258 reads

µjson — A minimal JSON parser and transformer in Go

by Oliver NguyenJanuary 9th, 2021

Too Long; Didn't Read

A software maker specialized in JavaScript and Go, I'm building OctoX - a browser extension for navigating GitHub. The package is a minimal JSON parser and transformer that works on unstructured (and trusted) JSON. It works by parsing input and calling the given callback function when encountering each item. The function is called when an object key/value or an array key is encountered. It receives 3 parameters in order: key, value, key, open bracket, value and bracketed character.

Company Mentioned

featured image - µjson — A minimal JSON parser and transformer in Go

µjson is a minimal JSON parser and transformer that works on unstructured (and trusted) JSON. It works by parsing input and calling the given callback function when encountering each item.

Motivation

Sometimes we just want to make some minimal changes to a JSON document or do some generic transformations without fully unmarshalling it. For example, removing blacklisted keys from response JSON. Why spend all the cost on unmarshalling into a

map[string]interface{}

just to immediately marshal it again. The following code is taken from StackOverflow:

{
  "responseHeader": {
    "status": 0,
    "QTime": 0,
    "params": {
      "q": "solo",
      "wt": "json"
    }
  },
  "response": {
    "numFound": 2,
    "start": 0,
    "docs": [
      { "name": "foo" },
      { "name": "bar" }
    ]
  }
}

With µjson, we can quickly write a simple transformation to remove

"responseHeader"

completely from all responses, once and forever. More on that later.

The original scenario that leads me to write the package is because of int64. When working in Go and PostgreSQL, I use int64 (instead of string) for ids because it’s more effective and has enormous space for randomly generated ids. It’s not as big as UUID, 128 bits, but still big enough for production use. In PostgreSQL, those ids can be stored as bigint and being effectively indexed. But for JavaScript, it can only process integer up to 53 bits (JavaScript has BigInt but that’s a different story, and using it will make things even more complicated).

So we need to wrap those int64s into strings before sending them to JavaScript. In Go and PostgreSQL, the JSON is

{"order_id": 12345678}

but JavaScript will see it as

{"order_id": "12345678"}

(note that the value is quoted). In Go, we can define a custom type and implement the json.Marshaler interface. But in PostgreSQL, that’s just not possible or too complicated. I wrote a service that receives JSON from PostgreSQL and converts it to be consumable by JavaScript. The service also removes some blacklisted keys or does some other transformations (for example, change

orderId

order_id

So I wrote a simple JSON parser and transformer. It can:

Print all keys and values in order
Reformat input
Remove all whitespaces
Remove blacklisted keys
Wrap int64s into strings before handing to JavaScript
Extract some values
… and more.

Important: Behavior is undefined for invalid JSON, use on trusted input only! For untrusted input, you might want to run it through json.Valid() before handing it to µjson.

Let’s see how µjson work by examples:

1. Print all keys and values in order

The callback function is called when an object key/value or an array key is encountered. It receives 3 params in order:

level

key

and

value

```
level
```
is the indentation level of the JSON, if you format it properly. It starts from 0. It increases after entering an object or array and decreases after leaving.
```
key
```
is the raw key of the current object or empty otherwise. It can be a double-quoted string or empty.
```
value
```
is the raw value of the current item or a bracket. It can be a string, number, boolean, null, or one of the following brackets:
```
{ } [ ]
```
. Values will never be empty.

It’s important to note that key and value are provided as raw. Strings are always double-quoted. It’s there for keeping the library fast and ignoring unnecessary operations. For example, when you only want to reformat the output JSON properly; you don’t want to unquote those strings and then immediately quote them again; you just need to output them unmodified. And there are ujson.Unquote() and ujson.AppendQuote() when you need to get the original strings.

