November 8th, 2021

Announcing .NET 6 — The Fastest .NET Yet

Richard Lander
Richard Lander
Product Manager

Welcome to .NET 6. Today’s release is the result of just over a year’s worth of effort by the .NET Team and community. C# 10 and F# 6 deliver language improvements that make your code simpler and better. There are massive gains in performance, which we’ve seen dropping the cost of hosting cloud services at Microsoft. .NET 6 is the first release that natively supports Apple Silicon (Arm64) and has also been improved for Windows Arm64. We built a new dynamic profile-guided optimization (PGO) system that delivers deep optimizations that are only possible at runtime. Cloud diagnostics have been improved with dotnet monitor and OpenTelemetry. WebAssembly support is more capable and performant. New APIs have been added, for HTTP/3, processing JSON, mathematics, and directly manipulating memory. .NET 6 will be supported for three years. Developers have already started upgrading applications to .NET 6 and we’ve heard great early results in production. .NET 6 is ready for your app.

You can download .NET 6 for Linux, macOS, and Windows.

See the ASP.NET Core, Entity Framework, Windows Forms, .NET MAUI, YARP,  and dotnet monitor posts for what’s new in a variety of scenarios.

Visual Studio 2022 is also releasing today. Read the announcement and watch the launch event to learn more about the release.

PowerShell 7.2 is also releasing today, built on .NET 6. PowerShell users get access to the same performance improvements and APIs as .NET developers.

.NET Conf is a free, three-day, virtual developer event that celebrates the major releases of .NET. It starts tomorrow and runs November 9-11 featuring speakers from our team, teams at Microsoft, and the broader community with over 80 sessions. Tune in to learn and engage with us.

Check out the new conversations posts for in-depth engineer-to-engineer discussions on the latest .NET features.

.NET 6 Highlights

.NET 6 is:

The release includes about ten thousand git commits. Even with the length of this post, it skips over many improvements. You’ll have to download and try .NET 6 to see everything that’s new.

Support

.NET 6 is a Long-term Support (LTS) release that will be supported for three years. It is supported on multiple operating systems, including macOS Apple Silicon and Windows Arm64.

Red Hat supports .NET on Red Hat Enterprise Linux, in collaboration with the .NET Team. On RHEL 8 and later, .NET 6 will be available for the AMD and Intel (x64_64), ARM (aarch64), and IBM Z and LinuxONE (s390x) architectures.

Please start migrating your apps to .NET 6, particularly .NET 5 apps. We have heard from early adopters that upgrading to .NET 6 is straightforward from .NET Core 3.1 and .NET 5.

.NET 6 is supported with Visual Studio 2022 and Visual Studio 2022 for Mac. It is not supported with Visual Studio 2019, Visual Studio for Mac 8, or MSBuild 16. If you want to use .NET 6, you will need to upgrade to Visual Studio 2022 (which is also now 64-bit). .NET 6 is supported with the Visual Studio Code C# extension.

Azure App Service:

Note: If you’re app is already running a .NET 6 Preview or RC build on App Service, it will be auto-updated on the first restart once the .NET 6 runtime and SDK are deployed to your region. If you deployed a self-contained app, you will need to re-build and re-deploy.

Unified and extended platform

.NET 6 delivers a unified platform, for browser, cloud, desktop, IoT, and mobile apps. The underlying platform has been updated to serve the needs of all app types and to make it easy to re-use code across all your apps. New capabilities and improvements are available to all apps at the same time, so that your code running in the cloud or on a mobile device behaves the same way and has the same benefits.

dotnet-unified-platform

The reach of .NET developers continues to widen with each release. Machine learning and WebAssembly are two of the most recent additions. For example, with machine learning, you can write apps that look for anomalies in streaming data. With WebAssembly, you can host .NET apps in the browser, just like HTML and JavaScript, or mix them with HTML and JavaScript.

One of the most exciting additions is .NET Multi-platform App UI (.NET MAUI). You can now write code — in a single project — that delivers a modern client app experience across desktop and mobile operating systems. .NET MAUI will be released a little later than .NET 6. We’ve put a lot of time and effort into .NET MAUI and are very excited to release it and see .NET MAUI apps in production.

Of course, .NET apps are also at home on Windows desktop — with Windows Forms and WPF — and in the cloud with ASP.NET Core. They are the app types we’ve offered for the longest and they continue to be very popular, and we’ve improved them in .NET 6.

Targeting .NET 6

Continuing on the theme of a broad platform, writing .NET code across all those operating systems is easy.

To target .NET 6, you need to use a .NET 6 target framework, like the following:

<TargetFramework>net6.0</TargetFramework>

The net6.0 Target Framework Moniker (TFM) gives you access to all the cross-platform APIs that .NET offers. This is the best option if you are writing console apps, ASP.NET Core apps, or reusable cross-platform libraries.

If you are targeting a specific operating system (like if writing a Windows Forms or iOS app), then there is another set of TFMs (that each target a self-evident operating system) for you to use. They give you access to all the APIs in net6.0 plus a bunch of operating-system-specific ones.

  • net6.0-android
  • net6.0-ios
  • net6.0-maccatalyst
  • net6.0-tvos
  • net6.0-windows

The version-less TFMs are each equivalent to targeting the lowest supported operating system version by .NET 6. You can specify an operating system version if you want to be specific or to get access to newer APIs.

The net6.0 and net6.0-windows TFMs are supported (same as .NET 5). The Android and Apple TFMs are new with .NET 6 and currently in preview. They will be supported with a later .NET 6 update.

There are no compatibility relationships between the OS-specific TFMs. For example, net6.0-ios is not compatible with net6.0-tvos. If you want to share code, you need to do that with source with #if statements or binaries with net6.0 targeted code.

Performance

The team has had a deep and growing focus on performance ever since we started the .NET Core project. Stephen Toub does an amazing job of capturing the progress of .NET performance with each release. If you haven’t had the chance, I recommend taking a look at his Performance improvements in .NET 6 post.

In this post, I’ve captured some heavy-hitter performance improvements that you’ll want to know about, including File IO, interface casting, PGO, and System.Text.Json.

Dynamic PGO

Dynamic Profile-guided Optimization (PGO) can markedly improve steady-state performance. For example, PGO gives a 26% improvement (510K -> 640K) in requests per second for the TechEmpower JSON “MVC” suite.

Dynamic PGO builds upon Tiered Compilation, which enables methods to first be compiled very quickly (referred to as “Tier 0”) to improve startup performance, and to then subsequently be recompiled (referred to as “Tier 1”) with lots of optimization enabled once that method has shown to be impactful. This model enables methods to be instrumented in Tier 0 to allow various observations to be made about the code’s execution. When these methods are rejitted at Tier 1, the information gathered from the Tier 0 executions is used to better optimize the Tier 1 code. That’s the essence of the mechanism.

Dynamic PGO will have slightly slower startup times than the default runtime, as there is extra code running in Tier 0 methods to observe method behavior.

To enable Dynamic PGO, set DOTNET_TieredPGO=1 in the environment where your application will run. You must also ensure that Tiered Compilation is enabled (it is by default). Dynamic PGO is opt-in because it is a new and impactful technology. We want a release of opt-in use and associated feedback to ensure that it is fully stress-tested. We did the same thing with Tiered Compilation. Dynamic PGO is supported and is already in use in production by at least one very large Microsoft service. We encourage you to try it.

You can see more on dynamic PGO benefits in Performance in .NET 6 post, including the following microbenchmark, which measures the cost of a particular LINQ enumerator.

private IEnumerator<long> _source = Enumerable.Range(0, long.MaxValue).GetEnumerator();

[Benchmark]
public void MoveNext() => _source.MoveNext();

Here’s the result, with and without dynamic PGO.

Method Mean Code Size
PGO Disabled 1.905 ns 30 B
PGO Enabled 0.7071 ns 105 B

That’s a pretty big difference, but there is also increased code size, which might surprise some readers. This is the size of assembly code generated by the JIT, not memory allocations (which is a more common focus). There is a good explanation for that from the .NET 6 Performance post.

One optimization common in PGO implementations is “hot/cold splitting”, where sections of a method frequently executed (“hot”) are moved close together at the beginning of the method, and sections of a method infrequently executed (“cold”) are moved to the end of the method. That enables better use of instruction caches and minimizes loads of likely-unused code.

As context, interface dispatch is the most expensive call type in .NET. Non-virtual method calls are the fastest, and even faster still are calls that can be eliminated via inlining. In this case, dynamic PGO is providing two (alternative) callsites for MoveNext. The first — the hot one — is a direct call to Enumerable+RangeIterator.MoveNext and the other — the cold one — is a virtual interface call via IEnumerator<int>. It’s a huge win if the hot one gets called most of the time.

This is the magic. When the JIT instrumented the Tier 0 code for this method, that included instrumenting this interface dispatch to track the concrete type of _source on each invocation. And the JIT found that every invocation was on a type called Enumerable+RangeIterator, which is a private class used to implement Enumerable.Range inside of the Enumerable implementation. As such, for Tier 1 the JIT has emitted a check to see whether the type of _source is that Enumerable+RangeIterator: if it isn’t, then it jumps to the cold section we previously highlighted that’s performing the normal interface dispatch. But if it is — which based on the profiling data is expected to be the case the vast majority of the time — it can then proceed to directly invoke the Enumerable+RangeIterator.MoveNext method, non-virtualized. Not only that, but it decided it was profitable to inline that MoveNext method. The net effect is that the generated assembly code is bit larger, but optimized for the exact scenario expected to be most common. Those are the kind of wins we intended when we started building dynamic PGO.

Dynamic PGO is discussed again in the RyuJIT section.

File IO Improvements

FileStream was almost completely re-written in .NET 6, with a focus on improving async File IO performance. On Windows, the implementation no longer uses blocking APIs and can be up to a few times faster! We’ve also made improvements to memory usage, on all platforms. After the first async operation (which typically allocates), we’ve made async operations allocation-free! In addition, we have made the behavior for edge cases uniform where Windows and Unix implementations were different (and it was possible).

The performance improvements of this re-write benefit all operating systems. The benefit to Windows is the highest since it was farther behind. macOS and Linux users should also see significantly FileStream performance improvements.

The following benchmark writes 100 MB to a new file.

private byte[] _bytes = new byte[8_000];

[Benchmark]
public async Task Write100MBAsync()
{
    using FileStream fs = new("file.txt", FileMode.Create, FileAccess.Write, FileShare.None, 1, FileOptions.Asynchronous);
    for (int i = 0; i < 100_000_000 / 8_000; i++)
        await fs.WriteAsync(_bytes);
}

On Windows with an SSD drive, we observed a 4x speedup and more than a 1200x allocation drop:

Method Runtime Mean Ratio Allocated
Write100MBAsync .NET 5.0 1,308.2 ms 1.00 3,809 KB
Write100MBAsync .NET 6.0 306.8 ms 0.24 3 KB

We also recognized the need for more high-performance file IO features: concurrent reads and writes, and scatter/gather IO. We introduced new APIs to the System.IO.File and System.IO.RandomAccess classes for those cases.

async Task AllOrNothingAsync(string path, IReadOnlyList<ReadOnlyMemory<byte>> buffers)
{
    using SafeFileHandle handle = File.OpenHandle(
        path, FileMode.Create, FileAccess.Write, FileShare.None, FileOptions.Asynchronous,
        preallocationSize: buffers.Sum(buffer => buffer.Length)); // hint for the OS to pre-allocate disk space

    await RandomAccess.WriteAsync(handle, buffers, fileOffset: 0); // on Linux it's translated to a single sys-call!
}

The sample demonstrates:

  • Opening a file handle using the new File.OpenHandle API.
  • Pre-allocating disk space using the new Preallocation Size feature.
  • Writing to the file using the new Scatter/Gather IO API.

The Preallocation Size feature improves performance since write operations don’t need to extend the file and it’s less likely that the file is going to be fragmented. This approach improves reliability since write operations will no longer fail due to running out of space since the space has already been reserved. The Scatter/Gather IO API reduces the number of sys-calls required to write the data.

