TOML and JSON converter

A clear answer. Here’s how.

Reads TOML 1.0.0 with a hand-written parser that follows the grammar published with the specification (the ABNF file) and was run against the toml-test conformance suite for 1.0.0: all 208 valid documents give the expected data and all 501 invalid ones are refused (nine of those are byte-level encoding errors that a text box can never deliver, so they are refused when a file is decoded). It handles tables, dotted keys, inline tables, arrays of tables, the four string forms (escapes, the trimmed first newline, line-ending backslashes), integers in four bases with underscores, floats including inf and nan, booleans, and offset date-times, local date-times, local dates and local times. Duplicate keys, a table defined twice and a key reused as a table are errors, reported with a line and column, and so are an integer outside 64 bits and a float too large for 64 bits (neither is silently rounded or turned into infinity). Keys named __proto__ or constructor are ordinary data. TOML 1.1 additions (the \e and \xHH escapes, optional seconds, multi-line inline tables) are not read. A byte-order mark at the start is ignored, a CRLF inside a multi-line string is read as one line feed, and input is limited to 1,000,000 characters, 128 levels of nesting and 200,000 values. JSON has no integer or date types, so the plain form writes integers as numbers up to ±(2^53 − 1) and larger ones as strings, floats with a decimal point (3.0), inf, -inf and nan as strings, and every date or time as its text. The typed form writes each value as {"type", "value"} in the toml-test format, which keeps all of that exactly and converts back to the same TOML data. JSON to TOML reads JSON exactly (number text is kept, duplicate keys are refused), writes objects as tables and arrays of objects as arrays of tables, and cannot express null, so it refuses it or leaves it out as you choose. Comments are dropped and key order is kept, except that TOML needs plain values before sub-tables. It does not validate a schema, format a document in place or keep comments.