When processing arrays and objects, first the open bracket (

) will be provided as value, followed by its children, and the close bracket (

). When encountering open brackets, you can make the callback function return

false

to skip the object/array entirely.

package main    

import "fmt"    
import "github.com/olvrng/ujson"    

func main() {   
    input := []byte(`{  
        "id": 12345,    
        "name": "foo",  
        "numbers": ["one", "two"],  
        "tags": {"color": "red", "priority": "high"},   
        "active": true  
    }`) 
    ujson.Walk(input, func(level int, key, value []byte) bool { 
        fmt.Printf("%2v% 12s : %s\n", level, key, value)    
        return true 
    })  
}

 0             : {
 1        "id" : 12345
 1      "name" : "foo"
 1   "numbers" : [
 2             : "one"
 2             : "two"
 1             : ]
 1      "tags" : {
 2     "color" : "red"
 2  "priority" : "high"
 1             : }
 1    "active" : true
 0             : }

0. The simplest examples

To easily get an idea on

level

key

and

value

, here are the simplest examples:

package main

import "fmt"
import "github.com/olvrng/ujson"

func main() {
	input0 := []byte(`true`)
	ujson.Walk(input0, func(level int, key, value []byte) bool {
		fmt.Printf("level=%v key=%s value=%s\n", level, key, value)
		return true
	})
	// output:
	//   level=0 key= value=true

	input1 := []byte(`{ "key": 42 }`)
	ujson.Walk(input1, func(level int, key, value []byte) bool {
		fmt.Printf("level=%v key=%s value=%s\n", level, key, value)
		return true
	})
	// output:
	//   level=0 key= value={
	//   level=1 key="key" value=42
	//   level=0 key= value=}

	input2 := []byte(`[ true ]`)
	ujson.Walk(input2, func(level int, key, value []byte) bool {
		fmt.Printf("level=%v key=%s value=%s\n", level, key, value)
		return true
	})
	// output:
	//   level=0 key= value=[
	//   level=1 key= value=true
	//   level=0 key= value=]
}

In the first example, there is only a single boolean value. The callback function is called once with

level=0

key

is empty and

value=true

In the second example, the callback function is called 3 times. Two times for open and close brackets with

level=0

, key is empty and value is the bracketed character. The other time for the only key with

level=1

key

"key"

and

value=42

. Note that the key is quoted and you need to call ujson.Unquote() to retrieve the unquoted string.

The last example is like the second, but with an array instead. Keys are always empty inside arrays.

2. Reformat input

In this example, the input JSON is formatted with correct indentation. As processing the input key by key, the callback function reconstructs the JSON. It outputs each key/value pair in its own line, prefixed with spaces equal to the param level. There is a catch, though. Valid JSON requires commas between values in objects and arrays. So there is ujson.ShouldAddComma() for checking whether a comma should be inserted.

package main

import "fmt"
import "github.com/olvrng/ujson"

func main() {
	input := []byte(`{"id":12345,"name":"foo","numbers":["one","two"],"tags":{"color":"red","priority":"high"},"active":true}`)

	b := make([]byte, 0, 1024)
	err := ujson.Walk(input, func(level int, key, value []byte) bool {
		if len(b) != 0 && ujson.ShouldAddComma(value, b[len(b)-1]) {
			b = append(b, ',')
		}
		b = append(b, '\n')
		for i := 0; i < level; i++ {
			b = append(b, '\t')
		}
		if len(key) > 0 {
			b = append(b, key...)
			b = append(b, `: `...)
		}
		b = append(b, value...)
		return true
	})
	if err != nil {
		panic(err)
	}
	fmt.Printf("%s\n", b)
}

{
	"id": 12345,
	"name": "foo",
	"numbers": [
		"one",
		"two"
	],
	"tags": {
		"color": "red",
		"priority": "high"
	},
	"active": true
}

There is a built-in method ujson.Reconstruct() when you want to remove all the whitespaces.

3. Remove blacklisted keys

This example demonstrates removing some keys from the input JSON. The key param is compared with a pre-defined list. If there is a match, the blacklisted key and its value are dropped. The callback function returns false for skipping the entire value (which may be an object or array). Note that the list is quoted, i.e.

