archived-code-advanced

[hybrid tundra]

I know what is eating memory

textures in the memory

but the main issue is not just these

the main issue is that these are stay in the memory even when the whole scene gets unloaded

[untold moth]

30-40% is not too bad for textures.

[hybrid tundra]

~200 textures left in the memory after I unload the complete level

[hybrid tundra]

well, based on memory profiler, the textures are the issue

and nothing else

[untold moth]

Then they should be 100%, no?

Or I don't know. How do you define if it's a problem or not.

[hybrid tundra]

well, its the most huge part of the memory

that's how I define its a problem

[untold moth]

So you have 60-70% occupied by many-many different things?

[hybrid tundra]

so, if scene unloading is not freeing up memory from textures, then what?

[hybrid tundra]

ignore that part pls

I wrote percentage wrongly

lets focus on textures left in memory

[untold moth]

It depends on what textures these are. If they are used by the objects in the scene(and loaded with them), then they should unload together.

[stuck plinth]

iirc TMP just uses Resources.Load internally, maybe UnloadUnusedAssets is what you need?

[hybrid tundra]

well, they aren1

*aren't

[untold moth]

How are they loaded? Are all 200 textures TMP related? That doesn't sound normal

I don't see any profiler data shared.

Also, when you're talking about 30-40%, how much is it in MB? Are we talking several GB? Several hundreds of MB?

[hybrid tundra]

game has 6gb memory usage

4gb is texture memory

what I see is that they stay referenced in materials

[untold moth]

Are we talking RAM, VRAM? Or is it unified memory platform?

[hybrid tundra]

so the textures are still referenced from materials

after I unload the scene

[untold moth]

Then the materials are still loaded

[hybrid tundra]

no idea, Memory Profiler only tells its memory

[hybrid tundra]

how could those be still loaded after I unloaded the whole scene?

[untold moth]

For example if you're instantiating them at runtime.

[ebon abyss]

It makes a distinction between graphics memory and.. not graphics memory

[untold moth]

Are you using the memory module in the regular profiler or something?

How are you even trying to optimize memory without knowing if it's RAM or VRAM? lol

[ebon abyss]

If it's mobile or other integrated graphics it doesn't really matter

But yeah use the actual memory profiler

[untold moth]

Investigate the chain of references. The issue is likely the objects that hold the reference to these textures.

[ebon abyss]

Xbox Series S haunts me

[hybrid tundra]

nope

[hybrid tundra]

why is that matter? game is crashing, most of memory is used by textures

and textures are leaking

[hybrid tundra]

it writes me gameobject names which are not even loaded at that point

at least Hierarchy window doesn't show them at all

[untold moth]

Well, then they are loaded. You'll need to investigate these objects.
There's not much else we can tell you unless you share the info that you're seeing.🤷‍♂️

Are these objects instantiated at runtime? Loaded in some non standard way? Are they DDOL? There are so many factors and you have barely shared any info.

[echo coral]

TMP textures often remain loaded always (especially if used in the global fallback list)

[untold moth]

If it's really what's happening, it's easy to confirm in the profiler.

[ebon abyss]

Are you using additive loading?

[hybrid tundra]

yeah, that part is pretty unusable

[hybrid tundra]

no, they are not

at least not in the Hierarchy

Memory Profiler shows me textures which are pointing to gameobjects, which are not existing in Hierarchy at all

[untold moth]

The hierarchy is not your whole game memory you know.

[hybrid tundra]

I know, but then where are these gameobjects

[ebon abyss]

Unused assets aren't unloaded until you tell them to
https://docs.unity3d.com/6000.1/Documentation/ScriptReference/Resources.UnloadUnusedAssets.html

[untold moth]

1. Are you profiling a build?
2. What are these gameobjects? How are they loaded? What does the memory profile show you when you select them?
3. Are you gonna share more info or not?

[hybrid tundra]

1. no, but the result was the same with builds

2. probably cached enemies, so maybe there's a script containing references to them, but I don't know then why those have real names, not just "null"

[untold moth]

In editor, you'll see a lot more assets loaded in memory. Things that you opened previously in the inspector, for example. This is very unreliable. Don't profile in the editor.

[hybrid tundra]

3. I'm still investigating, and I'm also under NDA

[untold moth]

As long as objects are in memory(not collected by GC), they are not null. And if they are referenced by something, they can't be collected by te GC. You'll need to look at what's referencing them.

You wouldn't see "null" entries in the memory profiler, as it would mean they don't exist, which would contradict them being in the profiler, right?

[hybrid tundra]

I know these

but I also know gameobjects like those are theoretically cleared up already

[untold moth]

Profile the build

And see what's referencing them. Or if they are marked as don't destroy

[hybrid tundra]

assetbundle

that's referencing a gameobject which not existing theoretically at that time in the build

[untold moth]

So there you have it. It's in the asset bundle.

How are you loading the asset bundle? Are you unloading it when you don't need it?

[hybrid tundra]

don't know the project yet, I thought these are not active, when addressables are active

[untold moth]

Then you'll need to research that. I'm not that knowledgeable with the asset bundles myself, so you'll need to look at the docs.

[stuck plinth]

addressables uses asset bundles, it's the same system underneath

[untold moth]

https://docs.unity3d.com/Packages/com.unity.addressables@1.21/manual/AssetDependencies.html#sidetoggle

[stuck plinth]

if bundles aren't unloading as expected, maybe you have a reference count that's not being decremented, make sure anything you're loading with LoadAssetAsync etc is being released properly

there's a profiler module for checking the reference counts

[hybrid tundra]

can you help me find that?

the reference count thing

[stuck plinth]

this one!

[hybrid tundra]

the addressable assets window is totally empty

during profiling

[echo coral]

you need to have done a local build with a report for it to work

The way i usually test using this is a local build then play using "Use existing bundles"

[jaunty compass]

although i change local rotations to match the player orientation and seat orientation, the rotations end up being different in the scene, messing up my player

is there some weird discrepency how unity handles rotations

[hybrid tundra]

it says Build report not found. Build report data is required to display correct frame data

[cold kernel]

Hello everyone, I would love to generate icon images from the icons of my computer files.
Is there anyone who knows Windows API and could help me in some way with it please?

[hybrid tundra]

what do I need to browse here?

[devout hare]

Those are the same rotation

[stuck plinth]

turn this on in the addressables preferences and look for the file in your addressables folder (in Library)

[echo coral]

you can enable build reports in preferences or it may ask you when you go to build bundles (2022 and later)

[hybrid tundra]

that was enabled for me

but how is that file called?

[echo coral]

it will find it itself but it should match the build you are profiling (if in editor should be fine if done recently)

this should probably move to #📦┃addressables

[hybrid tundra]

not sure if you can easily find that if you make builds for NDA platforms

[echo coral]

it will make the report when built for any platform

[stuck plinth]

i'm pretty sure it always goes into Library\com.unity.addressables no matter what

[echo coral]

you just need to add it to your reports window for use:

[jaunty compass]

idk maybe, but these values should be absolutely the same, which they are not.
Otherwise this mystery happens

[devout hare]

(270, 90, 0) and (-90, 0, 90) are absolutely the same rotation. The issue is somewhere else

Objects should generally be rotated to match each other using global rotation

You should also check what happens if both have rotation (0, 0, 0) and if they still mismatch then the objects' forward vector is wrong

[jaunty compass]

i initially did work with global rotation using
orientation.rotation = playerStateManager.CurrentSeat.transform.rotation;
but the rotation of the vehicle would impact the orientation of the player, and it would only be correct if the vehicle was pointing a certain direction

[devout hare]

Then either of those objects are wrong

[jaunty compass]

thanks i ended up looking elsewhere in my code and found the issue!
after setting the orientation, i set the transform.rotation which is a parent of orientation, which caused these strange issues!

[brisk spear]

anyone have a clue on why the second ReleaseMutex() onwards throw a not owned error?

[sly grove]

we're supposed to guess this without seeing any code?

[brisk spear]

im hoping this is a known bug but here is the code

[sly grove]

There are no known bugs in System.Threading.Mutex no

[brisk spear]

I can't debug inside Mutex since it uses external code

[sly grove]

it's 99.9% a bug in your code

not a bug in Mutex

[brisk spear]

ye i just cant put my finger on it

u can see from the screenshot i even log every wait & release event

[sly grove]

maybe there's more than one WinRecognizer in action?

[brisk spear]

considering changing to named pipe IPC instead of shared memory IPC

[brisk spear]

i only called it once at the start and close it onAppEnd but will check anw

[fathom geode]

Trying to make my procedural spider walk on 90 degree surfaces, but it gets stuck on them sometimes.

In the leg mover script, a raycast is fired from the cast position to the step targets to determine the normal of the surface and the new position of the step target. The average of the normals is calculated in the body script and the body rotates in accordance to that average. When the raycast doesn't detect any surface, this means the player is now at the edge of a 90 degree surface, and a raycast is fired backwards from the step target and detects the surface.

Function in the leg mover script responsible for setting the values of step targets: ```private void setStepTarget(){
bool hitWall=false;
if(pb.xInput!=0 && Mathf.Sign(pb.xInput)==wallRayDir){
var hit = Movement.castRay(castPos.position, wallRayDir*pb.transform.right, pb.wallLegRayLength,Color.red, pb.level);
if(checkHit(hit)) {
setHitValues(hit);
hitWall=true;

        }
    }   
    if(!hitWall) {
        var hit = Movement.castRay(castPos.position,(stepTarget.position-castPos.position).normalized,pb.groundLegRayLength,Color.red, pb.level );
    if(checkHit(hit)){
        setHitValues(hit);
    } else {
        hit = Movement.castRay(stepTarget.position,Quaternion.Euler(0,0,-wallRayDir*pb.stepRayRotation)*(stepTarget.position-castPos.position).normalized,pb.groundLegRayLength,Color.red, pb.level );
        if(hit.normal!=legNormal){
        setHitValues(hit);
        }
    }
    }
}    

Function in the body script for rotating and positioning the body: private void setBody(){ var hit = castRay(transform.position,-transform.up , groundBodyRayLength, Color.red, level); if (hit) { var targetRotation = Quaternion.LookRotation(Vector3.forward,averageLegs); transform.rotation=targetRotation; rb.position = hit.point + (Vector2)transform.up* bodyOffset; } }

(Both functions are in their respective Update functions)

SOMEHOW I FIXED IT BY PUTTING ONE FUNCTION IN FIXED UPDATE??

