archived-code-advanced

[undone coral]

you're stil rendering the previous frame when you start the next frame, in unity

[haughty niche]

"the type or namespace 'mirror' could not be found"

[undone coral]

so once you stopped ReadPixels you stopped waiting for the render thread to finish

[haughty niche]

why cant it find mirror anymore?

[undone coral]

not really though - you stopped waiting on the GPU to finish, which happens after render thread's job is done

that's why async gpu readback is talking about "latency"

they mean with respect to which game frame you are on at the moment you receive the callback

when you do async request on game frame 0, by the time the callback is called, you are usually on the main thread executing game thread 1 or 2

then again, if your main thread + render thread + gpu work is altogether less than 16ms, you will get the callback on the same frame.

i am on my macOS laptop so i just can't show you the GPU work bar in the hierarchy

[stuck onyx]

that would be awesome

but dont hurry

ill be here

[undone coral]

but i think you can see now

that suppose render thread was also gpu work

well they overlap

so don't wait for it

why is my scene so slow? i have like a million tris and a complex interaction with shadows

[stuck onyx]

here you mean GPU or CPU?

[undone coral]

render thread is CPU

but

[stuck onyx]

ah sorry

[undone coral]

what it really is is

"all the work here, there is no gameplay (meaning main thread) dependencies on it"

when you do readpixels

[stuck onyx]

aha

[undone coral]

as far as unity is concerned, you are using the texture for gameplay

main thread == gameplay

[stuck onyx]

got it

[undone coral]

part of this is that it was always a bad idea to let people use X which 99% of the time does not impact gameplay but because of the 1% of the 1% that does, it assumes all 100% impacts gameplay

exhibit A: physics

[stuck onyx]

so... basically... unity thought i was going to use that texture... and i was blocking the main thread because of it

because

it thought it was for gameplay

[undone coral]

yes

it assumed that it was

[stuck onyx]

shit but... yeah got it but, damn,

they tell you, " do this inside the lateupdate bla blña bla"

but dont tell you that

[undone coral]

when you do the other api calls they internally create a dependency graph

of operations

called command buffers (really appending to the current frame's command buffer)

that expresses inputs and outputs, like a visual programming tool / shader graph

you don't get access to any of the data on the main thread

that's the point

alright good luck out there

[stuck onyx]

just this last thing

i read and read again, as i said today is not my best day

this api call thing

AsyncGPUReadback.RequestIntoNativeArray

what we did with that is ignore the dependencies youre mentioning?

and simply tell unity... do this and shut the fuck up

is it like that?

[undone coral]

what we did is say we DO have a dependency but not on this frame

so rendering gameplay frame 0 might finish in the middle of computing gameplay frame 1 or 2

async gpu readback is a way to get rendered content of frame 0 in the "middle" of doing the gameplay for frame 2, for example

instead of in gameplay frame 0

because rendering overlaps with gameplay

clearly you still have a dependency - it's just limited to the code in the callback

instead of potentially everything that takes place while doing gameplay

[rugged radish]

I like seeing that "saved by batching" stat 😄

just wanted to confirm one thing, Graphics.RenderMeshInstanced in Update is not an anti-pattern right ?
my other high priority is to cut the gameobject count, even more than optimizing the rendering, that seems to help too

also need to reread your advice on shadows, noticed that there're no shadows being cast by those trees

[undone coral]

but are your trees actually static?

[undone coral]

no

you're adding a command to a big queue

[sly grove]

that's pretty standard

[undone coral]

yeah i meant no, it's not an anti-pattern

it's gucci

[rugged radish]

I'm very new to rendering, afaik trees are animated by vertex shaders and I'm not sure if that is considered static or not

[sly grove]

I was agreeing with you doc 😉

[sly grove]

As long as the GameObject is not moving (the Transform) it's static afaik

[rugged radish]

right now they are static, but I do want to add animation down the line

[tough knoll]

In the context of batching you have to mark it as static

and it doesn't care about vertex shaders

In fact, you could mark it as static and then move it, although the mesh itself won't move with the transform : p

[rugged radish]

still not sure if gpu instancing will fit my case, but it's definitely good information to learn and keep in mind 👍

[tough knoll]

Definitely test

I recently discovered that on older hardware my game is about 10fps slower with GPU instancing enabled : S

[rugged radish]

I'll need to see how fast it works when there're conditions involved
I shouldn't use instanced meshes when the player is nearby, which means I have to filter the array of transform matrices, which may be worse in the end

so yeah, need to benchmark different methods

[undone coral]

is this in Metal?

[tough knoll]

My uneducated guess is that since there's a small initial overhead when rendering GPU instanced objects and we weren't actually drawing that many objects, it came out worse. That probably is irrelevant to you rendering 6 million trees tho

[rugged radish]

metal as in metal workflow ?

[tough knoll]

Also, impostors

for vegetation is definitely an option

If you have like a LOT of it, from far away it probably won't be very noticeable

[strong harbor]

I had 2 questions hoping someone can answer, I have been reading the documentation a ton but cannot seem to find exactly what I am looking for.

1. Is there any way to know on a tilemap what tiles have something placed in them?
2. Is there any way to place tile grouping on the same tilemap

[rugged radish]

I have a lot, but not that much, it was just a technical test lol

[strong harbor]

I can elaborate far better for my exact situation

[undone coral]

metal as in macOS metal

[prime thorn]

can anyone help me i keep getting these errors

Assets\Scripts\GamePlayScripts\CollectableTrashScript.cs(5,14): error CS0101: The namespace '<global namespace>' already contains a definition for 'CollectableTrashScript'
Assets\Scripts\GamePlayScripts\CollectableTrashScript.cs(11,10): error CS0111: Type 'CollectableTrashScript' already defines a member called 'Update' with the same parameter types
Assets\Scripts\GamePlayScripts\CollectableTrashScript.cs(16,18): error CS0111: Type 'CollectableTrashScript' already defines a member called 'OnTriggerEnter2D' with the same parameter types

[undone coral]

it's telling you what hte issue is

#💻┃code-beginner you have duplicated the script or class

[undone coral]

something placed in them
you have to keep track of where your character or props are on a tilemap

[prime thorn]

yeah i just have no clue how to pin point the issue

[undone coral]

Is there any way to place tile grouping on the same tilemap
you can create your own array of indices if you need "groups". not sure what you mean by tile grouping though, maybe that's something specific to tilemaps

[undone coral]

i think if you're at 144 fps you're fine

change nothing

[strong harbor]

its specifically in the sense that, right now the way my system works is that i create these room prefabs that all have their own tile map and renderers, but i was trying to figure out if I can instantiate onto the same tilemap

[strong harbor]

and I see, i might need to figure out a system for trying to remember that then

[undone coral]

i think you can define a brush, but no, the unity built in tilemap isn't powerful enough for something like that out of the box

you can search for an asset store asset that might help you

[strong harbor]

I see I see. Thank you for the help

[gritty nacelle]

I do not know if this is advanced per se, but I'd like to find a random point along the edge of a box. I know I can use something like Random.randomOnUintSphere, but I can't think of the maths to get a distance from the center of a box to that direction intersecting with a face of said box

I guess that would not be exactly a random point though, as it would be more prevalant towards the centers of the faces

[undone coral]

when you say edge of a box

do you mean edge of a 2d rectangle?

edges of a 3d rectangular prism (i.e. scaled cube)...

you can sample a random point on the surface of a 3d shape using its UV mapping

for a 2d shape, you can do something similar using its outline. it's up to you to define the mapping

UVs do not guarantee the likelihood of sampling any point on the surface is the same

in principle, for 3d you can do something of the form

random triangle = float between 0...1 sampling a normalized map of triangle areas to triangles
random point on triangle = random barycentric normalized coordinate to world space

for 2d, you can think of it the same way. sort the line segments into an interval map (interval length is segment length) and sample it using a random number.

@gritty nacelle does that make sense?

[somber tendon]

🤔 random point on an edge of a box?

• have a list of edges (there should be 12, each connect the 2 of the 8 corners of the box)
• get random edge from the list
• interpolate between the edge's corners position using random value

[undone coral]

it's trickier if you want equal density of probability throughout the shape

otherwise the shape might as well not matter

your method will put way more points on the short edge than the long edge of a rectangle

that if you drew a rectangle over a field of random points and counted the ones that fell on every line

[crimson ruin]

How do people do serialization? All the tutorials I've found are super basic, they don't go into serializing object relations. For example if I have a list of GameObjects, I can fairly easily serialize the data in it, but the problem is instantiating the actual prefab for it.

Put everything in a Resource folder & then generate an association to it with its path in an editor script?

[novel plinth]

as you said, you already have a list of gos... then instantiate those? 🙂

I see zero correlation between serialization and instantiation

[somber tendon]

Yeah, I usually go that route, either using prefab name and load only the file with said name, or load all files in specific directory and get prefab with matching ID.
I usually use the later for in game items where I would mosy likely ends up loading multiple items

[crimson ruin]

The problem is that those GOs are created during gameplay, when the scene is loaded during loading they won't be there, so they must first be instantiated before being fed the deserialized data. Unless one wants to serialize all the components and all their associated data, which I don't think really is feasible.

[somber tendon]

you cant serialize gameobject into a plain string and expect that it can be deserialized back into the same gameobject

[crimson ruin]

Cool, then I know that I'm not the only one going down that route

[unkempt nova]

You are. They're doing something different I think, using prefabs

[rugged radish]

I use addressables to instantiate prefabs at runtime
the system has lots of nice things like async load, usage tracking, async loading for groups of resources by tag, and so on

not sure if that was your question though

[hushed fable]

I believe you have to come up with your own ID system to serialize references. All prefabs that you spawn and want to save probably should have some component (or external lookup) that has their prefab ID.

