System.Numerics Vector2/3/4 static methods possibly have incorrect return type, and lack documentation

[Quidney]

Description

Define:

[IsX refers to methods that start with Is indicating it checks whether 'it is' and then X is interchangeable based on the oparation]
IsX methods of Vectors have Vector return type, instead of the expected bool. (Let me clarify, I assume it returns a bool, because it has Is in the name, and as per convention, methods such as float.IsNaN(float) return a bool. I cannot confirm because both microsoft documentation and the metadata lack any useful information.)

Example:

If we take Vector2 for the demonstration, and inspect the Vector2.IsNaN(Vector2) method, we can see that it doesn't check if the value passed is NaN, it returns a new Vector2, where the NaN part is NaN, and the Non-NaN part is 0: (It looks like it is masking, although if that is the intended functionality is not clear due to aforementioned problem

Vector2 isNaNNaN = Vector2.IsNaN(Vector2.NaN);
Vector2 isNaNNonNaN = Vector2.IsNaN(Vector2.UnitX);
Vector2 isNaNOneNaN = Vector2.IsNaN(new Vector2(float.NaN, 3.14f));

Log(isNaNNaN);
Log(isNaNNonNaN);
Log(isNaNOneNaN);

Outputs:

<NaN. NaN>
<0. 0>
<NaN. 0>

The methods also lack public documentation.

Vector2 IsX methods for reference [from Metadata]:

 //
 // Parameters:
 //   vector:
 public static Vector2 IsEvenInteger(Vector2 vector);
 //
 // Parameters:
 //   vector:
 public static Vector2 IsFinite(Vector2 vector);
 //
 // Parameters:
 //   vector:
 public static Vector2 IsInfinity(Vector2 vector);
 //
 // Parameters:
 //   vector:
 public static Vector2 IsInteger(Vector2 vector);
 //
 // Parameters:
 //   vector:
 public static Vector2 IsNaN(Vector2 vector);
 //
 // Parameters:
 //   vector:
 public static Vector2 IsNegative(Vector2 vector);
 //
 // Parameters:
 //   vector:
 public static Vector2 IsNegativeInfinity(Vector2 vector);
 //
 // Parameters:
 //   vector:
 public static Vector2 IsNormal(Vector2 vector);
 //
 // Parameters:
 //   vector:
 public static Vector2 IsOddInteger(Vector2 vector);
 //
 // Parameters:
 //   vector:
 public static Vector2 IsPositive(Vector2 vector);
 //
 // Parameters:
 //   vector:
 public static Vector2 IsPositiveInfinity(Vector2 vector);
 //
 // Parameters:
 //   vector:
 public static Vector2 IsSubnormal(Vector2 vector);
 //
 // Parameters:
 //   vector:
 public static Vector2 IsZero(Vector2 vector);

Reproduction Steps

1. Start a .NET 10 Project,
2. Make any operations with static Vector2/Vector3/Vector3 IsX methods

Expected behavior

To be given a boolean

Actual behavior

It masks the given Vector2, which complicates checking 'IsNaN' for example as IEEE NaN != NaN.

Regression?

No response

Known Workarounds

No response

Configuration

.NET 10
C#14
Visual Studio 2026 [November 2025 Feature Update]

Other information

No response

[dotnet-policy-service]

Tagging subscribers to this area: @dotnet/area-system-runtime-intrinsics
See info in area-owners.md if you want to be subscribed.

[tannergooding]

IsX methods of Vectors have Vector return type, instead of the expected bool. (Let me clarify, I assume it returns a bool, because it has Is in the name, and as per convention, methods such as float.IsNaN(float) return a bool. I cannot confirm because both microsoft documentation and the metadata lack any useful information.)

These are Vector types and the convention for vector types (when doing comparisons using named methods) is to return a vector containing the per element result. It's been that way going back to the original preview of these types around 2014 in .NET Framework.

This is explicit because there are multiple values and it is confusing to users as to whether it should do a check for "all elements" or "any elements" satisfy the condition. It is also explicitly "less efficient" to perform a reduction to a boolean expression and typically not what you want to do when working with vectors.

we can see that it doesn't check if the value passed is NaN

It does this check and returns AllBitsSet (-qNaN with a "full payload") for true and Zero for false. This allows it to then be subsequently used with masking operations or operations like ConditionalSelect

I cannot confirm because both microsoft documentation and the metadata lack any useful information.)

This is indeed a bug and I'll get it resolved shortly. In the interim, you can check the docs for System.Numerics.Vector<T> or System.Runtime.Intrinsics.Vector128<T>, as those are identical in behavior.

To be given a boolean

You're meant to use these queries with the All, Any, and None APIs, as those are "reductions" across the full vector.

For example, if you wanted to check if any elements were NaN, you'd do:

return Vector2.AnyWhereAllBitsSet(Vector2.IsNaN(vector));

If you wanted to check if any were a specific value, using the standard operator == rules, you'd do something like:

return Vector2.Any(vector, 25);

[huoyaoyuan]

See also #107596