Before that, the function the raycasts was in the leg logic script's update function, and the setBody function was in the body logic script's update function, and instead i made setBody to be in fixedUpdate

How did this work?

[minor ruin]

var relativePos = transform.InverseTransformDirection(target.position);
relativePos.x = 0;
var targetPos = transform.TransformPoint(relativePos);

transform.LookAt(targetPos, transform.up);```

why does this only work when im moving the camera?

this is how the Hierarchy is setup

[long ivy]

well one glaring issue is that you're feeding a world position into a method that is looking for a direction

[sick rapids]

On fresh boot of a unity project, what is your memory footprint? Just looked and mine is 9gigs fresh, technically down from 11 at startup (after letting it settle). This seems abnormally high, and I'm curious if there's a way to see where all that memory is

The project is a 2d project only a month in, no use of the resources folder

I do have 64gb of memory on my pc, idk if it just eats more because it can

Perhaps someone can help me read this, or post their own for reference?

[hybrid tundra]

can the AsyncOperationHandles cause a memory leak, if I use Addressables.LoadAssetAsync and storing the handles only in local variables?

[compact ingot]

the handle itself is a struct, so it can't leak on its own, but you can use it in a way that causes a leak, likely pretty much everything else too.

[hybrid tundra]

well, if I just create them as local variables and not calling Release on them, then it will leak, right?

[compact ingot]

as long as it doesn't place itself or another resource in a scope that is not finalized when your local variable goes out of scope, then it wont leak

typically such types would implement IDisposable

[hybrid tundra]

can you rewrite this sentence with more simple words?

thanks

[stuck plinth]

AsyncOperationHandles don't implement IDisposable but still need to be released

[untold moth]

Looking at the processes, the editor itself only eats 2.5GB. 3.5 GB is your IDE. 960 - profiler, then some other extra editor processes for 1600, dotnet runtime, and some other minor stuff, like MSBuild and compilers.

[untold moth]

The editor itself looks pretty minimal.

[untold moth]

Untracked is usually reserved memory. If the editor releases some of the allocated memory, it would usually not release it to the system but keep it reserved for future allocations. Because requesting memory from the system can be expensive performance wise.

[ebon abyss]

Mine slowly increases to 50GB or so where it runs out of swap space and crashes

But yeah this doesn’t look like a lot

[compact ingot]

if you create the handle with new AsyncOperationHandle() and store them in a local variable and otherwise do nothing with them, you will not produce leaks, even if you do not release them since no resource is associated with that handle. If you get them from a member of the Addressables API, they will probably be associated with a resource that will need to be released when no longer needed to avoid a memory leak.

[hybrid tundra]

and that's what is happening

the latter part

@compact ingot can I use Addressables.Release with the asset name as well?

as I read, it works only with the actual object, not its name

[echo coral]

you mean address? I think you can only release with the asset or the handle.
My workflow is to release the assets, either by some component handling the load or a single handler that loaded many objects.

e.g. for loading a sprite to display:

icon = await Addressables.LoadAssetAsync<Sprite>(item.spriteAddress);
if(image != null)
{
   image.sprite = icon;
}
else
{
    //If UI image were destroyed mid load
    Addressables.Release(icon);
    icon = null;
}

[sick rapids]

Your unity process eats 50gb at worst? Damn! Do you think it's because we have higher amounts of memory to begin with?

What I was really concerned with is, at some point my designer is going to end up in the engine, and if he doesn't have more than 16gb of memory, 10gb is going to destroy him

[sick rapids]

Thanks for the help breaking it all down

[ebon abyss]

No it's because I'm making an open world game and also the engine has a memory leak 🙂

I also keep the editor open until it crashes, which usually takes 2 - 3 days

[gloomy nova]

Has anyone used Arch (https://github.com/genaray/Arch/tree/master) with Unity? if so, how was working with it?

[ebon abyss]

The downside I see here is that you still end up with GameObjects for basically anything... This is probably better in another engine.

[gloomy nova]

Nah, there's plenty of shit you can do without GameObjects

I pretty much only use gameobjects for representing things in the actual engine, all the game logic happens in code that's otherwise removed from Unity

[hybrid tundra]

do someone know how to use memory profiler?

[ebon abyss]

Yes

Someone knows how to use the memory profiler. Now ask your question.

[hybrid tundra]

if a gameobject has a Renderer, and that gameobject's own class holds a reference to it and its being stored, etc. But the gameobject gets destroyed once

but the material still be in the memory

can I somehow forcely release that memory?

should I override OnDestroy on the script in the gameobject

and when that gets called, should I set the Renderer's material to null?

[ebon abyss]

Unity doesn't release loaded assets until Resources.UnloadUnusedAssets is called.

[hybrid tundra]

ineffective

[ebon abyss]

Are you profiling a build or the editor?

[hybrid tundra]

build

but I gave this task up for now

based on the two profilers, there are unexisting assetbundles referring to some textures (addressable asset profiler doesn't show me those at all, memory profiler still referring to some weird blablah named addressable bundles)

• there are other references as well

like memory profiler tells me that there are gameobjects actively holding those materials inside ParticleSystemRenderers

but I manually cleared those, nullified those materials, called Resources.UnloadUnusedAssets, and still nothing was freed up

[dusty wigeon]

If you use addressable you need to release the handle

[hybrid tundra]

I know

[dusty wigeon]

If you use addressable, memory wont be release as long as another object in the bundle is still referenced.

[hybrid tundra]

also know

[echo coral]

Memory profiler helps to find what textures are loaded and also helps you find leaked "managed shell objects". A texture can get unloaded from native memory + gpu but the c# Texture2D object can sometimes hang around longer (because only the gc can delete it).
more info: https://docs.unity3d.com/Packages/com.unity.memoryprofiler@1.1/manual/managed-shell-objects.html

[hybrid tundra]

sorry for pinging you again, maybe I can do it, because you also did it.
But the issue is that I had like 13 addressable bundles stuck in memory, I cleared up those, I had only 3 (and those are containing only 6-7 materials. But the texture memory got increased.
I checked the memory profiler, and for some textures it shows that its referenced from 2-3 places, for some it doesn't show addressables, it just show me some gameobjects, which are not loaded anymore. Some of them has child objects with ParticleSystem components, their Renderer has the materials, which should have been unloaded. But even the gameobjects are invisible in Hierarchy view.
Tried to add a script to the parent to find all ParticleSystem children and set null to the materials and sharedmaterials of these child ParticleSystems, I also tried to use Resources.UnloadUnusetAssets (or whatever it is called), but the memory was still the same.
Some of these textures are also referenced by some asset bundles, which have a name with random letters and numbers, and those are not visible in the Addressable Asset Profiler at all. But only a small percentage of these textures are still referenced by these "ghost" addressable bundles.

[echo coral]

Resources unload won't help with addressable loaded things. Releasing loaded assets correctly is the real solution.
If object a is loaded and depends on texture b, releasing a will then unload b if nothing else needs b.

The only manual solution is to manually destroy the textures used by materials but this may break things later.

[hybrid tundra]

but based on what I wrote there are textures, which are yeah, used by some kind of unexisting addressable bundles, but they are also used by other things

I wanted to figure out how to get rid of those other things I mentioned

[echo coral]

Share screenshots of this if you can.
If you load a texture directly with addressables and keep a ref to it but release it, only the managed shell survives, the texture data is still unloaded from GPU memory.

So you need to not confuse yourself by looking at the managed shell only. The profiler should tell you if it's a leaked managed shell

[hybrid tundra]

I will, but I'm so tired to censor all images

only tomorrow

[echo coral]

Don't bother then. When I debugged similar issues I mostly used the profiler to tell what textures were loaded but I understood what was loading them in initially so releasing correctly was easier

[hybrid tundra]

I will bother, because I need to solve this issue asap

and yeah, it would be easy, if I would see through all of the project

and the million source codes

because I just got into the project, and don't really know most of the source code

[echo coral]

You can use a dependency viewer package to help find what uses a texture
E.g. https://github.com/yassy0413/UnityAssetDependencyViewer

There are editor functions to get asset dependencies too

[hybrid tundra]

I will check these tomorrow

I will go to sleep in 10 min

[echo coral]

Sometimes an asset you didnt think depended on something infact does (prefab variants have some weirdness with deps)

[quaint bloom]

Hi ! I have a question regarding cinemachine. Is it possible to create a freeloock camera, that focus on multiple target ?

and also, can I know via code if a target is inside or outisde the camera frustum ?

[quaint bloom]

nevermind I found out sorry sorry

[sly grove]

#🎥┃cinemachine for the future

[hybrid tundra]

@echo coral I have collected the main things I can see in Memory Profiler and Addressable Assets Profiler. Please go from page 1 to page 4, because my comments are mentioning parts from previous images.

[echo coral]

If you check the built addressable assets (Library/com.unity.addressables/aa/{platform here}) do you find a bundle with that filename? It doesnt make much sense for a bundle to magically exist and be in use that isnt real. You can change your group settings to make the local bundles keep their group name and not be hashed.

When you load your menu, go elsewhere then back to the menu, do the handle counts go up? Those are things that never get released properly and should be looked at

Remember you need to release any loaded asset or any InstantiateAsync() gameobject instance. Unity are dumb dumb and dont make these instances release themselves on destroy. It can be easier to load a prefab and instantiate the old way so you just have to release 1 thing (the prefab asset).

[hybrid tundra]

None of the addressable bundles have names starting with 8b4972 etc., all of their names are starting with normal english words. I can only see the bundles which are also in the Window - Asset Management - Addressables - Groups.
"You can change your group settings to make the local bundles keep their group name and not be hashed." where?
"do the handle counts go up?" You can also see that in the screenshot that yes, it goes up. But again, if you check my images I've sent, those are containing only those 7 materials, and that's needed for the main menu. It doesn't contain any gameplay related things. So I doubt this would be an issue. The Memory Profiler shows those FX stuff you can see on the images, that's much more suspicious.
About releasing, I theoretically made everything to be released. But there are like 30-40 different classes which does the addressable loading. Why? I don't know. I don't even know when should certain addressables loaded and when should be unloaded. Maybe I've setted the Release part to everything, but I can't be sure, because everything refers to everything.

