#49 deserialize $vector -> $binary

[a-random-steve]

Completes Pre-GA portion of #49 (and #67). Also might eliminate need for DataAPIVector given serialization options available.

1. $binary is properly deserialized for float[] properties

2. User can control how float arrays are serialized via attributes on the properties:

public class FloatArrayWriterObject
{
    [DocumentId]
    public string _id { get; set; }
    [JsonConverter(typeof(FloatArrayWriter))]
    public float[] Vector { get; set; }
}

public class FloatBinaryWriterObject
{
    [DocumentId]
    public string _id { get; set; }
    [JsonConverter(typeof(FloatBinaryWriter))]
    public float[] Vector { get; set; }
}

Fixes #49 .

[toptobes]

For clarity, could we not have [DocumentMapping(DocumentMappingField.Vector)] just make the field automatically use [JsonConverter(typeof(FloatBinaryWriter))] while using [JsonConverter(typeof(FloatArrayWriter))] for float[]s?

[a-random-steve]

For clarity, could we not have [DocumentMapping(DocumentMappingField.Vector)] just make the field automatically use [JsonConverter(typeof(FloatBinaryWriter))] while using [JsonConverter(typeof(FloatArrayWriter))] for float[]s?

@toptobes we could if $binary is only ever used (by other clients and cql) when inside the $vector structure. I thought that float arrays as a whole could be serialized/deserialized as $binary. The current structure allows the user to control that if desired, but also handles being able to deserialize any float array from $binary structure.

[toptobes]

@toptobes we could if $binary is only ever used (by other clients and cql) when inside the $vector structure. I thought that float arrays as a whole could be serialized/deserialized as $binary. The current structure allows the user to control that if desired, but also handles being able to deserialize any float array from $binary structure.

That's fair. So to be certain, if you don't provide an explicit type, and it's a vector field (whether by DocumentMappingField.Vector or ColumnVector, it'd use $binary, and otherwise it'll just use the normal float[] writer?

And if you wanted to override it manually you do something like this, or am I misunderstanding?

[JsonConverter(typeof(FloatArrayWriter))]
[DocumentMapping(DocumentMappingField.Vector)]
public float[] Vector { get; set; }

[stephenatsembit]

@toptobes we could if $binary is only ever used (by other clients and cql) when inside the $vector structure. I thought that float arrays as a whole could be serialized/deserialized as $binary. The current structure allows the user to control that if desired, but also handles being able to deserialize any float array from $binary structure.

That's fair. So to be certain, if you don't provide an explicit type, and it's a vector field (whether by DocumentMappingField.Vector or ColumnVector, it'd use $binary, and otherwise it'll just use the normal float[] writer?

And if you wanted to override it manually you do something like this, or am I misunderstanding?

[JsonConverter(typeof(FloatArrayWriter))]
[DocumentMapping(DocumentMappingField.Vector)]
public float[] Vector { get; set; }

Yep.

[skedwards88]

we could if $binary is only ever used (by other clients and cql) when inside the $vector structure

I'm not sure I fully grasp the resolution of this thread, but in case it's relevant: python client and curl at least do support $binary outside of $vector: https://docs.datastax.com/en/astra-db-serverless/api-reference/document-methods/insert-one.html

[sl-at-ibm]

I am afraid there is a regression in the form of plain lists of floats (not 'vectors') being serialized as $binary. This is something the Data API does not accept.

Observed and reproduced with the latest commit (aeecc0c) by running test Test_FloatingPoint_StringLiterals. The FloatingPointTest object has a float[] p_list_float field, that results in the following (faulty) payload:

{
    "insertOne": {
        "document": {
            "id": "1",
            "p_float_nan": "NaN",
            "p_float_pinf": "Infinity",
            "p_float_minf": "-Infinity",
            "p_double_nan": "NaN",
            "p_double_pinf": "Infinity",
            "p_double_minf": "-Infinity",
            "p_list_double": [
                -43.21,
                "NaN",
                "Infinity",
                "-Infinity",
                12.34
            ],
            "p_set_double": null,
            "p_list_float": {
                "$binary": "wizXCv/AAAB/gAAA/4AAAEFFcKQ="
            },
            "p_set_float": null
        }
    }
}

Suggestion: if not present, add a similar test that uses typed collections and a float[] object field similarly, just to cover both collections and tables.