the specific component that is being written to?

[rotund token]

the specific entity/component

[errant hawk]

so why does the CDFE need to be a container-type?

I did some digging in the source code, and I couldn't find anything

[rotund token]

what do you mean by container type?

[errant hawk]

it's a native container

[rustic rain]

is it?

I thought it's just wrapper for index type

and r/rw access type

[errant hawk]

why else would it require NativeDisableParallelForRestriction for parallel r/w?

[rustic rain]

because that's what safety is for?

[errant hawk]

yeah that's the thing

[rustic rain]

if you aim to do parallel r/w on it, you can get into race condition

unless you know for sure you don't

[errant hawk]

if it's not a container type needed to be initialized from the system, why are we not able to initialize a CDFE from within a job?

[rustic rain]

because you need system version

when you write to component

you also write version

[errant hawk]

Well, instead of using a CDFE, why not just a utility struct that you need to initialize for each job?

[rustic rain]

? what utility struct?

[errant hawk]

Hypothetically

[rustic rain]

I don't get it

[errant hawk]

var ecs = m_Access->EntityComponentStore;
ecs->AssertEntityHasComponent(entity, m_TypeIndex);

CheckComponentIsZeroSized();

void* ptr = ecs->GetComponentDataWithTypeRO(entity, m_TypeIndex, ref m_Cache);
UnsafeUtility.CopyPtrToStructure(ptr, out T data);

return data;

that is what is being done to read a component from a CDFE

[rotund token]

because you don't have any safety?

[rustic rain]

yeah, reading is very fast

but I still don't get what you meant 😅

[errant hawk]

What Im saying is why does there need to be an individual CDFE for each type in your job when it could be just some utility struct that also has safety like a CDFE

[rotund token]

because safety is per component

otherwise your utility struct would cause a dependency on every component and every job

[rustic rain]

yeah, if system doesn't know what component you'll use

your job will be isolated from any other

[errant hawk]

this safety shit is starting to really annoy me lmao

[rustic rain]

I think that's what dynamic type handle does

[rustic rain]

but you can disable it for your specific use case

if your solution is thread safe

[errant hawk]

yes it's thread safe, hence why I find it odd needing CDFE in the first place

absolutely nothing other-than a single thread is accessing the specific component, and no other system ever will write to it

[rotund token]

but your job is running in parallel?

otherwise it wouldn't be an error

[rustic rain]

but how else would you declare for other jobs that your job is using specific component

what if other job wants to write to that component

[errant hawk]

It wont lol

again, these are nested entities

[rotund token]

yes but entities doesn't know this

you could well be giving it non-nested components for all it knows

[errant hawk]

it does since I cached them in a DynamicBuffer 🤔

[rustic rain]

what nested means?

[errant hawk]

Is that rhetorical?

[rustic rain]

no, I don't know definition 😅

[errant hawk]

child entity, exactly the same as GameObjects being children to a parent GameObject

[rustic rain]

ah, what makes them special then?

[errant hawk]

ok so to be more specific, I have a component called a VehicleBase and a buffer DynamicBuffer<WheelBase> that is of type WheelBase. There are x amount of wheels per VehicleBase and each WheelBase has a reference to an Entity (along with other data)

it works perfectly fine the setup I currently have

but

I always questioned the requirement of CDFEs (disregarding safety) for my specific use case

when I could in theory just read/write to a pointer of that specific component

[rotund token]

which is all cdfe is wrapping for you?

[errant hawk]

that's what I mean, why is it even needed per-type when a simple utility struct can be used instead

[rotund token]

because safety

cdfe is just a utility sturct

[rustic rain]

because otherwise you don't write it to dependency

[rotund token]

and provides caching for speedup

[rustic rain]

you can make your own clone

[errant hawk]

idc about safety because I am r/w to isolated components that are never accessed outside of a single thread

[rustic rain]

or use Enzi's unsafe implementation

that won't write to dependency

and can access any component

with internal access pretty much anything is possible

[errant hawk]

Pretty much all I want is a utility struct that allows me to r/w to any component in complete disregard for safety, while not having to specify for each type.

