New in .NET Libraries with this release:
- Removal of
BinaryFormatteris complete - Enumerate over
ReadOnlySpan<char>.Split()segments Debug.Assertnow reports assert condition, by default.- Compression APIs now use
zlib-ng Guid.CreateVersion7enables creating GUIDs with a natural sort orderInterlocked.CompareExchangefor more types- AES-GCM and ChaChaPoly1305 algorithms enabled for iOS/tvOS/MacCatalyst
- Changes to X.509 Certificate Loading
- Support for XPS documents from XPS virtual printer
- Marking
Tensor<T>asExperimental - Introducing Runtime Metrics
- Introducing Environment CpuUsage
- Adding Metrics Measurement Constructor with TagList Parameter
- [
Microsoft.Bcl.MemoryCompatibility Package](# microsoft-bcl-memory-compatibility-package)
Libraries updates in .NET 9 Preview 7:
- Release notes
- What's new in .NET 9 documentation
.NET 9 Preview 7:
As announced earlier, starting with .NET 9, we no longer include an implementation of BinaryFormatter in the runtime (.NET Framework remains unchanged). The APIs are still present, but their implementation always throws an exception, regardless of project type.
Hence, setting the existing backwards compatibility flag is no longer sufficient to use BinaryFormatter.
- We published the BinaryFormatter migration guide. We'd appreciate if could give it a read and give us feedback by filling issues in the dotnet/docs repo.
- If you experience issues related to BinaryFormatter's removal not addressed in this migration guide, please file an issue in the dotnet/runtime repo and indicate that the issue is related to the removal of
BinaryFormatter.
The primary reason is that BinaryFormatter is unsafe. Any deserializer, binary or text, that allows its input to carry information about the objects to be created is a security problem waiting to happen.
There is a common weakness enumeration (CWE) that describes the issue: CWE-502 "Deserialization of Untrusted Data". BinaryFormatter is such a deserializer but this isn't specific to .NET; this applies to any deserializer. Another example is using eval in JavaScript to load JSON.
We also cover this in the BinaryFormatter security guide.
You have two options to address the removal of BinaryFormatter's implementation:
-
Migrate away from BinaryFormatter. We strongly recommend you to investigate options to stop using
BinaryFormatterdue to the associated security risks. The BinaryFormatter migration guide lists several options. -
Keep using BinaryFormatter. If you need to continue using
BinaryFormatterin .NET 9, you need to depend on the unsupported System.Runtime.Serialization.Formatters NuGet package, which restores the unsafe legacy functionality and replaces the throwing implementation.
string.Split is a very popular and convenient method for quickly partitioning a string with one or more supplied separators. For code focused on performance, however, the allocation profile of string.Split can be prohibitive, allocating a string for each parsed component and a string[] to store them all.
It also doesn't work with spans, so if you have a ReadOnlySpan<char> and want to use string.Split with it, you're forced to allocate yet another string, converting the ReadOnlySpan<char> to a string to be able to call the method on it.
In .NET 8, a set of Split and SplitAny methods were introduced for ReadOnlySpan<char>. Rather than returning a new string[], these methods instead accept a destination Span<Range> into which to write the bounding indices for each component.
This makes the operation fully allocation-free, as no new string object is required (each result is represented just as a Range within the original span) and no string[] needs to be allocated (the results can be written into the supplied destination buffer). This is great when the number of ranges is both known and small, yielding an API that's both efficient and almost as convenient as string.Split.
New overloads of Split and SplitAny have been added to allow incrementally parsing a ReadOnlySpan<T> with an a priori unknown number of segments. The new methods enable enumerating through each segment, which is similarly represented as a Range that can be used to slice into the original span.
public static bool ListContainsItem(ReadOnlySpan<char> span, string item)
{
foreach (Range segment in span.Split(','))
{
if (span[segment].SequenceEquals(item))
{
return true;
}
}
return false;
}Debug.Assert is commonly-used to help validate conditions that are expected to always be true, where failure typically indicates a bug in the code.
