#archived-dots

1 messages Β· Page 285 of 1

rotund token
#

As for your initialise seed you could either use a tag component to set it up or my new favourite thing, DidOrderChange to do it without a archetype change

solemn hollow
#

hah that was exactly what i was hoping to hear πŸ˜„ but how would you make sure it only ever triggers in first frame? those entities might move in chunks

#

man i have to say when you cache the handles for multiple components its quite some boilerplate

gentle harness
#

I am simply trying to change the scale of an entity with a spherical physics body. Anyone know how to do that?

#

I've tried

  • Setting LocalToWorld
  • Using Scale component
  • Using NonUniformScale component
safe lintel
#

physics doesnt respect the transform system

#

you have to modify the collider blobasset directly

gentle harness
#

There

#

there's only one instance of a blobasset at a time, right?

#

So if I want many shapes with different sized I'd need to create many blobassets?

safe lintel
#

theyre shared unless you set unique

#

in the authoring settings

rotund token
solemn hollow
#

hmm but that means im checking the whole chunk even if i only spawn one entity. i think i prefer the structural change here

rotund token
#

You can check 10 million entities in 0.17ms - literally what I've been testing recently

#

And it's only entities in that chunk

#

You can run the whole thing in ISystem due to no command buffer so main thread is faster

solemn hollow
#

i see. but you need to do the check for every component. while i need to only remove the Initialize tag once in any system. the other systems actually setting up the component can be ISystems

rotund token
#

Unless you are doing archetype changes every frame I personally don't see benefits of using a tag.
My real issue with tag is sure it's fine for this one case, then you add another case, and another. Next thing you have 10 archetype changes due to initialisation

rotund token
solemn hollow
#

each of your systems beeing responsible for setting up one component need to check all components of that type in the chunk.
with one shared tag on an entity that is only there for one frame each system only needs to setup exactly that entity with the tag and its component. and i only have 1 structural change no matter how many systems do some setup

#

better yet. if i spawn alot i know that all newly spawned entities to setup are contained in seperate chunks that i can iterate over quickly

rotund token
#

Yeah see that 1 shared component is already a deal breaker to me

#

I'm now coupling independent libraries together

solemn hollow
#

id have that in some core library

rotund token
#

And it still messes with systems setting up chunk components

#

I have to rebuild all the chunk components when the archetype changes

solemn hollow
#

yes i always wondered about how to manage that efficiently. guess you need to avoid practically all changes with chunk components

#

one reason i dont use them yet. they dont feel complete yet.

solemn hollow
rotund token
#

Components

#

Apart from that there is no communication

solemn hollow
#

but then how do you not couple?

rotund token
#

Seperate assemblies strictly prevent me from having circular dependencies

#

Entry piece of code I write, I write like it was for the asset store

solemn hollow
#

yes i do that too (not for asset store but assemblies i mean). but still my effects need to know about the speed attribute which also is read by my AI etc

rotund token
#

My actually ai is generic and standalone

solemn hollow
#

yes but how do you get the speed or health data into the AI? i needed codegen to do that

rotund token
#

It has no knowledge of a specific stat

#

That is purely setup in data

#

By a graph

#

If stat43 < x do y

#

The only thing that makes stat43 speed is some designer called it that

solemn hollow
#

but somehow that stat43 variable needs to be written to from outside the AI right? eg. sloweffect or anything on which the AI should react. how to not couple this?

rotund token
#

I could give you my Ai system right now

#

And you could use it

#

Without making a single file change in the assembly

solemn hollow
#

dont cause mine runs great too πŸ˜› i just need an explaination on how to decouple those Domains completely

#

i just dont get how you get data from ECS into the stat43 or out for that matter without knowing where to read from

#

the designer somewhere has to explain that that stat43 in the AI is the same as the speed stat on the Unit or sth?

rotund token
#

finally not on phone, can type easier

#

gone a bit off topic here

#

on the original idea, don't get more wrong using a tag component is fine for initialization. i was doing it until like a month ago.

#

my issue with shared initialization component is which system adds it?

#

or does all systems add it

#

if you have a standalone system adding it

#

well now when i write a unit test for any of these systems, i now have to initialization a second system

#

and then a 3rd for the ecbs

#

my single system is now no longer a nice little module that takes data, does work, spits out data

#

and that's basically what my goal is when writing any system. a nice little module

#

is it a requirement to make a good entities project? not at all

#

i've spoken a few times i enjoy trying to write idealistic software for myself

#

in reality this is often not feasible

solemn hollow
#

yes makes sense. i do want those little contained systems too but i do see a benefit in "infrastructure" systems handling one simple thing once instead of many systems having to do that thing. purely a design choice i think. in testing id would always have that system run in any test i guess. (but my testing experience is very limited)

rotund token
#

i'm a bit obsessive about architecture. i haven't written any code in 3 days but i've written 5 pages of ideas

#

on how to try get my condition triggers working really nicely

solemn hollow
#

the ones for effects?

rotund token
#

and i still haven't come up with a plan that i'm happy with

rotund token
solemn hollow
#

what type of conditions did you have in mind?

#

in my system i just got time and lifetime conditions. so stuff like trigger once every second. or trigger only once or trigger when destroyed.

rotund token
#

on hit
on block
full health
low health
daytime
frozen
burning
stat == y
standing in lava
etc

#

the problem is, everything

#

these don't have to all be handled the same way, ones that are on the actor vs world etc would be handled different

solemn hollow
#

so basically a buff on the character that returns damage on damage taken for example?

rotund token
#

these are the conditions

#

any (multiple) condition can be paired with an effect

#

effect might be
increase stat
deal damage
heal self
chill
apply debuff
create another effect
etc

#

so if you have the on block condition implemented

#

you can now pair it with
heal self on block
chill enemies on block
create burning ground on block

#

etc

#

then you implement low health condition now you have
heal on low health
chill on low health
create burning ground on low health

#

the idea of this whole system is anytime an engineer implements a new condition or effect

#

designers have whole new set of combinations they can apply throughout the game

solemn hollow
#

yeah we are back to poe πŸ™‚

rotund token
#

since everything uses this system

#

(well to be fair this is based off for honor talk but i'm just more familiar with poe mechanics so i think about those for conditions/effects)

solemn hollow
#

thats what i still need for my abilities (basically passive abilities are the same as your effect + condition id say). my effects dont trigger anything by themselves but are just something thats caused by abilities.

rotund token
#

the one thing i'm trying to avoid is condition query events

#

as i might have significantly more conditions than events

#

so ideally i'd like events to modify conditions

#

(and well i'm trying to avoid what people call 'events' as well but it's still an event in game)

#

a whole lot of going in circles

solemn hollow
#

is a condition a component on the effected entity?

rotund token
#

no that's the thing

#

my effects are separate entities

#

we've run into a lot of problems at work putting too much on the actual actor entity

#

getting to the point where we have chunks with 3-5 entities max

#

so i'm trying to avoid this

solemn hollow
#

yes but that also makes it hard to query a combination of effects

rotund token
#

(also it makes things a lot faster to calculate when i can fit 500 entities in a chunk and have 0 structural changes ever)

solemn hollow
#

i was thinking about splitting it the other way. having an extra entity per actor containing all effects

#

same with AI. but then again its more random access to read data from actor

rotund token
#

i was going to make each effect share a chunk with other effects targeting same actor

#

but it's a huge memory waste so i ditched that early

#

really infeasible for any moderate scale

solemn hollow
#

true. its a hard problem to solve with queries.

#

maybe having a hashmap containing all effects for every actor is ok

rotund token
#

also my effects strongly rely on change filters to not explode

rotund token
#

i think if i went this approach i'd just sit on a buffer instead

#

oh wait you said effect

#

not condition

#

nah that's not possible for the scale i'm trying atm

#

this system is basically an entire game

#

it eliminates like 50% of other systems a game would generally implement

solemn hollow
#

i know that feel :S. My AI controlls playerinput. whole game is basically AI doing stuff and you watch

rotund token
#

i'm falling a bit deep into OOP thinking atm

#

but i'm legit considering implementing like an unmanaged event

#

that conditions subscribe to >_>

solemn hollow
#

:S

rotund token
#

this is me in desperation

#

having ruled out everything sane

#

that said, the last time i went into insane with shared pointers

#

turned out amazing

solemn hollow
#

man i wonder if that would be a case for that one article talking about some sets of components you can query in ecs systems

rotund token
#

i do have some like, really solid ideas for this

#

but oddly enough i've ruled most of them out due to memory costs

#

though i probably shouldn't

solemn hollow
#

damn im not finding the article :S

#

cant remember which ecs it was

#

there it is

#

i think that relations feature would be a great solution for this case

#

though i have no clue how it is implemented and thus could be really slow. it would be just such a nice way to be able to get all effects related to an entity by a query without having to maintain a buffer with references and using chunksize for it.