Faster interface checking and casting

Interface casting performance has been boosted by 16% – 38%. This improvement is particularly useful for C#’s pattern matching to and between interfaces.

image

This chart demonstrates the scale of the improvement for a representative benchmark.

One of the biggest advantages of moving parts of the .NET runtime from C++ to managed C# is that it lowers the barrier to contribution. This includes interface casting, which was moved to C# as an early .NET 6 change. Many more people in the .NET ecosystem are literate in C# than C++ (and the runtime uses challenging C++ patterns). Just being able to read some of the code that composes the runtime is a major step to developing confidence in contributing in its various forms.

Credit to Ben Adams.

System.Text.Json Source Generators

We added a source generator for System.Text.Json that avoids the need for reflection and code generation at runtime, and that enables generating optimal serialization code at build time. Serializers are typically written with very conservative techniques because they have to be. However, if you read your own serialization source code (that uses a serializer), you can see what the obvious choices should be that can make a serializer a lot more optimal in your specific case. That’s exactly what this new source generator does.

In addition to increasing performance and reducing memory, the source generator produces code that is optimal for assembly trimming. That can help with producing smaller apps.

Serializing POCOs is a very common scenario. Using the new source generator, we observe that serialization is ~1.6x faster with our benchmark.

Method Mean StdDev Ratio
Serializer 243.1 ns 9.54 ns 1.00
SrcGenSerializer 149.3 ns 1.91 ns 0.62

The TechEmpower caching benchmark exercises a platform or framework’s in-memory caching of information sourced from a database. The .NET implementation of the benchmark performs JSON serialization of the cached data in order to send it as a response to the test harness.

Requests/sec Requests
net5.0 243,000 3,669,151
net6.0 260,928 3,939,804
net6.0 + JSON source gen 364,224 5,499,468

We observe an ~100K RPS gain (~40% increase). .NET 6 scores a 50% higher throughput than .NET 5 when combined with the MemoryCache performance improvements!

C# 10

Welcome to C# 10. A major theme of C# 10 is continuing the simplification journey that started with top-level statements in C# 9. The new features remove even more ceremony from Program.cs, resulting in programs as short as a single line. They were inspired by talking to people — students, professional developers, and others — with no prior C# experience and learning what works best and is intuitive for them.

Most of the .NET SDK templates have been updated to deliver the much simpler and more terse experience that is now possible with C# 10. We’ve heard feedback that some folks don’t like the new templates because they are not intended for experts, remove object orientation, remove concepts that are important to learn on day one of writing C#, or encourage writing a whole program in one file. Objectively, none of these points are true. The new model is equally intended and equally appropriate for students as professional developers. It is, however, different from the C-derived model we’ve had until .NET 6.

There are several other features and improvements in C# 10, including record structs.

Global using directives

Global using directives let you specify a using directive just once and have it applied to every file that you compile.

The following examples show the breadth of the syntax:

  • global using System;
  • global using static System.Console;
  • global using Env = System.Environment;

You can put global using statements in any .cs file, including in Program.cs.

Implicit usings is an MSBuild concept that automatically adds a set of global using directives depending on the SDK. For example, console app implicit usings differ from ASP.NET Core.

Implicit usings are opt-in, and enabled in a PropertyGroup:

  • <ImplicitUsings>enable</ImplicitUsings>

Implicit usings are opt-in for existing projects but included by default for new C# projects. For more information, see Implicit usings.

File-scoped namespaces

File-scoped namespaces enable you to declare the namespace for a whole file without nesting the remaining contents in { ... }. Only one is allowed, and it must come before any types are declared.

The new syntax is a single line:

namespace MyNamespace;

class MyClass { ... } // Not indented

This new syntax is an alternative to the three-lined indented style:

namespace MyNamespace
{
    class MyClass { ... } // Everything is indented
}

The benefit is a reduction indentation in the extremely common case where your whole file is in the same namespace.

Record structs

C# 9 introduced records as a special value-oriented form of classes. In C# 10 you can also declare records that are structs. Structs in C# already have value equality, but record structs add an == operator and an implementation of IEquatable<T>, as well as a value-based ToString implementation:

public record struct Person
{
    public string FirstName { get; init; }
    public string LastName { get; init; }
}

Just like record classes, record structs can be “positional”, meaning that they have a primary constructor which implicitly declares public members corresponding to the parameters:

public record struct Person(string FirstName, string LastName);

However, unlike record classes, the implicit public members are mutable auto-implemented properties. This is so that record structs are a natural grow-up story for tuples. For example, if you have a return type that is (string FirstName, string LastName) and you want to grow that up to a named type, you can easily declare the corresponding positional struct record and maintain the mutable semantics.

If you want an immutable record with readonly properties, you can declare the whole record struct readonly (just as you can other structs):

public readonly record struct Person(string FirstName, string LastName);

C# 10 also supports with expressions not just for record structs but for all structs, as well as for anonymous types:

var updatedPerson = person with { FirstName = "Mary" };

F# 6

F# 6 is about making F# simpler and more performant. This applies to the language design, library, and tooling. Our goal with F# 6 (and beyond) was to remove corner-cases in the language that surprise users or present hurdles to learning F#. We are very pleased to have worked with the F# community in this ongoing effort.

Making F# faster and more interoperable

The new task {…} syntax directly creates a task and starts it. This is one of the most significant features in F# 6, making asynchronous tasks simpler, more performant and more interoperable with C# and other .NET languages. Previously, creating .NET tasks required using async {…} to create a task and invoking Async.StartImmediateAsTask.

The task {…} feature is built on a foundation called “resumable code” RFC FS-1087. Resumable code is a core feature, and we expect to use it to build other high-performance asynchronous and yielding state machines in the future.

F# 6 also adds other performance features for library authors including InlineIfLambda and unboxed representations for F# active patterns. A particularly significant performance improvement is in the compilation of list and array expressions, which are now up to 4x faster and have better and simpler debugging as well.

Making F# easier to learn and more uniform

F# 6 enables the expr[idx] indexing syntax. Up until now, F# has used expr.[idx] for indexing. Dropping the dot-notation is based on repeated feedback from first-time F# users, that the use of dot comes across as an unnecessary divergence from the standard practice they expect. In new code, we recommend the systematic use of the new expr[idx] indexing syntax. We should all switch to this syntax as a community.

The F# community has contributed key improvements to make the F# language more uniform in F# 6. The most important of these is removing a number of inconsistencies and limitations in F#’s indentation rules. Other design additions to make F# more uniform include the addition of as patterns; allowing “overloaded custom operations” in computation expression (useful for DSLs); allowing _ discards on use bindings and allowing %B for binary formatting in output. The F# core library adds new functions for copy-and-update on lists, arrays, and sequences, plus additional NativePtr intrinsics. Some legacy features of F# deprecated since 2.0 now result in errors. Many of these changes better align F# with your expectations, resulting in fewer surprises.

F# 6 also added support for additional “implicit” and “type-directed” conversions in F#. This means fewer explicit upcasts, and adds first-class support for .NET-style implicit conversions. F# is also adjusted to be better suited to the era of numeric libraries using 64-bit integers, with implicit widening for 32-bit integers.

Improving the F# tooling

Tooling improvements in F# 6 make day to day coding easier. New “pipeline debugging” allows you to step, set breakpoints and inspect intermediate values for the F# piping syntax input |> f1 |> f2. The debug display of shadowed values has been improved, eliminating a common source of confusion when debugging. F# tooling is now also more performant with the F# compiler performing the parsing stage in parallel. F# IDE tooling is also improved. F# scripting is now even more robust, allowing you to pin the version of the .NET SDK used through global.json files.

Hot Reload

Hot Reload is another performance feature, focused on developer productivity. It enables you to make a wide variety of code edits to a running application, collapsing the time you need to spend waiting for apps to rebuild, restart, or to re-navigate to the same spot where you were after making a code change.

Hot Reload is available through both the dotnet watch CLI tool and Visual Studio 2022. You can use Hot Reload with a large variety of app types such as ASP.NET Core, Blazor, .NET MAUI, Console, Windows Forms (WinForms), WPF, WinUI 3, Azure Functions, and others.

When using the CLI, simply start your .NET 6 app using dotnet watch, make any supported edit, and when saving the file (like in Visual Studio Code) those changes will be immediately applied. If the changes are not supported, the details will be logged to the command window.

Hot Reload CLI experience

This image displays an MVC app being launched with dotnet watch. I made edits to both .cs and .cshtml files (as reported in the log) and both were applied to the code and reflected in the browser very quickly, in less than half a second.

When using Visual Studio 2022, simply start your app, make a supported change, and use the new “Hot Reload” button (displayed in the following image) to apply those changes. You can also opt to apply changes on save through the drop-down menu on the same button. When using Visual Studio 2022, Hot Reload is available for multiple .NET versions, for .NET 5+, .NET Core, and .NET Framework. For example, you will be able to make code-behind changes to an OnClickEvent handler for a button. It is not supported for the Main method of an application.

Hot Reload Visual Studio 2022 experience

Note: There is a bug in RuntimeInformation.FrameworkDescription that is demonstrated in that image that will be fixed shortly.

Hot Reload also works in tandem with the existing Edit and Continue capability (when stopped at a breakpoint), and XAML Hot Reload for editing an apps UI in real-time. It is currently supported for C# and Visual Basic apps (not F#).

Security

Security has been significantly improved in .NET 6. It is always an important focus for the team, including threat modeling, cryptography, and defense in depth mitigations.

On Linux, we rely on OpenSSL for all cryptographic operations, including for TLS (required for HTTPS). On macOS and Windows, we rely on OS-provided functionality for the same purpose. With each new version of .NET, we often need to add support for a new build of OpenSSL. .NET 6 adds support for OpenSSL 3.

The biggest changes with OpenSSL 3 are an improved FIPS 140-2 module and simpler licensing.

.NET 6 requires OpenSSL 1.1 or higher and will prefer the highest installed version of OpenSSL it can find, up to and including v3. In the general case, you’re most likely to start using OpenSSL 3 when the Linux distribution you use switches to it as the default. Most distros have not yet done that. For example, if you install .NET 6 on Red Hat 8 or Ubuntu 20.04, you will not (at the time of writing) start using OpenSSL 3.

OpenSSL 3, Windows 10 21H1, and Windows Server 2022 all support ChaCha20Poly1305. You can use this new authenticated encryption scheme with .NET 6 (assuming your environment supports it).

Credit to Kevin Jones for Linux support for ChaCha20Poly1305.

We also published a new runtime security mitigation roadmap. It is important that the runtime you use is safe from textbook attack types. We’re delivering on that need. In .NET 6, we built initial implementations of W^X and Intel Control-flow enforcement technology (CET). W^X is fully supported, enabled by default for macOS Arm64, and is opt-in for other environments. CET is opt-in and a preview for all environments. We expect both technologies to be enabled by default for all environments in .NET 7.

Arm64

There is a lot of excitement about Arm64 these days, for laptops, cloud hardware, and other devices. We feel that same excitement on the .NET team and are doing our best to keep up with that industry trend. We partner directly with engineers at Arm Holdings, Apple, and Microsoft to ensure that our implementations are correct and optimized, and that our plans align. These close partnerships have helped us a lot.

  • Special thanks to Apple who sent our team a bushel of Arm64 dev kits to work with prior to the M1 chip launching, and for significant technical support.
  • Special thanks to Arm Holdings, whose engineers code reviewed our Arm64 changes and also made performance improvements.

We added initial support for Arm64 with .NET Core 3.0 and Arm32 before that. The team has made major investments in Arm64 in each of the last few releases, and this will continue for the foreseeable future. In .NET 6, our primary focus was on supporting the new Apple Silicon chips and the x64 emulation scenario on both macOS and Windows Arm64 OSes.

You can install both the Arm64 and x64 versions of .NET on macOS 11+ and Windows 11+ Arm64 OSes. We had to make several design choices and product changes to make sure that worked.