There are many overloads of Debug.Assert, the simplest of which just accepts a condition:
Debug.Assert(a > 0 && b > 0);The assert fails if the condition is false. Historically, however, such asserts were void of any information about what condition failed. Now in .NET 9, if no message is explicitly provided by the user, the assert will contain the textual representation of the condition.
For example, for the above assert, rather than getting a message like:
Process terminated. Assertion failed.
at Program.SomeMethod(Int32 a, Int32 b)The message will now be like:
Process terminated. Assertion failed.
a > 0 && b > 0
at Program.SomeMethod(Int32 a, Int32 b)System.IO.Compression features like ZipArchive, DeflateStream, GZipStream, and ZLibStream are all based primarily on the zlib library. Now in .NET 9, these features instead all use zlib-ng, yielding more consistent and efficient processing across a wider array of operating systems and hardware.
Guid.NewGuid() creates a Guid filled mostly with cryptographically-secure random data, following the UUID Version 4 specification in RFC 9562.
That same RFC also defines other versions, including Version 7, which "features a time-ordered value field derived from the widely implemented and well-known Unix Epoch timestamp source".
In other words, much of the data is still random, but some of it is reserved for data based on a timestamp, enabling these values to have a natural sort order. In .NET 9, a Guid can be created according to Version 7 via the new Guid.CreateVersion7() and Guid.CreateVersion7(DateTimeOffset timestamp) methods. A Guid's version field can also be retrieved with the new Version property.
In previous versions of .NET, Interlocked.Exchange and Interlocked.CompareExchange had overloads for working with int, uint, long, ulong, nint, nuint, float, double, and object, as well as a generic overload for working with any reference type T.
Now in .NET 9, thanks to several contributions from @MichalPetryka, there are now also overloads for atomically working with byte, sbyte, short, and ushort.
Moreover, the generic constraint on the generic Interlocked.Exchange<T> and Interlocked.CompareExchange<T> overloads has been removed, so those methods are no longer constrained to only work with reference types. In addition, they can now work with any primitive type, which includes all of the aforementioned types as well as bool and char, as well as any enum type.
AesGcm.IsSupported and ChaChaPoly1305.IsSupported now return true when running on iOS 13+, tvOS 13+ or Mac Catalyst.
As with macOS, AesGcm on these environments only supports 16-byte (128-bit) tag values.
Since .NET Framework 2.0 (way back in 2005) the way to load a certificate has been new X509Certificate2(bytes). There have also been other patterns, such as: new X509Certificate2(bytes, password, flags), new X509Certificate2(path), new X509Certificate2(path, password, flags), and X509Certificate2Collection.Import(bytes, password, flags) (and its overloads).
Those methods all used content-sniffing to figure out if the input was something it could handle, and then loaded it if it could. For some callers this was the epitome of convenience, for others it was frustration that not every file format worked on every OS. For some it was a protocol deviation, and for a few it was a source of security issues.
.NET 9 introduces a new X509CertificateLoader class, which has a "one method, one purpose" design. In its initial version, it only supports two of the five formats that the X509Certificate2 constructor supported, but they are the only two formats that worked on all OSes.
The following snippets highlight the new API as a sort of exhaustive textbook example:
private static X509Certificate2 LoadSingleCertificate(
byte[] data,
X509ContentType format,
string pfxPassword,
X509KeyStorageFlags pfxFlags)
{
switch (format)
{
case X509ContentType.Cert:
return X509CertificateLoader.LoadCertificate(data);
case X509ContentType.Pkcs12:
//case X509ContentType.Pfx: //(same thing)
return X509CertificateLoader.LoadPkcs12(data, pfxPassword, pfxFlags);
case X509ContentType.Pkcs7:
SignedCms cms = new SignedCms();
cms.Decode(data);
return cms.SignerInfos[0].Certificate ?? throw new CryptographicException();
case X509ContentType.SerializedCert:
case X509ContentType.Authenticode:
// These two formats are only supported on Windows,
// and only from the obsolete constructors.
