@minor sluice it's like they are TRYING to mess this up
#archived-dots
trail burrow
dusky wind
But it's not that far off from just installing packages via git url
trail burrow
they keep going from bad to worse
dusky wind
Their solution for listing it though seems headass as fuck
And makes discovery impossible
minor sluice
until yesterday when timboc told me, I had no idea that the full package name is a valid git url you can add
because git url isn't indicative that it means this notation (too) imo
trail burrow
@solid flume also, for completenss sake... calling the high bit a "sign" bit is... technically wrong
minor sluice
scoped registries?
dusky wind
So this isn't too big of a departure for me
trail burrow
if you really want to get to splitting hairs and being technically correct level, yes the highest bit being set to 1 in signed numbers means the number is negative... but its not a sign bit... it's actually needed for calculations for sub/add/etc.
solid flume
Yeah true
dusky wind
Errr, you can add your own custom UPM packages if you set up your own NPM registries
I've been using it for some of my own shared libraries
trail burrow
@solid flume compared to floats, which have a proper sign bit - it's evne called that
dusky wind
Before that I just either git submoduled it
solid flume
I don't know floats
trail burrow
@solid flume floats is just base-2 scientific notation, with a sign bit in the front
dusky wind
Or subsequently used git urls in manifest.json
solid flume
Ah
dusky wind
I forget if it allows higher exponents or not
solid flume
Does Physics.OverlapSphere or something exist for ECS?
dusky wind
I haven't checked the standard in a while
trail burrow
@solid flume it does 11 bits
solid flume
So I can check vision radius
dusky wind
@solid flume nope, overlap queries only support AABB testing
I would wager a zero distance collider cast would work though
solid flume
I'm questioning why I'm even trying to use ECS for abm but it's not like I have anything better to do
dusky wind
It seems to work in my use case
@solid flume vision radius might be easier with CalculateDistsnces
trail burrow
checking radius is easy
solid flume
No idea what that is
dusky wind
@solid flume provided a collider geometry
It finds the closest point on nearby colliders
minor sluice
so the scoped registries are essentially the qualified names like com.unity.dots.... etc?
solid flume
I don't see how I would use that. I give it a sphere collider, it finds the closest point on every nearby collider?
dusky wind
@minor sluice for example, my GGPO port has the qualified name of "com.houraiteahouse.backroll"
It's hosted on https://upm.houraiteahouse.net
And my manifest.json says to fetch all packages starting with "com.houraiteahouse" from that registry
@solid flume yep
solid flume
But how do I find If there exists a plant in my vision radius?
dusky wind
You then filter it down to those that are under a maximum distance
The max vision radius you set
And that's the set of visible objects
Since we are operating in euclidian space, that distance automatically forms a sphere
solid flume
I'll have to try this and see
But thanks
I can do stuff like use a script to generate a bunch of cubes with the convert to entity component, and then run my game right? That'll work?
solid flume
Nah, this isn't a game just a simulation I'm doing for fun, so I can (probably) avoid (most of) the fancy stuff
dusky wind
Make sure to update the Translation component so they're not all stacked up on each other
solid flume
Or just don't and watch physics blow it apart?
dusky wind
If you are using Physics, sure
minor sluice
ah I see, thx,
at the moment, for my own projects at least, I'm more a fan of using git submodules than upm packages (maybe also because I haven't given the later an earnest try yet)
dusky wind
@minor sluice do actually try it, it's a lot less of a headache when working with people unfamiliar with source control
It also makes dependency management a much simpler problem
minor sluice
can imagine that, luckily (for this specific case at least) I work alone on my current projects. My main reason for the submodules was that I could edit the module in any project that uses it and push the changes then. Not sure if or how easy that works with packages
dusky wind
You generally need to republish it
But the intention is to have more than just you working with it remotely
minor sluice
but you as owner of the repository can treat the repo the package comes from as just any other git repo right? so that would mean you could have multiple local repositories of the package for multiple projects on your disk and just republish the package from whatever project you currently work in?
I think that would be an approach I'd like
dusky wind
Not in the same way submodules work
You will need to separately make changes to the repo, publish the package and update your package manifest.
It's not ideal for a single developer
But for a community or a large team, this separation is critical
minor sluice
ah, alright!
don't have much experience with working in large teams or for communities
trail burrow
https://forum.unity.com/threads/dots-sentiment.956936/#post-6238085 someone laying it out
amber flicker
haha wasn't that you who posted that @trail burrow ?
amber flicker
Had a lot of quotes similar to you that's why I assumed ๐ (e.g. https://discordapp.com/channels/489222168727519232/497874303463850004/747139543567958137)
trail burrow
@amber flicker yeah that quote has been circulated in a lot of places now apparently
someone sent a pic of it to me in PM, etc lol
trail burrow
ditch ecs, keep jobs/burst, switch to .net 5, improve GO/MB workflow, etc.
stop trying to re-write the entire engine sigh
frosty siren
Have we today built-in system for sync data between LocalToWorld DOTS component and Transform mono component?
minor sluice
wow, that thread gained quite some traction
rancid geode
well, HPC# allows more than just burst itself, it also allows then to do their safety checks to allow multi-threading in a safe way (the one thing that no game engine solved so far AFAIK)
trail burrow
@rancid geode no HPC# is only for burst compatibilty
HPC# is just crock of shit
it's just that they cant burst compile more than that
HPC# doesn't "allow" anything, it just allows burst to work
by restricting the feature set of the language
rancid geode
personally, I really like ECS-way of doing things (not the half-baked current hybrid workflow, but the way to structure the code ifself), so I prefer to wait until 1.0 to made my judgement about if it was worth their investment or not
trail burrow
to basically C
.NET 5 JIT performs basically as well as burst, and doesn't give up 85% of the language features...
rancid geode
HPC# doesn't "allow" anything, it just allows burst to work
@trail burrow if you dont do things the "Burstable" way, then the safety checks won't work and they can't prevent race conditions
they are actually linked
safe lintel
Given he paraphrases you without attribution I kinda wonder if he's just spewing other peoples gripes without any real experience in using it for himself ๐
rancid geode
it is just that dealing with classes in c# is a mess without having an option to manually control the memory the way we can in C++
solid flume
How do I store a target of something I want to move to?
trail burrow
@rancid geode HPC# exists for burst, no other reason. You can't burst code that isn't HPC#. The safety check system has nothing to do with HPC#, it could've been built without HPC# subset.
solid flume
In a component
rancid geode
so they need to rely on a constrained version of the language for that (while still allowing entire C# stuff out of Burst context)
trail burrow
@safe lintel I can assure you he doesn't, i know the guy, he knows what he's talking about
rancid geode
@rancid geode HPC# exists for burst, no other reason. You can't burst code that isn't HPC#. The safety check system has nothing to do with HPC#, it could've been built without HPC# subset.
@trail burrow I don't see how they could guarantee the thread-safety without the HPC#
solid flume
Like if I want to move to something, I check if my target isn't null and move toward it's transform.position
How does that work out in ECS language?
rancid geode
How do I store a target of something I want to move to?
@solid flume you can expose an Entity field and use GetComponent<Translation>(thatEntity)
safe lintel
@trail burrow thats not vblanco is it?
trail burrow
no
solid flume
I guess I could just add a bool to see if the target is valid and set that to false instead of nulling
rancid geode
Can I do null checks on it?
@solid flume if (thatEntity != Entity.Null)
violet cosmos
I really need to write on that thread... But, I'm tempted to just be lazy and say "DOTS is driving me to drink" ๐
solid flume
Oh that's nice
deft stump
I'm still wondering when will they support .NET 5, I know it won't be on release...
minor sluice
if your target is another entity and you have it stored in a component, I think the equivalent of checking if that entity target exist is by using EntityManager.Exists or if you pass a ComponentDataFromEntity group to a job, you also have an ComponentDataFromEntity.Exists method
trail burrow
@rancid geode the safety system can be in place without HPC#, you can still fuck it up despite their 'safety' system even now... all the safety system does is check for read/write conflicts... that system can be in place no matter HPC# or not.
minor sluice
for example, the ComponentDataFromEntity<Translation> group and you check if a the target exists with a translation component with doing the .Exists(targetEntity) method
rancid geode
@trail burrow by the way, Joachim (I believe) replied once about why the don't have ref return on the data access arrays of chunks, and he said that it was because they didn't have how to do their safety checks with a reference value
trail burrow
okey?
minor sluice
entity is a struct, did they include a null check? otherwise you probably have to check for if (entity == default) (for structs),
but it is better if you check if the entity actually exists in your world with entityManager.Exists
rancid geode
if you have reference values (class) you can't guarantee that it won't be modified in another place and create race conditions
minor sluice
oh but also, didn't know about Entity.Null (so please ignore what I said about default)
trail burrow
@rancid geode no sure.. but you can already do that with the "safety system" today in place, you can fuck up the thread safety and have it blow up despite all of their safety checks
it's very easy
so despite HPC#/Safety/Burst, you can still blow your leg off easy
rancid geode
well, yeah, but with the safety system you at least know what and how to fix
trail burrow
So whats the argument - that you COULD do something wrong that wasnt safe if the restriction wasn't there?
