mirror of
https://github.com/paperclipai/paperclip.git
synced 2026-10-28 20:21:50 +01:00
The candidate scanner found a maximal chain of base64url segments joined by dots, then offered every window of three consecutive segments in the chain. A neighbour word ending in a hyphen or underscore joined into the chain's first segment with no dot to mark the join, so the bearer's own first segment was never one of the chain's segments and the digest lookup never ran on it. The scanner now also offers a candidate that starts right after a hyphen or an underscore inside the chain's first segment. Only the first segment of a chain can carry this kind of glued prefix, since every later segment starts right after a literal dot. The extra candidate count per chain is bounded by the length of the first segment. The replacement now splices a match that starts inside a segment, so the literal text in front of it survives byte for byte. Also remove the internal tracker references the four preceding commits on this branch left in test names and comments. Co-authored-by: Paperclip <noreply@paperclip.ing>