The basic problem here is that Json.NET's DataTableConverter infers each DataColumn.DataType by looking at token values present in the first row only. It works this way because it streams the JSON for the table in rather than loading the entirety into an intermediate JToken hierarchy. While streaming gives better performance with reduced memory use, it means that null values in the first row may result in incorrectly typed columns.
This is a problem that comes up from time to time on stackoverflow, for instance in the question deserialize a datatable with a missing first column. In that case, the questioner knew in advance that the column type should be double. In your case, you have stated schema of datatable is dynamic, so that answer cannot be used. However, as with that question, since Json.NET is open source under the MIT License, one can create a modified version of its DataTableConverter with the necessary logic.
As it turns out, it's possible to correctly set the column type while retaining the streaming behavior by remembering columns with ambiguous data types, then replacing those columns with correctly typed columns when a proper type can be determined:
/// <summary>
/// Converts a <see cref="DataTable"/> to and from JSON.
/// </summary>
public class TypeInferringDataTableConverter : Newtonsoft.Json.Converters.DataTableConverter
{
// Adapted from https://github.com/JamesNK/Newtonsoft.Json/blob/master/Src/Newtonsoft.Json/Converters/DataTableConverter.cs
// Original license: https://github.com/JamesNK/Newtonsoft.Json/blob/master/LICENSE.md
/// <summary>
/// Reads the JSON representation of the object.
/// </summary>
/// <param name="reader">The <see cref="JsonReader"/> to read from.</param>
/// <param name="objectType">Type of the object.</param>
/// <param name="existingValue">The existing value of object being read.</param>
/// <param name="serializer">The calling serializer.</param>
/// <returns>The object value.</returns>
public override object ReadJson(JsonReader reader, Type objectType, object existingValue, JsonSerializer serializer)
{
if (reader.TokenType == JsonToken.Null)
{
return null;
}
DataTable dt = existingValue as DataTable;
if (dt == null)
{
// handle typed datasets
dt = (objectType == typeof(DataTable))
? new DataTable()
: (DataTable)Activator.CreateInstance(objectType);
}
// DataTable is inside a DataSet
// populate the name from the property name
if (reader.TokenType == JsonToken.PropertyName)
{
dt.TableName = (string)reader.Value;
reader.ReadAndAssert();
if (reader.TokenType == JsonToken.Null)
{
return dt;
}
}
if (reader.TokenType != JsonToken.StartArray)
{
throw JsonSerializationExceptionHelper.Create(reader, "Unexpected JSON token when reading DataTable. Expected StartArray, got {0}.".FormatWith(CultureInfo.InvariantCulture, reader.TokenType));
}
reader.ReadAndAssert();
var ambiguousColumnTypes = new HashSet<string>();
while (reader.TokenType != JsonToken.EndArray)
{
CreateRow(reader, dt, serializer, ambiguousColumnTypes);
reader.ReadAndAssert();
}
return dt;
}
private static void CreateRow(JsonReader reader, DataTable dt, JsonSerializer serializer, HashSet<string> ambiguousColumnTypes)
{
DataRow dr = dt.NewRow();
reader.ReadAndAssert();
while (reader.TokenType == JsonToken.PropertyName)
{
string columnName = (string)reader.Value;
reader.ReadAndAssert();
DataColumn column = dt.Columns[columnName];
if (column == null)
{
bool isAmbiguousType;
Type columnType = GetColumnDataType(reader, out isAmbiguousType);
column = new DataColumn(columnName, columnType);
dt.Columns.Add(column);
if (isAmbiguousType)
ambiguousColumnTypes.Add(columnName);
}
else if (ambiguousColumnTypes.Contains(columnName))
{
bool isAmbiguousType;
Type newColumnType = GetColumnDataType(reader, out isAmbiguousType);
if (!isAmbiguousType)
ambiguousColumnTypes.Remove(columnName);
if (newColumnType != column.DataType)
{
column = ReplaceColumn(dt, column, newColumnType, serializer);
}
}
if (column.DataType == typeof(DataTable))
{
if (reader.TokenType == JsonToken.StartArray)
{
reader.ReadAndAssert();
}
DataTable nestedDt = new DataTable();
var nestedUnknownColumnTypes = new HashSet<string>();
while (reader.TokenType != JsonToken.EndArray)
{
CreateRow(reader, nestedDt, serializer, nestedUnknownColumnTypes);
reader.ReadAndAssert();
}
dr[columnName] = nestedDt;
}
else if (column.DataType.IsArray && column.DataType != typeof(byte[]))
{
if (reader.TokenType == JsonToken.StartArray)
{
reader.ReadAndAssert();
}
List<object> o = new List<object>();
while (reader.TokenType != JsonToken.EndArray)
{
o.Add(reader.Value);
reader.ReadAndAssert();
}
Array destinationArray = Array.CreateInstance(column.DataType.GetElementType(), o.Count);
Array.Copy(o.ToArray(), destinationArray, o.Count);
dr[columnName] = destinationArray;
}
else
{
object columnValue = (reader.Value != null)
? serializer.Deserialize(reader, column.DataType) ?? DBNull.Value
