docling-parse 7.22.0__tar.gz → 7.22.1__tar.gz

{docling_parse-7.22.0 → docling_parse-7.22.1}/CHANGELOG.md

@@ -1,3 +1,10 @@
+## [v7.22.1](https://github.com/docling-project/docling-parse/releases/tag/v7.22.1) - 2026-09-28
+
+### Fix
+
+* Updated the page_item_sanitators/cells.h and its associated ground-truth ([#361](https://github.com/docling-project/docling-parse/issues/361)) ([`96960c3`](https://github.com/docling-project/docling-parse/commit/96960c3f13af88dd89037fd6fb681fdb5281a497))
+* **sanitizer:** Fold curly double quotes to `"`, not `'` ([#349](https://github.com/docling-project/docling-parse/issues/349)) ([`ea4a14d`](https://github.com/docling-project/docling-parse/commit/ea4a14d3ed1bab7fb3521b3e8f4db38f83a32eaf))
+
 ## [v7.22.0](https://github.com/docling-project/docling-parse/releases/tag/v7.22.0) - 2026-09-26
 
 ### Feature

{docling_parse-7.22.0 → docling_parse-7.22.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: docling-parse
-Version: 7.22.0
+Version: 7.22.1
 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

{docling_parse-7.22.0 → docling_parse-7.22.1}/docling_parse.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: docling-parse
-Version: 7.22.0
+Version: 7.22.1
 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

{docling_parse-7.22.0 → docling_parse-7.22.1}/docling_parse.egg-info/SOURCES.txt

@@ -767,6 +767,7 @@ tests/test_unit_load_failure.py
 tests/test_unit_nul_mapping.py
 tests/test_unit_page_count.py
 tests/test_unit_patterns.py
+tests/test_unit_quote_sanitization.py
 tests/test_unit_rotated_cell_contraction.py
 tests/test_unit_shading.py
 tests/test_unit_shadings.py

{docling_parse-7.22.0 → docling_parse-7.22.1}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "docling-parse"
-version = "7.22.0"
+version = "7.22.1"
 description = "Simple package to extract text with coordinates from programmatic PDFs"
 license = "MIT"
 keywords = ["docling", "pdf", "parser"]
@@ -46,7 +46,8 @@ requires = [
     "cibuildwheel>=2.19.2,<4.0.0; python_version >= '3.11'",
     "wheel>=0.43.0,<1.0.0",
     "delocate>=0.11.0,<1.0.0",
-    "cmake>=3.27.0,<4.0.0"
+    "cmake>=3.27.0,<5.0.0; sys_platform != 'win32'",
+    "cmake>=4.2.0,<5.0.0; sys_platform == 'win32'"
 ]
 build-backend = "setuptools.build_meta"
 
@@ -58,7 +59,8 @@ build = [
     "cibuildwheel>=2.19.2,<4.0.0; python_version >= '3.11'",
     "wheel>=0.43.0,<1.0.0",
     "delocate>=0.11.0,<1.0.0",
-    "cmake>=3.27.0,<4.0.0"
+    "cmake>=3.27.0,<5.0.0; sys_platform != 'win32'",
+    "cmake>=4.2.0,<5.0.0; sys_platform == 'win32'"
 ]
 dev = [
     "pytest>=9.0.3,<10.0.0",
@@ -155,6 +157,7 @@ max-complexity = 30
 combine-as-imports = true
 
 [tool.mypy]
+exclude = "^tests/data/"
 pretty = true
 no_implicit_optional = true
 namespace_packages = true

{docling_parse-7.22.0 → docling_parse-7.22.1}/src/parse/page_item_sanitators/cells.h

