archived-dots

[misty wedge]

@nova isle lets make a thread

[rotund token]

A

{
    public int Value;
}

public class TestBaker : Baker<TestAuthoring>
{
     public override void Bake(TestAuthoring authoring)
     {
     }
}```
B
```public class TestAuthoring : MonoBehaviour 
{
    public int Value;

    public class TestBaker : Baker<TestAuthoring>
    {
         public override void Bake(TestAuthoring authoring)
         {
         
         }
    }
}```
Opinions?!

Manual + DOTS_Guide use A
Samples use B

[rustic rain]

A

Cleaner solution

Considering bakers might be related to components from other assembly

Like Transform

[viral sonnet]

i use B. no idea what sticks yet

[rustic rain]

There's practically no difference though, no?
Just code styling

[viral sonnet]

not really, the bakers won't show up in intellisense they do 😄

[rotund token]

my bakers are in editor only authoring assemblies which nothing reference

keeps the old IS a bit smarter

but yeah was just curious which people prefer and if there was a clear majority

personally i've been doing A but not sure if this was just swayed by original documentation

there's something I like about separating the data and the logic

no one's doing this

{
    public int Value;

    public struct TestSystem : ISystem
    {
        // ...
    }
}```

[frosty siren]

I want to adapt my system to be compatible with SubScenes, but there's pure info about how SubScenes work with SCD. I want to bake some data to SCD with assets reference (Material, SOs). But i doubt about how subscene can serialize such data. Any advice? 🙂

[rotund token]

they can serialize it fine

alternatively to SCD they also work with class based IComponentData fine

[coarse turtle]

yea - I have a managed object in my subscene since it originally held some materials

[frosty siren]

Ok, thank you for clarification @rotund token @coarse turtle. Btw want to use SCD because it can be already an ID, i use to batch render process, no matter what it holds

[rotund token]

speaking of, is the correct way to add class IComponentData through bakers using AddComponentObject?

surprised there's no managed AddComponent overload

[whole gyro]

That's what I've been doing, but I would also like to know if this is correct.

[coarse turtle]

ah same, ive just been using AddComponentObject also

[whole gyro]

Has anyone had success with removing the transform components from an entity in a Baker? Is this possible?

I tried call AddTransformUsageFlags(TransformUsageFlags.ManualOverride) in a baker and it seems to cause all sorts of problems in the editor.

[rotund token]

com.unity.cinemachine.dots 👀
my cinemachine wrapper might not be needed soon

kind of a shame, it actually worked really well

[misty wedge]

maybe they stole your wrapper while you weren't looking

[viral sonnet]

someones enjoying the cat discord 😄

[viral sonnet]

i'll do the same. the cool thing is that the authoring dll can only exist in editor

[misty wedge]

It's kind of weird that they removed a specific job type for requiring the entity index

Since now you just kind of gotta do it yourself, it's not like every job computed it before, only the WithIndex ones

Especially since they could have then built the counting mechanism into ChunkEntityEnumerator

[rotund token]

You are aware there is a method to calculate it right?

[misty wedge]

Yes, of course. I just find it weird that they removed the automated version

[viral sonnet]

basically overhead 🙂

[misty wedge]

The automated version had overhead over computing it manually by scheduling the presum on the entityquery?

[viral sonnet]

wait, which job are you talking about? 🙂

[misty wedge]

IJobEntityBatchWithIndex

[viral sonnet]

oh, hm. yeah, i don't see the reason either

[viral sonnet]

maybe that's exactly what they wanted to prevent?

[misty wedge]

Why?

It just makes it less error prone and less code

[viral sonnet]

because you would need 2 jobs then. one that uses the enumerator with enabled/disabled and one that works on normal entities

[misty wedge]

Huh? They could just add an IJobChunkWithIndex that supplies the index and works with enabled/disabled components

[viral sonnet]

true, maybe they'll provide the job

or they want to reduce the job clutter of too many options. hard to tell what their thinking process is

[rotund token]

well they'd probably need
IJobChunkWithIndexWithFilter
IJobChunkWithIndexWithoutFilter

because the indices generation are different

and at this point it's just becoming a huge mess

i think it's rare enough used that simply calling a single line method to generate this is not a big deal

[misty wedge]

Wait what? I thought the method on the EntityQuery is the same regardless of whether there are enable-able components or not:
EntityQuery.CalculateBaseEntityIndexArrayAsync

I'm guessing it skips to the first enabled index

[rotund token]

CalculateBaseEntityIndexArray vs CalculateFilteredChunkIndexArray

[misty wedge]

Oh wow I didn't even see that

[rotund token]

it's probably for like shared filters etc

but yeah you need to factor this in (i havent tested enough yet)

[misty wedge]

Ah, so CalculateBaseEntityIndexArrayAsync is the correct method if you have enable-able components

Ok then, that makes more sense on why they kicked the job

[viral sonnet]

yeah, right. some chunks could be totally skipped with enable comps

[viral sonnet]

wait, isn't it the filtered one?

[proud jackal]

To check how it works try the [EntityInQueryIndex] on an IJobEntity and see the code it generates for both Schedule and ScheduleParallel :3

[proud jackal]

Wow that doc is great! Explains exactly what it does! o.o (didn't know we added it, thx for telling!)

[misty wedge]

Haha

[viral sonnet]

but you mentioned enable comps and the doc doesn't

[misty wedge]

That is true, I actually have no idea. I just started converting my job code 🤷

[proud jackal]

Well, guess then my answer is still valid Check IJobEntity :3

[misty wedge]

I'm guessing it's correct though, since the header for that section is Update IJobChunk to handle enableable components

[viral sonnet]

when would we use CalculateFilteredChunkIndexArray then? 🙂

[misty wedge]

^

Maybe shared filters? I have no clue

[viral sonnet]

hm, that's a bit confusing tbh. i'd expect filters/filtered in the context of enable comps

lets look at code! 😄

ignore me, i didn't read right, chunk index != entity index

[proud jackal]

Well, there are only two filters, Change and Shared 😛
Enabled comps is more masking 😆 (tho I have no clue what that specfic function does, I'll check nvm, seems you guys got it!)

[misty wedge]

Actually I think CalculateBaseEntityIndexArrayAsync does it all, since it knows about the filter (it is ran against an EntityQuery after all)

I'm assuming it skips until the first enabled component index, since otherwise that section of the docs doesn't make a lot of sense

[rotund token]

/// unfiltered list of chunks matched by this query. Most commonly, this is the chunkIndex parameter available
/// within <see cref="IJobChunk.Execute"/>.```

CalculateBaseEntityIndexArrayAsync specifically doesn't filter

[misty wedge]

I thought it just meant you are supposed to pass unfilteredChunkIndex

To read the starting entity from the NativeArray

[misty wedge]

At least that's what this comment reads like to me

It even mentions Execute

[rotund token]

        /// unfiltered list of chunks matched by this query. Most commonly, this is the chunkIndex parameter available
        /// within <see cref="IJobChunk.Execute"/>. For queries with no chunk filtering and no enableable components,
        /// array[N] will equal N.```
CalculateFilteredChunkIndexArrayAsync

it mentions enabled components

[misty wedge]

Yes, the unfiltered chunks, but not the unfiltered entities in the chunk

I think we are just speculating though, so someone will have to look at the actual implementation 🤷

[rotund token]

yeah but if there is enableable it implies array[N] != N

so confusing

[rotund token]

i am

[misty wedge]

If it doesn't that would mean the docs are wrong. I don't see any other function to calculate the entity index on entityquery though...

[viral sonnet]

i realized that i have IComps and authoring sometimes in the same file. that's not a good idea going forward

[rotund token]

yeah ok they are definitely a little different

[misty wedge]

So it gives you the correct index even if you have enable-able components?

meaning it could e.g. skip half a chunk

[rotund token]

that depends how the scheduling part of the job works

if it passes in chunks with no enabled elements or not

[misty wedge]

That makes sense, and if the enabled entities were 14-18, it would return 0, 14, 20

[rotund token]

wait no

i think the base version handles this

[misty wedge]

Why doesn't it handle filters?

[rotund token]

just if you use shared filters you need to use filtered version

from my 2min look at like 400 lines of unsafe code

[rotund token]

what do you mean?

[misty wedge]

Like it returns incorrect values if you set a sharedcomponent filter on the entity query?

I feel like it should completely ignore the filters, since you access the native array with unfilteredChunkIndex, implying it skips the filters for the creation of this index (i.e. just the chunk index)

[rotund token]

this is the different

(i deleted a bunch of identical variable setup at top)

[misty wedge]

Are you sure those two methods are intended for the same thing? The names aren't even the same

I feel like it would have to be named CalculateBaseEntityIndexArrayWithFilterAsync

(which again doesn't make any sense to me, since the filter is set on the entity query, so this function would know about it, meaning the same function could handle both cases...)

[rotund token]

just because it knows about the filter

doesn't mean it should use it

[misty wedge]

Yes, like I said earlier, it should ignore it, always

[rotund token]

there's a reason there is IsEmpty vs IsEmptyIgnoreFilter

[rotund token]

it should definitely not ignore it always - it's extremely useful to get the full chunk list sometimes
but yeah i don't know how this is handled in the job we haven't tested

[misty wedge]

For the use case of passing the returned array to a job for reading the first entity index for that chunk, it has to

Since the accessing chunk index is unfiltered

[rotund token]

i think testing it to see how it works is probably more prudent than just discussing it ^_^'

[misty wedge]

Definitely

[balmy thistle]

Many disputes can be resolved with a little empiricism

[misty wedge]

but what about the juicy speculation

[balmy thistle]

it's the hot takes that keep them coming back

[misty wedge]

If I'm switching over to ISystem, I need to mark all methods (and structs?) that I want to burst inside the ISystem with [BurstCompile], correct?

[rotund token]

make the system as [BurstCompile]

then any of the create/destroy/update that can be compiled as well

[misty wedge]

And if I call a function on a struct inside one of those methods, does the function / struct also need to marked with [BurstCompile]?

[viral sonnet]

pretty sure that doesn't need the tag

when you're calling a method from a burst compiled method it will always be burst compiled

[rotund token]

these methods do basically what i expect

2 components
Filter : ISharedComponentData
Enabled : IEnableableComponent
6 entiies
filter = 0, disabled
filter = 0, enabled
filter = 1, disabled
filter = 1, enabled
filter = 2, disabled
filter = 2, enabled

query.SetSharedComponentFilter(new Filter { Value = 1 });

above results

[misty wedge]

Yep, that makes sense

[rotund token]

let me throw it into a job to just confirm that part

[misty wedge]

So the docs are correct

[rotund token]

(i dont really understand what you mean by that since the docs don't mention filtering or anything)

[misty wedge]

The part about how you get the "old" firstEntityIndex

(since IJobEntityBatchWithIndex was removed)

[rotund token]

the old version ideally you should do the filter calc within the job

instead of on the query

(otherwise you got sync points)

[misty wedge]

That means they could add an IJobChunkWithFilter

[rotund token]

but yeah ok this works as expected either way

only index 1 is passed in

so the answer is, either works 🤣

let me just confirm with multiple chunks filtered

[misty wedge]

Is there any point in caching singletons, assuming they are never destroyed?

[rotund token]

i don't know why but CalculateFilteredChunkIndexArray looks faster to me

[misty wedge]

Do you keep your entity queries in the system and pass them to the struct that holds type handles, or store them in the struct?

[rotund token]

i have no type handles for this test

and just to confirm multi chunks of the same filter work

[misty wedge]

No I meant in general, as a code style 🙂

[rotund token]

oh funny you ask taht

historically stored them in the system

but i did just test this morning storing in the job

[misty wedge]

By job you mean the struct that holds all the type handles, and schedules the job?

