archived-dots

[rustic rain]

Yeah, but I don't have such menu

hold on, I'll restart Unity again

Nvm

[rustic rain]

yaaay, it finally works

[rotund token]

how convenient is it!

it will link your own new assemblies together as well

(dont judge the ui code too harshly, literally first thing i ever wrote in ui elements. was my learning case)

[rustic rain]

ok, normal renderpass works

now how to use it with collision texture

hm

Ok, I got it

was simpler than I expected kek

ok, it really is seems pretty simple

now, how do I get a color of pixel

of chosen viewport point?

[rotund token]

your texture size should match the dimensions of your resolution

[rustic rain]

I don't want it to be relevant

I do have a viewport point

upon clicking

I kind of want to just get that pixel disregarding what size texture is, based of viewport

[rotund token]

i don't really understand what your saying

your output texture neds to match screen resolution to have the right colors

[rustic rain]

Well, ok

How do I even get a texture tho

[rotund token]

does the tutorial not discuss the concept behind this?

[remote crater]

I successfully have .50 dots installed and my FPS issue is gone

That FPS issue crept in one of the september-nov unity upgrades

[rustic rain]

I wonder how I can make Material to have colour based off entity id

[viral sonnet]

I saw this a number of times now but I don't remember where. Certainly in the forum. I think a google search can lead you there.

[rustic rain]

any tip on keywords?

I have been looking for this info for hour already

best I got: source of RenderObject feature kek

[viral sonnet]

found something: https://forum.unity.com/threads/how-to-change-entitys-color-by-code-in-ecs.1045345/

keywords: "hybrid renderer color per entitiy" - hope you find what you're looking for

I've also read something about MaterialBlocks

[rustic rain]

nnnah, I'm not sure that's what I'm looking for

here's the thing

I need 2 renderers

1 for actual rendering

the other for collision rendering

where I render object's opaque basically

and then by reading pixel I determine what id of entity is inside that pixel

and I need to render 2nd without affecting first

Potentially maybe I just need another MeshRenderer for this

oh well, you can't add 2 MeshRenderers on same object

[safe lintel]

@rotund token is every folder inside that scripts folder also given an asmdef?

[round narwhal]

Hello everyone, I'm currently starting using DOTS and I cannot figure out how to create a Character Controller.
Can somebody give me a good example of what it can be ? (nb : it's for a FPS)

[rustic rain]

Here one. But it's broken by new version and pretty out dated
https://github.com/ssell/UnityDotsCharacterController

[round narwhal]

Thank you so much

[rustic rain]

Here my take on it without old struct jobs

but it's also outdated

haven't touched it since 0.50 release

[viral sonnet]

https://github.com/JonasDeM/EntitySelection.git does something similar for entity picking. maybe a good starting point

[rustic rain]

but it's for runtime, not for scene view

oj

oh

I see

it uses shader

Might be actually good, thanks

[viral sonnet]

np

[viral sonnet]

Getting a character controller right is a big pain so: https://assetstore.unity.com/packages/tools/physics/rival-dots-character-controller-195567

[rustic rain]

yep, that's actually seems like a real solution
My sincere appreciation, sir.

[hot basin]

how can I use generics in job decryption/entity query?

i.e. I have interface with method .Evaluate() and 4 components that implements it

can I make generic job with <T> where T:Interface?

how to query for that?

[viral sonnet]

what's generic? the job or the data?

[hot basin]

job, data just inherits interface

[viral sonnet]

huh, wasn't sure if it would work to just build an interface deriving from IJob or others.seems like someone already did the work https://nagachiang.github.io/unity-dots-creating-generic-systems-with-generic-jobs/#

[hot basin]

but it uses generic system also so there is still one system per component

I would like to do something like this:

var GenericQuery = GetEntityQuery(ComponentType.ReadOnly<IMyInterface>());

[viral sonnet]

hm, maybe the polymorphic struct from PhilSA can help?

then phils solution will help. although it will source gen a switch

[rotund token]

Yes

I like to keep my library / game very modular and reusable

Heavy use of asmdef ensures no circular dependencies

[rotund token]

This works fine what's the problem?

[hot basin]

it will?

[rotund token]

Sure why not?

You can use generics just not with entities.foreach/ijobentity (no code gen)

[viral sonnet]

@rotund token have you seen the post from dreaming in the events thread?

[rotund token]

nope, kind of stopped reading the thread 😅

why?

[viral sonnet]

Write key-value pairs to blocklists (NativeStream or similar) in parallel.
Copy the blocklist to a NativeList.
Sort the NativeList.
Allocate a second NativeList and store start offset and count pairs for matching keys
Iterate by key in parallel.

found that approach interesting

that's what he does in his animation package

certainly interesting when sorted by chunks especially

[rotund token]

well obviously now you need to implement it and compare 😄

[viral sonnet]

yeah ... 🤣

I don't think it's that fast though. Too many steps involved. Kind of the overhead when you need determinism

[rotund token]

yeah he loves determinism

which if you need is obviously important

[viral sonnet]

I'm gonna take the interesting parts out of it. Namely, getting rid of the NMHM, utilizing a NativeStream/UnsafeParallelBlockList for fast writes, transform into a NativeList in single thread (no add problem) sort by chunk ptr (if possible) or indexInChunk. building the start offset and counts this can be parallelized. could be noiiice 😄

[viral sonnet]

Is the default NativeList.Sort any good? Sorting 500k takes 8ms

[rotund token]

for a sort that sounds reasonably fast

[viral sonnet]

Well ... then case closed 🤣

shame though, sounded interesting without NMHM. maybe sorting by entity is a problem. The compare could certainly be written in a faster way

[rotund token]

you can pass in your own compare

but entity compare is already setup to only cmopare the index

        {
            return Index - other.Index;
        }

so not sure how you could make it faster since it already skips version so it's just an int compare

[viral sonnet]

ah thought it might compare version. never checked

[viral sonnet]

I'm just barely scratching the 60fps with using the new method of just writing to the NMHM key/values array

Pretty cool when this started with race conditions and Interlocked with 20ms. (Interlocked was the fast part, rest was slow)

[pulsar jay]

Anybody knows whats the state with reflection probes with URP and hybrid renderer v2? The reflection probes stop working when they are in a subscene

[misty wedge]

If I want to store a pointer to a struct on the managed heap, I would need to use GCHandle.Alloc(struct, GCHandleType.Pinned) first, then get the struct* or IntPtr to the struct, and dispose of the GCHandle when I don't need the pointer to the struct anymore. Is that right?

[rotund token]

yes

[misty wedge]

Is it possible to even get a pointer to the type of the struct (so for example Vector2*) outside of a fixed block?

Or do you need to use IntPtr

from e.g. GCHandle.AddrOfPinnedObject

[coarse turtle]

I guess stackalloc is another way of getting a pointer although it's only on the stack

[misty wedge]

Interesting, I've never seen that keyword. I assume this would let you create a stack array and manipulate it with a span without ever putting something on the heap?

[coarse turtle]

yea

[dense dew]

You can also use UnsafeUtility to convert a struct to a pointer and back. It will free you from writing everything inside a fixed block and disposing of GCHandle before leaving the scope. You can store it as either void* or IntPtr, it doesn't affect performance. Just use whatever seems cleaner for you (void* requires to write a lot of 'unsafe' keywords, storing IntPtr requires converting it to void* before passing to methods).
https://docs.unity3d.com/ScriptReference/Unity.Collections.LowLevel.Unsafe.UnsafeUtility.html

[pliant pike]

Basic question but does anyone know if dots physics is simple and robust enough even for an idiot like me to use currently

[viral sonnet]

There's a bit of a learning curve and it depends on your workflow. If you use subscenes and Hybrid Renderer it's quite easy to setup

just looking at the physics example helped me quite a lot to understand how it's working

[pliant pike]

Ok thanks I guess I'll give it a try

[calm edge]

why do you need subscenes for physics?

[misty wedge]

Are there any restrictions on using generic structs inside components?

[viral sonnet]

faster and easier conversion. it's not needed really but much of the setup is handled that way more conveniently

[hot basin]

should ComponentTypeHandle be used with IJobEntityBatch? (i'm still on 0.17)

is GetComponentTypeHandle<>() or batchInChunk.GetNativeArray(); heavy or something?

do I make less allocations with this compared to entityQuery.ToComponentDataArray<>()?

[coarse turtle]

Yes. When you do ArchetypeChunk.GetNativeArray(ComponentTypeHandle<T>) - Unity internally takes the pointer where the component data is and assigns it to the NativeArray's pointer (so its actually aliasing the pointer)

[hot basin]

so is this preferable way to iterate entities/entityqueries?

