How to split comma-separated key-value pairs with quoted commas

Question

I know there are a lot of other posts about parsing comma-separated values, but I couldn't find one that splits key-value pairs and handles quoted commas.

I have strings like this:

age=12,name=bob,hobbies="games,reading",phrase="I'm cool!"

And I want to get this:

{
  'age': '12',
  'name': 'bob',
  'hobbies': 'games,reading',
  'phrase': "I'm cool!",
}

I tried using shlex like this:

lexer = shlex.shlex('''age=12,name=bob,hobbies="games,reading",phrase="I'm cool!"''')
lexer.whitespace_split = True
lexer.whitespace = ','
props = dict(pair.split('=', 1) for pair in lexer)

The trouble is that shlex will split the hobbies entry into two tokens, i.e. hobbies="games and reading". Is there a way to make it take the double quotes into account? Or is there another module I can use?

EDIT: Fixed typo for whitespace_split

EDIT 2: I'm not tied to using shlex. Regex is fine too, but I didn't know how to handle the matching quotes.

Answer 1

You just needed to use your shlex lexer in POSIX mode.

Add posix=True when creating the lexer.

(See the shlex parsing rules)

lexer = shlex.shlex('''age=12,name=bob,hobbies="games,reading",phrase="I'm cool!"''', posix=True)
lexer.whitespace_split = True
lexer.whitespace = ','
props = dict(pair.split('=', 1) for pair in lexer)

Outputs :

{'age': '12', 'phrase': "I'm cool!", 'hobbies': 'games,reading', 'name': 'bob'}

PS : Regular expressions won't be able to parse key-value pairs as long as the input can contain quoted = or , characters. Even preprocessing the string wouldn't be able to make the input be parsed by a regular expression, because that kind of input cannot be formally defined as a regular language.

Answer 2

It's possible to do with a regular expression. In this case, it might actually be the best option, too. I think this will work with most input, even escaped quotes such as this one: phrase='I\'m cool'

With the VERBOSE flag, it's possible to make complicated regular expressions quite readable.

import re
text = '''age=12,name=bob,hobbies="games,reading",phrase="I'm cool!"'''
regex = re.compile(
    r'''
        (?P<key>\w+)=      # Key consists of only alphanumerics
        (?P<quote>["']?)   # Optional quote character.
        (?P<value>.*?)     # Value is a non greedy match
        (?P=quote)         # Closing quote equals the first.
        ($|,)              # Entry ends with comma or end of string
    ''',
    re.VERBOSE
    )

d = {match.group('key'): match.group('value') for match in regex.finditer(text)}

print(d)  # {'name': 'bob', 'phrase': "I'm cool!", 'age': '12', 'hobbies': 'games,reading'}

Answer 3

You could abuse Python tokenizer to parse the key-value list:

#!/usr/bin/env python
from tokenize import generate_tokens, NAME, NUMBER, OP, STRING, ENDMARKER

def parse_key_value_list(text):
    key = value = None
    for type, string, _,_,_ in generate_tokens(lambda it=iter([text]): next(it)):
        if type == NAME and key is None:
            key = string
        elif type in {NAME, NUMBER, STRING}:
            value = {
                NAME: lambda x: x,
                NUMBER: int,
                STRING: lambda x: x[1:-1]
            }[type](string)
        elif ((type == OP and string == ',') or
              (type == ENDMARKER and key is not None)):
            yield key, value
            key = value = None

text = '''age=12,name=bob,hobbies="games,reading",phrase="I'm cool!"'''
print(dict(parse_key_value_list(text)))

Output

{'phrase': "I'm cool!", 'age': 12, 'name': 'bob', 'hobbies': 'games,reading'}

---

You could use a finite-state machine (FSM) to implement a stricter parser. The parser uses only the current state and the next token to parse input:

#!/usr/bin/env python
from tokenize import generate_tokens, NAME, NUMBER, OP, STRING, ENDMARKER

def parse_key_value_list(text):
    def check(condition):
        if not condition:
            raise ValueError((state, token))

    KEY, EQ, VALUE, SEP = range(4)
    state = KEY
    for token in generate_tokens(lambda it=iter([text]): next(it)):
        type, string = token[:2]
        if state == KEY:
            check(type == NAME)
            key = string
            state = EQ
        elif state == EQ:
            check(type == OP and string == '=')
            state = VALUE
        elif state == VALUE:
            check(type in {NAME, NUMBER, STRING})
            value = {
                NAME: lambda x: x,
                NUMBER: int,
                STRING: lambda x: x[1:-1]
            }[type](string)
            state = SEP
        elif state == SEP:
            check(type == OP and string == ',' or type == ENDMARKER)
            yield key, value
            state = KEY

text = '''age=12,name=bob,hobbies="games,reading",phrase="I'm cool!"'''
print(dict(parse_key_value_list(text)))