#

well im off to bed.

viral sonnet
viral sonnet
rotund token
#

i'm only actually designing for ~3 cases atm

viral sonnet
#

like daytime, frozen are static effects with a duration. on hit is a trigger

rotund token
#

target, source, world

#

he just wanted to know what type of conditions i intend to support

viral sonnet
#

just saying, i've those trigger types split from the rest that are defined by effects. all triggers are bascially just an integer mask

#

having world is a weird abstraction? isn't that just another source with an infinite duration. like walking into a cold cave or smth which would be triggered by a physics trigger or smth

#

i've found that making hashmaps for easier lookup is quite nice. the problem is with effect amount and processing and how that scales. like, I don't feel comfortable for a game having 1000s of different effects and special code. most have a very generic subset that's used over and over again

rotund token
#

condition on this effect is it's night time, gain damage

#

most conditions are reactive like player takes damage or blocks damgae

#

but you can combine them

#

at night time + when you block damage, heal

#

(i support 8 conditions per effect simply due to bitmask, though i can't think of a use case for that many)

viral sonnet
#

your night time example just seems like a set of effects that gets applied at different times. a night/day system could handle that

rotund token
#

it could be a stat on a piece of gear

viral sonnet
#

i see, yeah ok, then it's an effect with a time condition.

#

i've several conditions in place. there can be really annoying effect types, like, gain 5% crit for a certain spell

#

pretty much all the time I can squeeze them in the code. but that's what I was meaning previously, at some point you can have a LOT of checks. another way of handling this was having an array of flags (basically they define which checks are made) which are iterated and the flag is just used for a switch. that way I can reduce the conditional checking to a minium

rotund token
#

at some point you can have a LOT of checks
yeah see that's what i'm trying to avoid

#

i don't want my conditions to check the state/events

#

i want the events to write true to the conditions

viral sonnet
#

just been on the balcony thinking about your approach. really interesting because we come from different angles, yours is effects, mine is abilities and if you look at them from another angle they are quite the same

#

like an ability is a set of effects - just the execution phase is different

#

i hope we get a proper 2021 soon. usually today is their release day. fingers crossed. i've gone back to 2020.3

rotund token
#

my 21 been stable as (now that i know the bugs to work around =D)

viral sonnet
#

the build workaround made me tilt πŸ˜„

#

and those UIElements bugs in BuildConfig and Inspector

rotund token
#

funny enough those UI bugs in build config were there in 0.17

#

i had really hoped they were fixed

viral sonnet
#

what? I never had them. you mean where the sub menus didn't appear and filters weren't working?

rotund token
#

yeah

#

i think it might have been only in 2022 or something and they backported the uielements updates

#

but i definitely had them a long time ago in 0.17

#

and i have a bug report open for it somewhere

viral sonnet
#

i never noticed in 2020.3

#

22 worker threads is fun to look at

#

as my pc died on sunday, i'm still looking at the cause for the slowdown on having many entities. like my single schedule job just doubles in time. i just find this not acceptable.

rotund token
#

its only 2021

#

its the new versions of uitoolkit that breaks it

viral sonnet
#

ah, and as your effects live as entities. this is where it differs to my abilities because creating short-lived entities has not been great

#

(would make so much easier though)

remote crater
#

Solve the problem, indeed addressables does not copy the data to the proper external build directory.

#

I have to find the error files, windows search em in my Unity Project Windows Explorer, and copy the whole directory to...

#

A directory under my external build directory, specified by a path.

#

It's super weird, complicated and arcane... It could be fixed easily tho.

rotund token
late mural
#

due to there being currently (as far as i know) no way to make entities interact with unity terrain, what would be the best tools to convert a unity terrain into a mesh?

#

unity terrain is just a height map right? Assuming there is a way to convert the height map into an image then i could just use blender!

#

apparently i just uninstalled my gpu, while trying to update my drivers, so if anyone replies i probably wont see it for a while

rotund token
#

i haven't used unity terrain in a good 5+ years but you can just export it as a mesh right?

late mural
#

or an asset store plugin which costs 40 bucks

rotund token
#

this doesn't work?

late mural
#

ooh ill try it, after i get my gpu and monitors not dying

#

ive already lost 1

#

and this 1 is now turning pink

#

good job windows

#

apparrently i cant plug my monitors into my usb ports, cause i uninstalled usb3 to dvi, somehow while updating my drivers

trim badge
#

Hi guys. I haven't use dots for a while. And I'm starting a new city builder project which have hundreds units. Should I start with dots?

severe nacelle
#

if so, yes

#

dots helps with having hundreds of entities running at the same time, it's not the only solution though

trim badge
#

I leant the basics. I'm worry if the whole structure might change when Entities 1.0 releases.

late mural
trim badge
late mural
#

np, i hope it goes well for you!

rotund token
#

@viral sonnet what do you think for writing my global conditions

# 
    {
        public bool Value;

        bool IGlobalCondition.Value => this.Value;
    }

    public struct IsNightTimeCondition : ICondition
    {
        public byte Index;

        byte ICondition.Index => this.Index;
    }

    [BurstCompile]
    public struct NightTimeConditionSystem : ISystem
    {
        private GlobalConditionImpl<IsNightTimeCondition, IsNightTime> impl;

        public void OnCreate(ref SystemState state)
        {
            this.impl.OnCreate(ref state);

            state.EntityManager.CreateEntity(typeof(IsNightTime));
            state.EntityManager.CreateEntity(typeof(IsNightTimeCondition), typeof(ConditionActive), typeof(ConditionAllActive));
        }

        public void OnDestroy(ref SystemState state)
        {
        }

        [BurstCompile]
        public void OnUpdate(ref SystemState state)
        {
            this.impl.OnUpdate(ref state, default);
        }
    }```
#

simplifies to the above per condition

#

Obviously there is no job included above to write to the IsNightTime singleton. as long as this is adheres to strict change filtering rules it'll only trigger condition changes on night time changes OR if a new condition is created or destroyed (just for that chunk)

#

setting up the index is automatically handled by authoring (unique is not unique per condition and will vary depending on number of conditions on effect)

#

concept works

unknown.png
#

1 million conditions / entities overhead (every frame early out)

unknown.png
#

time to write 1 million events on night change

unknown.png
solemn hollow
#

Not enzi but i like it. i dont fully understand how this byte Index works yet

rotund token
#

oh my conditions are stored in a bit field of length 8 (1 byte)

    {
        public BitArray8 Value;
    }```
#

each condition is given an index in this bit field

#

when bitfield == 255 (e.g. all conditions are true) then the effect triggers

#

for this specific case, bit 1-7 are always 1 and bit 0 is toggled since there's only 1 condition on this test effect

#

but yeah, no query, no Lookup, no ComponentFromEntity, no hashmaps just dependent on filters

#

i'm liking it. doesn't really solve my individual conditions problem i've been fretting about but i needed to get something done

solemn hollow
#

ah i see. i was confused about "allConditions" somehow thought you have all possible conditions ever in there

#

with individual you mean something not global like nightime right?

#

something like low health on an entity

rotund token
#

yes

#

i have 6 pages of designs for this now without something i like

#

i really want to do something like this that requires no lookups but i might be asking the impossible

minor sapphire
rotund token
#

Pen and paper

minor sapphire
#

that's even more hardcore πŸ˜„

rotund token
#

This way I can throw them out instead if maintaining documentation when done

minor sapphire
#

ah yes well played

rotund token
#

I miss my old 2 in 1 laptop

minor sapphire
#

I'd use my tablet more if I weren't sitting on a couch haha. Maybe when I move πŸ™‚

#

Well it's a bloody Christmas miracle. My daughter is asleep early enough for me to have enough motivation left to open unity

solemn hollow
rotund token
#

Yes but it's a bit more complicated than that

#

I'd need memory blocks for every target * potential effect

#

And then each block would need to also now what effect chunks need to update

#

Every condition of a type in a chunk might be looking at a different entity

#

there is a good chance i will use SetComponent for this case

devout prairie
#

So i've been following your convo and understand the problem, just trying to follow how the solution you've written above works.. could you explain how that's laid out as if i was 12?