[coarse turtle]

I think with ToComponentDataArray, there is a memcpy involved somewhere in the function chain, but I forget exactly where

[coarse turtle]

yea - I'd imagine so

[hot basin]

thanks!

is new NativeMultiHashMap<Entity, NativeArray<float>>() good or will it bite me later?

[viral sonnet]

personally I use this var translations = (Translation*) batchInChunk.GetComponentDataPtrRW(ref Translation_ReadHandle);

makes it possible to use a ref then ref var position = ref translations[i];

There's not much difference apart from the small overhead of checks Unity is doing in GetNativeArray. I just like it more that way.

reason why I started this was that burst has a much easier time to vectorize pointers

don't ask me why

some methods I just could not get it to work even though the code did 100% the same thing

[viral sonnet]

Uhm, I'm pretty sure NativeArray in a NMHM doesn't even compile

there's also no reason NativeMultiHashMap<Entity, float>() will have a float[] for every key entity. That's what the multi implies

[rotund token]

https://www.youtube.com/watch?v=xSAbPPFINlM
fun little rollback test. save system wasn't designed for this but handles it pretty well.
0.5ms to serialize, 0.4ms to deserialize

[hot basin]

it compiled that's why i'm asking 😄

[rotund token]

did it run though?

[hot basin]

nope

[hot basin]

sry my bad 😄

[hot basin]

nice!

[viral sonnet]

Very nice indeed. How much data does this need?

[rotund token]

im only saving translation/rotation/physicsvelocity in this particular demo

let me see what size it was

[viral sonnet]

is every frame saved or how does this work?

[rotund token]

yeah i was saving every frame for this

you could do whatever you want

for a 10min demo every frame is just easiest to implement

EntityManager.CreateEntity(typeof(Save));

is literally all you do to save a frame

[viral sonnet]

Braid done easy or what?

Or rather, how to code Braid in a one liner 😄

[rotund token]

cost of 24 frames in data

first frames clearly compressed better

probably a lot of 0 velocity

and all rotations would have been identical

so yeah im only compressing per component atm

if i do a final compression pass on the entire data it reduces size by another 20-25%

but im also only use lz4 so you could compress even better with something else

originally i compressed over whole set but thats 1 thread large time

[viral sonnet]

cool, maybe you can use the netcode delta compression.

[rotund token]

doing it per component means i can compress each component in parallel was much faster

[rotund token]

well i have no interest in taking this further

this is a save library not a rollback library

just thought i'd see how it did

makes an interesting/fun demo

[viral sonnet]

hehe, yeah. don't overdo it 😄

although, if you ever want to take it further that would be a sick feature

what are you planning with the save library? i.e asset store or not

[rotund token]

saving? 😄

oh you edited

not sure

it's pretty done i think

i'm just integrating with my demo game atm

to see pain points

the only thing i'm not sure about for usability is the prefab registration

if you're just saving subscenes / existing entities you don't need prefabs, it just gives everything a unique id and maps to it

but if you instantiate at runtime the system needs to find the prefab for it

which is handled, it searches for a matching prefab in the world

but how to uniquely ID these could be a bit awkward for a user

like if you have a prefab in 2 separate subscenes

they appear as the same thing

now personally in my project i have a single subscene with all my prefabs

then use weak asset references to avoid having duplicate prefabs

but do i want to ship or even enforce this on others

(it is how unity builds their demos as well though)

prefab per subscene is really bad memory use so its a preferred pattern i think though

[viral sonnet]

didn't unity add cross scene linking now?

[rotund token]

cross scene linking isn't so much the issue, it's if you link a prefab

each subscene gets its own SceneSectionData

which ties it to the subscene

[viral sonnet]

hm, yeah, I'd enforce saved entities in one subscene

maybe there's a good solution for it. I don't know but it sounds like quite the hassle

[rotund token]

i know my solution works really well but it requires users to use it

i was hoping to make this completely non-invasive

which apart from this it is

can drop it in any project and it should just work

[viral sonnet]

that's pretty awesome and honestly something I'd expect from a dots asset

[rotund token]

(well it'll just work saving translation, rotation, physics velocity)

if you want to save your own component need to add [Saved] to it

which is the point, it shouldn't save everything

[viral sonnet]

oh wow 0/10 too much work! /s

sounds all great, looking forward to it

I wrote a non-generic save system in 2019 (18?) and even though that was quite easy, lots of code was involved to add new comps and then instantiation. I would rather use something like this

[rotund token]

yeah this is the 3rd save system i've worked on

[viral sonnet]

does it use source gen?

[misty wedge]

Very nice! Is this still using the 3rd approach you mentioned?

[rotund token]

yes

[rotund token]

no

no code gen

i just use dynamictypehandle

with typeindex

[viral sonnet]

oh, and with that you get attributes?

[misty wedge]

How are you liking this approach compared to the second one? Since I also need to move away from SerializeWorld eventually 😢

[viral sonnet]

or is this reflection based?

[rotund token]

no reflection except to get the attribute at the start

[viral sonnet]

ah okay, was wondering how that would work

[rotund token]

            {
                if (type.TypeIndex == 0)
                {
                    continue;
                }

                if (type.Category == TypeManager.TypeCategory.ComponentData)
                {
                    if (type.Type.GetCustomAttribute(typeof(SaveAttribute)) != null)
                    {
                        this.AddComponentSaver(builder, type.TypeIndex);
                    }
                }
                else if (type.Category == TypeManager.TypeCategory.BufferData)
                {
                    if (type.Type.GetCustomAttribute(typeof(SaveAttribute)) != null)
                    {
                        this.AddBufferSaver(builder, type.TypeIndex);
                    }
                }
            }```

        {
            var stableTypeHash = TypeManager.GetTypeInfo(typeIndex).StableTypeHash;
            var saver = new ComponentDataSave(builder, stableTypeHash);
            this.savers.Add(saver.Key, saver);
        }```

just gets all components with the attribute from the typemanager

then from there it just uses dynamictypehandle to write out the data

went over it a bit more on the forums

[misty wedge]

And you migrate if any type hashes change?

[viral sonnet]

damn, sometimes burst needs like over a minute to compile -.-

not sure what that's about. it's sooo annoying that safety checks defaults to enabled and I have to turn it off every time

[rotund token]

yeah

[misty wedge]

I guess I will be going with something similar once I get around to it, this approach seems really solid

[pliant pike]

its really cool

[rotund token]

i wrote a couple of migrations depending on how comfortable user is with pointers (high performance) and just easy

[pliant pike]

it seems like its much simpler to do something like this in dots than monobehaviour

[rotund token]

    public class TestComponentForMigrationMigrator : ComponentDataMigrate<TestComponentForMigrationMigrator.Before, TestComponentForMigrationMigrator.After>
    {
        public override ulong From => 1392796844747678277;
        public override ulong To => 17841778420484547873;
    
        protected override void Migrate(ref After newComponent, in Before oldComponent)
        {
            newComponent.Value = oldComponent.value;
            newComponent.NewValue = 5;
        }
    
        public struct Before { public int value; }
        public struct After { public int Value; public int NewValue; }
    }```
is like the easy version for users

i wrote a typedatabase to record old types

so you can get info easily

so you can find your old components rather than just seeing a magic number

[misty wedge]

Awesome. Let me know if you ever make it public or anything, I'd happily pay for something like this

[viral sonnet]

the data oriented approach is what makes it easier really

[pliant pike]

yeah just having to add saved to the data is just ridiculously easy on the users end

[viral sonnet]

it is and while this could be done with monobehaviours, the strict DOD limitations are hardly followed in OOP approaches so it all gets kind of messy

[pliant pike]

it kind of amazes dots hasn't taken off years ago

dots seems way simpler and more logical to me than oop

[hot basin]

oop is to big in the industry

outside of game developement

to be used in broader context here

or I should say was

[viral sonnet]

kind of crazy, creating spellCreateData and directly executing the spell calculations -> 11.2ms writing the spellCreateData and then calculating it. 8.0ms + 6.85ms. for some reason one thread is completely idle. nativeThreadIndex is weird...

