fix(pdf): extract Adobe small capitals as letters

CHANGELOG.md

@@ -37,6 +37,8 @@ The release run heads these entries with the version and opens a fresh
 - A text page that holds a frame anchored to the page, such as a letter
   template, prints whole on one sheet, so the frame stays where the page puts
   it.
+- pdf text in Adobe small capitals and old-style figures (`Asmall`,
+  `oneoldstyle`) is found by a search and copies as letters and digits.
 
 ## v7.5.0 - 2026-10-06
 

src/odr/internal/pdf/pdf_page_extractor.cpp

@@ -64,6 +64,19 @@ struct ResolvedText {
   bool no_unicode{false};
 };
 
+/// Adobe's corporate-use small capitals and old-style figures (`Asmall` is
+/// U+F761, `zerooldstyle` U+F730) mirror ASCII and Latin-1 at U+F700. Their
+/// letters are what a search or a copy wants; the glyphs drawn stay the same.
+std::string fold_adobe_small_caps(const std::string &text) {
+  std::string result;
+  for (std::string_view rest = text; !rest.empty();) {
+    const char32_t c = util::string::next_utf8(rest);
+    util::string::append_c32(c >= 0xf721 && c <= 0xf7ff ? c - 0xf700 : c,
+                             result);
+  }
+  return result;
+}
+
 ResolvedText resolve_text(MarkedContentStack &marked, const Font *font,
                           const std::string &codes) {
   for (MarkedContent &entry : marked | std::views::reverse) {
@@ -78,7 +91,7 @@ ResolvedText resolve_text(MarkedContentStack &marked, const Font *font,
   if (font == nullptr) {
     return {codes, false}; // unknown font: historic raw-code passthrough
   }
-  std::string unicode = font->to_unicode(codes);
+  std::string unicode = fold_adobe_small_caps(font->to_unicode(codes));
   const bool no_unicode = unicode.empty() && !codes.empty();
   return {std::move(unicode), no_unicode};
 }

test/src/internal/pdf/pdf_page_extractor.cpp

@@ -636,6 +636,24 @@ Font type3_font(std::uint32_t first_char, const std::string &name, double width,
 
 } // namespace
 
+// Adobe's private-use small capitals and old-style figures fold back to their
+// letters and digits in the text, as pdf.js extracts them.
+TEST(PdfPageExtractor, adobe_small_caps_extract_as_letters) {
+  Font font = simple_font('A', {500, 500, 500, 500});
+  Encoding encoding(BaseEncoding::standard);
+  encoding.set_difference('A', "Asmall");
+  encoding.set_difference('B', "Bsmall");
+  encoding.set_difference('C', "oneoldstyle");
+  font.encoding = encoding;
+  Resources res;
+  res.font["F1"] = &font;
+
+  const std::vector<TextElement> text = run("BT /F1 10 Tf (ABCD) Tj ET", res);
+
+  ASSERT_EQ(text.size(), 1);
+  EXPECT_EQ(text[0].text, "ab1D");
+}
+
 // A Type3 glyph runs its char proc through the page machinery at the glyph
 // transform (`/FontMatrix` x size x `Tm` x CTM): the glyph paints as ordinary
 // path elements, and the shown run stays selectable but paints no visible text