Go How to deal with float infinity before converting to JSON

Question

I've come across a situation where I have some float64 fields that could be infinity/NaN and trying to marshal to JSON would result in an error regarding +Inf type isn't supported.

type Something interface {
  Id string `firestore:"id"`
  NumberA float64 `firestore:"numberA"`
  NumberB float64 `firestore:"numberB"`
  NumberC float64 `firestore:"numberC"`
}

This struct gets initially populated via another library (Google Firestore).

In reality this struct is much larger with a lot more fields that are floats.

I think I could use something like this loop below using reflect to to find them all, though I wonder if there is a cleaner way or more idiomatic approach.

v := reflect.ValueOf(structVar)
typeOfS := v.Type()
for i := 0; i< v.NumField(); i++ {
  if typeOfS.Field(i).Type.Kind() == reflect.Float64 && math.IsInf(v.Field(i).Interface().(float64), 1) {
    // ... some logic I'll put here
  }
}

I don't understand how to implement custom marshalling so maybe that could be an option to handle +Inf?

Create a new type e.g. `type MyFloat float64`, implement `MarshalJSON()` on `MyFloat`, and use the `MyFloat` type for the fields. — icza, Oct 24 '22 at 09:51

score 2 · Accepted Answer · answered Oct 24 '22 at 10:15

Custom handling of values can be done through custom types that implement the Marshaler interface. Your Something type, though, is malformed. It's defined as type Something interface{}, whereas that really ought the be a type Something struct:

type Something struct {
    Id      string    `firestore:"id"`
    NumberA JSONFloat `firestore:"numberA"`
    NumberB JSONFloat `firestore:"numberB"`
    NumberC JSONFloat `firestore:"numberC"`
}

type JSONFloat float64

func (j JSONFloat) MarshalJSON() ([]byte, error) {
    v := float64(j)
    if math.IsInf(j, 0) {
        // handle infinity, assign desired value to v
        // or say +/- indicates infinity
        s := "+"
        if math.IsInf(v, -1) {
            s = "-"
        }
        return []byte(s), nil
    }
    return json.Marshal(v) // marshal result as standard float64
}

func (j *JSONFloat) UnsmarshalJSON(v []byte) error {
    if s := string(v); s == "+" || s == "-" {
        // if +/- indiciates infinity
        if s == "+" {
            *j = JSONFloat(math.Inf(1))
            return nil
        }
        *j = JSONFloat(math.Inf(-1))
        return nil
    }
    // just a regular float value
    var fv float64
    if err := json.Unmarshal(v, &fv); err != nil {\
        return err
    }
    *j = JSONFloat(fv)
    return nil
}

That should do it

Looks great, where would the best place to store custom type and the marshal function? I suppose in the same file as the struct definition? — Galty, Oct 24 '22 at 12:30
@Galty yeah, the same package would be a good place, at least initially. You can always move it to another package that you can then easily reuse — Elias Van Ootegem, Oct 27 '22 at 08:58

score 0 · Answer 2 · answered Dec 26 '22 at 09:07

I created xhhuango/json to support NaN, +Inf, and -Inf.

type T struct {
    N  float64
    IP float64
    IN float64
}

func TestMarshalNaNAndInf(t *testing.T) {
    s := T{
        N:  math.NaN(),
        IP: math.Inf(1),
        IN: math.Inf(-1),
    }
    got, err := Marshal(s)
    if err != nil {
        t.Errorf("Marshal() error: %v", err)
    }
    want := `{"N":NaN,"IP":+Inf,"IN":-Inf}`
    if string(got) != want {
        t.Errorf("Marshal() = %s, want %s", got, want)
    }
}

func TestUnmarshalNaNAndInf(t *testing.T) {
    data := []byte(`{"N":NaN,"IP":+Inf,"IN":-Inf}`)
    var s T
    err := Unmarshal(data, &s)
    if err != nil {
        t.Fatalf("Unmarshal: %v", err)
    }
    if !math.IsNaN(s.N) || !math.IsInf(s.IP, 1) || !math.IsInf(s.IN, -1)     {
        t.Fatalf("after Unmarshal, s.N=%f, s.IP=%f, s.IN=%f, want NaN, +Inf, -Inf", s.N, s.IP, s.IN)
    }
}

Go How to deal with float infinity before converting to JSON

2 Answers2