Write Date (DateOnly) instead of Datetime

[rferraton]

The Problem

I'm having trouble writing parquet with data in date format (dateonly) to the generated parquets. It appear as INT96 parquet physical datatype (microsecond precision if i'm not wrong)

If there's a way to have INT32 physical type and Date as Logical type in the parquet writer? I'd love to hear it, if it is possible.

Currently, I'm using a DbDataReader as input.
I'm building a ParquetSchema from a list of DataFields.

	private static ParquetSchema BuildSchema(DbDataReader r)
	{		
		var fields = new List<DataField>(r.FieldCount);
		var schema = r.GetColumnSchema();
		for (int i = 0; i < r.FieldCount; i++)
		{
			fields.Add(new DataField(
				r.GetName(i),
				r.GetFieldType(i),
				schema[i].AllowDBNull));
		}
		return new ParquetSchema(fields);
	}

I retrieve the schema with a DbDataReader.GetColumnSchema(), which correctly retrieves the source datatypes (date, datetime2) with DbColumn.DataTypeName.

The current DataField constructor only takes the ClrType (DateTime) as input. It can therefore no longer know the DataTypeName (Date).

public DataField(string name, Type clrType, bool? isNullable = null, bool? isArray = null, string? propertyName = null)
   : base(name, SchemaType.Data) {

    Discover(clrType, out Type baseType, out bool discIsArray, out bool discIsNullable);
    ClrType = baseType;
    if(!SchemaEncoder.IsSupported(ClrType)) {
        if(baseType == typeof(DateTimeOffset)) {
            throw new NotSupportedException($"{nameof(DateTimeOffset)} support was dropped due to numerous ambiguity issues, please use {nameof(DateTime)} from now on.");
        } else {
            throw new NotSupportedException($"type {clrType} is not supported");
        }
    }

    IsNullable = isNullable ?? discIsNullable;
    IsArray = isArray ?? discIsArray;
    ClrPropName = propertyName ?? name;
           

The Question :

Would it be possible to add a little more precision to the date typing by providing the ability to add a DataTypeName to the DataField constructor and use it?

[rferraton]

Finally activating my brain and search :

	private static ParquetSchema BuildSchema(DbDataReader r)
	{		
		var fields = new List<DataField>(r.FieldCount);
		var schema = r.GetColumnSchema();
		for (int i = 0; i < r.FieldCount; i++)
		{
			if (schema[i].DataTypeName=="date")
			{
				fields.Add(new DateTimeDataField(r.GetName(i), DateTimeFormat.Date, isNullable: schema[i].AllowDBNull));
			}
			else
			{
				fields.Add(new DataField(
					r.GetName(i),
					r.GetFieldType(i),
					schema[i].AllowDBNull));
			}	
		}
		return new ParquetSchema(fields);
	}

Do the trick :-)