@@ -374,9 +374,9 @@ namespace pdflib
      * two-cluster k-means separates ordinary character gaps from dilated word
      * gaps. Sparse or insufficiently separated samples use a conservative
      * default threshold of 0.25; valid inferred thresholds are clamped to at
-     * least 0.18. Runs containing explicit spaces do not use this inferred
-     * threshold: those source-level boundaries are authoritative, with only
-     * jumps larger than one character advance allowed to add a boundary.
+     * least 0.18. Contraction caps the resulting threshold at 0.35 so a much
+     * larger gap elsewhere in a short line cannot hide a word boundary.
+     * Runs containing explicit spaces use the same cap for unmarked gaps.
      *
      * @param run One line of cells in source order.
      * @return Dimensionless normalized gap above which a new word begins.
@@ -408,9 +408,8 @@ namespace pdflib
      *
      * Builds line runs and treats explicit space cells as authoritative
      * boundaries. Runs without them infer a line-local cursor-placement gap
-     * threshold. Runs with them tolerate ordinary inner-word placement
-     * variation, while still splitting absolute-positioned fields separated
-     * by more than one neighboring character advance. Words are then merged
+     * threshold. Both kinds of runs split unmarked gaps above 0.35 neighboring
+     * character advances. Words are then merged
      * with normalized spaces into one line. Both levels retain maximum
      * enclosing rotation-aligned bounding boxes.
      * Visible glyphs without a recoverable Unicode mapping retain their U+FFFD
@@ -464,7 +463,7 @@ namespace pdflib
   inline bool page_item_sanitator<PAGE_CELLS>::is_semantic_space(
       const page_item<PAGE_CELL>& cell)
   {
-    return cell.is_pdf_word_space or
+    return (cell.is_pdf_word_space and cell.text_is_suppressed_glyph) or
            (not cell.text_is_suppressed_glyph and
             utils::string::is_space(cell.text));
   }
@@ -892,19 +891,18 @@ namespace pdflib
     contraction_result result;
     for(auto& run : collect_line_runs(cells))
       {
-        // Explicit PDF spaces are authoritative normal boundaries. On those
-        // lines only a jump larger than a complete neighboring advance may
-        // add another boundary; this still separates absolutely positioned
-        // table fields without splitting a word merely because its glyph ink
-        // or placement has an irregular space-sized gap.
-        const double word_gap = run.has_semantic_spaces
-          ? 1.0 : infer_word_gap(run);
+        // Explicit PDF spaces remain authoritative boundaries. Cap unmarked
+        // gaps as well: a large section-number gap can otherwise raise the
+        // inferred threshold past every word gap in a short heading.
+        const double word_gap = std::min(
+          0.35, run.has_semantic_spaces ? 1.0 : infer_word_gap(run));
         std::vector<page_item<PAGE_CELL>> words;
         page_item<PAGE_CELL>* previous_visible = nullptr;
         bool explicit_boundary = false;
 
-        for(auto& cell : run.cells)
+        for(std::size_t index = 0; index < run.cells.size(); ++index)
           {
+            auto& cell = run.cells[index];
             if(is_semantic_space(cell))
               {
                 explicit_boundary = true;
@@ -912,6 +910,19 @@ namespace pdflib
               }
 
             bool starts_word = words.empty() or explicit_boundary;
+            // TOC leaders may omit the boundary before their first dot while
+            // explicitly spacing every subsequent dot. Keep that first dot
+            // with the leader instead of attaching it to the heading.
+            if(not starts_word and cell.text == "." and
+               index + 4 < run.cells.size() and
+               is_semantic_space(run.cells[index + 1]) and
+               run.cells[index + 2].text == "." and
+               is_semantic_space(run.cells[index + 3]) and
+               run.cells[index + 4].text == "." and
+               previous_visible != nullptr and previous_visible->text != ".")
+              {
+                starts_word = true;
+              }
             if(not starts_word and previous_visible != nullptr)
               {
                 if(not word_baselines_compatible(*previous_visible, cell))

{docling_parse-7.22.0 → docling_parse-7.22.1}/src/parse/pdf_sanitators/constants.h

@@ -66,10 +66,10 @@ namespace pdflib
       {"\u2019", "'"},
       {"\u201A", ","},
       {"\u201B", "'"},
-      {"\u201C", "'"},
-      {"\u201D", "'"},
-      {"\u201E", "'"},
-      {"\u201F", "'"},
+      {"\u201C", "\""},
+      {"\u201D", "\""},
+      {"\u201E", "\""},
+      {"\u201F", "\""},
       
       {"\u2212", "-"},
     };

{docling_parse-7.22.0 → docling_parse-7.22.1}/tests/constants.py

@@ -1,7 +1,7 @@
 from tests.regression_page_selection import REGRESSION_PAGE_SELECTION
 
 HF_DATASET_REPO_ID = "docling-project/regression-dataset-for-docling-parse"
-HF_DATASET_REVISION = "65fd81c730024c5b989907bae9621603ba86b1b2"
+HF_DATASET_REVISION = "6d3458e2a6a729defd700a7b91e57b6cb2ce0290"
 
 REGRESSION_DIR = "tests/data/regression"
 

{docling_parse-7.22.0 → docling_parse-7.22.1}/tests/test_regression_cell_contraction.py

@@ -1,6 +1,7 @@
 #!/usr/bin/env python
 """Targeted regressions for adaptive character-to-word contraction."""
 