[echo coral]

groups can produce many asset bundles so stop looking for a 1 bundle for 1 group...
Those "leaking" materials depend on textures surely? It will cause their dependencies to stay loaded forever too

Click a group and change "Bundle Naming Mode" to adjust how the bundles a group makes are named:
https://docs.unity3d.com/Packages/com.unity.addressables@2.0/manual/ContentPackingAndLoadingSchema.html#advanced-options

[hybrid tundra]

"Those "leaking" materials depend on textures surely?"
what do you mean?

[echo coral]

Okay so what assets can you tell so far never get released? Is it those 7 materials?

Did they get loaded by you or are they dependencies of another asset?

e.g. you loaded a particle system prefab which uses a material

[hybrid tundra]

based on Addressable Assets Profiler, only those 7 materials.
based on Memory Profiler, its a lot more, like 500 textures

who to trust? I don't know

but Addressable Assets Profiler is experimental profiler module, so maybe I can trust more in Memory Profiler?

[echo coral]

TRUST BOTH

[hybrid tundra]

honestly, all I can see is the screenshots I've sent

I can't see anything else

and I don't really understand those screenshots to be honest

[echo coral]

Look i can load 1 material with addressables via Addressables.LoadAssetAsync<Material>("MyCoolMat") but it could use 10 textures.
Addressables is showing "MyCoolMat" has 1 handle and ofc the memory profiler shows the 10 textures loaded in for it.

So until i release "MyCoolMat", those 10 textures are never going to be unloaded

[hybrid tundra]

one thing is for sure, if you check the images, the gameplay materials, which should not be in the main menu, are not even referred by addressables

you can see that in the screenshots

this is why I'm confused

[echo coral]

look figure it out yourself ive said enough

[hybrid tundra]

O.o

am I look like I'm arguing with you?

[echo coral]

ive said all i want to and i cant help any more

[hybrid tundra]

don't really understand why are you pissed off, but ok

[hybrid tundra]

1 material, 10 textures... main menu materials don't contain gameplay textures in this project.

Addressable Assets profiler shows 7 materials, but those are only main menu materials

FX stuff are inside fully different materials

I've checked it

those materials name are starting with "fx_"

I have zero idea why you're offended...

I appreciate your help, and I'm thankful, and its just a discussion about what can be the issue... but I'm really trying to say that one profiler tells me only main menu materials and other profiler tells me "main menu only" materials AND "gameplay-only" materials

[sage wadi]

Is 150 lines for an infinite layers recursive process good?

[compact ingot]

Nobody knows

[sage wadi]

Real

[quaint bloom]

Hi ! I'm using cinemachine free look with a targetGroup, and I was wondering if it's possible to lock the camera in a certain orbit.
(like if I turn the camera from the side of thoses targets, it stay to the sides of thoses target ; and like if I turn the camera from behind one targets, it stay that way even when targets turn around each other)

here's a drawing to better explain it : (if it can help better vizualize what i'm going trought)
(here the camera stay & rotate at the side of thoses targets)

[compact ingot]

have you looked into target groups?

[quaint bloom]

the freelook camera has a targetgroup as a tracking target, so it work fine

it's just that I want to preserve the current orbit when targets move around

[compact ingot]

i have a feeling what you want is not specific enough to be automatable

how do you define what the current orbit is?

then how would you redefine that when the targets move?

[quaint bloom]

I tried to get the vector from target A - target b position and compare that to the world position, to then define an offset value to apply when it change.. but I couldn't make it

something like :

Vector3 targetVectors = targetGroup.Targets[0].Object.position - targetGroup.Targets[1].Object.position;
float moveAmout = Vector3.Angle(targetVectors,Vector3.forward);    

[untold moth]

Forget about addressables for a moment. Reproduce the state of objects being in memory whenever they shouldn't be, then dig into the reference hierarchy and check the info of each item. Ultimately, the deepest element in the hierarchy is the culprit. The info should tell you why this object is loaded, so focus on investigating that cause.

[hybrid tundra]

I would love to reproduce the state in an example project, if I would understand the reference hierarchy

can you check the page 4?

[untold moth]

Select each object in the hierarchy and look at it's info. Especially the deepest gameobject.

[hybrid tundra]

Okay, and what to check for?

in the info

[untold moth]

All of it. Basic info would tell you if it's referenced and why it's in the memory. Help would explain in more details. Advanced would have advanced info(don't remember what it contains).

[hybrid tundra]

what can you understand from this?

[untold moth]

First is that it's referenced by something. Is there something deeper in the hierarchy?
Then that there are fields on that script instance that could be referencing other objects and preventing the GC from collecting them.

[exotic trout]

Just random thought probably not relevant but any chance prefabs from the bundles (via direct loading or due to being referenced from direct loading) are still lingering in hideanddontsave land? Do those show up in the primary hierarchy in profiler?

[hybrid tundra]

from this FXCollection I can't go further.

[untold moth]

Then maybe try double clicking it or finding it in the memory profiler and selecting it there.

[hybrid tundra]

you mean in the center, in the Unity Objects area, right?

[untold moth]

Yes.

[hybrid tundra]

in here, the hierarchy is deeper

so what should I see in here?

[untold moth]

Same with this object. The Help basically tells you what's going on. You need to find what's referencing it.

[hybrid tundra]

I know that part, but where should I find what's referencing it?

is it visible from these screenshots?

can you see it?

[untold moth]

No. As it's the deepest in the hierarchy. Find it in the unity objects area and look at it's references.

[hybrid tundra]

ahh, so I should do it back and forth between the Unity Objects tab and the References?

[untold moth]

Yes.

[hybrid tundra]

good to know, it was not trivial

I can't find the ProcGenManager between the Unity Objects

so I guess that's the one referencing it?

[untold moth]

It's in the reference chain for sure. However the profiler says that it is referenced once by something.

[hybrid tundra]

ahh, because of the Referenced By: 1, right?

[untold moth]

Right.

Did you not read through the memory profiler manual?

[hybrid tundra]

I usually have very hard time understanding manuals, because they use very complex and long sentences

I usually just use youtube tutorials or anything else other than Unity manuals

[untold moth]

Well, if you're asking in this channel, you're expected to have done a minimal amount of research on the topic.

[hybrid tundra]

I mean I know English, but maybe not that much, like someone native speaker

[untold moth]

Many devs here are non native English speakers, yet they have no problem reading the docs. Anyways, you can learn in whatever way you like. I'm sure there are plenty if video tutorials on the memory profiler. Did you not look at any?

Anyways, if you can't find that object anywhere in the profiler, it's a good idea to see where it is used in code and whether it's reference is correctly nullified after it's destroyed.

[hybrid tundra]

I've seen some, but that didn't explain this part to me

[hybrid tundra]

okay

[hybrid tundra]

so to understand you correctly, if the Memory Profiler search field doesn't give results back for ProcGenManager class, then I should check the class itself, and especially that Dictionary the screenshot I've sent to you, mentioned?

[untold moth]

What dictionary?🤔

[hybrid tundra]

in here

you can see that

[untold moth]

It's the other way around. The dictionary is referenced in the ProcGenManager, not the other way around.

You need to look where in the code the ProcGenManager could be referenced.

[hybrid tundra]

isn't it about the ProcGenManager contains a dictionary with TileConfig, which contains the RitualTileData, which contains NetOjectController, which contains the FXCollection, which contains List<FX>? I have seen that from other textures that the FX is on gameobjects, which have ParticleSystemRenderer, and that contains materials, which seem to be still in memory somehow

ProcGenManager is referenced in 84 places in the entire project

[untold moth]

The core of the issue is that ProcGenManager is destroyed, yet still referenced somewhere, so whatever it references(eventually leading to your textures) can't be released either.

[hybrid tundra]

aaaaaaaaaaah

okay

[hybrid tundra]

and can't I write something into the ProcGenManager itself to release its own stuff?

and by the way

ProcGenManager is a MonoBehavior

if t:ProcGenManager filtering doesn't give back results when returning to main menu

then how could it still be referenced?

[untold moth]

I'm not sure if OnDestroy is called before the GC collects it, but if it does, then theoretically nulling all of it's fields in OnDestroy should work. But I'd rather make sure the ProcGenManager is not referenced after being destroyed instead.

[untold moth]

It's referenced in a script somewhere. Like a DDOL or there's a static reference to it.

[hybrid tundra]

I mean it will be just painful to check all 84 places where ProcGenManager is referenced

and even if I modify all 84 places

then still that won't be 100% sure to get a meaningful result in the memory

[untold moth]

Learn to look at the references window more efficiently. You only need to look at each separate class that references it. I'm sure there aren't 84 classes that reference it. Look at where it is assigned to class fields.

Ideally, make sure it's not assigned to class fields in many objects, just in one or a few at most.

[hybrid tundra]

I mean I have clicked in Visual Studio right click onto the ProcGenManager name, -> Find all references

and it gave me 71

[untold moth]

For example, here I have 13 references to a class in my project, but with a quick look at the list, I know that I only need to look at these 4 places.

[hybrid tundra]

but Unity itself wrote 84 references above the ProcGenManager class header

[untold moth]

What header? Are you doing C++ or something?

[untold moth]

Well, that's what I'm telling you - be smart about how you look at the references.

[hybrid tundra]

noo, not header, class declaration I meant

in here it writes 84

oh okay

from your image now I get it

[untold moth]

Again, it's all the places it's refrenced. Not all of them are relevant

[hybrid tundra]

what you've meant

okay

[hybrid tundra]

I've checked it. So I can only see these:
in million places: ProcGenManager.Instance. <- and its refering to the fields of the class

ProcGenManager.FromMenuSeed

ProcGenManager.TileSize

ProcGenManager.TravelPointData data = ...

there's only one place, where I can see
static ProcGenManager _instance,
but its in the ProcGenManager itself

maybe _instance should be destroyed, when returning to main menu

right?

[untold moth]

Ok, so it's a singleton. It has a static reference. You need to set it to null when the object is destroyed

If it's only supposed to be in one scene, I'd reconsider using a singleton pattern for it, but if you really want to, maybe try setting it to null in OnDestroy. Or if that doesn't work, on scene transition.

[hybrid tundra]

yeah, probably who wrote it also wanted it to make it as a singleton

I mean it seems to me that its just not used correctly

ok, thanks

one other thing I have recognized

is that some textures have setted Read/Write to Enabled

but the project doesn't contain Texture2D.SetPixel or SetPixels

so I guess these Read/Writes can be Disabled safely, right?