"numbers"

and

"active"

instead of

number

and

active

. For more advanced checking, you may want to run ujson.Unquote() on the key.

package main

import "bytes"
import "fmt"
import "github.com/olvrng/ujson"

func main() {
    input := []byte(`{
        "id": 12345,
        "name": "foo",
        "numbers": ["one", "two"],
        "tags": {"color": "red", "priority": "high"},
        "active": true
    }`)

    blacklistFields := [][]byte{
        []byte(`"numbers"`), // note the quotes
        []byte(`"active"`),
    }
    b := make([]byte, 0, 1024)
    err := ujson.Walk(input, func(_ int, key, value []byte) bool {
        for _, blacklist := range blacklistFields {
            if bytes.Equal(key, blacklist) {
                // remove the key and value from the output
                return false
            }
        }

        // write to output
        if len(b) != 0 && ujson.ShouldAddComma(value, b[len(b)-1]) {
            b = append(b, ',')
        }
        if len(key) > 0 {
            b = append(b, key...)
            b = append(b, ':')
        }
        b = append(b, value...)
        return true
    })
    if err != nil {
        panic(err)
    }
    fmt.Printf("%s\n", b)
}

{"id":12345,"name":"foo","tags":{"color":"red","priority":"high"}}

As you see in the output,

"numbers"

and

"active"

are removed.

4. Wrap int64 in string

This is the original motivation behind µjson. The following example finds keys ending with

_id"

(

"order_id"

"item_id"

, etc.) and converts their values from numbers to strings, by simply wrapping them in double-quotes.

For valid JSON, values are never empty. We can test the first byte of

value

(

value[0]

) to get its type:

```
n
```
: Null
```
f
```
,
```
t
```
: Boolean
```
0
```
...
```
9
```
: Number
```
"
```
: String, see ujson.Unquote()
```
[
```
,
```
]
```
: Array
```
{
```
,
```
}
```
: Object

In this case, we check

value[0]

within

…

to see whether it’s a number, then insert double-quotes.

package main

import "bytes"
import "fmt"
import "github.com/olvrng/ujson"

func main() {
	input := []byte(`{"order_id": 12345678901234, "number": 12, "item_id": 12345678905678, "counting": [1,"2",3]}`)

	suffix := []byte(`_id"`) // note the ending quote "
	b := make([]byte, 0, 256)
	err := ujson.Walk(input, func(_ int, key, value []byte) bool {
		// Test for keys with suffix _id" and value is an int64 number. For valid json,
		// values will never be empty, so we can safely test only the first byte.
		shouldWrap := bytes.HasSuffix(key, suffix) && value[0] > '0' && value[0] <= '9'

		// transform the input, wrap values in double quotes
		if len(b) != 0 && ujson.ShouldAddComma(value, b[len(b)-1]) {
			b = append(b, ',')
		}
		if len(key) > 0 {
			b = append(b, key...)
			b = append(b, ':')
		}
		if shouldWrap {
			b = append(b, '"')
		}
		b = append(b, value...)
		if shouldWrap {
			b = append(b, '"')
		}
		return true
	})
	if err != nil {
		panic(err)
	}
	fmt.Printf("%s\n", b)
}

{"order_id":"12345678901234","number":12,"item_id":"12345678905678","counting":[1,"2",3]}

After processing, the numbers in

"order_id"

and

"item_id"

are quoted as strings. And JavaScript should be happy now! 🎉 🎉

Recap

You can start using it by

import "

github.com/olvrng/ujson

. The source code is short and easy to read. Feedback is welcome 👋

Also published at https://olvrng.github.io/w/ujson

L O A D I N G
. . . comments & more!

About Author

Oliver Nguyen@olvrng

I’m a software maker working mostly in Go and JavaScript. Share knowledge and thoughts during my journey.

Read my stories About @olvrng

TOPICS

programming #golang #json #parser #go #mujson #minimal-json-parser-in-go #json-parser-go #parser-and-transformer-go

THIS ARTICLE WAS FEATURED IN...

Join HackerNoon

Latest technology trends. Customized Experience. Curated Stories. Publish Your Ideas