Our strategy is “pro native architecture”. We recommend that you always use the SDK that matches the native architecture, which is the Arm64 SDK on macOS and Windows Arm64. The SDK is large body of software. It is going to be much higher performance running natively on an Arm64 chip than emulated. We’ve updated the CLI to make that easy. We’re never going to be focused on optimizing emulated x64.

By default, if you dotnet run a .NET 6 app with the Arm64 SDK, it will run as Arm64. You can easily switch to running as x64 with the -a argument, like dotnet run -a x64. The same argument works for other CLI verbs. See .NET 6 RC2 Update for macOS and Windows Arm64 for more information.

There’s a subtlety there that I want to ensure is covered. When you use -a x64, the SDK is still running natively as Arm64. There are fixed points in the .NET SDK architecture where process boundaries exist. For the most part, a process must be all Arm64 or all x64. I’m simplifying a bit, but the .NET CLI waits for the last process creation in the SDK architecture and launches that one as the chip architecture you requested, like x64. That’s the process your code runs in. That way, you get the benefit of Arm64 as a developer, but your code gets to run in the process it needs. This is only relevant if you need to run some code as x64. If you don’t, then you can just run everything as Arm64 all the time, and that’s great.

Arm64 Support

The following are the key points you need to know, for macOS and Windows Arm64:

  • .NET 6 Arm64 and x64 SDKs are supported and recommended.
  • All in-support Arm64 and x64 runtimes are supported.
  • .NET Core 3.1 and .NET 5 SDKs work but provide less capability and in some cases are not fully supported.
  • dotnet test doesn’t yet work correctly with x64 emulation. We are working on that. dotnet test will be improved as part of the 6.0.200 release, and possibly earlier.

See .NET Support for macOS and Windows Arm64 for more complete information.

Linux is missing from this discussion. It doesn’t support x64 emulation in the same way as macOS and Windows. As a result, these new CLI features and the support approach don’t directly apply to Linux, nor does Linux need them.

Windows Arm64

We have a simple tool that demonstrates the environment that .NET is running on.

C:Usersrich>dotnet tool install -g dotnet-runtimeinfo
You can invoke the tool using the following command: dotnet-runtimeinfo
Tool 'dotnet-runtimeinfo' (version '1.0.5') was successfully installed.

C:Usersrich>dotnet runtimeinfo
         42
         42              ,d                             ,d
         42              42                             42
 ,adPPYb,42  ,adPPYba, MM42MMM 8b,dPPYba,   ,adPPYba, MM42MMM
a8"    `Y42 a8"     "8a  42    42P'   `"8a a8P_____42   42
8b       42 8b       d8  42    42       42 8PP"""""""   42
"8a,   ,d42 "8a,   ,a8"  42,   42       42 "8b,   ,aa   42,
 `"8bbdP"Y8  `"YbbdP"'   "Y428 42       42  `"Ybbd8"'   "Y428

**.NET information
Version: 6.0.0
FrameworkDescription: .NET 6.0.0-rtm.21522.10
Libraries version: 6.0.0-rtm.21522.10
Libraries hash: 4822e3c3aa77eb82b2fb33c9321f923cf11ddde6

**Environment information
ProcessorCount: 8
OSArchitecture: Arm64
OSDescription: Microsoft Windows 10.0.22494
OSVersion: Microsoft Windows NT 10.0.22494.0

As you can see, the tool is running natively on Windows Arm64. I’ll show you what that looks like ASP.NET Core.

.NET 6 ASP.NET Core running on Windows Arm64, Announcing .NET 6 — The Fastest .NET Yet

macOS Arm64

And you can see that the experience is similar on macOS Arm64, with architecture targeting also demonstrated.

rich@MacBook-Air app % dotnet --version
6.0.100
rich@MacBook-Air app % dotnet --info | grep RID
 RID:         osx-arm64
rich@MacBook-Air app % cat Program.cs 
using System.Runtime.InteropServices;
using static System.Console;

WriteLine($"Hello, {RuntimeInformation.OSArchitecture} from {RuntimeInformation.FrameworkDescription}!");
rich@MacBook-Air app % dotnet run
Hello, Arm64 from .NET 6.0.0-rtm.21522.10!
rich@MacBook-Air app % dotnet run -a x64
Hello, X64 from .NET 6.0.0-rtm.21522.10!
rich@MacBook-Air app %

This image demonstrates that Arm64 execution is the default with the Arm64 SDK and how easy it is to switch between targeting Arm64 and x64, using the -a argument. The exact same experience works on Windows Arm64.

x64 emulation on macOS Arm64 with ASP.NET Core

This image demonstrates the same thing, but with ASP.NET Core. I’m using the same .NET 6 Arm64 SDK as you saw in the previous image.

Docker on Arm64

Docker supports containers running with native architecture and in emulation, with native architecture being the default. This seems obvious but can be confusing when most of the Docker Hub catalog is x64 oriented. You can use --platform linux/amd64 to request an x64 image.

We only support running Linux Arm64 .NET container images on Arm64 OSes. This is because we’ve never supported running .NET in QEMU, which is what Docker uses for architecture emulation. It appears that this may be due to a limitation in QEMU.

Docker on macOS Arm64

This image demonstrates the console sample we maintain: mcr.microsoft.com/dotnet/samples. It’s a fun sample since it contains some basic logic for printing CPU and memory limit information that you can play with. The image I’ve show sets CPU and memory limits.

Try it for yourself: docker run --rm mcr.microsoft.com/dotnet/samples

Arm64 Performance

The Apple Silicon and x64 emulation support projects were super important, however, we also improved Arm64 performance generally.

Arm64 performance improvement

This image demonstrates an improvement in zeroing out the contents of stack frames, which is a common operation. The green line is the new behavior, while the orange line is another (less beneficial) experiment, both of which improve relative to the baseline, represented by the blue line. For this test, lower is better.

Containers

.NET 6 is better for containers, primarily based on all the improvements discussed in this post, for both Arm64 and x64. We also made key changes that will help a variety of scenarios. Validate container improvements with .NET 6 demonstrates some of these improvements being tested together.

The Windows container improvements and the new environment variable have also been included in the November .NET Framework 4.8 container update, releasing November 9th (tomorrow).

Release notes are available at our docker repositories:

Windows Containers

.NET 6 adds support for Windows process-isolated containers. If you use Windows containers in Azure Kubernetes Service (AKS), then you are relying on process-isolated containers. Process-isolated containers can be thought of as very similar to Linux containers. Linux containers use cgroups and Windows process-isolated containers use Job Objects. Windows also offer Hyper-V containers, which offers greater isolation through greater virtualization. There are no changes in .NET 6 for Hyper-V containers.

The primary value of this change is that Environment.ProcessorCount will now report the correct value with Windows process-isolated containers. If you create a 2-core container on a 64-core machine, Environment.ProcessorCount will return 2. In prior versions, this property would report the total number of processors on a machine, independent of the limit specified by the Docker CLI, Kubernetes, or other container orchestrator/runtime. This value is used by various parts of .NET for scaling purposes, including the .NET garbage collector (although it relies on a related, lower-level, API). Community libraries also rely on this API for scaling.

We recently validated this new capability with a customer, on Windows Containers in production on AKS using a large set of pods. They were able to run successfully with 50% memory (compared to their typical configuration), a level that previously resulted in OutOfMemoryException and StackOverflowException exceptions. They didn’t take the time to find the minimum memory configuration, but we guessed it was significantly below 50% of their typical memory configuration. As a result of this change, they are going to move to cheaper Azure configurations, saving them money. That’s a nice, easy win, simply by upgrading.

Optimizing scaling

We have heard from users that some applications cannot achieve optimal scaling when Environment.ProcessorCount reports the correct value. If this sounds like the opposite of what you just read for Windows Containers, it kinda-sorta is. .NET 6 now offers the DOTNET_PROCESSOR_COUNT environment variable to control the value of Environment.ProcessorCount manually. In the typical use case, an application might be configured with 4 cores on a 64-core machine, and scale best in terms of 8 or 16 cores. This environment variable can be used to enable that scaling.

This model might seem strange, where Environment.ProcessorCount and --cpus (via the Docker CLI) values can differ. Container runtimes, by default, are oriented in terms of core equivalents, not actual cores. That means, when you say you want 4 cores, you get the equivalent CPU time of 4 cores, but your app might (in theory) run on many more cores, even all 64 cores on a 64-core machine for a short period. That may enable your app to scale better on more than 4 threads (continuing with the example), and allocating more could be beneficial. This assumes that the thread allocation is based on the value of Environment.ProcessorCount. If you opt to set a higher value, it is likely that your app will use more memory. For some workloads, that is an easy tradeoff. At the very least, it is a new option you can test.

This new feature is supported for both Linux and Windows Containers.

Docker also offers a CPU groups feature, where your app is affinitized to specific cores. This feature is not recommended in that scenario since the number of cores an app has access to is concretely defined. We have also seen some issues with using it with Hyper-V containers, and it isn’t really intended for that isolation mode.

Debian 11 “bullseye”

We watch Linux distros lifecycle and release plans very closely and try to make the best choices on your behalf. Debian is the Linux distribution we use for our default Linux images. If you pull the 6.0 tag from one of our container repos, you will pull a Debian image (assuming you are using Linux containers). With each new .NET release, we consider whether we should adopt a new Debian version.

As a matter of policy, we do not change the Debian version for our convenience tags, like 6.0, mid-release. If we did, some apps would be certain to break. That means, that choosing the Debian version at the start of the release is very important. Also, these images get a lot of use, mostly because they are references by the “good tags”.

The Debian and .NET releases are naturally not planned together. When we started .NET 6, we saw that Debian “bullseye” would likely be released in 2021. We decided to take a bet on bullseye from the start of the release. We started releasing bullseye-based container images with .NET 6 Preview 1 and decided not to look back. The bet was that the .NET 6 release would lose the race with the bullseye release. By August 8th, we still didn’t know when bullseye would ship, leaving three months before our own release would go out, on November 8th. We didn’t want to ship a production .NET 6 on a preview Linux, but we held firm late to the plan that we’d lose this race.

We were pleasantly surprised when Debian 11 “bullseye” was released on August 14th. We lost the race but won the bet. That means that .NET 6 users get the best and latest Debian, by default, from day one. We believe that Debian 11 and .NET 6 will be a great combination for a lot of users. Sorry buster, we hit the bullseye.

Newer distro versions include newer major versions of various packages in their package feed and often get CVE fixes faster. That’s in addition to a newer kernel. Users are better served by a new distro version.

Looking a little further ahead, we’ll start planning support for Ubuntu 22.04 before long. Ubuntu is another Debian-family distro and popular with .NET developers. We hope to offer same-day support for the new Ubuntu LTS release.

Hat tip to Tianon Gravi for maintaining Debian images for the community and helping us when we have questions.

Dotnet monitor

dotnet monitor is an important diagnostics tool for containers. It has been available as a sidecar container image for some time, but in an unsupported “experimental” status. As part of .NET 6, we are releasing a .NET 6-based dotnet monitor image that is fully-supported in production.

dotnet monitor is already in use by Azure App Service as an implementation detail of their ASP.NET Core Linux diagnostics experience. This is one of the intended scenarios, building on top of dotnet monitor to provide higher-level and higher-value experiences.

You can pull the new image now:

docker pull mcr.microsoft.com/dotnet/monitor:6.0

dotnet monitor makes it easier to access diagnostic information — logs, traces, process dumps — from a .NET process. It is easy to get access to all the diagnostic information you want on your desktop machine, however, those same familiar techniques might not work in production using containers, for example. dotnet monitor provides a unified way to collect these diagnostic artifacts regardless of whether running on your desktop machine or in a Kubernetes cluster. There are two different mechanisms for collection of these diagnostic artifacts:

  • An HTTP API for ad-hoc collection of artifacts. You can call these API endpoints when you already know your application is experiencing an issue and you are interested in gathering more information.
  • Triggers for rule-based configuration for always-on collection of artifacts. You may configure rules to collect diagnostic data when a desired condition is met, for example, collect a process dump when you have sustained high CPU.

dotnet monitor provides a common diagnostic API for .NET apps that works everywhere you want with any tools you want. The “common API” isn’t a .NET API but a web API that you can call and query. dotnet monitor includes an ASP.NET web server that directly interacts with and exposes data from a diagnostics server in the .NET runtime. The design of dotnet monitor enables high-performance monitoring in production and secure use to gate access to privileged information. dotnet monitor interacts with the runtime — across container boundaries — via a non-internet-addressable unix domain socket. That model communication model is a perfect fit for this use case.