X509ContentType actualType = X509Certificate2.GetCertContentType(data);
if (actualType != format)
{
throw new CryptographicException();
}
#pragma warning disable SYSLIB0057
return new X509Certificate2(data);
#pragma warning restore SYSLIB0057
default:
throw new CryptographicException();
}
}
private static X509Certificate2Collection LoadCertificateCollection(
byte[] data,
X509ContentType format,
string pfxPassword,
X509KeyStorageFlags pfxFlags)
{
switch (format)
{
case X509ContentType.Pkcs12:
return X509CertificateLoader.LoadPkcs12Collection(data, pfxPassword, pfxFlags);
case X509ContentType.Pkcs7:
SignedCms cms = new SignedCms();
cms.Decode(data);
return cms.Certificates;
case X509ContentType.SerializedStore:
// Only supported on Windows, and only via the obsolete Import method
X509ContentType actualType = X509Certificate2.GetCertContentType(data);
if (actualType != format)
{
throw new CryptographicException();
}
X509Certificate2Collection coll = new X509Certificate2Collection();
#pragma warning disable SYSLIB0057
coll.Import(data);
#pragma warning restore SYSLIB0057
return coll;
default:
throw new CryptographicException();
}
}The X509CertificateLoader class is available for callers on older versions of .NET, and for callers on .NET Framework, via the Microsoft.Bcl.Cryptography compatibility package.
In .NET 8, we introduced OpenSSL specific APIs: SafeEvpPKeyHandle.OpenPrivateKeyFromEngine and SafeEvpPKeyHandle.OpenPublicKeyFromEngine.
They enable interacting with OpenSSL ENGINE components and utilize hardware security modules (HSM), for example.
.NET 9 introduces SafeEvpPKeyHandle.OpenKeyFromProvider which enables using OpenSSL providers and interacting with providers such as tpm2 or pkcs11.
Some distros have removed ENGINE support since it is now deprecated.
The following snippet shows basic usage:
byte[] data = ...;
// Refer to provider documentation you're using, i.e. https://github.com/tpm2-software/tpm2-openssl/tree/master
// specific values are just an illustrative example
using (SafeEvpPKeyHandle priKeyHandle = SafeEvpPKeyHandle.OpenKeyFromProvider("tpm2", "handle:0x81000007"))
using (ECDsa ecdsaPri = new ECDsaOpenSsl(priKeyHandle))
{
byte[] signature = ecdsaPri.SignData(data, HashAlgorithmName.SHA256);
// do stuff with signature created by TPM
// note that some providers i.e. tpm2 do not allow direct operations on public key (verify/encrypt), public key should be exported and re-imported into new ECDsa instance
}Performance improvements may be observed during the TLS handshake as well as improvements to interactions with RSA private keys using ENGINE components.
Windows 11 has added new APIs to help secure Windows keys with virtualization-based security (VBS). With this new capability, keys can be protected from admin-level key theft attacks with negligible effect on performance, reliability, or scale.
.NET 9 has added matching CngKeyCreationOptions flags.
The following snippet demonstrates how to use one of those flags:
using System.Security.Cryptography;
CngKeyCreationParameters cngCreationParams = new()
{
Provider = CngProvider.MicrosoftSoftwareKeyStorageProvider,
KeyCreationOptions = CngKeyCreationOptions.RequireVbs | CngKeyCreationOptions.OverwriteExistingKey,
};
using (CngKey key = CngKey.Create(CngAlgorithm.ECDsaP256, "myKey", cngCreationParams))
using (ECDsaCng ecdsa = new ECDsaCng(key))
{
// do stuff with the key
}In total 3 flags were added:
CngKeyCreationOptions.PreferVbsmatchingNCRYPT_PREFER_VBS_FLAGCngKeyCreationOptions.RequireVbsmatchingNCRYPT_REQUIRE_VBS_FLAGCngKeyCreationOptions.UsePerBootKeymatchingNCRYPT_USE_PER_BOOT_KEY_FLAG
XPS documents coming from a V4 XPS virtual printer could not be opened using the System.IO.Packaging library, due to lack of support for handling .piece files. Addressing this has been a long-standing request, but now in .NET 9 the gap has been addressed. Thanks to @edwardneal for the improvement!
