We Keep Coding

ASP.NET Web API 2 and partial updates

We are using ASP.NET Web API 2 and want to expose ability to partially edit some object in the following fashion:
HTTP PATCH /customers/1
{
"firstName": "John",
"lastName": null
}
... to set firstName to "John" and lastName to null.
HTTP PATCH /customers/1
{
"firstName": "John"
}
... in order just to update firstName to "John" and do not touch lastName at all. Suppose we have a lot of properties that we want to update with such semantic.
This is quite convenient behavior that is exercised by OData for instance.
The problem is that default JSON serializer will just come up with null in both cases, so it's impossible to distinguish.
I'm looking for some way to annotate model with some kind of wrappers (with value and flag set/unset inside) that would allow to see this difference. Any existing solutions for this?

I know that answers which are already given cover all aspects already, but just want to share concise summary of what we ended up doing and what seems to work for us pretty well.
Created a generic data contract
[DataContract]
public class RQFieldPatch<T>
{
[DataMember(Name = "value")]
public T Value { get; set; }
}
Created ad-hoc data cotnracts for patch requests
Sample is below.
[DataContract]
public class PatchSomethingRequest
{
[DataMember(Name = "prop1")]
public RQFieldPatch<EnumTypeHere> Prop1 { get; set; }
[DataMember(Name = "prop2")]
public RQFieldPatch<ComplexTypeContractHere> Prop2 { get; set; }
[DataMember(Name = "prop3")]
public RQFieldPatch<string> Prop3 { get; set; }
[DataMember(Name = "prop4")]
public RQFieldPatch<int> Prop4 { get; set; }
[DataMember(Name = "prop5")]
public RQFieldPatch<int?> Prop5 { get; set; }
}
Business Logic
Simple.
if (request.Prop1 != null)
{
// update code for Prop1, the value is stored in request.Prop1.Value
}
Json format
Simple. Not that extensive as "JSON Patch" standard, but covers all our needs.
{
"prop1": null, // will be skipped
// "prop2": null // skipped props also skipped as they will get default (null) value
"prop3": { "value": "test" } // value update requested
}
Properties
Simple contracts, simple logic
No serialization customization
Support for null values assignment
Covers any types: value, reference, complex custom types, whatever