Structured JSON logs

The JSON formatter is now the default console logger in the aspnet .NET 6 container image. The default in .NET 5 was set to the simple console formatter. This change was made in order to have a default configuration that works with automated tools that rely on a machine-readable format like JSON.

The output now looks like the following for the aspnet image:

$ docker run --rm -it -p 8000:80 mcr.microsoft.com/dotnet/samples:aspnetapp
{"EventId":60,"LogLevel":"Warning","Category":"Microsoft.AspNetCore.DataProtection.Repositories.FileSystemXmlRepository","Message":"Storing keys in a directory u0027/root/.aspnet/DataProtection-Keysu0027 that may not be persisted outside of the container. Protected data will be unavailable when container is destroyed.","State":{"Message":"Storing keys in a directory u0027/root/.aspnet/DataProtection-Keysu0027 that may not be persisted outside of the container. Protected data will be unavailable when container is destroyed.","path":"/root/.aspnet/DataProtection-Keys","{OriginalFormat}":"Storing keys in a directory u0027{path}u0027 that may not be persisted outside of the container. Protected data will be unavailable when container is destroyed."}}
{"EventId":35,"LogLevel":"Warning","Category":"Microsoft.AspNetCore.DataProtection.KeyManagement.XmlKeyManager","Message":"No XML encryptor configured. Key {86cafacf-ab57-434a-b09c-66a929ae4fd7} may be persisted to storage in unencrypted form.","State":{"Message":"No XML encryptor configured. Key {86cafacf-ab57-434a-b09c-66a929ae4fd7} may be persisted to storage in unencrypted form.","KeyId":"86cafacf-ab57-434a-b09c-66a929ae4fd7","{OriginalFormat}":"No XML encryptor configured. Key {KeyId:B} may be persisted to storage in unencrypted form."}}
{"EventId":14,"LogLevel":"Information","Category":"Microsoft.Hosting.Lifetime","Message":"Now listening on: http://[::]:80","State":{"Message":"Now listening on: http://[::]:80","address":"http://[::]:80","{OriginalFormat}":"Now listening on: {address}"}}
{"EventId":0,"LogLevel":"Information","Category":"Microsoft.Hosting.Lifetime","Message":"Application started. Press Ctrlu002BC to shut down.","State":{"Message":"Application started. Press Ctrlu002BC to shut down.","{OriginalFormat}":"Application started. Press Ctrlu002BC to shut down."}}
{"EventId":0,"LogLevel":"Information","Category":"Microsoft.Hosting.Lifetime","Message":"Hosting environment: Production","State":{"Message":"Hosting environment: Production","envName":"Production","{OriginalFormat}":"Hosting environment: {envName}"}}
{"EventId":0,"LogLevel":"Information","Category":"Microsoft.Hosting.Lifetime","Message":"Content root path: /app","State":{"Message":"Content root path: /app","contentRoot":"/app","{OriginalFormat}":"Content root path: {contentRoot}"}}

The logger format type can be changed by setting or unsetting the Logging__Console__FormatterName environment variable or via code change (see Console log formatting for more details).

After changing it, you will see output like the following (just like .NET 5):

$ docker run --rm -it -p 8000:80 -e Logging__Console__FormatterName="" mcr.microsoft.com/dotnet/samples:aspnetapp
warn: Microsoft.AspNetCore.DataProtection.Repositories.FileSystemXmlRepository[60]
      Storing keys in a directory '/root/.aspnet/DataProtection-Keys' that may not be persisted outside of the container. Protected data will be unavailable when container is destroyed.
warn: Microsoft.AspNetCore.DataProtection.KeyManagement.XmlKeyManager[35]
      No XML encryptor configured. Key {8d4ddd1d-ccfc-4898-9fe1-3e7403bf23a0} may be persisted to storage in unencrypted form.
info: Microsoft.Hosting.Lifetime[14]
      Now listening on: http://[::]:80
info: Microsoft.Hosting.Lifetime[0]
      Application started. Press Ctrl+C to shut down.
info: Microsoft.Hosting.Lifetime[0]
      Hosting environment: Production
info: Microsoft.Hosting.Lifetime[0]
      Content root path: /app

Note: This change doesn’t affect the .NET SDK on your developer machine, like with dotnet run. This change is specific to the aspnet container image.

Support for OpenTelemetry Metrics

We’ve been adding support for OpenTelemetry for the last couple .NET versions, as part of our focus on observability. In .NET 6, we’re adding support for the OpenTelemetry Metrics API. By adding support for OpenTelemetry, your apps can seamlessly interoperate with other OpenTelemetry systems.

System.Diagnostics.Metrics is the .NET implementation of the OpenTelemetry Metrics API specification. The Metrics APIs are designed explicitly for processing raw measurements, with the intent of producing continuous summaries of those measurements, efficiently and simultaneously.

The APIs include the Meter class which can be used to create instrument objects. The APIs expose four instrument classes: Counter, Histogram, ObservableCounter, and ObservableGauge to support different metrics scenarios. Also, the APIs expose the MeterListener class to allow listening to the instrument’s recorded measurement for aggregation and grouping purposes.

The OpenTelemetry .NET implementation will be extended to use these new APIs, which add support for Metrics observability scenarios.

Library Measurement Recording Example

    Meter meter = new Meter("io.opentelemetry.contrib.mongodb", "v1.0");
    Counter<int> counter = meter.CreateCounter<int>("Requests");
    counter.Add(1);
    counter.Add(1, KeyValuePair.Create<string, object>("request", "read"));

Listening Example

    MeterListener listener = new MeterListener();
    listener.InstrumentPublished = (instrument, meterListener) =>
    {
        if (instrument.Name == "Requests" && instrument.Meter.Name == "io.opentelemetry.contrib.mongodb")
        {
            meterListener.EnableMeasurementEvents(instrument, null);
        }
    };
    listener.SetMeasurementEventCallback<int>((instrument, measurement, tags, state) =>
    {
        Console.WriteLine($"Instrument: {instrument.Name} has recorded the measurement {measurement}");
    });
    listener.Start();

Windows Forms

We have continued to make key improvements in Windows Forms. .NET 6 includes better accessibility for controls, the ability to set an application-wide default font, template updates and others.

Accessibility improvements

In this release, we added UIA providers for CheckedListBox, LinkLabel, Panel, ScrollBar, TabControl and TrackBar that enable tools like Narrator, and test automation to interact with the elements of an application.

Default font

You can now set a default font for an application with Application.SetDefaultFont.

void Application.SetDefaultFont(Font font)

Minimal applications

The following is a minimal Windows Forms application with .NET 6:

class Program
{
    [STAThread]
    static void Main()
    {
        ApplicationConfiguration.Initialize();
        Application.Run(new Form1());
    }
}

As part of the .NET 6 release, we’ve been updating most of the templates to them more modern and minimal, including with Windows Forms. We decided to keep the Windows Forms template a bit more traditional, in part because of the need for the [STAThread] attribute to apply to the application entrypoint. However, there is more a play than immediately meets the eye.

ApplicationConfiguration.Initialize() is a source generated API that behind the scenes emits the following calls:

Application.EnableVisualStyles();
Application.SetCompatibleTextRenderingDefault(false);
Application.SetDefaultFont(new Font(...));
Application.SetHighDpiMode(HighDpiMode.SystemAware);

The parameters of these calls are configurable via MSBuild properties in csproj or props files.

The Windows Forms designer in Visual Studio 2022 is also aware of these properties (for now it only reads the default font), and can show you your application as it would look at runtime:

Application bootstrap, Announcing .NET 6 — The Fastest .NET Yet

Template updates

Windows Forms templates for C# have been updated to support the new application bootstrap, global using directives, file-scoped namespaces, and nullable reference types.

More runtime designers

Now you can build a general-purpose designer (for example, a report designer) since .NET 6 has all the missing pieces for designers and designer-related infrastructure. For more information, see this blog post.

Single-file Apps

In .NET 6, in-memory single file apps have been enabled for Windows and macOS. In .NET 5, this deployment type was limited to Linux. You can now publish a single-file binary that is both deployed and launched as a single file, for all supported OSes. Single files apps no longer extract any core runtime assemblies to temporary directories.

This expanded capability is based on a building block called “superhost”. “apphost” is the executable that launches your application in the non-single-file case, like myapp.exe or ./myapp. Apphost contains code to find the runtime, load it, and start your app with that runtime. Superhost still performs some of those tasks but uses a statically linked copy of all the CoreCLR native binaries. Static linking is the approach we use to enable the single file experience.

Native dependencies (like that come with a NuGet package) are the notable exception to single-file embedding. They are not included in the single file by default. For instance, WPF native dependencies are not part of the superhost, resulting in additional files beside the single file app. You can use the setting IncludeNativeLibrariesForSelfExtract to embed and extract native-dependencies.

Static Analysis

We’ve improved single-file analyzers to allow for custom warnings. If you have an API which doesn’t work in single-file publishing you can now mark it with the [RequiresAssemblyFiles] attribute, and a warning will appear if the analyzer is enabled. Adding that attribute will also silence all warnings related to single-file in the method, so you can use the warning to propagate warnings upward to your public API.

The single-file analyzer is automatically enabled for exe projects when PublishSingleFile is set to true, but you can also enable it for any project by setting EnableSingleFileAnalysis to true. This is helpful if you want to support a library being part of a single file app.

In .NET 5, we added warning for Assembly.Location and a few other APIs which behave differently in single-file bundles.

Compression

Single-file bundles now support compression, which can be enabled by setting the property EnableCompressionInSingleFile to true. At runtime, files are decompressed to memory as necessary. Compression can provide huge space savings for some scenarios.

Let’s look at single file publishing, with and without compression, used with NuGet Package Explorer.

Without compression: 172 MB

Announcing .NET 6 — The Fastest .NET Yet

With compression: 71.6 MB

Announcing .NET 6 — The Fastest .NET Yet

Compression can significantly increase the startup time of the application, especially on Unix platforms. Unix platforms have a no-copy fast-start path that can’t be used with compression. You should test your app after enabling compression to see if the additional startup cost is acceptable.

Single-file debugging

Single-file apps can currently only be debugged using platform debuggers, like WinDBG. We are looking at adding Visual Studio debugging with a later build of Visual Studio 2022.

Single-file signing on macOS

Single file apps now satisfy Apple notarization and signing requirements on macOS. The specific changes relate to the way that we construct single file apps in terms of discrete file layout.

Apple started enforcing new requirements for signing and notarization with macOS Catalina. We’ve been working closely with Apple to understand the requirements, and to find solutions that enable a development platform like .NET to work well in that environment. We’ve made product changes and documented user workflows to satisfy Apple requirements in each of the last few .NET releases. One of the remaining gaps was single-file signing, which is a requirement to distribute a .NET app on macOS, including in the macOS store.

IL trimming

The team has been working on IL trimming for multiple releases. .NET 6 represents a major step forward on that journey. We’ve been working to make a more aggressive trimming mode safe and predictable and as a result have confidence to make it the default. TrimMode=link was previously an opt-in feature and is now the default.

We’ve had a three-pronged strategy to trimming:

  • Improve trim-ability of the platform.
  • Annotate the platform to provide better warnings and to enable others to do the same.
  • Based on that, make the default trim mode more aggressive so that it is easy to make apps small.