[untold moth]

It depends. Some engine features might be requiring them on the CPU side.
Try disabling them and see if anything breaks or errors are thrown.

[hybrid tundra]

hmm, then maybe I should ask the others if those were setted intentionally, because it would take infinite time to test each part of a multiple hour long game

[untold moth]

Yeah. It's always good to ask whoever is responsible for it of their intention.

[hybrid tundra]

okay, thanks very much for today's help

I really appreciate it

can the textures enable Read/Write by default during importing btw?

or it required from someone to manually set some of these to Read/Write Enabled?

[untold moth]

I think it's off by default

[hybrid tundra]

:/

[hybrid tundra]

so I can be happy, because there are cases apparently, when its enabled by default

and then I could disable it

[austere kayak]

Hi guys, I got a problem, I got laser GO (through LineRenderer) + BoxCollider set to trigger, but the laser can push player when hit

the lasers are rotating around the Boss

[hybrid tundra]

is rigidbody added to the laser go?

[austere kayak]

nono

[hybrid tundra]

then add it 🙂

[somber charm]

Hey anyone free for a call today? I've got something really interesting

[plush hare]

Pretty sus, considering you just joined the unity discord today

these channels aren't for private help regardless.

[somber charm]

Not trying to be sus and I don't need help!

I've created a system in Unity that does not use procedural logic. It's a new way of thinking. I want to walk someone through my system and see if they can grasp the shift.

[plush hare]

You should just post it here

[somber charm]

Not what im looking to do that this stage. Looking for a one on one conversation. I know that asks alot, im sure lots of social anxiety here.

[plush hare]

No offense, but unity has hundreds of thousands of daily users. I doubt you've figured out a new and groundbreaking way to use the engine.

Unity uses C# which is typically used in OOP settings. So I'm not sure what you're talking about with procedural logic.

[somber charm]

Procedural logic = loops and conditionals.

[plush hare]

Just post your question. Nobody is going to ridicule you

[somber charm]

I don't have a question! I am trying to TALK to someone about my system. Not type. I want someone to feel the weight of what im saying. I am simply looking for someone who wants to understand something new.

[plush hare]

👨🏿

[thin mesa]

you are incredibly unlikely to find someone here willing to talk with you one on one about something you are being incredibly vague about. either describe this "system" here or find somewhere else to ask for private conversations at

[somber charm]

Thats fine. No who im looking for then. When I say "eliminate loops and conditionals" the right person will want to speak with me. That person may not be here and that's fine.

[plush hare]

That person probably is here, you're just choosing to be difficult.

[long ivy]

I can't speak for anyone else, but for me it just sounds like you have no idea what you're talking about

[plush hare]

Nobody wants to hop in a call with a random person. Especially people that do know their shit.

[somber charm]

I apologize if I come off that, purely want a conversion with someone.

[plush hare]

Good luck

[somber charm]

Thanks!

[thin mesa]

then have the conversation here in this channel. this server is not the place to seek private one on one conversations/help as the pretty much completely defeats the purpose of a community server.

[somber charm]

Thats cool brother, that's what the conversation is for. Have a good one

[somber charm]

Noted. Looks as if I am in the wrong place. Thanks

[plush hare]

You're in the best place to discuss unity topics. Again, you're just choosing to be difficult.

You will not find a better community to discuss unity.

[thin mesa]

intentionally being vague about this and refusing to elaborate publicly is going to pretty much guarantee nobody wants to discuss it with you

[somber charm]

All good, im in the wrong spot. Take care

[plush hare]

lmao

[plain abyss]

Is there actually a point to this or did you just want to show everyone you just learned about declarative programming

[robust pulsar]

This guy is so funny lmao
Shows up —> claims to have invented a new way to program —> refuses to elaborate —> resists all attempts to converse —> leaves

[untold moth]

Sounds like something a complete beginner might be interested in. Maybe try in #💻┃code-beginner

[tired fog]

https://www.youtube.com/watch?v=tMDmwD1cQPg&t=690

[cedar anvil]

you have to at least give people an idea of what it is you actually created

right now all youve said is "I created a new type of system", you should at least elaborate a little on what it really is

[plain abyss]

They're not even still here, they left a while ago

[cedar anvil]

shucks

I was going to ask what the motivation for eliminating conditionals is

what could possibly be a better solution than if(a < b) { ... }

[plain abyss]

It is a thing, Declarative Programming. Potentially they re-discovered functional programming through Linq and delegates:
https://en.m.wikipedia.org/wiki/Functional_programming

It is a useful paradigm for some applications but very well researched and definitely not a new discovery

The Lisp language is like the crab of the programming world.

At a certain level, every language evolves into it

[compact ingot]

depends on what your goals are. if you want to minimize complexity, everything is better that doesn't create a state machine. declarative programming's idea is that if you declare the desired outcome precisely enough, the solution of how to get there is obvious and deterministic. Meaning you don't need to write procedrual logic as everything that needs to happen is defined as necessary by declaring what solution you want. Note a declared solution is a function, not a value.

[peak badge]

There seems to be a bug in Unity's render pipeline where floor of negative 0 becomes -1 instead of 0. I think.

[compact ingot]

the floor of -0.5 is -1

[peak badge]

not -0.5. Floor of -0

[compact ingot]

same

[peak badge]

https://en.wikipedia.org/wiki/Signed_zero
it's not the same

-0 and +0 are supposed to be equivalent, which means floor(0) and floor(-0) should also be the same

[compact ingot]

its only a bug if what you see is different from what the language/API spec says that it should be. wikipedia has no authority over unity or C#

[peak badge]

I see. So you are saying that Unity spec is that floor(-0)==-1. Then which function would you use to first turn -0 to 0?

[compact ingot]

can you show the code that exhibits this behaviour?

[compact ingot]

i would assume that unity follows IEEE 754 and does its own thing only in edge cases where IEEE doesn't specify a behaviour, if you are talking about math or mathF libraries those should match the specifcation of hlsl.

[plain abyss]

I'm pretty sure C#'s float doesn't have signed zeroes

What context are you running into this from?

[peak badge]

const float x_y_factor = 1.5;// Normal hexes are sqrt3.
int xscale = resolution * 6;
int yscale = resolution * 2;
float x = uv.x * xscale;
float y = (1 - uv.y) * yscale;

float t = x_y_factor * x; // scaled y, plus phase
float temp1 = floor(t + y); // (y+x) diagonal, this calc needs floor
float temp2 = (t - y + 1); // (y-x) diagonal, no floor needed
float temp3 = (2 * y ); // scaled horizontal, no floor needed, needs +1 to get correct phase
float qf = (temp1 + temp3) / 3.0; // pseudo x with fraction
float rf = (temp1 + temp2) / 3.0; // pseudo y with fraction
int q = floor(qf); // pseudo x, quantized and thus requires floor
int r = floor(rf); // pseudo y, quantized and thus requires floor

int pX = q;
int pY = q - r;
int temp = floor(pY / float(resolution));
pY -= temp * resolution;
pX += temp * resolution;
if (pX < 0 || pX >= 3*resolution)
    return; // This exit condition is expected to be taken

This is my HLSL shader

[plain abyss]

Ah, shaders. not my division

Someone else can handle this one

[peak badge]

Notice in the bottom left corner, the pixels flicker on and off. I think this is likely due to Unity not consistently giving a 0 or -0 (even while paused).

[compact ingot]

are you making sure that your 0 is actually 0 and not 0 - float.epsilon?

[sage radish]

Unity shaders are just run through whichever graphics library you are using, like DirectX 11 or Vulkan. Unity doesn't implement the floor function or float operations in shaders, so this cannot be a Unity issue.

[untold moth]

Note that if you're relying on this kind of behavior(signed zero), you might get different results on different platforms too.

I also think that it might just be a floating point error issue.

[peak badge]

I actually want floor(0)==0. However unexpectedly, sometimes it has -0 for the same pixel, and this seems to result in floor(-0)==-1

[untold moth]

Did you actually check it with some kind of shader debugger?

I really doubt a signed zero is gonna appear somewhere naturally without you explicitly using it.

[peak badge]

I'm surprised that the pixels change values even while the game is paused. So unity is doing something unexpected

[teal escarp]

by "game is paused" do you mean "Time.timeScale == 0"? Because that doesn't affect everything 🙂

[untold moth]

Paused how? If it's still rendering every frame, floating error is very much possible.

[plain abyss]

If it's a shader issue, it's going to occur even in the editor, so it'll definitely occur in play mode while playback is paused

Shaders are always running, even if the game isn't

[untold moth]

Well, that doesn't stop rendering.

[peak badge]

Correct, but importantly, the camera isn't moving etc. So any rays that are being cast, I would assume would have the same floating point rounding error each frame

[untold moth]

Every time a shader executes it can get a different value due to a floating point error.

[untold moth]

Not necessarily.

If that was true, z fighting wouldn't exist

[teal escarp]

as I understand it you never get -0 unless you explicitly use it

[untold moth]

Anyways, you'll find all the answers if you just debug your shader.

[plain abyss]

This is probably not going to be something with a quick resolution, there's a whole lot of dark wizardry going on inside of the graphics library and you can't really assume anything to be true. You'd certainly expect something to be the same every frame if nothing moves, but shaders render at the whims of a devious trickster god whose methods are inscrutable and rational behavior is an accident rather than the rule.

[peak badge]

Well, adding a very small epsilon does fix the issue, but I was concerned this just means I'll find a different test cases where the issue reappears (which is now 1 epsilon shifted to the right).

int temp = floor(pY / float(resolution) + 0.00000001f); this seems to make it go away, at this resolution

[untold moth]

Again, if you're concerned, you should debug the shader. Perhaps there's an underlying issue that could lead to other problems.

[peak badge]

I have never debugged a shader at this level of detail. What would you suggest?

[untold moth]

On windows PIX or probably Nvidia Nsight(or one of the related tools). AMD has a tool as well I think. Other platforms have other tools. For me PIX is the most straightforward, though it might need you to use d3d12 as graphics API.

[untold moth]

You can debug it almost as any other code(with some shader and GPU peculiarities) - you can see the exact values of the variables at each step.

You'll probably need to research how to output the shader debug symbols in unity.

[peak badge]

I see, it doesn't look too hard, and I have NSight installed. However maybe I will do the practical thing and add an epsilon and move on.

