fix(pdf): place text by its own font, decode jpeg 2000, and clip to the crop box

CHANGELOG.md

@@ -32,6 +32,14 @@ The release run heads these entries with the version and opens a fresh
   white space wherever it stands, not the start of a token.
 - Text that decodes to half a surrogate pair costs that character a replacement
   mark instead of the whole document.
+- A pdf's text sits where the file puts it: a line was placed against the
+  browser's default font rather than its own, which dropped small text by
+  several points.
+- A pdf's cmyk colours are converted as Adobe converts them, so a process cyan
+  reads as one instead of as pure `#00ffff`.
+- A pdf page is the size of its crop box and shows what is on it and no more,
+  as a viewer shows it.
+- A pdf's JPEG 2000 images render. New dependency: `openjpeg`.
 
 ## v6.5.0 - 2026-08-10
 

CMakeLists.txt

@@ -57,6 +57,7 @@ find_package(pugixml REQUIRED)
 find_package(miniz REQUIRED)
 find_package(cryptopp REQUIRED)
 find_package(nlohmann_json REQUIRED)
+find_package(OpenJPEG REQUIRED)
 find_package(uchardet REQUIRED)
 find_package(utf8cpp REQUIRED)
 
@@ -216,6 +217,7 @@ set(ODR_SOURCE_FILES
         "src/odr/internal/pdf/pdf_filter.cpp"
         "src/odr/internal/pdf/pdf_function.cpp"
         "src/odr/internal/pdf/pdf_image.cpp"
+        "src/odr/internal/pdf/pdf_jpx.cpp"
         "src/odr/internal/pdf/pdf_graphics_operator_parser.cpp"
         "src/odr/internal/pdf/pdf_graphics_state.cpp"
         "src/odr/internal/pdf/pdf_object.cpp"
@@ -276,6 +278,7 @@ target_link_libraries(odr
         miniz::miniz
         cryptopp::cryptopp
         nlohmann_json::nlohmann_json
+        openjp2
         uchardet::uchardet
         utf8::cpp
 )

conan.lock

@@ -8,6 +8,7 @@
         "uchardet/0.0.8#6ab25e452021fcdb560f4e37f4a27bc1%1759735438.978",
         "pybind11/2.13.6#42746850cd4c68d1b1ea42de456c2182%1755673714.548",
         "pugixml/1.15#979e88f4fafbfe3585d2c0510a071cc7%1739435725.483",
+        "openjpeg/2.5.4#372fbc2b4348d45ab0c0a62a8475dc2f%1760446899.685",
         "nlohmann_json/3.12.0#2d634ab0ec8d9f56353e5ccef6d6612c%1744735883.94",
         "miniz/3.0.2#bfbce07c6654293cce27ee24129d2df7%1743673472.805",
         "gtest/1.14.0#f8f0757a574a8dd747d16af62d6eb1b7%1743410807.169",

conanfile.py

@@ -52,6 +52,7 @@ def requirements(self):
         self.requires("cryptopp/8.9.0")
         self.requires("miniz/3.0.2")
         self.requires("nlohmann_json/3.12.0")
+        self.requires("openjpeg/2.5.4")
         self.requires("uchardet/0.0.8")
         self.requires("utfcpp/4.0.9")
         if self.options.get_safe("with_http_server", False):

