qpdf_ruby 0.1.1 → 0.2.0

data/ext/qpdf_ruby/pdfua.hpp

@@ -0,0 +1,102 @@
+#pragma once
+#define POINTERHOLDER_TRANSITION 1
+
+#include <qpdf/QPDF.hh>
+
+#include <map>
+#include <optional>
+#include <string>
+
+// Fixes for the PDF/UA-1 gaps of Chromium's tagged PDF (print-to-PDF with generateTaggedPDF): it
+// writes no /Artifact sections, no LBody, no link descriptions, no XMP identification, and uses
+// PDF 2.0 structure types without a RoleMap. Each step only adds what is missing and leaves
+// everything else in place.
+namespace qpdf_ruby::pdfua {
+
+// Painting operators outside any marked content, per kind - what PDF/UA-1 7.1-3 rejects.
+struct UntaggedCounts {
+  long paths = 0;
+  long texts = 0;
+  long xobjects = 0;
+  long shadings = 0;
+  long inline_images = 0;
+  long total() const { return paths + texts + xobjects + shadings + inline_images; }
+};
+
+// Wraps every painting operation outside marked content in /Artifact BMC ... EMC: paths (with
+// their construction operators), text shows, XObjects, shadings, inline images. Content inside
+// marked content is never touched, and neither is a Form XObject that carries marked content of
+// its own (wrapping it would nest tagged content in an artifact, 7.1-2). Returns the counts.
+// Covers page content and the Form XObjects it draws - not annotation appearance streams (/AP),
+// tiling-pattern content or Type 3 glyph procedures, which Chromium does not emit today.
+UntaggedCounts mark_untagged_content_as_artifacts(QPDF& pdf);
+
+// The same walk without changing anything.
+UntaggedCounts count_untagged_content(QPDF& pdf);
+
+// Chromium tags CSS backgrounds and borders as content of the element they belong to - a Table's
+// cell backgrounds, an <article>'s card - and a grouping element (Document, Part, Art, Sect, Div,
+// BlockQuote, TOC, TOCI, Index, the table and list containers; after the RoleMap) must not hold
+// content of its own (PAC: "Content in an inadmissible location", "Invalid use of a TR"). Content
+// held directly by such an element that paints only shapes - no text, XObject or inline image, no
+// nested tagged content - becomes /Artifact BMC ... EMC, and leaves the element's kids and the
+// ParentTree. Content with text stays tagged. Returns how many pieces were turned into artifacts.
+long artifact_tagged_decorations(QPDF& pdf);
+
+// Gives each Link annotation without /Contents a description (7.18.1-2, 7.18.5-2): the text given
+// for its URI in `texts`, else its structure element's /Alt or /ActualText, else the URI itself
+// (a mailto: address without the scheme). An internal link uses the text given for "#<name>" of
+// its named destination (a name in the catalog's /Dests, a string in the /Names /Dests name tree),
+// else "Page N". Returns how many were described.
+long describe_links(QPDF& pdf, std::map<std::string, std::string> const& texts);
+
+// Moves an LI's content other than Lbl/LBody into a new LBody (7.2-20), structure elements as well
+// as marked-content references, with the ParentTree entries of the moved content updated.
+// Returns how many LBody elements were created.
+long wrap_list_bodies(QPDF& pdf);
+
+// Marked content and annotations whose ParentTree entry does not name the structure element that
+// holds them - MCIDs looked up in the stream they are numbered in (a reference's /Stm, else the
+// page), annotations through their /StructParent. Read-only; 0 for a consistent tree.
+long count_parent_tree_mismatches(QPDF& pdf);
+
+// Maps PDF 2.0 structure types Chromium emits (Aside, Strong, Em, ...) to PDF 1.7 standard types
+// in the RoleMap, unless the RoleMap already maps them (7.1-5). Returns how many were added.
+long map_nonstandard_roles(QPDF& pdf);
+
+// Chromium tags an HTML <figure> as a Figure without alternative text around the image's own
+// Figure (which carries the alt text) and its Caption (7.3-1). Such a grouping Figure - no /Alt or
+// /ActualText of its own, but a Figure below it - becomes a Div; nothing is lost, the caption stays
+// readable. Returns how many were retagged.
+long retag_grouping_figures(QPDF& pdf);
+
+// Figures that still have neither /Alt nor /ActualText - only a person can write those (7.3-1).
+long count_figures_without_alt(QPDF& pdf);
+
+// Identifies the file as PDF/UA-1 in its XMP metadata (7.1-8) and sets DisplayDocTitle and
+// Marked. An existing metadata stream keeps everything it has: pdfuaid:part and dc:title are each
+// added only if missing (read with namespace scoping, element or attribute form; comments and CDATA
+// do not count); an existing dc:title is never replaced. Without a stream, a new one is written.
+// The title is `title` if given (it also becomes the information dictionary's /Title), else that
+// /Title. Never throws for odd input: without a structure tree only the title and DisplayDocTitle
+// are set (Marked and pdfuaid would claim a tagged file), and an existing packet that is not
+// well-formed XML with an rdf:RDF element is left alone - apply() reports both as
+// unidentified_reason. Returns true if anything changed.
+bool add_pdfua_identification(QPDF& pdf, std::optional<std::string> const& title);
+
+struct Report {
+  long decorations = 0;
+  UntaggedCounts artifacts;
+  long links = 0;
+  long list_bodies = 0;
+  long roles = 0;
+  long figure_groups = 0;
+  long figures_without_alt = 0;
+  bool identified = false;
+  std::optional<std::string> unidentified_reason;  // why the file is not identified, if it is not
+};
+
+// All of the above, in the order that keeps each step's input intact.
+Report apply(QPDF& pdf, std::map<std::string, std::string> const& link_texts, std::optional<std::string> const& title);
+
+}  // namespace qpdf_ruby::pdfua