[rustic rain]

can do it yourself

[errant hawk]

I would, but idk how to actually access a component outside of a CDFE

[rustic rain]

just copy whatever [int] of cdfe does

it's pretty simple

considering internal implementation does it for you

[errant hawk]

I would but it uses an EntityComponentStore which is internal and I have no access to it. I really dont feel like reverse-engineering what the GetComponentDataWithTypeRW does

[rustic rain]

just get access

there's Assembly Reference asset

which gives you it

with internal access everything is so much brighter 😅

[errant hawk]

not sure what that even is or where to find it

[rustic rain]

https://learn.microsoft.com/ru-ru/dotnet/api/system.runtime.compilerservices.internalsvisibletoattribute?view=net-6.0

[errant hawk]

one thing I wont be doing is modifying packages

[rustic rain]

but it's totally fine

you don't modify package

you reference it in your own

and extend it from different assembly

Unity works with it

[errant hawk]

oh alright, I'll look into that eventually

[rustic rain]

create folder somewhere, create assembly reference asset
choose whatever assembly you want to extend this way
And then create class with that attribute for assembly

in attribute you can simply define string namespace

of your choice

sad part, that only works on initial developper

modders can't access it

BUT

modders have publicizer

which is even better, cause it makes even private fields public

kek

[errant hawk]

I imagine 1.0 will add more unsafe stuff

[rustic rain]

I only hope 1.0 will swap most private fields to internal

[errant hawk]

doesn't make sense if they only made NAs have unsafe alternative

[rustic rain]

allthough doubt

they want to add more unsafe

I have a feeling that at some point

they will even have some sort of safe wrapper on pointers

kek

[errant hawk]

well, I notice a lot of their methods are public, but struct/class is internal

my guess is they just want people to work in a safe context only for now

[rustic rain]

yeah

pretty much

[errant hawk]

so they dont get bombarded with issues/questions regarding unsafe context that unity devs dont want to answer right now

which makes perfect sense

[rustic rain]

But ppl demand more unsafe stuff 😅

[errant hawk]

so far the only issue with DOTS for me has been the physics engine lag that appears out of nowhere

and a few GUI glitches

[rustic rain]

not just lag...

it's glitching

somehow

in editor

[errant hawk]

yeah but they're nothing horrible

usually just a refresh or similar will work

On a more unrelated note: I'm finding Span<> to be a bit faster than a NA when allocated on the stack. Nothing massive, but a 0.1-0.3ms decrease is pretty good imo

[rustic rain]

how does it look?

any example?

I'v heard about it a lot

but never really saw

[errant hawk]

well, it's a stack allocation so it's always going to be faster than an array (which NAs are similar in preformance to IIRC)

plus, no need to manage memory as it does it for you

best way I can put it is "unsafe safe code"

better than NAs initialized in a job imo

[rustic rain]

I just ask for some screenshot of codeblock as example

in Unity ECS context

if possible

[errant hawk]

Span<ContactData> contacts = stackalloc ContactData[buffer.Length];

for contact solving for my wheels

[rotund token]

well you have limited stack

but yes it's just stackalloc with safety

[errant hawk]

and my ContactData is just a wrapper for some data:

        public readonly struct ContactData {
            public readonly WheelCollider wheelCollider;
            public readonly RaycastHit hit;
            public readonly bool collision;
            public readonly float force;

            public ContactData (WheelCollider wheelCollider, RaycastHit hit, bool collision, float force) {
                this.wheelCollider = wheelCollider;
                this.collision = collision;
                this.force = force;
                this.hit = hit;
            }
        }

[rotund token]

Alternatively unity has fixedlist etc

For stack allocating

[narrow scaffold]

Quick question:
When I want to represent a temporary state on an entity, for example a "poisoned" status effect. My first intuition would be to just add a component containing the relevant data to that entity as soon as the state is triggered. However I have read that adding and removing components during runtime is not the best practice performance-wise because of the chunk system. What would be the idiomatic approach for that problem? Should I just have all components that an entity might need in it's lifetime there from the start and activate/deactivate them by need?

[rustic rain]

It's totally fine if you do it not often (meaning, not every frame)

so feel free to add/remove comps

alternatively

you can come up with a system that does it without structural changes

but I bet, it won't be as extendable as components

[rotund token]

if you want a flexible system first question to ask is why is poisoned a special component

what about burning

or radiation

or acid

etc

are you just going to make 1 component for every effect? at end of day why are they any different

do the same thing, maybe slightly different tick rates and damage types but that's just data

[narrow scaffold]

Okay thats reassuring. I watched the interview with the v-rising developers and they said they rarely add or remove components on entities because of performance.
But you are right, for my example it would make much more sense to have a status effect component that can represent different effects if they are not too different

[rotund token]

(Personally my effect/condition system is completely static archetype. in fact my entire game is except for state components)

[rustic rain]

I assume you have dynamic buffer for DOTs?

