[API Proposal]: Expose Vector Dataype in SqlDbType

[apoorvdeshmukh]

Background and motivation

The System.Data.SqlDbType enum represents the datatypes supported by SQL Server and is used with SqlParameter to specify the column type to be used in SQL server operations while execution SqlCommand.

With the vector datatype being supported in SQL Server link there is a need to support the vector type in Microsoft.Data.SqlClient ADO.Net provider for SQL Server.

The API suggestion is aimed at adding an enum called Vector with value 36 in SqlDbType.
Once this enum is available Microsoft.Data.SqlClient (the SQL Server driver) can then leverage the enum value to allow vector operations using Microsoft.Data.SqlClient APIs.

namespace System.Data
{
    // Specifies the SQL Server data type.
    public enum SqlDbType
    {
        Vector = 36,
    }
}

The version of Microsoft.Data.SqlClient targeting .Net 10, will be able to use the enum SqlDbType.Vector to provide vector datatype support.

API Proposal

namespace System.Data
{
    // Specifies the SQL Server data type.
    public enum SqlDbType
    {
        Vector = 36,
    }
}```


### API Usage

```csharp
using Microsoft.Data.SqlClient;


SqlParameter param = new SqlParameter();

param.SqlDbType = System.Data.SqlDbType.Vector;

Alternative Designs

No response

Risks

No response

[dotnet-policy-service]

Tagging subscribers to this area: @cheenamalhotra, @David-Engel
See info in area-owners.md if you want to be subscribed.

[cheenamalhotra]

cc @roji

[dotnet-policy-service]

Tagging subscribers to this area: @roji, @ajcvickers
See info in area-owners.md if you want to be subscribed.

[bgrainger]

Vector is being supported across many different ADO.NET providers: https://github.com/pgvector/pgvector-dotnet, mysql-net/MySqlConnector#1549.

Is there value in adopting a consistent approach for this data type (or even adding some System.Data types to support it)?

[roji]

@bgrainger there's indeed some sort of vector support in most relational databases nowadays. However, the .NET type used to represent an embedding unfortunately varies considerably - pgvector has Vector/HalfVector/SparseVector (the first two are wrappers around ReadOnlyMemory, the latter is a custom sparse vector format - there's no universal way yet to represent a sparse vector in .NET).

There's a new extension package called Microsoft.Extensions.AI that's going to be released soon, which has an Embedding type which is a base class, and Embedding<T> which wraps ReadOnlyMemory<T>. These are very new, and their main reason for existing is to be the thing returned from the IEmbeddingGenerator abstraction included there as well. ADO.NET could standardize on these types (obsoleting current types), though that would require they take a reference on Microsoft.Extensions.AI, which may or may not be appropriate.

So basically at this point I'm not sure there's anything feasible to do type-wise... We could also consider adding a DbType.Vector value to the enum, but that doesn't uniquely identify a vector type (because there are multiple - float32, float16, sparse...). And with the actual .NET type varying across providers, I'm not sure there's much use for a common DbType value...

What do you think? Am I missing other idea here?

[dotnet-policy-service]

Tagging subscribers to this area: @cheenamalhotra, @David-Engel
See info in area-owners.md if you want to be subscribed.

[roji]

Note: this is basically the same as #103925 (which was about adding SqlDbType.Json).