No...
rancid geode
multi-threading is easy this way
in any case, HPC# / Burst / Jobs are all opt-in features
trail burrow
then you're not doing complex enough multi-threading
rancid geode
you can still do everything single-threaded with C# and/or multi-threaded with C# Tasks
trail burrow
if multi-threading means: batch processing of arrays of data, sure it stops the absolute noobs from fucking stuff up
that's kind of a moot point tho
rancid geode
you just have now an option to do thread-safe code without much more headache than limit yourself to a subset of the language
deft stump
multi-threaded with C# Tasks
that kinda defeats the purpose of writing safe multi threaded code tho
trail burrow
problem with using regular C# threads in unity is that the mono threading primitives are shit
and are basically unusable
for high frequency or low timing scenarios
rancid geode
@deft stump which brings back why HPC# and the Safety System are useful for us
solid flume
Can I find all entities inside a certain radius?
I can't seem to figure out how spherecasts work
rancid geode
if multi-threading means: batch processing of arrays of data, sure it stops the absolute noobs from fucking stuff up
@trail burrow well, yeah, but if you combine that with the ECS structure it becomes way damn powerful
solid flume
In ECS
rancid geode
which is way I am giving a shot on Unity's ECS and hoping that they will do that right
the current issue with Unity's ECS is not even the API itself, but the hybrid workflow that just doesn't work properly/the way we expected
trail burrow
@rancid geode batch processing of data is usually not needed when dealing with game logic outside of a few game types, actually the architecture they have picked hurts performance when you're implementing fine grained game logic which lots of conditions and checks and combinations of various operations
rancid geode
but it is not production-ready yet, so this can be something that is being worked on (but we don't now because they aren't clear about their roadmap)
deft stump
it was suppoed to be released in 2020.1
rancid geode
yes, but we are all developers and we know that timeline changes
trail burrow
the stuff you need to build an actual game
meanwhile people that need to build actual games on unity are left by the wayside, working with tech from 2016, because nothing is being done to fix old problems, bugs, etc.
rancid geode
this is a big change, I prefer it to take longer and right, and be released earlier and be useless
trail burrow
@rancid geode so far it's taking longer and wrong
rancid geode
don't see why wrong ๐ค but this could be personal taste
trail burrow
i mean they had 3 years so far, and it's still basically as useless as it was 3 years ago
deft stump
yes, but we are all developers and we know that timeline changes
@rancid geode true. however, they're a tool maker.
if a tool maker takes so long to make the tools, then it's better to just move on.
pliant pike
Unity better not give up on DOTS ๐
deft stump
we've got game schedules to keep
trail burrow
i was, for the record, super excited about dots when it was announced
minor sluice
it would be kinda neat if unity focused on a system to better distribute the load of regular threads, based on the end system the app runs on, if that was possible
pliant pike
we wouldn't have half the tools we do have if people gave up half way through making them just because they took to long
trail burrow
@pliant pike yeah sure if it was developed in a vacum that'd work
rancid geode
@deft stump I must disagree on that, they just showcased too early, most tool makers doesn't showcase their new tools until it is almost finished, even if it took 3-5 years to reach that state
tiny ore
The point is:
- a few experienced guys are feeling (at this point) that there are fundamental problems with Unity's specific approach...:)
- so the whole path is wrong (it's not only timing)...
low tangle
It's not that it was taking to long that I have the issue with, it's that they didn't tell us anything
rancid geode
the issue is that ECS shouldn't be showcased before late 2019 (which is when the API got barely usable)
low tangle
They just told us to wait till next unite for the next big update to entities or physics being released
rancid geode
we've got game schedules to keep
@deft stump for that reason they advertise that it is not production-ready
low tangle
Yeah, that was a recent change
minor sluice
I really like that they offer dots for us, even if I only have little experience with it,
think the main issue is transparency, communication and probably documentation too
trail burrow
@rancid geode and that would be FINE if they hadn't left the rest of the engine to die by the wayside lol.
while they working on their ivory tower dots thing
which is still, useless
low tangle
They didn't throw it in your face as much before, and they gave the impression it was only going to take one to two years for most of it to be ready for games
minor sluice
like james brought up really well, most of the time users go on forums or on here to find out about stuff that is sometimes basic but just not well documented and the ecs sample package isn't that accessible for beginners either
low tangle
They were showing off full demos (nordus RTS demo) in 2018
tiny ore
While that happens, we can live with 2015-ish renderer and (contested) animation tech for production-ready MB.
pliant pike
yeah they do need to properly document it
trail burrow
@minor sluice it's not even that, it's that it's buggy and incomplete. You can't actually build anything serious with it. I don't need a documentation or manual to do stuff unless it's really complex stuff... yeah sure docs suck, but you cant document shit that doesn't exist yet or is fully broken.
rancid geode
@rancid geode and that would be FINE if they hadn't left the rest of the engine to die by the wayside lol.
@trail burrow not sure why you feel that way, sure they haven't added anything BIG to GO world in the last years (beside some cool acquisitions like Cinemachine, ProBuilder and the new Bolt)
but we did have some cool progress like Addressables and InputSystem
low tangle
The documentation is the source code
pliant pike
only neo can read the source code
trail burrow
@rancid geode those are all new tools... and they both are in 'preview' stage... im talking about core bugs in the existing physics/animation/rendering/audio/etc. stuff, which goes unfixed for YEARS while they fiddle on dots
rancid geode
we could have more if there wasn't any DOTS to care about? maybe, or maybe the team size would be smaller because there won't be any DOTS stuff
trail burrow
things that actually block progress for real products
low tangle
There's a lot of standard comment (not XML autodoc) at the header of most files I've looked at in entities
minor sluice
I guess the few people that actually have a dots project in production spent tons of hours on the forums, and rolling a lot of custom systems that users would expect that unity offers itself?
for the documentation part, I think that is generally an issue with the new packages. I really liked that they pointed out that they work on documentation for how to convert shaders from standard to srp pipelines for example but... at the moment it's really pretty bad
trail burrow
@rancid geode oh and ProBuilder/Bolt? let's see, pro builder integration is still shit and its buggy af. Bolt 2.0 that was worked on for ages was scrapped and ditched, and they went back to Bolt 1.0 (unitys decision)
I'll give you cine machine is decent
adressables are still buggy, at least last time we tried to use them
minor sluice
another issue I see is that they market things as out of preview at times when a lot of users still seem to have problems with those packages that would indicate that they are still in alpha or beta state.
trail burrow
i dont know what kinds of companies you work for, but i prefer solid tools that do what says on the tin so we can schedule time and count on them, not have shit change under our feets 3 times per year.
low tangle
Cinemachine, ProBuilder and the new Bolt)
but we did have some cool progress like Addressables and InputSystem
Being honest, why were these not built in house, the engine feels like it's made of duct tape.
trail burrow
or have to drag 5-6 year old bugs around
minor sluice
I don't use bolt but some people said bolt 2 would have really been great and one even said that could have been a decent competitor solution to unreal's blueprints even
trail burrow
@minor sluice yeah i dont use it ither but they ditched all of it
lol
and if youre just a hobby user fiddling at home with unity personal edition, what you think of this... doesn't really matter, because you can take as long as you want, you have no budget, you can play with shit beause it's fun, etc.
but the people that have companies, staff and money on the line
need dependable shit
not ivory tower dreams
minor sluice
yeah, would prefer more stable features too, so I'm glad that they stated they want to work on stability and workflow improvement,
I still use all the old things, standard renderpipleine, monobehaviours etc. (and had few to no issues with them)
just that I played a few times with jobs/ecs
deft stump
i dont know what kinds of companies you work for, but i prefer solid tools that do what says on the tin so we can schedule time and count on them, not have shit change under our feets 3 times per year.
totally agree on this.