[rotund token]

(but this is not easy to do without a lot of experience)

[rotund token]

i have effects, that's it

+1 strength is the same as 50 poison damage/second

[narrow scaffold]

Out of curiosity, how do you handle state transitions? Like actions that should only be performed when first entering/leaving a state?

[muted star]

I think I just found the reason for my physics not behaving the way I want.
I just added a system into the FixedStepSimulationSystemGroup of the physics simulation world for debugging purposes and it turns out Time.DeltaTime is always different.
For instance Time.DeltaTime was 0,0128766 even though I called physicsSimulationWorld.SetTime(0, 1000) before calling physicsSimulationWorld.Update()

[rotund token]

yes because FixedStepSimulationSystemGroup has its own fixed update

this is what i'm saying

[muted star]

In my ECS bootstrapper I did as you suggested
physicsSimulationWorld.GetExistingSystem<FixedStepSimulationSystemGroup>().RateManager = null;

[rotund token]

i only have Start/Stop state transitions for application level states

my individual actors don't really use any

i do have limited states

and you can see how i handle states here: https://gitlab.com/tertle/com.bovinelabs.core/-/tree/master/BovineLabs.Core/States

[narrow scaffold]

Oh thanks that is very helpful. Still trying to wrap my head around thinking in systems so every example is great!

[rotund token]

this basically turns a bitfield into state components

without having to know the components

supports both only 1 state or multi states

[muted star]

yes because

[rustic rain]

I wonder if there's a way to quickly generate wrapper for bits

so you can have struct size of 4 bytes with 32 possible states

[rotund token]

you mean like a bitmask of any size you want?

[rustic rain]

practically

[rotund token]

https://gitlab.com/tertle/com.bovinelabs.core/-/blob/master/BovineLabs.Core/Collections/FixedBitMask.cs

        private struct Bytes4 : IFixedSize { }```
define a struct of any size you want

or if you want some huge size, feel free

        private struct Bytes1337 : IFixedSize { }```

https://gitlab.com/tertle/com.bovinelabs.core/-/blob/master/BovineLabs.Core.Tests/Collections/FixedBitMaskTests.cs

[rustic rain]

nnnah, I just wonder about using bitmasks

for sake of keeping size low

for multiple boolean fields

[rotund token]

https://gitlab.com/tertle/com.bovinelabs.core/-/blob/master/BovineLabs.Core/Collections/BitArray.cs
also have a bunch of masks 8 to 256

with a lot of functionality

[rustic rain]

for example my vanilla game input most probably won't use over 32 bits

so why bother creating huge struct

of 16 bools

[rotund token]

i use a lot of bitmasks

very important for networking

[rustic rain]

how so?

oh

keeping size low

right xD

[viral sonnet]

ouch! that's wild

[viral sonnet]

stop saying that. 😄 it only uses no safety in one obscure method where no system state is available. normal usage still sets reader/writer dependencies.

[rustic rain]

I just meant that in a way, that you came up with it long before anyone else I know

[viral sonnet]

yeah but it still requires a [NativeDisableParallelForRestriction] which you were talking about

[rustic rain]

not if you don't implement safety

which is even easier

kek

[misty wedge]

What's the fastest way to create a List from an UnsafeList? Is there some way to just set the list's internal array?

[rustic rain]

yeah, there is a way to add through pointer

lemme check if I can find it

        public static unsafe void AddRangeNative<T>(this List<T> list, void* arrayBuffer, int length)
            where T : struct
        {
            if (length == 0)
            {
                return;
            }

            var index = list.Count;
            var newLength = index + length;

            // Resize our list if we require
            if (list.Capacity < newLength)
            {
                list.Capacity = newLength;
            }

            var items = NoAllocHelpers.ExtractArrayFromListT(list);
            var size = UnsafeUtility.SizeOf<T>();

            // Get the pointer to the end of the list
            var bufferStart = (IntPtr)UnsafeUtility.AddressOf(ref items[0]);
            var buffer = (byte*)(bufferStart + (size * index));

            UnsafeUtility.MemCpy(buffer, arrayBuffer, length * (long)size);

