docling-parse 7.22.4__tar.gz → 7.23.1__tar.gz

{docling_parse-7.22.4 → docling_parse-7.23.1}/CHANGELOG.md

@@ -1,3 +1,26 @@
+## [v7.23.1](https://github.com/docling-project/docling-parse/releases/tag/v7.23.1) - 2026-10-26
+
+### Fix
+
+* Transcode non-UTF-8 widget names instead of losing the page ([#392](https://github.com/docling-project/docling-parse/issues/392)) ([`15c3f6a`](https://github.com/docling-project/docling-parse/commit/15c3f6a9609b00271cf7551a1ea216b2f91d1e01))
+* Fail a page whose content the truncated file no longer contains ([#415](https://github.com/docling-project/docling-parse/issues/415)) ([`7b7b273`](https://github.com/docling-project/docling-parse/commit/7b7b273339722541a6576631e95996a295df052c))
+* Keep the page when a damaged content stream breaks one operator ([#418](https://github.com/docling-project/docling-parse/issues/418)) ([`1357898`](https://github.com/docling-project/docling-parse/commit/1357898b1d749f975841c055a5dd194cc3f56b00))
+* Fall back from noncharacter ToUnicode mappings ([#395](https://github.com/docling-project/docling-parse/issues/395)) ([`3c91d2b`](https://github.com/docling-project/docling-parse/commit/3c91d2b372c9554e38aba6c14fda2b60ba0deee5))
+* Treat a space glyph mapped to a control character as a space ([#408](https://github.com/docling-project/docling-parse/issues/408)) ([`6d41900`](https://github.com/docling-project/docling-parse/commit/6d419002e6d2b4fab40a40f5e0048a9f1a74e4f3))
+* Ignore implausible vertical metrics in a font descriptor ([#412](https://github.com/docling-project/docling-parse/issues/412)) ([`2a9a6f8`](https://github.com/docling-project/docling-parse/commit/2a9a6f8b3a1f0d286da57e83911204bc3073f044))
+* **render:** Paint Type3 glyphs whose CharProc draws an image ([#422](https://github.com/docling-project/docling-parse/issues/422)) ([`dc69024`](https://github.com/docling-project/docling-parse/commit/dc690243ef79104c835facc0d776f15d5b31eec9))
+* Update pinned third-party libraries with known CVEs ([#423](https://github.com/docling-project/docling-parse/issues/423)) ([`bbfae48`](https://github.com/docling-project/docling-parse/commit/bbfae4839e31396ba06c1ebbd46362a4447269b1))
+
+### Documentation
+
+* Adding performance numbers ([#353](https://github.com/docling-project/docling-parse/issues/353)) ([`f8512ab`](https://github.com/docling-project/docling-parse/commit/f8512ab517428eea2c1e70d9b8db141bdbfa9ab8))
+
+## [v7.23.0](https://github.com/docling-project/docling-parse/releases/tag/v7.23.0) - 2026-10-22
+
+### Feature
+
+* Support for py3.15 ([#421](https://github.com/docling-project/docling-parse/issues/421)) ([`101de10`](https://github.com/docling-project/docling-parse/commit/101de108f935af64d046a819445a1d2157766371))
+
 ## [v7.22.4](https://github.com/docling-project/docling-parse/releases/tag/v7.22.4) - 2026-10-21
 
 ### Fix

{docling_parse-7.22.4 → docling_parse-7.23.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: docling-parse
-Version: 7.22.4
+Version: 7.23.1
 Summary: Simple package to extract text with coordinates from programmatic PDFs
 Author-email: Peter Staar <taa@zurich.ibm.com>, Christoph Auer <cau@zurich.ibm.com>, Michele Dolfi <dol@zurich.ibm.com>, Panos Vagenas <pva@zurich.ibm.com>, Maxim Lysak <mly@zurich.ibm.com>
 License-Expression: MIT
@@ -20,6 +20,7 @@ Classifier: Programming Language :: Python :: 3.11
 Classifier: Programming Language :: Python :: 3.12
 Classifier: Programming Language :: Python :: 3.13
 Classifier: Programming Language :: Python :: 3.14
+Classifier: Programming Language :: Python :: 3.15
 Requires-Python: >=3.10
 Description-Content-Type: text/markdown
 License-File: LICENSE

