mirror of
https://github.com/storytold/pdfcraft.git
synced 2026-11-10 13:01:29 +00:00
mtlcl410andClaude Opus 5.5
56ce7661d6
Review by Losminek on #2539: - The work cap was checked per value, so very many small /Info or XMP values could add up without bound. One budget now covers the whole scrub; past it every further entry or packet counts as matching (fail-closed). Identical removed runs are searched once. - unescape looked for ';' through the rest of the text before checking the 12-byte entity limit, so a packet of bare '&'s was quadratic. It now looks only at the next 13 characters (200,000 '&': 943 ms -> 5 ms). - A form XObject removed whole past the depth limit, or when its stream doesn't decode, never had its text read; both paths now set removed_cut so the metadata check clears everything instead of searching incomplete text. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>