Hmm, yeah I will have to debug it. Some of my known error states are fixed by the epsilon, but others aren't. I guess at least I know how to proceed, thanks.

[untold moth]

Tell us what it was if you figure it out

[peak badge]

int temp = floor ( pY / float ( resolution) ) ;

I used the debugger (and by the way, the hot-reloading of NSight is so much faster than trying to debug within Unity itself).

The issue seems to be this line. At resolution 33, for example, when pY==33, this floating point divide sometimes is 0 and sometimes is 1. Same thing for resolution 74, etc. As we all know, 1/33 is a repeating decimal, as is 1/74.

I'll take some recommendations of ways to make this more robust to floating point rounding issues. I'm trying to find how many multiples of resolution is in pY, including negative multiples (which is why I'm using floor).

[compact ingot]

make it work with absolute values only, later add the sign based on input values. 1/3 is not an irrational number in float representation. 1.0/3.0 ~ = 0.33333334f but i don't se how that matters here unless you add these up and try to get to 1... where you could again do some rounding.

[fathom geode]

How can I make the rotation of the procedural spider to be more natural? I'm currently using Quaternion.RotateTowards for the rotations. There's alsoa bug where the spider rotates inside a wall when the horizontal ray collides with a wall.

    {
        var targetRotation = Quaternion.LookRotation(Vector3.forward, averageNormal);
        transform.rotation = Quaternion.RotateTowards(transform.rotation,targetRotation,delta*rotationSpeed);
        rb.position = getAverage(stepTarget1.legPos, stepTarget2.legPos) + (Vector2)transform.up * bodyOffset;
    }```

private bool setStepWall(float delta)
    {
        if (pb.xInput != 0 && Mathf.Sign(pb.xInput) == wallRayDir)
        {
            var hit = Movement.castRay(castPos.position, wallRayDir * pb.transform.right, pb.wallLegRayLength, Color.red, pb.level);
            if (checkHit(hit))
            {
                 var targetRotation = Quaternion.LookRotation(Vector3.forward, hit.normal);
                pb. transform.rotation = Quaternion.RotateTowards(pb.transform.rotation,targetRotation,delta*pb.rotationSpeed);
                return true;
            }
        }
        return false;
    }```

[hybrid tundra]

Is there a way to make Unity to make a build for PC, when it was cloned from Perforce?
I have enabled Work offline.
No build folder checked out, no Library or anything like this checked out, but still it wants some files badly to checkout to make the build, but it can't since I'm not connecting to the server and build fails for this reason...
If I change the Project Settings - Version Control from Perforce to Hidden or Visible Meta Files, that doesn't solve the issue...

[sly grove]

What files is it complaining about?

Making a build is the same regardless of how you downloaded the project.

[peak badge]

I followed this idea, and came up with

    int temp = (pY + resolution * 10) / int(resolution)-10;

This feels like a very strange way of doing things, but it avoids the issues relating to the floor function with floating point discrepancies

10 resolution widths is well outside the visible screen, so I guess it's "fixed" now.

[hybrid tundra]

Assets/StreamingAssets/addressable_keys.json

[peak badge]

The problem is floor(int(3)/float(3)) sometimes results in 1 and sometimes 0, due to whether it feels like being 0.3333333f or 0.33333334f. I still don't know why the Unity renderer is not consistent about which one it is.

[sly grove]

What about that file?
Are you seeing a specific error related to it? When and where do you see the error and what does it say?

[hybrid tundra]

Version Control: Cannot check out files in offline mode
Assets/StreamingAssets/addressable_keys.json
UnityEngine.StackTraceUtility:ExtractStackTrace ()

[sly grove]

You're in offline mode apparently

[hybrid tundra]

I am

[plain abyss]

Well that would certainly explain why it says you're in offline mode

[hybrid tundra]

it was never a question that I'm in offline mode

[sly grove]

Well you need to check out all the essential files in the project before you can build.

And to do that you need to be online

[hybrid tundra]

well, I can only be online, when I actually work in the workplace

[sly grove]

Next time you're working in the workplace make sure you download everything you need for a build.

TBH I'm a little confused because is that Unity Version Control or is it Perforce?

I'm used to git where you have the whole project downloaded all the time

[hybrid tundra]

? everything is downloaded for build

its Perforce

this keys file gets generated in this way: PreBuildJSON:OnPreprocessBuild

AddressableKeysTool:GenerateJSON ()

AddressableKeysTool:Export ()

so its not something which is missing

[sly grove]

That sounds like a Unity Version Error, no?

[sly grove]

This

Are you using the Provider API in those scripts?

[hybrid tundra]

I don't know what is the Provider API, I didn't wrote those stuff

[sly grove]

Well maybe whoever wrote those scripts should be debugging them, or you should ask them for help debugging them.

[hybrid tundra]

well, those scripts have no problems other than that these make issues when its used away from Perforce server

[sly grove]

Yeah and?

My car doesn't have problems unless I drive it in the swamp.

[untold moth]

How do you calculate py? What makes you think it's the unity renderer?

[peak badge]

It only depends on the UV coordinate sent by the renderer, and the Resolution which is constant

[untold moth]

Is pv == uvs?
It would have nothing to do with the unity renderer. The uvs would be constant on the CPU side, on the GPU they pass through the vertex shader and get interpolated to each pixel. This has absolutely nothing to do with unity and is a quirk of the rendering process on the GPU.

You could try adding (0.5 / texture res) so that your uvs are in the middle of a sampled pixel.

This way the floating point error is not gonna matter.

Unless using a very huge texture.

The issue is that you're basically checking the corner of a pixel. You'd totally get into the same issue on CPU and in any other language or scenario that uses floats.

[peak badge]

I think you're misunderstanding the problem. The code is a conversion from [0..1] UV space to integer coordinate Hex space. https://www.redblobgames.com/grids/hexagons/implementation.html for a detailed description, and code snippets.

Adding 0.5 to the UV coordinate just shifts the problem that much to the right. The problem is not the relationship between UV and hex coordinate, the problem is the hex coordinate generated when the pY is equal to the Resolution (at which point, I want to wrap it by -Resolution, and sample a second texture). The wrapping is inconsistent, because sometimes when pY==Resolution, then int(Res)/float(Res) == 0 or 1, depending on some kind of quirk.

The hexagon's pY is correct, the problem is the float division to determine if we're ready to wrap. And only when the Resolution would create a certain kind of repeating decimal (1/33) (1/72), etc. The fix of shifting all the numbers positive, and doing integer division instead of float division, removes the quirky behavior.

pY is an integer, exactly equal to Resolution (33 in this case).

so it's 33/float(33) which produces the occilating result randomly each frame.

Since 33/float(33) has no variables, you should expect it would always produce the same result. It doesn't, which implies perhaps something about how this specific shader compiles, such that something like a carry bit isn't reset

if that was the case, then you'd be right that it's probably not Unity's fault, though I can't exactly figure out why it happens.

Unity controls the only variable: the UV coordinate

I don't think the UV coordinate has anything to do with the problem, and it's related to float precision.

[untold moth]

Is it actually 33? An integer? Is the type an integer? Why did you mention 0.3333 before then?

[untold moth]

And I'm telling you that you can solve that problem by adding half the pixel width. This would not be enough to shift to the next pixel, but enough to offset any floating point error.

Adding half pixel size in this kind of scenarios is a very common practice in shader code.

[peak badge]

Yup, I'll show you the relevant code:

float q = floor(qf); // pseudo x, quantized and thus requires floor
float r = floor(rf); // pseudo y, quantized and thus requires floor
    
int pX = q;
int pY = q - r;

int temp = floor(pY / float(resolution));

temp will get how many times resolution goes into pY. This tells us which texture to pull from (the 0th texture, 1st texture, etc).
You will notice q and r are floats, however they are quantized to an integer.

Adding 0.5 before quantization will only mean that the problem appears 0.5 cell units to the right. It still gets quantized, and thus would have the same problem.

remember, it's not a single pixel which has the problem, it's the entire cell, which is several pixels wide

[untold moth]

not 0.5. Please don't distort my words. Half of the pixel resolution.

[peak badge]

The output texture is 1920x1080 (it is just a rectangle which takes up the full screen). So say that I bump it by approximately 1/resolution in each direction.

[untold moth]

float pYMid = pY + (0.5 / resolution);
int temp = floor(pYMid / float(resolution));

[peak badge]

resolution in this case is the cell resolution, rather than the screen size. The Hex resolution is about 198x66

[untold moth]

I was never talking about screen resolution. It's the resolution of whatever discreet scale you're working on.

Anyways, try it out and see if it stops the jittering.

[peak badge]

This solution seems to also work

[untold moth]

Great. This is one way people deal with this kind of issues in shaders. Whether you do this or something else is up to you though.

[peak badge]

Yeah now I have 2 solutions. I will benchmark for which is faster

It was kind of a surprising sidetrack. I usually tested on resolution==100 but I tried resolution==300 which had this issue, as it does not represent well as 1/300.0 -> 0.0333334f (or maybe 0.033333f, randomly)

[peak badge]

It's very nearly identical, however adding 0.5 of the width is 1 instruction less. I think it's cleaner, I will use that.

Also big thanks goes to Red Blob Games for their great guide on hex shaders.

[tawny otter]

Anyone know if there is a way to inject an async operation at the front of the queue? Addressables.UnloadSceneAsync to be exact. In our project we need to preload levels with Addressables.LoadSceneAsync(..., activateOnLoad:false) and are running into the issue mentioned in the docs that this blocks the whole async queue until the levels are activated.

[old swallow]

How do I use resolveTarget and RenderBufferStoreAction.Resolve to resolve to frame buffer?

I am using RenderContext.BeginRenderPass(...) path to render eficient deferred shading with MSAA support, but currently it only works when rendering to textures with and without MSAA and frame buffer with no MSAA. I see that AttachmentDescriptor has a separate input for loadStoreTarget and resolveTarget. Could this mean that I render to memoryless buffers with MSAA (like my world normals and other gbuffers) and then resolve to the frame buffer?

I tried to set resolveTarget to BuiltinRenderTextureType.CameraTarget but just got warning "Ignoring color surface store action as it is memoryless". After making a temporary RT and plugging it into loadStoreTarget its rendering again, but only to the RT. No resolve is happening!

Is there any documentation on how to use RenderBufferStoreAction.Resolve or any examples???

