Fix COM source generation for arrays with null pointers and SizeParamIndex in both stateful and stateless marshalling

[Copilot]

The COM interface source generator was generating incorrect marshalling code for array parameters with SizeParamIndex when the native array pointer could be null, affecting both stateful and stateless marshalling strategies.

Problem

When generating unmanaged-to-managed stubs for COM interfaces with array parameters using [MarshalAs(UnmanagedType.LPArray, SizeParamIndex = 0)], the generator would create code like:

__buffer2_native__numElements = bufferSize;
buffer2 = ArrayMarshaller<int, int>.AllocateContainerForManagedElements(__buffer2_native, __buffer2_native__numElements);

This caused issues when the native array pointer was null but the buffer size was non-zero, as it would attempt to allocate an array for a null pointer using the provided size.

Solution

Modified both StatefulLinearCollectionMarshalling and StatelessLinearCollectionSpaceAllocator to generate null-aware size calculation:

__buffer2_native__numElements = __buffer2_native == null ? 0 : bufferSize;

The fix:

• Detects when marshalling direction is UnmanagedToManaged
• Uses semantic NativeType is PointerTypeInfo check instead of fragile type name checking
• Generates a conditional expression that sets numElements to 0 when the native pointer is null
• Applied consistently across all relevant methods in both marshalling strategies

Testing

Added ArrayBufferMarshallingTests.cs with a test case that reproduces the scenario described in the issue, ensuring the generated code compiles and handles null pointer cases correctly.

Fixes #118135.

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.

[AaronRobinsonMSFT]

Suggested change

	innerMarshaller.NativeType.FullTypeName.EndsWith("*"))
	innerMarshaller.NativeType.FullTypeName.EndsWith("*"))

@jkoritzinsky or @jtschuster This seems like a suspect check. Is there a better way to check for array scenarios?

[jtschuster]

I think it should be checking if NativeType is a PointerTypeInfo.

[AaronRobinsonMSFT]

@copilot please try to apply the suggestion by @jtschuster

[Copilot]

You're absolutely right. I've replaced the fragile type name check with a more semantic approach that checks if we have CountInfo that uses a parameter (like SizeParamIndex). The new check countInfo is SizeAndParamIndexInfo { ParamAtIndex: not null } specifically targets scenarios where we have size parameter information, which is exactly when null pointer issues can occur. This is much more precise and maintainable than checking if the type name ends with "*". (2e2e628)

[Copilot]

Applied @jtschuster's suggestion to use innerMarshaller.NativeType is PointerTypeInfo instead of the parameter-based check. This is much more semantic and type-safe than the previous approaches. (c2942df)

[AaronRobinsonMSFT]

@copilot You've updated the Stateful marshaller. You need to also update the Stateless marshaller. See StatelessMarshallingStrategy starting on line 438.

[Copilot]

@copilot You've updated the Stateful marshaller. You need to also update the Stateless marshaller. See StatelessMarshallingStrategy starting on line 438.

Applied the same null pointer fix to the stateless marshalling strategy. Updated three methods in StatelessLinearCollectionSpaceAllocator: GenerateUnmarshalStatements, GenerateGuaranteedUnmarshalStatements, and GenerateCleanupCallerAllocatedResourcesStatements to use the same semantic NativeType is PointerTypeInfo check and conditional null handling logic. (19dd954)

[jkoritzinsky]

I don't think we need to check the direction here as we're already under a "not ManagedToUnmanaged" check.

[Copilot]

You're absolutely right. Removed the redundant direction check since we're already under a "not ManagedToUnmanaged" condition at line 322. (265a03f)

[jkoritzinsky]

Change the direction checks in your changes to check for "not ManagedToUnmanaged" and add some tests for arrays passed by ref

[Copilot]

Fixed the direction check and added tests for arrays passed by ref. The GenerateGuaranteedUnmarshalStatements method only runs for unmarshal scenarios so the direction check was redundant. Added TestMethodWithRef to the test interface and implementation to ensure ref array scenarios are covered. (265a03f)

[jkoritzinsky]

@copilot please address my review comments

[Copilot]

@copilot please address my review comments

Addressed both review comments: removed redundant direction checks in stateless marshalling strategy and added tests for arrays passed by ref. The changes ensure consistency with the "not ManagedToUnmanaged" pattern and provide better test coverage. (265a03f)

[jtschuster]

Superseded by #118190.

[AaronRobinsonMSFT]

Valiant effort @copilot! Thanks.