[rugged radish]

it's complicated, my work station is way above the hardware I want to target
so I look at the resource load values and compare to games that have similar hardware requirements
it's a very rough estimate, but I don't have anything better rn

[crimson ruin]

Sweet I'll have a look at that

[crimson ruin]

Yeah I'll go with putting everything in resources & associating their resource path with a component on the prefab such that I can serialize that path & instantiate it again at load time to fill it up with the deserialized data

[unkempt nova]

Also might look into AssetDatabase, think it has a method for creating a prefab

Sounds like it could simplify what you're trying to do maybe

Nvm, PrefabUtility not AssetDatabase: https://docs.unity3d.com/ScriptReference/PrefabUtility.SaveAsPrefabAsset.html

[hushed fable]

Yea my advice was made with the assumption that you would need this in build

Editor should have more options and IDs available

[novel plinth]

Why not have a manager to handle those prefabs, so when the go intantiated, you'd want to keep the stats/state of that go e.g: guid, position, name, rotation etc.. just the stats, nothing else.. and serialize with whatever

Then later on, when loading your save file, you re-create those objects. which a common way to do for save system

you said Unless one wants to serialize all the components and all their associated data , not quite sure what you mean by saving a component.. all and all, just the important bits

[placid shell]

is there anyway to grab an enumerator out of a dll file or should i just manually recreate them in c#?

[novel plinth]

Man you're doing this blindly since yesterday 😫 recreate the enum on c# side, then do marshalling,,,

I bet you wont get this kind of answer here, you'll have much luck on c# discord instead

[unkempt nova]

Speaking of which, any chance I could get a link to that? I figure there are many

[novel plinth]

Marshalling?

[unkempt nova]

The C# discord, sorry

[novel plinth]

Sorry on my phone, dont know how to send invite

[placid shell]

ok thank you so much! Doing stuff blindly is the way i do most of my stuff in unity lol!

[unkempt nova]

Someone else got me, thanks @novel plinth

[novel plinth]

See.. the bot wont allow me to post c# discord link. Lol

[unkempt nova]

Ah yeah forgot it'll eat them

[novel plinth]

C# discord team will do a live conf soon regarding editor tooling for game development.

Im not sure if this is proper to post it here, if not then feel free to delete it

Might be useful to broaden your editor tooling skills

[unkempt nova]

Sounds useful, thanks

[thin mesa]

just FYI, the link to it is pinned in #archived-code-general if you ever need to redirect someone to it again

[stuck onyx]

is someone familiar with this?

the first 2 captures got that weird texture

for them im trying to use this

[unkempt nova]

I wanna say I've seen this before and the solution was clearing the GPU's buffer before changing the data in it

Question seems better suited for #archived-shaders imo

[stuck onyx]

thanks

[valid ore]

Collider[] hitColliders = Physics.OverlapSphere(center, (float)0.1);
how to use it in a thread that is not the main thread?

[mint kraken]

Unity physics doesn't support multithread at all

[valid ore]

And there is no way to trick it?

[sly grove]

you don't need to trick it

they do provide some tools to perform physics queries in other threads. e.g. https://docs.unity3d.com/ScriptReference/SpherecastCommand.html

[valid ore]

I don't understand how to change this Collider[] hitColliders = Physics.OverlapSphere(center, (float)0.1); to a SpherecastCommand

[sly grove]

start by looking at the example in the docs

[valid ore]

I do not understand this anymore

How to fix it?

@sly grove

[rugged radish]

it tells you in the error message

try TempJob

[valid ore]

And how to do it?

[sly grove]

yeah the error tells you exactly what to do

[valid ore]

Now it's like this

What to do?

[hoary pendant]

It says right there

[sly grove]

you should be starting your job from the main thread basically

[hoary pendant]

Main thread and dispose the NativeArray

[valid ore]

How to do that?

[sly grove]

well right now you're running your code in a separate managed thread:

        Thread thread1 = new Thread(checkChunksForever);
        thread1.Start();```

don't do that

[wind gyro]

Hiya, not sure if there's a channel for Unity Cloud Build specifically, so I think this is the next best channel? Is it possible to push a build into a docker repository with a post-build script in UCB?

[stuck onyx]

how to set the device so it pauses when we are looking at the profiler ?

[granite viper]

You can stop recording on the profiler

[stuck onyx]

yeah but i've seen its possible to set it so the device stops

[granite viper]

You can use breakpoints with a debugger

I've not seen pausing the game via the profiler when you're looking at a build tho

[stuck onyx]

yes, theres an option hidden somewhere called run in background that you turn off but i cant find it

but maybe ill do the breakpoint thing

[granite viper]

Hmm, I don't recall that option

[stuck onyx]

saw it on a video and seemed like it was in build settings but now i cant find the video or the option 🤣 anyway its ok

[hallow cove]

can anyone help with Search Windows?

it is very hot here, and I am struggling to find a solution

it has a lot of elements in it, and is very laggy, as the elements inside folders are also added, even if you don't open any of the folders

a solution I found was to open a new window every time you select a "fake" folder

however I have no idea what the best way to design this would be

[stuck onyx]

Is stopWatch a wrong way to measure performance on scripts? im seeing a huge difference of what the StopWatch calculated and what the profiler is showing

[somber swift]

It is good at getting rough idea of performance of certain parts of scripts but its not that good if you want to profile very small period of time or want very accurate results. For example if you want to know which of two methods would be faster, stopwatch should work well enough (if the methods takes small amount of time, just put them inside for loop for better measure)

[stuck onyx]

thanks AleksiH

[half marten]

I'm trying to write an algorithm that computes belief/likelihood based on uncertain evidence, but I'm having some trouble and would appreciate any help.

I have a list of 'event evidence' objects.

• Each of these evidence objects is a struct with a set of attribute variables that describe a specific hypothetical event (like 'time', 'place', 'subject', verb' etc..).
• Any of these attribute variables can be null, which would make the evidence object an incomplete description.
• Evidence objects conflict with one other when they have different (and not null) values for the same attribute. (eg. an evidence that contains place=school would conflict with an evidence that contains place=church).
• Each evidence object also has a 'source' variable that points to the specific person who gave the description, and a trust variable which denotes how much I trust this source.

I want to compute, based only on this list of evidence, the belief that at least one of the event descriptions in the list describes a true event.

• This belief should be higher if multiple different sources give descriptions that align(don't conflict) with each other. But if one source gives multiple aligning descriptions, this shouldn't count more than if they had just given one combined description.
• It's possible that none of the descriptions are true, meaning all the sources mistakenly shared false descriptions, and none of the events happened.
• It's also possible that multiple conflicting descriptions are all true, and refer to multiple different true events that all happened. The list of evidence objects could contain descriptions for any number of true events.

I've tried so many different approaches, but I can't find the right one. I think the answer may be in Bayesian Networks, fuzzy clustering algorithms, or Dempster-Shafer Belief functions. Can anyone help point me in the right direction? Thanks in advance.

[fickle inlet]

I want to create an effect with line renderer like in those fps games where you start throwing a granade and the line of where you are going to drop shows up

I tried with vector3.slerp but curvature direction is different if start position and end positions change

Could you recommend what to use?

I am not asking for the code, but just an input where to start

[sly grove]

you basically need to do the physics simulation. keep track of position and velocity of the imaginary thrown object and loop through each Time.fixedTimeStep and simulate how the position and velocity will change in each step, and save the positions to use for your line renderer

[fickle inlet]

I am not going to use physics in my game

I have preset start and end position

I only need a curvature between start and end points

[sly grove]

if there's no physics then what is causing a curvature

it's just a line right?

[undone coral]

you are looking for a bayesian network

[fickle inlet]

raycast finds end position

and start position is camera

and camera is always above the raycast hit position

[sly grove]

ok so set those as the two points on your LineRenderer, done.

[fickle inlet]

But I need it to be curved

[sly grove]

curved according to what rules

[undone coral]

is it always the first two captures? you might have to specify the graphics formats on both the destination texture and the redner texture. however in URP they should never change between camera renders...

[fickle inlet]

I don't know, could you elaborate more, I probably don't understand something. So you are saying that I need physics to make a curvature? Can't I just create mathematically curvature?

[sly grove]

Physics and math are one and the same

I'm saying you want to draw some curvature

presumably this is in advance of you actually throwing some object in your game right?

[fickle inlet]

Yes

[sly grove]

Is this curvature also what you will use to prescribe the path of that object?

[fickle inlet]

I need a trajectory where it will fall

yes

[sly grove]

Or does that object already have a way of moving a curved path from a to b?

[fickle inlet]

grenade will follow the line

[sly grove]

I mean you could use a bezier curve I guess

just generate a control point, maybe it's just directly above the target position, and use A, B and the control point to generate the curve

Honestly would recommend using a spline library to avoid doing the maths yourself

then you can sample the spline for your line renderer

and also sample it for your grenade path

[fickle inlet]

Bazier curve makes sense

It's also my learning project so I don't mind learning how to do it myself

i'll check spline library too

[half marten]

But how do I structure the network to account for different descriptions and different sources?

[bold slate]

On YouTube, there’s a channel called Tarodev, and he has a great tutorial on how to do this.

[undone coral]

i think if you read about it and look at an example it will be clear

[real mortar]

As a non-artist, I am interested in leveraging asset store equipment meshes (e.g. armor) and programmatically binding the mesh to my character rig by copying over bones, bone weights, poses, etc (in editor or at runtime). Essentially, eliminate the step of importing the armor mesh into Blender, rigging it to my character, and exporting it to Unity. Is this theoretically possible?

[half marten]

I've been reading about it for a while. I think bayesian networks don't work here because they require independent priors, but evidence from the same source is not independent.

[undone coral]

i think you can model "eivdence from the same source" as writing a "source is trustworthy" node and connecting it as part of the joint probability of the evidence nodes

does that make sense

the nodes can only be observations / facts

because they're modeling variables in a joint probability distribution.

so if this is important to your game

you would do something like "the npc is telling the truth" as a descendent of "the npc is trustworthy"

[half marten]

But the likelihood that the npc is telling the truth is also dependent on the testimonies from other sources

[undone coral]

(eg. an evidence that contains place=school would conflict with an evidence that contains place=church).
this is different. a probabilistic logic puzzle like this, as free form as you want, is hard to express. you can look for "probabilistic prolog" on google for something that has a limited form of expression that looks like this, with probabilities

[undone coral]

so... how would you express those testimonies as something that can be true or false?

"the testimony of this npc B is accurate" "this npc B is trustworthy" etc. etc.