[pliant pike]

I know there is a few obscure devs that use it though

[hot basin]

it's easier for companies to find someone with experience with oop than dod

so oop is economically more viable

[viral sonnet]

I dunno, we used MVC around 2008 for web development and that was already 100 times better. With long running engines you have the problem of technical debt so moving away from OOP is hard

And I also don't know what makes someone "good" in OOP. The spaghettis I've seen ... /shudder I consider someone good when he doesn't write endless amounts of spaghetti code. If he's capable of that, he's also capable of DOD

but let's face it. most programmers suck and should do something else

[pliant pike]

yeah from what I've seen most oop code is about obfuscating data and making things look more complicated than they are

[viral sonnet]

to their defense, corporate loves OOP. it's like the dream came true. have 100 people work on 100 different things and it all comes together. but yeah, that's what it is. a dream. the reality is that you have people just dedicated to clean up the mess.

when you can throw 100s of people at something management just doesn't care much. they rather see progress

[pliant pike]

I bet progress would be faster if they used something like dots though

[viral sonnet]

and progress can be pretty great in OOP. just code something. others will figure it out or write endless amounts of wrappers around it

[misty wedge]

I think having 100 people work on 100 separate things and having it come together at the end is hard no matter what you do

[viral sonnet]

exactly, it's impossible in my opinion. a few core people have to work on the core. when that's solid, and this takes time it gets more manageable

[misty wedge]

I think it depends very much on the type of application. With people moving away from monolithic applications in spaces like web development cooperation becomes easier as long as your interfaces are well defined

[viral sonnet]

tbh enterprise code today is pretty solid. it's not the mess that it was years ago

[misty wedge]

I think you will have very different experiences based on the company 😅

[viral sonnet]

DI and proper patterns that proofed solid have streamlined the process somewhat

that's for sure 😄

enterprise code as in, coding style.

[viral sonnet]

the nativelist block reserve is really one of the sickest optimizations I've done. Can't thank you enough @rotund token 😄 It even blows NativeStream and UnsafeParallelBlockList out of the water. As it's pre allocated it has nearly 0 impact

[rotund token]

yep

if you know your upper limit and code can handle allocating in chunks its about as fast as it can get

[viral sonnet]

I still have to figure out how I can handle the capacity in a better way. Right now I'm allocating much more than I need. I've casters, triggers and buffs that potentially calculate spells but all are unknown until it happens.

although this is much more of a problem in the stress tests I'm doing. In real scenarios it shouldn't be possible to hit the limit

uhm ScheduleParallel(testList.Length, 64, Dependency) how does this work in an IJobFor? There's only public void Execute(int index) - no inner loop count

guess I'm looking for IJobParallelForBatch

[rotund token]

inner is how the the scheduler works

it'll work on 64 elements per thread before trying to find a new range

its purely an optimization value

[viral sonnet]

6.04ms -> spellcastjob + write to list
7.32ms -> calc from list

instant calc method:
5.27ms -> spellcastjob
11.14ms -> spellcastjob + calc -> 5,87ms for calc``` this confuses me right now

calculating from the list is 1.45ms slower. I'd have expected it to be on par, maybe somewhat slower but not that much

aha! my block reserve size was too small. 4.58ms now. faster than the previous. noiiiice

i love when i lose some time but gain more 😄

[worldly isle]

I have been trying to find this but don't see it anywhere is there a way to say have a system with a group that have all the entities that have had a particular component added since last run. I don't know if I said this well or not...

Anyone have any ideas about this?

[haughty rampart]

you could add a tag to all the newly created and remove the tag afterwards

[muted field]

can jobs read data from static lists/arrays assuming it is only allow to read from them and not write ?

or do they also have to be native containers

[rotund token]

only static readonly

burst supports something like

public static readonly int[] Indices = new int[] {1,2,34,5};

[muted field]

thats fine

as long as i can read it its all i need

thanks!

[covert patrol]

which unity version is stable for the dots and some child packages not appering in 2021 i decided to learn dots which version should i use ?

[muted field]

actually, what if the array has a reference type will that still work ?

[rotund token]

probably not

[haughty rampart]

2020.3

[worldly isle]

So there is no built in way to do it then?

[haughty rampart]

no. why would there

[covert patrol]

there is no entitites packages in 2020.3 doesnt appear in package manager how can i install them

[haughty rampart]

you have to add it manually

[covert patrol]

thanks in some videos i see in unity registery so i confues i will add on git thanks

[muted field]

hmm thats a pain since i need to do jobs on a graph of nodes

😦

[rotund token]

#archived-dots message

i play graphs back in burst jobs

you dont need references to make a graph

[muted field]

well my node is a class and my edges are a class too

(pathfinding stuff)

[rotund token]

well yeah that wont work in a job

[muted field]

you saying i should use structs for them ?

[rotund token]

need to build it from structs and pointers

[muted field]

pointers ? so ill have to use unsafe c# to setup my graph data?

[haughty rampart]

'unsafe'

[muted field]

yes the context unsafe 😛

[rustic rain]

this is so annoying

some kind of compile error

in VS

in source of entities

[rotund token]

what version of vs

[rustic rain]

2022

[rotund token]

think i saw a few of these when i upgraded (in rider)

had to clear library to get rid of them

haven't come back though

[haughty rampart]

just wait for the release instead of fiddling with the code yourself tbh

[rotund token]

release of what?

[haughty rampart]

entities

[rotund token]

if you're waiting for 1.0 that's going to be a while...

any this issue happens just in the default packages folder nothing to do with changing the package

[haughty rampart]

a few more weeks for 0.51 for 2021 ....and 1 year for 1.0 2022

[rustic rain]

and what am I supposed to do now

kek

[haughty rampart]

not use 2022

[rustic rain]

uhm

[rotund token]

thats his version of visual studios...

not unity

[haughty rampart]

ohhh

[rustic rain]

yeah

[haughty rampart]

strange then

haven't experienced any issues with that

[hot basin]

Can someone explain to my why this job can't be vectorized?

Remark: ChunkDataUtility.cs:65:0: loop not vectorized: loop control flow is not understood by vectorizer
Remark: ChunkDataUtility.cs:65:0: loop not vectorized: loop control flow is not understood by vectorizer
Remark: ChunkDataUtility.cs:65:0: loop not vectorized: loop control flow is not understood by vectorizer
Remark: ChunkDataUtility.cs:65:0: loop not vectorized: loop control flow is not understood by vectorizer
Remark: EntityQueryManager.cs:752:0: loop not vectorized: loop control flow is not understood by vectorizer

[BurstCompile]
    public struct EvaluateAxisQuadratic : IJobEntityBatch
    {
        public ComponentTypeHandle<Axis> AxisTypeHandle;

        public void Execute(ArchetypeChunk batchInChunk, int batchIndex)
        {
            var axes = batchInChunk.GetNativeArray(AxisTypeHandle);

            for (var i = 0; i < batchInChunk.Count; i++)
            {
                var axis = axes[i];
                axis.value = (axis.m * math.pow(axis.input - axis.c, axis.k)) + axis.b;
                axes[i] = axis;
            }
        }
    }

is this because of floats? (all fields of Axis are of float type)

[rotund token]

what does the burst code look like

to start with

batchInChunk.Count

try use axes.Length

ArchetypeChunk .cs
public int Count => math.select(m_BatchEntityCount, m_Chunk->Count, m_BatchEntityCount == kUseChunkCount && m_Chunk != null);

[rotund token]

also this doesn't look like your actual look. it looks like the chunk code outside the job. if you go look at the burst inspector you'll see like 2500 lines of the code are not your job

[rustic rain]

bruh, this error remains

I can't use debugging because of it

since VS can't build it

I even tried 2019 VS

[rotund token]

you shouldn't have to build to attach a debugger

[haughty rampart]

Try rei stalling entities

[rotund token]

do you have a custom version of entities in your packages folder?

[rustic rain]

I tried reimport all

several times

[rotund token]

or is this in the package cache

did you delete your library or just do reimport all

[rustic rain]

all installed through dependency on hybrid renderer

[rustic rain]

just reimport all

[haughty rampart]

Just create a new project and move your assets over

[rotund token]

yeah dont bother with that just delete the library

[rustic rain]

all right

[hot basin]

that change added one Remark: ChunkDataUtility.cs:65:0: loop not vectorized: loop control flow is not understood by vectorizer

[rotund token]

yeah this has nothing to do with your loop

i have 15 of these for a job with 1 loop ^_^"

honestly the llvm ir optimisation diagnostics tab is pretty useless

i never use it

[hot basin]

so it's because of IJobEntityBatch per se not because of my code? should I even bother with IJobEntityBatch then (i'm still on Entities 0.17)

[rotund token]

why wouldn't it?

what are you proposing instead?

[hot basin]

get array outside and use IJobParallelFor instead

[haughty rampart]

Why are you not upgrading?

[rotund token]

for what benefit?

[rustic rain]

nnnnope, deleting library didn't help

[hot basin]

i'm bound to 2021 with my animations

[haughty rampart]

You should be using mecanim

[hot basin]

so are you saying that those lines in optimisation diagnostic are useless?

