Implement json.Marshaler
on your struct, which is to define a custom marshaling logic, which may double the DoubleMe
field.
See this example:
type Message struct {
DoubleMe int `json:"double_me"`
Message string `json:"message"`
}
func (m Message) MarshalJSON() ([]byte, error) {
m.DoubleMe *= 2
type Message2 Message
return json.Marshal(Message2(m))
}
func main() {
m := Message{5, "Hello, World!"}
data, err := json.Marshal(m)
fmt.Println(string(data), err)
fmt.Println("Original:", m)
}
Output (try it on the Go Playground):
{"double_me":10,"message":"Hello, World!"} <nil>
Original: {5 Hello, World!}
Some notes:
If you need this to work backwards too (e.g. when unmarshaling, divide the value by 2), also implement json.Unmarshaler
(left to the reader as an exercise).
Even though we doubled the field, after the marshaling, the original value did not change. This is because the method has a non-pointer receiver, so inside the method only a copy is acquired and modified, but not the original value. If we would've used pointer receiver, we would have to take care of restoring it.
Also note that a new type Message2
is created and used inside the MarshalJSON()
method. This is because we also used the json
package to do the "usual" marshaling once we applied our custom logic. But if we would pass m
to json.Marshal()
, that would be an endless "recursion", because the json
package would again call this MarshalJSON()
method. Instead we created a Message2
type having Message
as its underlying type. This means Message2
has zero methods (it does not implement json.Marshaler
), so passing a value of Message2
to json.Marshal()
will not call this method again, and since it has Message
as its underlying type, we can simply convert a value of type Message
to Message2
.