If you ship an application with Native AOT, a green managed test run leaves a gap. Native AOT compiles ahead of time, removes unused code, and requires alternatives to runtime code generation and unrestricted reflection. The published application can therefore behave differently from the code exercised by the managed test process.

Starting with MSTest 4.4, MSTest supports publishing and running test projects as Native AOT executables. Source generation records which tests exist and how to invoke them before trimming happens, without requiring developers to rewrite their test classes. The result is simple: test what you ship.

For teams with formal validation plans, including some in regulated environments, that native run provides more representative evidence. It doesn’t replace testing the final application artifact.

A managed pass can still hide a deployment failure

Consider an application that serializes a receipt with System.Text.Json and a test that covers that path:

[TestClass]
public class ReceiptFormatterTests
{
    [TestMethod]
    public void ReceiptIsSerialized()
    {
        var json = JsonSerializer.Serialize(new Receipt(42));

        StringAssert.Contains(json, "\"Total\":42");
    }
}

public sealed record Receipt(decimal Total);

The test passes in a normal managed run because System.Text.Json can discover the type through reflection. In a trimmed or Native AOT publish, reflection-based serialization is disabled by default. The same path throws:

System.InvalidOperationException:
Reflection-based serialization has been disabled for this application.

That failure is useful. It identifies an application deployment problem, not a testing-framework problem. The production code should provide generated JSON metadata, for example:

[JsonSerializable(typeof(Receipt))]
internal partial class AppJsonContext : JsonSerializerContext
{
}

var json = JsonSerializer.Serialize(
    new Receipt(42),
    AppJsonContext.Default.Receipt);

The System.Text.Json source-generation guidance describes this behavior and the available generation modes. Serialization is only one example; native testing can also expose unsupported runtime code generation, missing reflection metadata, or an incompatible dependency.

There are two separate responsibilities here. MSTest source generation keeps the test discoverable and runnable after trimming. The JSON source generator fixes the application path that the test exercises. MSTest doesn’t hide the application problem; it lets the native test process expose it before the product reaches deployment.

How MSTest makes the native test executable possible

MSTest’s first Native AOT preview arrived in April 2024. It proved that an MSTest project could become a native executable, but the experimental engine and source generator had limited coverage.

The new path moves source generation into the open MSTest toolchain and aligns it with MSTest 4.4. During compilation, the generator emits:

• A registry of the test classes in the assembly.
• Attribute data for supported test members.
• Delegates that construct test classes and invoke test methods.
• References that preserve discovered test classes and supported base classes when trimming runs.

The important result isn’t the generated code itself. The build records which tests exist and how to run them before trimming happens. Your tests remain ordinary [TestClass] and [TestMethod] code; the generator changes the build and execution path, not the programming model.

Configure one representative project

With the targeted release, the minimum project configuration is deliberately small:

<Project Sdk="MSTest.Sdk/4.4.0">
  <PropertyGroup>
    <TargetFramework>net10.0</TargetFramework>
    <PublishAot>true</PublishAot>
  </PropertyGroup>

  <!-- Keep your existing ItemGroup elements and project references. -->
</Project>

MSTest.Sdk uses Microsoft Testing Platform (MTP) by default. Setting PublishAot enables MSTest source generation and the native executable path. Projects that still use VSTest should review the VSTest-to-MTP migration guidance because command-line arguments, CI integration, and supported .runsettings entries differ.

Publish for the same operating system and architecture as the application:

dotnet publish ./MyProject.Tests/MyProject.Tests.csproj \
  -c Release -r linux-x64 -o ./artifacts/native-tests
./artifacts/native-tests/MyProject.Tests

The example uses the linux-x64 runtime identifier (RID). Replace it with the RID you deploy, such as win-x64 or osx-arm64; on Windows, run MyProject.Tests.exe. Replace the project path with your test project.

Then add a focused CI pilot:

1. Keep the existing managed test run.
2. Publish and run one representative test project as Native AOT.
3. Assert that both lanes discover the exact expected test count and outcomes.
4. Record native publish-and-run time separately from test execution time.
5. Expand only where the additional confidence justifies the CI cost.

This isn’t expected after a clean migration. It can happen when a test class can’t enter the generated registry—for example, because it only inherits [TestClass] or is inaccessible, file-local, static, or open generic—and the related diagnostics are ignored or suppressed. The registered subset can still pass and the process can exit successfully, so test-count parity is a release gate.

Start with a scheduled or release-validation job. Move the lane to every pull request only if its signal and publish time justify the added feedback cost.

Choose a project with meaningful deployment-sensitive paths: serialization, dependency injection, configuration binding, reflection-based plugins, or a dependency whose Native AOT support you need to prove. A project containing only arithmetic-style unit tests can demonstrate that the runner works, but it won’t tell you much about the application you ship.

Keep the boundaries clear

This isn’t equivalent to an end-to-end production validation. Configuration, operating system, architecture, external services, and packaging can still differ. It removes one important variable: the test and application can use the same trimming and ahead-of-time compilation model.

Source generation also doesn’t mean zero reflection. The default ReflectionFree mode uses generated attributes and delegates for supported construction and invocation, but some operations retain reflective fallbacks. For compatibility investigations, set:

<PropertyGroup>
  <MSTestSourceGenMode>Rooting</MSTestSourceGenMode>
</PropertyGroup>

Rooting preserves discovered test members but uses reflective execution.

The most important migration limits are:

Limitation	Migration guidance
A class only inherits [TestClass]	Declare the attribute directly; MSTEST0069 identifies this shape.
A test class is inaccessible, file-local, static, abstract, or open generic	Use a concrete, accessible, non-static, closed type. Abstract base fixtures remain supported through a concrete derived test class.
A test method is generic or has ref, out, or in parameters	Use a supported method signature.
[AssemblyFixtureProvider] is used	Replace it with a supported fixture pattern before relying on the native run.

Some MSTest SDK integrations, MTP extensions, and CI reporters aren’t available in the Native AOT path. TRX and Code Coverage remain supported. Treat analyzer and build diagnostics as migration gates rather than warnings to suppress, and check the MSTest SDK documentation for the current support matrix.

For a team, the change should stay deliberately narrow:

Keep	Add	Still required
Fast managed tests for everyday feedback	One published native test lane for selected projects	End-to-end validation of the final application artifact

That separation makes the rollout reversible. If the native lane costs more than the confidence it adds, change its frequency, choose a more representative project, or stop the pilot without disrupting the managed test suite.

Performance is evidence, not the premise

Source generation avoids the assembly-wide Assembly.GetTypes() scan and reflective construction and invocation for supported tests. That can reduce startup and discovery work, but it doesn’t guarantee a faster end-to-end run; test execution, process startup, publishing, and remaining reflection can dominate.

Production fidelity is useful even if the performance improvement is small. Performance is secondary, not the reason to test under the deployment model you ship.

Start with one project

Choose a project that exercises code you publish with Native AOT. Use the next deployment-only failure, test-count mismatch, or clean native run to evaluate whether the lane adds useful confidence. Then decide whether to expand, refine, or stop the pilot.

With MSTest 4.4 and a verified native publish path, the tests still look like MSTest while the executable behaves more like the application you actually ship.