At first I misunderstood the problem. As I was working with Xml I thought it was quite easy. Just add an attribute to the property and leave the property empty. But as I found out, Json doesn't work like that. Since I was looking for a solution that works for both xml and json, you'll find xml references in this answer. Another thing, I wrote this with a C# client in mind.
The first step is to create two classes for serialization.
public class ChangeType
{
[JsonProperty("#text")]
[XmlText]
public string Text { get; set; }
}
public class GenericChangeType<T> : ChangeType
{
}
I've chosen for a generic and a non-generic class because it is hard to cast to a generic type while this is not important. Also, for xml implementation it is necessary that XmlText is string.
XmlText is the actual value of the property. The advantage is that you can add attributes to this object and the fact that this is an object, not just string. In Xml it looks like: <Firstname>John</Firstname>
For Json this doesn't work. Json doesn't know attributes. So for Json this is just a class with properties. To implement the idea of the xml value (I will get to that later), I've renamed the property to #text. This is just a convention.
As XmlText is string (and we want to serialize to string), this is fine to store the value disregard the type. But in case of serialization, I want to know the actual type.
The drawback is that the viewmodel needs to reference these types, the advantage is that the properties are strongly typed for serialization:
public class CustomerViewModel
{
public GenericChangeType<int> Id { get; set; }
public ChangeType Firstname { get; set; }
public ChangeType Lastname { get; set; }
public ChangeType Reference { get; set; }
}
Suppose I set the values:
var customerViewModel = new CustomerViewModel
{
// Where int needs to be saved as string.
Id = new GenericeChangeType<int> { Text = "12" },
Firstname = new ChangeType { Text = "John" },
Lastname = new ChangeType { },
Reference = null // May also be omitted.
}
In xml this will look like:
<CustomerViewModel>
<Id>12</Id>
<Firstname>John</Firstname>
<Lastname />
</CustomerViewModel>
Which is enough for the server to detect the changes. But with json it will generate the following:
{
"id": { "#text": "12" },
"firstname": { "#text": "John" },
"lastname": { "#text": null }
}
It can work, because in my implementation the receiving viewmodel has the same definition. But since you are talking about serialization only and in case you use another implementation you would want:
{
"id": 12,
"firstname": "John",
"lastname": null
}
That is where we need to add a custom json converter to produce this result. The relevant code is in WriteJson, assuming you would add this converter to the serializer settings only. But for the sake of completeness I've added the readJson code as well.
public class ChangeTypeConverter : JsonConverter
{
public override bool CanConvert(Type objectType)
{
// This is important, we can use this converter for ChangeType only
return typeof(ChangeType).IsAssignableFrom(objectType);
}
public override object ReadJson(JsonReader reader, Type objectType, object existingValue, JsonSerializer serializer)
{
var value = JToken.Load(reader);
// Types match, it can be deserialized without problems.
if (value.Type == JTokenType.Object)
return JsonConvert.DeserializeObject(value.ToString(), objectType);
// Convert to ChangeType and set the value, if not null:
var t = (ChangeType)Activator.CreateInstance(objectType);
if (value.Type != JTokenType.Null)
t.Text = value.ToString();
return t;
}
public override void WriteJson(JsonWriter writer, object value, JsonSerializer serializer)
{
var d = value.GetType();
if (typeof(ChangeType).IsAssignableFrom(d))
{
var changeObject = (ChangeType)value;
// e.g. GenericChangeType<int>
if (value.GetType().IsGenericType)
{
try
{
// type - int
var type = value.GetType().GetGenericArguments()[0];
var c = Convert.ChangeType(changeObject.Text, type);
// write the int value
writer.WriteValue(c);
}
catch
{
// Ignore the exception, just write null.
writer.WriteNull();
}
}
else
{
// ChangeType object. Write the inner string (like xmlText value)
writer.WriteValue(changeObject.Text);
}
// Done writing.
return;
}
// Another object that is derived from ChangeType.
// Do not add the current converter here because this will result in a loop.
var s = new JsonSerializer
{
NullValueHandling = serializer.NullValueHandling,
DefaultValueHandling = serializer.DefaultValueHandling,
ContractResolver = serializer.ContractResolver
};
JToken.FromObject(value, s).WriteTo(writer);
}
}
At first I tried to add the converter to the class: [JsonConverter(ChangeTypeConverter)]. But the problem is that the converter will be used at all times, which creates a reference loop (as also mentioned in the comment in the code above). Also you may want to use this converter for serialization only. That is why I've added it to the serializer only:
var serializerSettings = new JsonSerializerSettings
{
NullValueHandling = NullValueHandling.Ignore,
DefaultValueHandling = DefaultValueHandling.IgnoreAndPopulate,
Converters = new List<JsonConverter> { new ChangeTypeConverter() },
ContractResolver = new Newtonsoft.Json.Serialization.CamelCasePropertyNamesContractResolver()
};
var s = JsonConvert.SerializeObject(customerViewModel, serializerSettings);
This will generate the json I was looking for and should be enough to let the server detect the changes.
-- update --
As this answer focusses on serialization, the most important thing is that lastname is part of the serialization string. It then depends on the receiving party how to deserialize the string into an object again.
Serialization and deserialization use different settings. In order to deserialize again you can use:
var deserializerSettings = new JsonSerializerSettings
{
//NullValueHandling = NullValueHandling.Ignore,
DefaultValueHandling = DefaultValueHandling.IgnoreAndPopulate,
Converters = new List<JsonConverter> { new Converters.NoChangeTypeConverter() },
ContractResolver = new Newtonsoft.Json.Serialization.CamelCasePropertyNamesContractResolver()
};
var obj = JsonConvert.DeserializeObject<CustomerViewModel>(s, deserializerSettings);
If you use the same classes for deserialization then Request.Lastname should be of ChangeType, with Text = null.
I'm not sure why removing the NullValueHandling from the deserialization settings causes problems in your case. But you can overcome this by writing an empty object as value instead of null. In the converter the current ReadJson can already handle this. But in WriteJson there has to be a modification. Instead of writer.WriteValue(changeObject.Text); you need something like:
if (changeObject.Text == null)
JToken.FromObject(new ChangeType(), s).WriteTo(writer);
else
writer.WriteValue(changeObject.Text);
This would result in:
{
"id": 12,
"firstname": "John",
"lastname": {}
}