Like I said before, Unity is a tool maker. We can't have tools that doesn't work out of the box.
they break dev scheduling
trail burrow
@minor sluice same, we use: Old Input, Old Renderer, Go/Mb system, "Old" UI system (not ui elements), etc.
minor sluice
or rather, there are always small issues somewhere, but they never felt like a pain point really
trail burrow
and they are still plauged by the bugs that were during during 5.x release cycle
because LITERALLY everything else they have made since 2016
is USELESS for a real project
because it's unstable, incompatible, incomplete
deft stump
Or atleast, I just want Unity to be honest with themselves.
Just make a giant public announcement and say "we fucked up, here's the ACTUAL state of things"
and my faith in them might get restored.
trail burrow
yeah i have zero faith anymore
low tangle
Honestly, with the whole hpc# and .net 5 stuff, I'm actually just heading back to cpp myself @trail burrow
trail burrow
@low tangle i have so much of my company/career tied to unity, i would jump ship if i could
in a fucking split second
minor sluice
the idea and the direction of the new features is really impressive I think,
dots, and also the problems that the new input solution would solve for example,
but I agree with your statement on production ready projects.
for inputs, I'll just use ReWired and let them handle whatever abstraction or interfacing they choose with the unity input
low tangle
Yep
trail burrow
and yeah ReWired is good
minor sluice
I haven't checked it for a few months, but they should still be on the old input system too
trail burrow
solid asset, been around for ages
minor sluice
ReWired seems like the goto input solution many unity studios choose for their games
trail burrow
yeah for good reason
low tangle
My prototype core in unity is using the native plugin interface right now, in cpp, and I honestly forgot why I hated it before. My 2yo unity game takes a long ass time to compile, and 20 min with il2cpp, which generates a gig of source
trail burrow
Yeah we just use unity for rendering, rest of game is pure standalone C#
low tangle
I can't stand il2cpp, really wish direct cpp was a option
trail burrow
I just really dont want to go back to C++ lol
I'm so happy in C# land
espc with .NET 5
low tangle
I am, but also tired of not being able to directly compile. .net native stuff has been killed off by Microsoft so many times
rancid geode
@rancid geode oh and ProBuilder/Bolt? let's see, pro builder integration is still shit and its buggy af. Bolt 2.0 that was worked on for ages was scrapped and ditched, and they went back to Bolt 1.0 (unitys decision)
@trail burrow Bolt 1 is fully functional as it was, also ProBuilder, it just didn't improved much since the acquisition, but I understand you as I was furious when they acquired Anima2D just to abandon it
trail burrow
@rancid geode I don't use any of those tools. I just know that when unity acquires something = over for it.
low tangle
Esp when you had just bought it full price recently
trail burrow
And bolt2 was VASTLY better than bolt1 from what i heard from ppl that actually use it
rancid geode
I tried bolt 2 (when was still with ludiq, right before the acquisition) and it was barely functional
tiny ore
What is the reason for abandoning Bolt2? Probably have them work on a DOTS 2.5 version that will never see light of day?
rancid geode
was at least 1-2 years away from being usable
trail burrow
@tiny ore they are 'consolidating' all their visual scripting tools, that is the official reasoning at least
rancid geode
but I think we are going a little offtopic here haha
tiny ore
I don't remember that they had an official visual scripting before
I mean, people use either Bolt or Playmaker
tiny ore
"consolidating" sounds like merging an internal team, probably
rancid geode
they are 'consolidating' all their visual scripting tools, that is the official reasoning at least
(this is actually a little funny, because they don't have a consolidated vs tool yet, but want to consolidated them all together already)
deft stump
god just reading through the thread and their response just makes me hurl
tiny ore
I read a bit, yeah, my expectations for the "consolidated" version are not high
low tangle
minor sluice
oh no, but unity recently acquired plasticSCM 
amber flicker
^consolidated
trail burrow
@minor sluice one of my fav products about to get 'reckt by unity
absolutely love plastic
low tangle
God are they becoming Google
minor sluice
someone told me it was perfectly fine to be used for unity before the acquisition too,
I speculate on some collab 2 rebrand or something too
north bay
Ye plastic is awesome, I'm a bit worried ๐ฆ
solid flume
[BurstCompile]
public struct RaycastJob : IJobParallelFor
{
[ReadOnly] public CollisionWorld world;
[ReadOnly] public NativeArray<RaycastInput> inputs;
public NativeArray<RaycastHit> results;
public unsafe void Execute(int index)
{
RaycastHit hit;
world.CastRay(inputs[index], out hit);
results[index] = hit;
}
}
``` What's the use of unsafe here?Got this directly from Unity.Physics documentation
Also, can I schedule a job (The above one) inside an entities.foreach?
trail burrow
@solid flume might be a mistake
low tangle
When you schedule it, results needs to be marked as disable
solid flume
@low tangle marked as disable?
low tangle
Yeah the uh, disable parallel for restriction
solid flume
Basically, I want to spherecast around entities that have a particular value of a specific component and then do something with the result
low tangle
The job is picking up on the inputs being safe, where only one touches one
solid flume
Ah, that might be why the unsafe was there
As an alternative to disable parallel for
low tangle
But it doesn't know that on the read write to the results
No
Unsafe is there because the physics code tends to use pointers often
Pointers to colliders and such
tiny ore
Do they change the semantics of what c# unsafe means?
low tangle
No
tiny ore
Because you only ened to mark the method as unsafe is there is pointer logic DIRECTLY in that method/class/struct
solid flume
Also, I realised spherecast will only hit one thing. I want to implement vision radius. Any way I can get entities within a radius?
outer swift
Anyone seen this funny performance in float4x4.TRS before?
tiny ore
If a method you call uses it, the calling one does not need to be marked as unssafe
low tangle
It just allows pointers, the person who wrote that likely copied or modified from another job that was using pointers
outer swift
7.72ms spent in float3x3 constructor looks wrong to me
tiny ore
It's a deep profiler... could be a slow machine (331 calls to the constructor)
outer swift
@tiny ore Nah, the reason I started profiling in the first place is because I was seeing slowdowns
So the performance is bad independently of the deep profiler
low tangle
It's also a fat struct to copy
outer swift
Plus it's running on a Ryzen 2600, so not that slow
low tangle
Is that in burst code or main thread
low tangle
Hm, it's only the constructors and matrix multiplys
outer swift
The weird thing is that all the other operations look fast, just the 3x3 construction is dog slow
low tangle
It's a 4x4 by the way
outer swift
I know, but the slow part appears to be a 3x3 ctor?
low tangle
Yeah
tiny ore
Testing here our own TRS
low tangle
If you expand the hiearchy a bit what's the rest of the call stack look like
amber flicker
have you checked in the timeline? it's not hiding a wait or smth?
tiny ore
Well
low tangle
Throw it into a job and burst it, then complete it
tiny ore
Deep Profiler in Editor/mono is shit
tiny ore
Just try a proper burst + il2cpp build
low tangle
Tossing stuff into a single job is a nice way to test if burst is going to help the code much
Has a lot less restrictions due to only bring one thread
outer swift
Thanks for the advice!
dusky wind
@outer swift IIRC TRS involves constructing 3 4x4s and multiplies them together
A good portion of the multiplies isn't vectorized due to their memory layout
If you know one or two of the components it might be easier to hand craft the matrix yourself
outer swift
Duly noted.
amber flicker
low tangle
Unity intends to list its common stock on the New York Stock Exchange under the ticker symbol โUโ.
wew
safe lintel
kind of a weird time to do it? where a lot of development on it seems in a weird state of flux with a somewhat grumbly userbase?
amber flicker
well.. someone else pointed out... "As the only game engine available on Apple products" ๐คฃ
safe lintel
frytakemymoney.gif
safe lintel
would be kinda hilarious(and tragic) if epic lost their lawsuit against apple and were barred from it completely
amber flicker
it's been in the works for a long time - probably were going to do it just as covid happened.. might have decided it's worse to keep putting it off and now with Unreal maybe it's as good a time as any... 100% speculation ๐
safe lintel
yeah true, just how the timing worked out doesnt seem to be on their side.
manic aurora
yep afaik itโs been planned, pretty sure I saw talk about it a while back
amber flicker
this is a slightly dumb question as I've been using lambdas & IJobChunks almost exclusively. Say I have a e.g. NativeArray<Entity> myEntities and I want to iterate over all of them and set their translations. What's the "best" way to do this? The two methods that spring to mind are either e.g. for() { EntityManager.SetComponentData} etc or for(Entity e, Translation t) { if(myEntities.contains(entity)) then... )} - neither of these feel very optimal? Feels like I'm missing a third way?
