json: marshal nil slice as null

[gopherbot]

by awreece:

What steps will reproduce the problem?
1. Marshal a struct that contains a nil []byte member.

For example,    
    type MyType struct {
        Buf []byte
        Ptr *int
        Map map[string]string
    }
    bytes, _ := json.Marshal(&MyType{nil, nil, nil})
    fmt.Println(string(bytes))

What is the expected output?
The member will be null. I expect to see:
j{"Buf":null,"Ptr":null,"Map":null}  

What do you see instead? The member is empty:
{"Buf":"","Ptr":null,"Map":null}

(A subsequent json.Unmarshal confirms this, and produces a value equivalent to
MyType{make([]byte,0,0), nil, nil} )

Which compiler are you using (5g, 6g, 8g, gccgo)?
8g

Which operating system are you using?
Ubuntu 10.04

Which revision are you using?  (hg identify)
253f02222ab0 tip

Attachments:

1. json_test.go (1049 bytes)

[rsc]

Comment 1:

In Go, a nil byte slice *is* an empty byte slice.
We could pick off nil as a special case, of course,
but then that will break code that expects it to
be handled like any other slice.  I'm not sure
there's a way to make everyone happy.
Russ

[gopherbot]

Comment 2 by awreece:

Please correct me if I'm wrong, but a nil byte slice is an empty byte slice while a nil
map is not an empty map? Why the difference?
~Alex

[rsc]

Comment 3:

Actually, a nil map is an empty map now.
Maybe I should fix its json output too.
Russ

[rsc]

Comment 4:

On second thought, maybe both should be null.  I'm not sure.

[gopherbot]

Comment 5 by [email protected]:

Unless "nil" byte slice is encoded as JSON "null", it is impossible to distinguish
between missing value and empty string/slice when serializing byte slice fields. Also,
the current behavior is inconsistent:
Suppose we have a JSON object X = {"foo": null}. If we apply
json.Marshal(json.Unmarshal(X)), the current implementation will produce {"foo": ""}.
Oops!
I ran into this issue when using json.Marshal on goprotobuf-genereated types. Goprotobuf
consistently treats "nil" as missing field, while as json.Marshal generates empty string
from it.

[robpike]

Comment 6:

Today they should be null, tomorrow they should be empty. But if both are distinctly
representable in the output, doesn't it make sense to present the distinction?

[rsc]

Comment 7:

Owner changed to @rsc.

Status changed to Accepted.

[rsc]

Comment 8:

Status changed to HelpWanted.

[gopherbot]

Comment 10 by awreece:

I would be willing to work on this, I was looking for something to get started on.

[rsc]

Comment 11:

This issue was closed by revision 48c75c5.

Status changed to Fixed.