[hallow cove]

quick question

I have a type

and I want to create an empty array from it

but I don't know how to do that

[undone coral]

it's up to you how many falsehoods and in what scope affect whether or not an "npc is trustworthy" - all we're doing is exposing a latent variable that happens to fit very neatly into a "if the npc is trustworthy, it has a very very strong impact on the probabilities of its descendents"

[hallow cove]

so I have a generic Type, and I want to make a new array of that Type

[undone coral]

you just gotta be creative with how you express these things

a bayesian net is actually really well suited for something like a common predecessor of many nodes that works like "this source of information is trustworthy"

you're just getting caught up int he meanings of the numbers rather than the nodes

@half marten but like wha'ts the gameplay?

[half marten]

The testimony of this npc B contributes to the belief in the specific statement they're expressing, eg. "Tim congratulated Tom". And NPC c says "Tim congratulated Tom on Monday" but NPC d says "Tim congratulated Tom on Saturday."

[undone coral]

i don't thin this stuff necessarily makes sense

there isn't any complexity here. a player will assume that a soon as an npc tells 1 lie, the npc isn't trustworthy

which is the most natural model

is the game logic puzzles?

is that hte idea?

[half marten]

That doesn't work for what I'm doing

[undone coral]

what is the gameplay

[half marten]

No, not a logic problem

I'm working on an experimental dialogue game. Not even really a game, just an idea

[undone coral]

contributes to the belief
i understand that you want to model it this way

these have to be bools*

i don'; tknow, it's pretty clear how to model what you are giving as an example

as a bayesian network

if you want to query "is npc C trustworthy?"

[half marten]

I'm trying to model belief of past events, based only on testimonies from sources

[undone coral]

which is latent, it is unknown

okay well i think you'll figure this out

the thing you need is a bayesian network

[half marten]

Are you thinking of something like this? https://i.imgur.com/tCpev5J.png

[undone coral]

the nodes are things like "the congratulation happened on a monday (true) or a saturday (false)"

@half marten but it should be obvious that, as soon as you observe that ANY ONE of the descendent facts from an NPC is a lie, then the NPC isn't trustworthy

and sure, yes, that would cause all the other "testimony" to be in doubt

but you don't need bayesian reasoning for that

there isn't going to be a situation where the precedent nodes "npc A is trustworthy" is 50% AND one of the descendent facts is observed

it's going to be 0% or 100% as soon as any descendent fact is observed

[half marten]

But not for my model. I want NPCs to be able to assert statements that might not be true (gossip) and still have their statements lend some credence even if they've lied in the past

[undone coral]

you can use prolog if you want to solve and express that kind of logic puzzle more simply

[undone coral]

you mean you want to show the player this number it calculates

for a probability that their statement is true? and that sometimes that probability is high even though the npc has lied in the past?

no probabilistic framework is going to give you this elegantly because it's a bad idea 🙂

[half marten]

Not the player, other agents

[undone coral]

they might as well play randomly then

if you seed it with garbage data it will act garbage

you'd have to add 100 facts that the npc has told the truth about

then 1 fact it lied about

and then yes, it will give you a nonzero probiblity the next statement is true

but that's so counterintuitive, and also, the agents will behave poorly if they did that

surely you see why - because you're contriving a bad situation for the agent to reason about

essentially with garbage data

anyway you should think about this more

you can do this very very easily

without any probabilistic framework

[half marten]

I think you've misinterpreted the system, and what I'm trying to do

[undone coral]

"if the npc has told more than N lies, not trustworthy."

that's it

[half marten]

You can forget about computing NPC trust, imagine that's a known number

[undone coral]

i think you can just program it to do exactly what you want

because a probabilistic reasoning framework, a scientific tool, will not meet your gameplay needs

i think you're imagining that there's some fuzziness to the behavior in the real world. there isn't, so the scientific thing is going to do an excellent job with finding the real answer, which isn't your intended gameplay

because right now i'm hearing

"i want the npc's dialogue to sometimes express doubt about what the player says, based on whom the player heard that from"

[half marten]

What if 2 somewhat trustworthy NPCs both share identical statements about an event? Those two testimonies should combine to give you a higher belief that the statement is true than if only one of the npcs said it, right?

[undone coral]

somewhat trustworthy is a Rhubarbist gameplay concept

[half marten]

Forget about the player for now

[undone coral]

do you see what i mean

Those two testimonies should combine to give you a higher belief that the statement is true than if only one of the npcs said it, right?
in a scientific sense? no

somewhat trustworthy? does that mean you don't know yet whether "NPC A is trustworthy" is true or false yet?

[half marten]

It's not a binary thing

[undone coral]

the facts aren't 0 to 1. the facts are true or false. the probability of those facts are 0-1

well there you go

i think that's where you're confused

anyway like i said

just think about it

[half marten]

Bayesian networks do use 0 to 1 probabilities though?

[undone coral]

in a bayesian network, the nodes are booleans

the directed arrows between nodes are expressing a relationship that the bool affects another bool

[half marten]

With a certain weight?

[undone coral]

in the sense that instead of say, a XOR gate, it's a probabistic logic gate.

maybe look carefully at what it is

wikipedia has a good example

https://en.wikipedia.org/wiki/Bayesian_network#Example

[half marten]

That example confirms what I was saying? The belief that it rained isn't 0 or 1, it's 0.2

[undone coral]

no

that's not what it's saying

it either rained or didn't rain. it can tell you the probability rain=true given you observed wet grass and that the sprinkler was off.

that's a query you can ask it

it doesn't mean it's 0.35 rained

or wahtever

that would be misinterpreting what it's trying to say

"What is the probability that it is raining, given the grass is wet"

so it would be incorrect that it rained 0.35 inches instead of 1 inch

do you see what i mean

i wouldn't interpret the probability as "how trustworthy is the npc" if you had a node "npc is trustworthy"

in the same way that i wouldn't interpret "how heads is a coin" as "0.5"

you certainly could interpret it that way but it would be counterintuitive and weird

the agent would behave poorly

a simple criteria would be, if the probability is greater than 50%, then the npc is trustworthy... but what you'll see is that most of the time, because of the way this is structured and if you use a learned value for the truth tables

that it will bel ike, 1% as soon as the npc tells one lie

as soon as that one lie is known

and it will be very close to 50% as long as it's an incomplete picture of their facts they've said

it will bounce around 50%

...essentially RANDOMLY

which is why i'm saying they might as well act randomly

or don't use the scientific criteria!

i don't think it's suitable for the gameplay you're trying to make

which is fictional world dialogue system

[half marten]

In my example, the observations are that certain people told you they believe a certain event happened. We also have the probability that each source will tell the truth about any event.

The source trust variable is essentially describing to what level should you take something that source says seriously. A trust of 0 means that the source's descriptions are completely random. It's possible they might sometimes say something that actually happened, but it shouldn't affect your belief at all.

[undone coral]

anyway i think it's pretty interesting

[half marten]

If two sources share identical descriptions of an event, our belief of that specific description should take in both of those observations.

[undone coral]

you can use a bayesian network to model this

and use "NPC A is trustworthy" as a precedent / parent node to everything the NPC A says

[regal olive]

How do I predict where the bullet will hit so i can make the ai aim at the player

i know the Velocity and traveltime

[undone coral]

you can query "what is the probability npc a is trustworthy" and it iwll give you a probability, and a good agent will probably treat 90% or better as true.

but who knows! maybe with a big network, >50% is true would make the agent better performing. that's an empirical question

[undone coral]

you can model this in a bayesian network as "npc A and npc B agree on on a description"

it would be a precedent of "the description happened this way"

or wahtever

as lon gas you express your things as bools, it works.

"probabilistic prolog" can help you express these things more succinctly the way they sound

but then you have to learn a programming language 🙂

[obsidian glade]

I think the answer you're looking for is really that it can be however you like, you're trying to establish some probability to an unknown from unknowns and without data, and the only thing you have to base anything off is the human and borderline philosophical point that if multiple people are saying the same thing it's probably more likely to be true, or that if they told a lie it puts doubt on their other statements - exactly how you quantify that social phenomenon really is up to you though

some models attempt to simulate it, some make assumptions, there's an entire field of ML devoted to taking data and building model weights out of it, you can make a very simple scoring system or try to incorporate some statistical theory, you can structure the data in a certain way, but there isn't a magic function of absolute truth to it

[undone coral]

agent better performing
in the sense that, if you had a game where the npc that gets the most facts right with the least amount of information.

[half marten]

But there are so many different ways that descriptions can overlap. Wouldn't you have to have a node for every single possible combination of agreements?

[undone coral]

i think maybe try something simple you can wrap your midn around

because "probabilistic prolog" is a million times more confusing lol

[undone coral]

i don't know yet

i'd have to try making this thing to find out

wait until you discover Pyro

your head is going to explode

[half marten]

I'm not really trying to implement ML, at least not for now

I get the feeling that Bayesian Networks can't solve this though, When you combine incomplete descriptions from different sources you quickly run into combinatorial explosion

[obsidian glade]

as a really simple idea as I'm a sucker for just trying ideas out without theory - how about you take the full set of statements from NPCs and build a set of events that have received predictions, then for each event you score it either positively or negatively based on a simple heuristic like:
1.0 trustworthy person said it's true -> +10 points
0.8 trustworthy person said it's false -> -8 points

this likely wont work exactly how you want, but you can iterate on additional factors as you go

[half marten]

Trustworthiness is the level that any given statement from a source correlates with the truth

So 1.0 trustworthiness = +infinity points

0.0 trustworthiness = no effect whatsoever

[obsidian glade]

sounds like you've already started on the iteration 👍

[half marten]

But I already know how I would implement this if all the sources have identical descriptions

It would use this formula: https://i.imgur.com/IhrMyfL.png

https://i.imgur.com/IhrMyfL.png

[undone coral]

nah

i think if you just did it it would be fine

your game is going to have 5 npcs in it...

you know what i mean?

i would start with something that has a pre-existing osftware library

that you can start building the game around

[half marten]

The issue is when descriptions overlap with each other

[half marten]

What kind of software library?

Prolog?