[rotund token]

yes

go look at the actual generated code

[haughty rampart]

Otherwise you won't even be able to upgrade with 1.0

[hot basin]

as long as unity doesn't provide anything remotely good I don't care, I'm using HR only because I don't want to write renderer for static objects right now

[haughty rampart]

Even 0.50 is already a lot better than 0.17

[hot basin]

i can wait until 0.51

[rotund token]

this is what your actual job code looks like

            {
                public float m;
                public float c;
                public float k;
                public float b;
                public Unity.Mathematics.float3 input;
                public Unity.Mathematics.float3 value;
            }```
assuming this was your struct

not sure if they were float or float3

[rustic rain]

Any other ideas how can I fix it?

[hot basin]

and when 1.0 hits I will rewrite my gpu animation instancer with this: https://forum.unity.com/threads/new-batchrenderergroup-api-for-2022-1.1230669/

[rotund token]

edit -> preferences -> external tools -> regenerate project files
otherwise download rider 😄

[haughty rampart]

Ok

[hot basin]

everything is a singular float

[rotund token]

makes sense

not sure why i thought they were float3

[hot basin]

it seems to vectorise? 😄

[rotund token]

not really it's all scalar

[rustic rain]

I guess I'll try rider xD

[rotund token]

student? it's free!

that's how they get you

next minute asking employer for it

[rustic rain]

not anymore, kek

[rotund token]

no sneaky email hanging around?! 😦

managed to use mine for like an extra 3 years after i finished uni

this was quite a while ago now though

(not for rider didnt exist)

[hot basin]

any idea why and how to change it?

[rustic rain]

no email in my uni at all xD

[rotund token]

dont think it likes pow

breaking it down

a quick google seems to imply yes it doesn't exist

if you konw your x,y are both positive you can use
exp(y*log(x))

doing this definitely simds the whole thing

like seriously the entire thing

ties all the way back into the schedule job code

[hot basin]

but I don't have y?

[rotund token]

pow(x,y)

[hot basin]

i'm calculating y there

oh that y

[rotund token]

dont actually know if this is faster though

you'll have to test it

[hot basin]

what assembly shows when it vectorize?

[rotund token]

you want to see lots of purple!

ps for wide
ss for scalar

go watch the burst talk on it

anyway enough optimising someone else code, need to optimise my own

[hot basin]

like this?

still ss

[rotund token]

cant do everything wide

also

if a loop has to be vectorized

you can use

Loop.ExpectVectorized()

UNITY_BURST_EXPERIMENTAL_LOOP_INTRINSICS

and it'll be a compile error if a loop isn't vectorized

very handy if you vectorize something to protect it from breaking in future

[hot basin]

ok thanks a lot!

[rotund token]

it's a bit finicky though

[hot basin]

there is also possibility to manually vectorise this with float4 everywhere

[rotund token]

often the only way to reliable get this to work

[hot basin]

but I wonder if I can easily reinterpret Axis to Axis4

[rotund token]

you know what, it's time to test IEnableableComponent, see what's broken

when i was looking at burst for your code it was doing a proper check for this

[dull copper]

Unity is apparently renaming the hybrid renderer to com.unity.entities.graphics

https://github.com/Unity-Technologies/Graphics/commit/13e6e83956f0b78d0fa44bbad8b624a054524279

I do welcome this change as it's now more obvious what it's for

[full epoch]

I updated dots related packages to the newest versions(0.50.1 entities etc) - everything works except mathematics 1.2.6 which should stay the old version (1.2.5 works for me)

[robust scaffold]

I just checked the changelog and it doesnt say anything about increasing the version of dependencies. But yea, if it compiles, it should work fine.

[robust scaffold]

https://docs.unity3d.com/Packages/com.unity.mathematics@1.2/changelog/CHANGELOG.html

Yea, nothing changed with 1.2.6 so you'll be fine

[robust scaffold]

"upgrade" to 2022.2 alpha... and burst crashes. FUUUUUUCK

thats what I get for trying to use Unity's automated upgrader. pain

[dull copper]

or.. maybe not use alpha

is current dots even compatible with it?

(haven't followed this part)

burst itself should work though

[robust scaffold]

Not using dots, just just base burst - jobs - collections

[dull copper]

having burst itself installed on latest alpha doesn't crash things here... so much be some script on your end that triggers it

why do you assume it's the auto-upgrader that messed it up? it's supposed to just adjust things for the changed API so things wouldn't break

if it straight up crashes, I'd try wiping the library folder in case there's some cached stuff from previous version that's incompatible (this wouldn't be the first time).. and then start looking at the scripts using Burst that might break it so you can pinpoint the source of the issue

[robust scaffold]

Yep yep, going through the process now. Happens every update yet I keep thinking the upgrader would work and the major update would work instantly upon launching.

[dull copper]

I can't say I've had similar experience on my end, never had issues with the autoupgrade scripts myself. of course this doesn't mean it wouldn't break for others.. maybe you've been just very unlucky for it to break on specific things you use it for

[robust scaffold]

Fuck. I think I narrowed down the crashing to somewhere in my shader stack. This is gonna be pain.

[remote crater]

What is the best way to do an array in a ComponentData struct?

I'm saying scew it and just making int a1;int a2;int a3; etc for now

I think I had this question for close to a year now... No one here or on the forums had a good answer that worked. Kinda silly, but its starting to hit my core architecture's foundation.

If I don't get help now, its going to be a costly refactor in the future.

I can get my game working, but I'm a fan of eloquent and beautiful looking code especially when it impacts functionality.

[whole gyro]

Do you need it to be in an IComponentData struct? If not, dynamic buffer is the recommended approach for having an "array" associated with an entity.

[muted field]

why cant jobs use reference types but regular threading can, whats the difference between jobs and threads?

[remote crater]

Thanks Graviton, I think you helped me, it's not pretty, but better than nothing.

[whole gyro]

I think you can technically put a FixedList* in an ICD, but you would need to have some guarantees about the number of elements and how many bytes each element requires.

Another option is to use a blob if the arrays are static and set up at conversion time

But dynamic buffers are definitely the easiest/cleanest of the 3 to use.

[viral sonnet]

many ways actually. Pointer and manual allocation/deallocattion. BlobAsset reference. DynamicBuffers. NativeMultiHashMap with entity as key.

[rotund token]

anyone know much about IJobBurstSchedulable

and any reason you wouldn't use this over IJob?

as far as I can tell it's IJob that can be schedule from burst code (ISystem)

which leads me to the question of there must be a difference otherwise why not just rewrite the IJob scheduler to support burst

[viral sonnet]

what information do we have about it? I just read it's removed in 0.50.

scheduling jobs from jobs was on my wish list. never heard of it before

haha, Unity is calling it 0.5 ... whoops 😄

[rotund token]

well i was not expecting it to be faster for me to write 10mill elements to a hashmap in single thread vs 24 threads
that locking kills it

in other news, wrote a new container NativeLookup which is basically just an alternate NativeMultiHashMap without the ability to remove elements and all my high performance batch operations built in

[upper tiger]

 CollisionFilter Filter = new CollisionFilter
                    {
                        BelongsTo = 0xffffffff,
                        CollidesWith = 1u << 3 | 1u << 5, // collide with terrain and world bounds
                        GroupIndex = 0
                    };

                    var pointDistanceInput = new PointDistanceInput
                    {
                        Position = localToWorlds[i].Position,
                        MaxDistance =  20f,
                        Filter = Filter
                    };

                    // this should get the closest hit
                    if (world.CalculateDistance(pointDistanceInput, out closestHit))
                    {
                          //... 
                    }                   

I cant seem to get my collision filter to collide with a layer 3 physics shape. It can get a distance hit with a simple cube just fine, but a complex mesh it never seems to trigger.

the "Belongs To" is mixed because I put it on both layer 3 and 5 out of frustration but didnt work

both the mesh object and the simple cube physics shape object have convert to entity with inject

[upper tiger]

I figured out the problem but not sure why. If you do calculate distance from within a mesh collider box, other colliders dont seem to register until you are out of the bounds of that collider

I want my objects to stay within the box but also avoid other things that are also in the box.

[rotund token]

not sure if related but from memory there is 1 big difference between unity regular physics (physx) and unity.physics and that is how collisions or at least raycasts work if you're within a collider

1 of them raycasts hit the other doesn't

[rustic rain]

thanks to your tips, guys, I am having a good progress

this is entityIdColor screenshot

certain problem tho is that it's not precise for some reason

entities are indeed rendered correctly, but left entity won't be calculated correctly

[rotund token]

Oh that's looking promising

If you're lookup is wrong from memory one issue is linear vs gamma

[rustic rain]

hm

thing is

look up is correct for some entities

but not for another

[rotund token]

Hmm

[rustic rain]

and also it gets ruined if I move my camera

I do recalculate camera matrix and etc

on each snapshot

hhhhmmm

I think I got the reason

I mixed colour bytes

and I can see jpeg like distortion

wait a secon

lol

this is so funny

I think I am reading pixel upside down

🤡

[rotund token]

Of yeah that's a common issue

[rustic rain]

sooo

to fix it

I just do resolution - positionOfPixel?

already testing

kek

[rotund token]

Yeah usually just flip the y

[rustic rain]

yyyep

it's good now

oh man, feels so good

Thank you so much

[rotund token]

Where'd you end up injecting this?

