What’s new in Unity 6? Overview of release updates and source code issues

Unicorn Developer
12 min readOct 18, 2024

--

Unity 6 is finally here! According to the developers, this is the most stable version in the game engine history. Why don’t we check it using a static code analyzer? We’ll also take a brief look at the key features and enhancements the update brings.

Introduction

Unity 6 is a global update that aims at improving engine performance and stability but goes far beyond that. The enhancements cover a variety of aspects: multiplayer tools, lighting system, rendering, post-processing, the XR-toolkit, some visual effects, and more. I’d also like to highlight a system for integrating AI models.

While the preview of Unity 6 has already been available since May 15, 2024, it seems that many people’s curiosity about what makes this update special has only recently arisen. So, in the first part of the article, I’ll give a brief overview of what I think are the most important enhancements. I won’t go into too much detail, though. Instead, I’ll provide links where you can find more information.

In the second part, you may want to test your code analysis skills by looking for potential errors found by the static analyzer in several code fragments of the open-source part of the engine. This exercise is a good way to learn from other people’s mistakes and minimize the chance of making them in your own code.

Enjoy reading the article!

What’s new in Unity 6?

Okay, let’s see what the Unity team has in store for us this time around.

AI integration based on artificial neural networks

The Sentis framework has been introduced. It enables developers to integrate pre-trained neural network models into the runtime environment and use them to implement features such as speech and object recognition, creation of intelligent NPCs, etc.

Toolkit extension for creating multiplayer projects

  • The new Multiplayer Center package serves as a starting point for creating a multiplayer game. It recommends multiplayer packages based on your project specifications and provides examples and educational materials.
  • The Unity team has introduced Multiplayer Play Mode. It enables users to simulate up to 4 players directly from the Unity Editor.
  • The devs have added the support for distributed authority topologies to the Netcode for GameObjects package. They’ve also added the Network Scene Visualization tool to the Multiplayer Tools package. It should help debug projects that use the new topology.
  • The team has introduced the Dedicated Server package to support the development of games and applications on a dedicated server platform.

Figure N1 — Multiplayer Play Mode in action

CPU load optimization

  • The new GPU Resident Drawer rendering system is now available. The system efficiency is directly related to the average number of object instances in a scene. The more instances of the same objects (e.g. vegetation created using SpeedTree) present, the more beneficial the system can be. It has some technical limitations. For example, it optimizes rendering of object instances with static meshes only. So, it doesn’t work with instances of particle systems and other effects.
  • The devs have added the new GPU Occlusion Culling feature. It enables Unity to use the GPU instead of the CPU to exclude objects from rendering when they’re occluded behind other objects.

GPU load optimization

  • The new Foveated Rendering feature is now available. It reduces the GPU load in XR projects by reducing the fidelity in the periphery of the user’s vision. The feature is equipped with two modes:
  • Fixed foveated renders the central region of the screen space for each eye at a maximum resolution and reduces the resolution in the peripheral areas;
  • Gaze-based foveated uses eye tracking to determine which regions of the screen space with maximum resolution.
  • The new feature helps reduce GPU load by automatically merging rendering steps in Render Graph. Users can further optimize the GPU by creating custom steps and native render passes. I’d also like to note that Render Graph is now available not only in HDRP but also in URP.

Better environment rendering quality

  • The developers have enhanced sky rendering for sunset and sunrise in HDRP. They’ve added the ozone layer support and atmospheric scattering to complement fog at long distances.
  • They’ve also improved the water rendering in HDRP by adding an underwater volumetric fog effect.

The left side of the image doesn’t have the ozone layer effect unlike the right one.

Underwater Volumetric fog

Miscellaneous enhancements

  • The developers have made significant improvements to the XR toolkit.
  • Unlike previous versions, the Unity 6 Editor now runs on Arm-based Windows devices.
  • The Unity team has enhanced the build window and added new build profiles.
  • They’ve also made some lighting-related enhancements, including a new way for building global illumination lighting.
  • The devs have fixed some known major issues and also improved Shader Graph.