[hybrid tundra]

I have checked that, but it didn't solve the optimization issues at all, unfortunately

the same textures are leaking

[peak badge]

Ondestroy: if not null, object.remove

I forget the exact function calls

If you don't, then you will leak GPU memory

[old swallow]

Usually its Dispose or Release

one of them is a call to the other

check the source its on the C# side

[hybrid tundra]

but this is what I wrote

[untold moth]

Then repeat the investigation. Does the manager still leak? If not, then there must be other issues.

I mean, it should be obvious at this point - rinse and repeat.

I'd also confirm with a debugger that OnDisable is actually being called in your scenario.

[coral schooner]

Question about Unity.Properties API with Code generation.

I noticed that the generated code for anything with an Attribute uses reflection to add the custom attributes in the property class contructor. But since public fields don't need any attributes to generate the property bag it avoids the reflection call.

However, properties do require the [CreateProperty] attribute, which then triggers the reflection call to add this as a custom attribute.

Is there any way to avoid this or should I just not care about the performance of these one time only reflection calls?

[hybrid tundra]

its still leaking

this is what I'm saying

if it would have been solved in this way, then I would not have any questions related to this

but it was not solved with this method I mentioned

but its 2 am here, I just woke up for a sec

[untold moth]

You mentioned textures and nothing about the manager object.
If it still leaks, then it's probably not collected by the GC properly.
Did actually confirm that OnDisable is getting called when you transition back to the menu scene?

[hybrid tundra]

so going back to sleep

[hybrid tundra]

its 99% gets called, because the manager gets destroyed

when going back to main menu

[untold moth]

This is an assumption. Place a breakpoint in the code and confirm with a debugger.

[hybrid tundra]

okay, I will check tomorrow

but please think for now that it gets called

because I have very limited time to deal with this issue

but its needed

so I would be happy i you would still provide me some suggestions

which I can test out during working hours

going back to sleep

[untold moth]

If you can confirm that it's still in the memory despite being set to null, then either something else holds a reference to it, or the gc didn't collect it for some reason. You could try forcing a GC collection manually and see if that helps

[thin mesa]

!collab 👇

[thorn flint]

:loudspeaker: Collaborating and Job Posting

We do not accept job or collab posts on Discord.
Please, use Discussions to promote yourself as job-seeking, advertise commercial job offers, or look for non-commercial projects to participate in:
• Collaboration & Jobs

[thin mesa]

make a game wit me
that's literally collaboration

[untold moth]

This is not a social chat server for making friends.

We're all enemies here.

[hard zephyr]

very cool

@thin mesa it dose not say that man

[thin mesa]

it literally did until you edited it. and even still, it currently says "make a game for me" which is still a collaboration request. take it to the discussions site like the rules state.

[hard zephyr]

yeah yeah save it for the judge

[hard zephyr]

thx for this tho

[hybrid tundra]

I have confirmed it, ProcGenManager's instance gets setted to null

[hybrid tundra]

@untold moth maybe this one particular thing was solved, seems to be missing from the current references

[untold moth]

Then continue investigating. Check the next last thing in the reference chain.

[hybrid tundra]

tried to set the TileConfig to null in MultiplayerManager

so far it didn't get removed

but now I make a new build with Resources.UnloadUnusedAssets added after it

[untold moth]

It has something that references it. Don't you see that there's something deeper in the hierarchy?

[hybrid tundra]

that was referenced from SaveDataManager, but that can't be cleared up, because that's existing in DontDestroyOnLoad

I think only the MultiplayerGame is the point where I can do something

[exotic trout]

If it's referenced in SaveDataManager then it needs to be cleared up if you want to clear it up, no?

[hybrid tundra]

well, SaveDataManager is existing all the time. And it should exist all the time

I still don't really get it then

[exotic trout]

Sure but before even getting to that problem, I've been lurking but wasn't this long convo about you not being able to trace down where its still being referenced?

if its being referenced there that's that question solved right?

[hybrid tundra]

I mean its referenced from SaveDataManager, but that should be referenced from there I think (not sure)

because SaveDataManager needs to save data from time to time

so the only thing which seems problematic to me is that the MultiplayerGame is existing only in gameplay

but not in main menu

so when we are in main menu, only the things referred by the MultiplayerGame should be cleared up

and TileConfig has only 1 reference

which is coming from MultiplayerGame

so MultiplayerGame should clear up TileConfig when MultiplayerGame itself gets destroyed

[exotic trout]

but if its referenced in savedatamanager, it should be in memory and therefor shouldn't be cleaned up?

[hybrid tundra]

MultiplayerGame itself could stay referenced by SaveDataManager, because that doesn't contain anything else related to textures, only TileConfig

[hybrid tundra]

I mean MultiplayerGame itself is just a script, its not a texture

if its referenced, I don't think its a problem

or I still don't understand some basic stuff

[echo coral]

Late to this convo, how much memory is this object using? is it worth fixing this?

[hybrid tundra]

not much, but there are 7500+ textures in the project, and lot of them are stays in the memory after gameplay

[exotic trout]

(heads up that you have been in this convo before :P)

[echo coral]

If its just the c# objects its not that big of a deal

[hybrid tundra]

the hugest textures are japanese and chinese font atlases

but those are not leaking

those are just huge textures constantly inside the game

[echo coral]

tmp?

[hybrid tundra]

yeah

my issue is that the game should reserve around 1.26 gb of memory

and its reserving 2.4 gb

because of its leaking (gameplay memory stays in game after returning to main menu)

[echo coral]

If used by fallback fonts they probably get loaded early on and will remain

Unity will allocate more managed mem but will hardly ever free it even when its empty

[hybrid tundra]

I just mean 2.40 gb is incredibly much more than it should be

and its causing issues

[untold moth]

If you want to cut the reference chain at the TileConfig, then you need to make sure it's not references anywhere else. Is it only referenced in the MultiplayerManager? What reference count does the profiler show for it?

[hybrid tundra]

yeah

when I select TileConfig

it says referenced by 1

MultiplayerGame is a child class, its parent class is referencing to TileConfig, but I've cleared that up by setting it null when MultiplayerGame gets destroyed

[untold moth]

Ok, so did you profile after setting it to null?

[hybrid tundra]

yeah

[fringe yarrow]

I downloaded the Unity Open Project Chop Chop, Upgraded it to 6.1 and I'm getting three errors.

Assets\Shaders\Unused\ScriptableRenderFeatures\DepthNormalsFeature.cs(94,5): error CS0619: 'RenderTargetHandle' is obsolete: 'Deprecated in favor of RTHandle'
Assets\Shaders\Unused\ScriptableRenderFeatures\DepthNormalsFeature.cs(17,17): error CS0619: 'RenderTargetHandle' is obsolete: 'Deprecated in favor of RTHandle'
Assets\Shaders\Unused\ScriptableRenderFeatures\DepthNormalsFeature.cs(31,67): error CS0619: 'RenderTargetHandle' is obsolete: 'Deprecated in favor of RTHandle'

However I have no experience and idea how to fix these errors.
Can I get some pointers on how to get rid of these errors?

[untold moth]

And what's coming up in the hierarchy below it?

[hybrid tundra]

so MultiplayerGame has reference to TileConfig. TileConfig references to RitualTileData. That refers to NetObjectController, that goes to FXCollection and that goes to List<FX>

MultiplayerGame itself is referenced by 9

[untold moth]

Ok, then you need to set it to null in MultiplayerGame.

[hybrid tundra]

already did that

MainTileConfig = null

and Resource.UnloadUnusetAssets()

[untold moth]

Then it didn't get set to null. At least not entirely.

[hybrid tundra]

inside OnDestroy

[untold moth]

You might be having several instances of MultiplayerGame.

Or it's not destroyed

[exotic trout]

(or one instance that remains during the menu?)

[echo coral]

if this is a texture remember that you can Destroy(myTexture) to unload it whenever you wish. The c# texture object will be gc'd later.

[hybrid tundra]

its inside a material

so should I rather modify FX class?

[echo coral]

if you do new Material() you need to destroy those material instances too manually

[hybrid tundra]

nope

[echo coral]

otherwise UnloadUnused should suffice for unity managed material instances

Addressables is different but I dont know if this applies here 🤔

[untold moth]

Maybe it's fine to delete these scripts? Considering they're in "unused" folder.

[untold moth]

They're trying to get rid of the managed objects. These wouldn't get unloaded, since they're referenced.

[exotic trout]

Might be worth confirming this is running (and on all the ones that exist?)

[echo coral]

and this is a seperate issue to their "texture" issue?

[untold moth]

The textures are referenced by managed objects.

[echo coral]

so only the "managed shell" is leaking?

[fringe yarrow]

I'll Comment the code inside the file.

[untold moth]

I don't remember. I think some objects in the chain are not even destroyed due to being referenced by shell objects.

[hybrid tundra]

ok maybe OnDestroy doesn't even get called for some reason

wtf

[exotic trout]

obviously we don't know the ins and outs of your multiplayer setup but a lot of multiplayer setups involve a relevant manager living throughout the main menu too

depending on what MultiplayerGame actually does

[hybrid tundra]

its not living there

I can see it disappearing from Hierarchy

when returning to main menu

[exotic trout]

did it move to another?

[hybrid tundra]

nope

t:MultiplayerGame

and it disappears

[echo coral]

Aha I wasn't aware that UnloadUnusedAssets() examines managed objects so this does matter. TIL 💡

[exotic trout]

based on my understanding of unity, you have said something wrong

[untold moth]

Hierarchy is not everything...

[exotic trout]

it either called ondestroy or it exists somewhere

[untold moth]

It could be a prefab of addressable or something

An asset basically

[exotic trout]

but they saw it leave the hierarchy

[untold moth]

Assets don't appear in the hierarchy

[exotic trout]

yes but

they said ondestroy wasnt called

regardless of the asset

the instance did exist

so it either still does or ondestroy was called, no?

[untold moth]

Perhaps it was never in the hierarchy in the first place. Hard to say for sure without more context

[hybrid tundra]

or I'm maybe attaching to a wrong Unity instance

with VS

or whatever

I have 2 PCs and 2 Unity editors in front of me

so its quite a complex setup

let me test it

[exotic trout]

thats fair, just based on what you've told us though seems like theres conflicting information. if you figure out which part is wrong that will likely help

[hybrid tundra]

okay

so OnDestroy gets called

but not all of its lines

[untold moth]

Is it an exception/error?