+from io import BytesIO
 from pathlib import Path
 
 from docling_parse.pdf_parser import DecodeConfig, DoclingPdfParser
@@ -8,6 +9,53 @@ from docling_parse.pdf_parser import DecodeConfig, DoclingPdfParser
 DATA = Path(__file__).parent / "data" / "regression"
 
 
+def _pdf_with_code_32_period() -> BytesIO:
+    """Map source byte 32 to a dot through /ToUnicode."""
+    cmap = (
+        b"/CIDInit /ProcSet findresource begin\n12 dict begin\nbegincmap\n"
+        b"/CMapName /Repro def\n/CMapType 2 def\n"
+        b"1 begincodespacerange\n<00> <FF>\nendcodespacerange\n"
+        b"1 beginbfchar\n<20> <002E>\nendbfchar\n"
+        b"endcmap\nCMapName currentdict /CMap defineresource pop\nend\nend\n"
+    )
+    content = b"BT /F1 24 Tf 72 700 Td <3820383331> Tj ET"
+
+    def stream(data: bytes) -> bytes:
+        return f"<< /Length {len(data)} >>\nstream\n".encode() + data + b"\nendstream"
+
+    objects = [
+        b"<< /Type /Catalog /Pages 2 0 R >>",
+        b"<< /Type /Pages /Kids [3 0 R] /Count 1 >>",
+        b"<< /Type /Page /Parent 2 0 R /MediaBox [0 0 595 842] "
+        b"/Resources << /Font << /F1 4 0 R >> >> /Contents 6 0 R >>",
+        b"<< /Type /Font /Subtype /Type1 /BaseFont /Helvetica /ToUnicode 5 0 R >>",
+        stream(cmap),
+        stream(content),
+    ]
+    pdf = bytearray(b"%PDF-1.7\n")
+    offsets = []
+    for number, obj in enumerate(objects, 1):
+        offsets.append(len(pdf))
+        pdf += f"{number} 0 obj\n".encode() + obj + b"\nendobj\n"
+    xref = len(pdf)
+    pdf += f"xref\n0 {len(objects) + 1}\n0000000000 65535 f \n".encode()
+    for offset in offsets:
+        pdf += f"{offset:010d} 00000 n \n".encode()
+    pdf += (
+        f"trailer\n<< /Size {len(objects) + 1} /Root 1 0 R >>\n"
+        f"startxref\n{xref}\n%%EOF\n"
+    ).encode()
+    return BytesIO(pdf)
+
+
+def test_source_code_32_can_decode_to_a_visible_glyph():
+    document = DoclingPdfParser(loglevel="fatal").load(_pdf_with_code_32_period())
+    try:
+        assert [cell.text for cell in document.get_page(1).word_cells] == ["8.831"]
+    finally:
+        document.unload()
+
+
 def _page(filename: str, page_no: int, *, keep_glyphs: bool = False):
     document = DoclingPdfParser(loglevel="fatal").load(
         path_or_stream=DATA / filename,
@@ -31,6 +79,39 @@ def test_newspaper_uses_local_dilation_for_word_boundaries():
     assert all(" " not in word for word in words)
 
 
+def test_short_headings_keep_inferred_word_boundaries():
+    page = _page("2305.03393v1-pg9.pdf", 1)
+    lines = [cell.text for cell in page.textline_cells]
+
+    assert "5.1 Hyper Parameter Optimization" in lines
+    assert "5.2 Quantitative Results" in lines
+
+
+def test_positioned_prose_and_headers_keep_word_boundaries():
+    paper = _page("2203.01017v2.pdf", 1)
+    assert any("custom OCR decoders" in cell.text for cell in paper.textline_cells)
+
+    details = _page("2203.01017v2.pdf", 6)
+    assert "5.1. Implementation Details" in [
+        cell.text for cell in details.textline_cells
+    ]
+
+    layout = _page("2206.01062.pdf", 3)
+    assert any(
+        "DocLayNet: A Large Human-Annotated Dataset for Document-Layout Analysis"
+        in cell.text
+        for cell in layout.textline_cells
+    )
+
+
+def test_first_toc_leader_dot_does_not_attach_to_heading():
+    page = _page("redp5110_sampled.pdf", 2)
+    words = [cell.text for cell in page.word_cells]
+
+    assert "Roles" in words
+    assert "Roles." not in words
+
+
 def test_elsevier_reference_words_do_not_contract_together():
     page = _page("elsevier-00.pdf", 9)
     words = [cell.text for cell in page.word_cells]