{docling_parse-7.22.4 → docling_parse-7.23.1}/cmake/extlib_freetype.cmake

@@ -19,7 +19,12 @@ else()
     file(MAKE_DIRECTORY ${EXTERNALS_PREFIX_PATH}/lib)
 
     set(FREETYPE_URL https://github.com/freetype/freetype.git)
-    set(FREETYPE_TAG VER-2-13-3)
+    # VER-2-13-3 is affected by CVE-2026-23865, an integer overflow in
+    # tt_var_load_item_variation_store (src/truetype/ttgxvar.c) leading to an
+    # out-of-bounds read when parsing HVAR/VVAR/MVAR tables of OpenType
+    # variable fonts. Fixed in VER-2-14-2; VER-2-14-3 is the latest patch
+    # release of that line.
+    set(FREETYPE_TAG VER-2-14-3)
     # GCC/MinGW -> libfreetype.a, MSVC -> freetype.lib (Windows arm64 uses MSVC).
     # DISABLE_FORCE_DEBUG_POSTFIX + Release avoid the "d" postfix, so the name is
     # stable per-toolchain.

{docling_parse-7.22.4 → docling_parse-7.23.1}/cmake/extlib_lcms2.cmake

@@ -33,7 +33,11 @@ else()
     file(MAKE_DIRECTORY ${EXTERNALS_PREFIX_PATH}/lib)
 
     set(LCMS2_URL https://github.com/mm2/Little-CMS.git)
-    set(LCMS2_TAG lcms2.17)
+    # lcms2.17 (and also 2.18) is affected by CVE-2026-41254 /
+    # GHSA-4xp6-rcgg-m9qq, an integer overflow in CubeSize (src/cmslut.c).
+    # Fixed in lcms2.19 (commits da6110b1, e0641b18); lcms2.19.1 is the
+    # latest release containing those fixes.
+    set(LCMS2_TAG lcms2.19.1)
 
     if(WIN32)
         # Windows (MSVC and MinGW): use vendored CMakeLists. Upstream has no

{docling_parse-7.22.4 → docling_parse-7.23.1}/cmake/extlib_openjpeg.cmake

@@ -18,7 +18,15 @@ else()
     file(MAKE_DIRECTORY ${EXTERNALS_PREFIX_PATH}/lib)
 
     set(OPENJPEG_URL https://github.com/uclouvain/openjpeg.git)
-    set(OPENJPEG_TAG v2.5.3)
+    # v2.5.3 (and also the latest tag v2.5.4) is affected by OSV-2025-219, a
+    # heap-buffer-overflow WRITE in opj_j2k_read_tile_header /
+    # opj_j2k_read_sod (OSS-Fuzz 403673832, severity HIGH). The fix is merged
+    # on master but no release tag contains it yet, so pin the fix commit
+    # directly: d33cbecc is the merge of uclouvain/openjpeg#1621 ("Fix heap
+    # buffer overflow in opj_j2k_read_sod") and contains the other OSV fixed
+    # commit 91d08b1 as a parent. Replace this with the v2.5.5+ tag once one
+    # is released.
+    set(OPENJPEG_TAG d33cbecc148d3affcdf403211fddc2cc5d442379)
     # GCC/MinGW -> libopenjp2.a, MSVC -> openjp2.lib (Windows arm64 uses MSVC).
     if(MSVC)
         set(OPENJPEG_IMPORTED_LIB ${EXTERNALS_PREFIX_PATH}/lib/openjp2.lib)

{docling_parse-7.22.4 → docling_parse-7.23.1}/docling_parse.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: docling-parse
-Version: 7.22.4
+Version: 7.23.1
 Summary: Simple package to extract text with coordinates from programmatic PDFs
 Author-email: Peter Staar <taa@zurich.ibm.com>, Christoph Auer <cau@zurich.ibm.com>, Michele Dolfi <dol@zurich.ibm.com>, Panos Vagenas <pva@zurich.ibm.com>, Maxim Lysak <mly@zurich.ibm.com>
 License-Expression: MIT
@@ -20,6 +20,7 @@ Classifier: Programming Language :: Python :: 3.11
 Classifier: Programming Language :: Python :: 3.12
 Classifier: Programming Language :: Python :: 3.13
 Classifier: Programming Language :: Python :: 3.14
+Classifier: Programming Language :: Python :: 3.15
 Requires-Python: >=3.10
 Description-Content-Type: text/markdown
 License-File: LICENSE

