CSV to JSON Converter - Header Row, Delimiter Detection, Quoted Fields

The reader is a small state machine written for this tool, not a library. It goes through the text one record at a time. A value that starts with a double quote runs to the next lone double quote, so it can contain the delimiter, line breaks and two double quotes in a row, which stand for one quote. Records end at CRLF, LF or a lone CR. A byte order mark at the start of the file is dropped, and one line break at the end does not create an empty row. Where real files break the rules the reader is forgiving and tells you: a quote inside an unquoted value is kept as text, and text after a closing quote is kept and added to the value.

Detect looks at the first records and picks the delimiter (comma, semicolon, tab or pipe) that splits them into the same number of columns most often, with at least two columns. That is how a file such as Product;Price;In stock, with a decimal comma in 850,50, is read correctly. You can also choose the delimiter yourself. With the header option on, the first row gives the keys. An empty name becomes column3 and so on, and a repeated name gets _2, _3 added. A row with fewer values than the header is padded with empty values, and a row with more gets extra keys called column4 and so on. If you would rather stop, choose to report the first uneven row as an error with its row and line number.

Conversion is off by default, because guessing types loses data. When it is on, only text written the way JSON writes numbers is converted: an optional minus sign, digits with no leading zero, an optional decimal part and an optional exponent. So 42, -3.5 and 1e3 become numbers, but 007, +5, .5, 1,234 and 0x1F stay text. A number with more than 15 significant digits also stays text, because a JSON number is a 64 bit float and would lose its last digits (a 16 digit card or account number is the usual victim). The number is written exactly as it appears, so 10.50 stays 10.50. The words true and false in any case become booleans and null becomes null. A separate option turns empty cells into null. Dates are never converted.

If a quoted value is never closed, the tool does not guess. It reports the row, the line and the column where the quote was opened, because everything after an open quote would otherwise be swallowed into one value. Empty lines are skipped, and they still count in row numbers so the numbers match what a spreadsheet shows. The JSON text is written by the tool itself, in the same order as the CSV columns, so a header such as 2024 stays where it is instead of moving to the front the way a JavaScript object would, and a header named __proto__ is an ordinary key. Choose 2 or 4 spaces, a tab or minified output.