JSON encoding and anonymous fields

[DisposaBoy]

it's certainly possible and I'm sure it used to support this use-case but that support was removed for go1. the rationale is somewhre. I remember reading so it's either on this group or golang-dev . IIRC John Asmuth forked the older pkg on github thus restoring that functionality

[Kyle Lemons]

XML and JSON had different opinions about how to encode anonymous fields.  Rather than try to normalize on a way to handle them consistently, I believe, the decision was to not handle them in either case in order to leave future freedom to decide upon a proper strategy.

[John Asmuth]

I didn't restore it - the original functionality used the name of the type for the name of the json object. My version just does it inline.

That is,

type W struct {

    X string

}

type Y struct {

    W

    Z string

}

would get a json object like { "X": "whatever", "Z": "whatever" } with my fork. With the original anonymous field handling, it'd be { "W": { "X": "whatever" }, "Z": "whatever" }.

You can get it at https://github.com/skelterjohn/json

[Joel Webber]

Thanks, all. I presumed it was something like that -- I can definitely see how it gets tricky to decide the right default behavior in this case. We've actually settled on using a different pattern that makes the JSON prettier and easier to switch on, which I've seen come up elsewhere on the group. FWIW, I actually really like it, even though it's a bit non-obvious at first:

type Message struct {

  Type string

  First *FirstSubtype `json:"omitempty"`

  Second *SecondSubtype `json:"omitempty"`

}

type FirstSubtype struct {

  Something string

}

type SecondSubtype struct {

  SomethingElse string

}

This is nice because you end up with slightly clearer JSON and you can switch on the type and dispatch more easily:

switch msg.Type {

  case "first":

    doSomethingWith(msg.First)

  case "second":

    doSomethingElseWith(msg.Second)