{docling_parse-7.22.4 → docling_parse-7.23.1}/docling_parse.egg-info/SOURCES.txt

@@ -615,6 +615,7 @@ src/parse/qpdf/annots.h
 src/parse/qpdf/logger.h
 src/parse/qpdf/object.h
 src/parse/qpdf/outline.h
+src/parse/qpdf/page_integrity.h
 src/parse/qpdf/qpdf_compat.h
 src/parse/qpdf/stream_decoder.h
 src/parse/qpdf/stream_filters.h
@@ -756,6 +757,9 @@ tests/test_unit_clipping.py
 tests/test_unit_cmap_encoding.py
 tests/test_unit_colorspace.py
 tests/test_unit_colorspaces.py
+tests/test_unit_control_char_word_space.py
+tests/test_unit_damaged_content_streams.py
+tests/test_unit_damaged_pages.py
 tests/test_unit_font_fallback.py
 tests/test_unit_font_name_resolution.py
 tests/test_unit_gid_glyph_names.py

{docling_parse-7.22.4 → docling_parse-7.23.1}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "docling-parse"
-version = "7.22.4"
+version = "7.23.1"
 description = "Simple package to extract text with coordinates from programmatic PDFs"
 license = "MIT"
 keywords = ["docling", "pdf", "parser"]
@@ -18,6 +18,7 @@ classifiers = [
     "Programming Language :: Python :: 3.12",
     "Programming Language :: Python :: 3.13",
     "Programming Language :: Python :: 3.14",
+    "Programming Language :: Python :: 3.15",
 ]
 readme = "README.md"
 authors = [
@@ -43,7 +44,7 @@ requires = [
     "setuptools>=77.0.3",
     "pybind11>=2.13.6",
     "cibuildwheel>=2.19.2,<3.0.0; python_version < '3.11'",
-    "cibuildwheel>=2.19.2,<4.0.0; python_version >= '3.11'",
+    "cibuildwheel>=4.3.0,<5.0.0; python_version >= '3.11'",
     "wheel>=0.43.0,<1.0.0",
     "delocate>=0.11.0,<1.0.0",
     "cmake>=3.27.0,<5.0.0; sys_platform != 'win32'",
@@ -56,7 +57,7 @@ build = [
     "setuptools>=77.0.3",
     "pybind11>=2.13.6",
     "cibuildwheel>=2.19.2,<3.0.0; python_version < '3.11'",
-    "cibuildwheel>=2.19.2,<4.0.0; python_version >= '3.11'",
+    "cibuildwheel>=4.3.0,<5.0.0; python_version >= '3.11'",
     "wheel>=0.43.0,<1.0.0",
     "delocate>=0.11.0,<1.0.0",
     "cmake>=3.27.0,<5.0.0; sys_platform != 'win32'",
@@ -76,7 +77,7 @@ perf = [
     "matplotlib>=3.10.8",
     "pdfplumber>=0.11.7",
     "pdfminer.six>=20240706",
-    "pymupdf>=1.26.4",
+    "pymupdf>=1.26.4,<2.0",  # 2.x hard-requires onnxruntime (no cp315 wheels yet)
     "pypdf>=5.1.0",
     "pypdfium2>=4.30.0",
     "huggingface-hub>=1.11.0",
@@ -162,7 +163,6 @@ pretty = true
 no_implicit_optional = true
 namespace_packages = true
 show_error_codes = true
-python_version = "3.10"
 
 [[tool.mypy.overrides]]
 module = [

{docling_parse-7.22.4 → docling_parse-7.23.1}/src/parse/pdf_decoders/document.h

@@ -14,6 +14,7 @@
 #include <qpdf/QPDFWriter.hh>
 
 #include <parse/qpdf/logger.h>
+#include <parse/qpdf/page_integrity.h>
 
 namespace pdflib
 {
@@ -78,6 +79,10 @@ namespace pdflib
 
     void ensure_annots_loaded();
 
+    // Throws missing_page_object_error when the page references an
+    // object the file does not contain (docling-parse#374).
+    void ensure_page_content_present(int page_number);
+
     void update_timings(pdf_timings& timings_, bool set_timer);
 
   private:
@@ -102,6 +107,11 @@ namespace pdflib
 
     int number_of_pages;
 