[undone coral]

i would search "github c# bayesian network"

[half marten]

See this graph for example

https://www.google.com/search?q=y+%3D+(x+*+0.6)+%2F(+(x+*+0.6)+%2B+((1-x)+*+0.4+)&rlz=1C1ONGR_en-GBGB981GB981&sxsrf=ALiCzsbn9sAA5kVdO_h7qkhXZ5aoWvm6bQ%3A1658869875603&ei=c1jgYtq2JJGhgAaun52ADQ&ved=0ahUKEwiavKftu5f5AhWREMAKHa5PB9AQ4dUDCA8&uact=5&oq=y+%3D+(x+*+0.6)+%2F(+(x+*+0.6)+%2B+((1-x)+*+0.4+)&gs_lcp=Cgdnd3Mtd2l6EANKBAhBGAFKBAhGGABQhAhYm2Jg1GNoAXAAeACAAUyIAdACkgEBNpgBAKABAcABAQ&sclient=gws-wiz

[obsidian glade]

reminds me of a philosophical question I was asked once: is finding a blue shoe evidence that all elephants are grey?

[obsidian glade]

I understand what you're saying - I'm deliberately not trying to calculate the exact weights for you because it's really up to you, there is no right answer here other than figuring out which factors you want to include and what model you'd like to build

[half marten]

Two sources in total agreement, one has trust 0.6, and the other has x. If x is 0.5, belief of the event is 0.6, because x isn't taken into account at all. If x is 0.6, total belief is 0.69

[obsidian glade]

I think doctorpangloss has given you a good high level approach to take and if you don't like that or think it doesn't fit maybe trying something simple and just getting to grips with the problem could help

🤷‍♂️

[half marten]

So 2 sources with trust=0.6 agreeing on the same statement gives the statement a belief of 0.69. Higher than if only one of the sources had said it.

[half marten]

It's not a question of weights, I have that figured out. But I don't think a Bayesian network can be built because the number of possible combinations of statements that could be referring to the same event is too high

[obsidian glade]

from the original question it doesn't seem obvious to me that that would be the case

especially if each attribute is independent

[half marten]

Take this example:

3 sources all have trustworthiness 0.6.

Source A says "Someone (null) congratulated Tom"
Source B says "Tim congratulated Tom on Monday"
Source C says "Mike congratulated Tom on Saturday"

B and C give completely different events, but they both agree with A.

If A, B, and C all agreed with each other, the belief of the event they're asserting would just be = (0.6 * 0.6 * 0.6) /( (0.6 * 0.6 * 0.6) + (0.4 * 0.4* 0.4 ) = 0.77142857142

So 3 sources with 0.6 belief agreeing would give 0.77 belief for the statement they agree on.

[half marten]

But here, B and C can't be referring to the same event. It's possible that both events happened, or just one, or neither.

[obsidian glade]

surely source C could be forgetting that it was Tim rather than Mike who congratulated Tom on Saturday, and source B could be forgetting that it was actually on Saturday rather than Monday, and it is indeed the same event?

[half marten]

For the time being, I'm ignoring that possibility.

[half marten]

And even if that was the case, it's still somewhat unlikely that they're referring to the same true event, and more likely that they're referring to different events altogether. Either way it just complicates things too much

[obsidian glade]

but you're just basing this off what feels right, trying to ignore very real possibilities as being too complicated, while seemingly wanting to account for other possibilities with scientific & mathematical rigour (which doesn't really exist, because the problem is so ill-defined to begin with) - it's all a bit strange to me

[half marten]

It just needs to feel right, but it needs to feel right in every given situation.

[wooden bronze]

is there anyway to stop an event from registering without having to unsubscribe it, like a function I can call to make it so it will not run any of the handlers, because the issue I am having currently is if the script or the object the script is on is destroyed, then the even somehow still runs the code from that script without it being in the scene (which would cause issues with it not being able to find its own gameobject and stuff). I could possibly just make it unsubscribe them during on destroy but since I'm using it in the new input system, the input system is destroyed first meaning I cant get a reference to it or something to unsubscribe them or something, I'm honestly confused. Anyway thanks for any help or guidance.

[fast pagoda]

i'm not sure why and how you would need to unregister during InputSystem destroy

event registering/unregister is usually done in OnEnable/OnDisable, the only thing you need when your "inputsystem is destroyed" is to disable every listeners i guess

[wooden bronze]

ok turns out you are right

apparently destroy does something different if I try to unregister in that, it throws the error, even if I am destroying the object OnDisable runs and unregistering it in there stops the issue.

Well, Thanks for your help, knew it had to be something simple

[strong harbor]

Would anyone happen to know why this isn't changing the color of the tiles despite getting the correct information for each tile in the loop?

[undone coral]

don't hesitate to share more on the chat as you make progress 🙂

it's a cool idea

[humble onyx]

could it be that the tileFlags are prohibiting you to set the color ?
https://docs.unity3d.com/ScriptReference/Tilemaps.TileFlags.LockColor.html
https://docs.unity3d.com/ScriptReference/Tilemaps.Tilemap.SetTileFlags.html

[strong harbor]

i have not messed with these settings at all so they shouldn't be

[unborn bramble]

Greetings! Not sure where i need to post my problem as it is not directly a code specific issue, so ill post it here for now.

Over the past couple of days, Unity has been crashing over and over. The behaviour in question has been happening whenever i was dragging an element to another sub window (like the Inspector), resulting in this screen:

This happens with every element of unity, from prefab editing to straight up variable assignment in the editor

For the past 20 minutes for instance, i had that crash error happen 10 times.

I can not work with a program that constantly crashes the second i move something in the editor

So i came here for help.

[tawny kraken]

I want to add a custom animation to a GameObject that contains an animation controller through .dll modding (BepInEx w Harmony) I made an asset bundle containing an updated animation controller (Animator) (so it can handle my new animation in a separate layer), an AvatarMask and the actual animation clip (animation) containing the keyframes, Im not that versed with unity modding so It quickly becomes a headache trying to tackle this problem

This is what I came up to load the assets, however Im not sure if what im doing even could work, because the array of Objects I loaded Are just Unity.Object And I havent figured out what im I supposed to do with these Objects to convert them in the types I want (Animator, Animation, AvatarMask)

// Variables that I want to hold the data from the bundle
public Animator kickController = new();     
public Animation kickAnimation = new();     
public AvatarMask kickMask = new();

private void Awake()
  {
    var kickAnimationBundleStream = Assembly.GetExecutingAssembly().GetManifestResourceStream("RavenM.assets.kickanim");
    var kickAnimationBundle = AssetBundle.LoadFromStream(kickAnimationBundleStream);
    var animationBundleObjects = kickAnimationBundle.LoadAllAssets(); //This is the array of GameObjects Im talking about
    foreach (Object obj in animationBundleObjects)
      {
        Plugin.logger.LogInfo($"{obj.name} type: '{obj.GetType()}' detected in the kick animation bundle");
      }     
  }

[misty glade]

@unborn bramble: Start with the obvious stuff - do you have any weird/unconventional packages/DLLs/editor scripts? Then handle the middle-tier stuff - reinstall unity, make sure it's completely clean (no leftover DLLs, directories, etc) - check for files that can't be deleted (hard drive issues) or dangling configuration files. Registry directories, etc. Kill it all with fire. Then install clean. If that doesn't work, dig into the crash logs if you're so inclined, and at a minimum, report the bug to Unity. I know it's a fair pain in the ass but.. Life is Pain

[tawny kraken]

Btw, here is the manifest of the bundle in case you need it

[stuck onyx]

correct, it was the rendertexture didnt have the same texture type than the destination texture

[unkempt moat]

Hello !
I have a question
i'm using Unity Netcode for Gameobject and i'm trying to disconnect my client.
Basically; it's giving me this error :

NullReferenceException: Before using the library you should call Init() and do not forget to call Shutdown() afterwards

I searched on google, and it's telling me to use this :

Networktransport.Init();

Now the problem, is, as you can see in the screen : I do not have this function.

I'm heritating from NetworkBehaviour in my script, not Monobehaviour.

Would someone know how to fix that please ?

[thin mesa]

not an advanced issue. but you have to call Init on the instance, it isn't a static method

https://docs-multiplayer.unity3d.com/netcode/current/api/Unity.Netcode.NetworkTransport/index.html#initializenetworkmanager

[unkempt moat]

Like this ?

I think I did something wrong, I do not have any Instance I guess

[thin mesa]

that's NetworkManager, not the NetworkTransport

[unkempt moat]

How do you do an instance of the network transport ?

[thin mesa]

how did you get an instance of the object to attempt to disconnect?

[unkempt moat]

With the client Id

[thin mesa]

okay well at this point you should redirect your issue to #archived-networking or to the netcode for gameobjects discord server which is also pinned in networking
maybe consider going through some of the docs/tutorials from unity for how to use it

[unkempt moat]

Okay thank you :)

[scenic helm]

hi guys, i was wondering if anyone knows of a way to take some previously allocated memory (such as array of chars) and create a string that uses that exact memory, rather than allocating new memory? For example the string constructor can take a char* to an array of chars, but it will always copy the values to a new location in memory, thus requiring an allocation.

I know it's not technically allowed as strings are meant to be immutable, but you could argue that its already possible to mutate strings, by getting a pointer to it using the fixed statement etc.

im doing some low level string stuff and want to avoid allocations. I don't mind if its super smelly as long as it works!

[novel plinth]

Proly what you want is Span<T> tho the behaviors aren't what you described above

it is possible to change the string but requires unsafe... keep in mind, YOU SHOULDN't!

unsafe
{
    char* str = (char*) gcHandle.AddrOfPinnedObject();
    for (int i = 0; i < someStr.Length; i++)
    {
        str[i] = someStr[i];
    }
}

untested, and I'm assuming you're used to this, so you should get the idea

to look for this sort of sorceries, your best bet is to c# discord... you won't find this sort of answer here unfortunately

also, Unity supports fastSpan now, you should go that route instead

[stuck onyx]

Does anybody know why stream.Close() took 16ms when i looked on the profiler? is there something wrong the way im using it? :

   await stream.WriteAsync(bytes, 0, bytes.Length);
   stream.Close();

[fresh salmon]

Disposables should be enclosed in a using statement, to avoid forgetting from closing the stream and/or disposing of it

Other than that, looks good

[stuck onyx]

but should that affect the performance so much?