And btw, I don't think it's wise to call unload unused assets in OnDestroy

[exotic trout]

I guess this is more a question for me than Tamas but in theory if this gets destroyed due to a scene unloading, the asset cleanup should probably run after all that scene stuff is done right?

I know the cleanup happens mostly on demand but surely it tries to clean up after a scene unload

[hybrid tundra]

yeap

attempting to use an invalid operation handle

because I call base.OnDestroy

[untold moth]

It depends if we're talking about managed or native side. I think native might be released separately.

[hybrid tundra]

that can be because I have added a lot of Addressables.Release previously, but probably placed some of them wrongly

because addressables assets are managed by at least 30-40 scripts

but that issue is independent from the texture issue staying in the memory now (I think)

[untold moth]

I mean the logic stays the same: investigate in the profiler - it tells you why it's in memory. Fix the issue, test and repeat if it still happens.

[hybrid tundra]

that's okay, I just thought it was setted to null, but still stayed in memory

but it seems it was not setted to null

the other one was setted (the ProcGenManager)

[hybrid tundra]

okay, that reference disappeared

but that was not much improvement

can't I just clear up the ParticleSystemRenderer's own material and sharedmaterial references correctly?

because setting those to null didn't make it disappear

[untold moth]

Check the profiler. They're either referenced by something or set to ddol.

[hybrid tundra]

why its still not enough to get rid of the FX reference?

[untold moth]

Because you need to read what the profiler is telling you...

[hybrid tundra]

so it seems to be that I still don't understand the profiler?

what the profiler says?

[untold moth]

Didn't we go over these many times by now?

[hybrid tundra]

its so not logical to me honestly

the whole References window

[untold moth]

Start by reading the details/info below the hierarchy

[hybrid tundra]

I don't get it... M_fx_ParticleSet_01_apb is a material. Its referenced by 9.
Checking those 9, these are 9 particlesystemrenderers...
those are ALL coming from 1 single place, which is the FX script.

what else should I see here?

[exotic trout]

(also just because it was relevant last time, did you confirm this code is running?)

[hybrid tundra]

yes, its running

because I have tested modifying this script weeks ago

this was the first thing I did

[exotic trout]

i dont understand why that reasoning confirms its running

(also i know your just double downing on trying to destroy things as much as possible but if its null before you destroy it the destroy is def. not doing anything)

[hybrid tundra]

if it would be running, then I would not have wasted literally entire weeks to solve this issue

I'm panicking, I'm stressful, I just want to finish this

[hybrid tundra]

honestly I don't even know what else should I do

I'm running out of ideas

[exotic trout]

you could try destroying the materials before setting them to null

[hybrid tundra]

ok, I will try that as well

[untold moth]

If they are coming from the script, they are referenced in one of it's fields

[hybrid tundra]

that's pretty obvious btw

[untold moth]

And I don't see you setting these fields to null anywhere

[hybrid tundra]

you think that I try to solve stuff which is pretty obvious

but this issue is not obvious at all

[hybrid tundra]

@untold moth

this is where I set those fields to null

well not literally

but indirectly

is that matter if I reach the same materials indirectly or not?

getting components or using the already popupated fields should be the same

[untold moth]

None of this code sets fields to null...

[hybrid tundra]

there are fields which I don't use there, yeah, but why would I need those in here?

[untold moth]

Looking at the hierarchy, there seems to be an array of renderers.

[dusty wigeon]

What is written here:

[hybrid tundra]

@untold moth those ParticleSystemRenderers are exactly the same (receiving from GetComponent) which are already cached in the script

[untold moth]

No it's not. You're just creating a new array when you get components.

There is an array field that holds a reference to these.

[dusty wigeon]

Look if any of those flags are set on reference of the object.

[hybrid tundra]

but I'm manipulating the exact same ParticleSystemRenderers

[untold moth]

That's not the point though

[hybrid tundra]

there is, yeah

[untold moth]

The point is that they can't get collected by the GC, while still referenced

[hybrid tundra]

so if two different variable refers to the same component, and only one of them sets stored material to null, then it won't get cleaned?

[exotic trout]

But i don't think he's trying to destroy the particle systems, just detatch the asset from the systems?

[dusty wigeon]

Texture usually does not needs to be collected by the GC to be freed because they are Graphics object. They needs to be released.

[untold moth]

No. And not just a variable, but a class/struct field.

[hybrid tundra]

I give back my degree

[exotic trout]

Why?

[unborn relic]

Hello,

I am using unity 6000 and terrain trees and infinite terrain.
I want to be able to add monobehaviors to my trees to be able to keep track of the tree's type, hp, state, make it fall, add rigidbody etc.

I know terrain tree's can't have monobehaviours, so what can I do?

My current thought is I need to have a tree manager that subscribes to OnTileGenerated, OnTileEnabled, OnTileDisabled and keep track of every tree instance based on location. I also need to switch terrain trees with prefab trees close to the player (5 meters - ish). Is this doable? WHat are better alternatives?

[hybrid tundra]

is it better?

now its a field

(but I'm still referring to the exact same components)

[untold moth]

No. You need to do array[I] = null obviously

[exotic trout]

but he's not trying to clear that?

[untold moth]

I mean, local fields count as well, but they don't live outside their scope.

[untold moth]

The GC won't collect the managed objects while they're still referenced.

[hybrid tundra]

why is that better than this?

[dusty wigeon]

You might want to try #archived-code-general

[hybrid tundra]

its pretty advanced topic

[untold moth]

Because all you do is assign null to a local variable.

[hybrid tundra]

not really trivial

[dusty wigeon]

You are flooding the channel

[exotic trout]

Yeah I thought he was trying to clear out the textures and materials no? he's not trying to get the particlesystems cleaned up

given thats a instanced component that can't be something to wanna clean up regardless?

if im wrong apologies i wont pretend im an expert by any means

[dusty wigeon]

Not sure in which situation you would want to keep a particle system with no material assigned.

[exotic trout]

I mean sure but that's not why dlich wanted to ensure they would be destroyed

[untold moth]

Textures are also linked to managed objects. I'm not entirely sure what's going on there, but for now it's clear that they're being part of a reference chain.

Anyways, looking at the profiler details would reveal more.

[exotic trout]

Fair I suppose, though in this case I think Tamas got confused because you advice was on how to destroy the particlesystems correctly and they thought it was critique on destroying those materials and textures better(?)

hope i didn't come off as trying to correct or anything just curious myself tbh 😄

[hybrid tundra]

weird

so this is why @untold moth said to manipulate them via the fields, not via the local variables

because apparently its very important not to create temporary variables for this

[untold moth]

I mean, this is basic level programming. Asigning a reference to a local variable is not gonna change the value of a field

[hybrid tundra]

pretty advanced though

I'm a programmer since 2009

received my degree in 2012

and didn't know about this

I've only used c# though

never learned c++

only in a hobbyist level

[untold moth]

What value is a gonna have in the end of the method?

private int a = 1;

void Method()
{
  int b = a;
  b = 0;
}

This is C#

[exotic trout]

dlich what reference did you want tamas to try and clear/unreference/destroy?

[hybrid tundra]

value of what?

[untold moth]

the on'es in the array

[exotic trout]

the particlesystemrenderers right

[untold moth]

what value is a gonna have?

[untold moth]

Yep.

[hybrid tundra]

one

because its a value type field

not a reference type

[exotic trout]

Tamas are you aware dlich is suggesting you destroy the particlesystemrenderers themselves and not the values inside of them

[untold moth]

The same happend in your code

[untold moth]

Reference would behave the same way

[hybrid tundra]

well, material and texture seems reference types

this is why I didn't understand

[untold moth]

The same behavior is true with refereneces

private MonoBehaviour a;//assigned somewhere else, not null

void Method()
{
  MonoBehaviour b = a;
  b = null;
}

same question

[hybrid tundra]

then a is also null

after the method

[untold moth]

no

[hybrid tundra]

why not?

its a reference

[untold moth]

References don't have sucha a behaviour

You just assign null to a local variable

[hybrid tundra]

I would be surprised

[untold moth]

Then try it out

Try it out:
https://dotnetfiddle.net/cTIWGZ

See what the console prints

[hybrid tundra]

aren't references memory addresses?

or pointing to memory addresses?

if I create a List<int>

and I give that to a function

[untold moth]

Sort of. But that doesn't change how asignment works.

[hybrid tundra]

and I manipulate that inside

then it will still manipulate the same list

I gave to method

well, I would be very much embarrassed, if I would be a programmer not knowing even the basic stuff

[untold moth]

When you assign a reference to another variable, it's as if you copy the address. So setting it to null just removes the address in that new variable.

[untold moth]

You're mixing up different concepts here.

[hybrid tundra]

both are references

and both are assignments

[untold moth]

When you change values in a list, you modify the contents of the list, not the reference to the list.

[hybrid tundra]

but I refer to the same list

the method's local variable gets assigned by the variable outside

[untold moth]

When you asign null or a reference to a reference type variable, you say that this variable would point to that object. If you have 2 variables, you can set both to point to the same object or to different objects. Modifying the value of these variables doesn't affect the object itself. Or the other variable.

[hybrid tundra]

same with List<int> a=b;

[untold moth]

This is same, but if you set a = null, that's not gonna affect list b. It's still gonna reference the same list as before.

Think of reference variables as just an in64 that contains an address(oversimplification) to the actual object in memory. All the same rules would apply to them as to a value type in this regard.

[hybrid tundra]

@untold moth so to understand, should I set ALL particlesystem references to null in the FX script?

[untold moth]

All the class(or struct if you have) fields, yes.

[hybrid tundra]

I'm too dumb for this task apparently, because its still can be seen in Memory Profiler

[untold moth]

Read the profiler data. Compare to before. Do you see a decrease in references count? What other metrics are saying? Read it all.

[hybrid tundra]

I don't even understand why there are 3 different "Referenced by" window for the same texture

[untold moth]

3 different windows?🤔 Not sure what you mean.

[hybrid tundra]

if I select the top T_fx_particleset_01, then it says a material and an assetbundle.
the middle one shows the same
the third one shows two materials

[untold moth]

Then perhaps you have 3 copies of this texture in the memory.

I don't know. It's hard to say anything without seeing what exactly you mean.

[untold moth]

Yeah, these are 3 separate objects in memory.