Here's my quick and inexpensive solution...
public static ObjectType Patch<ObjectType>(ObjectType source, JObject document)
where ObjectType : class
{
JsonSerializerSettings settings = new JsonSerializerSettings
{
ContractResolver = new CamelCasePropertyNamesContractResolver()
};
try
{
String currentEntry = JsonConvert.SerializeObject(source, settings);
JObject currentObj = JObject.Parse(currentEntry);
foreach (KeyValuePair<String, JToken> property in document)
{
currentObj[property.Key] = property.Value;
}
String updatedObj = currentObj.ToString();
return JsonConvert.DeserializeObject<ObjectType>(updatedObj);
}
catch (Exception ex)
{
throw ex;
}
}
When fetching the request body from your PATCH based method, make sure to take the argument as a type such as JObject. JObject during iteration returns a KeyValuePair struct which inherently simplifies the modification process. This allows you to get your request body content without receiving a deserialized result of your desired type.
This is beneficial due to the fact that you don't need any additional verification for nullified properties. If you want your values to be nullified that also works because the Patch<ObjectType>() method only loops through properties given in the partial JSON document.
With the Patch<ObjectType>() method, you only need to pass your source or target instance, and the partial JSON document that will update your object. This method will apply camelCase based contract resolver to prevent incompatible and inaccurate property names from being made. This method will then serialize your passed instance of a certain type and turned into a JObject.
The method then replaces all properties from the new JSON document to the current and serialized document without any unnecessary if statements.
The method stringifies the current document which now is modified, and deserializes the modified JSON document to your desired generic type.
If an exception occur, the method will simply throw it. Yes, it is rather unspecific but you are the programmer, you need to know what to expect...
This can all be done on a single and simple syntax with the following:
Entity entity = AtomicModifier.Patch<Entity>(entity, partialDocument);
This is what the operation would normally look like:
// Partial JSON document (originates from controller).
JObject newData = new { role = 9001 };
// Current entity from EF persistence medium.
User user = await context.Users.FindAsync(id);
// Output:
//
// Username : engineer-186f
// Role : 1
//
Debug.WriteLine($"Username : {0}", user.Username);
Debug.WriteLine($"Role : {0}", user.Role);
// Partially updated entity.
user = AtomicModifier.Patch<User>(user, newData);
// Output:
//
// Username : engineer-186f
// Role : 9001
//
Debug.WriteLine($"Username : {0}", user.Username);
Debug.WriteLine($"Role : {0}", user.Role);
// Setting the new values to the context.
context.Entry(user).State = EntityState.Modified;
This method will work well if you can correctly map your two documents with the camelCase contract resolver.
Enjoy...
Update
I updated the Patch<T>() method with the following code...
public static T PatchObject<T>(T source, JObject document) where T : class
{
Type type = typeof(T);
IDictionary<String, Object> dict =
type
.GetProperties()
.ToDictionary(e => e.Name, e => e.GetValue(source));
string json = document.ToString();
var patchedObject = JsonConvert.DeserializeObject<T>(json);
foreach (KeyValuePair<String, Object> pair in dict)
{
foreach (KeyValuePair<String, JToken> node in document)
{
string propertyName = char.ToUpper(node.Key[0]) +
node.Key.Substring(1);
if (propertyName == pair.Key)
{
PropertyInfo property = type.GetProperty(propertyName);
property.SetValue(source, property.GetValue(patchedObject));
break;
}
}
}
return source;
}