[fresh salmon]

It depends where the stream points to

[stuck onyx]

disk

[fresh salmon]

Memory, it should be pretty much instant, file, a bit longer

The actual contents aren't written to the disk until you call Flush, or close the stream which calls Flush internally

[stuck onyx]

yeah it pooints to Application.persistentDataPath

[stuck onyx]

just by adding using in the declaration is enough?

like...

using var stream = new FileStream(path, FileMode.OpenOrCreate);````

?

[fresh salmon]

using (var stream = new FileStream(...))
{
  // consume resource
}
// disposed here

[stuck onyx]

ah ok i remember this now

as you said is closed and disposed automatically if we do it like that

[thin mesa]

if you are using a unity version that supports this syntax, yes that would be enough. it's just sugar for the syntax spr2 provided in the following message

[fresh salmon]

Yep, and even if you return; early, or if an exception is thrown!

(the using block compiles to a try/finally block in the background...)

[stuck onyx]

just one question more, is there any way i can shorten the time the stream closes/is disposed?

maybe using something different than a stream? i used it because it allowed me to use WriteAsync

do I have an alternative?

[fresh salmon]

You can tell it that the IO is really async in the stream's constructor:

The IsAsync property detects whether the file handle was opened asynchronously. You specify this value when you create an instance of the FileStream class using a constructor that has an isAsync, useAsync, or options parameter. When the property is true, the stream utilizes overlapped I/O to perform file operations asynchronously. However, the IsAsync property does not have to be true to call the ReadAsync, WriteAsync, or CopyToAsync method. When the IsAsync property is false and you call the asynchronous read and write operations, the UI thread is still not blocked, but the actual I/O operation is performed synchronously.
Source MSDocs

[stuck onyx]

🙏 thanks a lot

[fresh salmon]

Oh, and it seems it also implements IAsyncDisposable, so you can do

await using FileStream stream = ...

The disposal will be awaited

[scenic helm]

@novel plinth thanks for the feedback.

I'm already using mostly span/readonlyspan on the implementation side, its just that unity and textmeshpro API's all require strings which is sad.

Ill check out fast span - is that something added in newer unity versions?

oh, and I feel like if I ask this sort of question on the C# discord I'll get burned at the stake for blasphemy or something ha

oh, fast span is an implementation detail of spans? I guess that means I'm already using fast spans for most of the implementation. I just need to somehow stuff it into a string at the end so I can hand it over to some other API lol.

[novel plinth]

If I were you, I'd just copy the implementation of Zstring 👀

also, fyi, String.Copy does exactly what you're looking for, for the muttable parts I mean... it still allocates for the creation of a new string of successful copy

here https://docs.microsoft.com/en-us/dotnet/api/system.string.copy?view=net-6.0

mind you, it's deprecated for a reason...

[stuck onyx]

where did you see it implements IAsyncDisposable?

[fresh salmon]

On MSDocs, it isn't explicitly told, but I scrolled down to the methods section and saw DisposeAsync being there

[stuck onyx]

hmmm maybe my c# version does not implement it

i think that must be c# 8 ?

[fresh salmon]

It's there for .NET Framework 4.8

Also there for .NET Standard 2.0

[stuck onyx]

hmmm how am i getting this then?

[stuck onyx]

intelligense shows FlushASync but not disposeAsync

[fresh salmon]

Alright, then keep the standard, non-async using

[stuck onyx]

but youre right, its here

https://docs.microsoft.com/en-us/dotnet/api/system.io.filestream.disposeasync?view=net-6.0

[fresh salmon]

Yeah, even on the .NET versions Unity uses, by switching to them using the dropdown at the top-left of the page

Oh well

Here's why

[stuck onyx]

😄

this is specific to unity?

any specific version or all of them?

[fresh salmon]

Yep, that's Unity docs

Any version that uses C# 8

[stuck onyx]

well.. thanks a lot

🙏

[sage radish]

2021.2 has full C# 8 support and partial C# 9 support.

[crimson ruin]

It's so unintuitive that destroyed GameObjects stick around a frame, why can't the API at least unparent them?

[novel plinth]

tho on the editor side, you can do so, but at the end of the frame... https://docs.unity3d.com/ScriptReference/Object.DestroyImmediate.html

[sly grove]

You could make your own function that destroys and unparents if that's what you want, and use that instead

[crimson ruin]

Yeah I'll do that and add it along side ExpectedUnparent 😀

[undone coral]

this is good new syntax

[undone coral]

the performance issues with textmesh pro / ugui aren't from copying even large strings

[undone coral]

anyway what is your objective? networking stuff? you can't marshal a char* to a managed string, full stop, without copying. you are expected to create ReadOnlySpans to achieve zero-copy buffers

[mint sleet]

Hello folks!

I got CORS issue from one of my two APIs. the other API works without flaw

but either way I could not handle the issue! Where is this fc bug coming from?

In the same application one API requires CORS the other one works fine.

and there is no difference!

[undone coral]

you might wanna ask people who are more familiar with web development

[undone coral]

ah. this is your issue. yes, do this on a unitask thread pool thread

you can also use await stream.FlushAsync with a main thread task

io to disk is always really slow. it just pretends to not be slow

[frosty elm]

hi dudes, i'm working on prototype for how MMO save the data and i need some help/information.
Any know any webpage, site or anything about how we can deal with store the player position, items, etc etc, during runtime?

[stuck onyx]

I didnt know FlushAsync did the same than Close or Dispose

thanks again doctor!

[undone coral]

see if there is a photon sample that is close to the gameplay you want, or search "unity mmo backend". there's also an asset store asset that is good for prototyping, and in my opinion very reasonable for 40-60 concurrent players with a typical intro skillset

[frosty elm]

any link? i cant find

[undone coral]

https://assetstore.unity.com/packages/templates/systems/ummorpg-remastered-mmorpg-engine-159401

[undone coral]

rpg maker might be more helpful to you - https://assetstore.unity.com/packages/tools/game-toolkits/rpg-builder-177657

[undone coral]

you should only destroy gameobjects that are used exclusively for rendering

[stuck onyx]

why is that ?

[undone coral]

SetActive(false) in contrast will immediately call OnDisable.

[stuck onyx]

hmm

[undone coral]

partly because of this problem - @crimson ruin is wrong that they stick around for a frame

it's that they stick around until a certain time that is counterintuitive

[stuck onyx]

what about memory? they must use some no?

or is it too little that doesnt matter?

[undone coral]

every game object that is Destroy ed will be in a phase where its code will have one last chance to execute before being removed from the scene... people will want it to be destoryed immediately but then... how would OnDestroy work?

[undone coral]

for non-rendering scenarios, where you might have 1 game object for all your little controller and manager components

yeah, too little to matter

[stuck onyx]

yeah that makes sense

[undone coral]

should the scene have ever been permitted to also process game logic? it's impossible to build an editor environment without that assumption.

hence DOTS not really changing things for the better*

[stuck onyx]

do i need to call .Close or Dispose after calling FlushAsync?

[undone coral]

dispose and close both call flush

dispose will always try to flush. if you've already flushed and there's no new data, it depends on the stream, but the most obvious implementation is it will see there's no unwritten data and dispose "instantly" - meaning without any io

so

UniTask.Void(async () => {
 using var fileStream = File.Open ( ... );
 await fileStream.FlushAsync();
});

@stuck onyx this is fine

the complete example:

[stuck onyx]

this is what i did, already inside a UniTask:

[undone coral]

// simplest: use a threadpool
UniTask.Void(async () => {
 await UniTask.SwitchToThreadPool();
 using var fileStream = new FileStream(...);
 fileStream.Write(...);
});

// on the main thread.
// must carefully never do blocking io
UniTask.Void(async () => {
 // await UniTask.SwitchToMainThread() is redundant here
 // file.open can still block indefinitely...
 var fileStream = new FileStream(...);
 try {
  await fileStream.WriteAsync(...);
 } finally {
  await fileStream.DisposeAsync();
 }
});

the best thing to do is use the thread pool

[stuck onyx]

the reason of using... 'using' is to dispose automatically of the stream no? if we are doing it manually whats the point?

[undone coral]

you can see in my main thread example i don't use using

because i'm not sure if your unity supports async disposables

if you are on the threadpool

you can use using var ...

[stuck onyx]

no it doesnt, it doesnt support them

[undone coral]

it's dangerous to create file streams on the main thread anyway

they can block indefinitely

meaning the constructor

i never write files using streams

i try to one-shot it in an async method

i suggest https://docs.microsoft.com/en-us/dotnet/api/system.io.file.writeallbytesasync?view=net-6.0

imo this is the only method for writing files that is safe enough to use on the main thread

because you don't actually care what the result is.

[stuck onyx]

i want to wait until the UnitaskVoid has finished, shall i launch it like

await UniTask.WhenAll(myTask);

[undone coral]

can you give me a little more context?

[stuck onyx]

im saving 16 textures on a loop

so i have to wait till 1 is finished to save the next one

1. render 1/16 part of the screen
2. move that portion from renderview to a native array
   3 Save that array in disc
   loop

the step 2 i did it as you told me using the GPUAsyncRequest

now when saving that into disc i saw there was a little delay when closing the stream thats why im back 😅

[undone coral]

use await File.WriteAllBytesAsync

and don't create a file stream, etc. etc.

just that one method

and save the files 1 at a time on the main thread, which is where async gpu readback gives you the frame bytes anyway

you need to be on the main thread

[stuck onyx]

thanks gonna try

hmmm

cant do mate

writeAllBytesAsync not available

i guess thats why i chose the stream

[undone coral]

hmm

it should be

[stuck onyx]

cant I writeAllBytes on a different thread and wait for it?