#

I like the bit field idea btw

#

So would there then be a conditions entity for every entity that requires conditions/effects?

rotund token
#

so yes, i'm writing a basically entire game unified effect system

#

effects are basically
condition -> action
actions are broken in 5 categories atm

  • modify stats
  • enable/disable capabilities (put things on cooldown)
  • add tags (stunned, invisible etc)
  • remove modifiers (poison etc)
  • create new effects
#

these can be applied to anything, equipment, skill trees, environment triggers, etc

#

and are intended to be extremely data driven

#

every time a new condition or action is implemented it can be used with all the existing ones added huge amounts of gameplay

#

now i have the whole effect workflow working amazing with cooldowns, durations, setting stats etc

#

but i'm working on now is trying to efficiently trigger the conditions

#

the above code is for triggering global conditions (time of day etc)

#

(each effect can have up to 8 conditions hence the bit array of 8 - though I think that amount will be highly unlikely and 90% of time i'd expect only 1 (or none) and rarely more than 2)

devout prairie
#

..just chewing on this over coffee, bear with me.. πŸ˜›

#

yeah that seems like it could apply to basically any aspect of cause/effect or events in a game

#

could potentially be a really efficient pipeline for anything that affects anything based on conditions right

rotund token
#

yes, my objective is to basically use this for everything

devout prairie
#

so you're using change filtering to avoid having to query for these things

rotund token
#

pretty much

#

though my implementation would work nothing like theirs

devout prairie
#

yeah i think i actually watched a good portion of that

#

just curious for this line:

unknown.png
#

why do you not just go var value = globalCondition[0].Value;

#

i mean i guess in the case of multiple globalConditions ( atm only one ) you would be reinterpreting the whole array to bools, so i guess that would be simpler

rotund token
#

probably because i copy pasted from the below reinterprets

#

and forgot to change it

rotund token
devout prairie
#

so say you have a million entities and 500 of them decide to fire the 'too hot' condition or something, how would they update the system

#

would they create a bunch of too hot entities and the system would read them in and parse them

#

or would they write to a fixed array of too hot bools or..

rotund token
#

well yeah this is the problem i haven't solved yet how i want to do it

#

and i've been investigating for days on how i want to do it

devout prairie
#

Yeah

rotund token
#

it's not /that/ hard to implement the problem is i'm testing this at

#

10,000,000 conditions/effects

devout prairie
#

It's a tough but interesting/important challenge

rotund token
#

my assumption is that there will be a lot more conditions than actors

#

and these type of events (hot, on hit, etc) will not be that frequent

#

therefore the events should write to the conditions

#

instead of conditions reading the state

#

but i just haven't decided the best way i want to handle this yet

devout prairie
#

Yeah

rotund token
#

i've already implemented 2 completely new patterns (at least i doubt anyone has ever used them before) while developing this

#

so i'm going a bit off the rails for this

devout prairie
#

Go for it imo

viral sonnet
rustic rain
#

public abstract partial class InteractionSystem<T> : SystemBase where T : InteractableBase

#

hmmm, I have such systems (that inherit)

#

I need to somehow create Dictionary of all systems that inherit from it and type T of that system

#

any idea how can I achieve that?

viral sonnet
#

with reflection πŸ™‚

coarse turtle
#

Scan the assemblies with reflection and grab all the systems that is a subclass of type T?

rustic rain
#

yeah, I got it, but now the other problem...

# 
            foreach (var interactableBase in fui.onFlyUpFinish)
            {
                var system = _interactionSystems[interactableBase.GetType()] as InteractionSystem<>;
            }

oh god, I hate OOP

#

but I couldn't find any other way

#

basically

#

abstract InteractionSystem<T> has abstract Process();

#

and I need to call it based on interactableBase type

#

looks like this

unknown.png
#

I hope I don't have to use reflection to call it

viral sonnet
#

after you have acquired the systems that match your needs you can mark them in some way, like RequiredForUpdate

muted star
#

Hello,
I don't want to interrupt the thread but I could use some feedback on something I'm currently stuck on.

I'm trying to implement a solution where
there is a simulation world which is updated slowly
It takes care of everything that is actually relevant to the game.

And the default world takes care of displaying everything in a pretty way.

To do this I would like to copy entities from the simulation world to the presentation world ever so often.

The only way I found to do this is using CopyEntitiesFrom() but the problem is,
that it doesn't replace the previously copied entities with the new copies.

So a ton of duplicates are created. What would be a good way to get rid of the duplicates and to preserve the new copies?
Could there even be a completely different solution to what I'm trying to achieve?

Btw moving the entities back and forth between the presentation world and the simulation world doesn't make sense in my case because that would prevent the presentation world to present the entities while the simulation world is calculating away at pathfinding etc.

rustic rain
rustic rain
#

why would you want to move huge data structures from 1 side of memory to another?

muted star
rustic rain
#

but that still would be a huge sync point

#

when you move entities

#

cause you can't copy entities without making sure nothing is getting written to them

#

which would be

#

CompleteAllJobs();

muted star
#

yes but the simulation world entities would be copied over at a low rate, like every 100ms for instance

rustic rain
#

but why?

solemn hollow
#

it sounds like you do a network game for 1 player lol

rustic rain
#

you will still have to run presentation every frame

#

or else nothing will be rendered

#

and your simulation world

#

will be limited to those same frames

#

and waiting until presentation from other world is finished

#

before next frame in Unity player even begins

muted star
#

yes but as I already said the advantage would be that the presentation world could have frequent sync points without disturbing the long-running jobs in the simulation world

rustic rain
#

but it's already like that

#

Simulation and Presentation groups are 2 different groups

#

which run one after another

#

it's like Update() and LateUpdate() afaik

muted star
#

so are system groups independent from each other regarding sync points?

rustic rain
#

simulation and presentation are

#

first runs simulation

#

only then presentation

#

after simulation is completely finished

muted star
#

Okay so if for instance the simulation system group just started a long-running job
and my custom system group, running right after the simulation system group, causes a sync point by creating a new entity.
Won't the simulation group jobs be forced to complete within this frame?

rustic rain
#

after simulation group? uugh

#

is that even possible?

muted star
#

okay maybe the sync point takes effect on the frame after idk. My point is the system groups aren't independent from each other when it comes to sync points.
If presentation group has sync point it's also a sync point for the simulation group.

rustic rain
#

I think we talk about different sync point, hehe

muted star
rustic rain
#

well yeah

#

and you shouldn't care about sync points in presentation

#

and after simulation

#

presentation is mostly managed anyway

#

so job stacking is impossible

muted star
rustic rain
#

sir

#

sync point is what is causing all jobs to finish

#

or some

muted star
#

yes

rustic rain
#

so best way to do it

#

in end of groups

#

at ECB

#

stack as many jobs as possible

#

and then run them all at once

#

utilizing as much threads as possible

#

after simulation is done, presentation comes in

#

and there is no job stacking there

#

all is run on main thread

#

because it uses managed code

muted star
#

sorry I still don't see how this could prevent presentation related systems from
causing simulation related systems to be forced to complete early and introduce lag into the gameplay experience but thanks for your help anyway πŸ‘

rustic rain
#

they can't, because they simply not run until simulation is finished

muted star
#

finished with jobs or with single-threaded code?

rustic rain
#

it's all tied together

viral sonnet
#

are you talking about procedural gen?

#

because there is already a streaming world. (utilized in the editor) what issue means is that you have a hard sync point when copying in the new rendermeshes

muted star
# viral sonnet are you talking about procedural gen?

No we were talking about how to have system groups which don't block each other with sync points.

On the one hand systems responsible for long-running simulation jobs (pathfinding for instance)
and on the other hand systems responsible for interpolating between simulation results with pretty presentation.

My idea was to have two separate worlds for simulation and presentation but apparently the goal can also be achieved without separate worlds according to Issue.

muted star
# rustic rain or some

In which case do sync points cause only some jobs to finish? Are there docs on this topic?

rustic rain
#

whenever you force run job that has dependencies

#

all dependencies will have to be completed first

#

and so if dependencies have their own dependencies...

#

it'll practically run all previously scheduled jobs

#

and you want to do that as rare as possible

muted star
#

okay that makes sense. Do structural changes force all jobs to complete?

rustic rain
#

yeah

#

sir, you are overthinking imo about sync points and etc

#

all you want

#

is just group all your schedulable jobs and systems that cause sync point away from each other

#

so your scheduled jobs stack will be as big as possible before popping

#

this will result in juicy threads utilisation

solemn hollow
#

Huh i just found out that removing a parent does not automatically remove the LocalToParent Component. Only completly destroying the parent does. why is that? does the localtoParent component do anything without a parent?

rustic rain
#

nah, it's just saved settings basically

solemn hollow
#

hmm i only have a LocalToWorld matrix (which contains scale) and i want to rotate it. what to do? i cannot use transform, rotation and scale components.

rustic rain
#

mm, you can grab rotation from it, manipulate it how you want and write back to same fields

#

assuming it has no translation, rotation or scale

#

and it's not parented - that should work

solemn hollow
#

yes that did work. but it has scale

#

and i cant get the scale out of the l2w matrix. since its somehow intermingled with rotation as i understand

rustic rain
#

uhm

#

I don't think so

solemn hollow
#

there is no localToWorld.Scale

rustic rain
#

ah, that's just helper method

#

you can write your own

#

why do you care about scale for rotation anyway?

solemn hollow
#

i just need to rotate the l2w matrix

rustic rain
#

scale matters for rotation only for children transforms

solemn hollow
#

ah ok. then i just need a way to rotate an arbitrary localToWorld

rustic rain
#

look up Rotation property source

#

and just write back to same values, hehe

solemn hollow
#

well that doesnt help my understanding of it sadly.

#

im sure there is a simple math function i can use to rotate a 4x4 transform matrix by a quaternion. is there not?

