Grzegorz MartensandClaude Opus 5.5
083191f970
Filters were charged only for what they handed a JBIG2 decoder, once they had handed it over: a chain that failed, or went past its image's limit, cost its time and nothing else, and images that weren't JBIG2 weren't charged at all. A 734 KB file of 200 images of Flate inside Flate took an image export 111 s, and that grew with the file. It now takes 12 s. Each byte read and each byte a stage decodes is now a step of the same allowance, paid as it happens, so no stage runs past what is left. That holds for every image read through ImageBudget, and a fax image pays for its rows and its code. JBIG2 globals pay when they have to be read, and kept ones don't. Optimize's two reads that went round the budget go through it, and merging fragments shares it. Optimize reads an image twice, to pack it and to resample it, so a refused image is now counted once a run rather than once a read. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01AmsdeUCdTaD72jkDw2pJKY
pdfcraft-redact
Layer L4 (may use edit): redaction, architecture §11.2.
// Marks are Redact annotations (pdfcraft-annot: Shape::Redact { quads, overlay }).
let marks: Vec<Mark> = marks(&doc); // page, areas (user space), fill, overlay text
let report = apply(&mut doc, None)?; // everything, or Some(&[pages])
clear_marks(&mut doc, None)?; // remove marks without applying
apply removes, under every marked area:
- text: glyphs are cut out of
Tj/TJ/'/"and replaced by aTJdisplacement of the same advance, so the rest of the line keeps its exact position. Widths come from/Widths,/W//DW(composite fonts, embedded CMaps with codespace and CID ranges), Type 3 font matrices, and approximations for the standard 14. A glyph goes when it overlaps an area by more than a fifth of its size (or a point) both ways; zero-size text goes when its origin is inside; - images: fully covered → removed; partly covered → a copy with the covered pixels cleared
(8/16-bit and 1/2/4-bit images, image masks; Flate/LZW/RL/A85/AHx, and CCITT fax or JBIG2 as
the last filter, decoded through
Document::decoded_image_within, a JBIG2 image with its/JBIG2Globals;/DecodeParmsgiven by reference, numbers inside included, count). The copy is Flate, and refers to no globals: when no other image uses them, they leave the file with the original on the full rewrite; when one does, they stay as they are, and may hold the shapes of marks under the area (not where they were). Codecs redaction doesn't decode (DCT, JPX) are removed whole (fail-closed), and so is a JBIG2 image whose page isn't its size (the renderer draws the whole page into the image, so the rows cleared wouldn't be the ones shown) or of more than one bit or component, and Group 3 on bytes (/K0 or more,/EncodedByteAlign): PdfCraft's renderer draws some of its layouts (the fill ending each end-of-line code on a byte) as other rows than the decoder reads, so a mark over what is shown needn't cover what is cleared; inline images under an area are removed. Images with separate/SMaskor/Maskdata, and images whose data decodes to more than three times their declared raster and a byte per row, plus 4 KiB (ASCIIHex text broken into lines under Flate is a little over twice; for a fax image's code, about a byte per pixel), are removed whole; cleared copies drop/Alternatesso alternate pixels cannot survive; A run that would decode more of its images than one redaction may (pdfcraft_cos::ImageBudget, filters and fax and JBIG2 decoders alike) is refused and changes nothing: such an image can't be cleared, and removing it whole would take what no mark covers with it. - vectors: covered paths are removed; partly covered paths and shadings are clipped so nothing paints inside the areas;
- form XObjects: rewritten recursively into new objects (pages sharing the original keep it);
- annotations whose rectangle overlaps (with their pop-ups and their replies,
/IRT), and form fields with a widget under a mark; - the marks themselves, replaced by boxes in their fill colour (and overlay text) drawn into the
page as a
/PCMark /Redactionstream.
A verification pass re-reads each redacted page and fails the operation if any glyph or inline
image is still under an area. It checks every content stream the page had, whatever its
dictionary calls it; only the overlay streams this apply added (Report::overlays) are skipped.
