问题
I am writing converter for json like this:
{
"datatable": {
"data": [
[
"A85002072C",
"1994-11-15",
678.9
]
],
"columns": [
{
"name": "series_id",
"type": "String"
},
{
"name": "date",
"type": "Date"
},
{
"name": "value",
"type": "double"
}
]
},
"meta": {
"next_cursor_id": null
}
}
At the moment my converter looks like this:
public class AbsToModelConverter : JsonConverter
{
public override bool CanConvert(Type objectType)
{
return objectType.Name.Equals("AbsFseModel");
}
public override object ReadJson(JsonReader reader, Type objectType, object existingValue, JsonSerializer serializer)
{
JArray array = JArray.Load(reader);
return new QuandlAbsModel
{
SeriesId = array[0].ToString(),
Date = array[1].ToObject<DateTime>(),
Value = array[2].ToObject<decimal?>()
};
}
public override void WriteJson(JsonWriter writer, object value, JsonSerializer serializer)
{
var orderItem = value as QuandlAbsModel;
JArray arra = new JArray();
arra.Add(orderItem.SeriesId);
arra.Add(orderItem.Date);
arra.Add(orderItem.Value);
arra.WriteTo(writer);
}
}
It works at the moment, but when i am using filters my json can contain not full data, for example:
"data":[["1994-11-15",678.9]]
And my JsonConverter stops working, because there is no element array[2] and it throws error. Problem is that elements in data array don't have names (i get JSON from web API, so i can't change json at all). Is there any way to make my converter deserialize json with filters?
I have column names in my json after the data table, maybe this will help. But i don't understand how i can use them atm. Any advices?
回答1:
JLRishe is correct that your problem is solvable without a custom converter. That's a good approach in many cases. If you're able to insert a translation over the JSON serializer/deserializer, it might be simpler to write, understand, and maintain than a custom JsonConverter. It's similar in spirit to the "serialization proxy pattern" used in the Java world. In essence, you're copying your data to a new serialization-specific object before serializing, and then doing the reverse to re-serialize.
This problem is solvable with a custom converter, and I've written an example to show that it can be done, but do consider using a translation proxy/layer first.
This example is a proof-of-concept; not production-ready code. I made very little effort to defend against malformed input or other errors. Its handling of the different fields/types is also very rudimentary--any changes to the fields/types will require changes to the converter. That sort of brittleness is likely to cause bugs and maintenance headaches over time.
To narrow down the problem a bit, I reduced the original question's sample JSON to its bare minimum:
{
"datatable": {
"data": [
"A85002072C",
"1994-11-15",
678.9
],
"columns": [
{
"name": "series_id"
},
{
"name": "date"
},
{
"name": "value"
}
]
}
}
For reference, here's the C# class definition I'm deserializing to:
public class Model
{
public string SeriesId { get; set; }
public DateTime Date { get; set; }
public Decimal? Value { get; set; }
}
And here's the proof-of-concept converter:
public sealed class ModelConverter : JsonConverter
{
public static readonly ModelConverter Instance = new ModelConverter();
private ModelConverter() {}
public override bool CanConvert(Type objectType) => objectType == typeof(Model);
public override object ReadJson(JsonReader reader, Type objectType, object existingValue, JsonSerializer serializer)
{
var obj = JObject.Load(reader);
var data = (JArray)obj["datatable"]["data"];
var columns = (JArray)obj["datatable"]["columns"];
if (data.Count != columns.Count)
throw new InvalidOperationException("data and columns must contain same number of elements");
var model = new Model();
for (int i = 0; i < data.Count; i++)
{
// A "switch" works well enough so long as the number of fields is finite and small.
// There are smarter approaches, but I've kept the implementation basic
// in order to focus on the core problem that was presented.
switch (columns[i]["name"].ToString())
{
case "series_id":
model.SeriesId = data[i].ToString();
break;
case "date":
model.Date = data[i].ToObject<DateTime>();
break;
case "value":
model.Value = data[i].ToObject<decimal?>();
break;
}
}
return model;
}
public override void WriteJson(JsonWriter writer, object value, JsonSerializer serializer)
{
var data = new JArray();
var columns = new JArray();
var model = (Model)value;
// Like the "switch" used in deserialization, these "if" blocks are
// pretty rudimentary. There are better ways, but I wanted to keep
// this proof-of-concept implementation simple.
if (model.SeriesId != default(string))
{
data.Add(model.SeriesId);
columns.Add(new JObject(new JProperty("name", "series_id")));
}
if (model.Date != default(DateTime))
{
data.Add(model.Date.ToString("yyyy-MM-dd"));
columns.Add(new JObject(new JProperty("name", "date")));
}
if (model.Value != default(Decimal?))
{
data.Add(model.Value);
columns.Add(new JObject(new JProperty("name", "value")));
}
var completeObj = new JObject();
completeObj["datatable"] = new JObject();
completeObj["datatable"]["data"] = data;
completeObj["datatable"]["columns"] = columns;
completeObj.WriteTo(writer);
}
}
I wrote a few unit tests to verify the serializer. The tests are based on xUnit.Net:
[Fact]
public void TestDeserializeSampleInputWithAllFields()
{
var json = File.ReadAllText(BasePath + "sampleinput.json");
var obj = JsonConvert.DeserializeObject<Model>(json, ModelConverter.Instance);
Assert.Equal("A85002072C", obj.SeriesId);
Assert.Equal(new DateTime(1994, 11, 15), obj.Date);
Assert.Equal(678.9M, obj.Value);
}
[Fact]
public void TestSerializeSampleInputWithAllFields()
{
var model = new Model
{
SeriesId = "A85002072C",
Date = new DateTime(1994, 11, 15),
Value = 678.9M,
};
var expectedJson = File.ReadAllText(BasePath + "sampleinput.json");
Assert.Equal(expectedJson, JsonConvert.SerializeObject(model, Formatting.Indented, ModelConverter.Instance));
}
And to prove that the serializer works without all fields present:
{
"datatable": {
"data": [
"B72008039G",
543.2
],
"columns": [
{
"name": "series_id"
},
{
"name": "value"
}
]
}
}
[Fact]
public void TestDeserializeSampleInputWithNoDate()
{
var json = File.ReadAllText(BasePath + "sampleinput_NoDate.json");
var obj = JsonConvert.DeserializeObject<Model>(json, ModelConverter.Instance);
Assert.Equal("B72008039G", obj.SeriesId);
Assert.Equal(default(DateTime), obj.Date);
Assert.Equal(543.2M, obj.Value);
}
[Fact]
public void TestSerializeSampleInputWithNoDate()
{
var model = new Model
{
SeriesId = "B72008039G",
Value = 543.2M,
};
var expectedJson = File.ReadAllText(BasePath + "sampleinput_NoDate.json");
Assert.Equal(expectedJson, JsonConvert.SerializeObject(model, Formatting.Indented, ModelConverter.Instance));
}
回答2:
You don't need a JsonConverter for this.
Define classes to represent the parts of the JSON you need:
class APIResponse
{
public DataTable DataTable { get; set; }
}
class DataTable
{
public object[][] Data { get; set; }
}
Use JsonConvert.DeserializeObject<T>() to deserialize the JSON:
var parsed = JsonConvert.DeserializeObject<APIResponse>(json);
Then get your values:
var rows = parsed.DataTable.Data.Select(r => new QuandLabsModel
{
SeriesId = Convert.ToString(r[0]),
Date = Convert.ToDateTime(r[1]),
Value = Convert.ToDecimal(r[2])
});
来源：https://stackoverflow.com/questions/58997236/newtonsoft-json-converter-for-json-with-filters