Trimming has previously been in preview because of the unreliable results for apps which use unannotated reflection. With trim warnings, the experience should now be predictable. Apps without trim warnings should trim correctly and observe no change in behavior when running. Currently, only the core .NET libraries have been fully annotated for trimming, but we hope to see the ecosystem annotate for trimming and become trim compatible

Reducing app size

Let’s take a look at this trimming improvement using crossgen, which is one of the SDK tools. It can be trimmed with only a few trim warnings, which the crossgen team was able to resolve.

First, let’s look at publishing crossgen as a self-contained app without trimming. It is 80 MB (which includes the .NET runtime and all the libraries).

image

We can then try out the (now legacy) .NET 5 default trim mode, copyused. The result drops to 55 MB.

image

The new .NET 6 default trim mode, link, drops the self-contained file size much further, to 36MB.

image

We hope that the new link trim mode aligns much better with the expectations for trimming: significant savings and predictable results.

Warnings enabled by default

Trim warnings tell you about places where trimming may remove code that’s used at runtime. These warnings were previously disabled by default because the warnings were very noisy, largely due to the .NET platform not participating in trimming as a first class scenario.

We’ve annotated large portions of the .NET libraries so that they produce accurate trim warnings. As a result, we felt it was time to enable trimming warnings by default. The ASP.NET Core and Windows Desktop runtime libraries have not been annotated. We plan to annotate ASP.NET service components next (after .NET 6). We’re hoping to see the community annotate NuGet libraries now that .NET 6 is released.

You can disable warnings by setting <SuppressTrimAnalysisWarnings> to true.

More information:

Shared with Native AOT

We’ve implemented the same trimming warnings for the Native AOT experiment as well, which should improve the Native AOT compilation experience in much the same way.

Math

We’ve improved Math APIs significantly. Some folks in the community have been enjoying these improvements already.

Performance-oriented APIs

Performance-oriented math APIs have been added to System.Math. Their implementation is hardware accelerated if the underlying hardware supports it.

New APIs:

  • SinCos for simultaneously computing Sin and Cos.
  • ReciprocalEstimate for computing an approximate of 1 / x.
  • ReciprocalSqrtEstimate for computing an approximate of 1 / Sqrt(x).

New overloads:

  • Clamp, DivRem, Min, and Max supporting nint and nuint.
  • Abs and Sign supporting nint.
  • DivRem variants which return a tuple.

Perf improvements:

BigInteger Performance

Parsing of BigIntegers from both decimal and hexadecimal strings has been improved. We see improvements of up to 89%, as demonstrated in the following chart (lower is better).

graph, Announcing .NET 6 — The Fastest .NET Yet

Credit to Joseph Da Silva.

Complex APIs now annotated as readonly

Various System.Numerics.Complex APIs are now annotated as readonly to ensure that no copy is made for readonly values or values passed by in.

Credit to hrrrrustic.

BitConverter now supports floating-point to unsigned integer bitcasts

BitConverter now supports DoubleToUInt64Bits, HalfToUInt16Bits, SingleToUInt32Bits, UInt16BitsToHalf, UInt32BitsToSingle, and UInt64BitsToDouble. This should make it easier to do floating-point bit manipulation when required.

Credit to Michal Petryka.

BitOperations supports additional functionality

BitOperations now supports IsPow2, RoundUpToPowerOf2, and provides nint/nuint overloads for existing functions.

Credit to John Kelly, Huo Yaoyuan, and Robin Lindner.

Vector<T>, Vector2, Vector3, and Vector4 improvements

Vector<T> now supports the nint and nuint primitive types, added in C# 9. For example, this change should make it simpler to use SIMD instructions with pointers or platform-dependent length types.

Vector<T> now supports a Sum method to simplify needing to compute the “horizontal sum” of all elements in the vector. Credit to Ivan Zlatanov.

Vector<T> now supports a generic As<TFrom, TTo> method to simplify dealing with vectors in generic contexts where the concrete type isn’t known. Credit to Huo Yaoyuan

Overloads supporting Span<T> were added to Vector2, Vector3, and Vector4 to improve the experience when needing to load or store vector types.

Better parsing of standard numeric formats

We’ve improved the parser for the standard numeric types, specifically for .ToString and .TryFormat. They will now understand requests for precision > 99 decimal places and will provide accurate results to that many digits. Also, the parser now better supports trailing zeros in the Parse method.

The following examples demonstrate before and after behavior.

  • 32.ToString("C100") -> C132
    • .NET 6: $32.0000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000
    • .NET 5: We had an artificial limitation in the formatting code to only handle a precision of <= 99. For precision >= 100, we instead interpreted the input as a custom format.
  • 32.ToString("H99") -> throw a FormatException
    • .NET 6: throws a FormatException
    • This is correct behavior, but it’s called here to contrast with the next example.
  • 32.ToString("H100") -> H132
    • .NET 6: throw a FormatException
    • .NET 5: H is an invalid format specifier. So, we should’ve thrown a FormatException. Instead, our incorrect behavior of interpreting precision >= 100 as custom formats meant we returned wrong values.
  • double.Parse("9007199254740997.0") -> 9007199254740998
    • .NET 6: 9007199254740996.
    • .NET 5: 9007199254740997.0 is not exactly representable in the IEEE 754 format. With our current rounding scheme, the correct return value should have been 9007199254740996. However, the last .0 portion of the input was forcing the parser to incorrectly round the result and return 9007199254740998.

System.Text.Json

System.Text.Json provides multiple high-performance APIs for processing JSON documents. Over the last few releases, we’ve added new functionality to further improve JSON processing performance and to mitigate blockers for people wanting to migrate from NewtonSoft.Json. This release includes continues on that path and is major step forward on performance, particularly with the serializer source generator.

JsonSerializer source generation

Note: Apps that used source generation with .NET 6 RC1 or earlier builds should be re-compiled.

The backbone of nearly all .NET serializers is reflection. Reflection is a great capability for certain scenarios, but not as the basis of high-performance cloud-native applications (which typically (de)serialize and process a lot of JSON documents). Reflection is a problem for startup, memory usage, and assembly trimming.

The alternative to runtime reflection is compile-time source generation. In .NET 6, we are including a new source generator as part of System.Text.Json. The JSON source generator works in conjunction with JsonSerializer and can be configured in multiple ways.

It can provide the following benefits:

  • Reduce start-up time
  • Improve serialization throughput
  • Reduce private memory usage
  • Remove runtime use of System.Reflection and System.Reflection.Emit
  • IL trimming compatibility

By default, the JSON source generator emits serialization logic for the given serializable types. This delivers higher performance than using the existing JsonSerializer methods by generating source code that uses Utf8JsonWriter directly. In short, source generators offer a way of giving you a different implementation at compile-time in order to make the runtime experience better.

Given a simple type:

namespace Test
{
    internal class JsonMessage
    {
        public string Message { get; set; }
    }
}

The source generator can be configured to generate serialization logic for instances of the example JsonMessage type. Note that the class name JsonContext is arbitrary. You can use whichever class name you want for the generated source.

using System.Text.Json.Serialization;

namespace Test
{
    [JsonSerializable(typeof(JsonMessage)]
    internal partial class JsonContext : JsonSerializerContext
    {
    }
}

The serializer invocation with this mode could look like the following example. This example provides the best possible performance.

using MemoryStream ms = new();
using Utf8JsonWriter writer = new(ms);

JsonSerializer.Serialize(jsonMessage, JsonContext.Default.JsonMessage);
writer.Flush();

// Writer contains:
// {"Message":"Hello, world!"}

The fastest and most optimized source generation mode — based on Utf8JsonWriter — is currently only available for serialization. Similar support for deserialization — based on Utf8JsonReader — may be provided in the future depending on your feedback.

The source generator also emits type-metadata initialization logic that can benefit deserialization as well. To deserialize an instance of JsonMessage using pre-generated type metadata, you can do the following:

JsonSerializer.Deserialize(json, JsonContext.Default.JsonMessage);

JsonSerializer support for IAsyncEnumerable

You can now (de)serialize IAsyncEnumerable<T> JSON arrays with System.Text.Json.The following examples use streams as a representation of any async source of data. The source could be files on a local machine, or results from a database query or web service API call.

JsonSerializer.SerializeAsync has been updated to recognize and provide special handing for IAsyncEnumerable values.

using System;
using System.Collections.Generic;
using System.IO;
using System.Text.Json;

static async IAsyncEnumerable<int> PrintNumbers(int n)
{
    for (int i = 0; i < n; i++) yield return i;
}

using Stream stream = Console.OpenStandardOutput();
var data = new { Data = PrintNumbers(3) };
await JsonSerializer.SerializeAsync(stream, data); // prints {"Data":[0,1,2]}

IAsyncEnumerable values are only supported using the asynchronous serialization methods. Attempting to serialize using the synchronous methods will result in a NotSupportedException being thrown.

Streaming deserialization required a new API that returns IAsyncEnumerable<T>. We added the JsonSerializer.DeserializeAsyncEnumerable method for this purpose, as you can see in the following example.

using System;
using System.IO;
using System.Text;
using System.Text.Json;

var stream = new MemoryStream(Encoding.UTF8.GetBytes("[0,1,2,3,4]"));
await foreach (int item in JsonSerializer.DeserializeAsyncEnumerable<int>(stream))
{
    Console.WriteLine(item);
}

This example will deserialize elements on-demand and can be useful when consuming particularly large data streams. It only supports reading from root-level JSON arrays, although that could potentially be relaxed in the future based on feedback.

The existing DeserializeAsync method nominally supports IAsyncEnumerable<T>, but within the confines of its non-streaming method signature. It must return the final result as a single value, as you can see in the following example.

using System;
using System.Collections.Generic;
using System.IO;
using System.Text;
using System.Text.Json;

var stream = new MemoryStream(Encoding.UTF8.GetBytes(@"{""Data"":[0,1,2,3,4]}"));
var result = await JsonSerializer.DeserializeAsync<MyPoco>(stream);
await foreach (int item in result.Data)
{
    Console.WriteLine(item);
}

public class MyPoco
{
    public IAsyncEnumerable<int> Data { get; set; }
}

In this example, the deserializer will have buffered all IAsyncEnumerable contents in memory before returning the deserialized object. This is because the deserializer needs to have consumed the entire JSON value before returning a result.

System.Text.Json: Writable DOM Feature

The writeable JSON DOM feature adds a new straightforward and high-performance programming model for System.Text.Json. This new API is attractive since it avoids needing strongly-typed serialization contracts, and the DOM is mutable as opposed to the existing JsonDocument type.

This new API has the following benefits:

  • A lightweight alternative to serialization for cases when using POCO types is not possible or desired, or when a JSON schema is not fixed and must be inspected.
  • Enables efficient modification of a subset of a large tree. For example, it is possible to efficiently navigate to a subsection of a large JSON tree and read an array or deserialize a POCO from that subsection. LINQ can also be used with that.

The following example demonstrates the new programming model.

    // Parse a JSON object
    JsonNode jNode = JsonNode.Parse("{"MyProperty":42}");
    int value = (int)jNode["MyProperty"];
    Debug.Assert(value == 42);
    // or
    value = jNode["MyProperty"].GetValue<int>();
    Debug.Assert(value == 42);

    // Parse a JSON array
    jNode = JsonNode.Parse("[10,11,12]");
    value = (int)jNode[1];
    Debug.Assert(value == 11);
    // or
    value = jNode[1].GetValue<int>();
    Debug.Assert(value == 11);

    // Create a new JsonObject using object initializers and array params
    var jObject = new JsonObject
    {
        ["MyChildObject"] = new JsonObject
        {
            ["MyProperty"] = "Hello",
            ["MyArray"] = new JsonArray(10, 11, 12)
        }
    };

    // Obtain the JSON from the new JsonObject
    string json = jObject.ToJsonString();
    Console.WriteLine(json); // {"MyChildObject":{"MyProperty":"Hello","MyArray":[10,11,12]}}

    // Indexers for property names and array elements are supported and can be chained
    Debug.Assert(jObject["MyChildObject"]["MyArray"][1].GetValue<int>() == 11);

ReferenceHandler.IgnoreCycles

JsonSerializer (System.Text.Json) now supports the ability to ignore cycles when serializing an object graph. The ReferenceHandler.IgnoreCycles option has similar behavior as Newtonsoft.Json ReferenceLoopHandling.Ignore. One key difference is that the System.Text.Json implementation replaces reference loops with the null JSON token instead of ignoring the object reference.