+    // qpdf had to repair the file while loading it (for example by
+    // reconstructing the cross-reference table). Set once at load time
+    // and read by the worker threads afterwards.
+    bool qpdf_reported_damage;
+
     nlohmann::json json_annots;
     bool annots_loaded;
 
@@ -120,6 +130,7 @@ namespace pdflib
     qpdf_pages({}),
 
     number_of_pages(-1),
+    qpdf_reported_damage(false),
 
     json_annots(nlohmann::json::value_t::null),
     annots_loaded(false),
@@ -139,6 +150,7 @@ namespace pdflib
     qpdf_pages({}),
 
     number_of_pages(-1),
+    qpdf_reported_damage(false),
 
     json_annots(nlohmann::json::value_t::null),
     annots_loaded(false),
@@ -292,6 +304,7 @@ namespace pdflib
 
 	qpdf_pages = qpdf_document.getAllPages();
 	number_of_pages = qpdf_pages.size();
+	qpdf_reported_damage = qpdf_document.anyWarnings();
 	
         LOG_S(INFO) << "#-pages: " << number_of_pages;	
       }
@@ -403,10 +416,38 @@ namespace pdflib
     return *src_afdh;
   }
 
+  void pdf_decoder<DOCUMENT>::ensure_page_content_present(int page_number)
+  {
+    std::string missing;
+    {
+      // Resolving the page's references may load objects into the shared
+      // qpdf document, which is only ever touched under this mutex.
+      std::lock_guard<std::mutex> lock(thread_safe_buffer_mutex);
+      missing = qpdf_page_integrity::find_missing_content(qpdf_pages.at(page_number));
+    }
+
+    if(missing.empty())
+      {
+        return;
+      }
+
+    std::string message = "page references " + missing + ", which is not in the file";
+    if(qpdf_reported_damage)
+      {
+        message += " (qpdf had to repair the document while loading it)";
+      }
+    message += ": the PDF is truncated or damaged";
+
+    LOG_S(ERROR) << message;
+    throw missing_page_object_error(message);
+  }
+
   pdf_decoder<DOCUMENT>::page_decoder_ptr
   pdf_decoder<DOCUMENT>::make_thread_safe_page_decoder(int page_number,
                                                        bool keep_qpdf_warnings)
   {
+    ensure_page_content_present(page_number);
+
     std::pair<int, std::shared_ptr<std::string> > result =
       get_thread_safe_page_buffer(page_number, keep_qpdf_warnings);
 
@@ -510,6 +551,7 @@ namespace pdflib
       else
         {
 	  LOG_S(INFO) << "decoding page thread-unsafe";
+          ensure_page_content_present(page_number);
           QPDFObjectHandle qpdf_page = qpdf_pages.at(page_number);
 
           page_decoder = std::make_shared<pdf_decoder<PAGE>>(qpdf_page, page_number);

{docling_parse-7.22.4 → docling_parse-7.23.1}/src/parse/pdf_decoders/page.h

@@ -1557,11 +1557,11 @@ namespace pdflib
     std::string text = "";
     QPDFObjectHandle value = field.getValue();
     if(value.isString()) { text = value.getUTF8Value(); }
-    else if(value.isName()) { text = value.getName(); }
+    else if(value.isName()) { text = name_to_utf8(value); }
 
     std::string appearance_state = "";
     QPDFObjectHandle appearance = annot.getKey("/AS");
-    if(appearance.isName()) { appearance_state = appearance.getName(); }
+    if(appearance.isName()) { appearance_state = name_to_utf8(appearance); }
 
     page_item<PAGE_WIDGET> widget;
     {

{docling_parse-7.22.4 → docling_parse-7.23.1}/src/parse/pdf_decoders/stream.h

@@ -93,6 +93,11 @@ namespace pdflib
     void do_form(const std::string& xobj_name,
 		 const xobject_subtype_name& xobj_subtype);
 
+    // Runs the Type3 CharProcs that the last show-string queued, see
+    // pdf_state<TEXT>::type3_procedure.
+    void paint_type3_glyphs();
+    void do_type3_glyph(const pdf_state<TEXT>::type3_procedure& glyph);
+
     void do_postscript(const std::string& xobj_name,
        const xobject_subtype_name& xobj_subtype);
 