docling_parse-7.22.1/tests/test_unit_quote_sanitization.py

@@ -0,0 +1,111 @@
+#!/usr/bin/env python
+"""ASCII folding of typographic quotes in the cell text sanitizer.
+
+`text_constants::replacements` (src/parse/pdf_sanitators/constants.h) folds the
+U+2018..U+201F quote block to ASCII before any text leaves the parser, and the
+exported `orig` field is a copy of the folded text, so the fold is the only
+form a consumer ever sees. The single quotes rightly become `'`; the double
+quotes U+201C, U+201D, U+201E, U+201F must become `"`, not `'`: folding them
+to `'` turns `24<U+201D> x 36<U+201D>` (inches) into `24' x 36'` (feet)
+(docling-parse#335, docling-project/docling#1433).
+
+The PDF is built in memory with core-14 Helvetica and /WinAnsiEncoding and no
+embedded font program, so the glyph decode is not in question: 0x91..0x94 are
+quoteleft, quoteright, quotedblleft, quotedblright and 0x84 is quotedblbase in
+WinAnsi (PDF 32000-1, Annex D.2).
+"""
+
+from tests.pdf_builder import (
+    build_pdf,
+    content_stream,
+    parse_page,
+    simple_page_pdf,
+    stream_object,
+)
+
+HELVETICA_WINANSI = (
+    "<< /Type /Font /Subtype /Type1 /BaseFont /Helvetica /Encoding /WinAnsiEncoding >>"
+)
+
+
+def _extract(text: str) -> list[str]:
+    pdf = simple_page_pdf(
+        f"BT /F1 18 Tf 10 100 Td ({text}) Tj ET",
+        resources="/Font << /F1 5 0 R >>",
+        media_box="[0 0 400 200]",
+        extra_objects=[HELVETICA_WINANSI],
+    )
+    page = parse_page(pdf)
+    return [cell.text for cell in page.textline_cells]
+
+
+def test_double_quotes_fold_to_double_quote():
+    # 24<U+201D> x 36<U+201D>  (quotedblright, WinAnsi 0x94)
+    assert _extract(r"24\224 x 36\224") == ['24" x 36"']
+
+
+def test_single_quotes_fold_to_apostrophe():
+    # it<U+2019>s  (quoteright, 0x92) and a literal ASCII apostrophe stay `'`
+    assert _extract(r"it\222s, 10' poles") == ["it's, 10' poles"]
+
+
+def test_quote_family_folds_by_shape():
+    # U+2018 a U+2019, U+201C b U+201D, U+201E c U+201C  ->  'a' "b" "c"
+    assert _extract(r"\221a\222 \223b\224 \204c\223") == ['\'a\' "b" "c"']
+
+
+def test_orig_matches_folded_text():
+    pdf = simple_page_pdf(
+        r"BT /F1 18 Tf 10 100 Td (36\224) Tj ET",
+        resources="/Font << /F1 5 0 R >>",
+        media_box="[0 0 400 200]",
+        extra_objects=[HELVETICA_WINANSI],
+    )
+    page = parse_page(pdf)
+    (cell,) = page.textline_cells
+    assert cell.text == '36"'
+    assert cell.orig == cell.text
+
+
+# U+201B and U+201F have no WinAnsi code, so the full block goes through a
+# non-embedded Identity-H font whose /ToUnicode maps code 0x0001..0x0008 to
+# U+2018..U+201F in order.
+TO_UNICODE_QUOTES = """/CIDInit /ProcSet findresource begin
+12 dict begin
+begincmap
+/CMapName /Test-Quotes def
+/CMapType 2 def
+1 begincodespacerange
+<0000> <FFFF>
+endcodespacerange
+1 beginbfrange
+<0001> <0008> <2018>
+endbfrange
+endcmap
+CMapName currentdict /CMap defineresource pop
+end
+end"""
+
+
+def test_whole_quote_block_via_tounicode():
+    content = "BT /F1 20 Tf 20 100 Td <00010002000300040005000600070008> Tj ET"
+    objects = [
+        "<< /Type /Catalog /Pages 2 0 R >>",
+        "<< /Type /Pages /Kids [3 0 R] /Count 1 >>",
+        "<< /Type /Page /Parent 2 0 R /MediaBox [0 0 300 200] "
+        "/Resources << /Font << /F1 5 0 R >> >> /Contents 4 0 R >>",
+        content_stream(content),
+        "<< /Type /Font /Subtype /Type0 /BaseFont /Test-Quotes "
+        "/Encoding /Identity-H /DescendantFonts [6 0 R] /ToUnicode 7 0 R >>",
+        "<< /Type /Font /Subtype /CIDFontType2 /BaseFont /Test-Quotes "
+        "/CIDSystemInfo << /Registry (Adobe) /Ordering (Identity) /Supplement 0 >> "
+        "/FontDescriptor 8 0 R >>",
+        stream_object("", TO_UNICODE_QUOTES.encode("latin-1")),
+        "<< /Type /FontDescriptor /FontName /Test-Quotes /Flags 4 "
+        "/FontBBox [0 -200 1000 800] /ItalicAngle 0 /Ascent 800 /Descent -200 "
+        "/CapHeight 700 /StemV 80 >>",
+    ]
+    page = parse_page(build_pdf(objects))
+    text = "".join(cell.text for cell in page.textline_cells)
+    # U+2018 U+2019 U+201A U+201B U+201C U+201D U+201E U+201F
+    assert text == "'" + "'" + "," + "'" + '"' + '"' + '"' + '"'