You can see the behavior of ReferenceHandler.IgnoreCycles in the following example. In this case, the Next property is serialized as null since it otherwise creates a cycle.

class Node
{
    public string Description { get; set; }
    public object Next { get; set; }
}

void Test()
{
    var node = new Node { Description = "Node 1" };
    node.Next = node;

    var opts = new JsonSerializerOptions { ReferenceHandler = ReferenceHandler.IgnoreCycles };

    string json = JsonSerializer.Serialize(node, opts);
    Console.WriteLine(json); // Prints {"Description":"Node 1","Next":null}
}

Source build

With source build, you can build the .NET SDK on your own machine from source with just a few commands. Let me explain why this project is important.

Source build is a scenario and also infrastructure that we’ve been working on in collaboration with Red Hat since before shipping .NET Core 1.0. Several years later, we’re very close to delivering a fully automated version of it. For Red Hat Enterprise Linux (RHEL) .NET users, this capability is a big deal. Red Hat tells us that .NET has grown to become an important developer platform for their ecosystem. Nice!

The gold standard for Linux distros is to build open source code using compilers and toolchains that are part of the distro archive. That works for the .NET runtime (written in C++), but not for any of the code written in C#. For C# code, we use a two-pass build mechanism to satisfy distro requirements. It’s a bit complicated, but it’s important to understand the flow.

Red Hat builds .NET SDK source using the Microsoft binary build of the .NET SDK (#1) to produce a pure open source binary build of the SDK (#2). After that, the same SDK source code is built again using this fresh build of the SDK (#2) to produce a provably open source SDK (#3). This final binary build of the .NET SDK (#3) is then made available to RHEL users. After that, Red Hat can use this same SDK (#3) to build new .NET versions and no longer needs to use the Microsoft SDK to build monthly updates.

That process may be surprising and confusing. Open source distros need to be built by open source tools. This pattern ensures that the Microsoft build of the SDK isn’t required, either by intention or accident. There is a higher bar, as a developer platform, to being included in a distro than just using a compatible license. The source build project has enabled .NET to meet that bar.

The deliverable for source build is a source tarball. The source tarball contains all the source for the SDK (for a given release). From there, Red Hat (or another organization) can build their own version of the SDK. Red Hat policy requires using a built-from-source toolchain to produce a binary tar ball, which is why they use a two-pass methodology. But this two-pass method is not required for source build itself.

It is common in the Linux ecosystem to have both source and binary packages or tarballs available for a given component. We already had binary tarballs available and now have source tarballs as well. That makes .NET match the standard component pattern.

The big improvement in .NET 6 is that the source tarball is a now a product of our build. It used to require significant manual effort to produce, which also resulted in significant latency delivering the source tarball to Red Hat. Neither party was happy about that.

We’ve been working closely with Red Hat on this project for five+ years. It has succeeded, in no small part, due to the efforts of the excellent Red Hat engineers we’ve had the pleasure of working with. Other distros and organizations have and will benefit from their efforts.

As a side note, source build is a big step towards reproducible builds, which we also strongly believe in. The .NET SDK and C# compiler have significant reproducible build capabilities.

Libraries APIs

The following APIs have been added, in the addition to the ones already covered.

WebSocket Compression

Compression is important for any data transmitted over a network. WebSockets now enable compression. We used an implementation of permessage-deflate extension for WebSockets, RFC 7692. It allows compressing WebSockets message payloads using the DEFLATE algorithm. This feature was one of the top user requests for Networking on GitHub.

Compression used with encryption may lead to attacks, like CRIME and BREACH. It means that a secret cannot be sent together with user-generated data in a single compression context, otherwise that secret could be extracted. To bring a user’s attention to these implications and help them weigh the risks, we named one of the key APIs DangerousDeflateOptions. We also added the ability to turn off compression for specific messages, so if the user would want to send a secret, they could do that securely without compression.

The memory footprint of the WebSocket when compression is disabled was reduced by about 27%.

Enabling the compression from the client side is easy, as you can see in the following example. However, bear in mind that the server can negotiate the settings, such as, requesting a smaller window or denying compression completely.

var cws = new ClientWebSocket();
cws.Options.DangerousDeflateOptions = new WebSocketDeflateOptions()
{
    ClientMaxWindowBits = 10,
    ServerMaxWindowBits = 10
};

WebSocket compression support for ASP.NET Core was also added.

Credit to Ivan Zlatanov.

Socks proxy support

SOCKS is a proxy server implementation that can process any TCP or UDP traffic, making it a very versatile system. It is a long-standing community request that has been added to .NET 6.

This change adds support for Socks4, Socks4a, and Socks5. For example, it enables testing external connections via SSH or connecting to the Tor network.

The WebProxy class now accepts socks schemes, as you can see in the following example.

var handler = new HttpClientHandler
{
    Proxy = new WebProxy("socks5://127.0.0.1", 9050)
};
var httpClient = new HttpClient(handler);

Credit to Huo Yaoyuan.

Microsoft.Extensions.Hosting — ConfigureHostOptions API

We added a new ConfigureHostOptions API on IHostBuilder to make application setup simpler (for example, configuring the shutdown timeout):

using HostBuilder host = new()
    .ConfigureHostOptions(o =>
    {
        o.ShutdownTimeout = TimeSpan.FromMinutes(10);
    })
    .Build();

host.Run();

In .NET 5, configuring the host options was a bit more complicated:

using HostBuilder host = new()
    .ConfigureServices(services =>
    {
        services.Configure<HostOptions>(o =>
        {
            o.ShutdownTimeout = TimeSpan.FromMinutes(10);
        });
    })
    .Build();

host.Run();

Microsoft.Extensions.DependencyInjection — CreateAsyncScope APIs

The CreateAsyncScope API was created to handle disposal of IAsyncDisposable services. Previously, you might have noticed that disposal of a IAsyncDisposable service provider could throw an InvalidOperationException exception.

The following example demonstrates the new pattern, using CreateAsyncScope to enable safe use of the using statement.

await using (var scope = provider.CreateAsyncScope())
{
    var foo = scope.ServiceProvider.GetRequiredService<Foo>();
}

The following example demonstrate the existing problem case:

using System;
using System.Threading.Tasks;
using Microsoft.Extensions.DependencyInjection;

await using var provider = new ServiceCollection()
        .AddScoped<Foo>()
        .BuildServiceProvider();

// This using can throw InvalidOperationException
using (var scope = provider.CreateScope())
{
    var foo = scope.ServiceProvider.GetRequiredService<Foo>();
}

class Foo : IAsyncDisposable
{
    public ValueTask DisposeAsync() => default;
}

The following pattern was the previous suggested workaround to avoid the exception. It is no longer required.

var scope = provider.CreateScope();
var foo = scope.ServiceProvider.GetRequiredService<Foo>();
await ((IAsyncDisposable)scope).DisposeAsync();

Credit to Martin Björkström.

Microsoft.Extensions.Logging — compile-time source generator

.NET 6 introduces the LoggerMessageAttribute type. This attribute is part of the Microsoft.Extensions.Logging namespace, and when used, it source-generates performant logging APIs. The source-generation logging support is designed to deliver a highly usable and highly performant logging solution for modern .NET applications. The auto-generated source code relies on the ILogger interface in conjunction with LoggerMessage.Define functionality.

The source generator is triggered when the LoggerMessageAttribute is used on partial logging methods. When triggered, it is either able to autogenerate the implementation of the partial methods it’s decorating or produce compile-time diagnostics with hints about proper usage. The compile-time logging solution is typically considerably faster at runtime than existing logging approaches. It achieves this by eliminating boxing, temporary allocations, and copies to the maximum extent possible.

There are benefits over manually using LoggerMessage.Define APIs directly:

  • Shorter and simpler syntax: Declarative attribute usage rather than coding boilerplate.
  • Guided developer experience: The generator gives warnings to help developers do the right thing.
  • Support for an arbitrary number of logging parameters. LoggerMessage.Define supports a maximum of six.
  • Support for dynamic log level. This is not possible with LoggerMessage.Define alone.

To use the LoggerMessageAttribute, the consuming class and method need to be partial. The code generator is triggered at compile time and generates an implementation of the partial method.

public static partial class Log
{
    [LoggerMessage(EventId = 0, Level = LogLevel.Critical, Message = "Could not open socket to `{hostName}`")]
    public static partial void CouldNotOpenSocket(ILogger logger, string hostName);
}

In the preceding example, the logging method is static and the log level is specified in the attribute definition. When using the attribute in a static context, the ILogger instance is required as a parameter. You may choose to use the attribute in a non-static context as well. For more examples and usage scenarios, visit the compile-time logging source generator documentation.

System.Linq — Enumerable support for Index and Range parameters

The Enumerable.ElementAt method now accepts indices from the end of the enumerable, as you can see in the following example.

Enumerable.Range(1, 10).ElementAt(^2); // returns 9

An Enumerable.Take overload has been added that accepts Range parameters. It simplifies taking slices of enumerable sequences:

  • source.Take(..3) instead of source.Take(3)
  • source.Take(3..) instead of source.Skip(3)
  • source.Take(2..7) instead of source.Take(7).Skip(2)
  • source.Take(^3..) instead of source.TakeLast(3)
  • source.Take(..^3) instead of source.SkipLast(3)
  • source.Take(^7..^3) instead of source.TakeLast(7).SkipLast(3).

Credit to @dixin.

System.Linq — TryGetNonEnumeratedCount

The TryGetNonEnumeratedCount method attempts to obtain the count of the source enumerable without forcing an enumeration. This approach can be useful in scenarios where it is useful to preallocate buffers ahead of enumeration, as you can see in the following example.

List<T> buffer = source.TryGetNonEnumeratedCount(out int count) ? new List<T>(capacity: count) : new List<T>();
foreach (T item in source)
{
    buffer.Add(item);
}

TryGetNonEnumeratedCount checks for sources implementing ICollection/ICollection<T> or takes advantage of some of the internal optimizations employed by Linq.

System.Linq — DistinctBy/UnionBy/IntersectBy/ExceptBy

New variants have been added to the set operations that allow specifying equality using key selector functions, as you can see in the following example.

Enumerable.Range(1, 20).DistinctBy(x => x % 3); // {1, 2, 3}

var first = new (string Name, int Age)[] { ("Francis", 20), ("Lindsey", 30), ("Ashley", 40) };
var second = new (string Name, int Age)[] { ("Claire", 30), ("Pat", 30), ("Drew", 33) };
first.UnionBy(second, person => person.Age); // { ("Francis", 20), ("Lindsey", 30), ("Ashley", 40), ("Drew", 33) }

System.Linq — MaxBy/MinBy

MaxBy and MinBy methods allow finding maximal or minimal elements using a key selector, as you can see in the following example.

var people = new (string Name, int Age)[] { ("Francis", 20), ("Lindsey", 30), ("Ashley", 40) };
people.MaxBy(person => person.Age); // ("Ashley", 40)

System.Linq — Chunk

Chunk can be used to chunk a source enumerable into slices of a fixed size, as you can see in the following example.

IEnumerable<int[]> chunks = Enumerable.Range(0, 10).Chunk(size: 3); // { {0,1,2}, {3,4,5}, {6,7,8}, {9} }

Credit to Robert Andersson.

System.Linq — FirstOrDefault/LastOrDefault/SingleOrDefault overloads taking default parameters

The existing FirstOrDefault/LastOrDefault/SingleOrDefault methods return default(T) if the source enumerable is empty. New overloads have been added that accept a default parameter to be returned in that case, as you can see in the following example.

