How to extract multiple JSON objects from one file?

Update: I wrote a solution that does not require reading the entire file in one go. It is too big for a stackoverflow answer, but can be found here jsonstream.

You can use json.JSONDecoder.raw_decode to decode arbitarily big strings of "stacked" JSON (so long as they can fit in memory). raw_decode stops once it has a valid object and returns the last position where was not part of the parsed object. It is poorly documented [1] (see footer), but you can pass this position back to raw_decode and it start parsing again from that position. Unfortunately, the Python json module doesn ot accept strings that have prefixing whitespace. So we need to search to find the first non-whitespace part of your document.

from json import JSONDecoder, JSONDecodeError
import re

NOT_WHITESPACE = re.compile(r'\S')

def decode_stacked(document, idx=0, decoder=JSONDecoder()):
    while True:
        match = NOT_WHITESPACE.search(document, idx)
        if not match:
            return
        idx = match.start()
        
        try:
            obj, idx = decoder.raw_decode(document, idx)
        except JSONDecodeError:
            # do something sensible if there's some error
            raise
        yield obj

s = """

{"a": 1}  


   [
1
,   
2
]


"""

for obj in decode_stacked(s):
    print(obj)

prints:

{'a': 1}
[1, 2]

Note About Missing Documentation

The current signature of raw_decode() dates from 2009, when simplejson was ported into the standard library. The documentation for raw_decode() in simplejson mentions an optional idx argument that can be used to start parsing at an offset. Given that the signature of raw_decode() has not changed since 2009, I think it is fair to assume the API is fairly stable. Especially as decode() uses the idx argument of raw_decode() to ignore prefixing whitespace when parsing a string. And this is exactly what this answer is using the idx argument for too. The documentation of raw_decode() in simplejson is:

raw_decode(s[, idx=0])

Decode a JSON document from s (a str or unicode beginning with a JSON document) starting from the index idx and return a 2-tuple of the Python representation and the index in s where the document ended.

This can be used to decode a JSON document from a string that may have extraneous data at the end, or to decode a string that has a series of JSON objects.

JSONDecodeError will be raised if the given JSON document is not valid.

So, as was mentioned in a couple comments containing the data in an array is simpler but the solution does not scale well in terms of efficiency as the data set size increases. You really should only use an iterable object when you want to access a random item in the array, otherwise, generators are the way to go. Below I have prototyped a reader function which reads each json object individually and returns a generator.

The basic idea is to signal the reader to split on the carriage character "\n" (or "\r\n" for Windows). Python can do this with the file.readline() function.

import json
def json_reader(filename):
    with open(filename) as f:
        for line in f:
            yield json.loads(line)

However, this method only really works when the file is written as you have it -- with each object separated by a newline character. Below I wrote an example of a writer that separates an array of json objects and saves each one on a new line.

def json_writer(file, json_objects):
    with open(file, "w") as f:
        for jsonobj in json_objects:
            jsonstr = json.dumps(jsonobj)
            f.write(jsonstr + "\n")

You could also do the same operation with file.writelines() and a list comprehension:

...
    json_strs = [json.dumps(j) + "\n" for j in json_objects]
    f.writelines(json_strs)
...

And if you wanted to append the data instead of writing a new file just change open(file, "w") to open(file, "a").

In the end I find this helps a great deal not only with readability when I try and open json files in a text editor but also in terms of using memory more efficiently.

On that note if you change your mind at some point and you want a list out of the reader, Python allows you to put a generator function inside of a list and populate the list automatically. In other words, just write

lst = list(json_reader(file))

Added streaming support based on the answer of @dunes:

import re
from json import JSONDecoder, JSONDecodeError

NOT_WHITESPACE = re.compile(r"[^\s]")


def stream_json(file_obj, buf_size=1024, decoder=JSONDecoder()):
    buf = ""
    ex = None
    while True:
        block = file_obj.read(buf_size)
        if not block:
            break
        buf += block
        pos = 0
        while True:
            match = NOT_WHITESPACE.search(buf, pos)
            if not match:
                break
            pos = match.start()
            try:
                obj, pos = decoder.raw_decode(buf, pos)
            except JSONDecodeError as e:
                ex = e
                break
            else:
                ex = None
                yield obj
        buf = buf[pos:]
    if ex is not None:
        raise ex

This is what I do. It assumes there will be a newline between each object, but allows each object to span multiple lines.

import json
def json_reader(filename):
    with open(filename) as f:
        text = ""
        error = None
        for line in f:
            text += line
            try:
               yield json.loads(text)
               text = ""
               e = None
            except e:
               error = e
        if error is not None:
            raise error

It isn't super efficient, since it attempts to parse the partial json text multiple times, but it is often better than loading the entire file into memory, and avoids adding another dependency.

The exception was thrown in the JSONDecoder, so maybe we can reuse that.

def json_iter(text: str):
    whitespace = re.compile(r'[ \t\n\r]*', re.VERBOSE | re.MULTILINE | re.DOTALL).match
    decoder = json.JSONDecoder()
    end = -1
    while True:
        end = whitespace(text, end).end()
        if end == len(text):
            return
        pos = end
        obj, end = decoder.raw_decode(text, idx=pos)
        yield obj


for idx, obj in enumerate(json_iter("[1][2][3][true][false]true false 1 2 3 4 5 6 7 8 9 10 {}")):
    sys.stdout.write(f"[{idx}] => {obj}\n")

[0] => [1]
[1] => [2]
[2] => [3]
[3] => [True]
[4] => [False]
[5] => True
[6] => False
[7] => 1
[8] => 2
[9] => 3
[10] => 4
[11] => 5
[12] => 6
[13] => 7
[14] => 8
[15] => 9
[16] => 10
[17] => {}