Handling shapechanging json in go.


There quite a few great libraries for Go when handling JSON, however, handling json that might change shape isn’t something those packages handle well.

What do you mean by json that changes shape

Working with APIs that emit JSON is usually easy, however, in some cases, a single endpoint might return a dictionary one case, an array in another. This is very hard to deal with in a typed language like Go. digjson attempts to make working with these “changing shapes” a bit easier.


Turn a json path into a reasonable value that you can work with.


Performance beyond the standard libraries json.decode (I’ll make it as fast as I can with the feature set).

Is this actually a problem?

For years, I never saw anything like this, but in the course of the last 6 months I’ve run into two APIs (one public and one private) that have this behavior. There are some other packages that solve some of the issues, but I wanted something that fit my use case exactly.

I still don’t understand, can you give me an example

Absolutely, the tests show some examples, but lets use something simple.

To start, the JSON looks like this:

jsondata = { "user": { "username: "user1" } }

var u string
was_found, err := Dig(jsondata, "user.username", &u)

The value “user1” is now in u.

Dig also works with structs:

jsondata = { "user": { "username: "user1" } }

type User struct {
    Username string `json:"username"`

var u User
was_found, err := Dig(jsondata, "user", &u)

So far, this isn’t doing much for you compared to the normal JsonDecoder other than some nice json path access, lets look at three examples I’ve run into in the real world.

jsondata_with_obj = { "user": { "username": "user1" } }
jsondata_with_nil = { "user": null }
jsondata_with_array = { "user": [{ "username": "user1"}, {"username": "user2"} ] }

This is very hard to deal with, because during the decode, you don’t know what the type will be, You can choose to decode to map[string]interface{} but this is very cumbersome with deep Json Documents.

Dig assumes you want the datatype you’re giving it, if it’s a list, and there is only a object, it will wrap it in a list (one item), if there is a null, you’ll get back a empty list. It’s easier if i just show you.

type User struct {
    Username string `json:"username"`

var users []User
was_found, err := Dig(jsondata_with_obj, "user", &users)

// users == [ User{ Username: "user1" } ]

was_found, err := Dig(jsondata_with_nil, "user", &users)

// users == []

was_found, err := Dig(jsondata_with_array, "user", &users)

// users == [ User{ Username: "user1" }, User{ Username: "user2"} ]

This seems super terrible


Why would anyone ever build an API like this?

I believe the instances where I’m experiencing this are related to a platform that returns XML and someone has written a XML to JSON converter. In these cases, it’s … understandable that things work the way they do, but it’s still very painful to consume the data.

Can I change the separator to ‘.’ if it doesn’t work for me?

Not yet, but that is a good idea. (PR welcome)

Does it implicitly convert “123” (string) to 123 (int) if I give it a struct field with the type int.

Yes, there is also DigStrict that will not do that and return an error. I recommend checking out the tests for a clearer description here.

Are there tests?

Some, but I’m sure we could write more (PR Welcome).

Is it obvious how this works?

GoLang has a reflect module that is very powerful, but it is pretty weird to use if you’re not familiar with how Go works, I encourage you to read the code!

Why are there so many questions in this README?

Good point ?


View Github