low tangle
do you have a set of entites, or do you have a archtype you can iterate?
amber flicker
set of entities - their archetype isn't unique
low tangle
for { if HasComponent<T>() -> GetComponent<T>(e) }
HasComponent compares off the archtype of the entity, and is pretty dang fast
amber flicker
duh... yea that makes total sense.. thanks @low tangle - I somehow thought I had to be in a lambda to do that
low tangle
if you have a set, theres no way to filter it with a query, which sucks because you could just ask the query for a set of the components in a array, or to set all of them with a array of components<T>
amber flicker
yea
low tangle
If your not within a system base, then you just need to do what its doing for you. If within a job, get a GetComponentDataFromEntity<T>() and use the Exist function on that, or maybe it has a Has function now, forget. otherwise if your main thread, just use EntityManager.HasComponent<T>() and EntityManager.GetComponentData<T>() etc
amber flicker
yea so if I'm in SystemBase and I do for(.. myEntities.Length){ GetComponent, SetComponent} etc, do I need to do anything to make it burstable I wonder
amber flicker
but if I didn't do that, it wouldn't be right?
low tangle
it wont be bursted yeah, burst can only run on methods, like the ones within lambdas
solid flume
Any way I can get all entities in X unit radius? Maybe within a job
low tangle
how many entities
low tangle
how many subjects will you be looking at
if the set is small, say less than 10k protentional, then just burst a foreach and radius check each.
solid flume
Oh less than 10k is small I like that
low tangle
if its more than that, try the boids hashmap bucket approach and quantize each position into a grid point hash, and then brute force a smaller set
tiny ore
You are telling guys to do an O(n^2) loop on 10k entities? Or a single O(n)?
tiny ore
I'd go for the buckets approach if you need to check that from different places
If it's just a single run per update, then should be fine
low tangle
store the location your checking from and use one point within the job for all protentional, its not as slow as you think
you should always start with the naive / brute force approach. you can and will likely optimize it later if it was ever a problem
tiny ore
I mean, I few.months ago I wrote a sample for a game with 10k zombies and 6 cars
For switch (the demo)
low tangle
very nice
tiny ore
So in the end, the best solution (that scaled) was more.the inverted idea:
low tangle
hows the switch hardware feel using ecs/burst?
tiny ore
Instead.of the cars checking for the zombies closeby, I made the main zombie updates to look for the close cars
hows the switch hardware feel using ecs/burst?
Photon Quantum's own ECS
low tangle
yeah, inversion of logic is usually a good way to structure things with ECS
tiny ore
But the general principles are similar
low tangle
yeah same thing honestly, just assumed you were referring to it
tiny ore
Ye... I mean, same idea would work perfectly with unity's, that's why I'm saying
And I agree you don't need to add the whole kitchen sync (like a broadphase) for this every time
I needed to update the zombies anyway for their movement, so I scheduled the car proximity checks to happen in subsequent ticks)
low tangle
oh it for sure does, but I get the feeling crypticax hasn't learned a ton of ecs patterns yet, so I didn't want to give them a complex solution when a simple and easy to understand one will at least get them onto the next phase of their game
solid flume
I've literally just picked up ECS because I wanted to do agent based modeling and this seemed like a great application
tiny ore
In the end it's 6 X 10k checks (over the scheduling tick rate) anyway, but it felt better from the gameplay with the invertion
low tangle
its great for that yeah, anything simulation heavy really
tiny ore
In the end I did not add a broadphase (even for the cars, that would be fast to add) because it worked well already
dusky wind
Though there are some unfortunate game types that require strictly sequential processing
low tangle
yeah and there a real pain to decompose into jobs
tiny ore
I did NOT used our parallel tasks for this case, lol
I wanted to measure the limits within the main thread
solid flume
I still don't know how to do the iteration since I need all entities with a component of type X to look for the nearest component of type Y and move randomly if it can't find it
dusky wind
@solid flume you can do bucketing
Its a common parallelization technique that puts common components in small buckets based on some function
dusky wind
It's not perfect but it splits up your objects roughly by their base location
You then run the actual distance check only in the buckets
Or the bucket and all nearby buckets
tiny ore
Is that documented as a pattern in unity ECS samples?
solid flume
Still no idea how to do that. ECS is confusing
tiny ore
I mean a couple of bucketing implementations
dusky wind
Using a bucket size roughly equal to the average vision radius is probably going to be sufficient
@tiny ore unfortunately not
solid flume
Bucketing would involve some sort of hashmap from coordinate to list?
dusky wind
I only know this because I do big data analysis for my day job
low tangle
yeah its what the boids demo does, its simple in practice once you understand it
tiny ore
Ye, any voids demo must probably do that
dusky wind
Which is just DOTS but scaled up to millions of data points and thousands of machines
tiny ore
Boids
low tangle
sucks it hasn't been turned into a proper documentation anywhere
dusky wind
Actually
Is there generic interest in this kind of stuff?
I can probably write up some common techniques I have been using
low tangle
eventually ecs and dots needs some proper patterns written up to direct people to
tiny ore
Definitely
low tangle
theres a lot of useful tools/patterns out there
tiny ore
I can say that's one of the things we do more
It's to recognize patterns (including when customers ask questions), document and point
dusky wind
I've done a lot of systems programming
And think very much in terms of share nothing parallelization
solid flume
https://github.com/hecomi/UnityECSBoidsSimulation Where is the hashing in this I can't find it?
dusky wind
And pure functional programming
solid flume
Ah thanks
low tangle
theres a comment above that line I linked as well explaining whats going on
tiny ore
Yep, that does it
low tangle
thats the exact position hash at that line
yeah. the unifiy wiki has the old recomendation on it
Unity currently recommends only using the new ECS package for experimentation or if you're just starting your project and plan to release towards the end of 2019 or later. Since the package should be quite stable by then.
I've been saying two years, but I knew they said it was only going to take a year
might be able to find the form post as well
solid flume
I don't get how that hashmap works. Different boids could have the same hash?
trail burrow
not if the entity index + version is used as a key, that's a 'perfect' 8 byte hash
dusky wind
@solid flume yes
low tangle
the position is being hashed, into a axis aligned grid, of whole (integer) numbers
trail burrow
oh
dusky wind
I'm not entirely sure either but it would put multiple in the same bucket (hash)
low tangle
if a boid is 'close' to another, its position will hash to the same cell and be in the same bucket
solid flume
So how do I get entities in a bucket?
solid flume
Aren't hashmaps basically dictionaries?
low tangle
which is that line I linked
dusky wind
Probably the misleading name
solid flume
Ugh ECS ain't easy
dusky wind
Haha anything involving multiple threads isn't easy
But that what also makes it interesting/fun
solid flume
So every update loop basically I calculate this hashmap? And it's only available to this system, so anything that needs it should run in this system?
low tangle
yes, but you can do it on change and expose it
solid flume
English please
solid flume
Static?
low tangle
then other systems can grab it from that system
doesn't have to be static, as you can always World.GetExistingSystem<hashmapsystem>().TheHashMap
solid flume
Ah
low tangle
but static is fine yeah
dusky wind
This is fine assuming you don't have multiple Worlds with the sam system
low tangle
hashing every frame is fine if there isn't too many boids, I think that demo handled 50k or so? on some macbook
solid flume
Yeah I'm way under 50k
dusky wind
Yeah it's not particularly heavy of an operation
solid flume
So how do I ensure that this will run before any system that needs the hashmap?
low tangle
ideal is if you can do whatever work you need with the hashmap, within the same system thats going to use it. So if your hashing your npcs to avoid them, then doing a movement operation, you would want to place these within the same system. if that doesn't work for you then it gets a bit more complicated or slow if you want to keep it simple
amber flicker
I have a System that I want to Update manually... from an extension method... It doesn't seem to be possible as I can't e.g. get the default World from a static method so then I can't get the System to run. Is this crazy talk?
low tangle
World.DefaultGameobject blah blah doesnt work?
amber flicker
yea - Burst error BC1016: The managed function Unity.Entities.World.get_DefaultGameObjectInjectionWorld() is not supported
I guess maybe I should be using IJobFor or something
instead of trying to do this with a system
low tangle
burst compiling your static method?
amber flicker
I think it's trying - but adding [BurstDiscard] to the extension method it errors about not being able to do that as the method takes ref NativeArray<>
low tangle
yeah it needs to be a job struct if your going to compile it, otherwise you need to look at the documentation for burst function pointer stuff, I haven't had to use them yet
can't link to a section due to a lack of anchors
solid flume
What's the entity in query index?