I know I'm a little bit late on this answer, but I think I have a solution that doesn't require having to change serialization and also doesn't include reflection (This article refers you to a JsonPatch library that someone wrote that uses reflection).
Basically create a generic class representing a property that could be patched
public class PatchProperty<T> where T : class
{
public bool Include { get; set; }
public T Value { get; set; }
}
And then create models representing the objects that you want to patch where each of the properties is a PatchProperty
public class CustomerPatchModel
{
public PatchProperty<string> FirstName { get; set; }
public PatchProperty<string> LastName { get; set; }
public PatchProperty<int> IntProperty { get; set; }
}
Then your WebApi method would look like
public void PatchCustomer(CustomerPatchModel customerPatchModel)
{
if (customerPatchModel.FirstName?.Include == true)
{
// update first name
string firstName = customerPatchModel.FirstName.Value;
}
if (customerPatchModel.LastName?.Include == true)
{
// update last name
string lastName = customerPatchModel.LastName.Value;
}
if (customerPatchModel.IntProperty?.Include == true)
{
// update int property
int intProperty = customerPatchModel.IntProperty.Value;
}
}
And you could send a request with some Json that looks like
{
"LastName": { "Include": true, "Value": null },
"OtherProperty": { "Include": true, "Value": 7 }
}
Then we would know to ignore FirstName but still set the other properties to null and 7 respectively.
Note that I haven't tested this and I'm not 100% sure it would work. It would basically rely on .NET's ability to serialize the generic PatchProperty. But since the properties on the model specify the type of the generic T, I would think it would be able to. Also since we have "where T : class" on the PatchProperty declaration, the Value should be nullable. I'd be interested to know if this actually works though. Worst case you could implement a StringPatchProperty, IntPatchProperty, etc. for all your property types.

JSON Newtonsoft C# Good Practice for Serialize/ Deserialize Lists of Objects

I've readed others posts here about this question
Serializing a list of Object using Json.NET
Serializing a list to JSON
Merge two objects during serialization using json.net?
All very useful. Certain, I can serialize in one json two lists, but I cant deserialize it.
I´m working with Json Newtonsoft, C#, MVC5, framework 4.5. This is the scenario:
C# CODE
public class User
{
public int id { get; set; }
public string name { get; set; }
}
public class Request
{
public int id { get; set; }
public int idUSer{ get; set; }
}
List<User> UserList = new List<User>();
List<Request> RequestList = new List<Request>();
string json= JsonConvert.SerializeObject(new { UserList, RequestList });
JSON RESULT
{
"UserList":[
{
"id":1,
"name":"User 1"
},
{
"id":2,
"name":"User 2"
},
{
"id":3,
"name":"User 3"
}
],
"RequestList":[
{
"id":1,
"idUSer":1
},
{
"id":2,
"idUSer":1
},
{
"id":3,
"idUSer":1
},
{
"id":4,
"idUSer":2
}
]
}
C# DESERIALIZE
I dont Know how configure the settings of Json.Deserialize< ?, Settings>(json) for indicate what types of objects are being deserialized.
Change of approach
So that, change of approach, I've created a new class "Cover" in order to put the lists together and serialize one object
public class Cover
{
private List<User> user = new List<User>();
private List<Request> request = new List<Request>();
public List<User> User
{
get { return user;}
set { User = value;}
}
public List<Request> Request
{
get {return request;}
set {Request = value;}
}
}
SERIALIZE
string json = JsonConvert.SerializeObject(cover);
JSON The json result is the same.
DESERIALIZE
Cover result = JsonConvert.DeserializeObject<Cover>(json, new
JsonSerializerSettings { TypeNameHandling = TypeNameHandling.Auto });
It's work fine. My situation is resolved but I have doubts about concepts, in my mind something is not clear:
MY QUESTIONS ARE:
For the first aproach:
Do you think that there is a way to deserialize a json with different lists of objects? Is not a good practice?
Second aproach: Why jsons are equals for first situation?