Enumerable.Empty<int>().SingleOrDefault(-1); // returns -1

Credit to @Foxtrek64.

System.Linq — Zip overload accepting three enumerables

The Zip method now supports combining three enumerables, as you can see in the following example.

var xs = Enumerable.Range(1, 10);
var ys = xs.Select(x => x.ToString());
var zs = xs.Select(x => x % 2 == 0);

foreach ((int x, string y, bool z) in Enumerable.Zip(xs,ys,zs))
{
}

Credit to Huo Yaoyuan.

PriorityQueue

PriorityQueue<TElement, TPriority> (System.Collections.Generic) is a new collection that enables adding new items with a value and a priority. On dequeue the PriorityQueue returns the element with the lowest priority value. You can think of this new collection as similar to Queue<T> but that each enqueued element has a priority value that affects the behavior of dequeue.

The following sample demonstrates the behavior of PriorityQueue<string, int>.

// creates a priority queue of strings with integer priorities
var pq = new PriorityQueue<string, int>();

// enqueue elements with associated priorities
pq.Enqueue("A", 3);
pq.Enqueue("B", 1);
pq.Enqueue("C", 2);
pq.Enqueue("D", 3);

pq.Dequeue(); // returns "B"
pq.Dequeue(); // returns "C"
pq.Dequeue(); // either "A" or "D", stability is not guaranteed.

Credit to Patryk Golebiowski.

Faster handling of structs as Dictionary values

CollectionsMarshal.GetValueRef is a new unsafe API that makes updating struct values in Dictionaries faster. The new API is intended for high performance scenarios, not for general purpose use. It returns a ref to the struct value, which can then be updated in place with typical techniques.

The following example demonstrates using the new API:

ref MyStruct value = CollectionsMarshal.GetValueRef(dictionary, key);
// Returns Unsafe.NullRef<TValue>() if it doesn't exist; check using Unsafe.IsNullRef(ref value)
if (!Unsafe.IsNullRef(ref value))
{
    // Mutate in-place
    value.MyInt++;
}

Prior to this change, updating struct dictionary values could be expensive for high-performance scenarios, requiring a dictionary lookup and a copy to stack of the struct. Then after changing the struct, it would be assigned to the dictionary key again resulting in another look up and copy operation. This improvement reduces the key hashing to 1 (from 2) and removes all the struct copy operations.

Credit to Ben Adams.

New DateOnly and TimeOnly structs

Date- and time-only structs have been added, with the following characteristics:

  • Each represent one half of a DateTime, either only the date part, or only the time part.
  • DateOnly is ideal for birthdays, anniversary days, and business days. It aligns with SQL Server’s date type.
  • TimeOnly is ideal for recurring meetings, alarm clocks, and weekly business hours. It aligns with SQL Server’s time type.
  • Complements existing date/time types (DateTime, DateTimeOffset, TimeSpan, TimeZoneInfo).
  • In System namespace, shipped in CoreLib, just like existing related types.

Perf improvements to DateTime.UtcNow

This improvement has the following benefits:

  • Fixes 2.5x perf regression for getting the system time on Windows.
  • Utilizes a 5-minute sliding cache of Windows leap second data instead of fetching with every call.

Support for both Windows and IANA time zones on all platforms

This improvement has the following benefits:

  • Implicit conversion when using TimeZoneInfo.FindSystemTimeZoneById (https://github.com/dotnet/runtime/pull/49412)
  • Explicit conversion through new APIs on TimeZoneInfo: TryConvertIanaIdToWindowsId, TryConvertWindowsIdToIanaId, and HasIanaId (https://github.com/dotnet/runtime/issues/49407)
  • Improves cross-plat support and interop between systems that use different time zone types.
  • Removes need to use TimeZoneConverter OSS library. The functionality is now built-in.

Improved time zone display names

Time zone display names have been improved on Unix:

  • Removes ambiguity from the display names in the list returned by TimeZoneInfo.GetSystemTimeZones.
  • Leverages ICU / CLDR globalization data.
  • Unix only for now. Windows still uses the registry data. This may be changed later.

The following additional improvements were also made:

  • The UTC time zone’s display name and standard name were hardcoded to English and now uses the same language as the rest of the time zone data (CurrentUICulture on Unix, OS default language on Windows).
  • Time zone display names in Wasm use the non-localized IANA ID instead, due to size limitations.
  • TimeZoneInfo.AdjustmentRule nested class gets its BaseUtcOffsetDelta internal property made public and gets a new constructor that takes baseUtcOffsetDelta as a parameter. (https://github.com/dotnet/runtime/issues/50256)
  • TimeZoneInfo.AdjustmentRule also gets miscellaneous fixes for loading time zones on Unix (https://github.com/dotnet/runtime/pull/49733), (https://github.com/dotnet/runtime/pull/50131)

Improved support for Windows ACLs

System.Threading.AccessControl now includes improved support for interacting with Windows access control lists (ACLs). New overloads were added to the OpenExisting and TryOpenExisting methods for EventWaitHandle, Mutex and Semaphore. These overloads — with “security rights” instances — enable opening existing instances of threading synchronization objects that were created with special Windows security attributes.

This update matches APIs available in .NET Framework and has the same behavior.

The following examples demonstrate using these new APIs.

For Mutex:

var rights = MutexRights.FullControl;
string mutexName = "MyMutexName";

var security = new MutexSecurity();
SecurityIdentifier identity = new SecurityIdentifier(WellKnownSidType.BuiltinUsersSid, null);
MutexAccessRule accessRule = new MutexAccessRule(identity, rights, AccessControlType.Allow);
security.AddAccessRule(accessRule);

// createdMutex, openedMutex1 and openedMutex2 point to the same mutex
Mutex createdMutex = MutexAcl.Create(initiallyOwned: true, mutexName, out bool createdNew, security);
Mutex openedMutex1 = MutexAcl.OpenExisting(mutexName, rights);
MutexAcl.TryOpenExisting(mutexName, rights, out Mutex openedMutex2);

For Semaphore

var rights = SemaphoreRights.FullControl;
string semaphoreName = "MySemaphoreName";

var security = new SemaphoreSecurity();
SecurityIdentifier identity = new SecurityIdentifier(WellKnownSidType.BuiltinUsersSid, null);
SemaphoreAccessRule accessRule = new SemaphoreAccessRule(identity, rights, AccessControlType.Allow);
security.AddAccessRule(accessRule);

// createdSemaphore, openedSemaphore1 and openedSemaphore2 point to the same semaphore
Semaphore createdSemaphore = SemaphoreAcl.Create(initialCount: 1,  maximumCount: 3, semaphoreName, out bool createdNew, security);
Semaphore openedSemaphore1 = SemaphoreAcl.OpenExisting(semaphoreName, rights);
SemaphoreAcl.TryOpenExisting(semaphoreName, rights, out Semaphore openedSemaphore2);

For EventWaitHandle

var rights = EventWaitHandleRights.FullControl;
string eventWaitHandleName = "MyEventWaitHandleName";

var security = new EventWaitHandleSecurity();
SecurityIdentifier identity = new SecurityIdentifier(WellKnownSidType.BuiltinUsersSid, null);
EventWaitHandleAccessRule accessRule = new EventWaitHandleAccessRule(identity, rights, AccessControlType.Allow);
security.AddAccessRule(accessRule);

// createdHandle, openedHandle1 and openedHandle2 point to the same event wait handle
EventWaitHandle createdHandle = EventWaitHandleAcl.Create(initialState: true, EventResetMode.AutoReset, eventWaitHandleName, out bool createdNew, security);
EventWaitHandle openedHandle1 = EventWaitHandleAcl.OpenExisting(eventWaitHandleName, rights);
EventWaitHandleAcl.TryOpenExisting(eventWaitHandleName, rights, out EventWaitHandle openedHandle2);

HMAC one-shot methods

The System.Security.Cryptography HMAC classes now have static methods that allow one-shot calculation of HMACs without allocations. These additions are similar to the one-shot methods for hash generation that were added in a previous release.

DependentHandle is now public

The DependentHandle type is now public with the following API surface:

namespace System.Runtime
{
    public struct DependentHandle : IDisposable
    {
        public DependentHandle(object? target, object? dependent);
        public bool IsAllocated { get; }
        public object? Target { get; set; }
        public object? Dependent { get; set; }
        public (object? Target, object? Dependent) TargetAndDependent { get; }
        public void Dispose();
    }
}

It can be used by to create advanced systems, such as sophisticated caching systems or customized versions of the ConditionalWeakTable<TKey, TValue> type. For instance, it will be used by the WeakReferenceMessenger type in the MVVM Toolkit to avoid memory allocations when broadcasting messages.

Portable thread pool

The .NET thread pool has been re-implemented as a managed implementation and is now used as the default thread pool in .NET 6. We made this change to enable all .NET applications to have access to the same thread pool independent of whether the CoreCLR, Mono, or any other runtime was being used. We have not observed or expect any functional or performance impact as part of this change.

RyuJIT

The team has made many improvements to the .NET JIT compiler this release, documented in each of the preview posts. Most of these changes improve performance. A few of the RyuJIT highlights are covered here.

Dynamic PGO

In .NET 6 we’ve enabled two forms of PGO (Profile Guided Optimization):

  • Dynamic PGO uses data gathered from the current run to optimize the current run.
  • Static PGO relies on data gathered from past runs to optimize future runs.

Dynamic PGO was already covered in the performance section, earlier in the post. I’ll provide a re-cap.

Dynamic PGO enables the JIT to collect information at runtime about the code paths and types that are actually used for that specific run of the app. The JIT can then optimize the code in terms of those code paths for sometimes very significantly improved performance. We’ve seen healthy double-digit improvements in both testing and production. There are a set of classic compiler techniques that are not possible with either a JIT or ahead-of-time compilation without PGO. We’re now able to apply those techniques. Hot/cold splitting is one such technique and devirtualization is another.

To enable Dynamic PGO, set DOTNET_TieredPGO=1 in the environment where your application will run.

As stated in the performance section, dynamic PGO delivers a 26% improvement (510K -> 640K) in requests per second for the TechEmpower JSON “MVC” suite. That’s an amazing improvement with no code changes.

Our ambition is to enable Dynamic PGO by default in a future release of .NET, hopefully with .NET 7. We strongly encourage you to try Dynamic PGO in your applications and give us feedback.

Full PGO

To get the full benefit of Dynamic PGO, you can set two extra environment variables: DOTNET_TC_QuickJitForLoops=1 and DOTNET_ReadyToRun=0. This ensures that as many methods as possible participate in tiered compilation. We call this variant Full PGO. Full PGO can provide larger steady-state performance benefits than Dynamic PGO but will have slower startup times (since more methods must be jitted at Tier 0).

You would not want to use this option for a short-running serverless application, but it could make sense for a long-running one.

In future releases we plan to streamline and simplify these options so that you can get the benefits of full PGO more simply and for a wider range of applications.

Static PGO

We currently use Static PGO to optimize .NET Libraries assemblies like System.Private.CoreLib that ship with R2R (Ready To Run).

The benefit of static PGO is that is that optimizations are made when assemblies are compiled to R2R format with crossgen. That means that there is runtime benefit with no runtime cost. That’s very significant and the reason why PGO is important with C++, for example.

Loop alignment

Memory alignment is a common requirement for various operations in modern computing. In .NET 5, we started aligning methods at the 32-byte boundary. In .NET 6, we have added a feature to perform adaptive loop alignment that adds NOP padding instructions in a method having loops such that the loop code starts at mod(16) or mod(32) memory address. These changes improve and stabilize the performance of .NET code.

In the following Bubble sort chart, data point 1 represents the point where we started aligning methods at 32-byte boundary. Data point 2 represents the point where we started aligning inner loops that as well. As you can see, both the performance and stability of the benchmark improve considerably.

bubble sort chart

Hardware-accelerated structs

Structs are an important part of the CLR type system. In recent years, they have been frequently used as a performance primitive throughout the .NET libraries. Recent examples are ValueTask, ValueTuple and Span<T>. Record structs are a new example. In .NET 5 and .NET 6, we’ve been improving performance for structs, in part by ensuring that structs can be held in ultra-fast CPU registers when they are locals, arguments or return values of methods). This is particularly beneficial for APIs that compute with vectors.