#

yeah well found a thread that explains it. you need to pull out the scale before you can safely rotate the l2w... too much hassle. guess ill need to keep the scale component around and reconstruct the l2w from there.

rustic rain
#

Create it from 000 pos, desired rotation delta and 111 scale

solemn hollow
#

thanks! i think i found another way to handle my problem though.

devout prairie
#

Stupid question but i've never did this before:

#

How do i get a managed component from an entity

#

CompanionLink seems to be internal so i can't get that component to get a reference to the gameobject that it links to

north bay
#

If I understand the code correctly you can directly use EntityManager.GetComponentObject<T> on the entity (and with some luck unity injects all MonoBehaviours)
But I've never tried using the companion stuff
I don't even know how an entity ends up having a companion 🀷

solemn hollow
#

dont pin me down on the exact api but GetComponentObject<T> sounds about right.

rotund token
#

if you want to query managed components you can actually just use entities.foreach
Entities.ForEach((MyManagedComponent m) => {}).WithoutBurst().Run();
otherwise as said, use GetComponentObject

solemn hollow
#

CompanionObjects are for HybridComponents like spriterenderer or sth that has no pure conversion yet. those are real monobehaviours on hidden gameobjects

devout prairie
#

GetComponentData was complaining

#

just doing something hacky at the moment for debug purposes

#

basically, applying a TextMesh to my prefab

#

it renders, by the miracle of modern medicine, but was trying to access it to update it's text value

viral sonnet
#

so, funky question: let's say i have an archetype of 128 bytes. it would need 2 cache lines to read the whole. and for simplicity, let's say this archetype only has 4 comps with a size each of 32. comp a, b, c, d. does the order in which i read these comps then matter?

#

what's confusing me is this. ecs lays out the data as SoA. so reading comp a actually reads comp A of entity 1/2. every comp access is actually a cache miss if you need only entity1 and are not SIMDing

#

so my conclusion is that the order doesn't matter anyway because every other comp read is a cache miss anyway

#

which makes it pretty bad (if you are not SIMDing) to have any comps smaller than 64 bytes

late mural
#

imma try il2cpp, im wondering what components do you need to add to the build asset to make it do it?

viral sonnet
#

interesting, i just merged some comps and that made it substantially worse. job takes now ~3ms instead of 2.8ms. main comp increase from 16b to 36b. I don't get it. That's a pretty big increase

rotund token
#

i'm confused

#

why is that a cache miss

viral sonnet
#

before I continue any thinking, entities is SoA, right?

rotund token
#

i hate acronyms

#

ideally all components are loaded into cache so why would order matter

viral sonnet
# 
struct Entity{
    float3 position
    float3 velocity 
    int health 
}

Entity[] myEntities 



// SOA
struct Entities{
   float3[] positions
   float3[] velocities
   int[] healthBars
}

Entities myEntities;``` good thing I've this reddit post still open. For some reason no Unity doc ever says what it is
rotund token
#

yes ok it's soa

#

i still don't know what it stands for ^_^'

viral sonnet
#

structure of arrays and array of structs πŸ™‚

#

it confused me for the longest time but tbh, graphics where I don't know how it relates to the linear memory confuse me even more

rotund token
#

but i don't get why either case would actually miss

#

take my cpu, 3900X
I have 64KB L1 per core, 512KB L2 per core (and 64MB shared in L3)
a chunk is 16KB

#

i can load, and ideally it does load, the entire chunk into the cache easily

viral sonnet
#

hm, yeah, you are right. i'm probably mix it up with pre-fetching too

#

all nice in theory, then i make a test and it's not working out in practice

#

like why does it matter how i merge my struct

#

and it's about 20 something bytes

rotund token
#

you've probably broken some simd operations

#

look at the difference burst has generated

#

(intel has even more per core cache, current gen is 80KB, 1.25MB and also 30MB shared)

viral sonnet
#

i'm pretty certain none of the code has any simd but i'll check. it's the best reason i can think of

rotund token
#

are you using all the data in your job?

#

(the after merged data)

viral sonnet
#

hm, good call. i'll check.

#

do you know if temp allocated memory in jobs end up in l1 cache of the core?

#

afaik it's allocated in global memory. would be nice. maybe stack alloc?

#

and yes, every merged data was used

#

i was pretty sure to begin with but i rechecked as the result is really confusing me.

#

and the simd part is too hard to find out right now. it's mostly scalar (the faster path)

rotund token
#

just from my experience, smaller the components faster the job

#

as a general rule

#

and adding unnecessary data to a job is definitely noticeable for slowing down

viral sonnet
#

what I can't say: how much one cycle of a job thread pushes the L1 cache, so how much pre-read data gets pushed out by other data. i'd have to count every stack alloc made. - what makes no sense nor reason to me is splitting or merging the same kind of data resulting in different timings. like, okay, i started a test that makes no sense in the context of cache lines or the L1 cache size. fine, but then to get slower. eh, that's just weird and i hope it's a simd thing i'm not seeing. i've to rewrite a lot for these tests so i think it's best if i start a focused one. - one other remark, with the size of L1 cache, it should matter very little how I micro optimize shuffling data around. pretty sure I don't exceed 16k per cycle. oh btw. 3900x has 32k, same as mine, I have a 5900x now

#

oh wait, 768k - per core, not thread, so yeah 64k

#

buuut πŸ˜„ 32k data, 32k instruction

#

cpuz isn't lying ^^

remote crater
#

Can you change the layer of an entity in runtime?

molten flame
#

Having some trouble with conversion in NetCode, hoping someone can help.
I've got some GUI GameObjects which are regular UGUI buttons and such that I'd like to be part of a prefab that I can spawn in.
However because they are GameObjects, I don't want them converted, I just want them to spawn when the rest of the prefab does and be connected to the prefab with a managed component reference on the prefab root.
So the setup is pretty simple, I have my prefab referenced in a "Ghost Collection" prefab holder that'll get converted and I put on the game object part of the prefab a convert to entity stop thinking it would do that.
But it seems this "convert to entity stop" doesn't work for prefabs and the gameobjects get converted anyway.
Does anyone have experience with this kind of thing? How did you solve it?

#

This is the setup

unknown.png unknown.png unknown.png
safe lintel
#

hm I believe the plan (albeit not executed yet to my knowledge in master) is to kill build configs completely for 1.0. We agree the idea is useful, but we also don't have the people to make it work properly. Sorry.

#

would have thought build configs were the future of building projects and/or they would be more serious about expanding the team working on the future core of unity πŸ₯²

rustic rain
#

Aren't build settings part of project settings?

rotund token
#

But it seems this "convert to entity stop" doesn't work for prefabs and the gameobjects get converted anyway.
yes for some reason convert to entity stop doesn't work in subscenes

rustic rain
#

You can make your own script that destroys whole hierarchy of entities in conversion

#

That should achieve same result in the end

safe lintel
#

@rustic rain my assumption was that build settings override project settings (might have been a quirk here or there) but I guess it’s all going back to project settings in 1.0

molten flame
rotund token
#

not sure i follow sorry

#

what does, "inject GameObjects from a prefab" mean?

molten flame
#

I guess I want the GameObject part of a GameObject prefab to spawn when I spawn an Entity prefab.
I sort of expected the GameObject part of the prefab to be serialized normally and just get instantiated by some built in system when the entity prefab was instantiated, I think this is supported with the addComponentObject method

rotund token
#

subscenes don't really contain gameobjects once converted

#

imo the easiest (and probably best) way to implement this is to just setup a reference to the gameobject prefab and simple instantiate it when the entity is created

molten flame
#

Its just pain cos you have to set up all the references from code instead of the editor

#

Yeesh I've fallen far lol, younger me would have never said that

rotund token
#

simply create a class IComponentData, with a GameObject field, let's call it SpawnGameObject

#

and query when the entity doesn't have a Gameobject but has the SpawnGameObject component
can re-use this for anything that needs a gameobject presentation

molten flame
#

I'm just being a lil bitch XD
I've got this ShipManagedComponents which was going to have a bunch of references to the GameObjects in the GameObject prefab.
But instead, I can just set up all those references on the GameObject prefab and point to that with a single reference as you are describing

#

Thanks for talking it through tertle πŸ™‚

nocturne dust
#

quick question, can I use Dots just for one part of my game. For example fish flocks. But everything else would be gameboject logic? or it is not recomended?

late mural
molten flame
#

You absolutely can and this is what Unity is expecting most people to do.

nocturne dust
#

but is it worth it? That is question, as it is very small part of project, and I am scared it will effect build size and etc.

#

while I could use just particles in more stupid way.... tho Dots would make it more realistic my tenfold

molten flame
#

No, its not worth it lol.
Unless you wanna learn how DOTS works and you are super keen.

#

There are a couple packages on the asset store for ECS flocks but they're not free

nocturne dust
#

Yea I saw, there are some on github as well using ECS. Current asset I use uses shaders for fishes heavily, which comes to 22mb RAM usage. O wonder how much it would be in ECS style

rustic rain
#

I haven't really started to figure it

#

but I have this idea in mind:

#
  1. try to create managed component that will store prefab instances
  2. then cut those prefab instances from authoring components
#
  1. expect SubScene to serialize that GameObject reference
nocturne dust
late mural
rotund token
#

i think you need to define what you mean by DOTS

nocturne dust
rotund token
#

what you want to do can be done quite nicely with jobs/burst

#

without touching entities

#

and if it's only a small snippet of your app that is probably the approach i would take

late mural
nocturne dust
rotund token
#

from memory it's really not worth it

late mural
nocturne dust
#

well, everything on webgl is sometimes πŸ˜„

#

mostly never :"d

late mural
#

from googling around it seems that some versions of dots work better or worse with it, and some unity versions work with dots and webgl better than others

nocturne dust
#

oh yea... i totally forgot.... that Unity 2022 have problems with dots O.o

late mural
solemn hollow
nocturne dust
#

isnt jobs to use multithreading? while webgl is living in one main thread world?

late mural
late mural
nocturne dust
#

so my knowledge were old O.o thank you for showing me this

late mural
#

no problem, wish you luck in getting stuff working!

late mural
rustic rain
#

hmm

#

I assume during OnCreate cycle

#

not all systems are created yet

#

as in, literally instantiated as object

#

so annoying tbh

#

I want to make a mini helper system that would let me serailize certain fields in systems, but outside of systems themselves

#

so I'd just create a Mono<T> where T : system

#

and just write to fields of my choice

#

after they are created

rotund token
rustic rain
#

wha

rotund token
#

hence you should always use GetExistingSystem<T> not GetOrCreateSystem

rustic rain
#

relaly?