[undone coral]

on a thread pool thread yes

await UniTask.SwitchToThreadPool()

inside your task

https://github.com/dotnet/runtime/blob/96b4b53d4a13dfdf644343beeb67b57367559d18/src/libraries/System.Private.CoreLib/src/System/IO/File.cs#L1012

this is thei mplemetation but it's not easy to port

why does this shit matter? on mobile devices, the operating system isn't reliable with files

they don't maintain a contract

stat can take forever. you have to wrap anything that touches files with CancelAfterSlim

otherwise you can torch thread pool threads

if you do File.WriteAllBytes instead of File.WriteAllBytesAsync

do you see what i am saying

[stuck onyx]

im putting all my effort but nott completely, just that WriteAllBytes on a thread might f** ** me hard

[undone coral]

i'm not sure even WriteAllBytesAsync helps you much.

it will throw an exception if you have a problem with the file path

[stuck onyx]

well the filepath is starndard for all it shouldnt give a problem no? its the ApplicationDataPath/subfolder/

only thing is if they have no space

[undone coral]

i'm saying ios might decide, in an afternoon

[stuck onyx]

ah

okay, yeah got it

[undone coral]

"apps only get 1MB to write to, if we feel like it"

[stuck onyx]

yeah, wouldnt be surprising

[undone coral]

do you see what i mean

no contracts

you can ask for temporary files, that has worked the best in my experience

[stuck onyx]

what does this mean? 😅

[undone coral]

https://docs.unity3d.com/ScriptReference/Application-temporaryCachePath.html

[stuck onyx]

oh, i didnt even know that existed

oh but its readonly

anyway, leaving aside the location of the files... ill try to make use of the script you sent me

[undone coral]

the property is read only

[stuck onyx]

in order to use this method in https://github.com/dotnet/runtime/blob/96b4b53d4a13dfdf644343beeb67b57367559d18/src/libraries/System.Private.CoreLib/src/System/IO/File.cs#L1012

do I need the whole library?

[regal olive]

Hello. Is there a way to capture each frame produced by the rendered as int32 color arrays? Ideally a callback where I could script whatever I need to with that array.

onRenderImage cannot work in that case because it doesn't get called on HDRP pipelines.

[fresh salmon]

There's ScreenCapture.CaptureScreenshotAsTexture that gives out a Texture2D you can manipulate further. No idea about the performance of that method though

https://docs.unity3d.com/ScriptReference/ScreenCapture.CaptureScreenshotAsTexture.html

[stuck onyx]

take a look at this

@regal olive

https://github.com/keijiro/AsyncCaptureTest/blob/master/Assets/AsyncCapture.cs

first you set up a renderTexture, asign it to the cam you want

[regal olive]

Thanks that's very helpful.

[stuck onyx]

call AsyncGPUReadback.RequestIntoNativeArray
(ref _buffer, _rt.flip, 0, OnCompleteReadback);

and you have it

[regal olive]

wow that's slow

one second per call

is that expected?

oh nvm

there's a one second timer lol

[alpine adder]

when using async and await in unity for things like api calls do I have to worry about thread lock?

[misty glade]

Anyone know why you can't have prefabs with items that are reordered? It's not a problem for me but I'm curious behind the reasoning:

(like when trying to add a child gameobject in the middle of an instance of a prefabs heirarchy)

[sly grove]

by default async methods in Unity all run on the main thread

[sly grove]

it's just not one of the features they chose to support. They have support for changed fields and for added components. But not for rearranged or removed components. Probably because it's just a lot easier to implement those in the scene/prefab file format than deletions/rearranges

[undone coral]

tasks (async await) are not threads

you can read a lot more about it here - https://stackoverflow.com/questions/4130194/what-is-the-difference-between-task-and-thread

[tough knoll]

neat

[undone coral]

no i think it's not worth it

too hard to just grab

i mean, that particular piece of code and its dependencies

i think we're overthinking this 🙂 just write the files in a task on the thread pool

[tough knoll]

Hey while you're here

What are cases where synchronously writing files would block indefinitely?

[alpine adder]

Oh I thought I read in a forum it wasn't safe and thats why unity people don't use it

[tough knoll]

@undone coral

[undone coral]

lol

[undone coral]

you can use Tasks, it's easier to use UniTask

UniTask helps you deal with unity + Task idiosyncracies

i use tasks a lot now

they're more robust in my opinion that coroutines, and they are useful with unirx

[alpine adder]

I agree so the only reason why unity developres don't use it then is because of lack of knowledge?

[undone coral]

well i use it and i'm a unity developer 🙂

[rugged radish]

they do have some drawbacks, people may choose coroutines intentionally

[alpine adder]

depending on the situation through right? coroutines have their place but with async/await we can use those with apis and such

[rugged radish]

sure, tasks are very convenient, especially for async APIs
for me, coroutines may be better suited for driving lots of monobehaviors, tasks are kinda heavy to launch and track lots of them continuously

[alpine adder]

totally and completely agree. I was just under the impression asyn/await and tasks where a NONO in unity for whatever reason

[rugged radish]

well, Unity documentation used to have a "Avoid async/await" in bold letters not that long ago 🙂

might be why

[undone coral]

unity is going to integrate unitask and unirx

coroutines were created before Tasks

[hushed fable]

Tasks are also problematic on WebGL platform due to use of System.Threading, which UniTask addresses.

[sharp moon]

i just came across https://www.drdobbs.com/forward-difference-calculation-of-bezier/184403417?pgno=5 this solution for approximating points of a bezier curve. I'm not good at math and was wondering if there was a way to modify it to instead work on the first derivative of a bezier curve to approximate the tangents instead of the points?

[regal olive]

what is the best solution for having a websocket server in an unity game as of now?

[novel plinth]

Tasks become problem if you're using threadPool, if you're not doing context switching you'll be fine... UniTasks won't work in webGl too if it uses threadpool UniTask.RunOnThreadPool

tbf, it's WebGl issue not Unity or tasks... webgl supports for multi threading is just non-existent

[hushed fable]

Sure, but UniTask won't break with simple things like delaying tasks

It's not obvious when the async API uses unsupported System.Threading features

[novel plinth]

it is due to jasvascript ... https://www.tutorialspoint.com/why-doesn-t-javascript-support-multithreading

webGpu tackled this issue

[hushed fable]

You can multithread on the web as the link you posted says, it's just not quite that simple with all the mechanics combined with webassembly and potential complexities of Unity's runtime.

Not aware of WebGPU changing this

[novel plinth]

it's a feature that later added, not natively built from the start or thought when they 1st built it.. *I dont know how to say this in english, not my 1st language

[novel plinth]

it will, it has privilege to lower level api, basically what desktop can do, webgpu can too

https://www.w3.org/TR/webgpu/

[hushed fable]

Yes, at least on a GPU

[novel plinth]

same approach on cpu side, lower level api than webgl... if you're interested, take a time to read the implementation in the link above

[novel plinth]

even the old Unity docs stated the reason why the webgl won't work with threading https://docs.unity3d.com/2019.3/Documentation/Manual/webgl-gettingstarted.html

they later removed the notes, but the issue still there

proly they though it was bad for marketing purposes

[hushed fable]

It's likely just a bit poorly worded and doesn't necessarily mean what you think it does. Again, I don't think WebGPU will implement new multithreading paths. WebGPU is most likely not a magical lower access granter like the old school browser plugins were, but very specific graphics related features implemented in the browser sandbox.

Unity's engine for a brief moment was multithreaded on browsers, but there were a lot of issues.

[novel plinth]

either way, it's a fresh technology, only known high-tech companies adapted the technology and I honestly think Unity wouldn't adapt it soon 😃

..I don't think WebGPU will implement new multithreading paths they do have a brand new implementation of it called WebWorkers

it is also mentioned in the linked doc

[hushed fable]

You sure web workers is new? Web workers were mentioned in your first link and the spec was published in 2009.
The WebGPU spec says that it can be used from web workers.

Adopting WebGPU for the browser platform today would do nothing since none of the browsers have it enabled yet. https://caniuse.com/?search=webgpu

[novel plinth]

it a common term for background workers for web browsers.. Mozilla did the same thing with it's firefox

give me a sec

here https://developer.mozilla.org/en-US/docs/Web/API/Web_Workers_API/Using_web_workers

as for the implementation, they are different from one to another

[scenic helm]

Zstring looks pretty good I'll admit, we have our reasons for needing to roll our own - we are mainly targetting playstation consoles and there are some really annoying memory related behaviours that occur when dynamically allocating memory - mainly large amounts of memory fragmentation and wasted blocks that we absolutely have to avoid, as memory is never truly released on these platforms.

We have two systems currently - one which allocates a fixed buffer during startup that we use to store string data, with a custom type that acts largely like a string for all of our internal logic - the problem being that we sometimes need to stringify this. The other one is simply a library for performing non-allocation string analysis and manipulation - again for similar reasons.

String.Copy looks promising but as you say its deprecated so I wouldn't want to use it.

Either way thanks again for your feedback - I'll have a look into Zstring to see if it would work with a fixed buffer (doesn't look like it would be too far a leap).

[hushed fable]

Where does the WebGPU spec talk about new web workers implementation?

[novel plinth]

you can read around the GPUBuffer or GPUQueue sections

[hushed fable]

You sure they aren't just talking about how you can use WebGPU APIs from existing web workers

[novel plinth]

yes, what I suggested is for you to copy the implementation of it and ofcourse adapt it to your needs

[novel plinth]

nope, pretty sure

[novel plinth]

the source code is on github and you do exact the same

and with some adjustment obviously, to your needs

[scenic helm]

I'm referring to memory related problems on playstation consoles - stringifying isn't a slow operation, but it causes dynamic allocations which we need to avoid at all costs (see my above message).

I really only have one objective here - avoiding dynamic allocations of strings during runtime. Ultimately I may end up having to write some native code that can allocate strings using a fixed buffer or some memory that I can actively release.

[novel plinth]

as for String.Copy yeah, they deprecated it for a very obvious reason, which that obvious reason is to muttate internal string, which I supposed is what you want .. lol,, my english is funny

[scenic helm]

yeah I wont deny that what I'm trying to do is pretty stinky

unfortunately this sort of thing is common when optimizing for console

at least we can create native plugins.

[novel plinth]

that's the correct way to say it when one asking how to mutate a string.. lol

[novel plinth]

There's another option if you're looking for a cheap way of allocating strings, StringBuilder has ValueStringBUilder.. but sadly this is internal

internal ref struct ValueStringBuilder
{
    private char[]? pools;
    private Span<char> chars;
}

and i doubt if it will perform better than zstring

[scenic helm]

looks like it uses ArrayPool as well, unless you pass it a stackalloced buffer - so its probably similar in performance to Zstring?

I'm actually already using stackalloc'd buffers as scratchpads for the non-alloc string manipulations/analysis that I am doing - so I don't think it would change much in terms of performance for me.