Stabilize performance measurements

There is a tremendous amount of engineering systems work on the team that never appears on the blog. That will be true for any hardware or software product you use. The JIT team took on a project to stabilize performance measurements with the goal of increasing the value of regressions that are auto-reported by our internal performance lab automation. This project is interesting because of the in-depth investigation and the product changes that were required to enable stability. It also demonstrates the scale at which we measure to maintain and improve performance.

perf-#43227, Announcing .NET 6 — The Fastest .NET Yet

This image demonstrates unstable performance measurements where performance fluctuated between a slow and fast in successive runs. The x-axis is the date of the test and y-axis is the time of the test in nanoseconds. By the end of the chart (after these changes were committed), you can see that measurements stabilized, with the best result. This image demonstrates a single test. There are many more tests that are demonstrated to have similar behavior at dotnet/runtime #43227.

Ready-to-run Code / Crossgen 2

Crossgen2 is a replacement of the crossgen tool. It is intended to satisfy two outcomes:

  • Make crossgen development more efficient.
  • Enable a set of capabilities that are not currently possible with crossgen.

This transition is somewhat similar to the native code csc.exe to managed-code Roslyn compiler. Crossgen2 is written in C#, however, it doesn’t expose a fancy API like Roslyn does.

There are perhaps a half-dozen projects we had/have planned for .NET 6 and 7 that are dependent on crossgen2. The proposal for vector instruction default is a great example of a crossgen2 capability and product change that we wanted to make for .NET 6 but is more likely .NET 7. Version bubbles are another good example.

Crossgen2 enables cross-compilation (hence the name “crossgen”) across operating system and architecture dimensions. That means that you will be able to use a single build machine to generate native code for all targets, at least as it relates to ready-to-run code. Running and testing that code is a different story, however, and you’ll need appropriate hardware and operating systems for that.

The first step is to compile the platform itself with crossgen2. We completed that with .NET 6, for all architectures. As a result, we were able to retire the old crossgen this release. Note that crossgen2 only applies to CoreCLR and not to Mono-based applications (which have a separate set of code generation tools).

This project — at least at first — is not oriented on performance. The goal is to enable a much better architecture for hosting the RyuJIT (or any other) compiler to generate code in an offline manner (not requiring or starting the runtime).

You might say “hey … don’t you have to start the runtime to run crossgen2 if it is written in C#?” Yes, but that’s not what is meant by “offline” in this context. When crossgen2 runs, we’re not using the JIT that comes with the runtime that crossgen2 is running on to generate ready-to-run (R2R) code. That won’t work, at least not with the goals we have. Imagine crossgen2 is running on an x64 machine, and we need to generate code for Arm64. Crossgen2 loads the Arm64 RyuJIT — compiled for x64 — as a native plugin, and then uses it to generate Arm64 R2R code. The machine instructions are just a stream of bytes that are saved to a file. It can also work in the opposite direction. On Arm64, crossgen2 can generate x64 code using the x64 RyuJIT compiled to Arm64. We use the same approach to target x64 code on x64 machines. Crossgen2 loads a RyuJIT built for whatever configuration is needed. That may seem complicated, but it’s the sort of system you need in place if you want to enable a seamless cross-targeting model, and that’s exactly what we want.

We hope to use the term “crossgen2” for just one release, after which it will replace the existing crossgen, and then we’ll go back to using the “crossgen” term for “crossgen2”.

.NET Diagnostics: EventPipe

EventPipe is our cross-platform mechanism for egressing events, performance data, and counters, either in-process or out-of-process. Starting with .NET 6, we’ve moved the implementation from C++ to C. With this change, Mono uses EventPipe as well. This means that both CoreCLR and Mono use the same eventing infrastructure, including the .NET Diagnostics CLI Tools.

This change also came with small reduction in size for CoreCLR:

lib after size – before size diff
libcoreclr.so 7037856 – 7049408 -11552

We’ve also made some changes that improve EventPipe throughput while under load. Over the first few previews, we’ve made a series of changes that result in throughput improvements as high as 2.06x what .NET 5 was capable of:

Announcing .NET 6 — The Fastest .NET Yet

Higher is better for this benchmark. .NET 6 is the orange line and .NET 5 is the blue one.

SDK

The following improvements were made to the .NET SDK.

CLI install of .NET 6 SDK Optional Workloads

.NET 6 introduces the concept of SDK workloads. Workload are optional components can be installed on top of the .NET SDK to enable various scenarios. The new workloads in .NET 6 are: .NET MAUI and Blazor WebAssembly AOT workloads. We will likely create new workloads (possibly from the existing SDK) in .NET 7. The biggest benefit of workloads is size reduction and optionality. We want to make the SDK smaller over time and enable installing just the components you need. This model will be good for developer machines and even better for CI.

Visual Studio users don’t really need to worry about workloads. The workloads feature has been designed so that an installation orchestrator like Visual Studio can install workloads for you. Workloads can be managed directly via the CLI.

The workloads feature exposes multiple verbs for managing workloads, including the following ones:

  • dotnet workload restore — installs the workloads required by a given project.
  • dotnet workload install — installs a named workload.
  • dotnet workload list — lists the workloads you have installed.
  • dotnet workload update — updates all installed workloads to the newest available version.

The update verb queries nuget.org for updated workload manifests, updates local manifests, downloads new versions of the installed workloads, and then removes all old versions of a workload. This is analogous to apt update && apt upgrade -y (used on Debian-based Linux distros). It is reasonable to think of workloads as a private package manager for the SDK. It is private in the sense that it is only available for SDK components. We may reconsider that in future.

The dotnet workload commands operate in the context of the given SDK. Imagine you have both .NET 6 and .NET 7 installed. The workloads commands will provide different results for each SDK since the workloads will be different (at least different versions of the same workloads).

Note that dotnet workload install copies the workloads from NuGet.org into your SDK install so will need to be run elevated or use sudo if the SDK install location is protected (meaning at an admin/root location).

Built-in SDK version checking

To make it easier to track when new versions of the SDK and Runtimes are available, we’ve added a new command to the .NET 6 SDK.

dotnet sdk check

It tells you if there is a newer version available for any of the .NET SDKs, runtimes, or workloads you have installed. You can see the new experience in the following image.

sdkcheck

dotnet new

You can now search NuGet.org for new templates with dotnet new --search.

Other improvements to template installation include support for the --interactive switch to support authorization credentials for private NuGet feeds.

Once CLI templates are installed, you can check if updates are available via --update-check and --update-apply.

NuGet Package Validation

Package Validation tools enables NuGet library developers to validate that their packages are consistent and well-formed.

This includes:

  • Validate that there are no breaking changes across versions.
  • Validate that the package has the same set of publics APIs for all runtime-specific implementations.
  • Determine any target-framework- or runtime- applicability gaps.

This tool is part of the SDK. The easiest way to use it is by setting a new property in your project file.

<EnablePackageValidation>true</EnablePackageValidation>

More Roslyn Analyzers

In .NET 5, we shipped approximately 250 analyzers with the .NET SDK. Many of them already existed but were shipped out-of-band as NuGet packages. We’ve adding more analyzers for .NET 6.

By default, most of the new analyzers are enabled at Info level. You can enable these analyzers at Warning level by configuring the analysis mode like this: <AnalysisMode>All</AnalysisMode>.

We published the set of analyzers we wanted for .NET 6 (plus some extras) and then made most of them up-for-grabs. The community has added several implementations, including these ones.

Contributor Issue Title
Newell Clark dotnet/runtime #33777 Use span-based string.Concat
Newell Clark dotnet/runtime #33784 Prefer string.AsSpan() over string.Substring() when parsing
Newell Clark dotnet/runtime #33789 Override Stream.ReadAsync/WriteAsync
Newell Clark dotnet/runtime #35343 Replace Dictionary<,>.Keys.Contains with ContainsKey
Newell Clark dotnet/runtime #45552 Use String.Equals instead of String.Compare
Meik Tranel dotnet/runtime #47180 Use String.Contains(char) instead of String.Contains(String)

Thanks Meik Tranel and Newell Clark.

Enable custom guards for Platform Compatibility Analyzer

The CA1416 Platform Compatibility analyzer already recognizes platform guards using the methods inOperatingSystem and RuntimeInformation, such as OperatingSystem.IsWindows and OperatingSystem.IsWindowsVersionAtLeast. However, the analyzer does not recognize any other guard possibilities like the platform check result cached in a field or property, or complex platform check logic is defined in a helper method.

For allowing custom guard possibilities we added new attributes SupportedOSPlatformGuard and UnsupportedOSPlatformGuard for annotating the custom guard members with the corresponding platform name and/or version. This annotation is recognized and respected by the Platform Compatibility analyzer’s flow analysis logic.

Usage

    [UnsupportedOSPlatformGuard("browser")] // The platform guard attribute
#if TARGET_BROWSER
    internal bool IsSupported => false;
#else
    internal bool IsSupported => true;
#endif

    [UnsupportedOSPlatform("browser")]
    void ApiNotSupportedOnBrowser() { }

    void M1()
    {
        ApiNotSupportedOnBrowser();  // Warns: This call site is reachable on all platforms.'ApiNotSupportedOnBrowser()' is unsupported on: 'browser'

        if (IsSupported)
        {
            ApiNotSupportedOnBrowser();  // Not warn
        }
    }

    [SupportedOSPlatform("Windows")]
    [SupportedOSPlatform("Linux")]
    void ApiOnlyWorkOnWindowsLinux() { }

    [SupportedOSPlatformGuard("Linux")]
    [SupportedOSPlatformGuard("Windows")]
    private readonly bool _isWindowOrLinux = OperatingSystem.IsLinux() || OperatingSystem.IsWindows();

    void M2()
    {
        ApiOnlyWorkOnWindowsLinux();  // This call site is reachable on all platforms.'ApiOnlyWorkOnWindowsLinux()' is only supported on: 'Linux', 'Windows'.

        if (_isWindowOrLinux)
        {
            ApiOnlyWorkOnWindowsLinux();  // Not warn
        }
    }
}

Closing

Welcome to .NET 6. It is another huge .NET release, with near-equal servings of performance, functionality, usability, and security improvements. We hope you find many improvements that end up making you more efficient and capable in your every-day development and incease performance or reduce the costs of your apps in production. We’ve already started hearing good things from those among you have already started to use .NET 6.

At Microsoft, we’re also in the early phases of .NET 6 deployment, with a few key apps already in production and many more soon to come in the coming weeks and months.

.NET 6 is our latest LTS release. We encourage everyone to move to it, particularly if you are using .NET 5. We’re expecting it to be the fastest adopted .NET release ever.

This release is the result of at least 1000 people (but probably a lot more). That includes .NET team from Microsoft and many more in the community. I’ve tried to include many community-contributed features in this post. Thank you for taking the time to create those and work through our process. I hope that the experience was a good one and that even more people will contribute.

This post is the result of a collaboration of many talented people. The contributions includes the feature content that the team provides throughout the release, significant new content created for this final post, and a ton of technical and prose corrections that are required to bring the final content up to the quality you deserve. It’s been a pleasure crafting it and all the other posts for you.

Thanks for being a .NET developer.

Category
.NETFeatured

Author

Richard Lander
Richard Lander
Product Manager

Richard Lander is a Program Manager on the .NET team. He works on making .NET work great in memory-limited Docker containers, on Arm hardware like the Raspberry Pi, and enabling GPIO programming and IoT scenarios. He is part of the design team that defines new .NET runtime capabilities and features. Favourite fantasy: Dune and Doctor Who. He grew up in Canada and New Zealand.

77 comments

Discussion is closed. Login to edit/delete existing comments.

Newest
Newest Popular Oldest

Read next