@@ -405,7 +410,26 @@ namespace pdflib
               }
             LOG_S(INFO) << " --> " << std::setw(12) << inst.key << " | " << inst.val;
 
-            execute_operator(inst, parameters);
+            // An operand of the wrong type, a resource that fails to load, a
+            // malformed inline image: in a damaged stream any of these can
+            // surface as an exception from one operator. That operator is
+            // lost either way; the ones after it need not be. Log it and go
+            // on, as mupdf does ("ignoring error during interpretation").
+            // Resource exhaustion (std::bad_alloc) is not caught here.
+            try
+              {
+                execute_operator(inst, parameters);
+              }
+            catch(const std::logic_error& e)
+              {
+                LOG_S(ERROR) << "operator " << inst.val << " failed (" << e.what()
+                             << ") => skipping the operator";
+              }
+            catch(const std::runtime_error& e)
+              {
+                LOG_S(ERROR) << "operator " << inst.val << " failed (" << e.what()
+                             << ") => skipping the operator";
+              }
 
             parameters.clear();
           }
@@ -428,11 +452,8 @@ namespace pdflib
   {
     if(stack.size()==0)
       {
-        std::stringstream message;
-        message << "stack-size is zero in " << __FILE__ << ":" << __LINE__;
-
-        LOG_S(ERROR) << message.str();
-        throw std::logic_error(message.str());
+        LOG_S(ERROR) << "graphics state stack is empty: starting from a fresh state";
+        ensure_stack();
       }
 
     pdf_state<GLOBAL>& state = stack.back();
@@ -508,14 +529,16 @@ namespace pdflib
 
   void pdf_decoder<STREAM>::Q()
   {
-    if(stack.size()>0)
+    // The bottom entry is the state the stream started in, not something a
+    // `q` pushed: a `Q` without its `q` (a truncated stream, a damaged one)
+    // must not pop it, or every operator after it has no state to run in.
+    if(stack.size()>1)
       {
         stack.pop_back();
       }
     else
       {
-        LOG_S(ERROR) << "invoking 'Q' on empty stack!";
-        //throw std::logic_error(__FILE__);
+        LOG_S(ERROR) << "invoking 'Q' without a matching 'q': keeping the base state";
       }
   }
 
