Format certain JSON objects on one line

Question

Consider the following code:

>>> import json
>>> data = {
...     'x': [1, {'$special': 'a'}, 2],
...     'y': {'$special': 'b'},
...     'z': {'p': True, 'q': False}
... }
>>> print(json.dumps(data, indent=2))
{
  "y": {
    "$special": "b"
  },
  "z": {
    "q": false,
    "p": true
  },
  "x": [
    1,
    {
      "$special": "a"
    },
    2
  ]
}

What I want is to format the JSON so that JSON objects that have only a single property '$special' are rendered on a single line, as follows.

{
  "y": {"$special": "b"},
  "z": {
    "q": false,
    "p": true
  },
  "x": [
    1,
    {"$special": "a"},
    2
  ]
}

I have played around with implementing a custom JSONEncoder and passing that in to json.dumps as the cls argument, but the two methods on JSONEncoder each have a problem:

• The JSONEncoder default method is called for each part of data, but the return value is not a raw JSON string, so there doesn't appear to be any way to adjust its formatting.

• The JSONEncoder encode method does return a raw JSON string, but it is only called once for the data as a whole.

Is there any way I can get JSONEncoder to do what I want?

Answer 1

The json module is not really designed to give you that much control over the output; indentation is mostly meant to aid readability while debugging.

Instead of making json produce the output, you could transform the output using the standard library tokenize module:

import tokenize
from io import BytesIO


def inline_special(json_data):
    def adjust(t, ld,):
        """Adjust token line number by offset"""
        (sl, sc), (el, ec) = t.start, t.end
        return t._replace(start=(sl + ld, sc), end=(el + ld, ec))

    def transform():
        with BytesIO(json_data.encode('utf8')) as b:
            held = []  # to defer newline tokens
            lastend = None  # to track the end pos of the prev token
            loffset = 0     # line offset to adjust tokens by
            tokens = tokenize.tokenize(b.readline)
            for tok in tokens:
                if tok.type == tokenize.NL:
                    # hold newlines until we know there's no special key coming
                    held.append(adjust(tok, loffset))
                elif (tok.type == tokenize.STRING and
                        tok.string == '"$special"'):
                    # special string, collate tokens until the next rbrace
                    # held newlines are discarded, adjust the line offset
                    loffset -= len(held)
                    held = []
                    text = [tok.string]
                    while tok.exact_type != tokenize.RBRACE:
                        tok = next(tokens)
                        if tok.type != tokenize.NL:
                            text.append(tok.string)
                            if tok.string in ':,':
                                text.append(' ')
                        else:
                            loffset -= 1  # following lines all shift
                    line, col = lastend
                    text = ''.join(text)
                    endcol = col + len(text)
                    yield tokenize.TokenInfo(
                        tokenize.STRING, text, (line, col), (line, endcol),
                        '')
                    # adjust any remaining tokens on this line
                    while tok.type != tokenize.NL:
                        tok = next(tokens)
                        yield tok._replace(
                            start=(line, endcol),
                            end=(line, endcol + len(tok.string)))
                        endcol += len(tok.string)
                else:
                    # uninteresting token, yield any held newlines
                    if held:
                        yield from held
                        held = []
                    # adjust and remember last position
                    tok = adjust(tok, loffset)
                    lastend = tok.end
                    yield tok

    return tokenize.untokenize(transform()).decode('utf8')

This reformats your sample successfully:

import json

data = {
    'x': [1, {'$special': 'a'}, 2],
    'y': {'$special': 'b'},
    'z': {'p': True, 'q': False}
}

>>> print(inline_special(json.dumps(data, indent=2)))
{
  "x": [
    1,
    {"$special": "a"},
    2
  ],
  "y": {"$special": "b"},
  "z": {
    "p": true,
    "q": false
  }
}

Answer 2

I found the following regex-based solution to be simplest, albeit … regex-based.

import json
import re
data = {
    'x': [1, {'$special': 'a'}, 2],
    'y': {'$special': 'b'},
    'z': {'p': True, 'q': False}
}
text = json.dumps(data, indent=2)
pattern = re.compile(r"""
{
\s*
"\$special"
\s*
:
\s*
"
((?:[^"]|\\"))*  # Captures zero or more NotQuote or EscapedQuote
"
\s*
}
""", re.VERBOSE)
print(pattern.sub(r'{"$special": "\1"}', text))

The output follows.

{
  "x": [
    1,
    {"$special": "a"},
    2
  ],
  "y": {"$special": "b"},
  "z": {
    "q": false,
    "p": true
  }
}