I'm facing a strange bug, where .NET Core 2.1 API seems to ignore a JSON body on certain cases.
I have something like the following API method:
[Route("api/v1/accounting")]
public class AccountingController
{ sometimes it's null
||
[HttpPost("invoice/{invoiceId}/send")] ||
public async Task<int?> SendInvoice( \/
[FromRoute] int invoiceId, [FromBody] JObject body
)
{
// ...
}
}
And the relevant configuration is:
public IServiceProvider ConfigureServices(IServiceCollection services)
{
services
.AddMvcCore()
.AddJsonOptions(options =>
{
options.SerializerSettings.Converters.Add(new TestJsonConverter());
})
.AddJsonFormatters()
.AddApiExplorer();
// ...
}
Where TestJsonConverter
is a simple converter I created for testing why things doesn't work as they should, and it's simple as that:
public class TestJsonConverter : JsonConverter
{
public override object ReadJson(JsonReader reader, Type objectType, object existingValue, JsonSerializer serializer)
{
var token = JToken.Load(reader);
return token;
}
public override bool CanRead
{
get { return true; }
}
public override bool CanConvert(Type objectType)
{
return true;
}
public override void WriteJson(JsonWriter writer, object value, JsonSerializer serializer)
{
throw new NotImplementedException("Unnecessary (would be neccesary if used for serialization)");
}
}
Calling the api method using Postman works, meaning it goes through the JSON converter's CanConvert
, CanRead
, ReadJson
, and then routed to SendInvoice
with body
containing the parsed json.
However, calling the api method using HttpWebRequest (From a .NET Framework 4, if that matters) only goes through CanConvert
, then routes to SendInvoice
with body
being null.
The request body is just a simple json, something like:
{
"customerId": 1234,
"externalId": 5678
}
When I read the body directly, I get the expected value on both cases:
using (var reader = new StreamReader(context.Request.Body))
{
var requestBody = await reader.ReadToEndAsync(); // works
var parsed = JObject.Parse(requestBody);
}
I don't see any meaningful difference between the two kinds of requests - to the left is Postman's request, to the right is the HttpWebRequest:
To be sure, the Content-Type
header is set to application/json
. Also, FWIW, the HttpWebRequest
body is set as follows:
using(var requestStream = httpWebRequest.GetRequestStream())
{
JsonSerializer.Serialize(payload, requestStream);
}
And called with:
var response = (HttpWebResponse)request.GetResponse();
Question
Why does body
is null when used with HttpWebRequest
? Why does the JSON converter read methods are skipped in such cases?
The problem was in the underlying code of the serialization. So this line:
JsonSerializer.Serialize(payload, requestStream);
Was implemented using the default UTF8
property:
public void Serialize<T>(T instance, Stream stream)
{
using(var streamWriter = new StreamWriter(stream, Encoding.UTF8) // <-- Adds a BOM
using(var jsonWriter = new JsonTextWriter(streamWriter))
{
jsonSerializer.Serialize(jsonWriter, instance); // Newtonsoft.Json's JsonSerializer
}
}
The default UTF8
property adds a BOM character, as noted in the documentation:
It returns a UTF8Encoding object that provides a Unicode byte order mark (BOM). To instantiate a UTF8 encoding that doesn't provide a BOM, call any overload of the UTF8Encoding constructor.
It turns out that passing the BOM in a json is not allowed per the spec:
Implementations MUST NOT add a byte order mark (U+FEFF) to the beginning of a networked-transmitted JSON text.
Hence .NET Core [FromBody]
internal deserialization failed.
Lastly, as for why the following did work (see demo here):
using (var reader = new StreamReader(context.Request.Body))
{
var requestBody = await reader.ReadToEndAsync(); // works
var parsed = JObject.Parse(requestBody);
}
I'm not very sure. Certainly, StreamReader
also uses UTF8
property by default (see remarks here), so it shouldn't remove the BOM, and indeed it doesn't. Per a test I did (see it here), it seems that ReadToEnd
is responsible for removing the BOM.
For elaboration: