如何在System.Text.Json的Converter中修改JsonSerializerOptions的WriteIndented行为

在使用System.Text.Json进行序列化时，自定义JsonConverter<Point>后，在Write方法中创建新的JsonSerializerOptions并设置WriteIndented = false，但该设置会被忽略，序列化结果仍保持缩进格式；同时设置的PropertyNamingPolicy却能正常生效。需要实现Converter内正确修改WriteIndented的行为。

复现代码 #

open System
open System.Text.Json
open System.Text.Json.Serialization

type Point = { x: int; y: int }

JsonSerializer.Serialize ({ x=3; y=4 })

type JsonConverterPointNotIndented () =
    inherit System.Text.Json.Serialization.JsonConverter<Point>()

    override _.Read (_, _, _) = raise <| NotImplementedException()

    override c.Write (writer, value, options) =

        let o = JsonSerializerOptions ()
        o.WriteIndented <- false // 该设置被忽略，仍会输出缩进格式
        o.PropertyNamingPolicy <- JsonNamingPolicy.SnakeCaseUpper // 此设置生效

        JsonSerializer.Serialize(writer, value, o)

let o = JsonSerializerOptions()
o.WriteIndented <- true
o.Converters.Add(JsonConverterPointNotIndented())
JsonSerializer.Serialize ({ x=3; y=4 }, o)

// 实际输出：
val it: string = "{
  "X": 3,
  "Y": 4
}"

// 期望输出：
val it: string = "{"X":3,"Y":4}"

问题原因 #

当在Converter的Write方法中调用JsonSerializer.Serialize(writer, value, o)时，传入的Utf8JsonWriter已经被外部的JsonSerializerOptions（设置了WriteIndented = true）初始化，writer自身的缩进状态已经开启。新创建的JsonSerializerOptions中的WriteIndented设置无法覆盖writer已有的状态，而PropertyNamingPolicy属于序列化逻辑配置，不受writer状态影响，因此能正常生效。

解决方法 #

方法1：直接修改Utf8JsonWriter的缩进状态（推荐） #

在Write方法中临时修改writer的缩进属性，序列化完成后恢复原有状态，避免影响后续序列化逻辑：

type JsonConverterPointNotIndented () =
    inherit System.Text.Json.Serialization.JsonConverter<Point>()

    override _.Read (_, _, _) = raise <| NotImplementedException()

    override c.Write (writer, value, options) =
        // 保存writer原有缩进状态
        let originalIndentation = writer.Indentation
        let originalWriteIndented = writer.WriteIndented
        try
            // 设置为无缩进
            writer.WriteIndented <- false
            writer.Indentation <- 0
            
            let o = JsonSerializerOptions ()
            o.PropertyNamingPolicy <- JsonNamingPolicy.SnakeCaseUpper
            JsonSerializer.Serialize(writer, value, o)
        finally
            // 恢复原有状态
            writer.WriteIndented <- originalWriteIndented
            writer.Indentation <- originalIndentation

方法2：先序列化为字符串再写入writer #

通过先将对象序列化为无缩进的字符串，再直接写入writer，这种方式不受外部writer状态影响，但会额外产生字符串对象，性能略逊于方法1：

type JsonConverterPointNotIndented () =
    inherit System.Text.Json.Serialization.JsonConverter<Point>()

    override _.Read (_, _, _) = raise <| NotImplementedException()

    override c.Write (writer, value, options) =
        let o = JsonSerializerOptions ()
        o.WriteIndented <- false
        o.PropertyNamingPolicy <- JsonNamingPolicy.SnakeCaseUpper
        // 先序列化为无缩进字符串
        let jsonStr = JsonSerializer.Serialize(value, o)
        // 直接写入原始JSON字符串
        writer.WriteRawValue(jsonStr)

效果验证 #

修改后，执行序列化代码会得到预期的无缩进结果：

val it: string = "{"X":3,"Y":4}"

内容的提问来源于stack exchange，提问作者citykid