[rustic rain]

I pretty much copied that tool that other guy sent

[rotund token]

But yeah it's a great technique

[rustic rain]

overall looks like this

[rotund token]

It's also nice how accurate the picking is vs colliders

[rotund token]

Oh that's easier than I expected

Nice

[rustic rain]

I guess the only problem

it doesn't render sprites

since they don't have RenderMesh component

[rotund token]

Can add them in a separate loop though?

[rustic rain]

Probably

[karmic basin]

Nice 👍

[remote crater]

Can you test if an entity has an IBufferElementData attached?

I'm able to use them, but if I try and access it form an entity that does not have one attached, it errors out: ArgumentException: A component with type:TriggerIgnoreBufferElement [Buffer] has not been added to the entity.

My code: DynamicBuffer<TriggerIgnoreBufferElement> beA = World.DefaultGameObjectInjectionWorld.EntityManager.GetBufferfrome <TriggerIgnoreBufferElement>(entityA);

[rotund token]

you can use has component

[remote crater]

oh thanks tert

you rool

Worked like a charm, now I can use arrays properly with some semblance of a solid architecture.

[viral sonnet]

What am I looking at, default HashMap vs your lookup implemenatation?

[rotund token]

yeah pretty much

just interesting that single thread is faster than multi thread

for normal hash map

[viral sonnet]

dat interlocked stall. same problem with nativeList

[rotund token]

too many threads and interlock destroys me

that said, i'm still using interlocked for my implementation

[viral sonnet]

do you convert the keys/values from the hashmap to a list and reserve?

[rotund token]

            [NoAlias] this NativeHashMap<TKey, TValue>.ParallelWriter hashMap,
            [NoAlias] TKey* keys,
            [NoAlias] TValue* values,
            int length)
            where TKey : unmanaged, IEquatable<TKey>
            where TValue : unmanaged
        {
            var writer = hashMap.m_Writer;
            var data = writer.m_Buffer;

            var newLength = Interlocked.Add(ref data->allocatedIndexLength, length);
            var oldLength = newLength - length;

            var keyPtr = ((TKey*)data->keys) + oldLength;
            var valuePtr = ((TValue*)data->values) + oldLength;

            UnsafeUtility.MemCpy(keyPtr, keys, length * UnsafeUtility.SizeOf<TKey>());
            UnsafeUtility.MemCpy(valuePtr, values, length * UnsafeUtility.SizeOf<TValue>());

            var buckets = (int*)data->buckets;
            var nextPtrs = ((int*)data->next) + oldLength;

            for (var idx = 0; idx < length; idx++)
            {
                var hash = keys[idx].GetHashCode() & data->bucketCapacityMask;
                var index = oldLength + idx;
                var next = Interlocked.Exchange(ref UnsafeUtility.ArrayElementAsRef<int>(buckets, hash), index);
                nextPtrs[idx] = next;
            }
        }```

i added an AddBatch method for ParallelWriter

which is pretty much what my above NativeLookup is using

[viral sonnet]

mh, interesting. you can get rid of the memcpy with the method I suggested a few days ago

[rotund token]

i don't want to get rid of the memcpy though

[viral sonnet]

why not? 🙂

[rotund token]

because the memory i'm writing from is not safe to keep a reference to

because chunks can change while my jobs are running

[viral sonnet]

i mean, it would be another extension method. this one is perfectly fine. what's the data type for the keys/values you use? temp allocated nativeList?

[rotund token]

keys/values is a pointer to <entity,components> in chunks

i'm just direct copying entity data in

per chunk

[viral sonnet]

ah I see, ok, now I got you why you are not doing it 😄

that's clever 🙂

yeah, seems like the most convenient way and fastest way to index chunk data. pretty cool

[rotund token]

Inspired me to turn his into a scene view entity picking option again

I used to have this a few years ago and I missed it so finally adding it back is good

I can now just select entities with most, such a nice simple concept

[rustic rain]

Can anyone remind me how to get component data so I can use it in ForEach loop?

[shell berry]

You can use EntityManager.GetComponentData() before the foreach.

[viral sonnet]

So, this? https://github.com/JonasDeM/EntitySelection

what's nice about it, is that it doesn't need physics, colliders or raycasts.

[rustic rain]

yyyep

camera based selection

[rotund token]

ive implemented this technique a few times since i first saw the medium post in 2017.
that particular repo does a DrawMesh per entity though so kind of falls apart at high counts

[viral sonnet]

you always go one step beyond 🙂

any idea to write better profileMarker code? I hate this

[rotund token]

i always use Auto personally - having to do an end is horrid

but i never use that many markers

1 per method max

[rotund token]

what i mean by this is use DrawMeshInstanced instead

Something like

            {
                var mesh = chunk.GetSharedComponentData(renderMeshType, this.EntityManager);
                var entities = chunk.GetNativeArray(entityType);
                var localToWorlds = chunk.GetNativeArray(localToWorldType);

                this.matrices.Clear();
                this.colors.Clear();
                propertyBlock.Clear();

                foreach (var entity in entities)
                {
                    var color = IndexToColor(entity.Index);
                    map.Add(color, entity);
                    this.colors.Add(color);
                }

                propertyBlock.SetVectorArray(colorID, this.colors);

                this.matrices.AddRange(localToWorlds.Reinterpret<Matrix4x4>());
                var matrixArray = NoAllocHelpers.ExtractArrayFromListT(this.matrices);

                cmd.DrawMeshInstanced(mesh.mesh, mesh.subMesh, mat, 0, matrixArray, this.matrices.Count, propertyBlock);
            }```

[viral sonnet]

yep, I'm using Auto now too. Much cleaner now.

I'm always amazed how much broad changes you can make in DOD. Like enabling multiple code paths for one feature just because the paths already exist. In OOP that was hardly doable.

[muted field]

so with dots everything has to be value type not reference type but what if you need a struct with an array of points like a polygon shape with vertices

since arrays are reference types how can i represent it in a struct

plus i need an array of each polygon so i need up with array of polygons which have array of vertices

so how can i setup the data like that using structs 🤔

[viral sonnet]

first of all, no, not everything has to be a value type. it can be pointer too. second, where do you want to store this data?

[muted field]

i need to keep a persistent list of all my polygons

but im doing jobs on all the data

so i cant use classes

[viral sonnet]

You can keep that data in a NativeList/NativeArray in one of your systems then

[muted field]

i dont seem to have native list in my options only native array

is it in a separate package ?

[viral sonnet]

it's all in Unity.Collections

[muted field]

its not for me

[viral sonnet]

Do you have this package installed? https://docs.unity3d.com/Packages/com.unity.collections@1.2/manual/collection-types.html

[muted field]

let me double check - should have since native array has worked fine

ah okay so native array is built in the rest is in a package

that explains it

[viral sonnet]

if you don't have burst installed, also add that 🙂

[muted field]

so its allowed to have native lists in a struct then

does that mean when you pass a native list (or array) in methods the whole collection is copied ?

so i need to use ref every time

[viral sonnet]

while you can use ref, the only thing that is copied is the pointer, length and capacity + some safety checks

[muted field]

oh

[viral sonnet]

the actual data is in unmanaged memory

[rotund token]

native containers wrap a pointer so effectively behave like a class not a struct

unsafe containers don't do this though and act like a struct

[muted field]

i see. interesting new way to think about code design compared to my old oop ways

[viral sonnet]

start with some easy burst examples and enjoy the speed 🙂

[rotund token]

got back from shops, did 1 more quick pass on this entity selection bursting the setup
@rustic rain if performance is something you need have a look at this - bursted setup + mesh instancing

you won't have access to like half the extensions though by default 🤣
e.g. AddRangeNative, NoAllocHelpers.ExtractArrayFromListT, AddBatchUnsafe

[muted field]

wait it seems i cannot use my struct it says cannot use non nullable in native list ... is that because my struct has a native array in it ?