Why is that what we add to each hash
Also, if I don't do .Schedule does the entities.foreach actually happen?
low tangle
it does not yeah, you have to .Schedule .Run or .ScheduleParalell
coarse turtle
I believe if you don't specify a run or schedule, Unity will throw a compilation error in the console stating it must end with a Run or Schedule
low tangle
have you dug though the ecs samples packages yet? and got the hang of systems?
amber flicker
So I'm trying to use a job struct but I don't know how to set a CDFE without running into the same issues :/
low tangle
you have to get CDFE's from a system
there a 'object' ||struct|| that you need to get from a entity manager
amber flicker
to get a system or an entity manager I guess it has to be passed in to the extension method?
low tangle
I think you might be approaching your problem in a way that isn't compatible with burst/jobs
can you explain a bit more what your trying to do
amber flicker
I think I want to be able to say:
public myStruct: IComponentData { int id = 100; }
myStruct.FindEntities(ref resultList);```
Then the FindEntities extension method runs the FindSystem, passing the list to be written to. The system finds all entities with that struct and returns all entities with a matching id for example.wait so
you want to be able to run a filter, for a id=100 on that struct type?
storing the result into a nativelist?
amber flicker
Essentially - but there's blobassetrefs involved etc
amber flicker
the id is a blobassetref rather than int, which contains a BlobArray<Hash128>
amber flicker
yea I'm cool with that, no worries
solid flume
So this is how I iterate over a multihashmap? foreach (int ind in buckets.GetValuesForKey(hash))
low tangle
manually yeah, but theres a job for iterating it
these are also useful jobs for filtering
solid flume
What's the filter mean? I know the normal parallelfor
low tangle
been awhile since I used it, let me find a test case for it
solid flume
Also, can I get a random value from a multihashmap?
Like, a random value from the hashset corresponding to a specific key
coarse turtle
oh the Filtering is quite nice - you can either append indices of some collection to a native list
low tangle
ah yeah, the filter job is a filter you run on the native array
struct FilterMod7Job : IJobParallelForFilter
{
public bool Execute(int index)
{
return index % 7 == 0;
}
}
struct FilterAllJob : IJobParallelForFilter
{
public bool Execute(int index)
{
return true;
}
}
or prune out elements from some collection
low tangle
yep, used to fill up a list
which you can then run a fordefered on that list, and get a index per job elem as well
solid flume
So my solution for a random value is generate a random number from 0 to number of values mapped to hash -1, go over it in a foreach and break out at that value
Horrible solution I know
low tangle
struct GetArrayValuesJobParallel : IJobParallelForDefer
{
[ReadOnly]
public NativeArray<int> array;
public void Execute(int index)
{
}
}
{
var list = new NativeList<int>(Allocator.TempJob);
list.Resize(42, NativeArrayOptions.UninitializedMemory);
var setValuesJob = new GetArrayValuesJobParallel { array = list.AsDeferredJobArray() };
var jobHandle = setValuesJob.Schedule(list, 3);
}
So. Many. Jobs.
low tangle
allowing you to run a filter on the input to a filter job, then the list your appending to, into a defered job
you're meant to chain jobs
if you want speed that is
solid flume
Actually working first, speed later is usually how I do things
First make it do something, then see why its slow and fix it
low tangle
then just do it within a systembase, use entitymanager, use .run on your foreach lambdas
everythings allowed there
.withstructurechanges on the foreach lambda if you need to entitymanager.add remove destroy within
solid flume
I don't think I'll actually need too many of the big performance things apart from the occasional bursted job, since the number of entities that actually do something will not be too many in the grand scheme of things
Also does ECS have a way of logging values to a file? Can I just use standard c# filestreams? Because that's basically a requirement in what I'm doing
low tangle
yes, just use a filestream
can't use those within burst, so put .withoutburst if its a foreach
solid flume
Aight, thanks. Can I use an Entity.Index value to get the actual entity?
Sorry for spamming questions xD
rancid geode
What do you mean with actual entity? for logging purpose?
solid flume
I have the entity index. How do I get the entity?
I've basically hashed some entities to buckets
So I get a value from a bucket
Which is the index of an entity in that bucket
rancid geode
The usual recommendation is to not store an Entity reference outside IComponentData and/or IBufferElementData
solid flume
Hmm
trail burrow
there is no actual entity
it's just a index into a lookup data structure
Entity is all there is
solid flume
Then I can get the Entity struct somehow?
solid flume
I can have a hashmap to entities?
trail burrow
yes
solid flume
Well that simplifies things
trail burrow
it sure does
hollow sorrel
@low tangle btw i tried to use JobsUtility.JobWorkerCount = 0 in build and it just crashes on startup lmao, dunno if that's because i'm doing something weird or unity, log just says invalid address
yet JobWorkerCount = 1 does work in build
so seems like even though they added 0 in 2020 it still doesn't work properly so yer not missing anything
coarse turtle
Modifying LocalToWorld.Position moves my entity right?
@solid flume No LocalToWorld.Position is a readonly property
low tangle
hollow sorrel
also fun fact my sim loop takes 0,15ms with 1 jobworkerthread, and 0,50ms with default
aaa
trail burrow
@hollow sorrel their job system is kinda meh ๐
it has so much inter-job overhead
its an absolute PITA to get consistent perf out of
tiny ore
also fun fact my sim loop takes 0,15ms with 1 jobworkerthread, and 0,50ms with default
aaa
Sure, let's take the pain out of parallelism and replace with overhead... Sorry,.could not resist
rancid geode
LocalToWorld is to store the transformation matrix for the given entity, if you have a Translation component attached too it will get overriden every frame inside the TransformSystemGroup
rancid geode
so you can either:
- Use the Translation component (recommended approach)
- Modify directly the LocalToWorld (which is a float4x4 and involves some math work)
if doing the second option, ensure to remove the Translation component, or else the LocalToWorld will get overriden again on the following frame
solid flume
So I just feed translation what I want to move by and it does it?
rancid geode
you feed it with the actual position you want it to have
solid flume
Oh
So I use the localtoworld to get the position, add whatever I want and assign it to translation.value
Man things are weird with ECS
rancid geode
example:
Entities.ForEach((ref Translation t) => t.Value += math.float3(1, 0, 0)).Schedule();
you will only use LocalToWorld directly if you are doing something very special to it or if you have a parented Entity which you want to get the actual world position (as Translation is always the "local position")
you can see LocalToWorld.Position as transform.position and Translation.Value as transform.localPosition
solid flume
I just saw the boids demo use it and I was like sure okay I guess this is how this works
Makes sense now though
So I just wrote my first system and if this works Ima be very surprised
Ayyyy it works
Not as intended, but I know why
mint iron
also fun fact my sim loop takes 0,15ms with 1 jobworkerthread, and 0,50ms with default
i pretty much just use Run() for everything, its generally faster unless you're doing 5K+ operations.
dusky wind
@mint iron does that still support Burst for enabled jobs?
I sort of wish there was an adaptive performance check
safe lintel
yeah you can burst .Run stuff, same limitations regarding no managed things
solid flume
A collider cast says it can return all intersections for a given shape
Does that mean everything inside it too, or just the stuff the shape surface actually intersects with?
tiny ore
I have not checked the source code of the dots physics package, but I assume they have both sphere-cast and sphere-overlap
the overlap test should definitely consider things "inside"
A cast I'm not sure, would have to check the code (but I have not done so)
solid flume
I actually need vision radius, so an aabb is not what I'm going for
Although I guess it could work out if I approximate a sphere as a cube
Could it work if I have a trigger on everything that needs a vision radius, and implement on enter and on exit callbacks?
tiny ore
I think we all explained you should use a spatial hash yesterday
Do not try to place this in a complete physics engine (overkill) if the only thing you need is a distance (radius) check
BTW, a distance/radius check is cheaper than the AABB
You just compare the Sqr radius with the Sqr distance
solid flume
The spatial hash isn't working
solid flume
I mean theoretically. Anything moving randomly will only end up staying inside it's own hashed area
tiny ore
That implementation is just one example, you can do however you want
solid flume
And it only detects something inside it's own hashed area
So anything even on the border doesn't work
tiny ore
That's the bread and butter of distance checks in a stateless system:
- broadphase pass: inject objects into the spatial hash data structure (whatever you chose to use)
- queries: test only stuff inserted in cells close enough to the desired origin
And it only detects something inside it's own hashed area
That'รs easy to solve
You need to also get the hashes for anything in the right/left/forward/backward (maybe also up/down) direction from the desired point, using the radius
If you make the hash cells big ENOUGH, you only need to check 2 extra ones
if the cell size is Radius * 2, you only need to check 3 cells (not 2, sorry) besides the one you're in
assuming a 2D hashing... So 4 cells in total
If your radius varies (and you do not have a reasonable maximum), you need to procedurally find all cells/hashes that your position+radius reach...