            NoAllocHelpers.ResizeList(list, newLength);
        }

        /// <summary> Resize a list.  </summary>
        /// <typeparam name="T"><see cref="List{T}" />.</typeparam>
        /// <param name="list">The <see cref="List{T}" /> to resize.</param>
        /// <param name="size">The new length of the <see cref="List{T}" />.</param>
        public static void ResizeList<T>(List<T> list, int size)
        {
            if (list == null)
            {
                throw new ArgumentNullException(nameof(list));
            }

            if (size < 0 || size > list.Capacity)
            {
                throw new ArgumentException("Invalid size to resize.", nameof(list));
            }

            if (size == list.Count)
            {
                return;
            }

            if (_resizeListDelegates == null)
            {
                var ass = Assembly.GetAssembly(typeof(Mesh)); // any class in UnityEngine
                var type = ass.GetType("UnityEngine.NoAllocHelpers");

                var methodInfo = type.GetMethod("Internal_ResizeList", BindingFlags.Static | BindingFlags.NonPublic);

                if (methodInfo == null)
                {
                    throw new Exception("Internal_ResizeList signature changed.");
                }

                _resizeListDelegates = (Action<object, int>)methodInfo.CreateDelegate(typeof(Action<object, int>));
            }

            _resizeListDelegates.Invoke(list, size);
        }

_resizeListDelegates
that should be a field in your static class

[misty wedge]

Where are these functions?

Or is this not part of e.g. collections

[muted star]

Hey guys, I would like to set the update rate of the FixedStepSimulationSystemGroup to something really fast, like 1 ms for instance.
At the same time I would like the deltaTime used by the systems within that system group to be 33 ms.

This is so I can manually call .Update() on the world every ~33 ms without having to wait up to another 33ms for the system group to be updated with the 'correct' deltaTime.

Is this possible?
If not the only other option I see currently would be to make my own version of the systems within that update group to fit my needs and to put them in my own update group.

physicsSimulationWorld.GetExistingSystem<FixedStepSimulationSystemGroup>().RateManager = null didn't work btw.
It would of course be ideal if the rate manager could be ignored completely and all the systems within the group be updated manually (without having to wait for a rate manager at all)

[rustic rain]

it's extension from tertle

[viral sonnet]

as this is very specific, take a look at the fixedsimulation group implementation and copy it

i think the code should answer your questions

[muted star]

okay, thank you!

[viral sonnet]

and yeah i would suggest setting up your own group for this

what you can also do is call .Update on the group every frame. basically ticking it yourself. (ah you already do that)

[muted star]

I didn't know that was a thing. Currently I call Update() on the world

[merry vapor]

Hey everyone. Started messing with ECS this weekend. Certainly liking the idea of it. But I'm also running into complications and all the usual setup issues. I'm thinking for the project in question I probably dont need URP either.

I really just need a subscene type setup. Thats what I was setting up with ECS for - to be able to break up levels into sections and toggle them on and off as needed.

Given I'm uncertain about URP and aiming for a pretty graphically simple game, do you think better off making a simple custom system to just activate/deactivate sections of the level as needed instead?

[viral sonnet]

yeah, it's pretty convenient. i was utilising this to implement a timelapse/fast forward for the game. just needed a custom group and called update on it once. the group itself called update on itself then a bunch of times which triggers the update of the systems. just need to handle scheduled jobs because those need to be completed before the next update call

[dense storm]

To be specific, I only use dots for bullet movement and bullet collision detections in a bullet hell. But I am not sure how to inform GameObject’s component to take damage if the bullet hits.

Both bullet and player have entity representation

[viral sonnet]

on conversion you can use AddComponentObject and your transform. the transform will be then on the entity and can be used to GetComponent a monobehaviour. you either whitelist your manual MB components or make a hashmap/dictionary for faster lookup to not always have to call GetComponent

[dense storm]

Okay that sounds good. Will give it a try :) Thanks!

[devout prairie]

So i think i'm failing to see the wood for the trees here.. I'm getting this error:

Appears to be coming from BuildPhysicsWorld, complaining that the Projectiles job is reading from collision world

This is the projectiles job:

Which runs inside fixed step, before build physics world

I have another system which also runs before build physics world ( right after this one ) which also passes in 'world' as read only, but it doesn't throw any complaints about that one

[viral sonnet]

have you registered in onStart?