Text is placed with the font and size set by Tf or by an ExtGState's /Font through gs, and a
form XObject starts from the text state of the content that draws it. Content that can't be
decoded makes the operation fail rather than leave it unredacted, and so does an inline image the
renderer's tokenizer reads no image for (RedactError::Unreadable): where the page's operators
resume after it would be a guess. A Redact annotation with neither a usable /QuadPoints array nor a usable
/Rect makes it fail with RedactError::UnreadableMark before anything changes, instead of
skipping that mark. The count covers the marks on the pages being applied.
Before changing any page, apply prepares the redaction appearances with a shared 8 MiB
budget for their uncompressed PDF content across the selected pages. It bounds repeated-text
counts and working line buffers before allocating them, and measures and encodes each repeated
line once. Invalid or excessive appearances return RedactError::OverlayLimit without changing
the document. Reduce the marked area, disable repeated text, or apply fewer pages at a time.
This limits appearance generation; it is not a limit on the input PDF or the whole process's memory.
Structure cleanup resolves named MCIDs and indirect content references. Form content
references are updated to the cleaned copy; unchanged sibling descriptions remain. The structure
tree's object references (/OBJR) to the annotations and form fields apply removes are dropped,
as Sanitize drops them, so a full rewrite of a tagged PDF can't keep a removed comment or field value.
A tree where that can't be done (an element written in place, or one too large to walk) makes
apply fail with RedactError::Tags rather than leave the annotation reachable.
Replacement descriptions (/Alt, /ActualText, /E) are removed from marked-content spans
changed by cleanup, from every structure element above them (found through both /P and /K),
and from elements that lose an annotation reference, which also lose /T. An element written in
place above one whose descriptions go makes apply fail with RedactError::Tags. Unchanged
spans retain their descriptions; unused property entries are retired during redaction
and hidden-content cleanup, including aliases in shared/inherited resources. Descriptions still
used by surviving pages or forms are kept.
A redacted page that inherits /Resources from the page tree gets them pinned (by reference;
a direct inherited dictionary is made one shared object first) before it is rewritten. The parent's
/Resources is then retired unless another page still inherits it, so it cannot keep a removed
XObject reachable. Surviving form XObjects lose the resource entries for removed XObjects their own
content doesn't draw, for the same reason. Pages that aren't redacted are left untouched.
Remove hidden information / Sanitize
let found = sanitize::scan(&doc); // [(Hidden, count)] per category
sanitize::remove_hidden(&mut doc, &[Hidden::Comments])?; // chosen categories
sanitize::sanitize(&mut doc)?; // all of them
Categories: metadata (/Info, XMP), file attachments, comments, form fields (flattened so the
values stay visible), hidden text (render modes 3 and 7, or wholly off the page), hidden layers
(content of off optional-content groups, then the groups), bookmarks, links/actions/JavaScript,
and /PieceInfo private data. XMP metadata, private data, associated files (/AF) and
additional actions (/AA) are removed from every object reachable from the catalog that owns
them (images, fonts, Forms, annotations and fields as well as the catalog and pages), never from
name maps such as resource dictionaries. Content is read strictly: a page whose content can't be
read in full is refused, not cleaned in part. Both make the next save a full rewrite.
patterns finds phone numbers, e-mail addresses, credit card numbers (Luhn-checked), US Social
Security numbers and dates for Search & Redact.
A phone label such as Tel. or Fax: also enables Italian international geographic numbers with a slash, such as Tel. +39 0426/123456 or Fax: +390426/123456 (#3205 follow-up).
The country code and national number are matched together, with the significant initial 0 and 6 to 11 national digits described in the AGCOM numbering plan published by ITU.
This narrow case accepts at most one space after +39 and one slash immediately after the area code; international slash numbers without a phone label, or with country codes other than +39, remain unsupported.
Patterns are heuristic: dates, protocol identifiers and longer number runs are retained, and users still need to review the marks before applying them.
codes holds the redaction code sets (U.S. FOIA and U.S. Privacy Act exemptions):
CodeSet::from_id("foia")?.overlay(&["(b)(6)"]) gives the overlay text for the picked codes.
Custom code sets aren't supported yet.
Not yet: pattern locales, struct-tree/alt-text cleanup, metadata and hidden-information sanitising (M8.4), re-encoding of DCT images (they are removed instead), and storing a cleared black-and-white image as CCITT Group 4 or JBIG2 again (it is Flate, larger; the optimizer makes it smaller).