I found the mentioned changes to be the most significant, but this is by no means a complete list. You can find a more comprehensive catalog of improvements, as well as their detailed descriptions, here.

Now that we’ve learned about the enhancements, it’s time to see whether the update has brought new bugs to the engine source code.

Dissecting new bugs in game engine code

So, I’ve equipped myself with the PVS-Studio code analyzer and investigated the C# part of the current engine version (6000.0.21f1), uncovering some interesting potential and explicit errors. However, instead of simply describing them, I’d like to invite you to test your code-analysis skills. Challenge yourself to find errors in the given code fragments and figure out how to fix it. Each snippet has a full answer below it, so you can check your guesses.

Let the Hunger Games begin, and may the programmer’s wits be ever in your favor! By the way, it’d be awesome if you could share your findings in the comments :)

First test

public bool propagationHasStopped { get; }
public bool immediatePropagationHasStopped { get; }
public bool defaultHasBeenPrevented { get; }
public EventDebuggerCallTrace(IPanel panel, EventBase evt,
int cbHashCode, string cbName,
bool propagationHasStopped,
bool immediatePropagationHasStopped,
long duration,
IEventHandler mouseCapture): base(....)
{
this.callbackHashCode = cbHashCode;
this.callbackName = cbName;
this.propagationHasStopped = propagationHasStopped;
this.immediatePropagationHasStopped = immediatePropagationHasStopped;
this.defaultHasBeenPrevented = defaultHasBeenPrevented;
}

The answer is here

The analyzer message:

V3005. The ‘this.defaultHasBeenPrevented’ variable is assigned to itself. EventDebuggerTrace.cs 42.

For some reason, the defaultHasBeenPrevented property is assigned to itself during initialization. Since the property has no setter, its value will never change again and will always have the default value of false.

You can fix the issue by adding a new parameter to the constructor to initialize the property.

Too easy for you? Let’s call this one a warm-up and see how you do next!

Second test

public void ParsingPhase(....)
{
....
SpriteCharacter sprite =
(SpriteCharacter)textInfo.textElementInfo[m_CharacterCount]
.textElement;

m_CurrentSpriteAsset = sprite.textAsset as SpriteAsset;
m_SpriteIndex = (int)sprite.glyphIndex;

if (sprite == null)
continue;

if (charCode == '<')
charCode = 57344 + (uint)m_SpriteIndex;
else
m_SpriteColor = Color.white;
}

The answer is here

The analyzer message:

V3095. The ‘sprite’ object was used before it was verified against null. Check lines: 310, 312. TextGeneratorParsing.cs 310.

The sprite variable is dereferenced before being checked for null:

m_CurrentSpriteAsset = sprite.textAsset as SpriteAsset;

m_SpriteIndex = (int)sprite.glyphIndex;

if (sprite == null)

continue;

If sprite is indeed null, this inevitably leads to an exception. Perhaps the developers didn’t consider the importance of operation order when adding string dereferencing to the method. To avoid the issue, you can move the dereferences under the null check:

if (sprite == null)

continue;

m_CurrentSpriteAsset = sprite.textAsset as SpriteAsset;

m_SpriteIndex = (int)sprite.glyphIndex;

Third test

private static void CompileBackgroundPosition(....)
{
....
else if (valCount == 2)
{
if (((val1.handle.valueType == StyleValueType.Dimension) ||
(val1.handle.valueType == StyleValueType.Float)) &&

((val1.handle.valueType == StyleValueType.Dimension) ||
(val1.handle.valueType == StyleValueType.Float)))
{
.... = new BackgroundPosition(...., val1.sheet
.ReadDimension(val1.handle)
.ToLength());

.... = new BackgroundPosition(...., val2.sheet
.ReadDimension(val2.handle)
.ToLength());
}
else if ((val1.handle.valueType == StyleValueType.Enum)) &&
(val2.handle.valueType == StyleValueType.Enum)
....
{

}

The answer is here

The analyzer message:

V3001. There are identical sub-expressions to the left and to the right of the ‘&&’ operator. StyleSheetApplicator.cs 169.