src/odr/internal/html/pdf_file.cpp

@@ -14,6 +14,7 @@
 #include <odr/internal/html/common.hpp>
 #include <odr/internal/html/html_service.hpp>
 #include <odr/internal/html/html_writer.hpp>
+#include <odr/internal/pdf/pdf_color.hpp>
 #include <odr/internal/pdf/pdf_document.hpp>
 #include <odr/internal/pdf/pdf_document_element.hpp>
 #include <odr/internal/pdf/pdf_document_parser.hpp>
@@ -316,14 +317,11 @@ std::string device_color_to_css(const pdf::GraphicsState::Color &color) {
     b = to255(color.rgb[2]);
     break;
   case pdf::ColorSpace::device_cmyk: {
-    // Naive CMYK -> RGB (no ICC).
-    const double c = color.cmyk[0];
-    const double m = color.cmyk[1];
-    const double y = color.cmyk[2];
-    const double k = color.cmyk[3];
-    r = to255((1 - c) * (1 - k));
-    g = to255((1 - m) * (1 - k));
-    b = to255((1 - y) * (1 - k));
+    const std::array<double, 3> rgb = pdf::cmyk_to_rgb(
+        color.cmyk[0], color.cmyk[1], color.cmyk[2], color.cmyk[3]);
+    r = to255(rgb[0]);
+    g = to255(rgb[1]);
+    b = to255(rgb[2]);
     break;
   }
   case pdf::ColorSpace::unknown:
@@ -2196,7 +2194,7 @@ class HtmlServiceImpl final : public HtmlService {
         .m = m,
         .invisible = invisible,
         .is_matrix = is_matrix,
-        .asc = ascent_em(text),
+        .asc = ascent_em(text.font),
         .scale = is_matrix ? 1.0 : m.a,
         .ox = m.e,
         .baseline = m.f,
@@ -2255,11 +2253,17 @@ class HtmlServiceImpl final : public HtmlService {
 
   template <typename AddClass>
   static PageBox begin_page(const pdf::Page &page, AddClass &&add_class) {
-    const pdf::Array &page_box = page.media_box.as_array();
-    const double box_x0 = page_box[0].as_real();
-    const double box_y0 = page_box[1].as_real();
-    const double width = page_box[2].as_real() - box_x0;
-    const double height = page_box[3].as_real() - box_y0;
+    // The crop box is what a viewer shows (14.11.2); it falls back to the media
+    // box. Its corners may be given in either order (7.9.5).
+    const pdf::Array &page_box = page.crop_box.as_array();
+    const double box_x0 =
+        std::min(page_box[0].as_real(), page_box[2].as_real());
+    const double box_y0 =
+        std::min(page_box[1].as_real(), page_box[3].as_real());
+    const double width =
+        std::max(page_box[0].as_real(), page_box[2].as_real()) - box_x0;
+    const double height =
+        std::max(page_box[1].as_real(), page_box[3].as_real()) - box_y0;
 
     std::string classes = "p";
     {
@@ -2358,11 +2362,15 @@ class HtmlServiceImpl final : public HtmlService {
     // side margin is part of that width, so a phone screen keeps a gutter.
     out.out() << ".d{display:flex;flex-direction:column;align-items:center;"
                  "gap:16px;padding:16px 0;width:max-content;min-width:100%}";
+    // `overflow:hidden` clips to the crop box, as a viewer does: content may
+    // sit outside it (a bleed, or an InDesign spread's other page).
     out.out() << ".p{position:relative;margin:0 16px;background:#fff;"
-                 "box-shadow:0 1px 4px rgba(0,0,0,.5)}";
+                 "overflow:hidden;box-shadow:0 1px 4px rgba(0,0,0,.5)}";
     // `.t`: shared base for all absolutely-positioned line blocks.
+    // `font-size:0` collapses its strut, which outranks the run it holds and
+    // would take the line box's baseline.
     out.out() << ".t{position:absolute;left:0;top:0;transform-origin:0 0;"
-                 "white-space:pre;line-height:1;font-kerning:none;"
+                 "white-space:pre;line-height:1;font-size:0;font-kerning:none;"
                  "font-variant-ligatures:none}";
     write_mode_css();
     // SVG overlay covering the page box (visual graphics layer).
@@ -2472,11 +2480,15 @@ class HtmlServiceImpl final : public HtmlService {
     }
     const std::string url = file_to_url(reencoded, "font/ttf");
     const std::string n = std::to_string(index + 1);
-    font_faces += "@font-face{font-family:'odr-f";
-    font_faces += n;
-    font_faces += "';src:url(";
-    font_faces += url;
-    font_faces += ");}";
+    // The overrides sum to one em, so `line-height:1` puts the baseline at
+    // exactly the `ascent_em` a run's `top` is derived from.
+    const double ascent = ascent_em(&font);
+    std::ostringstream face;
+    face << "@font-face{font-family:'odr-f" << n << "';src:url(" << url
+         << ");ascent-override:" << round2(ascent * 100.0)
+         << "%;descent-override:" << round2((1.0 - ascent) * 100.0)
+         << "%;line-gap-override:0%}";
+    font_faces += std::move(face).str();
     const auto rule = [&](const char *cls, const char *color) {
       font_styles += '.';
       font_styles += cls;
@@ -2495,19 +2507,20 @@ class HtmlServiceImpl final : public HtmlService {
     }
   }
 
