dbc*_*dbc 16 c# json system.text.json
I am trying to deserialize some JSON that contains a value that is sometimes an array, and sometimes a single item.  How can I do this with System.Text.Json and JsonSerializer?  (This question is inspired by this question for Json.NET by Robert McLaws.)
I have received the following JSON:
[
  {
    "email": "john.doe@sendgrid.com",
    "timestamp": 1337966815,
    "category": [
      "newuser",
      "transactional"
    ],
    "event": "open"
  },
  {
    "email": "jane.doe@sendgrid.com",
    "timestamp": 1337966815,
    "category": "olduser",
    "event": "open"
  }
]
And I want to deserialize it to a list of the following type:
class Item
{
    public string Email { get; set; }
    public int Timestamp { get; set; }
    public string Event { get; set; }
    public List<string> Category { get; set; }
}
Using the following code:
var options = new JsonSerializerOptions
{
    PropertyNamingPolicy = JsonNamingPolicy.CamelCase,
};
var list = JsonSerializer.Deserialize<List<Item>>(json, options);
However, when I do I get the following exception:
Run Code Online (Sandbox Code Playgroud)[ { "email": "john.doe@sendgrid.com", "timestamp": 1337966815, "category": [ "newuser", "transactional" ], "event": "open" }, { "email": "jane.doe@sendgrid.com", "timestamp": 1337966815, "category": "olduser", "event": "open" } ]
The exception occurs because the value of "category" is sometimes a single string, and sometimes an array of strings.  How can I deserialize such a property with System.Text.Json?
dbc*_*dbc 24
As inspired by this answer by Brian Rogers and other answers to How to handle both a single item and an array for the same property using JSON.net, you can create a generic JsonConverter<List<T>> that checks whether the incoming JSON value is an array, and if not, deserializes an item of type T and returns the item wrapped in an appropriate list.  Even better, you can create a JsonConverterFactory that manufactures such a converter for all list types List<T> encountered in your serialization graph.
First, define the following converter and converter factory:
public class SingleOrArrayConverter<TItem> : SingleOrArrayConverter<List<TItem>, TItem>
{
    public SingleOrArrayConverter() : this(true) { }
    public SingleOrArrayConverter(bool canWrite) : base(canWrite) { }
}
public class SingleOrArrayConverterFactory : JsonConverterFactory
{
    public bool CanWrite { get; }
    public SingleOrArrayConverterFactory() : this(true) { }
    public SingleOrArrayConverterFactory(bool canWrite) => CanWrite = canWrite;
    public override bool CanConvert(Type typeToConvert)
    {
        var itemType = GetItemType(typeToConvert);
        if (itemType == null)
            return false;
        if (itemType != typeof(string) && typeof(IEnumerable).IsAssignableFrom(itemType))
            return false;
        if (typeToConvert.GetConstructor(Type.EmptyTypes) == null || typeToConvert.IsValueType)
            return false;
        return true;
    }
    public override JsonConverter CreateConverter(Type typeToConvert, JsonSerializerOptions options)
    {
        var itemType = GetItemType(typeToConvert);
        var converterType = typeof(SingleOrArrayConverter<,>).MakeGenericType(typeToConvert, itemType);
        return (JsonConverter)Activator.CreateInstance(converterType, new object [] { CanWrite });
    }
    static Type GetItemType(Type type)
    {
        // Quick reject for performance
        if (type.IsPrimitive || type.IsArray || type == typeof(string))
            return null;
        while (type != null)
        {
            if (type.IsGenericType)
            {
                var genType = type.GetGenericTypeDefinition();
                if (genType == typeof(List<>))
                    return type.GetGenericArguments()[0];
                // Add here other generic collection types as required, e.g. HashSet<> or ObservableCollection<> or etc.
            }
            type = type.BaseType;
        }
        return null;
    }
}
public class SingleOrArrayConverter<TCollection, TItem> : JsonConverter<TCollection> where TCollection : class, ICollection<TItem>, new()
{
    public SingleOrArrayConverter() : this(true) { }
    public SingleOrArrayConverter(bool canWrite) => CanWrite = canWrite;
    public bool CanWrite { get; }
    public override TCollection Read(ref Utf8JsonReader reader, Type typeToConvert, JsonSerializerOptions options)
    {
        switch (reader.TokenType)
        {
            case JsonTokenType.Null:
                return null;
            case JsonTokenType.StartArray:
                var list = new TCollection();
                while (reader.Read())
                {
                    if (reader.TokenType == JsonTokenType.EndArray)
                        break;
                    list.Add(JsonSerializer.Deserialize<TItem>(ref reader, options));
                }
                return list;
            default:
                return new TCollection { JsonSerializer.Deserialize<TItem>(ref reader, options) };
        }
    }
    public override void Write(Utf8JsonWriter writer, TCollection value, JsonSerializerOptions options)
    {
        if (CanWrite && value.Count == 1)
        {
            JsonSerializer.Serialize(writer, value.First(), options);
        }
        else
        {
            writer.WriteStartArray();
            foreach (var item in value)
                JsonSerializer.Serialize(writer, item, options);
            writer.WriteEndArray();
        }
    }
}
Then add the the converter factory to JsonSerializerOptions.Converters before deserialization:
var options = new JsonSerializerOptions
{
    Converters = { new SingleOrArrayConverterFactory() },
    PropertyNamingPolicy = JsonNamingPolicy.CamelCase,
};
var list = JsonSerializer.Deserialize<List<Item>>(json, options);
Or add a specific converter either to options or to your data model directly using JsonConverterAttribute:
class Item
{
    public string Email { get; set; }
    public int Timestamp { get; set; }
    public string Event { get; set; }
    [JsonConverter(typeof(SingleOrArrayConverter<string>))]
    public List<string> Category { get; set; }
}
If your data model uses some other type of collection, say ObservableCollection<string>, you can apply a lower level converter SingleOrArrayConverter<TCollection, TItem> as follows:
    [JsonConverter(typeof(SingleOrArrayConverter<ObservableCollection<string>, string>))]
    public ObservableCollection<string> Category { get; set; }
Notes:
If you want the converter(s) to apply only during deserialization, pass canWrite: false to the parameterized constructor:
Converters = { new SingleOrArrayConverterFactory(canWrite: false) }
The converter will still get used, but will unconditionally generate a default serialization.
The converter is not implemented for jagged 2d or nD collections such as List<List<string>>.  It is also not implemented for arrays and read-only collections.
According to Serializer support for easier object and collection converters #1562, because JsonConverter<T> lacks an async Read() method, 
A limitation of the existing [JsonConverter] model is that it must "read-ahead" during deserialization to fully populate the buffer up to the end up the current JSON level. This read-ahead only occurs when the async+stream
JsonSerializerdeserialize methods are called and only when the current JSON for that converter starts with a StartArray or StartObject token.
Thus using this converter to deserialize potentially very large arrays may have a negative performance impact.
正如在同一个线程中所讨论的,转换器 API 可能会在 System.Text.Json - 5.0 中重新设计,以完全支持转换器对数组和对象的反序列化,这意味着该转换器可能会受益于.NET 5async时的重写(不再标记为“核心”)最终发布。
演示小提琴在这里。