That's it
solid flume
Huh
Also, how would interactions work? Like if entity A needs to consume a resource in entity B
amber flicker
Just my 2 cents - Erick is correct... but I'd probably go with the physics approach unless performance is your concern right now - I suspect Unity.Physics is fast enough not to be your bottleneck unless you're doing 1000s of collider casts a frame or something
(but I may be missing context from earlier convos)
solid flume
Context: I need a vision radius. I see the approach Erick is talking about, and I want to explore different approaches as well and see what works how
Also, yeah performance is not really a concern for me
Its a simulation
So if everything isn't 60fps isn't a big deal
amber flicker
imo using physics (at least for now) means you can quickly experiment with different casts and reduce the code you have to write right now - it's also supposedly very well optimised code (though I haven't used it myself)
but obviously, it's good to try out these different options and learn / see for yourself
depending on how much time and appetite you have for such things ๐
solid flume
Quick questions: Do collider casts return what's inside it, if it doesn't intersect with the surface. And is there anything for on trigger exit? I can only find stuff that works on trigger enter
Also apparently there's visual scripting for entities? Do my eyes deceive me?
com.unity.visualscripting.entities
amber flicker
I haven't tried it, but yea it exists. I would be mildly surprised if it supported raycasts, collecting results and processing them in a way you want already but if it does, ๐
tiny ore
I agree with Timboc
My perspective should always be taken with a grain of salt (I'm a performance freak)
amber flicker
My expectation (untested) is the collider cast would return what's inside - though easy to test
tiny ore
Go for the physics engine to have a quick result
solid flume
Ookay, thanks
tiny ore
But @amber flicker this is one of the points that make Tim Sweeney and many other experienced developer look at ECSs with a grain of salt.
We also deliver a ECS-only gameplay engine (Photon Quantum), and I know the struggles of developers to "think" in ECS ways, and to develop gameplay, etc
It enables a lot of cool stuff, but it does have an impact in teams... It all boils down (IMHO) to what is your target audience... ECSs are great, but they are not really easy for the vast majority of devs... We REQUIRE an ECS approach due to performance reasons for multiplayer, but we try to EASE a bit to the regular developer with:
- lots of samples
- docs
- flexibility in the APIs (you can even use a semi OO approach to help you organize ideas).
It's worked well for the past 3 years... But I understand the criticism of the overall approach
amber flicker
Is this in reference to the post on the forums @tiny ore ? Or something I said here to Cryptic?
tiny ore
To the post, but in reference to the questions...
It's not only about the ECS libs being "incomplete". Doing things with ECSs is many times a bit more convoluted
amber flicker
Completely agree
tiny ore
And you only get the real benefits if:
- you have enough complexity to begin with
- you rely on well designed data parallelism
For most cases of "small" single player mobile games (the VAST majority of the Unity market share), ECSs are not really really necessary... And alienates most of the developer community they built.
And this is probably the reason why they will NEED to keep supporting 2 engines (Dots vs MB)
They just can not afford to "deprecate" MBs... Maybe never.
So we only hope they can fully support both engine flavors going forward.
We're all part of the unity community.
We all want it to succeed... And I in particular (I think my colleagues agree with me on this) would love to move to a pure ECS rendering + animation to fit Quantum even tighter (we will jump to it as soon as it's stable and complete).
But I just can't guarantee at all they (Unity) will be able to do both things well... Lets see... We're in this for the long run.
amber flicker
Sounds like we're in agreement basically? I wonder if something I said made you think otherwise.
For additional context, I'm most of the way through making an animation package for Unity (tween library with a Timeline for authoring). It was complex and a lot of work but the authoring workflow in editor is consistent if working with GameObjects or entities. It is all ecs under the hood, meaning decent performance but from a user perspective they can choose the paradigm they want to use the API in. It isn't perfect but the reason I bring it up is I see that as part of the solution to the problem of ecs being sometimes overly complex and not a great fit for many teams, is better tooling and to just keep working on getting features over to ecs land.
amber flicker
Not at all - just don't want to come across as though I'm evangelical haha
solid flume
So casts do return everything inside
tiny ore
nice then
solid flume
Also, any way I can have a nativearray of native lists?
I can burst all the collider casts I need to do inside of a job, but every collider cast needs to output a list
amber flicker
@tiny ore 1 sec, posted in here before.. finding a link
stone osprey
How do we realise the following in a ecs -> "Player should work to entity, once collided a timer should start a countdown" ? We already have a "MoveTo", "Position", "Collider" e.g. component... we simply have no clue how we should chain them in order to spawn a timer once collided... any ideas ?
solid flume
I guess I can have a work around with native multihashmaps instead of nested lists
amber flicker
tiny ore
Create a Timer component... Add it to entity on collision callback
Create a system to run timer
Thanks... will take a look now
Oh, it's for tweening...
You have no interest/opinion on character animation timeline editors?
It's a hot topic given Unity's dead state in MB-land (mecanim and playables are "abandoned"), and no sign of maturity status for everything DOTS
solid flume
If a native collection is created and disposed inside a job, I should use the temp allocator right?
amber flicker
Yea... well.. I know Unity have their dots timeline.. I don't want to get into rigging and state machines etc.. I trust Unity will do it fairly well in the end
tiny ore
I do not...:)
amber flicker
haha fair
tiny ore
Given what Mecanim was
amber flicker
notice I said.. in the end
tiny ore
and Playables (promising API-wise) is implemented as a Mecanim instance underneath,
In the end, possibly... I hope so
But what I've seen for (studios with enough resources) is a lot of custom tools (mostly around Playables)
If the new tooling has an API similar to Playables (with a good implementation underneath), I bet custom tooling will still be a thing
Hence opportunities for tool developers
amber flicker
Most likely.
On a smaller scale, atm I'm close to releasing a very modest asset (https://www.boxofclicks.com/btagged - half finished web page) that also attempts to bridge the workflow gaps a little. I think there's an opportunity for tool develops here to help pull the quality of the ecs experience up.. but of course, it still relies on Unity making a lot more progress.
solid flume
For the same entity, does the entityInQueryIndex stay the same across all identical queries?
amber flicker
Not if you have any structural changes in between the jobs running. Generally you shouldn't rely on it.
solid flume
Gotit. I'll rely on the entity.index then
amber flicker
that's the same thing ๐ - well, not the same thing but the same issue
just rely on entity
tiny ore
I guess the struct/type Entity is the thing... That's the unique identifier
I assume Unity never reuses the entity unique "instance" (with a versioned property)
solid flume
So my solution is queuing up collider casts in one foreach, doing it in a bursted parallel for job, then going through the results in the next foreach
solid flume
Does a nativemultihashmap need to be preallocated if I want to write to it in a parallel bursted job?
amber flicker
you don't need to give it a size but you need to allocate it - in your case probably Allocator.TempJob.
solid flume
So I don't need to use a parallel writer, I can just use a nativedisableparallelfor?
amber flicker
you should use the parallel writer version of it
I'm not sure if you need the nativedisable... if you're only writing to it
I have a feeling you don't
spark glade
Argh, shit updated Burst to 1.4.0.preview4 and getting like a million warnings Burst warning BC1370: An exception was thrown from a function without the correct [Conditional("ENABLE_UNITY_COLLECTIONS_CHECKS")] guard. which appear to come from within the Entities package itself 0.11.0-preview.7.
Anyone knows starting which entities version (or Unity version 2020 maybe?) those will disappear?
amber flicker
I guess with burst 1.4 you probably want entities 0.14 in Unity 2020.1.0f1 or later?
spark glade
Everything seems to run (on Android at least) so far...
It didn't complain otherwise
I feel backed into a corner... ๐
Can't upgrade to 2020 due to apparent Hybrid Renderer issues ๐ข
Want to upgrade to 2020 because entities beyond 0.11 is only 2020 and higher ๐ข
<sacrasm> I heard that Dots is secretly fully complete and they'll unleash it with a BANG right before the IPO ๐ </sarcasm>
amber flicker
sooner or later, trying to stick to 0.11 is going to be a headache. What are the apparent render issues you mention?
stone osprey
I simply cant find a nice solution for this scenario : Event occurs ( Button click e.g. ), Player moves to Entity, on collision spawn another entity. Only once... how would you implement this ?
