Introduction
Structures, or structs, are used to collect multiple pieces of information together in one unit. These collections of information are used to describe higher-level concepts, such as an Address composed of a Street, City, State, and PostalCode. When you read this information from systems such as databases, or APIs, you can use struct tags to control how this information is assigned to the fields of a struct. Struct tags are small pieces of metadata attached to fields of a struct that provide instructions to other Go code that works with the struct.
What Does a Struct Tag Look Like?
Go struct tags are annotations that appear after the type in a Go struct declaration. Each tag is composed of short strings associated with some corresponding value.
A struct tag looks like this, with the tag offset with backtick ` characters:
type User struct {
Name string `example:"name"`
}
Other Go code is then capable of examining these structs and extracting the values assigned to specific keys it requests. Struct tags have no effect on the operation of your code without additional code that examines them.
Try this example to see what struct tags look like, and that without code from another package, they will have no effect.
package main
import "fmt"
type User struct {
Name string `example:"name"`
}
func (u *User) String() string {
return fmt.Sprintf("Hi! My name is %s", u.Name)
}
func main() {
u := &User{
Name: "Sammy",
}
fmt.Println(u)
}
This will output:
[secondary_label Output]
Hi! My name is Sammy
This example defines a User type with a Name field. The Name field has been given a struct tag of example:"name". We would refer to this specific tag in conversation as the “example struct tag” because it uses the word “example” as its key. The example struct tag has the value "name" for the Name field. On the User type, we also define the String() method required by the fmt.Stringer interface. This will be called automatically when we pass the type to fmt.Println and gives us a chance to produce a nicely formatted version of our struct.
Within the body of main, we create a new instance of our User type and pass it to fmt.Println. Even though the struct had a struct tag present, we see that it has no effect on the operation of this Go code. It will behave exactly the same if the struct tag were not present.
To use struct tags to accomplish something, other Go code must be written to examine structs at runtime. The standard library has packages that use struct tags as part of their operation. The most popular of these is the encoding/json package.
Encoding JSON
JavaScript Object Notation (JSON) is a textual format for encoding collections of data organized under different string keys. It’s commonly used to communicate data between different programs as the format is simple enough that libraries exist to decode it in many different languages. The following is an example of JSON:
{
"language": "Go",
"mascot": "Gopher"
}
This JSON object contains two keys, language and mascot. Following these keys are the associated values. Here the language key has a value of Go and mascot is assigned the value Gopher.
The JSON encoder in the standard library makes use of struct tags as annotations indicating to the encoder how you would like to name your fields in the JSON output. These JSON encoding and decoding mechanisms can be found in the encoding/json package.
Try this example to see how JSON is encoded without struct tags:
package main
import (
"encoding/json"
"fmt"
"log"
"os"
"time"
)
type User struct {
Name string
Password string
PreferredFish []string
CreatedAt time.Time
}
func main() {
u := &User{
Name: "Sammy the Shark",
Password: "fisharegreat",
CreatedAt: time.Now(),
}
out, err := json.MarshalIndent(u, "", " ")
if err != nil {
log.Println(err)
os.Exit(1)
}
fmt.Println(string(out))
}
This will print the following output:
[secondary_label Output]
{
"Name": "Sammy the Shark",
"Password": "fisharegreat",
"CreatedAt": "2019-09-23T15:50:01.203059-04:00"
}
We defined a struct describing a user with fields including their name, password, and the time the user was created. Within the main function, we create an instance of this user by supplying values for all fields except PreferredFish (Sammy likes all fish). We then passed the instance of User to the json.MarshalIndent function. This is used so we can more easily see the JSON output without using an external formatting tool. This call could be replaced with json.Marshal(u) to print JSON without any additional whitespace. The two additional arguments to json.MarshalIndent control the prefix to the output (which we have omitted with the empty string), and the characters to use for indenting, which here are two space characters. Any errors produced from json.MarshalIndent are logged and the program terminates using os.Exit(1). Finally, we cast the []byte returned from json.MarshalIndent to a string and passed the resulting string to fmt.Println for printing on the terminal.
The fields of the struct appear exactly as named. This is not the typical JSON style that you may expect, though, which uses camel casing for names of fields. You’ll change the names of the field to follow camel case style in this next example. As you’ll see when you run this example, this won’t work because the desired field names conflict with Go’s rules about exported field names.
package main
import (
"encoding/json"
"fmt"
"log"
"os"
"time"
)
type User struct {
name string
password string
preferredFish []string
createdAt time.Time
}
func main() {
u := &User{
name: "Sammy the Shark",
password: "fisharegreat",
createdAt: time.Now(),
}
out, err := json.MarshalIndent(u, "", " ")
if err != nil {
log.Println(err)
os.Exit(1)
}
fmt.Println(string(out))
}
This will present the following output:
[secondary_label Output]
{}
In this version, we’ve altered the names of the fields to be camel cased. Now Name is name, Password is password, and finally CreatedAt is createdAt. Within the body of main we’ve changed the instantiation of our struct to use these new names. We then pass the struct to the json.MarshalIndent function as before. The output, this time is an empty JSON object, {}.