#

oh well

#

hehehe

rotund token
#

unity considers GetOrCreateSystem a mistake

rustic rain
#

it never told me so

rotund token
#

if you want to initialize something in a system before OnCreate use the constructor

#

just make sure you put [Preserve] on it to stop it stripping in builds

#

(btw this wasn't always the behaviour, but was changed a few years ago)

rustic rain
#

[Preserve]?

#

for what?

rotund token
#

[0.9.0] - 2020-04-09
Changed
Systems are now constructed in two phases. First, ECS creates a new instance of all systems and invokes the constructor. Then, it invokes all OnCreate methods. This way, you can now use World.GetExistingSystem<OtherSystem>() from inside OnCreate().

rotund token
#

because it's created from reflection (activator)

#

so nothing references it

rustic rain
#

ah, I don't use constructors

rotund token
#

yes i was just saying if you wanted to initialize something in a system before oncreate use the constructor

#

that's all

rustic rain
#

all right, looks like I made my project a bit faster

#

hmm

#

now how can I inject into before OnCreate phase

#

is it even possible?

#

looks like it is

rotund token
#

Can use constructor like I just said πŸ’©

rustic rain
#

but I want to do that outside of systems

rotund token
#

That said this sounds like you're going back to entities 0.0.11

#

Back when things were injected into systems

rustic rain
#

basically define fields as public or with [SerializeField] and then this field will get written value from some mono in scene, if it exists

muted star
#

Hello, I get an exception for this code. What am I doing wrong?

unprocessedEntitiesQuery = GetEntityQuery(ComponentType.ReadOnly<IsMirroredInPresentationalWorldTag>());
unprocessedEntitiesQuery.SetSharedComponentFilter(new WasProcessedByPresentation {Value = false}); // Throws exception```

InvalidOperationException: Trying to get iterator for ECS.Components.Tags.WasProcessedByPresentation but the required component type was not declared in the EntityQuery.

rotund token
#

the error is pretty descriptive for this one

#

you're using WasProcessedByPresentation on your filter

#

but you haven't included it in your query

#

add it to the query

muted star
#

yes I just realized :p
thank you!

#

I should have told my rubber duck about the exception

wanton apex
#

Hello !

Is there a way to add an element to a DynamicBuffer using an ECB command ? I would like to be able to add an entity (wrapped into a IBufferElementData) to a DynamicBuffer after the ECB is playback. Currently I get an invalid reference to the entity when using Buffer.Add(entity) because it's a temporary index created from the ECB.

rotund token
#

Addbufferelement

wanton apex
#

It is working with Parallel Writers ? I can't find it

rotund token
#

sorry

#

AppendToBuffer

wanton apex
#

Thank you very much !!

#

I'm not sure about my architecture, I didn't play with ECS for long time.
I'm trying to create as an exercice a simple Particle System with ECS.
So I have an Emitter entity that it responsible for spawning the particle entities. I would like to be able to track the spawned particles from the Emitter. Is it a good practice to store them in a DynamicBuffer even if the count of this buffer could be very huge (let say up to 100 000 element). Or should I find better architecture ?

devout prairie
#

Generally what you'd want to do with something like particles is each particle is an entity, and each particle entity has a Translation/LocalToWorld component which stores position..

#

You then use a SystemBase to loop over all of the particle entities and do whatever on them.

#

For more complex behavioural setups you maybe want to look at the unity Boids example, which to be fair is quite complex

#

Regards tracking which particles are spawned from that emitter, the approach would generally be apply a 'tag' component to each one which identifies it as a particle, so for example 'ParticleTag' and then loop over all entities that have this.

rustic rain
wanton apex
#

@rustic rain Thank you very much for the answer ! What you describe is what I'm working on :

  • One system that update the Emitters, responsible for spawning the entities and setup their components
  • Based on components on the particles, different systems update the particles (Ex : If ColorOverSpeedComponent is on a particle, a dedicated system will query the speed and the color of the particle and update it.
    This system looks nice because it's easily extensible.

For tracking the particles from an emitter, what I understand is that I have three choice :

  • Using a hashmap as a System Member to associate each emitters with their entity (could be complicated to track the emitter deletion, etc...)
  • Get rid of this tracking while it's a very specific usage
  • Store the Emitter on the particle. Iterate over all particles and filter them to get all the particles that have been spawned from a specific emitter.

Am I right ?

rustic rain
#

and then you can get a query of all entities that belong to that same shared component

#

and in case you need to remove all at once

#

it'll be simply through shared filter

wanton apex
#

I never use SharedComponent, I need to look at the documentation to see how it works, but it looks very promising, thanks !

safe lintel
#

I actually had my particles in a dynamic buffer prior to just refactoring them all out and as actual entities this past week(they werent entities in the buffer, just transform and color data)

#

what I can say is that a single particle emitter(my entity that contained a dynamic buffer of particles), would kind of kill the framerate with between 10-50k particles

#

but if I ran it with hundreds of emitters with small particle counts, I could easily get over 100k without issue

#

anyway I used project tiny's particle system as inspiration for it, all the code is in c# so its worth checking out whatever you decide to do

devout prairie
safe lintel
#

I actually reverted it to entities not for the particle count issues but because using DrawMeshInstancedIndirect meant I had to sort transparency myself, and I figured it was less work to just have entity particles than to research z sorting

devout prairie
#

so you have say three emitter entities, a million particles using three shared components to group them into three groups

devout prairie
#

or hashmap or whatever

pliant pike
#

dynamicbuffers are just slow in general

safe lintel
#

yeah, although with low particle counts(I guess what tiny expected) the relationship between the emitter entity and its buffer is a little more palatable than when you just have a really high quantity of particles

wanton apex
#

Thank you very much all for the answers

safe lintel
#

tbh it was perfectly fine for anything that I would have used the old shuriken for, though as entities you can kind of get closer to vfx graph

devout prairie
#

yeh it kinda makes sense for that many emitter with small particle counts really, before going to the stage of super optimization.. it's a shame vfx graph is such a goddamn pain to use, i've been using particle systems for years doing animation work and i find vfx graph to be a bit of a black box in terms of the order of nodes and operators and them not doing exactly what you'd expect etc etc

safe lintel
#

my experience is quite minimal with particle scripting so I usually figure its my own incompetence but I do wish they'd post the c++ source to a lot more of their subsystems on github, would love to glean more ideas on particle setup and the math behind some behaviours

#

btw @rotund token should we replace all GetOrCreateSystem<T> with GetExistingSystem then for caching systems in OnCreate? is GetOrCreateSystem slated for deprecation?

rustic rain
#

No it's not

#

But yes you should replace

coarse turtle
muted star
#

I have an unexpected excpetion when I run this code

var queryDesc = EntityManager.UniversalQuery.GetEntityQueryDesc();
queryDesc.None = new[] {ComponentType.Exclude<SimulationSystemEntityTag>()}; // This causes an exception...

query = GetEntityQuery(queryDesc); // ...here!
#

Is queryDesc.None broken?

karmic basin
#

First time I see em.UniversalQuery πŸ‘€

#

But I'd say it's the Exclude that messes up

#

Try that ? None = new ComponentType[] { ComponentType.ReadOnly<SimulationSystemEntityTag>() }

muted star
#

yep that worked πŸ™‚

rustic rain
#

what

#

UniversalQuery?

#

I want to know more about that

#

xD

#

is that like, query for all entities?

coarse turtle
#

yea

rustic rain
#

I wonder what would be application for that

#

I assume that's for excluding some components

coarse turtle
#

i mainly use it to reset my world entirely

rustic rain
#

yeah, I guess it's one way

coarse turtle
#

it does nuke WorldTime iirc so you'd need to recreate that if you ever do something like that πŸ€”

whole gyro
#

Is there a way to prevent an Entity field from being remapped during a call to Instantiate()?
I'd like each of my entities to have a component referencing the prefab they came from. But due to how instantiate works, any field containing an Entity reference in the local hierarchy will be remapped to the newly instantiated version of that entity. I'd rather not have to update this component on every newly instantiated entity.