In JSON.NET you need to specify the type you're about to deserialize, by supplying its name as a type argument to DeserializeObject.
However, in this line:
string json= JsonConvert.SerializeObject(new { UserList, RequestList });
you create anonymous object and then serialize it - new { UserList, RequestList }. So there is the catch - you cannot use anonymous type as type arguments.
To handle such situations, JSON.NET provides DeserializeAnonymousType<>. It doesn't require you to supply the type argument; actually you can't as you going to deserialize anonymous type. Instead it is inferred from the type of the second argument, passed to the method. So you just create a dummy, anonymous object, without data and pass it to this method.
// jsonData contains previously serialized List<User> and List<Request>
void DeserializeUserAndRequest(string jsonData)
{
var deserializedLists = new {
UserList = new List<User>(),
RequestList = new List<Request>()
};
deserializedLists = JsonConvert.DeserializeAnonymousType(jsonData, deserializedLists);
// Do your stuff here by accessing
// deserializedLists.UserList and deserializedLists.RequestLists
}
Of course this all works fine, but this approach suggests that you already know the structure of the serialized data. If this structure doesn't match the structure of the initialized by you anonymous type you'll get nothing after the DeserializeAnonymousType method. And this is valid not just for the type of the properties of the anonymous type, but for their names too.

To your first question:
I would consider the option with the Cover class the 'best practice' as you are using the same model for serialization and deserialization and it's all up to Json.NET to figure out how to do the (de)serialization magic.
If for some reason you don't want to use this approach, there are two other options:
Anonymous types: http://www.newtonsoft.com/json/help/html/DeserializeAnonymousType.htm
Deserializing into a dictionary: http://www.newtonsoft.com/json/help/html/DeserializeDictionary.htm
To your second question - Are you sure the generated JSON is absolutely the same with both approaches? (You can use a tool like www.diffchecker .com to verify)
With your second approach the top-level names should be different - it should be 'Users' instead of 'UserList' and 'Requests' instead of 'RequestList'

C# Parse/Deserialize JSON partially with Newtonsoft

I have to extract a part of json-string using .net or newtonsoft json.
JSON:
var json = "{\"method\":\"subtract\",\"parameters\":{\"minuend\":\"SOME_CUSTOM_JSON_OBJECT_DIFFERENT_FOR_EACH_METHOD\",\"subtrahend\":23}}";
C# Class:
class MyJson{
public string method { get; set; }
//public string parameters {get; set;}
public object parameters {get; set;}
}
I do not need to parse all the children of "parameters" json-object. "parameters" could be a very big object ([{obj1}...{obj1000}], objX of 1000 fields), parse which would be not performant.
I would like i.e. to pass it exactly as it is on some point, so conversion "string-C#object-string" would be redundant.
I do not want use Regexp or string transformations (string.Substring, Split and co), because of error margin, I know that all .net and newtonsoft string transformations based.
Question 1: if I define a property of type "object", how newtonsoft will handle this? (Documentation is worse than msdn, so I'm looking for the input from you, who already tried this).
static void Main(string[] args)
{
var json = "{\"method\":\"subtract\",\"parameters\":{\"minuend\":42,\"subtrahend\":23}}";
var data = JsonConvert.DeserializeObject<MyJson>(j);
// what internal representaion of data.parameters?
// How is it actually converted from json-string to an C# object (JObject/JsonObject).
}
In perfect case:
"parameters" is a string and calling
ExtractMyJson(jsonString)
gives me the json string of parameters.
Basically I need the newtonsoft version of
string ExtractMyJson(jsonString){
var p1 = jsonString.Split(",");
// .. varios string transformations
return pParams;
}
Note: please don't reference "dynamic" keyword or ask why no string transformations, it's the very specific question.

If you know that your parameters are unique you can do something like this:
class MyJson
{
public string method { get; set; }
public Dictionary<string,object> parameters { get; set; }
}
................
string json = "{\"method\":\"subtract\",\"parameters\":{\"minuend\":{\"img\": 3, \"real\": 4},\"subtrahend\":23}}";
var data = JsonConvert.DeserializeObject<MyJson>(json);
If you let it as object is going to receive the type Newtonsoft.Json.Linq.JObject.