[rotund token]

definitely reduces a bit of boilerplate

yeah

[misty wedge]

What if two "jobs" use the same entity query?

Just create it once in OnCreate and pass it to both?

I never cached my type handles, component lookups, etc since I didn't know you could do that, so now I'm cleaning a lot of stuff up

I saw that ISystem even logs a warning if you don't cache them

[viral sonnet]

what are you two talking right now? where to store ComponentTypeHandles?

[misty wedge]

No, where to store the entity query for the struct that schedules the job

(assuming you store the component type handles in this struct)

(since you could also just store those in the system if you wanted)

[rotund token]

{
    private EntityQuery query; // todo
    private ComponentTypeHandle<Translation> translationHandle;

    public void OnCreate(ref SystemState state)
    {
        translationHandle = state.GetComponentTypeHandle<Translation>();
    }

    public void OnUpdate(ref SystemState state)
    {
        this.translationHandle.Update(ref state);
        state.Dependency = new TestJob
            {
                TranslationHandle = this.translationHandle,
            }
            .Schedule(this.query, state.Dependency);
    }

    private struct TestJob : IJobChunk
    {
        public ComponentTypeHandle<Translation> TranslationHandle;
    }
}```

what i'd traditionally have done

{
    private EntityQuery query; // todo
    private TestJob job;

    public void OnCreate(ref SystemState state)
    {
        this.job = new TestJob
        {
            TranslationHandle = state.GetComponentTypeHandle<Translation>(),
        };
    }

    public void OnUpdate(ref SystemState state)
    {
        job.TranslationHandle.Update(ref state);
        state.Dependency = this.job.Schedule(this.query, state.Dependency);
    }

    private struct TestJob : IJobChunk
    {
        public ComponentTypeHandle<Translation> TranslationHandle;
    }
}```

what i tested this morning just from curiousity

[misty wedge]

Is it safe to reuse the job struct like that?

[rotund token]

why not

[misty wedge]

I don't know, you know all of the inner workings, not me 🤷

Just making sure 😅

[rotund token]

struct is copied onto the JobChunkWrapper

but yeah it breaks down if you have 2 jobs in same system using same handle

but it does save you a bunch of lines if you only have the 1 job

[misty wedge]

You can't give them the same handle?

[viral sonnet]

ok, you posted what i expected from the conversation 😄 yeah, i dunno. doesn't seem too bad. reducing the list of parameters in the system is pretty good

[rotund token]

you'd just need to update them both separate

[misty wedge]

Wouldn't it be faster to copy the updated handle to the other jobs? Or does that break everything

[viral sonnet]

still gonna stick to the traditional method. makes it obvious at least what's all used on a quick glance

[misty wedge]

Also, if you had a job with a lot of component type handles do you think (from a style perspective) wrapping the job in another struct with a single update method that updates all handles, componentlookups, etc, is a better pattern to keep the system cleaner?

[rotund token]

thats actually what i was testing this morning

because i had something similar setup that a single job had like X sub jobs

[viral sonnet]

thanks for testing @rotund token. i'm gonna need to read all you've posted later. i'm moving namespaces and creating files for authoring -.-

[rotund token]

and each sub job was setting itself up

[misty wedge]

I think you used something along the lines of what I mean in the PositionBuilder file you posted some time ago

[rotund token]

oh that part of my spactial builder

https://gitlab.com/tertle/com.bovinelabs.core/-/blob/0.9/BovineLabs.Core/Spatial/PositionBuilder.cs

this thing

[misty wedge]

Yep

[rotund token]

this is really useful for re-usable code

[misty wedge]

Since it basically wraps the job you are calling, and updates all handles

[rotund token]

i dont write like this for regular systems though, just if i want some chunk of logic usable in multiple places

[misty wedge]

Doesn't the OnUpdate method get kind of cluttered otherwise though?

e.g. if you have

job.Handle1.Update();
job.Handle2.Update();
job.Handle3.Update();
...

I guess you could just put it in a method, basically just where to put the code 🤷

[rotund token]

my on updates are pretty empty for the most part

i mean, it's a few Updates()

then a job schedule

and that's it 😄

        public void OnUpdate(ref SystemState state)
        {
            this.effectHandle.Update(ref state);
            this.statEffectHandle.Update(ref state);
            this.statValueHandle.Update(ref state);
            this.attributeTargetChunkCacheHandle.Update(ref state);
            this.statValueBufferHandle.Update(ref state);

            state.Dependency = new UpdateActiveChangesJob
                {
                    EffectActiveHandle = this.effectHandle,
                    StatEffectHandle = this.statEffectHandle,
                    StatValueHandle = this.statValueHandle,
                    AttributeTargetChunkCacheHandle = this.attributeTargetChunkCacheHandle,
                    StateValueBufferHandle = this.statValueBufferHandle,
                    SystemVersion = state.LastSystemVersion,
                }
                .ScheduleParallel(this.query, state.Dependency);
        }```

typical system

[misty wedge]

Yeah, but if you have 3+ jobs in that system, it's quite a bit more lines 😅

[rotund token]

i always group all my updates at the top

not that bad though

[misty wedge]

Just trying to find a style I like

I think I'll go with this though

[rotund token]

why am i doing an IsEmpty check on this system hmm

i hate when i just randomly grab some code

[misty wedge]

Also, since you are checking if the query is empty here, is that better than adding a RequireForUpdate?

Ah 😅

[rotund token]

and something puzzles me

[rotund token]

well IsEmpty is for including filtering

RequireForUpdate does not include filtering

this is why i'm confused

because i have no filtering on this system

but now i'm wondering if it got lost when i changed it to a EntityQueryBuilder

[misty wedge]

Also, is using [RequireMatchingQueriesForUpdate] fine if you aren't using any filters?

I thought I read somewhere it is worse performance wise

[rotund token]

it just gives old behaviour back

i actually have to go through and check all my systems at one point

and add those tags where required

[misty wedge]

I'm not sure if I should be using RequireForUpdate vs [RequireMatchingQueriesForUpdate] if I have no filters

[rotund token]

it just iterates all queries and does a IsEmptyIgnoreFilter, if any return true, stops then and runs system

but a lot of them time you don't need to check all queries for a system

[rotund token]

they do different things if you have more than 1 query

[misty wedge]

[RequireMatchingQueriesForUpdate] is more like RequireAnyForUpdate

So maybe RequireForUpdate for "minimum" requirements, and [RequireMatchingQueriesForUpdate] or RequireAnyForUpdate for specific parts of the system? 🤷

[wraith hinge]

fwiw, i am frequently also confused when working with filters on queries. what i don't know is if this is just a fundamentally always confusing subject, or if there's some other design we could have gone with that would have the same performance but been less confusing

[viral sonnet]

i think it is confusing with how many nuances there are. nothing that a good doc and a small cheatsheet with a bunch of use cases couldn't explain though.

top tier would probably be a static analzyer - if that's possible

to eleborate, something that warns us, hey you have 2 queries but you're using alwaysUpdate. no idea how viable that really is.

[proud jackal]

it is something our team has thought about before, however for self-evident reasons we chose to wait 1. Can't analyze an API that wasn't stabilized yet (that's about to change) 2. Performance is likely gonna be questionable, as a result it probably has to be a once every now again check you can use as a diagnostic tool when stuff isn't working rather than the normal usecase of an analyzer which is to keep things working. This being said, our SourceGen will throw error at you if you do something you were not supposed to do, as any good compiler should :3 (we've also thought about adding so it warns about lack of BurstCompile on ISystem and IJobEntity, tho that's actually not especially performant inside an ISystem, since we don't know which syntax is managed without doing type analysis on everything, it's a little easier with IJobEntity as that only really has one way to get managed data into the execute in the first place, which is pretty easy to analyze for) Anywho none of this will happen in any near future

[whole gyro]

@errant hawk Hey, did you ever figure out how to prevent this from happening? Same thing is happening to me

Meant to reply to this message

[errant hawk]

nope, still happens. It's a bug and other users have reported on it afaik

[whole gyro]

Got it, thanks. Looks like I can work around it by closing the subscene. Seems like it also works when I disable live baking (using the "internal" trick to see that checkbox)

Is it normal that when I click on an entity in the scene view, it opens up the inspector with the authoring data instead of the entity data? There is also no circle at the top right of the inspector to switch the view mode

[rotund token]

there should be a circle o_O

[gentle gyro]

Did you ever get ECS working on the Quest 2?

What is the 'build asset?'

[rotund token]

right there

[whole gyro]

Yeah, it just doesn't show for entities that I click on in the scene view. It shows that circle when I click in the entity hierarchy though.

[rotund token]

odd

[gentle gyro]

When I build ECS 1.0 on the Quest 2, I get this error log. The build works fine on Rift S and the editor...
Anyone know what might be causing this?

[brave field]

Will PhysicsCategoryTags be forced to be authoring only in future dots physics release? Currently I still use it at runtime

[stiff mesa]

Hey everyone! Can anyone share their experience combining monobehaviors and entities? I've made a system that loads SubScenes and it looses all non-ECS data. Can I somehow combine SubScenes with MonoBehaviors? Or can I use regular scenes with ECS? And what would you prefer?

[rotund token]

only entities can exist in subscenes (there are a few built in companion objects setup though)

[stiff mesa]

So, only scene loading + converting objects to ECS is supported? Can I benefit from SubScene features like blob storages if I use simple additive scenes?

[brave field]

For 1.0 Baker, wat's the conversionSystem.GetEntities() replacement?

[gentle gyro]

Could this il2cpp bug explain why my ECS Quest build works in mono but crashes with il2cpp?

[balmy thistle]

Almost certainly

[gentle gyro]

🤔 , it only starts crashing when the renderer kicks in apparently...

[balmy thistle]

We'll update the known issues in this post: https://forum.unity.com/threads/experimental-entities-1-0-is-available.1341065/ when they become available

[balmy thistle]

One of the issues is a discrepancy in calculation of the size of an object between il2cpp and other sources of object code

A crash isn't the only possible outcome, and it's dependent on the links of memory references that are followed at runtime

(if it's indeed the same problem).

[gentle gyro]

@balmy thistle, have you had any luck finding Anyone who has successfully built ECS, either 0.51 or 1.0, to the Quest 2?

[balmy thistle]

No, thanks for reminding me. If it is related to the il2cpp issue cited earlier, it would be dependent on the editor version specifically.

[gentle gyro]

Ok, I tried 2021.3.11, 2021.3.7, and 2022.2.0...

Ok, Thanks!

[viral sonnet]

if you have a blobassetreference with other data in an IComp, 100%

if your build works in mono, wait for next unity 2022.2 release

[gentle gyro]

I basically have a hello world setup. One GameObject, one subscene with one "GameObject and nothing else. No systems or anything...

[gentle gyro]

Any ballpark on when 2022.2 might release?

[rotund token]

betas tend to release every 1-3 weeks

[viral sonnet]

might be an unrelated quest problem. hello world setups are working in il2cpp (at least on windows)

[gentle gyro]

Excellent, Thanks!

[viral sonnet]

fingers crossed we get one on Wednesday.

[gentle gyro]

The same project works fine with il2cpp on Rift S and Editor. It's only a problem on the Quest 2 with il2cpp. It even works on the Quest 2 with mono...

For what it's worth: I just realized the crash happens as soon as my entity authored by my subscene is spawned...

[viral sonnet]

SubScenes never stored their Unity assets, such as Textures, in uncompressed formats. They use AssetBundles with compression turned on, and going forward they will use the new ContentArchive, which is also compressed. Huh, ever heard of ContentArchive before?

[gentle gyro]

Nope, VERY new to DOTS...

