Crash while chaning contentType for protobuf response

Ktor Version and Engine Used (client or server and name) 1.3 Server

Describe the bug When trying to change contentType framework throws exception: io.ktor.http.UnsafeHeaderException: Header Content-Type is controlled by the engine and cannot be set explicitly

To Reproduce Steps to reproduce the behavior:

1. I Wrote Protobuf converter for Kotlin serialization library and it works great when client asks for application/protobuf, but response doens’t contain any content type

 class ProtobufConverter() : ContentConverter {
    override suspend fun convertForSend(
            context: PipelineContext<Any, ApplicationCall>,
            contentType: ContentType,
            value: Any
    ): Any? {
        context.context.response.header(HttpHeaders.ContentType,"application/protobuf")
        return ProtoBuf.dump(serializerForSending(value) as KSerializer<Any>, value)
    }

    override suspend fun convertForReceive(context: PipelineContext<ApplicationReceiveRequest, ApplicationCall>): Any? {
        val request = context.subject
        val channel = request.value as? ByteReadChannel ?: return null
        val serializer = serializerByTypeInfo(request.typeInfo)
        val content = channel.readRemaining().readBytes()
        return ProtoBuf.load(serializer, content)
    }

}

1. When I am trying to change contentType with: context.context.response.header(HttpHeaders.ContentType,“application/protobuf”) framework throws exception
2. In docs i found that ApplicationResponse class should have function “contentType” but it is not there in Ktor 1.3

Expected behavior Any way to change contentType manually when application asks for application/protobuf

[cy6erGn0m]

Most likely Protobuf support will be provided via kotlinx.serialization. We just need to design a converter config API for that purpose.

[Coneys]

Don’t you think that lack of popularity comes from fact, that serializing and deserializing Protobuf was really hard without Kotlin? Now it is really straightforward, and we get way smaller output (in our company, some responses became 3x smaller)