The analyzer prompts that two identical expressions are executed using the && operator in one of the conditions:

if (((val1.handle.valueType == StyleValueType.Dimension) ||
(val1.handle.valueType == StyleValueType.Float)) &&
((val1.handle.valueType == StyleValueType.Dimension) ||
(val1.handle.valueType == StyleValueType.Float)))

Obviously, this is an error, and the fix is fairly straightforward. Look at the rest of the code — it performs similar operations consistently, first using val1 and then val2. For example, the condition that follows the one in question, consists of two similar checks:

else if ((val1.handle.valueType == StyleValueType.Enum)) &&
(val2.handle.valueType == StyleValueType.Enum)

The only difference between the two is that the first one uses the val1 value, and the second one uses val2.

So, to fix the error, most likely, the devs need to replace all val1 with val2 in the repeated expression of the incorrect condition:

if (((val1.handle.valueType == StyleValueType.Dimension) ||
(val1.handle.valueType == StyleValueType.Float)) &&
((val2.handle.valueType == StyleValueType.Dimension) ||
(val2.handle.valueType == StyleValueType.Float)))

Fourth test

public partial class BuildPlayerWindow : EditorWindow
{
....
internal static event Action<BuildProfile>
drawingMultiplayerBuildOptions;
....
}

internal static class EditorMultiplayerManager
{
....
public static event Action<NamedBuildTarget> drawingMultiplayerBuildOptions
{
add => BuildPlayerWindow.drawingMultiplayerBuildOptions +=
(profile) => ....;
remove => BuildPlayerWindow.drawingMultiplayerBuildOptions -=
(profile) => ....;
}
....
}

The answer is here

The analyzer message:

V3084. Anonymous function is used to unsubscribe from ‘drawingMultiplayerBuildOptions’ event. No handlers will be unsubscribed, as a separate delegate instance is created for each anonymous function declaration. EditorMultiplayerManager.bindings.cs 48.

This is a simple yet common error. Anonymous functions are used here for subscribing to and unsubscribing from an event. No matter how visually similar they are, these functions are still different objects. As a result, subscribing to an event is performed correctly, but unsubscribing fails.

One solution is to implement a full-fledged method instead of an anonymous function and use for both subscribing to/unsubscribing from the event.

Fifth test

public void DoRenderPreview(Rect previewRect, GUIStyle background)
{
....
Matrix4x4 shadowMatrix;
RenderTexture shadowMap = RenderPreviewShadowmap(....);

if (previewUtility.lights[0].intensity != kDefaultIntensity ||
previewUtility.lights[0].intensity != kDefaultIntensity)
{
SetupPreviewLightingAndFx(probe);
}

float tempZoomFactor = (is2D ? 1.0f : m_ZoomFactor);
previewUtility.camera.orthographic = is2D;
if (is2D)
previewUtility.camera.orthographicSize = 2.0f * m_ZoomFactor;
....
}

private void SetupPreviewLightingAndFx(SphericalHarmonicsL2 probe)
{
previewUtility.lights[0].intensity = kDefaultIntensity;
previewUtility.lights[0].transform.rotation = ....;
previewUtility.lights[1].intensity = kDefaultIntensity;
....
}

The answer is here

The analyzer message:

V3001. There are identical sub-expressions ‘previewUtility.lights[0].intensity != kDefaultIntensity’ to the left and to the right of the ‘||’ operator. AvatarPreview.cs 721.

The condition of the first if statement in the DoRenderPreview method consists of two identical subexpressions. This could be either a bug or simply redundant code left out due to oversight.

The implementation of the SetupPreviewLightingAndFx method, which uses not only previewUtility.lights[0] but also previewUtility.lights[1], hints at an error here.

So, this is how you can fix the error:

if (previewUtility.lights[0].intensity != kDefaultIntensity ||  
previewUtility.lights[1].intensity != kDefaultIntensity)
{
....
}

Sixth test

void UpdateInfo()
{
....

var infoLine3_format = "<color=\"white\">CurrentElement:" +
" Visible:{0}" +
" Enable:{1}" +
" EnableInHierarchy:{2}" +
" YogaNodeDirty:{3}";

m_InfoLine3.text = string.Format(infoLine3_format,
m_LastDrawElement.visible,
m_LastDrawElement.enable,
m_LastDrawElement.enabledInHierarchy,
m_LastDrawElement.isDirty);

var infoLine4_format = "<color=\"white\">" +
"Count of ZeroSize Element:{0} {1}%" +
" Count of Out of Root Element:{0} {1}%";

m_InfoLine4.text = string.Format(infoLine4_format,
countOfZeroSizeElement,
100.0f * countOfZeroSizeElement / count,
outOfRootVE,
100.0f * outOfRootVE / count);
....
}

The answer is here

The analyzer message:

V3025. Incorrect format. A different number of format items is expected while calling ‘Format’ function. Arguments not used: 3rd, 4th. UILayoutDebugger.cs 179.

Take a look at the second string.Format(….). Its format string (the first argument) has four slots for insertion. There are also four values passed for insertion (from the second to the fifth argument). The issue is that the slots contain only 0 and 1. As a result, only the first and second values are inserted, while the other two aren’t used at all.

The corrected format string looks as follows:

var infoLine4_format = "<color=\"white\">" +
"Count of ZeroSize Element:{0} {1}%" +
" Count of Out of Root Element:{2} {3}%";

Seventh test

protected static bool IsFinite(float f)
{
if ( f == Mathf.Infinity
|| f == Mathf.NegativeInfinity
|| f == float.NaN)
{
return false;
}

return true;
}

The answer is here

The analyzer message:

V3076. Comparison of ‘f’ with ‘float.NaN’ is meaningless. Use ‘float.IsNaN()’ method instead. PhysicsDebugWindowQueries.cs 87.

The error here relates to a not-so-obvious behavior. A comparison of two values that are equal to NaN is always false. So, as the analyzer prompts, we need to use float.IsNaN(f) instead of the f == float.NaN expression here.

Eighth test

public readonly struct SearchField : IEquatable<SearchField>
{
....
public override bool Equals(object other)
{
return other is SearchIndexEntry l && Equals(l);
}

public bool Equals(SearchField other)
{
return string.Equals(name, other.name, StringComparison.Ordinal);
}
}

The answer is here

The analyzer message:

V3197. The compared value inside the ‘Object.Equals’ override is converted to the ‘SearchIndexEntry’ type instead of ‘SearchField’ that contains the override. SearchItem.cs 634.

As the analyzer message states, in the first Equals method, the other parameter is incorrectly converted to the SearchIndexEntry type instead of SearchField. This results in the same method overload being invoked when Equals(l) is called later. However, if other is really of the SearchIndexEntry type, the code loops. So, the StackOverflowException will be thrown.

Ninth test

private void DrawRenderTargetToolbar()
{
float blackMinLevel = ....;
float blackLevel = ....;
float whiteLevel = ....;
EditorGUILayout.MinMaxSlider(....);
float whiteMaxLevel = ....;

if (blackMinLevel < whiteMaxLevel && whiteMaxLevel > blackMinLevel)
{
m_RTBlackMinLevel = blackMinLevel;
m_RTWhiteMaxLevel = whiteMaxLevel;

m_RTBlackLevel = Mathf.Clamp(blackLevel,
m_RTBlackMinLevel,
whiteLevel);

m_RTWhiteLevel = Mathf.Clamp(whiteLevel,
blackLevel,
m_RTWhiteMaxLevel);
}
}

The answer is here

The analyzer message:

V3001. There are identical sub-expressions ‘blackMinLevel < whiteMaxLevel’ to the left and to the right of the ‘&&’ operator. FrameDebuggerEventDetailsView.cs 364.