[brave field]

1.0 Baker API replacement questions

[late mural]

wondering with the new tranform system, will there be any easy way of creating rectangular prisms (like how before uniform scale existed we could just make one axis have a larger scale than another on a cube), or will we have to create it externally in something like blender?

[rotund token]

actual game with netcode has been a bit rough to upgrade compared to my libraries

[devout prairie]

Trying to do a hacky fix for something but this doesn't seem to be working:

Hard to tell from the error i'm getting from the ecb further down the chain..

In theory is the above possible?

actually no it makes sense right, as i'm not creating a native memory allocation for the SpawnPointBufferElement right, it's just a local struct

actually no, that's what the native reference should be doing isn't it

hmm

[rotund token]

well... what's not working?

[devout prairie]

i'm not sure it's throwing a null entity in the ecb after the job that takes this native ref..

just grabbing some lunch so i'll investigate a bit more, just checking if there was anything obviously wrong with this

such as 'can i create a native ref of a buffer element'.. but it's just a struct anyway right

[rotund token]

i have no idea what you're doing because the above code does not have any reference to ecb or an entity

[misty wedge]

That sounds like the issue isn't the NativeReference but something you're doing with the ECB, no?

[devout prairie]

yeah i think possibly yep.. will double check as soon as i'm back.. just checking there wasn't anything obviously wrong with that

my other version of this same code runs perfectly but it's grabbing an actual buffer element, so i thought maybe there was some issue with how i'd written that native ref, just started using those recently

[misty wedge]

So any consensus yet on whether to put the baker in the authoring component or just in the same file? I know it doesn't really matter but I want to try and keep everything consistent...

[rustic rain]

just do whatever fits you styling I guess 🤔

I doubt bakers will be stripped of build

even though conversion will

and it's a good thing

[misty wedge]

Yeah it's more about code style, I just wanted to know what everyone else was doing

[rustic rain]

I think

keeping conversion stuff in build will be a good thing for modding

[misty wedge]

Why? It doesn't run at runtime anyways

[rustic rain]

because how else will modders figure out how conversion works

besides

they will be able to use it too

in their builds

[misty wedge]

I have my modding set up very differently so I don't think this is an issue for me

[rustic rain]

I want to allow modding for literally everything

[misty wedge]

I explicitly don't 😅

[rustic rain]

depends on a game, right

[misty wedge]

Also a security issue

[rustic rain]

don't care about it, singleplayer game 😅

[misty wedge]

I meant more in the context of downloading a mod that installs a bitcoin miner on your PC

[rustic rain]

I'v worked with RimWorld game for a while and it has literally 0 security.
And after about thousands of mods installed

not a single virus

[misty wedge]

Not a multiplyer security context

[rustic rain]

allthough there was a story about guy who crippled game (not PC) if you were using some 18+ mod 🤣

[misty wedge]

Anecdotes don't really matter, it's still a potential security flaw

[rustic rain]

I get it, I just trust that people will use trusted mod shop

[misty wedge]

That doesn't really matter if someone gets a virus and sues you because they used your official mod shop

[rustic rain]

or at least some antivirus software, kek

[misty wedge]

It's an issue about covering your legal bases as well

If people really want to make a crazy mod for your game they can just use something like BepInEx after all, and then everything goes 🤷

[rustic rain]

nah, modding is way bigger than that

implementing new mechanics, game modes or even multiplayer kek

[misty wedge]

You can do all of that with BepInEx

[rustic rain]

some dudes even implemented multithreading into RW

it sucks, but it works !

[misty wedge]

Is there any documentation on modifying the drawn inspector for components?

[viral sonnet]

no doc sadly. i think there should be a supported way, right now all we can do is lend some implementation. previously it was Inspector<T> now its PropertyInspector<T>

i still haven't updated my inspectors to 1.0 😄

[misty wedge]

Where are these types located?

[viral sonnet]

Unity.Platforms.UI

[misty wedge]

Aline is a real performance hog if you are debugging a lot of separate entities...

[viral sonnet]

ok, that was weird. yesterday i moved all my authoring comps to a new assembly def. checked some files if they are working during the process. they did. today i finished the process and my prefabs all have missing authoring scripts. for no apparent reason. now i ticked allow unsafe code in the def and it works now. yet i have no unsafe code. nor was there any compiler error. i can untick unsafe code again and it somehow fixed itself.

eh, now some other authoring scripts are not working. must be some cache problem -.-

reimported the file and works again. hm, maybe this is a 2022.2 bug. i usually don't move so much files around. 🙂

[misty wedge]

I just remade all of my prefabs, since I was to lazy to attempt to port the old values

[viral sonnet]

That would have bummed me out 😄 I already had all authoring data in MBs so it was just a matter of writing the bakers and now moving everything to seperate namespaces and asm defs. But I realized that the effort was not that great because if I mark the asmdef editor only, I can't use the authoring MBs. (errors, can't put editor scripts on a GO) - the plan was to not even have the authoring dll in a build. that didn't worked out.

[sage kestrel]

would anyone know how to convert a NativeList<float3> to a Vector3[]?

[oblique cosmos]

Two quick questions: I'm currently trying to learn DOTS (and ECS in particular), so I'm going through some basic things. Currently: Moving an object to another one. I finally got the actual ECS part of it working, but I'm running into some strange issues. Namely, the chasing object arrives at an offset what looks an awful lot like 1 unit, so I'm wondering if there is some kind of offset involved in gameobject -> entity? Furthermore, is there any difference between mathf.normalize and math.normalize? Because the latter appears to produce nonsensical results for me?!

[viral sonnet]

no offset between GO and entity. @rustic rain has a method to convert without a memcpy if i remember correctly. haven't seen a difference of normalize. what are you normalizing, float3?

[rustic rain]

🤔

[oblique cosmos]

Hm. Odd. It's just a cube and a sphere and, just tested it, the sphere (the chaser) ends up exactly 1 unit off course and I can't seem to explain it. And yes, I'm normalizing float3, simple direction stuff. I can't even explain the behavior I'm seeing with it, it feels almost like the function can't produce negative values or something, it kinda sorta works on positive axes and then just stops chasing entirely on negative values. Or... something.

[rustic rain]

you sure you pinged the right person?

Cause I have no idea what you're talking about 😅

[viral sonnet]

oh, sorry then 😄 i thought you posted a method to convert NativeArray<T> to T[]

[rustic rain]

actually vice versa, and it's originally was tertle's

[viral sonnet]

ok 🙂

[wraith hinge]

i would like to point out that that method has unsafe in the name for a reason, and if stuff moves around under you, you'll crash the editor

[rustic rain]

It uses GC pin

and you have to unpin it manually

[oblique cosmos]

Any reason why this would produce nonsense? (More specifically, the normalizing appears to not produce negative values, because it works when the object is in front (as in in a positive direction) but ceases to work in the other direction. Granted, I've not programmed nor done math in a while, but... that's not how that should work, is it? Or am I just going crazy?)

[viral sonnet]

i see no errors here

normalize does Returns a normalized version of the float3 vector x by scaling it by 1 / length(x). so negative values are fine

[oblique cosmos]

Yea... me neither. Problem is, the rest of the setup is as simple as it gets, you can infer the entire codebase from that screenshot, it's the simplest components and, well, the system setting up the job. In Unity, there's a subscene with a box (the target) and two spheres (the chasers) and that's it :|

It works without the normalizing (but, you know, then the speed thing is kinda pointless), but then I somehow get this odd 1 unit offset. I'm so damn puzzled...

[viral sonnet]

i was about to ask, what if you don't normalize. yeah the speed is off but other than that it works? hm. are you using transform system v2? as i'm using physics i can't say much about it. such a simple usecase should work though. same goes for transformAspect

[oblique cosmos]

Uhhh, v2 would be the new 1.0 thing, yes? I don't even know, I just started learning and then 1.0 dropped so I'm all sorts of confused lol. But yea, I'm using whatever is standard in the experimental 1.0.

That's the result without normalizing

[viral sonnet]

can you try math.normalizesafe?

[oblique cosmos]

Which, ironically, is 1 unit off, so some kind of normalized vector thingy appears to sneak in there or... something oO

Uh, sure, can try

[viral sonnet]

does the job always run?

[oblique cosmos]

No clue, jobs and I are still on a need to know basis, but I see no reason why it wouldn't

Noramlizesafe produces that...

Now I'm even more confused.

[viral sonnet]

you are getting closer! 😄

[oblique cosmos]

Well, one of the spheres is, anyway :D

[sage kestrel]

PathFindJob pathFindJob = new PathFindJob
        {
            gridSize = gridSize,
            startPosition = startPosition,
            endPosition = endPosition,
            resultPath = path
        };

        JobHandle jobHandle = pathFindJob.Schedule();
        jobHandle.Complete();

The UNKNOWN_OBJECT_TYPE PathFindJob.resultPath has not been assigned or constructed. All containers must be valid when scheduling a job.
so i keep getting this error even tho i assigned a value to the resultPath,any idea why?

[viral sonnet]

@oblique cosmos can you post the system where the job is scheduled?

[coarse turtle]

how is path constructed/declared?

[oblique cosmos]

@viral sonnet

[sage kestrel]

        NativeList<float3> path = new NativeList<float3>();

[oblique cosmos]

That's basically it, ignoring the empty OnCreate and Destroy

[coarse turtle]

ah you need to pass in arguments (length and Allocator). The default constructor would create an invalid native list

[sage kestrel]

it is a list that is used in a job,so i basically pass in TempJob?

[coarse turtle]

yes

[sage kestrel]

ok it works now

thx

[misty wedge]

I kind of dislike the name CleanupComponent... It feels weird to store data there

Maybe that's just me

[viral sonnet]

@oblique cosmos yeah, sorry that all looks fine to me. the only thing that could be off now are the values itself. open the dots hierarchy and look if you find anything odd. math and job is all correct so it's really the only thing that could cause issues

and maybe give it a try without using TransformAspect

i'm not too sure about this. from TransformAspect.Position: LocalPosition = HasParent() ? math.transform(WorldToParentMatrix, value) : value;

doesn't seem like your entity has a parent though

and give setting LocalPosition a try

[oblique cosmos]

Hmmm. So, firstly, I have no idea how this Entities Hierachy is supposed to work now. I seem to recall that looking different in the previous versions I started with (like a few days ago). Now it just seems like a copy of the normal inspector (just with all the other entities in there) and it displays nothing useful, it doesn't even update stuff?!

Pretty sure I already tried LocalPosition, actually, just as a reality check but nothing changed. I'll try again though, to be sure. Throwing out the Aspect thing might also be worth a try, yea.

[misty wedge]

I need to mark the ISystem with [BurstCompile] if I want to burst any of the 3 callback methods right?

[viral sonnet]

there's a circle at the top right. make sure it's fully orange to see the actual entities

[viral sonnet]

yes

[misty wedge]

Is there any documentation on ISystem anywhere? I couldn't find any...

[oblique cosmos]

Ohhhh, so that's what that was about 🤔 I'll look into it, thanks!

Okay, so LocalTransform and not using the Aspect produces identical outcomes, as expected. Guess I'll check out those values, see what the hell is going on there.

Looked around in the values, it all looks good, as far as I can tell. IDs match up, positions are as they are displayed, all checks out. Just the math doesn't. Thing is, when I move the target around in a certain way, the chaser hit perfectly. It's got something to do with certain axes, I'm not sure. It's weird. I don't see how this could be happening, other than something being up with math.normalize, but I figure that would've been spotted already, given how common that function is. Puzzling.

That's what's happening.