-  static double ascent_em(const pdf::TextElement &text) {
+  /// Baseline offset below a line block's `top`, in em. Capped at one em so
+  /// the `@font-face` descent can make the two sum to it.
+  static double ascent_em(const pdf::Font *font) {
     double em = 0.8;
-    if (text.font != nullptr && text.font->descriptor_ascent) {
-      em = *text.font->descriptor_ascent;
-    } else if (text.font != nullptr && text.font->embedded_font != nullptr) {
-      const std::uint16_t units = text.font->embedded_font->units_per_em();
+    if (font != nullptr && font->descriptor_ascent) {
+      em = *font->descriptor_ascent;
+    } else if (font != nullptr && font->embedded_font != nullptr) {
+      const std::uint16_t units = font->embedded_font->units_per_em();
       if (units != 0) {
-        em = static_cast<double>(
-                 text.font->embedded_font->bounding_box().y_max) /
+        em = static_cast<double>(font->embedded_font->bounding_box().y_max) /
              units;
       }
     }
-    return std::clamp(em, 0.5, 1.2);
+    return std::clamp(em, 0.5, 1.0);
   }
 
   static std::string glyph_run_str(const pdf::Font &font,

src/odr/internal/pdf/AGENTS.md

@@ -146,10 +146,12 @@ Things the code won't shout at you:
   is unlocked with the empty password first so `/Info` decrypts). It is
   all-or-nothing: a malformed structure leaves `document_type` at `unknown` rather
   than half-filling the fields. XMP is not parsed — the strings are `/Info`-only.
-- **Image codecs are deliberately not decoded** in the filter framework
+- **Image codecs are not decoded** in the filter framework
   (DCTDecode/JPXDecode/CCITTFaxDecode/JBIG2Decode): `decode()` stops and hands
   back the still-encoded payload for the image path; `read_decoded_stream` treats
-  them as an error. `Crypt` passes through only as `Identity`.
+  them as an error. A JPEG then passes through to the browser and a JPEG 2000
+  goes to `pdf_jpx` (openjpeg) to be re-encoded as PNG like any other raster;
+  CCITT and JBIG2 remain undecodable. `Crypt` passes through only as `Identity`.
 - **Inherited page attributes** (`Resources`/`MediaBox`/`CropBox`/`Rotate`, Table
   30) are resolved by threading an accumulator down the `Pages` recursion — *not*
   by a `Parent` walk. Lenience (all with a `Logger` warning): `CropBox` ←

src/odr/internal/pdf/pdf_color.cpp

@@ -11,12 +11,6 @@ namespace {
 
 double clamp01(const double v) { return std::clamp(v, 0.0, 1.0); }
 
-/// Naive DeviceCMYK -> RGB (no ICC), matching the HTML emitter's conversion.
-std::array<double, 3> cmyk_to_rgb(const double c, const double m,
-                                  const double y, const double k) {
-  return {(1 - c) * (1 - k), (1 - m) * (1 - k), (1 - y) * (1 - k)};
-}
-
 /// sRGB gamma encode of a linear component (IEC 61966-2-1).
 double linear_to_srgb(const double c) {
   const double v = clamp01(c);
@@ -183,6 +177,38 @@ std::vector<double> ColorSpaceDef::initial_components() const {
 