#

Is there some kind of [DontRemap] attribute I'm missing? I looked through the remapping code and couldn't find anything like that.

rotund token
viral sonnet
whole gyro
#

from EntityRemapUtility.cs

        public static Entity RemapEntityForPrefab(Entity* remapSrc, Entity* remapDst, int remappingCount, Entity source)
        {
            // When instantiating prefabs,
            // internal references are remapped.
            for (int i = 0; i != remappingCount; i++)
            {
                if (source == remapSrc[i])
                    return remapDst[i];
            }
            // And external references are kept.
            return source;
        }
#

Btw, I think this prefab remapping is generally a very useful feature. It allows you to have a prefab hierarchy with entities referencing other entities in the hierarchy. 99% of the time, you want all these remapped. I just wish there was a way to turn it off in the 1% of cases where you want to maintain the references to the prefab.

coarse turtle
#

ah πŸ€” sucks Burst doesn't exactly support unsafe function pointers yet

rotund token
#

How so

coarse turtle
# rotund token How so

internal static delegate* unmanaged[Cdecl]<uint, void> UnloadSoundHandler; It's more in regards to this syntax since I'm loading a native dll dynamically in the editor so I'm grabbing the function pointer and assigning it to delegate*

#

Burst doesn't recognize delegate* unmanaged[Cdecl]<> if you try to stick it into a bursted job πŸ˜…

rotund token
#

Oh not what I thought you meant

#

Hmm I'm pretty sure I wrangled with something like this a while ago, but it was marshalling a managed method

#

I feel like I managed to get it to work somehow but I can't remember

coarse turtle
#

o if there's a way that'd be nice to know since it'd be nice to keep writing jobs consistent while maintaining and adding more features to a library incrementally

#

but otherwise oh well, I don't really mind slapping compilation definitions onto jobs for editor stuff, I was more or less just hoping that it would work out of the box once I moved over to 2021 Β―_(ツ)_/Β―

rotund token
#

i can't remember it was a while ago

#

however there is a definitely a round about way of being able to call marshaled stuff

#

by writing a tiny little c++ library

coarse turtle
#

o - i'll take another look at the forums - I feel like I remember seeing someone from Unity comment about it a while ago but can't really remember

rotund token
#

take my little logger that I pass to native libraries to allow them to use the unity log

    {
        public static readonly SharedStatic<IntPtr> Ptr = SharedStatic<IntPtr>.GetOrCreate<RecastLogger>();

        [RuntimeInitializeOnLoadMethod]
        private static void Init()
        {
            Ptr.Data = Marshal.GetFunctionPointerForDelegate<DelegateLog>(DoLog);
        }

        [UnmanagedFunctionPointer(CallingConvention.StdCall)]
        private delegate void DelegateLog(LogCategory category, string msg);

        private enum LogCategory
        {
            Progress = 1,
            Warning,
            Error,
        }

        [MonoPInvokeCallback(typeof(DelegateLog))]
        private static void DoLog(LogCategory category, string msg)
        {
            switch (category)
            {
                case LogCategory.Progress:
                    Debug.Log(msg);
                    break;
                case LogCategory.Warning:
                    Debug.LogWarning(msg);
                    break;
                case LogCategory.Error:
                    Debug.LogError(msg);
                    break;
            }
        }
    }```
# 
        public static extern Context CreateContext(IntPtr log, bool state);```
I pass this to a library
#

if you simply made a little library that looked like

        public static extern void SetMethod(IntPtr method);
        
        [DllImport(DLL)]
        public static extern void ExecuteMethod();```
#

you could set a method in the library, store it, then call ExecuteMethod from burst πŸ˜„

#

very round about

coarse turtle
#

lol, yea i actually have something similar to that in my dll lol

#

Maybe i'll do that universally haha

remote crater
#

Is it possible to change the layer of an entity at runtime? I want to display one on the UI for a moment to signify pickup.

#

I have it tween up, look attractive and dissapear, but it also shows in game too because the layer does not change to UI only.

rotund token
#

i assume you're talking rendering not physics layers

#

RenderMesh has a layer field

remote crater
whole gyro
rotund token
#

it's passed to the batch render group, not sure why it wouldn't work

whole gyro
#

Sorry, I think I'm thinking of the layerMask field which is the equivalent of Renderer.renderingLayerMask. Pretty sure this used to be broken. But I think it may have been fixed in a recent hybrid update based on some forum posts I found

remote crater
molten flame
remote crater
#

The above code is in case someone wants to set a layer, code easily made thanks to tertle's great help again.

#

I have a whole pool of entities in my game with Disabled/Disabled Rendering tags

#

And I just clone em, and change stuff

#

Hey, I'll give more examples

#

The code won't compile for anyone right off the bat since it uses some easily removed stuff.

#

But the premise of simply removing <disabled> and <disabledrendering> is awesome for all to use.

#

Assuming you don't mind 1 copy of prefabs floating in memory which is normally a non issue, this is awesome for object pooling.

#

Some fancy addressables moves and you can load and unload tho if that is an issue.

#

MY DOTS/ECS project I'm on now is like a symphony

#

I wish I could just show everyone how easy it is to use... But any time I try and make a new ECS/DOTS project, it just doesn't run, lolololol.

whole gyro
remote crater
rotund token
# 
rm.layer = 3;
ecb.SetSharedComponent<RenderMesh>(e, rm);```
late mural
#

i dont wanna necrobump a necrobump, as such ill ask my question again, sorry lol, which components do you need on the build asset to make it do il2cpp as i wanna mess around with it?

rotund token
#

also i don't really understand why you're using a ECB when you're using EntityManager right there

remote crater
rotund token
#

render mesh is a struct

remote crater
#

I did some freaky deeque things like that and it was setting the parents data

#

kk, pass by object

whole gyro
rotund token
#

unity's advice is not to pool

whole gyro
#

Instantiating a prefab of an existing archetype is supposed to be very fast. Might actually be better than the structural change of removing Disabled and DisabledRendering tags from existing entities

remote crater
#

Its the fastest way of instantiating

#

... that I know of

#

...the above line is for legal reasons I cannot speak of

#

..that indemnify me in case I do not know some black box inner workings of dots/ecs

#

..which I probably don't. therefore do not sue

late mural
#

ecb works good enough in most cases, on decent computers (one of my friends) you can istantiate 1000s of entities (all with rigid bodies aswell) per second and still get over 60 fps

remote crater
#

I use ecb and ecbp depending on use

whole gyro
# remote crater When I say pool, it means just to clone an entity

Your code appeared to be removing tags like Disabled, DisabledRenderering, and something called DOTS_OBJECT_POOLER from existing entities. But there is EntityManager.Instantiate() and EntityCommandBuffer.Instantiate() that are designed to create a new entity from an existing entity.

#

And there is a Prefab tag that works in a similar way to the Disabled tag. It is automatically filtered from all entity queries by default and these entities won't render. But Instantiate() will automatically remove this Prefab tag from the entity for you.

whole gyro
remote crater
whole gyro
whole gyro
# late mural i dont wanna necrobump a necrobump, as such ill ask my question again, sorry lol...

I haven't tried il2cpp recently but I saw quite a few forum posts since 0.51 came out talking about some new setting you have to change in the build settings: https://forum.unity.com/threads/il2cpp-dots-entities-arent-rendered.1303950/

Unity Forum
Question - (IL2CPP + DOTS) = Entities aren't rendered?

If I turn IL2CP on im just getting a blank, black screen.

This option is important for me as I remember it always added a lot of FPS to my...

late mural
viral sonnet
#

such a bummer that chunks are 16k. i'd have a really good usage of sharedComp but it would also require to make a chunk much much smaller :/

#

kind of wish for a better implementation for sharedcomps. on one hand they are just a grouping mechanism

rotund token
#

on one hand they are just a grouping mechanism
there is no other hand πŸ˜„

viral sonnet
#

so now I'm going deep into DBs because when sharedComps are not sufficient. entities and DBs are the next best grouping mechanism πŸ˜„

molten flame
#

@rotund token how do you handle cleanup of entities when transitioning scenes or states, e.g. going from in-game to the main menu?

rotund token
#

i only have 1 scene

#

as for going back to main menu, i have a client (+server) world

#

and i just dispose of them

molten flame
#

Ah ok, where do you put common stuff, like a loading screen?

#

Default world?

rotund token
#

my current architecture is a bit weird actually

#

my menu/loading/etc is all part of client world

#

after disposing existing client world i just create anohter

#

i used to do that all in default world

#

but it annoyed me the state transitions across worlds

molten flame
#

Ah ok, yes me too!

molten flame
remote crater
#

I ran into a peculiar error where if I do not assign disabled to every entity in a .foreach, I get weird errors:

#

If you'll note in the paste.bin I arbitrarily don't run a ecb.addcomponent on every entity and it sends those errors. If I run it on every component it doesn't error out.