Wow. Okay then, I figured out what's been going on. Got nothing to do with DOTS, got everything to do with weird background file management stuff. It would appear that, somehow, a previous test (which I deleted entirely) somehow got restored and that had another speed system in it. Why exactly nothing complained about the exact same file and struct names is not entirely clear to me, but after sanitizing the entire project it appears to actually work. ... Now I just feel even dumber :)

Either way, at least I got a "looks good" out of it, so at least I know I'm not being entirely off track here. Thanks for the attempt to help, @viral sonnet

[viral sonnet]

np, glad you figured it out

[oblique cosmos]

Wish I also knew how absolutely nothing complained about that ghost file existing, yet it still actually working somehow. Mystery in itself, that one 🤔

[wraith hinge]

Is there any documentation on

[viral sonnet]

i was testing sentris problem myself in a new project because i found it odd but i got hold up by setup problems 😄 setup with entities.graphics could be better handled. i mean, you install it and then. nothing. confusing for any newcomer because you now need to know to install either urp or hdrp. and then you install urp and again, nothing. i remember some kind of wizard appearing? anyway, here you need to know to start the urp converter wizard. after i've done that. crash. 😄

[oblique cosmos]

On the bright side, the DOTS guide thingy posted on the forums actually tells you about that. So that's nice. Anywho, I for one will go sit in a corner now and feel ashamed for getting bested by file management!

[harsh owl]

A UnsafeMultiHashMap is just a faster NativeMultiHashMap so long they are both read only?

[misty wedge]

It doesn't have any leak detection

Or does it now that that's built into the allocator? I'm actually not sure

[viral sonnet]

there's no real speed differences to speak of

[harsh owl]

but one has thread safety and the other does not

there is no way thread safety is zero cost

[viral sonnet]

you could say there's some overhead in Native because of safety system but that can be turned off and isn't in builds

[gentle gyro]

Delayed waiting...

[viral sonnet]

i'll have to check, there were some rewrites in the hashmaps.

but not for multiHashMap

yeah, nothing changed in multiHashMap. NativeMultiHashMap uses UnsafeMultiHashMap. UnsafeParallelHashMapBase uses atomic operations via Interlocked. what you are maybe referring to is NativeHashMap and NativeParallelHashMap. those were rewritten to work either with only single thread or multi threaded.

[rotund token]

As enzi said this only exists in editor (and even then can be turned off in burst jobs). In a build you should expect no difference it's all stripped

[gentle gyro]

ahh

[harsh owl]

I guess I will use the normal one

[rotund token]

In general (but not all) the native version just holds a pointer to the unsafe version

[raw mica]

Indeed. Native prefix is to signal there is safety checks for lifetime and r/w safety with the job system.
Leak checking is handled elsewhere via the memory system.

Performance should be nearly identical between Unsafexxx and Nativexxx when safety checks are off. Safety checks are on by default in the Editor and unavailable in standalone player builds. However, we have added a slew of safety checks that do not rely on the engine job safety system (primary around bounds checking) that can be forced on in standalone player builds via defining UNITY_DOTS_DEBUG

Note there is a bug in Burst compiled builds where some of our safety methods may be removed by Burst in standalone player builds. So if you have a crash in standalone player builds originating from a DOTS package, a good procedure for now is to disable burst and define UNITY_DOTS_DEBUG

[misty wedge]

Are there any use cases for unsafe collections besides nesting?

[rotund token]

I have a few but they are in the realm of super hacky things

[misty wedge]

I meant for normal people 😅

[rotund token]

Like a shared static with an unsafe hash map pinned kind of hacky - something that bypasses the job safety

[gentle gyro]

! Would this bug crash in Mono or just IL2CPP???

[raw mica]

Less than there were historically. The primary reason for Unsafe vs Native was that Native containers contained a managed safety type called DisposeSentinel which made using Native containers in any codepaths expecting a generic constraint of where t: unmanaged impossible.

We no longer rely on DisposeSentinel so Native and Unsafe containers are all unmanaged

Use cases now would be you really want to avoid safety checks for various "I know what I'm doing" codepaths

[misty wedge]

Yeah I know the history, that's why I was asking now that dispose sentinel is gone and all the non parallel collections are unmanaged and can be put in components

[raw mica]

You can disable safety via attributes but that isn't always possible based on what you're trying to do

[misty wedge]

Something else I was wondering, is there any performance benefit in using IJobChunk over IJobEntity if you are only processing entities? (converting old Entities.ForEach to job structs)

Since it's a lot more boilerplate

I guess if the system shares component type handles for multiple jobs in that system they only need to call update once?

[raw mica]

Exceptions thrown from Bursted codepaths in standalone player builds will abort the application and present the exception message in the player.log be it a mono or il2cpp method. The bug I'm referring to only applies to bursted codepaths that crash on a safety message that should be thrown only when UNITY_DOTS_DEBUG is defined. Without defining that, there is no safety checks in collections so you may crash and not get any messages but just a stack trace

[gentle gyro]

So just to confirm: if its this bug it should crash in Mono as well?

[raw mica]

The bug in burst doesn't prevent or cause crashes. The bug is that today you might crash due to a problem in your code and not get any messages about what went wrong due to a lack of safety checks -- mono and il2cpp. You can add in extra safety checks by defining UNITY_DOTS_DEBUG in standalone player builds. The bug I am referring to is that those extra safety checks may not be enabled so you're back to the original case of getting a crash due to a problem in your code and no nice message

[gentle gyro]

Got it

[raw mica]

But if you disable Burst and define UNITY_DOTS_DEBUG you should get a nicer message (for standalone player builds -- this isn't required in editor). We have some more improvements in this area that will be in the next release, but it should cover most code in the current release

[viral sonnet]

i'm trying to use aspects in IJobChunk. i've got the most covered but I don't know how to write back. ```[BurstCompile]
public partial struct ChaseJob2 : IJobChunk
{
public TransformAspect.TypeHandle TransformAspect_WriteHandle;
[ReadOnly] public ComponentTypeHandle<Speed> Speed_ReadHandle;
[ReadOnly] public ComponentTypeHandle<Target> Target_ReadHandle;

        [ReadOnly] public ComponentLookup<LocalToWorld> LocalToWorld_Lookup;
        public float dt;

        public void Execute(in ArchetypeChunk chunk, int unfilteredChunkIndex, bool useEnabledMask, in v128 chunkEnabledMask)
        {
            var speeds = chunk.GetNativeArray(Speed_ReadHandle);
            var targets = chunk.GetNativeArray(Target_ReadHandle);
            var ltws = TransformAspect_WriteHandle.Resolve(chunk);
            
            for (int i = 0; i < chunk.Count; i++)
            {
                var ltw = ltws[i];
                var speed = speeds[i];
                var target = targets[i];
                
                var targetPos = LocalToWorld_Lookup[target.target].Position;
                var chaserPos = ltw.Position;
                var dir = targetPos - chaserPos;
                dir = math.normalizesafe(dir);
                
                ltw.Position += dir * speed.speed * dt;

                ltws[i] = ltw; // has no setter
            }
        }
    }```

[gentle gyro]

Do you know of any bugs that would ONLY crash on IL2CPP for Quest 2 and work fine on Rift S and Editor. Seems to be related to the new rendering pack

[viral sonnet]

the code generated IJobEntity version just uses var ltws = TransformAspect_WriteHandle.Resolve(chunk); and sends a ref to the Execute. so i thought cool, seems we don't need to write back but it's not working ^^

[misty wedge]

Anyone that uses rider know if you can set the struct member autocompletion to only show missing members, and not all? Visual studio only showed missing ones, which was nice 😦

[raw mica]

Something else I was wondering is there

[frosty siren]

does 0.50+ use regular build pipeline?

[wraith hinge]

no

(the 1.0 prerelease probably will, though. it works in master right now.)

[misty wedge]

I thought 1.0.0-exp.8 already supports the normal build pipeline?

[wraith hinge]

you know, i should check.

[misty wedge]

Otherwise I'm not sure why my builds using the normal pipeline have worked 😅

[viral sonnet]

i've used File->Build Settings->Build with 1.0exp - guess that's what tony meant with regular pipeline

[wraith hinge]

i see. reading up on it, it's a somewhat under-tested setting in preferences in 1.0-exp, but if it works, great! or maybe il2cpp builds are just crashing sooner than you could notice that there was a problem 😅

[coarse turtle]

i think so far the only bug i noticed with il2cpp builds have been the SceneSectionStreamingSystem (although I did this with the previous build config assets) 🤔

0x00007FF8F2FC0221 (GameAssembly) [some_path\cpp\Unity.Entities9.cpp:20720] DotsSerialization_CreateReader_mD449775FCAA012FBD6CF131CB7C488037D473745 
0x00007FF8F33D68A4 (GameAssembly) [some_path\cpp\Unity.Scenes.cpp:13013] AsyncLoadSceneOperation_UpdateAsync_m25C4533BE332945DB9ACFCF80AB913B132653253 
0x00007FF8F33E9CF1 (GameAssembly) [some_path\cpp\Unity.Scenes.cpp:25911] SceneSectionStreamingSystem_UpdateLoadOperation_mC3CC60795D148C4D5B4DBC1BAB083B129ABECEE7 
0x00007FF8F33E92B3 (GameAssembly) [some_path\cpp\Unity.Scenes.cpp:25646] SceneSectionStreamingSystem_ProcessActiveStreams_mE7326AB99FF0AC55EDB08634A5B9449DA849E7FE 
0x00007FF8F33E9141 (GameAssembly) [some_path\cpp\Unity.Scenes.cpp:26863] SceneSectionStreamingSystem_OnUpdate_m48E550F61E3B9720FA4CBC7B99BE33FD55750A85 
0x00007FF8F2F9F4A7 (GameAssembly) [some_path\cpp\Unity.Entities7.cpp:27485] SystemBase_Update_m795994EE60C054DFA78C313A60ED025B2668C9B4 
0x00007FF8F2EF92CF (GameAssembly) [some_path\cpp\Unity.Entities1.cpp:24782] ComponentSystemGroup_UpdateAllSystems_m78B1AA9EAF07133497E9BA556B592AC93935441C 
0x00007FF8F2F9F4A7 (GameAssembly) [some_path\cpp\Unity.Entities7.cpp:27485] SystemBase_Update_m795994EE60C054DFA78C313A60ED025B2668C9B4 
0x00007FF8F2EF92CF (GameAssembly) [some_path\cpp\Unity.Entities1.cpp:24782] ComponentSystemGroup_UpdateAllSystems_m78B1AA9EAF07133497E9BA556B592AC93935441C 
0x00007FF8F2F9F4A7 (GameAssembly) [some_path\cpp\Unity.Entities7.cpp:27485] SystemBase_Update_m795994EE60C054DFA78C313A60ED025B2668C9B4 
0x00007FF8F27DBD06 (GameAssembly) [2022.2.0b8\Editor\Data\il2cpp\libil2cpp\vm\Runtime.cpp:608] il2cpp::vm::Runtime::InvokeWithThrow 
0x00007FF8F27DB7A9 (GameAssembly) [2022.2.0b8\Editor\Data\il2cpp\libil2cpp\vm\Runtime.cpp:593] il2cpp::vm::Runtime::Invoke

[viral sonnet]

have you read the notice about havok dots physics where you need Unity Pro+?

This confuses me 😄 a lot! personal and plus is free. at pro it's not? what am i missing?

pro now for consoles?

oh i think i answered this myself. pro is needed for consoles

[raw mica]

This is a known issue. It's fixed internally, but won't be out until at least 2022.2.0b10 (but it might be 0b11)

[rotund token]

Think pro has been needed for consoles since 2021

[viral sonnet]

yeah i was just a bit confused in regards to havok

[rotund token]

I think the change mentioned on forums for havok in 1.0 is its now included with a unity pro licence

No longer need a seperate subscription

[safe lintel]

hope its still available for us poors 🙂

[safe lintel]

oh dear I guess it isnt

hard to showcase demand for a product if all you can do is beg on the forums(given how prevalent application metrics are)

[rotund token]

yeah unfortunately the announcement did not sound like it was available outside of pro, but we'll see i guess

[safe lintel]

gotta say I think gigaya getting canned was one of the largest community outcries and that was for nought so dont really have expectations on this decision changing

[rotund token]

quick question before i start breaking things, anyone tried to manually create multiple ISystems of the same type in a single world?