@@ -1777,6 +1800,155 @@ namespace pdflib
 
   }
 
+  void pdf_decoder<STREAM>::paint_type3_glyphs()
+  {
+    auto& pending = current_text_state().pending_type3_procedures;
+    if(pending.empty())
+      {
+        return;
+      }
+
+    // do_type3_glyph() pushes a copy of the current state, so take the queue
+    // out of it first.
+    std::vector<pdf_state<TEXT>::type3_procedure> glyphs;
+    glyphs.swap(pending);
+
+    for(const auto& glyph : glyphs)
+      {
+        do_type3_glyph(glyph);
+      }
+  }
+
+  void pdf_decoder<STREAM>::do_type3_glyph(const pdf_state<TEXT>::type3_procedure& glyph)
+  {
+    LOG_S(INFO) << __FUNCTION__;
+
+    // A CharProc can show text in a Type3 font whose CharProc leads back to
+    // it, directly or through a form XObject.
+    constexpr int max_depth = 8;
+    static thread_local int depth = 0;
+    if(depth >= max_depth)
+      {
+        LOG_S(WARNING) << "Type3 glyph procedures nest deeper than " << max_depth
+                       << " levels, the glyph is not painted";
+        return;
+      }
+
+    std::vector<qpdf_stream_instruction> insts;
+    try
+      {
+        qpdf_stream_decoder decoder(insts);
+        QPDFObjectHandle charproc = glyph.charproc;
+        decoder.decode(charproc);
+      }
+    catch(const std::exception& e)
+      {
+        LOG_S(WARNING) << "could not decode a Type3 glyph procedure: " << e.what();
+        return;
+      }
+
+    // 9.6.5: names in the CharProc resolve against the font's /Resources,
+    // and against the page's when the font has none.
+    auto page_fonts_       = std::make_shared<pdf_resource<PAGE_FONTS>>(page_fonts);
+    auto page_grphs_       = std::make_shared<pdf_resource<PAGE_GRPHS>>(page_grphs);
+    auto page_colorspaces_ = std::make_shared<pdf_resource<PAGE_COLORSPACES>>(page_colorspaces);
+    auto page_shadings_    = std::make_shared<pdf_resource<PAGE_SHADINGS>>(page_shadings);
+    auto page_patterns_    = std::make_shared<pdf_resource<PAGE_PATTERNS>>(page_patterns);
+    auto page_xobjects_    = std::make_shared<pdf_resource<PAGE_XOBJECTS>>(page_xobjects);
+
+    QPDFObjectHandle resources = glyph.resources;
+    if(resources.isDictionary())
+      {
+        if(resources.hasKey("/Font"))
+          {
+            QPDFObjectHandle fonts = resources.getKey("/Font");
+            page_fonts_->set(fonts, timings);
+          }
+        if(resources.hasKey("/ExtGState"))
+          {
+            QPDFObjectHandle grphs = resources.getKey("/ExtGState");
+            page_grphs_->set(grphs, timings);
+          }
+        if(resources.hasKey("/ColorSpace"))
+          {
+            QPDFObjectHandle colorspaces = resources.getKey("/ColorSpace");
+            page_colorspaces_->set(colorspaces);
+          }
+        if(resources.hasKey("/Shading"))
+          {
+            QPDFObjectHandle shadings = resources.getKey("/Shading");
+            page_shadings_->set(shadings);
+          }
+        if(resources.hasKey("/Pattern"))
+          {
+            QPDFObjectHandle patterns = resources.getKey("/Pattern");
+            page_patterns_->set(patterns);
+          }
+        if(resources.hasKey("/XObject"))
+          {
+            QPDFObjectHandle xobjects = resources.getKey("/XObject");
+            page_xobjects_->set(xobjects, timings);
+          }
+      }
+
+    // The glyph is not page content: its text cell already exists, and its
+    // paths and images are ink, not shapes or pictures. Only the render
+    // instructions are kept.
+    page_item<PAGE_CELLS>  glyph_cells;
+    page_item<PAGE_SHAPES> glyph_shapes;
+    page_item<PAGE_IMAGES> glyph_images;
+
+    this->q();
+    current_global_state().cm(glyph.matrix);
+    current_bitmap_state().paints_type3_glyph = true;
+    {
+      pdf_decoder<STREAM> new_stream(config,
+
+                                     page_dimension,
+                                     glyph_cells,
+                                     glyph_shapes,
+                                     glyph_images,
+
+                                     page_fonts_,
+                                     page_grphs_,
+                                     page_colorspaces_,
+                                     page_shadings_,
+                                     page_patterns_,
+                                     page_xobjects_,
+
+                                     instructions,
+
+                                     timings);
+
+      bool updated_stack = new_stream.update_stack(stack, stack_count);
+
+      std::vector<qpdf_stream_instruction> parameters;
+
+      depth += 1;
+      try
+        {
+          new_stream.interprete(insts, parameters);
+        }
+      catch(...)
+        {
+          depth -= 1;
+          throw;
+        }
+      depth -= 1;
+
+      if(updated_stack)
+        {
+          new_stream.Q();
+        }
+
+      for(auto item:new_stream.get_unknown_operators())
+        {
+          unknown_operators.insert(item);
+        }
+    }
+    this->Q();
+  }
+
   void pdf_decoder<STREAM>::do_postscript(const std::string& xobj_name,
                                           const xobject_subtype_name& xobj_subtype)
   {
@@ -2356,6 +2528,7 @@ namespace pdflib
         {
           LOG_S(INFO) << "executing " << to_string(name);
           current_text_state().Tj(parameters, stack_count);
+          paint_type3_glyphs();
         }
         break;
 
@@ -2363,6 +2536,7 @@ namespace pdflib
         {
           LOG_S(INFO) << "executing " << to_string(name);
           current_text_state().TJ(parameters, stack_count);
+          paint_type3_glyphs();
         }
         break;
 
@@ -2376,6 +2550,7 @@ namespace pdflib
 
           std::vector<qpdf_stream_instruction> Tj_params = {parameters[0]};
           current_text_state().Tj(Tj_params, stack_count);
+          paint_type3_glyphs();
         }
         break;
 
@@ -2395,6 +2570,7 @@ namespace pdflib
 
           std::vector<qpdf_stream_instruction> Tj_params = {parameters[2]};
           current_text_state().Tj(Tj_params, stack_count);
+          paint_type3_glyphs();
         }
         break;