#

I should be able to do a workaround... But I wonder what's at play.

rotund token
#

looks unrelated to the ecb

#

that just caused the issue to appear

#

seems you have bad dependencies somewhere (read error)

drowsy pagoda
#

Here is my simple code:

public void LoadScene(Hash128 guid)
        {
            var system = World.GetOrCreateSystem<SceneSystem>();
            var loadParameters = new SceneSystem.LoadParameters
            {
                AutoLoad = true
            };
            var entity = system.LoadSceneAsync(guid, loadParameters);
        }

The guid parameter is pulled from a MonoBehaviour SubScene inspector field. It's property: subScene.SceneGUID. That is what is passed to the LoadScene method on Awake.
The LoadScene method is in a SystemBase.
When loading the game, the scene does load and I verified the name matches. But it doesn't become parented to World, and it's greyed out and doesn't open, always (closed). What am I missing here?

rotund token
#

are you sure the entities aren't loaded?

drowsy pagoda
#

No, I click on them and the inspector shows their component datum. But they entities do not appear in game view. There is only one cube in the sub scene. Also the sub scene that is referenced in the MB is a Prefab asset and not in scene when hitting play. Not sure if that matters or not.

rotund token
#

i ask because these are actually open scenes for me and entities in world

unknown.png
#

just the inspector is broken

rotund token
#

are you trying to open a subscene that's not actually a subscene

drowsy pagoda
#

So I created an actual sub scene. Added a cube. Dragged the scene to prefabs assets, deleted the game object sub scene. Referenced the Prefab sub scene to MB that lead me to write my issue.

rotund token
#

ah, put your subscene in the actual scene

drowsy pagoda
#

Yeah I think that’s the part I’m having trouble with. Never done scene loading before in dots.

rotund token
#

they can't be prefabs

#

the OnEnable method needs to execute on the subscene script

#

to setup before you can load it

# 
        {
            GameObjectSceneUtility.RegisterSubScene(gameObject.scene, this);```
drowsy pagoda
#

Ah. I see. Ok. I’ll run some more tests tomorrow. Thanks for the help.

rotund token
#

it registers the scene

#

all your subscenes should exist in your primary scene

drowsy pagoda
rotund token
#

untick the auto load field

drowsy pagoda
#

Oh…lol

#

Thx

molten flame
#

Ran into an odd bug today using netcode.
I had a ghost that was travelling along all fine at constant velocity but as soon as it reached around 2150 units along any axis from the origin it would snap to -2150 units on that axis and not move anymore.
Turned out that happened was the quantization I set was way too high and it was overflowing the float.
So if this bizarre situation happens to you, just make sure you use a lower quantization value or the unquantized variant.

rotund token
#

how much did you quantize that to o_O

rustic rain
#

float overflow?

#

is that even a thing?

rotund token
#

netcode requires quantization of floats for serialization

wanton apex
#

Hi !

Is there a simple way to initialize a PhysicsBody / PhysicsShape via code when instantiating an entity, I would like to be able to initialize it or not depending on a boolean value in a component ? I look at the PhysicsBody Authoring, but it looks like it add lots of components. Is there any helper function somewhere to do this ?
Thanks !

late mural
wanton apex
late mural
wanton apex
#

Thank you ! I let you know when i will work on it !

wanton apex
late mural
wanton apex
#

Does EntityCommandBuffer support AddSharedComponent() in a burst job ? I get some strange errors that refer to it :
(0,0): Burst error BC1051: Invalid managed type found for the field `EqualFn` of the struct `Unity.Entities.FastEquality.TypeInfo`.: the type `System.Delegate` is a managed type and is not supported

glad epoch
#

where can i find -s WASM_MEM_MAX in unity

rustic rain
#

but shared components are considered managed

#

from what I'v seen in manual

wanton apex
#

Thanks for the answer. I need to find a workaround to keep performance good... Thanks !

rustic rain
#

For this case I usually just keep special group that runs right before ECB

#

in which I do managed code, that causes sync points

wanton apex
#

In my case I instantiate Entities and add some Component and Shared Component (Physics Setup) to it in the same System. You think I would need to separate it in two system to keep burst for the larger part of the code ?

rustic rain
#

you might want to just move it to non-bursted system

#

alltogether

#

you'll get rid of useless buffer/creation/filling/unpacking

#

and will be able to freely manipulate creation of your entities

wanton apex
#

Thanks, I will do it. But I have some performance Issue regarding instantiation. I need to investigate it. Thank you for the answer !

rustic rain
#

but it's all done on main thread anyway

#

the reason to use buffer - is to be able to "schedule" structural changes in bursted system

#

ECB itself is running on main thread

wanton apex
#

I try it using main thread, thanks !

misty wedge
#

Can you get a reference to a native array element without using a pointer?

rustic rain
misty wedge
#

local, since that's the only place where ref is valid

#

I didn't find any native array api that returns a ref struct

rustic rain
misty wedge
#

That's using pointers

rustic rain
#

but that gives you reference

misty wedge
#

Well yes, but I asked if it was possible without pointers

rustic rain
#

ref var kek = ref array.GetAsRef(i);

#

well

misty wedge
#

basically

rustic rain
#

why pointers are a problem in the first place?

misty wedge
#

It's not, but ref is generally safer, and I wanted to know if native array exposes any API that can do it

rustic rain
#

but it's literally that, kek

misty wedge
#

?

#

ref is not the same as an unsafe pointer

rustic rain
#

and in editor you'll have safe checks

misty wedge
#

Since when does unity have safety mechanisms for C# pointers

rustic rain
# 
    [Conditional("ENABLE_UNITY_COLLECTIONS_CHECKS")]
    private static unsafe void CheckIndexInRange(int index, int length)
    {
        if (index < 0 || index > length)
            throw new ArgumentException( $"Index {index} is out of range of length {length}" );
    }
#

that part

misty wedge
#

When is that method called?

rustic rain
#

When you want it to xD

    public static unsafe ref T GetAsRef<T>( this NativeArray<T> array, int index ) where T : struct
    {
        CheckIndexInRange( index, array.Length );

        return ref UnsafeUtility.ArrayElementAsRef<T>( array.GetUnsafePtr( ), index );
    }
misty wedge
#

Wait, that's a method on NativeArray?

#

I thought you were joking

rustic rain
#

it's not built in

misty wedge
#

...

rustic rain
#

UnsafeUtility is

#

just drop it into utility class and it'll work like a charm

misty wedge
#

So to answer my original question, no, there is no native functionality on NativeArray πŸ˜…

rustic rain
#

yep

misty wedge
#

I'm well aware of how to write it with pointers

#

If I were to access the struct from the automatically dereferenced pointer, I would need to store it as a ref struct though, right? Since otherwise it would copy it locally

rustic rain
#

I don't quite get what you mean

#

ah

#

now I do, yeah

misty wedge
#

e.g.

int* ptr = ...;
ref var p = ref ptr[someIndex];
rustic rain
#

why do smth fancy like this though

misty wedge
#

I wouldn't really call that fancy, it's just getting a reference to an element in an array

rustic rain
#

but you can do it without actually touching pointers yourself

#

ref UnsafeUtility.ArrayElementAsRef<T>( array.GetUnsafePtr( ), index );

#

this is all you need basically

misty wedge
#

Ah, didn't know that methods exists, cool

rustic rain
#

and that would act like a class

rustic rain
misty wedge
#

Well that was when we were talking about if it was possible natively, so I saw unsafe and didn't continue reading πŸ˜… my bad

#

Thanks for the pointers (hehe)

lusty otter
#

Will Burst optimize away property getters/setters so they are basically zero cost comparing to a field?

misty wedge
lusty otter
#

Nice thanks.

dense crypt
#

How do you guys handle input? Do you read inputs per system, or do you have components that contain their respective values?

viral sonnet
#

best to gather inputs up in a nice input struct

hot basin
rustic rain
#

hmm

#

I wonder if I can keep pointer reference

#

to singleton component

#

assuming it'll never change archetype

elfin spire
#

Is this the right way to execute some mainthread code after a scheduled job is complete?

viral sonnet
#

it works but it's not really the intended way, you'd utilize the dependency chain and another job that calls Unpack... method

#

remember you can make managed jobs without burst

elfin spire
#

oh, other jobs can execute main thread stuff?

viral sonnet
#

well I don't know what's in this method. there are some calls that unity complains about when not being called by mainthread so that's out the window but for everything else, it's possible

elfin spire
#

texture.Apply

viral sonnet
#

the problem with LateUpdate is that it could be called a frame later (because the jobHandle.IsCompleted state is true a frame later) and make problems. if you are really concerned use .Run or jobHandle.Complete()

elfin spire
#

oh yes, that's why it's complaining about accessing deallocated stuff, i'll move it to update and let is handle .IsComplete, that should work

viral sonnet
#

WithoutBurst and Run is 100% on mainthread so you can do all you want - the problem is that it has no dependency as input so it doesn't help in your case

#

yes, I'd use jobHandle.Complete too. it's the cleanest solution imo and maybe you can get around to burst compile the Unpack method/texture manipulation and then you can schedule both. I haven't used any Texture2D, someone else would need to answer how they use it and if it's possible to burst. (I don't think so as it's a managed class)

elfin spire
#

ugh nativearray deallocated error still

#

I don't understand these allocators so i'll revert back to .Run() and burst

gentle harness
#

Did you dispose your nativearray?

gentle harness
elfin spire
#

yes mono, it's not ECS, I was disposing the nativearray in the main thread, in that method that's called after the job IsComplete

safe lintel
#

Im not sure that you can reliably use IsComplete tbh, might need to use handle.Complete() to force it finished at some point

elfin spire
#

ok, thanks

#

would I so: schedule from main thread, then wait a constant amount of time like 0.1 second then call complete then do the main thread work?

gentle harness
#

Several frames has likely passed in 0.1 seconds

#

If you don't dispose your nativearrays before then you will get the warnings

#

Call Complete() at the point in the frame when you need to use the data and dispose of things

safe lintel
#

@elfin spire might be able to schedule in update and then complete in lateupdate. this is kind of the pain point of why I personally dislike using jobs outside of ecs

elfin spire
#

maybe i should turn this into a compute shader.. all I'm trying to do here is update a 2000x2000 control texture (within a radius R of a bunch of coordinates do some simple math to set the pixel to the value 0,0.5 or 1) to show an influence map in the game around certain objects, what do you all think?