Again, we have the if statement with the condition consisting of two essentially identical expressions, which differ from each other only in the operand order. It should probably be something else instead of blackMinLevel in the second expression. A review of the surrounding code suggests that whiteLevel would be the most logical option. So, the fixed condition could look as follows:

if (blackMinLevel < whiteMaxLevel && whiteMaxLevel > whiteLevel)
{
....
}

Tenth test

internal static IEnumerable<Sample> FindByPackage(PackageInfo package, ....)
{
if (string.IsNullOrEmpty(package?.upmReserved) &&
string.IsNullOrEmpty(package.resolvedPath))
{
return Enumerable.Empty<Sample>();
}
try
{
IEnumerable<IDictionary<string, object>> samples = null;
var upmReserved = upmCache.ParseUpmReserved(package);
if (upmReserved != null)
samples = upmReserved.GetList<....>("samples");
....
}
....
}

The answer is here

The analyzer message:

V3042. Possible NullReferenceException. The ‘?.’ and ‘.’ operators are used for accessing members of the ‘package’ object. PackageSample.cs 102.

The check at the beginning of the method aimed to handle several cases at once:

· when package.upmReserved is null or an empty string;

· when package.resolvedPath is null or an empty string;

· when package is null.

However, it doesn’t work correctly in most cases:

· if package is null, the NullReferenceException is thrown during dereferencing in the second subexpression of the check;

· if either package.upmReserved or package.resolvedPath is null or an empty string (but not both at once), the method doesn’t exit as expected.

The developers may have used the operator && instead of || by mistake. So, the fixed check can look like this:

if (string.IsNullOrEmpty(package?.upmReserved) || 
string.IsNullOrEmpty(package.resolvedPath))
{
return Enumerable.Empty<Sample>();
}

Eleventh test

[RequiredByNativeCode]
internal static void InvalidateAll()
{
lock (s_Instances)
{
foreach (var kvp in s_Instances)
{
WeakReference wr = kvp.Value;
if (wr.IsAlive)
(wr.Target as TextGenerator).Invalidate();
}
}
}

The answer is here

The analyzer message:

V3145. Unsafe dereference of a WeakReference target. The object could have been garbage collected between checking ‘IsAlive’ and accessing the ‘Target’ property. TextGenerator.cs 140.

Despite the small amount of code, this is probably the most complex error in the article. Here are the reasons why:

1. Not everyone is familiar with the concept of WeakReference. The garbage collector can always remove the object referenced by a weak reference, despite the presence of that reference.

2. The WeakReference.IsAlive property enables you to check whether the object referenced by the weak reference still exists. However, this code ignores the possibility that the object can be cleaned up after passing the check but before the reference is dereferenced. As a result, the NullReferenceException may be thrown anyway.

3. One would assume that the lock operator could protect the object from the garbage collector. However, after reproducing the case, I learned that this wasn’t true.

How can you protect yourself from the NullReferenceException, then? To avoid the NullReferenceException for sure, one needs to create a strong reference to the object. Then, the garbage collector will no longer delete it as long as the reference is relevant. In other words, create a simple local variable referring to the object and work with it. The safe method can look like this:

[RequiredByNativeCode]
internal static void InvalidateAll()
{
lock (s_Instances)
{
foreach (var kvp in s_Instances)
{
WeakReference wr = kvp.Value;
var target = wr.Target;

If (target != null)
(target as TextGenerator).Invalidate();
}
}
}

Conclusion

So, how’d you do? Aren’t you tired yet? Okay, now it’s time to take a break, as the article is coming to an end. I hope you found it not only useful but also enjoyable.

Looking for errors even in such small code fragments can be challenging, but it’s almost impossible to detect them in the code of such a large project as the Unity Editor. Unless, of course, you use special software tools.

You can try the tool I used to detect the potential issues described for free: just request a trial version on the PVS-Studio website.

By the way, you can use the PVS-Studio code analyzer to check not only the Unity engine but also games developed with it. More information on this you can find in the documentation.

See you in the next articles!

--

--

Unicorn Developer
Unicorn Developer

Written by Unicorn Developer

The developer, the debugger, the unicorn. I know all about static analysis and how to find bugs and errors in C, C++, C#, and Java source code.