And of course I have to stringify it at some point.

[novel plinth]

it will allocate more once you do tmpTxt.SetText(); while zstring literally won't... never personally tried it, just raw guesses

for a trivia, there's a proposal to expose ValueStringBuilder... but I don't think they would do that lol

[scenic helm]

fair enough. and yeah it doesn't seem like they would expose it any time soon.

[novel plinth]

here just incase you want to track the progress https://github.com/dotnet/runtime/issues/25587

the proposal was there since 2018 lol...

[scenic helm]

thanks

[regal olive]

Hello, what library should I use to have a websocket server in my unity game?

[compact ingot]

you can use the .NET builtin HttpListener. If that’s the best thing to do depends on your actual requirements.

[placid shell]

when importing stuff from a dll is there any way to get error handling and stop unity from crashing entirely if something isn't right?

[placid shell]

it appears that using try and catch doesn't work sadly

[compact ingot]

Only if the dll is written without unsafe code

[placid shell]

is there anyway of telling? I'm using a closed source nvidia api

[compact ingot]

Certainly using unsafe code if it’s written in C or C++

[placid shell]

ah, welp that is unfortunate

so i guess trial and error is my only hope, and there is no way of stopping unity from crashing entirely?

[compact ingot]

If you get segmentation faults and stack overflow from them, then no

[placid shell]

i dont get anything, no response,

it just crashes unity on hitting play

[compact ingot]

Sounds like segmentation fault

[placid shell]

unfortunate

what usually causes them?

[compact ingot]

accessing memory that the OS has not assigned to the process

[placid shell]

aha that is unfortunate, all im tryna do is run a function currently, is there anyway i can get it to not do that then?

[compact ingot]

No, but if you get it while calling a native ‚extern’ method, your address for the method is likely wrong

[placid shell]

oh ok then, how i can i make the address for the method correct then?

[compact ingot]

You need to know the correct signature . You can find a guide and examples on MSDN

[placid shell]

thankyou! That would make sense, as i do not know if i have the type correct yet!

[thick rain]

Anybody here working with netcode for gameobjects?

[thin mesa]

#archived-networking

[thick rain]

oh thanks

[placid shell]

btw if in the dll something is private, but when i recreate it in c sharp i make it public, will the dll not like this?

[untold moth]

Something?

[placid shell]

?

[untold moth]

If something is private in dll you won't be able to access it, and thus make it public.

[placid shell]

unfortunately im talking about structs, which cannot be grabbed from dll's as far as im aware, but from the nvidia docs i reckon they are private structs, so when im recreating them in c# i'm uncertain about whether to match the dll or make them public

[untold moth]

Well, then they're not the same structs anymore.

[placid shell]

oh, that is annoying, ill have to figure out how to create private structs, something ive never succeeded in doing before lol, oh well im sure ill find some way to make it work with being private then!

[untold moth]

Just mark them with a private accessor.🤷‍♂️

[placid shell]

aha, unfortunately i have another private struct, that takes that struct as part of its constructor

[untold moth]

Then don't make them private.🤷‍♂️

[placid shell]

but they are private in the dll? Do you think it would be ok?

[untold moth]

I don't see the issue.

[untold moth]

They're not the same struct, so it's totally fine.

[placid shell]

im just scared that if i dont set things up in c# the way it is set up in the dll that it won't accept the structs as being the same

[untold moth]

Yours defined in your assembly(potentially namespace). Theirs in their.

[placid shell]

im afraid i dont know what that means?

[untold moth]

It's like you can have 2 classes with the same name and layout in different namespaces. Doesn't make them the same class.

C# will still treat them as 2 different objects.

[placid shell]

oh ok that makes sense, so if my dll is expecting the private dll struct, but i pass it my public c# struct, it wont raise an error?

[untold moth]

It will. Because it expects something else.

[placid shell]

oh, well how can i get it to be happy then?

[untold moth]

Use their struct.

[placid shell]

how, it is a private struct in the dll? How do i pull it out?

[untold moth]

Then you can't.🤷‍♂️

[placid shell]

how would i pull out a public struct in a dll then, let us start with that

[untold moth]

Unless you get the source code, make it public and recompile the dll

[untold moth]

Same as anything else I guess.

[placid shell]

unfortunately it is not

dll import doesn't work for structs

there are functions in the api to create the private structs that save as public structs (hope that makes sense), so assuming i can find some way to get/create the public struct then perhaps i can just deal with not having access to the private struct, although trying to get/create any struct from the dll is a pain lol

[untold moth]

maybe something like this?
https://www.codeproject.com/Questions/1352299/Reading-a-complex-struct-from-Cplusplus-DLL-to-Csh

[untold moth]

That doesn't make sense.

https://docs.microsoft.com/en-us/dotnet/api/system.runtime.interopservices.marshal.ptrtostructure?view=net-6.0

Is it returned as a pointer? Hard to say anything without seeing the code in question.

[placid shell]

sorry for late reply, (i shall continue late replying sorry lol), i am unfortunately eating dinner, but all those links you have sent do look fascinating, so ill check them out after dinner, thanks!

[undone coral]

unfortunately no one here knows anything about optimizing for playstation, i'm sorry

[undone coral]

it is substantially, vastly easier to use an arena allocator for strings if this is essential to you. then you are welcome to manually memory manage strings.

zstring looks like it has some arena allocation features. @scenic helm without more context, it is hard to say what you should do. if you have to manipulate strings for routine gameplay, isn't it for things like dialogue? you will not have framerate issues in that situation, and besides, tough cookie, it's a text based game!

[final heath]

I want to make a vr combat system with swords but I dont know how to detect swing length and force. Can anybody help?

[stuck onyx]

short question about cancelation tokens....
In this function...

 async UniTask MyUnitask(CancellationToken ctoken)
    { 
    // my stuff 1
    UniTask.NextFrame(ctoken);
    //my stuff 2
    }````

when the cancellation token is triggered, does myStuff 2 executes?

or i need to do....

[compact ingot]

you need to await that next frame

[stuck onyx]

yeah but... the Unitask.NextFrame(ctoken) is enough for the rest of the script NOT continue executing or do i have to do... if(ctoken.HasBeenCanceled)return ?

[compact ingot]

if you need to make sure it doesn't run you need to guard it with ct.IsCancellationRequested or ct.ThrowIfCancelationResuested()

[stuck onyx]

async UniTask MyUnitask(CancellationToken ctoken)
    { 
    // my stuff 1
    UniTask.NextFrame(ctoken);
  if(ctoken.IsCancellationRequested) return;
    //my stuff 2
    //my stuff 3
  UniTask.NextFrame(ctoken);
  if(ctoken.IsCancellationRequested) return;

    //my stuff4
     //mystuff 5
    }````

so i'd need to explicitly ask if it has been requested to avoid the task to continue, like that

because in this case i want the task to execute complete, i dont want to stop at certain point and continue

if it is interrupted, i want it to start again

would be that the corrrect way ?

[compact ingot]

iirc you'd use ct.ThrowIfCancellationRequested(); to actually cancel all subsequent execution immediately

[rugged radish]

check out the cancel exception workflow
i prefer it more because of the finally block, very convenient to reset / dispose decencies and stuff like that

[compact ingot]

if you just check the token, nothing is actually cancelled but what you manually skip

[stuck onyx]

hmmm okay lests see...

i think ThrowIfCancellationRequested is enabled on UniTask by default

so...

[compact ingot]

that NextFrame will not throw if you cancel it

it just returns a completed task

[stuck onyx]

ohh, okay

if i use the AttacheExternalCancellation, the task will stop the flow then ?

like...

async UniTask MyUnitask(CancellationToken ctoken)
    { 
    // my stuff 1
    UniTask.NextFrame();
    //my stuff 2
    //my stuff 3
  UniTask.NextFrame(ctoken);
    //my stuff4
     //mystuff 5
    }````

[compact ingot]

idk, never used that

[stuck onyx]

okay ill do some tests

[compact ingot]

its a discouraged API

[stuck onyx]

discouraged?

what do you mean?

[compact ingot]

you're not supposed to use it

but its there if you run into situations where you have to

[stuck onyx]

i've heard doctorpangloss say unity is going to include it

unitask and UniRx

[compact ingot]

UniTask does this a lot, the longer the method name the more "discouraged" its use is

[stuck onyx]

hmm, thanks a lot Anikki!!!

btw , aniki comes from the japanese or some sort of alien stuff

[compact ingot]

but its only discouraged based on the opinion of its author ofc

[compact ingot]

its jap. for "big brother" but misspelled

[stuck onyx]

yeah, callme oyabun

XD

[compact ingot]

its also a regular norwegian/finnish name

[stuck onyx]

ohh, youre from norway?

[compact ingot]

nope

[stuck onyx]

cool, because it sucks

[compact ingot]

no dissing of other cultures!

[stuck onyx]

lol i dont diss the culture, i just dont like the country XD

[compact ingot]

yeah, this is not the place for discussing that kind of opinion 😄

[stuck onyx]

true

[compact ingot]

btw the one thing to remember about async is that its not magic and if you have a fundamental understanding how threads work, you can guess most of its behaviour, but if you are talking to people the discussion often revolves around implicit and quirky parts of it that are very easy to misunderstand and often people have opinions about what they mean.

[stuck onyx]

yeah i understand till certain point how threads work, thats why im not multithreading, just using async/await

[compact ingot]

if you want to master async in C# you need to understand synchronization contexts

and when in doubt, you can always do your own async with primitives (and a lot of bugs), or in unity (and other single threaded systems) by tracking state variables

[stuck onyx]

my own async with primitives? what do you mean ?