The difficult part is the "once" here... Only execute that action once
amber flicker
one option is to create an event entity that is destroyed/consumed by the spawner?
stone osprey
@amber flicker How do you mean this ? The problem here is... how do we tell the ecs "If that player collides, only spawn stuff once... next time he collides, ignore it"
dense storm
Hello, anyone know if we can use Kinematica with dots project?
Since there is no animation system
dusky wind
@stone osprey alternatively
Use singletons
And just edit the target instead of spawning a new one
amber flicker
@stone osprey So.. the answer imo depends on the data. Why should it only spawn stuff once? If it's because the player doesn't have a sword yet and the spawner makes a sword, that might mean the spawn system checks if the player already has one for example.
stone osprey
@dusky wind @amber flicker Thanks for your answers. Lets use the following example to describe my problem more detailed. When we press on the "Chop tree" button a event gets spawned... our player listens to that event and should move to a tree. Once he collides with that tree we wanna spawn a "progress entity" which takes care about the chopping part. Thats what i need to realise inside the ecs. With monobehaviours i could simply chain actions/functions/delegates to realise this easily ( onCollision += spawnProgressAndRemoveFromCollisionLaterOn )... But this a ecs, so i have no idea how i could tell the ecs that it should only execute a certain task one single time on collision and next time we collide it ignores that
amber flicker
I guess there are many directions to approach this from. One way would be to add a 'ReadyToChop' tag to the player. The tree system sees a player with that tag has collided and sends another event or manipulates things directly. Or the player could check for trees and do a similar thing. The key is around the player having that tag and when the action is performed, the tag is removed.
A variation on that is to have a ReadyToChop tag that has a .Enabled bool that you set (instead of adding and removing the component).
stone osprey
@amber flicker Thanks a lot ! Now we are talking ๐ That makes sense... never thought its that easy.
amber flicker
ok awesome ๐ - I wouldn't say making all these systems is going to be easy but I'm glad that helps. One thing to keep in mind is if a player is told to 'chop' many times, only one tag is added until he chops and the tag is removed. This is great if it's what you want. If it's not, you can put a counter inside which you increment or you can look into using Buffer elements instead.
For the more general case you could make an array (dynamic buffer) of actions you want the player to perform and remove them as they're performed.
stone osprey
@amber flicker Most of the systems are already implemented ^^ another thing im struggling with is how i fire the collisions... my current approach is that i check all entities each frame and once a collision occurs i fire a "event-entity", that includes the data of entities collided... but instead we could also mark those both entities as "collided" directly ( which would involve a structural change )... What would you go with ?
amber flicker
these decisions (structural changes vs bools vs entities) are so dependent on the scales and systems involved - there's no general answer in terms of speed other than to profile with your use-case. As for style, I don't think it matters too much. Either way you have to deal with what if something collides multiple times in quick succession or what happens if both things collide with each other and both try to tag each other etc.
stone osprey
@amber flicker Alright, thanks ^^ than im gonna go with my approach till i really need to change it... because there so many ways of ecs-architectures im mostly not that sure what is considered clean and what not ๐
amber flicker
I don't think there's really any authority on the subject - still working it out ourselves. Stick with what you have until you find a reason to change sounds like a good strategy ๐
solid flume
The spherecast needs a sphere collider. As far as I know, I add colliders through PhysicsShape components. How do I get the collider through a entities.foreach?
solid flume
Do I have to dispose of native collections I create in OnUpdate?
There's something wrong with my code. Every time I press play, unity starts using 4GB RAM and stops responding
Apparently disposing of collections I create in onupdate was doing this
zinc plinth
I see the entities/hybrid renderer with addressables bug still isn't fixed, nice ๐
did I just lose a whole ms of cpu time by updating hdrp 
zinc plinth
depends, are you on built in render pipeline ?
zinc plinth
then should be a breeze
solid flume
This is what's taking up my CPU time
zinc plinth
be careful that you don't have accidently added physics colliders to your cubes
solid flume
That's probably it
zinc plinth
it's waiting 11.4ms for jobs from other systems, not the rendering itself
solid flume
Nope it wasnt that. I removed all colliders bar one, and it's still showing the same thing
Worse, if anything
zinc plinth
then show. the. jobs. running.
zinc plinth
there's a jobs tab in the frame debugger
zinc plinth
show me the components you have on your cube entities
zinc plinth
show me the components you have on your cube entities
solid flume
How?
zinc plinth
in the entity debugger
solid flume
zinc plinth
are you doing this on purpose ?.. https://i.imgur.com/fNZNuIA.png
zinc plinth
that's alot of chunks, do you have alot of types of entity in your world or just the renderer ones ?
solid flume
The green is one type of entity, blue is one, and theres a tiny little sphere
The only difference between the green and the blue is a material and a component
Also what is meant by chunks
There are 10k total cubes, so maybe that's why so many chunks?
amber flicker
is your material set to be instanced?
also, the hybrid renderer does take quite a long time for 10k dynamic cubes - probably around 4-5ms at a guess depending on machine
coarse turtle
Think of a chunk as a span of memory defined for an archetype (if you have an archetype of xyz, then memory gets allocated to store xyz)
solid flume
I don't know about instancing, but this is the material
amber flicker
don't think that shader's supported
solid flume
It's not shader graph. I noticed removing the light from my scene improved frame time, so I just made all the materials unlit
I dont know if ECS works with 2D or else I would've used that. I have no practical use for 3 dimensions
zinc plinth
isn't project tiny for 2d ?
deft stump
Oh it works alright.
Save for... Animating
deft stump
Curios. What game or poc are you trying to do?
solid flume
I'm trying agent based modeling
Think this https://www.youtube.com/watch?v=r_It_X7v-1E
I figured it out
I generate maps in edit mode. Turns out old cubes weren't getting disposed
Its better, but that culling job is still taking like 10ms
zinc plinth
it's somehow not doing culling per chunk but per entity, which shouldn't happen
amber flicker
switch to standard shader and enable material instancing to see if that helps
zinc plinth
and btw if you're going for what I think it is, you can use vertex colors instead of materials
you could also merge your meshes but thats a bit more advanced to get right
solid flume
I'm creating the landscape in editor, so I could do it. It's basically just a big mesh with however many quads I need and vertex coloured
tiny ore
@solid flume I'll play devil's advocate here. I have some experience with simulation (and other stuff you with a PhD in CS).
solid flume
@amber flicker no difference
tiny ore
It looks like you want to simulate something grid based, like finite elements, discrete integrations, etc
solid flume
Yes
tiny ore
Why the heck are you using entities then?
This makes no sense IMHO...:)
You should just use a native array
solid flume
It looked like a good way to learn
zinc plinth
@tiny ore ??????
solid flume
I see your point though
tiny ore
You have a Nail, and you are trying to use a screwdriver, honestly
solid flume
Store state in a struct array, update it in a bursted job?
solid flume
I don't
tiny ore
OR even better run this in CUDA or similar
solid flume
Never heard of this
tiny ore
Do you have any idea that pretty much all the fastest super computers actually run processing data on GPUs?
solid flume
Yes
tiny ore
Well, that's GPGPU (general processing on GPUs)
solid flume
I should do this through compute shaders?
tiny ore
CUDA is the most prevalent of the programming libraries for this (NVidia specific)
zinc plinth
@tiny ore and you're throwing a way too complicated answer and omega overkill at someone who just want to try stuff out
solid flume
I see that it would be way better for what I want to do, but I want small scale and learn ECS and try stuff out
zinc plinth
this is the #archived-dots channel
rancid geode
well, it is not, in an very optimized scenario
tiny ore
Yes, I'm throwing the industry standard...
rancid geode
for learning purposes I don't see why not
solid flume
And don't want to learn an entirely new tool just for a project I'm doing to stop me from being bored
tiny ore
DOTS is not ECS only
A burst compiled parallel job on top of native arrays of structs is nice
Sorry, but I'm not wrong
solid flume
I know you aren't. I'm not denying it
rancid geode
yes, but you surely can do that with ECS too if that is what you are trying to learn
tiny ore
I'm not a fanboy of any particular tech... I just pointed to the right options...:)
solid flume
I know I'm not using the right tool for the job
tiny ore
The problems you are facing are mostly because you are going around and using the wrong tool
solid flume
I just thought that trying to simulate a couple hundred agents might be a decent use case to learn ECS
I mean, it taking ages to cull is not my fault
tiny ore
You are designing stuff for entities semantics, where entity semantics are not a good way to express this problem
Specially archetype ECS (Unity's memory model)
You are neither benefiting much from data locality (not nearly as close as you'd be with a spatial buffer) neither making your life easier (in terms of expressive power/semantics)
solid flume
I'm not trying to offend anyone here, or try say anything is wrong. I'm not in a position to do that, but can we get back to the topic?
tiny ore
Sorry, I had to say this...