We announced the addition of Tensor<T> earlier in .NET 9. Adding new built-in types for exchanging tensor data across libraries and allowing accelerated handling for core operations is an important but large undertaking.
The work done in .NET 9 currently encompasses 8 types and nearly 600 new public APIs, many of which are generic and can support a T that is arbitrary or constrained to one of the generic math interfaces.
Due to the size of this work and the recognized importance of it to the long term capability of the .NET ecosystem, it has been decided to mark these new APIs as [Experimental] (see https://learn.microsoft.com/dotnet/api/system.diagnostics.codeanalysis.experimentalattribute) for .NET 9 and to plan on supporting them officially in .NET 10 instead.
This is being done to allow time for additional feedback from the community and important libraries in the .NET ecosystem that plan on taking advantage of the provided functionality. It also gives us additional time to coordinate with improvements in the C# language to ensure a great end-to-end experience.
Please try out the new APIs and provide feedback on the overall experience. This includes (but is not limited to) the naming or shape of APIs, any core functionality that appears to be missing or that behaves unexpectedly, and how the general user experience was.
Some additional callouts include:
- Despite being marked
[Experimental]much of the surface is relatively stable and isn't expected to change. There isn't much you can get wrong for an API likeAdd(x, y)after all. This isn't a guarantee that we won't change such APIs, but the expected churn for these ones is minimal. - There's known missing functionality such as no operator support given that we want
Tensor<T>to work over allTbutoperator +can only be defined for types that implement the generic mathIAdditionOperatorsinterface. This requires language support forextension operators. - The
TensorSpan<T>types can wrap native allocations, but we don't have a non-ref struct type equivalent. Such a type would need to be disposable and needs additional design consideration around how ownership/lifetime tracking works. If this is an important scenario to you, we'd like to hear about it. - The names/concepts used for some APIs can be very inconsistent across the entire machine learning ecosystem (both inside and outside .NET) and in many cases we opted choose the name that was most consistent with existing .NET concepts or behavior. This decision is typically matching the same semantics given to other major tensor libraries (both in .NET and in other language ecosystems), but if there are any quirks or unexpected behaviors found then we'd like to hear about it.
As a final note, the TensorPrimitives class which we shipped in .NET 8 is stable and has been expanded in .NET 9 with additional API surface that is also considered stable. It is not marked as [Experimental]. It is the class that contains most of the accelerated algorithms that underpin the Tensor<T> type and so they can still be used to accelerate your code where applicable.
There are many potential applications for these algorithms including in machine learning/AI, image processing, games, and beyond.
.NET has long supported System.Runtime Counters. With the introduction of the Metrics feature, it became a natural step to expose runtime counters as metrics. This enhancement allows users to collect runtime metrics more flexibly and enables support for telemetry platforms like OpenTelemetry.
The detailed semantic conventions for runtime metrics can be found here.
Users or tools can collect runtime metrics by using the System.Diagnostics.Metrics event source provider to listen to the meter named System.Runtime. Below is an example of how to use the dotnet-counters tool to listen and display runtime metrics for a specific process ID:
dotnet-counters monitor --process-id 29104 --counters System.Runtime
The output of this command will be like the following:
Press p to pause, r to resume, q to quit.