[rotund token]

you can't store a native container inside another native container

you can store unsafe containers inside native containers just know by default burst will always compile this the 'slow' way (still fast)

[muted field]

oh jeez this is getting complicated quickly

[rotund token]

also there are Fixed* containers where you don't have to worry about memory

FixedList64Bytes<> for example

which is really just a chunk of fixed memory that behaves like a list

[muted field]

why wont they allow a container of structs with containers =/

[viral sonnet]

what's your code? I thought you just have a struct with a NativeList/Array inside it? that should be fine

[muted field]

    public readonly struct Area
    {
        private readonly NativeArray<float2> vertices;
        public readonly NativeArray<float2> Vertices => vertices;
        public Area(in NativeArray<float2> vertices)
        {
            this.vertices = vertices;
        }
    }

    private NativeList<Area> _areas = new NativeList<Area>(); // denied

[viral sonnet]

ah, yeah that won't work

[muted field]

what would be the right way to setup my data for this ?

[viral sonnet]

the quickest I could think of is with pointers and manual allocation/deallocation. another is to flatten the whole struct or use 3 flat arrays and get rid of the area struct alltogether

[rotund token]

need to know what your actual problem is

[muted field]

my goal is to check if areas overlap another area

planning to use parallelfor job for it

[rotund token]

how large are your polygons

number of points

[muted field]

some could have 1000+

[rotund token]

yeah ok if so large/variable you probably need unsafe container or pointers

i.e. just change your NativeArray<float2> vertices to UnsafeList<float2>

just remember unsafelist behaviours like a struct and needs write back

[muted field]

what do you mean by write back

[viral sonnet]

private NativeList<int> verticesLengths``` one way to flatten. you'd have a linear list of all the areas and their vertices and another array that has the length of each area

[rotund token]

        {
            var nativeList = new NativeList<int>(0, Allocator.Temp);
            var unsafeList = new UnsafeList<int>(0, Allocator.Temp);
            
            Add(nativeList);
            Add(unsafeList);
            
            Debug.Log(nativeList.Length); // 1
            Debug.Log(unsafeList.Length); // 0
        }

        private void Add(UnsafeList<int> list)
        {
            list.Add(1);
        }

        private void Add(NativeList<int> list)
        {
            list.Add(1);
        }```

[rotund token]

i was about to suggest this actually

i'd even consider vertices instead being 2 values, start + length

[muted field]

why does that give 0 ? 🤔

[rotund token]

        {
            var i = 0;
            Add(i);
            Debug.Log(i); // 0
        }

        public void Add(int i)
        {
            i += 1;
        }

[muted field]

i might opt for that though if unsafelist works it does make the code slightly cleaner

[rotund token]

for the same reason this is 0

[muted field]

so then how do you actually add to an unsafe list if .Add(1) doesn't actually add to its length

[rotund token]

pass it by ref

and write it back if it's stored in a native container

[viral sonnet]

or return it in the method but ref is better

[rotund token]

yeah or that

just imagine it's an int

[muted field]

so just do list.Add(ref number)

[rotund token]

no

        {
            list.Add(1);
        }```

the container has to be by ref

[muted field]

oh so unsafe containers do not get passed like references like native containers do

aka they dont wrap a pointer to it

[rotund token]

            var list = lists[0];
            list.Add(1);
            lists[0] = list;```

just remember this as well

usually i wouldn't even bring up pointers/unsafe containers for someone newish to dots unless they had a solid c/c++ background

[rotund token]

and i would seriously consider this solution instead if your polygons are reasonably stable once created/destroyed

[muted field]

got to learn it eventually 😛

[rotund token]

not really

i'd say only 20% of our devs know anything about low level code at work

if you're into engine development then yes 100%

but if you're just a gameplay dev not really

[muted field]

well for unity sure but if you use c++ it would be nice to learn

[rotund token]

obviously if you're solo and have to do everything then yeah it'll probably creep in at some point

as you don't have anyone else to come in to optimize it

[muted field]

i am indeed solo

-_-

var list = lists[0];
list.Add(1);
lists[0] = list;

why can't we do:

lists[0].Add(1);?

[rotund token]

lists[0] doesn't return by ref

it's a local copy

you can add a ref return method

            where T : struct
        {
            return ref UnsafeUtility.ArrayElementAsRef<T>(array.GetUnsafePtr(), index);
        }```

[muted field]

ok so i need to think of it like a struct

[rotund token]

then you could do
lists.ElementAt(0).Add(1)

[muted field]

i might add some extension methods

to make my life easier

would be nice if unity did that so noobs can use them without having to think about all this too 😛

[rotund token]

i think unity's preference is users don't need to touch unsafe code

btw though they do have some of these

like dynamicbuffer has elementat method built in

just nativearray is missing it

[muted field]

true but even my case is a simple one that many people might need to do

with the other suggestion i could just opt for 1 array with [size][0]..[n][size][0]..[n] i guess

[viral sonnet]

InvalidOperationException: The Unity.Entities.EntityCommandBuffer has been declared as [ReadOnly] in the job, but you are writing to it. How odd is that? I never had any issues with it and the command buffer isn't read only. Yet Unity is complaining all of a sudden. hm

[rotund token]

are you passing it in a weird way

where its hidden from the scheduler so the safety system can't inject into it

i.e. passing it in via a pointer or something

[viral sonnet]

indeed I was 😄 was located in a struct that I accidentally marked ReadOnly

I've made such drastic changes today. I hope not everything comes crashing down lol

if it works out I've streamlined the shit out of the spell system 😄

aw man, I've changed duration from ushort to float and now all my authoring data is gone. Yet I was pretty sure as long as the name stays the same the data in the monoBehaviour persists

[rotund token]

you cant interpret ushort to float

if you had changed ushort to int

it would have kept

[viral sonnet]

huh? why is the cast not possible? float has enough range?

[rotund token]

unity doesn't convert them

the type isn't even stored on disk

[viral sonnet]

lazy unity ... 😄 yeah, the serialized file is 1:1 with ushort or float. glad I had a backup

must be some oversight. there's no technical reason I can think of

[rotund token]

if no . not float

😄

(but nah it doesn't work like that )

[viral sonnet]

some legacy UnityJS code 🤣

[rotund token]

that said, i just converted an int to a float field

and it works fine

data still there

[viral sonnet]

yep, I did that pretty often over the years. either it's special for ushort or something else tripped up

[rotund token]

was it a list/array of ushorts?

[viral sonnet]

nope

[molten flame]

Does anyone know if there is an extension or something to access the hit child collider of a compound collider in a collision query?
I'm just working off the advanced collider parade demo which has everything I need but I can't help but feel that this is obvious functionality that has some extension hiding somewhere.

[minor sapphire]

I do this, let me check

if (collisionWorld.CastRay(raycastInput, out RaycastHit hit) &&
    !(projectile.Owner == hit.Entity && elspsedTime - birth.Time < 0.1f) &&
    !projectile.Destroyed)
{
    projectile.Destroyed = true;
    collisionWorld.Bodies[hit.RigidBodyIndex].Collider.Value.GetLeaf(hit.ColliderKey, out ChildCollider childCol);
    Collider* childColPtr = childCol.Collider;
    Collider projCol = *col.ColliderPtr;

If there is an obvious extension lying around let me know but otherwise... make an extension? 😄

I'm not aware of an easier way at the moment

[molten flame]

Thanks @minor sapphire! I did stumble across your posts on the forums discussing this topic.
I'm implementing a destructible object system too.

[minor sapphire]

are you by any chance generating voxels?

[rotund token]

Do you also have crazy cool lights?!

[molten flame]

No voxels or crazy cool lights for me.
I'm just using Rayfire to generate some high detail destructible walls.. its the one over the top ECS thing I want to do lol.

[minor sapphire]

I've been spending a stupid amount of time trying to optimise the CompoundCollider.Create call

made my own version of it, removing as many allocations and memcpy as I can, doing the BoundingVolumeHierechy generation upfront so I can pre allocate the blob memory and write directly to that lol

[muted field]

im getting this error:
Allocator Unity.Collections.AllocatorManager+AllocatorHandle must not be None or Invalid

not sure why though

theres no way to set a allocator to unsafe list from what i can tell

oh wait

i found it

[rotund token]

You still need to allocate and deallocate all unsafe containers

The difference is they are unsafe so they don't have safety to tell you

[rotund token]

I'm actually doing something similar

[muted field]

if they persist for lifetime of the game do you still need to dispose when you exit the game or will that dispose auto ?

[rotund token]

Are you using ray fire just to create the fractures

Or integrating the whole thing

[rotund token]

Yes you need to dispose

[muted field]

okay

i wonder if any one has released a nice wrapper around unsafe collections that handles some of the pains of dealing with them

[molten flame]

Oh neat!
I'm just doing the fractures at authoring time using the Shatter script.
Would be neat to have the Unyielding feature though, I really liked how you could build something that was somewhat structurally sound.