[devout prairie]

Yeah

[viral sonnet]

sometimes a dependency problem doesn't occur because it's finished before. at one point i had several when i only tested with 1 entity and no errors with 100k

hm, looks fine to me on first glance

[devout prairie]

Only difference with the other system which runs after this is, it's a job schedule rather than an Entities.ForEach that i'm passing world into

But yeah no complaints on that other job

[viral sonnet]

oh you sure its coming from world?

[devout prairie]

Seems to be stemming from build physics

[viral sonnet]

hm yeah

[devout prairie]

this is projectile system here:

TriggerFX as i say also uses readonly world to do i think an overlap sphere and has no issues

i had a similar issue before which arose when i changed ordering of certain jobs that lean on build physics, where i had an error pop up that actually wasn't an error, it was actually masking a different error from another job that was running

so it was basically impossible to debug with a lot of f***ery

[devout prairie]

I think it might be related to using ECB during physics step

[rotund token]

Does this only happen during double ticks?

[devout prairie]

and/or setting translations etc ( although the entities themselves are not physics )

[devout prairie]

i doubt it because it's really simple test case

Simply swapping the order or Projectiles system and TriggerFX exposes this error:

I wanted the projectile system inside step physics because although they are not physics ( i'm translating them manually ) i wanted smooth motion and accurate raycasts for hits

moving the system out to simulation system fixes the error, but i'm pretty sure last time i had this outside of fixed step it was missing collisions

although technically, i think if my fixed step is reigned in and running 1:1 with the frame that shouldn't really happen right

[rotund token]

https://gitlab.com/tertle/com.bovinelabs.core/-/blob/master/BovineLabs.Core/Extensions/ListExtensions.cs

[misty wedge]

Why can I do .WithDisposeOnCompletion on a NativeReference but I can't put it in a job with [DeallocateOnJobCompletion]...

[rotund token]

only native array supports DeallocateOnJobCompletion

(honestly they should remove this attribute)

[misty wedge]

What's the correct way to dispose from a job then?

[rotund token]

container.Dispose(Handle)

[misty wedge]

Does that not incur any scheduling cost?

[rotund token]

the same cost .WithDisposeOnCompletion does

[misty wedge]

Ah, I thought they did something trickier, I never checked

Thanks!

[rotund token]

if you don't want allocation/dispose costs

just don't allocate/dispose it every frame

just store in the system as persistent

[misty wedge]

That's what I usually do (now), but this is pretty old code where I wasn't doing that yet

I'm trying to add some functionality, I'll refactor it eventually (i.e. never)

This is also kind of weird code tbh, I should probably change it, but I'm not sure if that would make it slower. I still have no idea how long scheduling takes.

Basically what the job does is calculate a bunch of stuff so that I can then set tiles on a tilemap at the end. It gets a bunch of world chunks as an input, but instead of scheduling it as an IJobParallelFor I schedule a bunch of IJobs

The reason I did it this way originally is that I don't need a lot of the chunks immediately, since the enqueued chunks take up more than the camera's viewport, and chunks currently calculating are only forced to completion if the chunk enters the camera's viewport

And you can't complete only some of an IJobParallelFor

[rustic rain]

Wouldn't that be slower?

Rather than temp allocator

[rotund token]

why?

also it's not temp allocator if you're allocating it outside of the job - it's going to be TempJob

[rustic rain]

Huh

[misty wedge]

Anything to watch out for when using Nullable<T> in jobs?

[rustic rain]

Ain't that just wrapper for value and Boolean of whether it is null?

[rotund token]

TempJob vs Persistent

[misty wedge]

Yes

[rustic rain]

Huh

[rotund token]

main thread difference

[rustic rain]

What's the difference between tempjob and persistent then?

[rotund token]

0.43ms for allocating the 1mill element tempjob every frame

vs 0.006ms on the system re-using persistent

[misty wedge]

I thought it was where the memory was taken from

[rotund token]

allocation speed

[rustic rain]

I see

[misty wedge]

There's a really good and interesting article on it, although I'm not sure how much of it is still relevant: https://www.jacksondunstan.com/articles/5406

[rotund token]

allocation speed

if i just change persistent to allocate / dispose every frame

[viral sonnet]

so, tertle. your 128 entities test is pretty concerning

you should bring that up in the forum

how did you do the test to get 128 entities cap? pad the data?

or manual job queuing with 128 length?

i still have hopes this is opt-in and not a general rule now

[rotund token]

i've already brought it up privately

[rotund token]

yeah just added a 53 byte component

[viral sonnet]

oh you have dat connections 😄

i see, is padding a fair test?