from what i understand this is not supported but I kind of have an idea I want to try

[viral sonnet]

i don't see how this can work 🙂

[rotund token]

why not

[viral sonnet]

aren't there way too many cases where the type is cached somewhere?

(and no, i've never tried 🙂 )

[rotund token]

there's nothing in the code as far as i can tell on creating a system

that checks or would break

internal SystemHandle CreateUnmanagedSystem(Type t, bool callOnCreate)

anyway let me explain what i'm thinking

my saving currently works by simply adding an attribute [Saved] to a component

It finds all these, creates 1x ComponentDataSave (or buffer version etc)
public sealed unsafe class ComponentDataSave : ComponentSave
per component

the only difference between these on a per component is a ulong stableTypeHash

there's no generics

what i'm wonder is if I could turn this ComponentDataSave into an ISystem

1 per [saved]

it's basically just a subsystem atm, sets up handles, schedules jobs

in theory there is nothing in here that would stop this being an ISystem

so what i'd like to do is create 1 per [saved] attribute and put them under a single system group

and simply stop the system group updating unless it needs to save/load

this way you could more naturally inject new save alternatives into it in a more dots fashion instead of inheriting my ISaver in some class and i find it via reflection

[viral sonnet]

certainly worth a try.

[rotund token]

(i'm basically procrastinating updating my game library because too many changes =D)

[viral sonnet]

hehe. contrary to what i said before - it should be possible. although i hardly, if ever, saw a system getting added twice in a world so that's where my hesitation comes from

[rotund token]

i can totally do this in 0.51 with systembase

[viral sonnet]

havok should be free for unity personal/plus

[rotund token]

i hope so (even though I have a pro license) but there was no specific mention of it in the announcement

A new cohesive experience for Havok Physics for Unity is being rolled out, and will be included in Unity Pro, Unity Enterprise and Unity Industrial Collection offerings.

https://forum.unity.com/threads/experimental-entities-1-0-is-available.1341065/

[rotund token]

the one tricky bit is writing a value to each ISystem

[viral sonnet]

what you're trying to do is the closest we can possibly get to a generic ISystem

got an answer in the forum We will continue to evaluate the uptake of Havok Physics for Unity upon release to understand how users are interacting with it, whether there's significant demand for it in Free/Plus tiers and ultimately, drive for better user experience. https://forum.unity.com/threads/dots-development-status-and-next-milestones-september-2022.1341077/#post-8489882

[rotund token]

Havok Physics for Unity's exclusive availability in Unity Plus, Enterprise and UIC upon release with ECS for Unity 2022
I feel like that is meant to say Unity Pro

also that does sound like atm no plans for free

[safe lintel]

yeah its pro/enterprise/uic only now

[viral sonnet]

right, we have to see if there's demand in free/plus tier ...

i don't know what i'm paying for in plus ... haha

anyway, var statePtr = ResolveSystemState(GetExistingUnmanagedSystem(TypeManager.GetSystemTypeHash<T>())); that could be a bit difficult. if (_unmanagedSlotByTypeHash.TryGetFirstValue(typeHash, out ushort handle, out _)) { var block = _stateMemory.GetBlock(handle, out var subIndex); return new SystemHandle(block->States[subIndex].SystemHandle.m_Entity, handle, block->Version[subIndex], (uint)SequenceNumber); }

they rely on hashes and types a lot

really, plus is the shit tier for games in development 😄

[rotund token]

but if you notice

_unmanagedSlotByTypeHash

is actually a multi hash map

it just returns the first

but it can hold more than 1 of the same type

[viral sonnet]

oh, good catch

[rotund token]

i don't need to look this up either

i'll have the specific handle when i create it

[viral sonnet]

and the hashmap is private

well, the handle is all you'd need anyway

the downside on the system handle, i've not looked into this. how to get a ref from the system struct

[rotund token]

i can get the systemstate easy enough

there used to be a way to get a ref<T> of an ISystem

but i think it was removed

[viral sonnet]

a unity dev said they introduced the handle to prevent writing to systems from outside. but that's exactly what you need now

[rotund token]

(if anyone else is reading this, do not do what i'm trying to do. just ignore this entire conversation)

[viral sonnet]

an option would be to create the system struct yourself and register it

[rotund token]

i could, i know how it's allocated

it would require a lot of extensions and probably modification of the entities package

which i tend to refuse to do

(extensions fine, modifications no)

[viral sonnet]

yeah

[rotund token]

                var sysHandle = new SystemHandle<T>(handle, block->Version[subIndex], (uint) SequenceNumber);
                void* ptr = (void*)(IntPtr)block->SystemPointer[subIndex];
                return new SystemRef<T>(ptr, sysHandle);```

ok this is how it used to work

in 0.51

let me diff

[misty wedge]

It didn't seem to break anything when I tried it, but there could be edge cases 🤷

[viral sonnet]

oh right, i still have the unanswered question. how to add an ISystem that has [DisableAutoCreation] - i think i never found out how to do that in a non weird way

[rotund token]

can't you just call CreateSystem<T>

then add it to whatever system group you want

            default(ComponentSystemGroup).AddSystemToUpdateList(systemHandle);```

something like that

(obviously don't use default)

SystemHandle CreateSystem<T>(this World self)

/// This can result in multiple instances of the same system in a single World, which is generally undesirable.
///```

hey it's commented you can have multiple systems fine

1 less concern

maybe this is doable

[viral sonnet]

damn, i can't remember anymore where i got stuck. it was with the generic ISystem test i did which had a bunch of problems 😄

wait, i said, how to write to the system but that's dumb anyway now that we have a system entity

[misty wedge]

If you really need to you can just memcpy values into the struct

[viral sonnet]

and you can write your hashes to these entities. so seems to be just a case of manually creating the systems with the method you posted

how would you memcpy without the ptr?

[misty wedge]

You can get a pointer to the struct

But it requires internal access

[viral sonnet]

if we have the ptr we can also get a ref struct. no need for memcpy

[misty wedge]

What type are you going to return?

[viral sonnet]

type is known. i dunno, whatever the system is called 🙂

[misty wedge]

You'd probably need to make generic methods using reflection though

[rotund token]

yeah that might be the best approach

i can just use the entity

avoid a lot of hacks

sadly work starting so now i'm going to think about this all day

[viral sonnet]

well it will work and the implementation seems even easier than expected

now i'm thinking of going back to the generic stat system

i just need to save all that would be generic in the system entity comp

generic jobs already work fine

[misty wedge]

I was thinking of something else, I iterate over all systems and call a method on the ISystem, so I don't know the type there at compile time 👍

[patent kelp]

Hiya, can you think of why this job isn't working? Basically, when debugging, the function exists on the marked line

Entities
.ForEach(
(Entity e, ref LocalToWorldTransform position, in MoveToDestinationComponent move) =>
{
float3 dist = move.dest - position.Value.Position; <<<<<<<<<<<<<<<<<<<<<<<< F10 exits here
float len = length(dist);
dist = normalize(dist);
position.Value.Position += dist * deltaTime * speed;

                if(len < 0.2f)
                {
                    ecb.SetComponentEnabled<MoveToDestinationComponent>(e, false);
                }
            }
        ).Schedule();

[rotund token]

what do you mean by exists?

as in you can't step through with a debugger?

because that line is very much not doing anything exciting

[patent kelp]

exits... sorry

[rotund token]

are you verifying this with a debug log or something?

[patent kelp]

F5... F10

[rotund token]

so yeah the debugger from my experience does not work reliably in entities.foreach

[patent kelp]

Debug.Log it is ... thanks

[misty wedge]

Wasn't there some way to go to the generated code using rider?

[rotund token]

rider only wants to take me to the generated code most of the time >_>

[misty wedge]

I had that issue before, now I can't get there 😅

[rotund token]

you can try toggle include source generated files

i can't remember if that's it

[misty wedge]

😦

[rotund token]

;'(

[misty wedge]

IJobEntity really removes a lot of boilerplate if you don't require chunk level optimizations

[patent kelp]

For logging in an Entity.Foreach... is there something special for this to work? I don't see any logging
Entities
.ForEach(
(Entity e, ref LocalToWorldTransform position, in MoveToDestinationComponent move) =>
{
Debug.Log("0");
float3 dist = move.dest - position.Value.Position;
Debug.Log("1");
float len = length(dist);
dist = normalize(dist);
position.Value.Position += dist * deltaTime * speed;
Debug.Log("2");
if (len < 0.2f)
{
ecb.SetComponentEnabled<MoveToDestinationComponent>(e, false);
}
}
).Schedule();

[rotund token]

if it's not logging then the job is never entering

[rotund token]

i just need an optional

{
}```
i can implement

and i could replace 90% of my ijobchunks

i should probably request this

{
    return chunk.DidOrderChange(LastSystemVersion);
}```

and if false it does not call the Execute()
or

private NativeList<bool> buffer;

bool ChunkExecute(in ArchetypeChunk chunk, int unfilteredChunkIndex)
{
    if (!buffer.isCreated)
        buffer = new NativeList<bool>(0, Allocator.Temp);
    return true;
}```

[viral sonnet]

the source gening is pretty good, just look at this var __Unity_Transforms_TransformAspectTypeHandleArray = __Unity_Transforms_TransformAspectTypeHandle.Resolve(batch); var speedData = InternalCompilerInterface.UnsafeGetChunkNativeArrayReadOnlyIntPtr<Speed>(batch, ref __SpeedTypeHandle); var targetData = InternalCompilerInterface.UnsafeGetChunkNativeArrayReadOnlyIntPtr<Target>(batch, ref __TargetTypeHandle); int chunkEntityCount = batch.ChunkEntityCount; int matchingEntityCount = 0; if (!useEnabledMask) { for (int i = 0; i < chunkEntityCount; ++i) { var __Unity_Transforms_TransformAspectTypeHandleArrayArray = __Unity_Transforms_TransformAspectTypeHandleArray[i]; ref var speedData__ref = ref InternalCompilerInterface.UnsafeGetRefToNativeArrayPtrElement<Speed>(speedData, i); ref var targetData__ref = ref InternalCompilerInterface.UnsafeGetRefToNativeArrayPtrElement<Target>(targetData, i); Execute(ref __Unity_Transforms_TransformAspectTypeHandleArrayArray, in speedData__ref, in targetData__ref); matchingEntityCount++; } } actually uses ptrs and refs even for in

[misty wedge]

I mean in may technically be slower than ref, but yes it is very well done

[viral sonnet]

it's much better than the usual chunk.GetNativeArray and local struct copies

[misty wedge]

There's not much they can do about that though since in and ref is also used to define the dependencies

[rotund token]

is it in burst?

[misty wedge]

I actually have no idea 😅

[rotund token]

last i checked i thought burst treated them exactly the same

[viral sonnet]

nah, never seen a difference

[misty wedge]

Then disregard what I said 👍

[viral sonnet]

it probably is for mono (maybe)

[misty wedge]

Even outside of burst it should be the same speed if you don't accidentally copy the struct

[rotund token]

in theory i assume a compiler could even optimize an in somehow

[misty wedge]

How would it? It could technically break your code if the compiler doesn't copy the struct

Unless it does some magic to avoid that possibility somehow

[viral sonnet]

what concerns me a bit is the amount of boilerplate for enabledMask

IJobChunk jobs i've seen with enabled support don't have all that masking. i dunno if that's needed

[misty wedge]

It's the normal code that is also used to check the mask iirc

[rotund token]

this looks like a near duplicate of the utility method

[misty wedge]

^

I guess it's just inlined to be faster? 🤷

[viral sonnet]

i'm aware of the methods existence. ok, kind of relieved then 😄

[misty wedge]

How do you get a dynamic buffer in a query?