[rotund token]

That's literally the native collections

They wrap an unsafe collection

[muted field]

yeh but cant have them in a struct that goes in a native collection

[rotund token]

I.e nativelist has a pointer to unsafelist

[muted field]

if thats all they are im surprised they wont allow collections that have structs containing collections

unless its on their roadmap to allow it some how

[rotund token]

Btw even when they finish removing dispose handle they won't allow nested native containers

[viral sonnet]

what's going on here? I have to complete the combined handles. otherwise unity complains about it, yet newHandle is a dependency for the job? anything obvious I'm missing here?

[rotund token]

They will have static analysis to prevent it

[muted field]

i dont know what you mean by that

i know you cant have NativeList<NativeList<>> but NativeList<MyStruct> which has a NativeList i feel like that should be possible if i can do it with an unsafe list and Native Lists are just wrappers

[rotund token]

You can't have them nested because dispose handle is a class

They are removing this

In theory this means they can be nested

However they will add a static analysis to prevent it

I.e make it a compile error

[muted field]

why would they prevent it - seems like a nice idea to allow it surely?

[rotund token]

What's the actual error

[rotund token]

Because it's super costly to scan pointers of pointers of pointers to inject the safety handle

You could have 1000 lists inside each other

[muted field]

is it not costly to use an unsafe list like im doing though ?

[viral sonnet]

InvalidOperationException: The previously scheduled job UpdateCombatEffectsJob writes to the Unity.Collections.NativeList`1[NZSpellCasting.CreateBasicSpellDataComplex] UpdateCombatEffectsJob.JobData.casterCalulateSpellsList. You must call JobHandle.Complete() on the job UpdateCombatEffectsJob, before you can read from the Unity.Collections.NativeList`1[NZSpellCasting.CreateBasicSpellDataComplex] safely.

[rotund token]

There's no safety in unsafe

[muted field]

oh so the safety is what made it slower

[rotund token]

Btw I do actually have a list that you want

Think I call it unsafelistptr

[viral sonnet]

the system in question is handle3

[rotund token]

Acts like a native list without safety

On a tram will send you it when i get home

[muted field]

oh nice. ping me when you have it i will take a read of it 🙂

thanks!

[rotund token]

I'd need to see the full source but what is caster calculate spells list

Is it a native list

[viral sonnet]

yes

[rotund token]

And you're reading length

On main thread

When job could be writing to it?

[viral sonnet]

ah, good catch!

[rotund token]

If this is the issue you probably just need deferred job

Hard from me to be certain standing in a crowded tram with limited source code =D

Back in a bit

[minor sapphire]

excuses 😄

[viral sonnet]

seems like there is no defer job for IJobParallelForBatch :/

eh, completing at this stage is okay. was just wondering why that happens. sometimes the job dependency system still trips me up completely 😄

[rotund token]

Think it has batch option built in

It hasn't been update for new standards

Still used schedule for parallel

[viral sonnet]

just index 😦

[rotund token]

Just have to calculate it yourself 💩

Bit of a pain

[minor sapphire]

Everything will be ok T_T

[rotund token]

Nope, better start refactor 71 enzi

[muted field]

that phrase should be slogan for all of DOTS 😛

[rotund token]

🤷‍♂️

I don't find it a pain

Just need years of experience first!

[minor sapphire]

it is a bit of a pain though 😛 just one that's worth it haha

[rotund token]

Then can do the basics

[viral sonnet]

I rather deal with pain than the Mono brick wall 😅

thanks for your help tertle! have a good night all o/

[muted field]

🥲

i would like my game complete before then lol

[minor sapphire]

nice joke

It's the same humour I use every day

[rotund token]

Why release when there are still optimisations to be had!

[muted field]

lol so true

i think this is where im getting an error:

        public void Add(in Vector2 vertex)
        {
            float2 v = new float2(vertex.x, vertex.y);
            vertices.Add(v);
        }

ObjectDisposedException: The UNKNOWN_OBJECT_TYPE has been deallocated, it is not allowed to access it Unity.Collections.LowLevel.Unsafe.AtomicSafetyHandle.CheckWriteAndBumpSecondaryVersion (Unity.Collections.LowLevel.Unsafe.AtomicSafetyHandle handle) (at <823fb226a3f9439cb41fdcb61f9c86a1>:0)

but am not sure how to fix it

very hard to decipher the error's origin

[minor sapphire]

vertices?

[muted field]

vertices is an unsafe list

[minor sapphire]

check IsCreated

[muted field]

hmm strange maybe thats not the function causing it then

ok it might be when im adding to my native list:private NativeList<Polygon> _areas = new NativeList<Polygon>();

but that is also created

oh actually its not created 🤔

so i can't declare a new native list as a class level field ?

can they only be created within functions

hm it works if i create the native list in a function but not if its declared as a class field

[rustic rain]

so...
I assume there is no graphic solutions in DOTS yet, so in order to draw some lines and etc, I need to use GameObjects?

[rotund token]

i have my own draw library but cant share it atm unfortunately

but there's one in the store that's even better https://assetstore.unity.com/packages/tools/gui/aline-162772

highly recommend something like this (grab it on sale)

[pliant pike]

I've used that its pretty good, although it does bog down your framerate if you try and render over 100,000 cubes or so

[rustic rain]

yeah, sadly I'm not really in position to purchase anything

Is there any free drawing library?

[pliant pike]

I mean its not that hard to use gameobject stuff instead

[rustic rain]

Well, that would require quite some sacrifices

[pliant pike]

yeah I guess it depends on what you want to use it for, if its just for debug purposes then it shouldn't matter surely

[rich tusk]

Is there any way I can find the problem why my capsule doesn't render after I set up my new project? I'm using URP with Entities cause I found it will render nothing at all if I don't use URP or HDRP (According to the Unity forum).

Not sure if there is something wrong with my capsule or something, I recreated one and it's still not rendered. the chunk I checked is normal to me but I'm new to ECS so don't know is there something missing.

[pliant pike]

are you sure its not fallen through the floor or flying off somewhere

[rich tusk]

It stays in the position in the scene view when play mode is on.

And when I tried another cube I removed the collider, didn't add the RigidBody, so it will not fall I guess.

[pliant pike]

have you set it up exactly like in here https://docs.unity3d.com/Packages/com.unity.rendering.hybrid@0.50/manual/creating-a-new-hybrid-renderer-project.html

[rich tusk]

Problem solved. I did what I miss in the guide, then created a material to solve the shader error, and the capsule rendered as I expected. Thx for the help!

[rustic rain]

How to do it correctly?
Basically, I need to access CompanionLink component of entity after it was added only once in order to add some other component.

Do I just create simple system where I target query which doesn't have yet the component I want to add?

[solemn hollow]

I dont know your exact usecase but you should probably use managed IComponentData on the entity instead of modifying the companion if all you need is to use managed Data for the entity

If you really need to access the Companion you can as you suggested query it. but i dont think you can query self added components directly on the companion. keep track of which companions are already processed by adding a tag to their entity which you can then exclude in the query.

[rustic rain]

I use companion in order to avoid having another companion game object system

I simple just add my other game object as child

and do whatever I want with it through my own component with reference to GO component I need

LineRenderer in this case

it's all irrelevant though

All I need is to somehow add component to entity only once, and just asking whether there is intended way for it

since I need to add it only after CompanionLink component is added

[solemn hollow]

i see. Just have a system that queries for CompanionLink and exclude the component with the reference to your own gameobject. Once your system did the processing and added your own "Companion" component it will never run again. You could use IsystemStateComponents for that to make cleanup easier

[haughty rampart]

@robust scaffold how's the source generator coming along? i'm curious

[robust scaffold]

bit of an issue with my data source files. I'm now overhauling the entire system

(basically I accidentally overwrote the data files, and now I have no clue how to get it back. So ive gotta figure out what I did a loooong time ago and get my data files remade)

[haughty rampart]

Data files?

[haughty rampart]

And literally get the files back from your version control if you've accidentally overwritten them

[pliant pike]

damn, your reminding me I'm need to sort out some kind of source/version control I've had something similar happen

[robust scaffold]

Too many files and copyrighted by another company. They aren't my data source files. I'm just... borrowing them for this hobby project.

[haughty rampart]

And what kind of data source is it? Couldn't be code

[robust scaffold]

Just text files and large pngs

the PNGs are the main problem. Big 16k map textures

[haughty rampart]

How'd you break that with a source generator though?

[robust scaffold]

I have the original file, I did some sort of remapping in order to change the coordinate system.

[robust scaffold]

Nah, this was another thing I did. That I fucked up badly

[haughty rampart]

Ahhh. Ok

[robust scaffold]

Yea, this was my own stupidity without backing it up.

[haughty rampart]

Version control system......

[robust scaffold]

Yea... next time eh?