You can.
Won't comment.
You can always go the wrong way...:)
solid flume
I understand what I'm doing makes no sense, but I want to do it anyway
zinc plinth
kindly fuck off now, you're getting omega annoying
he wants to use dots and ecs, let him be @tiny ore
solid flume
I mean, there are nicer ways to say that
tiny ore
kindly fuck off now, you're getting omega annoying
Sorry... You can do whatever you want... Including call annoying someone telling you the right way to do stuff...:)
rancid geode
people, lets calm down please and go back to the topic
deft stump
kindly fuck off now, you're getting omega annoying
there's a proper way to say that
rancid geode
@solid flume is just trying to get some help to learn ECS, lets help him
this is what this server is for
solid flume
I'm storing data in native collections, and using it in a bursted job to do a bunch of collider casts. If I don't dispose of the collections in on update, memory leak. If I do, nothing happens
rancid geode
@solid flume how is your code?
tiny ore
I honestly just started with a proper DOTS way (name of the channel).
solid flume
make collections
fill collections in entities.foreach
burstedParallelFor.Schedule.Complete
use bursted results in entities.foreach (different one)
dispose collections
Basically my code^
Burst is turned off, so there's that
deft stump
this make collection part...
I'm guessing it's a nativeArray of sorts.
and you're filling it with stuff from components right?
zinc plinth
wait, why do you have burst turned off
solid flume
Because I'm using foreach loops and burst doesn't like that
rancid geode
dispose collections you are passing the Dependency as the parameter?
zinc plinth
just [BurstDiscard] it then no ?
solid flume
What dependency?
In effect, I want to run an entities.foreach, then a bursted job, then an entities.foreach, then dispose in that order
rancid geode
Dependency = MyJob.Schedule(Dependency);
MyNativeStuff.Dispose(Dependency);
yes, the Dependency property
I mean, if you Entitiess.ForEach is Run it doesn't make difference, but if you are using Schedule it does
thus why seeing the actual code would make easier to help
solid flume
So,```cs
var dep = new JobHandle;
dep = Entities.ForEach(...).Schedule(dep);
dep = burstedJob.Schedule(dep);
dep = Entities.ForEach(...).Schedule(dep);
dispose passing dep as parameter
(return dep/complete dep?)
Why does onupdate require me to return a jobHandle
rancid geode
you are using JobComponentSystem, use SystemBase instead
ComponentSystem and JobComponentSystem shouldn't be used anymore
but yes, your logic is right now
(return dep in your case if you want multithreading, complete if you want to finish everything right away on main thread)
solid flume
I can't use a GetComponentDataFromEntity inside an entities.foreach?
I want a component on an entity that I got from a component
coarse turtle
You can,
var someData = GetComponentDataFromEntity<SomeType>(true); // True for read only data, false if you need to write back
Entities.ForEach((Entity e) => {
var data = someData[e]; // Do w/e you need with data
}).Schedule();
you can even skip the GetComponentDataFroEntity altogether and just use GetComponent , HasComponent or SetComponent inside the ForEach
solid flume
Oh. That's much better
The GetComponentData thing throws an error when I play
The has component throws an error
amber flicker
literally GetComponent not GetComponentData if you want to replace the functionality of a CDFE
solid flume
Figured it out. Apparently I can't use those if the Foreach takes the same component as a ref parameter
amber flicker
yes, correct ๐
tiny ore
Honest question: is this from the same entity, or for example from another referenced entity?
In the loop iteration. The limitation of not being possible to Get the component
solid flume
Okay, now when I did the dependency thing, it's throwing an error that I need to call complete on the job that writes to the native collection before using said native collection, which is what I want to happen anyway but i guess isn't happening
@tiny ore another referenced entity
tiny ore
hmm
trying to understand the reasons why, that's it (has to do with the memory model). Thanks
solid flume
You're welcome
amber flicker
@tiny ore I think it's not strictly necessary but probably very verbose to define what you're trying to do is safe (if it is). When you do ForEach..(Entity e, ref Thing thing), the thing is in the chunk (as I guess you know). When you do GetComponent<Thing>(..) it's random access. By ensuring you remove the ref Thing from foreach, it means they can easily detect if you try to write to thing in parallel and if it's safe to do so. What would be nice in some future version is to have the expressivity to say something like "I want to write to all the Thing's in the iterated chunk and only ever read using CDFE". I didn't explain that well but I guess you're smart enough to work out what I mean haha
tiny ore
I assumed it is to protect concurrent modification, yes.
I just thought you could retrieve a copy of the original data, and modifications could be done with copy-on-write semantics under the hood...
But maybe I assumed wrong
Then it makes sense to not give access, because the loop may run in parallel, so you can guarantee no concurrent modifications happen to the same instance of the component
amber flicker
yea - plus, if you're sure anyway you can just disable the safety check
tiny ore
got it. ty
kind of obvious, ye, sorry... REF means you do want to write to it (at least the copy from the looped entity)
Say I do want to modify in parallel, but I want to use data (read) from the same component of other entities... What is the existing approach for it?
a) I manually create a copy of these so I can read the original data
b) any proper safe way to do it from the ECS API as it is now?
variation of a) could also mean I do one iteration with read only saving the data I want to modify in a buffer, then do a second parallel loop to apply modifications.
coarse turtle
the variation of a - exists to a degree - you can read the data initially - do some modification as a copy, and then apply the changes later through their command buffer
the changes aren't applied in parallel though, it's played back on the main thread
tiny ore
ok, so there's an exposed command buffer (that's one of the common solutions for this)
amber flicker
yea - there are existing ones (e.g. at start and end of frame) that are sync points you can use for this - you can also create your own
or yea, you could write to a queue or a nativearray in parallel then process
one technique I kinda use is to attach a SharedComponentData to a group of entities - then I can easily filter the entities at the chunk level and process them in order - useful if there's a hierarchical relationship with many entities
tiny ore
makes sense.
tardy spoke
I might frame this. I've been using HasComponent and was wondering if there were other ones, haha.
amber flicker
I still think everyone should be forced to write an IJobChunk before using the lambdas
tardy spoke
They abstract a lot away
I wonder if many people just use IJobChunks instead of the SystemBases so it's more explicit
amber flicker
yea - it's awesome.. but I can only imagine how magical it feels not knowing what is being abstracted
I don't think many people do (IJC instead of SB for explicitness) - it's just a horrible amount of code.
tardy spoke
I actually think this would be a lot cleaner and easier to understand as an IJobChunk
amber flicker
fyi your WithNone can be written WithNone<A, B>()
tardy spoke
I just find with the level of abstraction all the variables got named virtually the same thing which makes it fairly confusing
Also I'm not really sure exactly what "GetComponentTypeHandle" and "GetEntityTypeHandle" are doing
Well, kind of more how it's doing or why it's required
amber flicker
It's not what you're talking about but does VS's autocomplete/intellisense drive anyone else mad? I have a lot of Translation translation type of things and I swear 90% of the time it thinks I want Translation which is the exact inverse relationship of how often I do haha
GetComponentTypeHandle etc is all to do with HPC limitations I had assumed
tardy spoke
HPC?
rancid geode
I wonder if many people just use IJobChunks instead of the SystemBases so it's more explicit
well, I am one of those people haha
amber flicker
(High Performance C# ๐คฎ ๐ฌ ) the subset of C# that Unity support burst etc for - structs n stuff
tardy spoke
interesting
amber flicker
Gaaah I'm spending dayzzzz making examples, documentation, videos and tutorials for something that might sell zero copies. Much work, no fun. Bleh.
tardy spoke
MVP it. Throw it out there. Document later.
lol
DOTS people are used to no docs, it'll be fine
Besides, most the people that buy it are probably in this channel and will just ask you how to use it, hahaha
amber flicker
I actually hope it has a substantial non-dots audience ๐ค