: DBNull.Value;
dr[columnName] = columnValue;
}
reader.ReadAndAssert();
}
dr.EndEdit();
dt.Rows.Add(dr);
}
static object RemapValue(object oldValue, Type newType, JsonSerializer serializer)
{
if (oldValue == null)
return null;
if (oldValue == DBNull.Value)
return oldValue;
return JToken.FromObject(oldValue, serializer).ToObject(newType, serializer);
}
private static DataColumn ReplaceColumn(DataTable dt, DataColumn column, Type newColumnType, JsonSerializer serializer)
{
var newValues = Enumerable.Range(0, dt.Rows.Count).Select(i => dt.Rows[i]).Select(r => RemapValue(r[column], newColumnType, serializer)).ToList();
var ordinal = column.Ordinal;
var name = column.ColumnName;
var @namespace = column.Namespace;
var newColumn = new DataColumn(name, newColumnType);
newColumn.Namespace = @namespace;
dt.Columns.Remove(column);
dt.Columns.Add(newColumn);
newColumn.SetOrdinal(ordinal);
for (int i = 0; i < dt.Rows.Count; i++)
dt.Rows[i][newColumn] = newValues[i];
return newColumn;
}
private static Type GetColumnDataType(JsonReader reader, out bool isAmbiguous)
{
JsonToken tokenType = reader.TokenType;
switch (tokenType)
{
case JsonToken.Integer:
case JsonToken.Boolean:
case JsonToken.Float:
case JsonToken.String:
case JsonToken.Date:
case JsonToken.Bytes:
isAmbiguous = false;
return reader.ValueType;
case JsonToken.Null:
case JsonToken.Undefined:
isAmbiguous = true;
return typeof(string);
case JsonToken.StartArray:
reader.ReadAndAssert();
if (reader.TokenType == JsonToken.StartObject)
{
isAmbiguous = false;
return typeof(DataTable); // nested datatable
}
else
{
isAmbiguous = false;
bool innerAmbiguous;
// Handling ambiguity in array entries is not yet implemented because the first non-ambiguous entry in the array
// might occur anywhere in the sequence, requiring us to scan the entire array to determine the type,
// e.g., given: [null, null, null, 314, null]
// we would need to scan until the 314 value, and do:
// return typeof(Nullable<>).MakeGenericType(new[] { reader.ValueType }).MakeArrayType();
Type arrayType = GetColumnDataType(reader, out innerAmbiguous);
return arrayType.MakeArrayType();
}
default:
throw JsonSerializationExceptionHelper.Create(reader, "Unexpected JSON token when reading DataTable: {0}".FormatWith(CultureInfo.InvariantCulture, tokenType));
}
}
}
internal static class JsonSerializationExceptionHelper
{
public static JsonSerializationException Create(this JsonReader reader, string format, params object[] args)
{
// Adapted from https://github.com/JamesNK/Newtonsoft.Json/blob/master/Src/Newtonsoft.Json/JsonPosition.cs
var lineInfo = reader as IJsonLineInfo;
var path = (reader == null ? null : reader.Path);
var message = string.Format(CultureInfo.InvariantCulture, format, args);
if (!message.EndsWith(Environment.NewLine, StringComparison.Ordinal))
{
message = message.Trim();
if (!message.EndsWith(".", StringComparison.Ordinal))
message += ".";
message += " ";
}
message += string.Format(CultureInfo.InvariantCulture, "Path '{0}'", path);
if (lineInfo != null && lineInfo.HasLineInfo())
message += string.Format(CultureInfo.InvariantCulture, ", line {0}, position {1}", lineInfo.LineNumber, lineInfo.LinePosition);
message += ".";
return new JsonSerializationException(message);
}
}
internal static class StringUtils
{
// Adapted from https://github.com/JamesNK/Newtonsoft.Json/blob/master/Src/Newtonsoft.Json/Utilities/StringUtils.cs
public static string FormatWith(this string format, IFormatProvider provider, object arg0)
{
return format.FormatWith(provider, new[] { arg0 });
}
private static string FormatWith(this string format, IFormatProvider provider, params object[] args)
{
return string.Format(provider, format, args);
}
}
internal static class JsonReaderExtensions
{
public static void ReadAndAssert(this JsonReader reader)
{
if (reader == null)
throw new ArgumentNullException("reader");
if (!reader.Read())
{
throw JsonSerializationExceptionHelper.Create(reader, "Unexpected end when reading JSON.");
}
}
}
Then use it like:
var settings = new JsonSerializerSettings { Converters = new[] { new TypeInferringDataTableConverter() } };
DataTable tbl1 = Newtonsoft.Json.JsonConvert.DeserializeObject<DataTable>(jsonTable1, settings);
DataTable tbl2 = Newtonsoft.Json.JsonConvert.DeserializeObject<DataTable>(jsonTable2, settings);
Do not set NullValueHandling = NullValueHandling.Ignore since nulls are now handled correctly.
Prototype fiddle
Note that, while this class handles re-typing of columns with null values, it doesn't handle re-typing of columns containing array values where the first array item is null. For instance, if the first row of some column has the value
[null, null, null, 314, null]
Then the column type inferred would ideally be typeof( long? [] ), however that is not being implemented here. Likely it would be necessary to completely load the JSON into a JToken hierarchy to make that determination.