[rotund token]

i don't really want to put fuel on fire and create speculation on a forum post, I was hesitant to even bring it up before 1.0 released without testing

[rotund token]

why not?

this component never queried

it wasn't an existing component, just a new one

[viral sonnet]

ah ok, then it's fair

[rotund token]

imo it represents the empty space in the chunk that will exist in 1.0

[viral sonnet]

yep

let me know if you get any kind of response 🙂

[rotund token]

i couldn't say, but i don't expect anything to change

at least not near future

not expecting them scrap 2 years of work ^_^'

just because i'm not excited on enable components i think a lot of people/studio will use them a lot

[viral sonnet]

well, if they hardcoded this i don't know what to say ...

archetypes and chunks should support both forms. uncapped and capped with enabled/disabled comps

[rotund token]

i'll probably just do my best to find use for these 50 bytes

maybe add support for my effects to do multiple things using a buffer instead of 1:1

etc

[viral sonnet]

will it perform better then? 🙂

[rotund token]

than it currently does? doubtful

than it will, probably

i could significantly reduce entity count if instead of a piece of gear creating 6 entities for 6 stats

just creating 1 entity that can set 6 stats

so i should be able to make up a be bit of the loss

[viral sonnet]

good example why development shouldn't be internal for so long. but then again, we voiced our concerns when it was first talked about so. i dunno. bad move (in case there really is no uncapped chunks anymore)

maybe the results are better with the new job system. after all, this is where the most overhead comes from

[rotund token]

i understand the reasoning cort posted on forums

it's more just sadness for potential

in reality i don't think it will change much

[viral sonnet]

yeah and i see the point, it just can't be generalised so much to ignore every possibility

like sure, the big gameplay entities have a massive size but what about the others

specialized ones more importantly

so now we have hardcoded chunk size and hardcoded entities cap. ugh

this is clearly not built for the future lmao

[rotund token]

while I would really like small chunks especially with unmanaged shared components coming in 1.0, i wouldn't that far to say it's not designed for the future

[viral sonnet]

you see chunk size of 16k and 128 entities cap in 5-8 years? it's not even appropriate for todays hardware.

[rotund token]

sure why not

they're nice numbers

[viral sonnet]

640k memory was also a really nice number 🙂

[rotund token]

128 is simd register

i can see it slowing down if you increase this value

and 16kb fits nicely in cache

now would it be nice to have dynamic sizes per cpu

sure

is it needed right now? i hardly think so

would also breaks determinism which is something unity seems keen on maintaining

[viral sonnet]

anyway, as i said, they should have made effort to make these numbers tweakable very easily. especially chunk size was one of the first major questions i saw. and as our tests show, more fitting chunk sizes can mean drastic improvements. that goes for any archetype and entities should be able to make every archetype configurable how much data it allocates.

the 128 simd register is only relevant with the enabled/disabled feature, right? the 16k was stated that it's not about caches. sure it fits in nicely, cache sizes will be bigger at one point though

and to have optimised a feature so aggressively over a complete framework only to be optional is ... not good to put it mildly

eh i'll shut up now. maybe it's really optional and not even a problem. i'll wait and see what's what

[rotund token]

i should say that in 1.0, systems are [alwaysupdatesystem] by default, and you have to sign up if you want to require matching queries for update. this is because everybody was always going insane because their systems weren't running and they forgot to say alwaysupdatesystem.
been thinking about this for a couple of days

this is such a scary change (for old code)

i think this kills any hope of updating our existing project to 1.0

though the chance was already little to none

[viral sonnet]

doesn't it just require to flip? [alwaysupdate] -> remove - no attribute add [requirequery] or whatever it will be called

but yeah, it is quite the change. probably made because of beginners. i certainly will have to get used to it.

[rotund token]

not exactly

if you have 2 queries

previously either one having entities would make system run

you can't just make requirefor for both queries otherwise it would only run now if queries had entities

not either

you have to do early outs on each query individually on what they use

[viral sonnet]

are you sure that will be the case with the new attribute? i think with the new attribute it will work just like the current behaviour

and you have to sign up if you want to require matching queries for update
basically this