[rotund token]

you don't

😄

[misty wedge]

[rotund token]

(don't think it's supported yet)

i assume at some point

[misty wedge]

I also wish it supported destroying entities, but it seems that breaks the iterator

[viral sonnet]

really, they mentioned somwhere, less things should break the iterator. hm

[misty wedge]

One of the devs said you can't call DestroyEntity a few days ago, I never tested it though

[viral sonnet]

it was kind of annoying how careful you have to be when adding buffers and filling them with data

at least a local ecb works

[rotund token]

just use a local ecb

damn too fast enzi

😄

[viral sonnet]

😄

[misty wedge]

Yeah I know it's possible guys 😅 I just meant I wish it code generated it

[viral sonnet]

SystemAPI.Query is kind of in a meh state. it doesn't quite know what it wants to be yet. right now i haven't found much usage. most where i wanted to use it I've then converted to an entityQuery

codegen for it is quite complex

[misty wedge]

My main use would be for cleaning up + getting rid of clean up entities on world disposing

[misty wedge]

This works for me, not sure if that's intended:

[viral sonnet]

seems intended. so DBs work after all.

[misty wedge]

I guess you just can't get a read only dynamic buffer

[rotund token]

oh there we go

[safe lintel]

ahhhh this seems like new behaviour? instantiating a prefab from an entity that sits inside of a hierarchy will parent the new entity to the same hierarchy level?

[viral sonnet]

was that not happening before when the parent also gets converted?

[safe lintel]

prefab shouldnt be part of a hierarchy but let me see

nope, prefab def not part of that same hierarchy(just a root space entity)

[proud jackal]

With how the generator works if it works it's most likely intended.

So currently Query works with RefRO, RefRW, Managed Comps, SharedComp, EnabledRefRW, EnabledRefRO, DynamicBuffer and Aspects

[viral sonnet]

so the only thing that's missing is monobehaviour and read only dynamic buffers.

[viral sonnet]

how can i get the ref from RefRO then?

ok, i'm not supposed to get a ref from RO. makes sense. in case it doesn't return a local struct copy all is good. 🙂 haven't tested that, seems there's a struct copy involved though

in which case i have to ask, why use a RefRO in the first place

[proud jackal]

Quick heads up, as I can see you're using the thing, .WithEntityQueryOptions will be renamed .WithOptions.

As for RefRO, it's essentially just a pointer to the component memmory, but ReadOnly so no copies here, well, until you copy values over that is 😛

[viral sonnet]

ok cool, thanks!

SystemAPI.GetSingleton or SystemAPI in general is not supported in MonoBehaviour code? it errors InvalidOperationException: No suitable code replacement generated, this is either due to generators failing, or lack of support in your current context

haha, well, i guess it's called SystemAPI for a reason 😅

[rotund token]

what world / EM would it even be getting it from

[viral sonnet]

right 😄

[late mural]

tryna optimize a for loop using the parallel for loop job thing, wondering though what for the 2nd parameter in .Schedule(ForLoopLength, ItemsPerProccessingBatch), is there a guide on what to set it to in order to get optimal speed? My for loop can operate on anything from 0 fluid particles to over 30 thousand incase that helps

[wraith hinge]

basically, the batch size is the amount of work that is scheduled at a time. you're trading off between the overhead of scheduling more units of work (if you have a small batch size) and risking that one core will run longer than the others (if you have a large batch size). the only real way to decide is to try it on your target hardware and see what's best (or if it even matters)

[late mural]

ah ok, im guessing it also depends on what work im doing in the for loop then?

[wraith hinge]

yes

if i wasn't sure what to do, i would probably just set it to something arbitrary like 1/128 my total workload and forget about it, and come back to it maybe never, or maybe if i looked at the profiler and saw a million tiny jobs of this thing dominating the profile

[rotund token]

I usually use a value of 1 or 256

[late mural]

ok ill mess around with it, and see what works best!

[rotund token]

1 is more of a special case where you have really heavy work and not many indices

So you want each element to have its own worker to go as wide at possible

[late mural]

ok, im not really doing much heavy work (just checking each particle's list of 6 contact planes and seeing if any of them match any of the shapes that are checking for collisions), Thanks for all the info yall!

[rotund token]

But generally from my benchmarks on high index counts anything above 64 isn't that noticeable performance difference

Hence I just use a nice number as a default unless my profiling points to an issue

[late mural]

ok awesome!

[rotund token]

There are a couple of other values that might make sense depending on your data

If you're using a grid packed in a single array

It might make sense to use width as the value

Therefore each worker gets 1 row

You can do this to be smart and minimise bound checks

Actually that's probably more useful for ijobforbatch

[late mural]

ok that sounds clever, ill mess around and see what i can get working best!

[viral sonnet]

hm, i currently have a case for a system with RequireForUpdate<SingletonStruct> - the SingletonStruct exists on a system entity. yet the system that has the RequireForUpdate doesn't run. (it also has the RequireMatchingQueriesForUpdate tag)

well this is weird. filtering with c:SingletonStruct in the hierarchy has 0 results. clicking on the system entity, the comp is there.

are system entities not picked up or smth?

[late mural]

hmm it appears that the parallelfor job does not like me trying to write to a NativeList, is there anyway to make the NativeList have parallel writing?

[viral sonnet]

NativeList<T>.ParallelWriter

[late mural]

aha thanks, and where would i put this?

[viral sonnet]

in the job, instead of the NativeList you have now

[late mural]

oh ok thanks!

[viral sonnet]

it can't resize though, so set the list.Capacity before scheduling the job

[late mural]

ok and making certain that the capacity is seperate to the length right, so i dont end up with tonnes of ints equalling 0?

[rotund token]

Capacity is just how much the list can store

Length will be how many elements were added

[late mural]

ok good i think lol

[late mural]

ok, yas, thanks!

[rotund token]

You need to make sure you set capacity high enough it can add all the elements you want

If you are iterating 3000 things and each can add 1, set capacity to 3000

[viral sonnet]

whelp, i have set some debug logs to see if the system is running. changed it to alwaysUpdate and put a SystemApi.HasSingleton in place. now it's working. but before that it error'ed on some other stuff that i know works. (couldn't find a spell) - there are some very weird caching issues going on

[late mural]

ok ill just set that to the max particle limit the container has, cause if ever the amount of particles goes above that it instead just overwrites the beginning, so should work great, thanks!

[viral sonnet]

with RequiredUpdate the system still won't run. which is funny because HasSingleton clearly works

__query_175196122_0 = this.CheckedStateRef.GetEntityQuery(new Unity.Entities.EntityQueryDesc{All = new Unity.Entities.ComponentType[]{Unity.Entities.ComponentType.ReadOnly<NZSpellCasting.SpellLoaderMetaData>()}, Any = new Unity.Entities.ComponentType[]{}, None = new Unity.Entities.ComponentType[]{}, Options = Unity.Entities.EntityQueryOptions.Default | Unity.Entities.EntityQueryOptions.IncludeSystems}); THERE IT IS. Unity.Entities.EntityQueryOptions.IncludeSystems! This is clearly missing from RequireForUpdate<T> preventing a system from running when the singleton is on a system entity. @wraith hinge would you kindly forward this 🙂 or is this intended behaviour?

[rotund token]

i haven't been following but based off your result is this a fail on RequireForUpdate with a system entity?

follow up question, why would you do this 😆
i.e. when would you not expect the system component to exist?

[proud jackal]

Well if you did some setup stuff in another systems OnStartRunning, then the component would surely be missing until everything has loaded in 😆

[rotund token]

OnStartRunning shouldn't execute until all Systems are created unless you were manually creating systems?

[viral sonnet]

yep, it's setup stuff and the reason is that the system which does the setup has to wait for a subscene to load

[rotund token]

oh subscene loading

i feel like thats a common issue

but i disliked putting RequireForUpdate on all my settings files i all my systems

so i simply turn off Simulation/Presentation Updating until my 3 (client/server/client+server) subscenes have loaded

therefore i can guarantee all settings are loaded onto entities before any system starts running

avoids 1+ query check on nearly every system....

[viral sonnet]

yeah true, i also don't feel like having so many RequireForUpdate is a great solution. but it's pretty clean when it comes to unloading/reloading the game. don't have to do any management of system groups

[rotund token]

i have 1 system that handles it, it's pretty much hidden from the world

it oes other work like only like scene X, Z into client, Y, W into server, A,B,C into both

setup load rules such as only load subscenes in range

this is the 0.51 version, it's pretty simple. currently deciding how to handle 1.0 netcode group changes

[viral sonnet]

nice stuff 👌 and when the player logs out you disable all the simulation groups?

[rotund token]

when a user is disconnected i destroy client world and re-created it

i always have a client world (except in dedicated server)

but it's disposed and re-created every time a game ends

[viral sonnet]

i'm not doing any multiplayer (yet) but i can surely clean this up better to take advantage of enabling/disabling system groups to prevent such very redundant checks. thanks

[rotund token]

i used to just put requireforupdate

it just bothered me so much after a while

[viral sonnet]

yeah, i see it the same way. i make 1 sytem now handling the sub system groups. i'm okay querying something once a frame.

[rotund token]

i'm a huge fan of system groups that limit updates these days

(very happy system groups became SystemBase in 1.0)

[viral sonnet]

haven't realized until now. that's pretty cool

does RequireForUpdate also require the [RequireMatchingQueriesForUpdate] tag? i don't think it does

[rotund token]

no

(also they do different things)

[viral sonnet]

yeah true. just wanted to make sure 😊

[last sandal]

Hey guys I am interested in exploring dots for my new project. Do you have some good tutorials, docs I can look at.
I plan on building a settlers game in a pherical procedural generated world which mean I would need navmesh and collisions physics to work. Is there any good material on these subjects?

[rotund token]

navmesh will take you a little work (it's doable)

physics is pretty solid

probably should start with manual

but they also released a new dots guide as part of the samples

https://github.com/Unity-Technologies/EntityComponentSystemSamples

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

i haven't really looked at it too much but from a quick glance gives a basic start point

[rotund token]

not sure how but i kind of broke live conversion =S

[brave field]

🥲 Until now subscene conversion still kind of broken

[dense storm]

Anyone know what component datas I need to render sprites in pure dots?

[narrow scaffold]

Are pointlights supposed to work in Entities Graphics & URP? For me only directional lights work at the moment.

[muted star]

Hey everyone, is 1.0 working okay? Is it time to upgrade from 0.51?

[narrow scaffold]

I managed to get my project upgraded but there are still a lot of issues. For me it's basically crashing every time I save my scene. Subscene conversion is kinda broken. The new baking system is missing some features I needed (For example removing components). But if you want to get started you can definitely try. Just make sure you create a backup/new branch.

[rustic rain]

Can't say

too much critical bugs (at least for me)

can't even keep working

[rotund token]

if i ignore netcode and the complete rework of how systems/groups get assigned to different worlds

honestly it's been pretty smooth

but netcode is bending me over backwards atm

[brave field]

🥲 Still slowly upgrading project

[rotund token]

starting to get the hang of this though

huge pain to update everything but it's starting to make sense

and it can do some really useful things that solve some major issues i had before

as in i can finally setup my libraries to work with or without netcode without the need of a dependency without having to write a bunch of conditional code

[viral sonnet]

has something drastically changed for netcode or in general?

[dense crypt]

How does il2cpp/burst handle integer overflows in a job?