[haughty rampart]

XD

[rotund token]

Just bind ctrl+s to commit

Side note - riders local history is great

[muted field]

I'm getting this error but don't know how to fix it:

_result is not declared [ReadOnly] in a IJobParallelFor job. The container does not support parallel writing. Please use a more suitable container type.

I have it marked as write only because thats intentional:

        private NativeList<int2> _result;```

in the execute function i have:

if (result[0])
{
    var writer = _result.AsParallelWriter();
    writer.AddNoResize(new int2(index, i));
}

i was told this is how you are meant to write to a native list in parallel jobs yet it wont work

[molten flame]

Not really DOTS related but for some reason my active scenes aren't saved in the editor when I save and close. It always gives me a default scene when I load my project back up.
This isn't normal right?

It's not a big deal, my scenes are still saved, they just aren't active in the hierarchy and I have to reload them. Just seems wrong..

[muted field]

you can declare which one is default but yeh its not dots related at all

[molten flame]

How do I do this for the editor? I have done this for my play mode scene.

[muted field]

#💻┃unity-talk ask in the right place

[rotund token]

just replied on forums but you can't schedule jobs from jobs

[muted field]

yeh i was just about to reply. wish the error would say that because the error is not helpful lol

[pliant pike]

isn't it also that all imported nativearrays will come up with that error 🤔

[muted field]

define imported?

you mean created within a job ?

[pliant pike]

yeah if its local to the onupdate

[rotund token]

the error is useful actual

_result.AsParallelWriter();

[muted field]

whats wrong with that line?

i read on the forum that you had to use that to write to a list in parallel jobs

[pliant pike]

a single array into a parrallel job is liable to have problems

[rotund token]

this is true

[muted field]

thats why i was confused i had to declare it read only

because i wanted to write to it

[rotund token]

note that you can't increase capacity in parallel

you have to pre-allocate the capacity

[muted field]

is there no resize feature ?

for the writer*

[rotund token]

not in parallel

[muted field]

i guess that makes sense

[rotund token]

resizing requires reallocating to a new pointer

[muted field]

yeh i can imagine a race condition in that situation

[rotund token]

there are containers that do support resizing

such as NativeStream

[muted field]

ill have to keep that first job on the main thread and only do vertex overlap test on a job then

[rotund token]

as each thread can get its own stream

[pliant pike]

I see so that's what nativestream is for

I havent found a need for that yet I guess

[muted field]

how did you figure all this out - its tricky enough as it is with the current documentation details

[pliant pike]

personally I started this when there was literally almost no docs

I don't even know how I got anywhere with it

[muted field]

that would've frustrated me no end lol

[rotund token]

pretty much haven't ready any documentation

[pliant pike]

mostly I learned from the few tutorials that existed and help from the forums and this discord

[rotund token]

i have read probably the entire source code though

[muted field]

i would never have thought native stream existed to look for such a thing

[pliant pike]

yeah that's a bit beyond me

[muted field]

they should hire you to write the docs you'd probably be good at explaining it all 😄

[pliant pike]

I do sometimes look through the scripting api as well

[rotund token]

hell

no

they could not pay me enough to write documentation

[pliant pike]

some of the stuff doesn't even have vague descriptions of what they do though

or even if they do, theres no how to use them so its a lot of trial and error

[muted field]

i would never had known about parallel writer if it wasnt for searching the forums for a particular error that some one else had already had

[pliant pike]

yeah there's probably(definitely) a lot I still don't know about dots

but there's no need to learn everything at once

[muted field]

but the more i think about it, my approach wouldn't work due to race conditions anyway

[pliant pike]

if you keep to the safety systems then there shouldn't be much of a problem

[muted field]

how come a nativestream can be resized. but a native list cannot even though lists are known to be resizeable.

in parallel context anyway

[pliant pike]

because each thread gets their own unique nativestream

[muted field]

oh so you get a collection of streams at the end

rather than all bundled into one collection like a list would be

[pliant pike]

yeah I'm not sure how it works to be honest, is it all collected into one single array at the end perhaps

[rotund token]

nope

you can read it back in parallel

or you can merge it into a single stream/list if you want

ideally you don't need to do that though

[pliant pike]

ok that makes sense

[rotund token]

btw NativeQueue.ParallelWriter can also resize in parallel

[muted field]

oh so they support queue but not list 🤔

that seems even more strange to me

[viral sonnet]

NativeQueue is really the only container I haven't used much. From my limited test it wasn't that great but now I'm reading the source, it doesn't look too bad. Do you have more experience with it?

[rotund token]

it's /fine/

much faster than default nativelist for example

honestly it's a great first container for most people

solves a lot of thread issues etc

while performing well

i don't use it much anymore but it still has its uses for convenience

[viral sonnet]

I can't remember what the issue was. Seems there's no parallel reader

[rotund token]

there's not

that said a parallel queue reader wouldn't make any sense?

[viral sonnet]

Guess that was it, from source reading seems limited

[rotund token]

it's a queue it's meant to be ordered

[viral sonnet]

yeah, good point 🙂

[muted field]

some weird inconsistencies with their collections the hash set Count is a method yet in list its a property

[rotund token]

so that hybrid thing we hate enzi

i got my workflow working amazing

[viral sonnet]

tell me more 🙂

[rotund token]

i have skin meshed renders in the world but dont show up in hierarchy

when you select them in scene view they select the enitty

(i.e. i added skinned mesh renderers into the entity picking stuff)

for normal workflow you can legit not tell they are gameobjects

it just feels pure

the whole setup is automatic

simply add a

hybrid component authoring script to an entity

[north bay]

They changed it to be a method some time ago to clearly indicate that it has to compute the count

[rotund token]

with the gameobject/any components

[muted field]

they cant keep track when adding ?

[rotund token]

and then my setup automatically adds component references i need

[viral sonnet]

sounds great. anything to share? your comp certainly has the same logic as mine

[rotund token]

what i mean

yeah i'll probably share this whole thing soon~

it's very simple actual

should have turned off gizmos before i recorded this 😄

[viral sonnet]

This is what I'm doing for hybrid

[rotund token]

oh i dont have anything like this

[viral sonnet]

I've written this some time ago. Could be simplified I guess. I was never too sure how to handle subscenes/normal scenes and conversion vs. no conversion

[rotund token]

i instantiate at runtime

[viral sonnet]

same here.

this is just the comp that does the setup + drawing so I can see a mesh

[rotund token]

            {
                var entity = this.GetPrimaryEntity(hybrid);
                this.DstEntityManager.AddComponentData(entity, new HybridComponent { Prefab = hybrid.Prefab });

                switch (hybrid.CopyMode)
                {
                    case HybridComponentAuthoring.CopyModes.None:
                        break;
                    case HybridComponentAuthoring.CopyModes.CopyTransformToGameObject:
                        this.DstEntityManager.AddComponent<CopyTransformToGameObject>(entity);
                        break;
                    case HybridComponentAuthoring.CopyModes.CopyTransformFromGameObject:
                        this.DstEntityManager.AddComponent<CopyTransformFromGameObject>(entity);
                        break;
                    default:
                        throw new ArgumentOutOfRangeException();
                }
            });```

my entire authoring

all my authoring scripts exist in editor asmdef

to strictly disallow runtime authoring

[viral sonnet]

does this run in Convert?

[rotund token]

you mean convert to entity?

no that is runtime authoring

strictly disallowed 😅

[viral sonnet]

the code you posted. not a system, right?

[rotund token]

oh

sorry

    {
        protected override void OnUpdate()
        {
            this.Entities.ForEach((HybridComponentAuthoring hybrid) =>
            {
                var entity = this.GetPrimaryEntity(hybrid);
                this.DstEntityManager.AddComponentData(entity, new HybridComponent { Prefab = hybrid.Prefab });

                switch (hybrid.CopyMode)
                {
                    case HybridComponentAuthoring.CopyModes.None:
                        break;
                    case HybridComponentAuthoring.CopyModes.CopyTransformToGameObject:
                        this.DstEntityManager.AddComponent<CopyTransformToGameObject>(entity);
                        break;
                    case HybridComponentAuthoring.CopyModes.CopyTransformFromGameObject:
                        this.DstEntityManager.AddComponent<CopyTransformFromGameObject>(entity);
                        break;
                    default:
                        throw new ArgumentOutOfRangeException();
                }
            });
        }
    }```

it's just a GameObjectConversionSystem

originally i had a bunch of other logic

which is why it's a system

but i should probably move it to a IConvertGameObjectToEntity now that it's not required

[viral sonnet]

ah, I see. I remember having some trouble with getting the correct entity index. either there was some bug or limitation but I had a weird way of working around it

[rotund token]

but yeah i dont create a separate presentation entity

the components are attached directly to the actual entity

but this is only done on the client