Camel casing fields properly requires that the first character be lower-cased. While JSON doesn’t care how you name your fields, Go does, as it indicates the visibility of the field outside of the package. Since the encoding/json package is a separate package from the main package we’re using, we must uppercase the first character in order to make it visible to encoding/json. It would seem that we’re at an impasse. We need some way to convey to the JSON encoder what we would like this field to be named.
Using Struct Tags to Control Encoding
You can modify the previous example to have exported fields that are properly encoded with camel-cased field names by annotating each field with a struct tag. The struct tag that encoding/json recognizes has a key of json and a value that controls the output. By placing the camel-cased version of the field names as the value to the json key, the encoder will use that name instead. This example fixes the previous two attempts:
package main
import (
"encoding/json"
"fmt"
"log"
"os"
"time"
)
type User struct {
Name string <^>`json:"name"`<^>
Password string <^>`json:"password"`<^>
PreferredFish []string <^>`json:"preferredFish"`<^>
CreatedAt time.Time <^>`json:"createdAt"`<^>
}
func main() {
u := &User{
Name: "Sammy the Shark",
Password: "fisharegreat",
CreatedAt: time.Now(),
}
out, err := json.MarshalIndent(u, "", " ")
if err != nil {
log.Println(err)
os.Exit(1)
}
fmt.Println(string(out))
}
This will output:
[secondary_label Output]
{
"name": "Sammy the Shark",
"password": "fisharegreat",
"preferredFish": null,
"createdAt": "2019-09-23T18:16:17.57739-04:00"
}
We’ve changed the struct fields back to be visible to other packages by capitalizing the first letters of their names. However, this time we’ve added struct tags in the form of json:"name", where "name" was the name we wanted json.MarshalIndent to use when printing our struct as JSON.
We’ve now successfully formatted our JSON correctly. Notice, however, that the fields for some values were printed even though we did not set those values. The JSON encoder can eliminate these fields as well, if you like.
Removing Empty JSON Fields
It is common to suppress outputting fields that are unset in JSON. Since all types in Go have a “zero value,” some default value that they are set to, the encoding/json package needs additional information to be able to tell that some field should be considered unset when it assumes this zero value. Within the value part of any json struct tag, you can suffix the desired name of your field with ,omitempty to tell the JSON encoder to suppress the output of this field when the field is set to the zero value. The following example fixes the previous examples to no longer output empty fields:
package main
import (
"encoding/json"
"fmt"
"log"
"os"
"time"
)
type User struct {
Name string `json:"name"`
Password string `json:"password"`
PreferredFish []string `json:"preferredFish<^>,omitempty<^>"`
CreatedAt time.Time `json:"createdAt"`
}
func main() {
u := &User{
Name: "Sammy the Shark",
Password: "fisharegreat",
CreatedAt: time.Now(),
}
out, err := json.MarshalIndent(u, "", " ")
if err != nil {
log.Println(err)
os.Exit(1)
}
fmt.Println(string(out))
}
This example will output:
[secondary_label Output]
{
"name": "Sammy the Shark",
"password": "fisharegreat",
"createdAt": "2019-09-23T18:21:53.863846-04:00"
}
We’ve modified the previous examples so that the PreferredFish field now has the struct tag json:"preferredFish,omitempty". The presence of the ,omitempty augmentation causes the JSON encoder to skip that field, since we decided to leave it unset. This had the value null in our previous examples’ outputs.
This output is looking much better, but we’re still printing out the user’s password. The encoding/json package provides another way for us to ignore private fields entirely.
Ignoring Private Fields
Some fields must be exported from structs so that other packages can correctly interact with the type. However, the nature of these fields may be sensitive, so in these circumstances, we would like the JSON encoder to ignore the field entirely—even when it is set. This is done using the special value - as the value argument to a json: struct tag.
This example fixes the issue of exposing the user’s password.
package main
import (
"encoding/json"
"fmt"
"log"
"os"
"time"
)
type User struct {
Name string `json:"name"`
Password string `json:"<^>-<^>"`
CreatedAt time.Time `json:"createdAt"`
}
func main() {
u := &User{
Name: "Sammy the Shark",
Password: "fisharegreat",
CreatedAt: time.Now(),
}
out, err := json.MarshalIndent(u, "", " ")
if err != nil {
log.Println(err)
os.Exit(1)
}
fmt.Println(string(out))
}
When you run this example, you’ll see this output:
[secondary_label Output]
{
"name": "Sammy the Shark",
"createdAt": "2019-09-23T16:08:21.124481-04:00"
}
The only thing we’ve changed in this example from previous ones is that the password field now uses the special "-" value for its json: struct tag. In the output from this example that the password field is no longer present.
These features of the encoding/json package — ,omitempty, "-", and other options — are not standards. What a package decides to do with values of a struct tag depends on its implementation. Because the encoding/json package is part of the standard library, other packages have also implemented these features in the same way as a matter of convention. However, it’s important to read the documentation for any third-party package that uses struct tags to learn what is supported and what is not.
Conclusion
Struct tags offer a powerful means to augment the functionality of code that works with your structs. Many standard library and third-party packages offer ways to customize their operation through the use of struct tags. Using them effectively in your code provides both this customization behavior and succinctly documents how these fields are used to future developers.