[viral sonnet]

it doesn't. it overflows 🙂

[raw mica]

re Burst overflow: Specifically: https://forum.unity.com/threads/does-burst-support-the-unchecked-keyword.761495/#post-5072381

[robust scaffold]

Ugh. Burst spoils me. I need deterministic floating point math in compute shaders... fixed point is pain

[olive kite]

I keep getting a crash and I think it is due to this error: "System.IndexOutOfRangeException: Index {0} is out of range in DynamicBuffer of '{1}' Length." any idea why that might happen?

[wraith hinge]

@olive kite is it in a build or editor? if editor, you can disable burst on the code in question to see a better error. if in a build, builds are just in bad shape right now, need patch release

[devout prairie]

Whatcha working on now Mr Flaks?

[olive kite]

in Editor, I'll try that. It is in ScheduleParallel but didn't think I was modifying any values in the buffer. I'll also try not running in parallel

[wraith hinge]

also i assume it isn't literally an editor crash, but just an exception? if the editor itself is crashing to desktop, that is likely unrelated and something that needs another editor beta or two

[olive kite]

oh yes, it is an exception. I am trying to trouble shoot a build crash and am having a hard time doing so, I thought this might be what is causing it

[wraith hinge]

i mean, it could be. but as i said, builds are not in good shape anyway, so if it were me i would just wait

[olive kite]

Next Fest demo is live 🙂

[wraith hinge]

¯_(ツ)_/¯

well, at least try to stick to mono builds then

and turn off stripping

(these two things work around a bunch of our known bugs)

[olive kite]

yeah I am doing mono, though I don't believe I did anything to turn off stripping

thanks, i'll look into it

[wraith hinge]

i have a pr open to master now that makes things work again with high stripping, but it hasn't landed yet and then i need to backport it to the experimental branch

[olive kite]

oh cool, rough eta?

[wraith hinge]

#1026945628753891368 message

[olive kite]

yeah I have stripping disabled and using Mono. I'll try to disable Burst although it only seems to happen with 1500+ pathfinding agents active and it's rare, so maybe i'll leave it running overnight

[wraith hinge]

oof

[zenith stirrup]

hey guys. is there any trick to get a Baker to run?
my Baker<T> is neither called for prefabs, nor for scene GameObjects

[remote crater]

I have a mesh who's at a wrong front facing angle. In GameObjectUnity I just make it a child of an empty object and adjust its rotation vector,but DOTs doesn't convert it right if it is a child of an empty. How do I adjust the angle of a Mesh in DOTS so when scripts change the angle, it doesn't nuke my offset rotation?

My DOTS ECS game is finally nearing close to being super super fun. Thanks guys for all the help over the year and a half.

https://youtu.be/1jEwSdsVyeY

[proud jackal]

No Runtime baking, so Scene GameObjects can't be baked, but SubScenes can, so put your GO in a SubScene and you're all set! :3 (To make prefabs work, you can simply call GetEntity in your baker from your prefab field)

[rotund token]

Yeah no more update in world or custom simulation groups for client server etc

Now uses flags from entities

[viral sonnet]

oh, last time i used netcode was in 0.6 😄 care to explain or have some code snippets?

[remote crater]

Is it possible to rotate an Entity that has a Mesh rotated inside it?

I haven't had a question in weeks since DOTS / ECS was in such a good spot. My only other question is: How could I put graffiti paint on a mesh in HDRP DOTS/ECS? In regular, you just skin it again, but DOTS/ECS doesn't let you skin on the fly.

What's cool is that my game would be super super fun and have a patch out if I got these two very basic questions answered.

Am I articulating it correctly? Do people know what I mean by a badly centered mesh that you fix by rotating it?

Like the initial mesh doesn't align 0,0,0, so you by hand rotate it.

But when you put it in a DOTS/ECS Systembase, the rotation of forward facing there, nukes your offset since you're using the translation.

[rotund token]

For painting I'd probably be going this with shaders not actually changing the mesh

[remote crater]

I haven't looked much into shaders, but now you give me reason to. Thank you. I might do like I did with particles and make a random shader generator.

[rotund token]

Translation is local space you probably need to use the local to world to get the actual world forward

[remote crater]

I have a script that already works for other ships.

I'm trying to align another mesh with it, and they work and shoot like a ship.

But they don't face forward.

Because the mesh itself is off centered.

I could do an ugly hack where I test the entity type in the loop,slowing the game down and readjusting it rotationally.

[rotund token]

Why not just fix the mesh

[remote crater]

In GameObjectUnity it allows you to make an Empty GameObject and then place a GameObject as a child then rotate it there.

I did not make the mesh myself. It came from Unity Asset Store.

[rotund token]

I hate offset meshes, position or rotation

Just an extra hierarchy call for no reason =/

[remote crater]

I didn't know I could edit a mesh.

[rotund token]

I haven't opened a modelling app in ages but from memory it's actually super easy

[remote crater]

It is not possible to edit a mesh in Unity. Ok. So I should try blender I suppose.

This is super valuable information.

I would have never thought of that. You'd think Unity would have some rudimentary mesh editor built in by now.

or an offset rotation for DOTS/ECS

Ok this is great, I thought it was a trivial issue, but it was not. I'm super glad I asked you guys about it.

God bless ya tertle,ain't no way I would fix this on my own,except hacky in line script

I may resort to that if the proper way is not working.

[remote crater]

ahahaha, it workd Tertle and it looks so freaking tacky and dumb like I wanted!

[rotund token]

Did you just change the model?

[remote crater]

https://youtu.be/BgXVP8tiC-k

Yah, I got blender and changed the model...Now I'm dangerous, I can make my own models.

You might not laugh, but this made me laugh, came out just as I wanted

Space is boring,right? So I'm doing the old Oregon Trail thing of an event every 30seconds to 3 minutes

Absurdity trail. Weird stuff and funny stuff,dialogue, trading, anaomolies to research and normal fights and strange fights

Like space dragons, looks so tacky, I love it!

https://youtu.be/BgXVP8tiC-k

[rotund token]

you use [WorldSystemFilter]

for example
[WorldSystemFilter(WorldSystemFilterFlags.ClientSimulation | WorldSystemFilterFlags.ServerSimulation)]

but there's also a cool little feature called ChildDefaultFilterFlags

which is what worlds your systems go into if you add them to this system group without specifying a group itself

[viral sonnet]

have they built this or is it core entities?

[rotund token]

core entities

this is not part of netcode which is the other useful thing (get to that in a sec)

so you can do something like this

    public partial class DebugSystemGroup : ComponentSystemGroup```

bit of a mouthful

[viral sonnet]

nice, thanks. have to look into it

[rotund token]

but basically this system group will be added to all worlds, including editor

and by systems that use this system group will by default be added to all worlds except editor

so yeah the other really useful thing, in particular for me who writes a lot of standalone libraries

is this is built into entities

so i can setup my systems to run either for a single world game or specify into client/server where it makes sense

without having a reference to netcode and/or conditionally compile it out

(or the better approach where i previously wrote a copy of the updateinworld in my core library that was compiled out if netcode was present)

[patent kelp]

Is there a trick to making a baker work? I have code nearly identical to this and a cube in the world with a MyAuthoring component attached... but the baker code never runs. Do I need to create an instance of the baker and run it?
https://docs.unity3d.com/Packages/com.unity.entities@1.0/api/Unity.Entities.Baker-1.html

[rotund token]

your cube is in a subscene yes?

[patent kelp]

The main scene

[rotund token]

that doesn't work

you must use subscenes

[robust scaffold]

Trying to do some deterministic multiplayer real time fluid simulation.

[patent kelp]

Is that documented somewhere? I Thought that I had been thorough...?

[rotund token]

i haven't read the documentation in a long time, but a quick search through it does feel like there is a missing page on subscenes

[proud jackal]

With no runtime conversion the rule is nice and simple.
SubScenes = Entities only
Main Scene = GameObjects only

(this being said, gameobjects can get references to the world, and as a result add entities, likewise, entities can have references to managed data, and even instantiate GameObjects)

But yeah, docs, samples etc. Those are still WIP

[patent kelp]

So, I just created a scene, added a cube with that MyAuthoring component in it, and put the scene under my main scene... main scene is in Red, and the cube is in blue... is this a good setup?

[rotund token]

create 2 scenes, 1 main, 1 your subscene for entity data

in your main scene create a gameobject, add a component called subscene

put your second scene in the field on the subscene script

but your entities inside that second scene

you can open/close the subscene with the buttons in the inspector or the toggle in the hierarchy

[patent kelp]

Wow... I would never have guess that...

[rotund token]

ideally have it closed when entering play mode

[whole gyro]

I've always just right clicked my main scene and click "New Empty Sub Scene"

[rotund token]

or that ^

[whole gyro]

Didn't even know you could manually add a subscene component or even that a subscene was registered via a component

[proud jackal]

Two ways to create a SubScene in the Editor, one is New Empty Sub Scene, the other is Empty GO with a SubScene comp, New Empty Sub Scene is just short hand for the latter

[rotund token]

you should see something like this peon (ignore the SubSceneLoadConfig script, that's my own thing)

[patent kelp]

Alright... I have that

[gentle gyro]

Have you seen this DOTS 1.0 tutorial?
https://github.com/Unity-Technologies/EntityComponentSystemSamples/tree/master/DOTS_Guide/ecs_tutorial

[rotund token]

ah there is a page with the info

[gentle gyro]

Very quick to set up (if your not building for Quest 2)

[patent kelp]

That is new to me.

[rotund token]

i couldn't see it when i had a very quick look

[patent kelp]

Thank you David777, Tertle, and Dani

[rotund token]

there's meant to be a link from the ecs package page to this

but i noticed it's broken

https://docs.unity3d.com/Packages/com.unity.entities@1.0/manual/concepts-intro.html#further-information

DOTS Cookbook: Information about data-oriented design and the fundamentals of Unity's ECS implementation.
DOTS Tutorial: A tutorial that guides you through creating a simple scene with Entities.

both of these just send you to dead links, seems like it should send you here instead
https://github.com/Unity-Technologies/EntityComponentSystemSamples/tree/master/DOTS_Guide

[gentle gyro]

I can't remember how I found this page...

[gentle gyro]

But yah, I do remember now coming across the broken link...

[rotund token]

pretty sure a unity guy posted it here when they added it post 1.0

but it's just part of the samples so it shouldn't be too hard to stumble across

hmm my 1 complaint atm is
ClientServerWorldExtensions
IsThinClient()
IsClient()
IsServer()
are part of Netcode not Entities, even though it only uses entities flags

so if i want to do something conditional for like a thin client in a library that works with or without netcode i either need to wrap it or manually check the flags

just going to make a copy and put it in my core library, just annoying

[strange wagon]

Hi! what is the easiest way to render 1k simple animation (just idle) meshes (800vertx) on mobile now? I found vertex animation plugin https://github.com/maxartz15/VertexAnimation but it seems outdate and maybe doesn't support mobile cause have a lot of error inside shader graph

[whole gyro]

Depends what you mean by easiest. But for the best performance, I think vertex animation textures (like the library you linked) are the way to go if just want basic idle animation playback. There are quite a few different libraries out there that use vertex animation textures and I'm not sure if any of them are actively maintained. So you'll probably need to update one of them yourself. There might also be some assets on the store (like this: https://assetstore.unity.com/packages/tools/animation/flipnote-texture-based-skeletal-animation-130904). I haven't tested any of these libraries or assets myself recently, but I did experiment with vertex animation textures with ECS a couple of years ago and got something working.

[gentle gyro]

Have you been able to build DOTS to mobile?

[wraith hinge]

builds in general are in bad shape, as we've said. however, david it occurred to me that i wonder if mono works on android with stripping disabled, since a lot of the issues are related to il2cpp and/or stripping

[devout prairie]

Ohh, nice

[devout prairie]

Are you meshing it?

[robust scaffold]

nah, just 2d integer (fixed point) texture and the lagrangian iterations acted upon it.