Have you tried JTOKEN?
It is a rather simple solution to partially read basic or nested JSONs as described in this post.
For a nested JSON
{
"key1": {
"key11": "value11",
"key12": "value12"
}
"key2": "value2"
}
it would look like this
JToken token = JToken.Parse(json);
var value12 = token.SelectToken("key1.key12");
to get the element of the key "key12.
I think this could go nicely with your problem.

Well Objects are treated the same way your parent object is treated. It will start from the base of the graph. So if you have something like:
Person
{
Address Address {get;set;}
}
The Json will start Deserializing Address and then add in the Person object.
If you want to limit thesize of the graph depth you can use a setting like :
JsonConvert.DeserializeObject<List<IList<IList<string>>>>(json, new JsonSerializerSettings
{
MaxDepth = 2
});
For more configurations of the JsonSerializer check JsonSerializerSettings
If your field is an object then that object will have the KeyValuePair of every property that it holds, based on that when you cast that field you can access that type.(the behaviour is the same as assigning a type to an object in C#).
Update: So if you question using JsonObject or type, well JObject is and intermediary way to construct the json format in a generic format. But using the Type deserializatin means you can ignore properties you are not interested in. Mapping to a json with a type makes more sense because it creates a new object and dismisses the old JObject.

How to get object json when deserializing array

I have an incoming JSON, which consists array of some objects, say, Foo. I deserialize them with
result = JsonConvert.DeserializeObject<List<Foo>>(message);
Now i want to add a string property to Foo, which will store it's JSON (which i received), so that Foo'll look like:
public class Foo
{
public int MyInt { get; set; }
public bool MyBool { get; set; }
public string JSON { get; set; }
}
But i don't know how can i say JSON.Net the way it can populate such a field..
UPD
I'll clearify what i want. Say i receive JSON:
[{"MyInt":1,"MyBool":0},{"MyInt":2,"MyBool":0},{"MyInt":3,"MyBool":1}]
Here is array of 3 objects and i want, when deserializing, to add corresponding part of json to object, so that:
First object will contain {"MyInt":1,"MyBool":0}
Second object will contain {"MyInt":2,"MyBool":0}
Third object will contain {"MyInt":3,"MyBool":1}
in their JSON Property
I'll be gratefull for any help!

This is one way to do it, but it doesn't maintain the exact original JSON - but it does provide a static record of the original JSON (but without the exact format of the original values - i.e. Bool maybe be 0/1 or true/false):
var message = #"[{""MyInt"":1,""MyBool"":0},{""MyInt"":2,""MyBool"":0},{""MyInt"":3,""MyBool"":1}]";
var foos = JsonConvert.DeserializeObject<List<Foo>>(message);
var t = JsonConvert.SerializeObject(foos[0]);
foos = foos.Select(s => new Foo() { MyBool = s.MyBool, MyInt = s.MyInt, JSON = JsonConvert.SerializeObject(s) }).ToList();
If you are dealing with a lot of Foos, then you might want to find a more efficient way. There might be a way to 'update' using linq, rather than creating a new list.

Okay, i found an answer. I didn't know that i can deserialize message into JArray and then enumerate it (good job, newtonsoft:) ). Here is what i endede up with:
if (tokenType is JArray)
{
var arr = JsonConvert.DeserializeObject(message) as JArray;
foreach (var item in arr)
{
try
{
var agentParameter = item.ToObject<Foo>();
agentParameter.JSON = item.ToString();
result.Add(agentParameter);
}
catch (Exception)
{
LogProvider.Error(string.Format("Failed to Deserialize message. Message text: \r\n {0}", item.ToString()));
}
}
}