[rugged radish]

try
{
...
}
catch (OperationCanceledException) when (tokenSource.Token.IsCancellationRequested)
{
...
}
finally
{
...
}```
what I use in my code to cancel tasks
they're regular c# Tasks, but UniTask is very similar afaik

and I don't have any checks in the awaited task

[compact ingot]

locks, semaphores, monitors, and most fundamentally Interlocked.CompareExchange

[stuck onyx]

oh, okay, I got it

[compact ingot]

there is a lot of nuance to it, but its all there, no secret sauce needed

[undone coral]

okay so cancellation tokens are confusing

because it's just copying java

and i happen to know a lot about how it works in java 🙂

so consider the following categories of methods:

1. i do a lot of cpu work on the thread i am called on and
   a. i expect to finish and be done, OR
   b. i can be gracefully interrupted and cleanup my work
2. i do something that either succeeds or fails, and i do no CPU work but it takes a long time, and
   a. i have no way of knowing if i succeeded or failed until the long time is over
   b. i can query and potentially interrupt the execution of the thing i'm waiting on
3. i do a mix of the categories 1 and 2

if you are 1, the implementation should check if a passed token has is cancellation requested "often enough" to give the cpu time back to the canceller. if you are 1a, you should throw operation cancelled exception because you didn't finish and it would be a programming error to cancel. if you are 1b, you should not throw anything.

if you are 2, which is typical of making a web request, you should 2a throw an operation cancelled exception or 2b do not throw the exception. in 2a, cancellation is always implemented as something similar to a timeout. because for example, you can't actually cancel a web request. do you see what i mean? it will still happen even if you decide to cancel an await UnityWebRequest.Post - the thing you are waiting on in 1a is just time.

if you are 3, you should use a return type that properly communicates what happens or method overloading to let the user decide. this is common because you will compose things that cannot really be cancelled (web requests) with things that can (creating an image in unity from a web request).

@stuck onyx WHY this isn't documented anywhere i don' tknow

OperationCanceledException will always be thrown by category 2a, which is by and far the most common category people interact with. it makes sense why!

you cannot actually cancel a web request. so in my opinion you should actually not pass the cancellation token to a web request, and instead after the request check if cancellation is requested and gracefully exit without throwing (i.e., category 3 method, you are a method that composes things that can and can't be cancelled)

some methods, like UniTask.Delay, they can be cancelled gracefully but unitask throws an operation cancelled exception anyway. that's because category 2b is too confusing of an API distinction, so nobody deals with it

for this there is the don't throw if cancellation requested extension method in UniTask, so that you can handle methods you know are category 2b

[stuck onyx]

i can know if they have been finished or not but saving cpu is more important so i rather 1b

[undone coral]

you can't actually cancel a file writing operation

[stuck onyx]

so stop everything and start again when possible

[undone coral]

you can design a file writing operation that can be cancelled

yes

so for example, writing buffer-by-buffer you would check if cancellation is requested in your while loop

this is fine

[stuck onyx]

hmmmm, no i dont think I want to cancel an operation like that.... as you suggested... I must pass the CancellationToken

and check if it has been triggered, correct?

so this...

[undone coral]

if it's indeed on a different thread. if it's on the main thread and you want to cancel from the main thread, you would have to await UniTask.Yield() (which is correct anyway)

so for example

[stuck onyx]

im confused about this:
``` await TakeSnapShotAsync(i, false).AttachExternalCancellation(cancelAsyncSnaps);````

is this supposed to attach a cancellation token that cancels the task on any stage?

[undone coral]

it can't do that no

[stuck onyx]

or this is sh**t

and i must pass the cancellation token by parameter

[undone coral]

you have to make it a method argument in order to pass it down

[stuck onyx]

and check

[undone coral]

yes

[stuck onyx]

aha okay thats what im doing wrong

and then on the key parts of the process ask if it has been triggered and stop it myself

[undone coral]

while (bytesLeft && !cancellationToken.IsCancellationRequested) {
  // observe the file stream methods don't accept cancellation tokens
  // because they cannot actually be cancelled
  bytesLeft -= await WriteSmallChunk(someBytes);
  // this ensures you can cancel "a task executing
  // on the main thread" from the main thread
  await UniTask.Yield(); 
}

[stuck onyx]

yeah

thats how i saw one guy did it with normal tasks

but i saw this in Unitask and thought... this must be some advanced shit

[undone coral]

yuou would then check how many bytes are left

if it's greater than zero and yo're outside the while loop... i guess delete the file

the screenshot

however... consider that if you quit the app which gracefully cancels this stuff

you won't have a chance to delete the file

remember: you can't actually cancel a file writing operation

[stuck onyx]

hmmm no is not my case... if i cancel the process its because i want the cpu to be free for heavier operations, but its not like i want to cancel INMEDIATLY

[undone coral]

i think all you need then is UniTask.Yield

[stuck onyx]

you know, like when player moves, i set the FPS to 60

and when player is idle i set to 30

[undone coral]

if you want to yield more time

[stuck onyx]

so thats when i do this operations

[undone coral]

yeah

that makes sense

it's a little tricky to detect when you are idle

you can estimate how much time you have

[stuck onyx]

if player moves i cancel, but its not like im gonna stop in the middle of writting the snap into disk

[undone coral]

this is why in practice, you pick a fixed amount of work and spread it out

[stuck onyx]

nah it doesnt have to be exact, right now the script is damn quick

[undone coral]

you don't try to measure when you're idle

[stuck onyx]

ill fix the cancel thing

[undone coral]

cancellation makes the most sense for tasks that compose a bunch of steps

[stuck onyx]

im pretty proud of what the script does and since you helped me id like to share it with you

[undone coral]

cool

yeah i'm glad to spread the joy of tasks

[stuck onyx]

well im sure there must be stuff that can be inmensely improved though 😅

[undone coral]

we can look at the source code of WithExternalCancellation

@stuck onyx i think all it does is turn any method into a 2a - which can be cancelled in the sense that you can stop waiting for it

so it will correctly throw an operation cancelled exception and nothing else

[stuck onyx]

hmmm

why this does not work ?

i throw the OperationCanceled

and im catching it here

when i launch it

but seems it's not catching it

[undone coral]

you're throwing the exception in the callback

[stuck onyx]

oh shit

[undone coral]

the way to do this is

async UniTask TakeSnapshot(CancellationToken token=default) {
 var readback = new UniTaskCompletionSource<byte[]>();
 AsyncGPUReadback.Request(cb => {
  readback.TrySetResult(cb.GetData<byte>());
 });
 // cannot actually be cancelled!!!
 // we cannot pass a token to cancel a async gpu readback request
 var data = await readback;
 if (token.IsCancellationRequested) {
  return;
 }
 File.WriteAllBytes(data);
}

@stuck onyx do you see?

[stuck onyx]

you converted the callback into a task

[undone coral]

yeah, in a way

i made a way to wait for the callback

that is ergonomic

[stuck onyx]

yeah, i wanted to do something like that for so many parts of my code

[undone coral]

well

now you know 🙂

that's because instead of TaskCompletionSource it should have been called Future or Promise but microsoft didn't want to use any words that could get them in copyright jeopardy or some completely irrelevant legalistic horseshit that would impact exactly 0 people

you can blame microsoft's preposterously bad lawyers

[stuck onyx]

Promise is from js correct?

[undone coral]

yes

it's older than that i'm sure

but you see what i'm saying

[stuck onyx]

I see, Its a bit new to me but i get the grip of it

great @undone coral now the tasks are launched and cancelled exactly how i want and the order i want and the cpu is optimized 👏

only thing is i couldnt convert the callback

[undone coral]

i think you can figure it out from the example

i don't remember what the task completion source is called

[stuck onyx]

yep, googling the error

[undone coral]

okay

iut's jsut the types

i don't know if async gpu readback gives you the result on the main thread

[stuck onyx]

thanks a lot again once is finished ill send you the script

[undone coral]

i assume it does

[misty glade]

I need to pick a resolution for images in our game - they're "important" (they're character portraits), so high res seems better, but it's for mobile so to some extent, low res is fine. I know the ratio I want (1:2) and the designer is doing the work in vector so they can generate any size images just fine. How can I evaluate and pick an ideal size for these images that I'm going to put in the game? The only thing I'm aware of is that they should be a multiple of 4 so they can be compressed to DXT5 but.. I'm not sure how to see the imported texture sizes after unity does a build for a specific image.

ie - something like this https://github.com/aschearer/unitysizeexplorer but geared specifically for images - I also don't know if this repo/project is still functional, last commit over 4 years ago

[pale zinc]

you can use different compression settings and check file size and quality in the editor. If you switch to that platform, Unity will apply compression in the editor

https://docs.unity3d.com/Manual/class-TextureImporterOverride.html
I get pretty good results with ASTC 4x4 block on android and iOS

those don't need POT textures by the way

In most cases crunch compression can dramatically reduce memory footprint without being noticeable

[misty glade]

Where could you see the size in the editor? I am obviously able to change the resolution of the PNG but ... I can't really see the effect on the crunched file size after the build

even on debug mode

or is that it? complete image size?

8.3mb? jesus

(granted I haven't sized this image as a multiple of 4, or downsized the resolution but that seems stupidly large?)

[pale zinc]

drag out the preview window

on the texture, not the sprite

uncompressed NPOT texture of that size can easily be 8mb

[misty glade]

yeah lol, i made that exact same texture 2 pixels wider:

1/4 the size

[pale zinc]

for DXT5 that matters a lot yeah, but if it's impractical to make the texture POT, you can use different compression techniques

[misty glade]

how do you enable that preview window? Dumb question I know but I can't seem to locate it.

[pale zinc]

should just be at the bottom of the inspector (not in debug mode)

[misty glade]

I don't even think it has to be perfect POT - tooltip says any multiple of 4

Mine doesn't have that panel.. digging, though.. i'll find it

[pale zinc]

looks like this if it's folded in; just click it or drag up

[misty glade]

oh it only shows up if you have the "popup" window thing

like the inspector when it's embedded in a layout doesn't show it

[pale zinc]

oh yeah

wait, what

lol

[misty glade]

Got it - the debug mode "Complete Image Size" is the same value it appears, so .. I'll just look there (i usually am working in debug mode in my inspector anyway)

[pale zinc]

I just have my inspector embedded like normal

[misty glade]

preview window/widget of the inspector.. there's gotta be a way to enable it for the .. uh.. embedded view

[pale zinc]

can you take a screenshot of your entire window

it's weird if it's not there

[misty glade]

oh

hahaha i found it

alllllllllll the way hiding at the bottom

i would have assumed it would appear "after" the other components

derp.

[pale zinc]

yeah it's just the general preview window; it also shows materials, prefabs, etc

[misty glade]

What do I look like, an artist? Gimme code or gimme death. 😉

(As an aside the preview window for scripts is .. terrible)