#

update frequency, about 0.1 second, can be less

#

think grass growing around wet areas and desert expanding around dry areas

safe lintel
#

eh for me easier just to do it as a job in update/lateupdate, more like if you start to schedule a lot of jobs from monobehaviours might need to think of strats for handling dependencies(which ecs basically does for you). think I was just fretting needlessly πŸ˜…

elfin spire
#

oh I get it

#

I always trip over allocator lifecycle

#

I used Ninja and coroutine, which was doing threading for me, no burst though

viral sonnet
#

maybe split the texture data from the actual data and after you are done computing, generate/update the texture

#

you could do all that in a big 2000x2000 array

#

must be even possible to vectorize

#

if you split the load into quadrants you can even parellel it

#

2k is quite big though

#

for updating ever 0.1 sec you can use variable updaye group

#

that way you can change the update rate without much changes

viral sonnet
#

how does this work? An EntityCommandBuffer can give us a DB with AddBuffer. Yet, the entity isn't even registered in a chunk and when the DB is inlined in the chunk, how does it know where to write the data to? or is it just a temporary buffer that gets copied to the chunk when the ECB gets applied?

rotund token
#

yes addbuffer is just local memory inside the ecb

viral sonnet
#

alright, thanks

#

I have quite the predicament, on one hand I want to use write handles, also to keep the safety system intact, on the other I want to bump the change version manually because I have conditional writes. Seems I need to write an Accessor that's not bumping the version. It would be great if we have a simple parameter for that.

#

well, easy enough to add to my lib

rotund token
#

haha wrote something similar

unknown.png
#

turns out i didn't need it though

#

i wrote a lot of hacks for my current library

#

and i've managed to avoid using most of them

viral sonnet
#

do you still use the temp memory and memcmp?

rotund token
#

yeah i have quite a few memcmp

#

well 4

viral sonnet
#

i just don't know how to get around that. for example, a buffer where spell ids are saved for cooldowns. not every spell has a cooldown, so how to exactly go about that for a memcmp?

rotund token
#

my cooldown is a component and effects without a cooldown just don't have the component

viral sonnet
#

oh you make zero compares. hm. is that a list or what are you using?

rotund token
#

actually just spent all morning writing tests for all my states for activation

unknown.png
#

found 2 bugs

viral sonnet
#

ah i see

rotund token
#

oh so yeah im going to write up a little thing about my timer

# 
{
    remainings[i] = math.max(0, remainings[i] - this.DeltaTime);
}```
#

a timer like like this is hugely problematic with change filters right

#

because while remaining > 0 you need it to execute

#

so you can't just change filter it

#

so usually it just runs every frame

#

and i'm going for no operations executing when nothing triggers

#

i only solved this last night

viral sonnet
#

glad you solved it πŸ™‚ i have to ask though, why not use ticks?

rotund token
# 
if (batchInChunk.DidChange(this.RemainingHandle, this.SystemVersion))
{
    var remainings = batchInChunk.GetNativeArray(this.RemainingHandle).Reinterpret<float>();
    for (var i = 0; i < remainings.Length; i++)
    {
        remainings[i] = math.max(0, remainings[i] - this.DeltaTime);
    }
}

// System 2
if (batchInChunk.DidChange(this.RemainingHandle, this.SystemVersion))
{
    var remainings = batchInChunk.GetComponentDataPtrRO(ref this.RemainingHandle);

    var length = UnsafeUtility.SizeOf<float>() * batchInChunk.Count;
    if (UnsafeUtility.MemCmp(remainings, this.Zeros, length) != 0)
    {
        batchInChunk.SetChangeFilter(this.RemainingHandle);
    }
}```
rotund token
viral sonnet
#

simulation ticks. so you just save a start tick and end tick for timers

#

heh, that's a smart memcmp πŸ™‚

rotund token
#

because
a) i don't want to be limited to fixed update
b) that still causes the system to update every frame which is what i'm trying to avoid

#

i will say the float subtraction was by far the slowest part of my whole loop

viral sonnet
#

well, having float times is a huge pain in networking and general effects that tick with a certain interval

#

you could still have remaining ticks with an integer

#

so the memcmp still works

rotund token
#

this is true but again i'm not sure i want to be restricted to a fixed update

#

it makes the library less universal

#

and i'm already testing this with more entities than anyone has probably ever used

viral sonnet
#

what's this.Zeros? a matching nativeArray of zero values?

rotund token
# 
            [NativeDisableUnsafePtrRestriction]
            public void* Zeros;```
#

it's just a chunk of memory of 0s

#

since i'm in an ISystem i can't store native arrays

#

so literally just allocated a chunk of 16KB memory in the system

# 
            UnsafeUtility.MemClear(this.zeroes, maxSize);```
really nothing exciting
viral sonnet
#

lol, what a workaround πŸ˜„ what does Unity intend to use in ISystem? (now I know why my ISystem wasn't working with the nativecontainers)

rotund token
#

native containers will work fine in 1.0

#

with the removal of dispose sentinel

#

they just currently don't so limited to unsafe containers

viral sonnet
#

noice

rotund token
#

anyway i can actually do this change filtering thing in the same system

#

i wrote a quick
public static unsafe void SetChangeFilter<T>(this ArchetypeChunk chunk, BufferTypeHandle<T> handle, uint version)

#

where i could specify the version specifically instead of using global

#

so i could just use global version + 1

#

but i decided i'd rather just a second system instead of another hack

#

i'm avoiding this extension

rotund token
#

reducing tick rate during heavy simulations

#

(though I would like the performance)

#

i did consider using quantized floats as ints but decided against it for now

#

not confident i can give up precision atm

viral sonnet
#

as you are using read handles, this doesn't trip anything up?

rotund token
#

System 1 uses a write handle

#

actually

#

both systems use write handles

viral sonnet
#

i see

rotund token
#

batchInChunk.SetChangeFilter(this.RemainingHandle);
i could just change that to batchInChunk.GetNativeArray(this.RemainingHandle);

#

and it'd do the same thing

viral sonnet
#

yeah makes sense to just get the ro ptr. i can probably reduce some handles myself

rotund token
#

i just thought calling SetChangeFilter was clearer what it was doing

viral sonnet
#

yep, i also like thay approymuch better. more control over the change filtering is very welcome

rotund token
#

but yeah a system checking a change filter will not trigger another change filter on itself hence the need for the second system (though as mentioned I could use the +1 hack above but decided against it)

#

hmm you indirectly notified me of a potential problem

#

SetChangeFilter should only be called on ComponentTypeHandle that are marked as write

#

time to throw in a check for that

viral sonnet
#

i probably should also add that check πŸ™‚

#

i need to find out what's actually happening with refs. the difference between ref var input = ref inputs[i]; and var input = inputs[i]; one should put the pointers on the stack, the other the actual data, right?

rotund token
#

depends what mood burst is in πŸ˜„

#

but yes I think in theory

viral sonnet
#

hm, so I should really avoid refs for just reads. i saw some improvements for bigger structs. even down to anything > integer

#

main reason, no memcpy

#

(puts another test on his todo)

#

in theory putting the data on stack should be vastly superior

rotund token
#

again it depends a lot on burst!

#

you could just end up loading the memory directly into simd registers

#

but no expert on this~

viral sonnet
#

hm, so only in the case burst decides to simd it?

#

(which I know it doesn't) πŸ˜„

late mural
#

do entity and their rigidbodies obey regular time scale, or do they have a seperate one?

late mural
#

oh also how do i build with il2cpp, it appears i need some kind of tool installed?

unknown.png
gentle gyro
late mural
late mural
late mural
# gentle gyro

i would assume that is in player settings, although where i cannot find

#

oh think i found it?

gentle gyro
late mural
#

i notice you have incremental garbage destruction enabled, just curious is this better than the default usually

late mural
gentle gyro
late mural
#

oh, mine was default off

late mural
#

im not clever

gentle gyro
late mural
gentle gyro
late mural
gentle gyro
#

Looking...

late mural
#

ok thanks so much!

gentle gyro
#

Did you add the Windows Build Support (IL2CPP) module when you installed the version of Unity for this project?

late mural
gentle gyro
#

I hope not...because I'm running out of ideas... πŸ™‚

late mural
#

lol

late mural
#

it is installed

#

welp

gentle gyro
#

Hum...

late mural
#

any other ideas?

#

managed to find that info on the build failed thing

gentle gyro
#

what editor are you using?

late mural