Status: Running
Name Current Value
[System.Runtime]
dotnet.assembly.count ({assembly}) 16
dotnet.gc.collections ({collection})
gc.heap.generation
gen0 0
gen1 0
gen2 0
dotnet.gc.heap.total_allocated (By) 1,655,208
dotnet.gc.pause.time (s) 0
dotnet.jit.compilation.time (s) 0.245
dotnet.jit.compiled_il.size (By) 81,019
dotnet.jit.compiled_methods ({method}) 754
dotnet.monitor.lock_contentions ({contention}) 0
dotnet.process.cpu.count ({cpu}) 16
dotnet.process.cpu.time (s)
cpu.mode
system 0.031
user 0.156
dotnet.process.memory.working_set (By) 31,395,840
dotnet.thread_pool.queue.length ({work_item}) 0
dotnet.thread_pool.thread.count ({thread}) 0
dotnet.thread_pool.work_item.count ({work_item}) 0
dotnet.timer.count ({timer}) 0
.NET has long supported retrieving CPU usage for the current process via properties like Process.TotalProcessorTime, PrivilegedProcessorTime,
and UserProcessorTime. However, these properties require a dependency on the System.Diagnostics.Process library and involve calling Process.GetCurrentProcess() to retrieve the current process.
Additionally, since these properties are designed to work with any system process, they introduce extra performance overhead.
The new Environment.CpuUsage property provides a more efficient way to retrieve CPU usage for the current process, eliminating the need to create a Process object.
Here is an example of how to use the Environment.CpuUsage property:
Environment.ProcessCpuUsage usage = Environment.CpuUsage;
Console.WriteLine($"Total CPU usage: {usage.TotalCpuUsage}");
Console.WriteLine($"User CPU usage: {usage.UserCpuUsage}");
Console.WriteLine($"Kernel CPU usage: {usage.KernelCpuUsage}");The Measurement<T> class in the System.Diagnostics.Metrics namespace has been updated to include a new constructor
that accepts a TagList parameter.
This new constructor allows users to create a Measurement object with a TagList that contains associated tags.
Previously, users who employed TagList and called the Measurement constructor that accepted an IEnumerable<KeyValuePair<string,object?>> had to allocate extra boxing objects, which negated the performance benefits of using TagList. With this update, users can now pass TagList directly to the Measurement constructor, avoiding unnecessary overhead and improving performance.
var tags = new TagList() { { "Key1", "Value1" } }
var measurement = new Measurement<int>(10, tags);The Microsoft.Bcl.Memory compatibility package provides compatibility for the Base64Url, Index and Range types in .NET Framework and .NET Standard 2.0. The package is useful for projects that need to target .NET Framework or .NET Standard 2.0 and want to use such types.
.NET 9.0 introduces the new Base64Url class in the System.Buffers.Text namespace. Additionally, the types Index and Range were introduced in the System namespace starting with .NET 5.0.
However, these types are not supported in .NET Framework or .NET Standard 2.0. To use them in those environments, you can leverage the Microsoft.Bcl.Memory compatibility package.
The Index and Range types simplify slicing operations on collections, while Base64Url enables URL-safe encoding for data in .NET Framework and .NET Standard 2.0.
Here’s an example that implicitly uses the Index type:
string[] words = ["The", "quick", "brown", "fox", "jumps", "over", "the", "lazy", "dog"];
// Use Index to reference the last element
Console.WriteLine(words[^1]);
// Output: "dog"Here is example of Base64Url which is encoding in a URL-safe version of Base64, commonly used in web applications, such as JWT tokens.
using System.Buffers.Text;
using System.Text;
// Original data
byte[] data = Encoding.UTF8.GetBytes("Hello World!");
Span<byte> encoded = new byte[Base64Url.GetEncodedLength(data.Length)];
Base64Url.EncodeToUtf8(data, encoded, out int _, out int bytesWritten);
string encodedString = Base64Url.EncodeToString(data);
Console.WriteLine($"Encoded: {encodedString}");
// Encoded: SGVsbG8gV29ybGQh
Span<byte> decoded = new byte[data.Length];
Base64Url.DecodeFromUtf8(encoded[..bytesWritten], decoded, out _, out bytesWritten);
string decodedString = Encoding.UTF8.GetString(decoded[..bytesWritten]);
Console.WriteLine($"Decoded: {decodedString}");
// Decoded: Hello World!