You can even see that one of them has smaller memory footprint. Maybe a mip level or just a lower res texture?

[hybrid tundra]

okay, I think in this way I won't be able to solve this task

I'm thinking of searching or making an example test project

because its unproductive to wait 20 min for every change I make

(building takes like 20 minutes even with incremental build)

even the scripts are multiple thousand lines long

in this project

so its even hard to find the fields

[untold moth]

They're usually at the top. But you can also view all the class members in visual studio(or some other ide).

[hybrid tundra]

I mean if it takes me 1+ week to maybe decrease 1-2 MB of memory and its still like 1 gb I would need to free up, this whole thing is not effective

I'm not effective in this task

until now I just wasted time with this

maybe I will see something in a test project

[untold moth]

This is why you need to investigate properly and find the root cause. There's likely only a few lines of code or assets that are responsible for all of this.

[hybrid tundra]

yeah, I know FX class is responsible for a big chunk of memory

and it seemed to be that some addressable bundles as well

but no idea which one is reserved, because Memory Profiler only tells me some random character named assetbundles

and I can't identify which one is that

because its not the real name

[untold moth]

I don't know a lot about addressable so can't say anything about it, but you can usually make some assumptions and infer what asset it corresponds to from other data(like assets that it references/includes and stuff).

[dusty wigeon]

You potentially have duplicated assets(due to asset bundles) if you are not expecting to have 3 of it.

[hybrid tundra]

I'm not expecting anything

I have expected to at least reduce the number of references, but so far I couldn't achieve it

[dusty wigeon]

Do you have multiple texture called T_fx_Particles_01 in the project ?

[hybrid tundra]

because apparently reference counts didn't even decrease

[hybrid tundra]

nope

[dusty wigeon]

Do you use Addressable

[untold moth]

Reference counts of what?

[hybrid tundra]

yes

[hybrid tundra]

anything

honestly

this project has so much textures and references

[dusty wigeon]

Have you look at the Addressable Report

[hybrid tundra]

I would have been happy if I could have reduced at least a few

[untold moth]

Did you not manage to remove the manager thing from memory yesterday?

[hybrid tundra]

I could remove that, but that was like 1 out of the 10 thousand references

or even more

[dusty wigeon]

What is the status here after your build

[untold moth]

ok, so the "I couldn't achieve it" is totally wrong.

[hybrid tundra]

I'm not happy with my own results, that's what I wanted to say

and deadline is close

and probably there will be other tasks unrelated to this one

anyway, making a test project now

hopefully I will understand something from that

[untold moth]

Then focus on understanding the issue. Blindly testing different things is just gonna be a waste of time.

[exotic trout]

Not the best solution and I know its roughly 1ish gb of potentially savable memory but is this a make or break problem for you considering your deadline?

If it is fair but if theres bigger tasks for your deadline than reducing the ram might be worth considering

[hybrid tundra]

yeah, the last 5-6 hours were very effective in understanding the issue

as you could see

anyway, don't want to spam the channel

[untold moth]

I can't see. I only have seen that you had a misunderstanding about a very basic thing. I didn't see any effort trying to read the info in the profiler or track the root cause.

[dusty wigeon]

You should really look into addressable duplicated. It has the potentially to save you from a shit load ton of memory with minimal work.

[hybrid tundra]

even for addressables I would need a test project, because I have worked with addressables million years ago, and don't really remember how they work

but seemed pretty complex

[dusty wigeon]

If you do a seperate project you would not know if you have an issue with duplicated in the current...

[exotic trout]

I mean yeah you are dealing with relatively complex stuff here

[untold moth]

I must correct what I've said about this code. I didn't notice that you're accessing info.particles. I thought you were setting info = null.
However, if info type is a struct, then what I said is still correct to a degree and that wouldn't get rid of the reference.

[echo coral]

Checking for duplicated assets is a good idea, you can use the built in analyser and fixes to quickly fix them:

Addressable build reports also show duplications so you can check that also and focus on duplicated textures (this would ensure it doesnt load duplicates of the same texture into GPU memory!)

[hybrid tundra]

not a struct, its a class, and I know a struct would be a value type

[hybrid tundra]

even if these textures are duplicated, that doesn't matter, because these should not exist at all at this point of the game

the things I mentioned starting with "fx_"

those are gameplay related

and I was in main menu, when I made the memory profiler snapshots

this is why I wanted to just purge somehow out of the references from FX script to at least see those disappaering from the memory profiler

but they're not

[echo coral]

If you say spawn a particle system in your gameplay scene and release it correctly (via addressables) when going back to the main menu scene, the textures will be unloaded for you.

[exotic trout]

dumb question but i don't think you've shown the particlesystemrenderers in inspector yet, that material isn't assigned to ParticleSystemRenderer's .trailMaterial right? you destroyed .material and .sharedMaterial previously and should maybe be destroying the renderers regardless but figured worth asking

[echo coral]

If Resources unload un used is called and this particle system game object was destroyed then we should presume it would have its textures be unloaded right? (Unless asset bundle assets and therefore addressable assets are excluded from this process)

my point being if this still does not unload the textures the solution is correct releasing or destroying all textures manually and praying its fine

[hybrid tundra]

the materials are in Renderer's Material field

[hybrid tundra]

currently I'm not trying to fully unload them, for now I just wanted to check if I could get rid of the FX reference

because there are a lot of materials referenced by FX

getting rid of them probably won't unload the textures, but at least I wouldn't see those million references coming from FX appearing in Memory Profiler

[hybrid tundra]

the FX class still refers to the ParticleSystemRenderers holding the materials

I can't find anything else to set to null there

literally nothing

its unbelievable

[echo coral]

if you destroy the particle system game object it should solve the problem? Its not like this is going to work properly after your current adjustments

[hybrid tundra]

the parent gameobject gets destroyed

that's there in the Hierarchy during gameplay

but when I return to main menu, it disappears

[echo coral]

then some dangling references are preventing the components from being GC'd

[hybrid tundra]

well, currently

https://discussions.unity.com/t/how-to-unload-everything-currently-loaded-by-addressables/843869/7

I have tried this

and even this didn't clean up all assetbundles, just to see if that would solve the issue

(I know it wouldn't be a final solution, I just wanted to see if this would get rid of references)

but even this was not effective

[echo coral]

c# objects leaking is not related to addressable asset releasing

you can correctly unload assets but keeping references to the c# objects will mean they dont get GC'd

[hybrid tundra]

I know, I just have zero idea how to track all references of the multiple hundred textures

if 1-2 texture releasing also takes for hours

its an extremely huge codebase

1-2 script have multiple thousand lines of code

so I really hope I find 1 solution which solves at least half of the memory issues

otherwise I won't be able to finish this optimization in time

I'm pretty sure about it

[echo coral]

to unload a texture just grab them from the materials and Destroy()
It will break shit later as things will presume the texture is loaded still but it will half solve the issue

[hybrid tundra]

what do you mean by "grab them from the materials"

[echo coral]

Your previous messages imply the textures are used mostly by particle systems?

[hybrid tundra]

there are some of them used by particle systems, like maybe a quarter of them, or even more. But I tried to call Destroy on them. I tried to clean up those materials like at least 4-5 way and it just can't die

[echo coral]

you want to destroy the Texture2D objects too

again, this will cause problems later

[hybrid tundra]

if there's a one line solution to destroy Texture2D objects, I will add it

[echo coral]

Destroy(material.mainTexture)

[hybrid tundra]

well, materials are on everything, maybe some of them are assigned from script, some are not, infinite amount of prefabs are in front of me

so this one line would be needed to add to million scripts

infinite amount of work would be to solve this

[echo coral]

If say you did this to spawn them, it would be easy to release them all too:

private async UniTask LoadPrefab()
{
    particleSystemPrefab = await Addressables.LoadAssetAsync<GameObject>("CoolParticle");

    for (int i = 0; i < 5; i++)
    {
        Instantiate(particleSystemPrefab, transform);
    }
}

private void OnDestroy()
{
    Addressables.Release(particleSystemPrefab);
    particleSystemPrefab = null;
}

[hybrid tundra]

I've counted with the Asset window, there's 4400 Mono scripts

in this project

nope, that's just one little script

which spawns some of them

[scenic forge]

There's no magic bullet to memory management, if your existing code has no care in the world about using and instantiating resources, then you are kind of screwed as you have discovered.

[hybrid tundra]

but that's all get released

problem is the million other scripts, which don't manage releasing well

[hybrid tundra]

then I will fail

[echo coral]

My example works easily because addressables only needs the prefab to be released for things to get unloaded

[hybrid tundra]

because in 1 month, I won't be able to track down all 4400 scripts

[echo coral]

that sucks but it wont magically work because you need it too

Perhaps you need to re read how addressable loading/releasing works incase you can infact correct things with less effort

[hybrid tundra]

honestly I'm thinking of asking someone to get assigned to help in the next month, and maybe 2 of us could reduce 5% of leaking

in 1 month

no idea

when I see that 1 material is referenced from 16 different places, scripts, Particle Systems, etc.

and that's just 1 material

and that's a long line of references

because that can go even further

[echo coral]

when talking about addressable assets, what the memory profiler says does not matter

[hybrid tundra]

well, if this can't forcely release all of these, I'm not even sure if everything is held by addressables

https://discussions.unity.com/t/how-to-unload-everything-currently-loaded-by-addressables/843869/7

[echo coral]

As shown in my example, if we load 1 thing, we just release that 1 thing and it will handle the dependency unloading correctly
#archived-code-advanced message

[hybrid tundra]

I called this

and still it didn't clear up nothing

[echo coral]

Okay do you use LoadAssetAsync() to load prefabs? Or do you use InstantiateAsync() ?

[hybrid tundra]

LoadAssetAsync, InstantiateAsync, there are even Utils classes to do these, multiple one, I have counted last week like at least 40 different scripts which are loading addressables in some way

and some of them are monobehaviors, some of them are not

I don't know which one needs to release assets in which moment

or in which method

so its not that easy

if everything would have been in 1-2 scripts, then yeah, it would be very easy

[echo coral]

Then that spaghetti needs to be removed to simplify things so you can actually release things properly

[hybrid tundra]

this is why I'm panicking

[echo coral]

You can get all references to these functions and start reducing or refactoring

[hybrid tundra]

there are textures even from different games in the memory, because of its an inherited code

or legacy code