Fix deserialization of (typed-table) similarity when reading rows

[sl-at-ibm]

The "$similarity" field, which correctly carries a JsonIgnore attribute for a typed table, fails to be deserialized into the field when a table is read (even for a vector search with includeSimilarity=true).

This PR changes the condition for reading fields, adding an exception to the JsonIgnore rule for this case.

A test that would fail on current main and now passes is as follows:

public class SZBook
{
  [ColumnPrimaryKey(1)]
  [ColumnName("title")]
  public string Title { get; set; } = null!;

  [ColumnPrimaryKey(2)]
  [ColumnName("author")]
  public string Author { get; set; } = null!;

  [ColumnVectorize(
    provider: "nvidia",
    modelName: "nvidia/nv-embedqa-e5-v5",
    dimension: 1024
  )]
  [ColumnName("summary_genres_vector")]
  public object? SummaryGenresVector { get; set; }

  [ColumnIgnore]
  [DocumentMapping(DocumentMappingField.Similarity)]
  public double? Similarity { get; set; }
}

[...]


    [Fact]
    public async Task CheckSimilarityTypedTable()
    {

        var table = await fixture.Database.CreateTableAsync<SZBook>("szbooks");
        await table.CreateVectorIndexAsync("blu", b => b.SummaryGenresVector);
        await table.InsertManyAsync(new SZBook [] {
            new SZBook(){
                Title = "t1",
                Author = "a1",
                SummaryGenresVector = "summary 1"
            },
            new SZBook(){
                Title = "t2",
                Author = "a2",
                SummaryGenresVector = "summary 2"
            },
        });

        // Find a row
        var findOptions = new TableFindOptions<SZBook>()
        {
            Sort = Builders<SZBook>.TableSort.Vector(
                b => b.SummaryGenresVector,
                new float[] { 0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f,0.1f }
            ),
            IncludeSimilarity = true,
        };
        var result = await table.FindOneAsync(findOptions);
        Assert.NotNull(result.Similarity); // NOW PASSES!

        // Find a row, the vectorize way
        var findOptions2 = new TableFindOptions<SZBook>()
        {
            Sort = Builders<SZBook>.TableSort.Vectorize(
                b => b.SummaryGenresVector,
                "something"
            ),
            IncludeSimilarity = true,
        };
        var result2 = await table.FindOneAsync(findOptions2);
        Assert.NotNull(result2.Similarity); // FAILS currently because of https://github.com/stargate/data-api/issues/1949
    }

Note that the vector search must be done with a vector (and not a vectorize string) because, due to a still-open Data API issue, in the latter case the $similarity field does not get returned.

[sl-at-ibm]

@stephenatsembit Please, would you take a look? I'm not 100% confident mine is the right approach.

[toptobes]

Is this supposed to be using DocumentMapping for a table?

[sl-at-ibm]

Is this supposed to be using DocumentMapping for a table?

Good question, @toptobes . But on closer inspection the answer seems, somewhat surprisingly, yes, judging from the way that attribute is verified within the GetPropertyName method of the RowConverter class.

Which perhaps calls for an agnostic renaming of DocumentMappingAttribute into MappingAttribute.

Or, even better, creation of a separate attribute for tables (there's no such thing that I could find).
This table attribute would only cover the similarity case (whereas the docMapping has a range of uses: vector/ize,lexical,hybrid).

I'm keeping on hold but if you agree I can do the split in this PR.

[toptobes]

I wouldn't mind separate DocumentMapping and ColumnMapping attributes (I know "document" and "column" aren't "equivalents" but seems more intuitive than FieldMapping and more consistent w/ the other attrs than TableMapping)