Json parser

clean_json_string(json_str)

Clean a JSON-like string so that it can be parsed by the built-in json module.

1. Remove single-line (//...) and multi-line (/.../) comments outside of string literals.
2. Convert single-quoted string literals to double-quoted ones, preserving internal apostrophes.
3. Replace Python-specific boolean/null literals (True, False, None) with their JSON equivalents (true, false, null).
4. Remove trailing commas in objects and arrays.

Parameters:

Name	Type	Description	Default
json_str	str	The raw JSON-like string to clean.	required

Returns:

Type	Description
str	A cleaned JSON string suitable for json.loads.

Source code in dynamiq/utils/json_parser.py

def clean_json_string(json_str: str) -> str:
    """
    Clean a JSON-like string so that it can be parsed by the built-in `json` module.

    1. Remove single-line (//...) and multi-line (/*...*/) comments outside of
       string literals.
    2. Convert single-quoted string literals to double-quoted ones, preserving
       internal apostrophes.
    3. Replace Python-specific boolean/null literals (True, False, None) with their
       JSON equivalents (true, false, null).
    4. Remove trailing commas in objects and arrays.

    Args:
        json_str: The raw JSON-like string to clean.

    Returns:
        A cleaned JSON string suitable for json.loads.
    """
    json_str = _remove_comments_outside_strings(json_str)

    pattern = r"'((?:\\'|[^'])*)'"
    json_str = re.sub(pattern, single_quoted_replacer, json_str)

    json_str = re.sub(r"\bTrue\b", "true", json_str)
    json_str = re.sub(r"\bFalse\b", "false", json_str)
    json_str = re.sub(r"\bNone\b", "null", json_str)

    json_str = re.sub(r",\s*(\]|\})", r"\1", json_str)

    return json_str

extract_json_string(s)

Extract the first JSON object or array from the string by balancing brackets. The function looks for '{' or '[' and keeps track of nested brackets until they are balanced, returning the substring that contains the complete JSON.

Parameters:

Name	Type	Description	Default
s	str	The input string potentially containing a JSON object or array.	required

Returns:

Type	Description
str | None	The extracted JSON string if found and balanced, otherwise None.

Source code in dynamiq/utils/json_parser.py

def extract_json_string(s: str) -> str | None:
    """
    Extract the first JSON object or array from the string by balancing brackets.
    The function looks for '{' or '[' and keeps track of nested brackets until
    they are balanced, returning the substring that contains the complete JSON.

    Args:
        s: The input string potentially containing a JSON object or array.

    Returns:
        The extracted JSON string if found and balanced, otherwise None.
    """
    bracket_stack: list[str] = []
    start_index: int | None = None
    in_string = False
    escape = False

    for i, char in enumerate(s):
        if char == '"' and not escape:
            in_string = not in_string
        elif char == "\\" and not escape:
            escape = True
            continue

        if not in_string:
            if char in "{[":
                if not bracket_stack:
                    start_index = i
                bracket_stack.append(char)
            elif char in "}]":
                if bracket_stack:
                    opening_bracket = bracket_stack.pop()
                    if (opening_bracket == "{" and char != "}") or (opening_bracket == "[" and char != "]"):
                        return None
                    if not bracket_stack and start_index is not None:
                        return s[start_index : i + 1]
                else:
                    return None
        escape = False

    return None

parse_llm_json_output(response)

Attempt to parse the received LLM output into a JSON object or array. If direct parsing fails, looks for the first balanced JSON substring, then tries corrections.

Parameters:

Name	Type	Description	Default
response	str	The raw output from the LLM.	required

Returns:

Type	Description
dict[str, Any] | list[Any]	A Python dict or list representing the parsed JSON.

Raises:

Type	Description
ValueError	If the output cannot be parsed into valid JSON.

Source code in dynamiq/utils/json_parser.py

def parse_llm_json_output(response: str) -> dict[str, Any] | list[Any]:
    """
    Attempt to parse the received LLM output into a JSON object or array.
    If direct parsing fails, looks for the first balanced JSON substring,
    then tries corrections.

    Args:
        response: The raw output from the LLM.

    Returns:
        A Python dict or list representing the parsed JSON.

    Raises:
        ValueError: If the output cannot be parsed into valid JSON.
    """
    try:
        return json.loads(response)
    except json.JSONDecodeError:
        pass

    json_str = extract_json_string(response)
    if not json_str:
        raise ValueError(f"Response from LLM is not valid JSON: {response}")

    try:
        return json.loads(json_str)
    except json.JSONDecodeError:
        pass

    cleaned_json_str = clean_json_string(json_str)
    try:
        return json.loads(cleaned_json_str)
    except json.JSONDecodeError as e:
        raise ValueError(f"Failed to parse JSON after corrections: {e}")

single_quoted_replacer(match)

A helper function for clean_json_string to replace single-quoted JSON-like string literals with double-quoted equivalents, preserving internal apostrophes.

Parameters:

Name	Type	Description	Default
match	Match[str]	The regular expression match object for a single-quoted string.	required

Returns:

Type	Description
str	The corresponding double-quoted string literal.

Source code in dynamiq/utils/json_parser.py

def single_quoted_replacer(match: Match[str]) -> str:
    """
    A helper function for clean_json_string to replace single-quoted JSON-like
    string literals with double-quoted equivalents, preserving internal
    apostrophes.

    Args:
        match: The regular expression match object for a single-quoted string.

    Returns:
        The corresponding double-quoted string literal.
    """
    content = match.group(1)
    content = content.replace("\\'", "'")
    content = content.replace('"', '\\"')
    return f'"{content}"'