mirror of
https://github.com/storytold/pdfcraft.git
synced 2026-11-14 06:42:20 +00:00
Grzegorz MartensandClaude Opus 5.5
b1f2216115
content(), which reads each JBIG2 dictionary to know copies of it, charged the
run budget for what its filters made at the end, a unit for each 4 KiB. That
says nothing of the work: three Flates can inflate a 204-byte stream to
32 MiB, a 1-bit TIFF predictor run over all of it, and ASCIIHex skip it as
white space to nothing. The Resources review of f802198f4 measured 1.33 s for
one unit, apply returning Ok after 26.6 s with 20 of them in 11 KB, about 50 to
70 minutes a MB, never refused; remove_hidden and Optimize's discard too.
It now decodes through the run's ImageBudget (decode_filters), which pays for
what every stage reads and makes, as the images stored again already are. Past
it the run is refused with ImagesPastAllowance, which remove_hidden gives as
SharedJbig2Budget. The run budget still counts the decoded size.
a_dictionary_pays_for_every_stage_of_its_filters: two Flates of a MiB of zero
bytes under ASCIIHex spend one unit of the run and more than a MiB of the
allowance; with a MiB of allowance content() and unshare are refused.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AmsdeUCdTaD72jkDw2pJKY