[robust scaffold]

I'm trying to get a hold of the source code of this: https://www.youtube.com/watch?v=_3eyPUyqluc and combine it with an existing codebase this: https://github.com/diwi/PixelFlow and do so in a massive scale (8000x8000 grid at most) in a multiplayer deterministic environment.

[devout prairie]

ah that should be cool.. i'm curious why you're creating a massive fluid sim..

have you looked at piping the point data into vfx graph?

[robust scaffold]

Just for fun. And to see if I can.

[robust scaffold]

Ideally, this fluid sim will be overlaid on a 2D tilemap with physics based static walls and, a very far stretch goal, dynamic rigidbodies.

[devout prairie]

sounds like a fun project 🙂

[robust scaffold]

hrm, i think maybe with burst and a distributed update sequence, I might be able to do the fluid sim using deterministic floats on the CPU...

Just update 32x32 or however many I can get away with chunks per frame, distribute the workload of updating the entire map's fluid data over the course of hundreds of frames. Maintaining 60fps is key, hrm. But first, I need to see if my lighting system explodes when updating to 1.0

[devout prairie]

do you have it actually up and running on the gpu so far?

[robust scaffold]

I have a form of copy pasted version, very old project of mine. https://github.com/diwi/PixelFlow/blob/master/src/com/thomasdiewald/pixelflow/java/fluid/DwFluid2D.java

Not that repo of course

but a copy of it ported to hlsl in unity

It was for an old version of unity and was pretty clunky, back when I was dipping my toes into compute shaders, but it worked back then. Dont know where my screenshots went. The fundamentals aren't hard. Just requires massive numbers of iterations to converge on a usable approximation. 40 for pressure projection and 20 for each diffuse property (60 in total for velocity, temperature, and density). So 100 iterations in total. Very expensive. That youtube link promises to cut down on these jacobi iterations significantly with a separable filter approximation but the source code is locked behind Ubisoft corporate copyright. Which is pain.

[devout prairie]

ahhh.. do they talk about how they achieved the reduction beyond just saying that no?

not that i know much about it tbh, but it's something i was always interested in trying out

that and meshing an actual fluid

[robust scaffold]

They have research papers and pseudo code but not enough to recreate it. Especially since they only describe the mathematical derivation of the poisson filter construction (which apparently requires use of neural networks?) but not the finalized per iteration mathematical 2d filter values.

https://dl.acm.org/doi/10.1145/3528233.3530737

[robust scaffold]

Yea, this is just numerical transport with color coded value ranges. Not actual meshed fluid interactions.

[devout prairie]

that's some fairly heavy duty maths stuff 😛

[robust scaffold]

Derivation of all this stuff is pretty crunchy but the utilization is simple enough. A set of constant filters to sample and sum together (I think) multiple times in expanding concentric rings for a set distance.

It's like poisson blurs, the faster alternative estimation of a gaussian blur.

[devout prairie]

just skimming through the paper, the math is way over my head but they do mention machine learning as a possible avenue to help create the depth field ( section 4 ) where basically the simulation is 1d and they fake the depth to avoid doing multiple sims on different dimensions

[robust scaffold]

machine learning is the newest academic buzzword for magic

[devout prairie]

haha yeah

it's interesting though i wonder if a neural network could be trained on fluid data and then used to simulate it faster 😐

[robust scaffold]

https://dl.acm.org/action/downloadSupplement?doi=10.1145%2F3528233.3530737&file=supplementary.pdf

Thats the key formulas for reproducing but it lacks the application and starting points. Ugh, they were suppose to publish a sample project using it but last I heard from the pipeline a month or so back was that it was locked in paperwork for unclassification.

What

IJC overthrows its successor

They even purged the section relating to IJC in the documentation. What is going on?

[rotund token]

full circle

IJobChunk is back (documents not really updated)

[proud jackal]

https://tenor.com/view/the-circle-of-life-gif-9789392

[robust scaffold]

Yea. We were really spoiled with 0.50 release with accurate documentation on release. Not comprehensive but at least largely accurate. Ugh. Gotta shift gears back to the old unity ecs trial and error days.

[rotund token]

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

just read this documentation instead

[robust scaffold]

Nice, good find. Thanks. I just managed to scrape together enough free time to dive into 1.0 right now so this is all completely new for me

[proud jackal]

Yeah, sorry about that one! deadlines are slightly skewed for docs vs features, so we don't have to e.g. write docs for the old thing then while next weeks task is called remove the thing and make docs to that now instead, tho any question can be asked here or on the forums and we do try to answer them! :3

[robust scaffold]

Is there a version that includes IJobFilter? I dont think it changed (outside of a rename) in 1.0 but it's odd that DOTS makes no mention of it.

[rotund token]

works fine

they just never mention the specialty jobs

[robust scaffold]

Isnt IJF the only specialty job now that the NMHM variants got purged?

[rotund token]

IJobFilter
IJobParallelForBatch
IJobParallelForDefer
IJobParallelForTransform

[robust scaffold]

Oh right. Huh. I even use the defer version. Transform hasnt changed right? I need to find their implementation of GO-Entity syncronization and reverse it with the new transform changes that messed my implementation up.

[rotund token]

also 6 in physics

[robust scaffold]

Oh right, the physics jobs. I wish tiny got revived with 1.0. All their 2d stuff made my life easier

They didnt expose the companion component type addition (still internal). I'm concerned about the radio silence surrounding hybrid dots. But that is just unity in the end.

And with the runtime conversion being deprecated, I need to find a way to add monobehaviors to the engine physics loop somehow.

ugh, i would've sworn that DOTS Netcode guy said that runtime conversion wasn't getting chopped. Well that was a lie.

[wraith hinge]

usually when we say things that are wrong they were true at the time

we're trying to get better about not saying things that might be false in the future, though

[robust scaffold]

Yea, it's just me venting.. I had a lot riding on that and now I need to figure out a hybrid workaround.

[rotund token]

to be fair to the netcode guys

At the moment are speaking, ConvertToClientServerEntity it is still present in 1.0. A final decision is still in progress for what I know.
I will check fore more up-to-date status, but it is safe to assume it will be still there.

is what i suspect you're thinking of

and then followed up with

The component has been deprecated. A new baking flow for the ConvertToClientServerEntity will be provided for the final 1.0 release but there is not a replacement for it right now in 1.0 preview.

[robust scaffold]

My fault with bad reading comprehension. Thats what I am also remembering now. I just took their statement but ran with it.

[wraith hinge]

in general, my feeling has been that we were correct to focus on a) deleting most of the things that we knew were permanent nightmares in one way or another (with the exception of entities.foreach), and b) fixing all the things that users absolutely positively could never fix under any circumstances (e.g. job system overhead, isystem not working properly, most jobs not being schedulable from burst, etc)

correct me, but so far my observation has been that users can staple their way around hybrid interop, it's just pretty annoying. and that is much easier to fix in a patch release than e.g. conversion just permanently being unreliable

[rotund token]

personally i find hybrid fine, it took like an hour and a single system/component to solve all my issues

and i can have gameobjects + components referenced/managed from subscenes

[robust scaffold]

For what I call "static" monobehaviors. Basically everything except physics and I got them working with an assembly exposed baking system. However I'm tied to the engine's box2d physics which requires interplay and recognition between the monobehavior and engine. Baking prevents the Physics2D from recognizing MB scripts and components as part of the physics world.

[rotund token]

it'd be better if an official solution existed

but it's hardly a blocking issue

[robust scaffold]

Well, not baking. The 0.5X GOCSs. Havent tried with the 1.0 baker yet but I doubt its different

The alternative of course is inversion of the entities - GO data structure. Which I think is the way forward.

[wraith hinge]

2d is indeed a bit behind the times. that said, i'm surprised a 2d game has sufficient scaling issues to create interest in dots

[robust scaffold]

I like DOTS and the linear data structure makes CPU-GPU-CPU communication a lot more streamlined.

[wraith hinge]

true, it is nice for that

[balmy thistle]

Oh not to brag but I've made 2d Unity games with performance problems

[robust scaffold]

I've tried to make my own physics engine, which was a great learning experience into really in depth data layout burst optimization, but what it really made me appreciate was how robust unity's physics engine was. Sure, it's not the fastest but it is rock solid and robust.

Mine had bodies overflowing velocity just trying to calculate friction vectors. I know there was some math issue somewhere inside the polygon body but in the end i just went back to the built in version.

[balmy thistle]

Friction, really? Nice. What friction model were you using?

[robust scaffold]

Just basic single scalar friction on the contact point application. Something was borked with the tangent calculation. I had everything hyperoptimized using vector4 wide math operations and debugging why the orthogonal vector was not normalized was just too much work in the end

Everything else worked, kinda. Multiple manifold / contact bodies didnt seem to resolve in a logical manner, how I was doing it at least assuming every pair contact was independent of another.

Theoredically that shouldve worked but what it turned out to be was a lot of jittering. Might have been an issue with resolving cases of multiple contact points per pair contact independently in parallel resulted in extreme values.

It was just basic box 2d but vectorized. Fun project overall

Huh, does anyone know what happened to the dots tab? Im trying to enable live conversion

Hrm, GO positions no longer reset when play is cycled?

Or is that a side effect of domain reload disabled

[rotund token]

ok so the settings are all in preferences -> entities

however it seems that the live baking option in 1.0 was hidden by accident

i believe it's on by default but if it was off when you upgraded it will be off and hidden

you can go into the help -> about

type internal

and it will enable the option in the preferences -> entities tab

[robust scaffold]

huh, that is weird. I dont trust Unity's upgrader for major versions so this is a manually ported project and that option is also hidden

[robust scaffold]

Yea, that worked. Any way to disable the internal again?

Or will a restart do it

[rotund token]

type it again

pretty sure it stays on between sessions (it just sets an editor pref)

[robust scaffold]

ugh, yea

[robust scaffold]

Neat. The boxes around the component types are missing though. May be a broken UI which was why it was hidden

I wish all the components were listed, one component per row, instead of compressed together. But that was how it was since live conversion was a thing so eh.

[signal phoenix]

I've often thought the same. For me at least the readability would be better

[robust scaffold]

I guess, is it really needed anyways? With the inspector view, this tiny preview is largely redundant.

[signal phoenix]

that's true, hadn't even thought of that

[robust scaffold]

I am liking this burstable builder API. Makes query construction a lot more compact. Real nice.

[west lichen]

If I have a bunch of jobs that are going to all perform actions on a single float value, is it safe to parallelize it? Like a float like position. If I want 50 jobs to all add X amount to the position, is it safe to do something like:

public void Execute(int i)
    {
        float blah = obj.x;
        blah += something[i];
        blah -= something[i];
        blah += something[i];
        blah *= something[i];
        obj.x = blah
    }

If the operation isn't instant, say it takes several lines of code. Could the jobs potentially step on eachother's toes?

[robust scaffold]

50 to 1 can not be parallelized. That's a classic race condition situation. You must either do it in sequence (singlethreaded) or use interlocked commands.

[west lichen]

ah, duh. Thanks

[late mural]

wondering do we have any dates as to when unity physics will support the new transform system?

[robust scaffold]

Soooooooon^tm

[late mural]

ah, welp atleast soon is better than never!

[robust scaffold]

Welcome to DOTS where waiting is the only constant around here.

[late mural]

lol

[robust scaffold]

And no optional components in IJE, shame. I have a perfect use for it, if only there was optional components...

[late mural]

i dont even know what IJE is, but sounds cool!

[robust scaffold]

IJobEntity

[late mural]

oh ok cool!

[robust scaffold]

@rotund token Do you know why Unity made the transform system AoS instead of the old SoA? Seems like jobs wanting just the position will waste a lot of cache line bringing in scale and rotation in.

I know there's the option to go back but it's getting deprecated.

[rotund token]

think they wrote up a post about it somewhere