Split a CSV line with quoted fields

line.split(",") works right up until a product is called Mug, large. CSV has a small quoting rule, and almost every broken import breaks on it. Write the parser.

Write parse_csv_line(line) that returns the fields as a list of strings:

• Fields are separated by commas. An empty line is one empty field: [""].
• A field that starts with " is quoted. Inside it, commas are plain text and "" stands for one ". The field ends at the next lone ".
• After a closing quote, the next character must be a comma or the end of the line. Anything else raises ValueError.
• A quote that is never closed raises ValueError.
• In an unquoted field, a " that isn’t the first character is just a character.
• Don’t strip whitespace. Don’t use the csv module.

>>> parse_csv_line('7,"Mug, large","say ""hi""",')
['7', 'Mug, large', 'say "hi"', '']

Approach

Parse field by field. At the start of each field, decide its mode from the first character.

def parse_csv_line(line):
    fields = []
    i, n = 0, len(line)
    while True:
        buf = []
        if i < n and line[i] == '"':
            i += 1
            while True:
                if i >= n:
                    raise ValueError("unterminated quote")
                ch = line[i]
                if ch == '"':
                    if i + 1 < n and line[i + 1] == '"':
                        buf.append('"')
                        i += 2
                        continue
                    i += 1
                    break
                buf.append(ch)
                i += 1
            if i < n and line[i] != ",":
                raise ValueError(f"unexpected character after closing quote at {i}")
        else:
            j = line.find(",", i)
            if j == -1:
                j = n
            buf.append(line[i:j])
            i = j
        fields.append("".join(buf))
        if i >= n:
            return fields
        i += 1  # skip the comma

The loop always appends a field before it checks for the end. That one choice gives you [""] for an empty line and a trailing empty field for a,, with no special cases.

Unquoted fields use str.find to jump straight to the next comma. That is faster than a character loop, and correct, because quotes have no meaning inside an unquoted field.

Complexity

O(n) in the length of the line. Each character is looked at a constant number of times, and fields are built with "".join rather than repeated +=.

Why this matters

CSV is still how a lot of data moves between systems, and the quoting rule is where imports fail. Some writers emit \" instead of "". Some don’t quote at all. Each mistake shifts every column after it by one. A strict parser that raises on bad input is better than a lenient one that silently loads shifted columns into your warehouse. Real CSV also allows newlines inside quotes, so a “line” is not always a line. That is why production readers work on a stream of characters, not on str.splitlines().