 namespace odr::internal {
 
+std::array<double, 3> pdf::cmyk_to_rgb(const double c, const double m,
+                                       const double y, const double k) {
+  const double r =
+      255 +
+      c * (-4.387332384609988 * c + 54.48615194189176 * m +
+           18.82290502165302 * y + 212.25662451639585 * k - 285.2331026137004) +
+      m * (1.7149763477362134 * m - 5.6096736904047315 * y -
+           17.873870861415444 * k - 5.497006427196366) +
+      y * (-2.5217340131683033 * y - 21.248923337353073 * k +
+           17.5119270841813) +
+      k * (-21.86122147463605 * k - 189.48180835922747);
+  const double g =
+      255 +
+      c * (8.841041422036149 * c + 60.118027045597366 * m +
+           6.871425592049007 * y + 31.159100130055922 * k - 79.2970844816548) +
+      m * (-15.310361306967817 * m + 17.575251261109482 * y +
+           131.35250912493976 * k - 190.9453302588951) +
+      y * (4.444339102852739 * y + 9.8632861493405 * k - 24.86741582555878) +
+      k * (-20.737325471181034 * k - 187.80453709719578);
+  const double b = 255 +
+                   c * (0.8842522430003296 * c + 8.078677503112928 * m +
+                        30.89978309703729 * y - 0.23883238689178934 * k -
+                        14.183576799673286) +
+                   m * (10.49593273432072 * m + 63.02378494754052 * y +
+                        50.606957656360734 * k - 112.23884253719248) +
+                   y * (0.03296041114873217 * y + 115.60384449646641 * k -
+                        193.58209356861505) +
+                   k * (-22.33816807309886 * k - 180.12613974708367);
+  return {pdf::clamp01(r / 255.0), pdf::clamp01(g / 255.0),
+          pdf::clamp01(b / 255.0)};
+}
+
 std::shared_ptr<pdf::ColorSpaceDef>
 pdf::parse_color_space(const Object &object, const ColorSpaceContext &context) {
   const Object resolved = context.resolve(object);

src/odr/internal/pdf/pdf_color.hpp

@@ -73,6 +73,10 @@ struct ColorSpaceContext {
   std::function<std::shared_ptr<ColorSpaceDef>(const std::string &)> named;
 };
 
+/// DeviceCMYK -> sRGB without an ICC engine: pdf.js's polynomial fit of Adobe's
+/// transform. The naive `(1-c)(1-k)` reads pure cyan as `#00ffff`.
+std::array<double, 3> cmyk_to_rgb(double c, double m, double y, double k);
+
 /// Build a colour space from its PDF object — a name (`/DeviceRGB`, …) or an
 /// array (`[/ICCBased 5 0 R]`, `[/Separation …]`, …). Returns `nullptr` for an
 /// unsupported or malformed definition.

src/odr/internal/pdf/pdf_document_element.hpp

@@ -161,8 +161,8 @@ struct XObject final : Element {
   // --- image (`/Subtype /Image`) ---
   /// Browser-ready bytes: a `DCTDecode` JPEG passed through, or a raster
   /// re-encoded as PNG (with any `/SMask`/`/Mask` composited into RGBA). Empty
-  /// for an undecodable codec (JPX/CCITT/JBIG2) and for a stencil, so `Do`
-  /// skips it.
+  /// for an undecodable codec (CCITT/JBIG2) and for a stencil, so `Do` skips
+  /// it.
   std::string image_data;
   std::string image_mime;
 

src/odr/internal/pdf/pdf_document_parser.cpp

@@ -749,7 +749,7 @@ std::vector<std::uint8_t> resolve_mask_alpha(DocumentParser &parser,
   DecodeResult result =
       decode(filter, decode_parms, parser.read_object_stream(object));
   if (result.stopped_at_filter.has_value()) {
-    return {}; // an image codec we cannot decode (CCITT/JBIG2/JPX)
+    return {}; // an image codec we cannot decode (CCITT/JBIG2)
   }
   return decode_mask_alpha(
       result.data, image_int(parser, dictionary, "Width", 0),
@@ -841,10 +841,15 @@ void parse_image_data(DocumentParser &parser, const Dictionary &dictionary,
     }
   }
 
+  // A JPX codestream may carry its own opacity; `/SMaskInData` says whether it
+  // counts (8.9.5.4).
+  const std::int32_t smask_in_data =
+      image_int(parser, dictionary, "SMaskInData", 0);
+
   if (std::optional<EncodedImage> encoded =
           encode_image(parser.read_object_stream(object), filter, decode_parms,
                        width, height, bits_per_component, color_space.get(),
-                       decode_array, alpha, color_key)) {
+                       decode_array, alpha, color_key, smask_in_data)) {
     x_object.image_data = std::move(encoded->data);
     x_object.image_mime = std::move(encoded->mime);
   }