Json.net deserializing list gives duplicate items

I have just started using Newtonsoft.Json (Json.net). In my first simple test, I ran into a problem when deserializing generic lists. In my code sample below I serialize an object, containing three types of simple integer lists (property, member var and array).
The resulting json looks fine (the lists are converted into json-arrays). However, when I deserialize the json back to a new object of the same type, all list items are duplicated, expect for the array. I've illustrated that by serializing it a second time.
From searching around, I've read that there may be a "private" backing field to the lists that the deserializer also fills.
So my question is: Is there a (preferably simple) way to avoid duplicate items in following case?
Code
using System;
using System.Collections.Generic;
using Newtonsoft.Json;
namespace JsonSerializeExample
{
public class Program
{
static void Main()
{
var data = new SomeData();
var json = JsonConvert.SerializeObject(data);
Console.WriteLine("First : {0}", json);
var data2 = JsonConvert.DeserializeObject<SomeData>(json);
var json2 = JsonConvert.SerializeObject(data2);
Console.WriteLine("Second: {0}", json2);
}
}
public class SomeData
{
public string SimpleField;
public int[] IntArray;
public IList<int> IntListProperty { get; set; }
public IList<int> IntListMember;
public SomeData()
{
SimpleField = "Some data";
IntArray = new[] { 7, 8, 9 };
IntListProperty = new List<int> { 1, 2, 3 };
IntListMember = new List<int> { 4, 5, 6 };
}
}
}
Resulting output
First : {"SimpleField":"Some data","IntArray":[7,8,9],"IntListMember":[4,5,6],"IntListProperty":[1,2,3]}
Second: {"SimpleField":"Some data","IntArray":[7,8,9],"IntListMember":[4,5,6,4,5,6],"IntListProperty":[1,2,3,1,2,3]}
There may be some overlap here with Json.Net duplicates private list items. However, I think my problem is even simpler, and I still haven't figured it out.

That is because you are adding items in the constructor. A common approach in deserializers when processing a list is basically:
read the list via the getter
if the list is null: create a new list and assign via the property setter, if one
deserialize each item in turn, and append (Add) to the list
this is because most list members don't have setters, i.e.
public List<Foo> Items {get {...}} // <=== no set
Contrast to arrays, which must have a setter to be useful; hence the approach is usually:
deserialize each item in turn, and append (Add) to a temporary list
convert the list to an array (ToArray), and assign via the setter
Some serializers give you options to control this behavior (others don't); and some serializers give you the ability to bypass the constructor completely (others don't).

I'm pretty sure that this post is not relevant anymore, but for future reference, here a working solution.
Just need to specify that ObjectCreationHandling is set to Replace, i.e. Always create new objects and not to Auto (which is the default) i.e. Reuse existing objects, create new objects when needed.
var data = new SomeData();
var json = JsonConvert.SerializeObject(data);
Console.WriteLine("First : {0}", json);
var data2 = JsonConvert.DeserializeObject<SomeData>(json, new JsonSerializerSettings() { ObjectCreationHandling = ObjectCreationHandling.Replace });
var json2 = JsonConvert.SerializeObject(data2);
Console.WriteLine("Second: {0}", json2);

I encountered a similar issue with a different root cause. I was serializing and deserializing a class that looked like this:
public class Appointment
{
public List<AppointmentRevision> Revisions { get; set; }
public AppointmentRevision CurrentRevision
{
get { return Revision.LastOrDefault(); }
}
public Appointment()
{
Revisions = new List<AppointmentRevision>();
}
}
public class AppointmentRevision
{
public List<Attendee> Attendees { get; set; }
}
When I serialized this, CurrentRevision was being serialized too. I'm not sure how, but when it was deserializing it was correctly keeping a single instance of the AppointmentRevision but creating duplicates in the Attendees list. The solution was to use the JsonIgnore attribute on the CurrentRevision property.
public class Appointment
{
public List<AppointmentRevision> Revisions { get; set; }
[JsonIgnore]
public AppointmentRevision CurrentRevision
{
get { return Revision.LastOrDefault(); }
}
public Appointment()
{
Revisions = new List<AppointmentRevision>();
}
}

How to apply ObjectCreationHandling.Replace to selected properties when deserializing JSON?
Turns out (I'm in 2019), you can set the list items in your constructor as you were doing in your question. I added the ObjectCreationHandling.Replace attribute above my declaration of the list, then serialising should replace anything stored in the list with the JSON.