AIKIDO-2026-430108

TL;DR

parse() has a quadratic time path in parseKey, which on every key line and table header line scans from the current key position to the end of the whole document for the next dot. On an ordinary flat TOML document with many dot free key lines, or many repeated [[a]] tables, that scan repeats over the remaining document on each line, so parse time grows quadratically with document size on default options. Because parsing is synchronous, a multi megabyte externally supplied document can block the event loop for a minute or more. The fix replaces parseKey with a strictly linear implementation.