{docling_parse-7.22.0 → docling_parse-7.22.1}/tests/test_unit_rotated_cell_contraction.py

@@ -40,10 +40,9 @@ def _rotated_text_pdf() -> bytes:
 
 
 def _explicit_space_with_irregular_inner_gap_pdf() -> bytes:
-    # A and B have a 3.336 pt gap even though they belong to one word. This
-    # models the varying side bearings exposed by tight Type 3 ink boxes. The
-    # same line contains a genuine PDF space between B and C, which is the
-    # authoritative word-boundary signal.
+    # Helvetica A and B have a 3.336 pt cursor gap, the same width as the
+    # explicit PDF space between B and C. The A/B boundary is ambiguous, but
+    # the source space must separate B and C.
     content = " ".join(
         [
             "BT /F1 12 Tf 100 50 Td",
@@ -144,7 +143,7 @@ def test_facing_edges_contract_at_45_degrees():
     assert word_bounds == pytest.approx(char_bounds, abs=1e-6)
 
 
-def test_explicit_spaces_override_irregular_inner_word_gaps():
+def test_explicit_space_remains_a_word_boundary():
     document = DoclingPdfParser(loglevel="fatal").load(
         path_or_stream=BytesIO(_explicit_space_with_irregular_inner_gap_pdf())
     )
@@ -153,5 +152,7 @@ def test_explicit_spaces_override_irregular_inner_word_gaps():
     finally:
         document.unload()
 
-    assert [cell.text for cell in page.word_cells] == ["AB", "CD"]
-    assert [cell.text for cell in page.textline_cells] == ["AB CD"]
+    words = [cell.text for cell in page.word_cells]
+    assert words[-1] == "CD"
+    assert "".join(words[:-1]) == "AB"
+    assert [cell.text for cell in page.textline_cells] == [" ".join(words)]

{docling_parse-7.22.0 → docling_parse-7.22.1}/uv.lock

@@ -2,9 +2,12 @@ version = 1
 revision = 3
 requires-python = ">=3.10"
 resolution-markers = [
-    "python_full_version >= '3.12'",
-    "python_full_version == '3.11.*'",
-    "python_full_version < '3.11'",
+    "python_full_version >= '3.12' and sys_platform != 'win32'",
+    "python_full_version >= '3.12' and sys_platform == 'win32'",
+    "python_full_version == '3.11.*' and sys_platform != 'win32'",
+    "python_full_version == '3.11.*' and sys_platform == 'win32'",
+    "python_full_version < '3.11' and sys_platform != 'win32'",
+    "python_full_version < '3.11' and sys_platform == 'win32'",
 ]
 
 [[package]]
@@ -292,7 +295,8 @@ name = "cibuildwheel"
 version = "2.23.3"
 source = { registry = "https://pypi.org/simple" }
 resolution-markers = [
-    "python_full_version < '3.11'",
+    "python_full_version < '3.11' and sys_platform != 'win32'",
+    "python_full_version < '3.11' and sys_platform == 'win32'",
 ]
 dependencies = [
     { name = "bashlex" },
@@ -315,8 +319,10 @@ name = "cibuildwheel"
 version = "3.3.1"
 source = { registry = "https://pypi.org/simple" }
 resolution-markers = [
-    "python_full_version >= '3.12'",
-    "python_full_version == '3.11.*'",
+    "python_full_version >= '3.12' and sys_platform != 'win32'",
+    "python_full_version >= '3.12' and sys_platform == 'win32'",
+    "python_full_version == '3.11.*' and sys_platform != 'win32'",
+    "python_full_version == '3.11.*' and sys_platform == 'win32'",
 ]
 dependencies = [
     { name = "bashlex" },
@@ -365,6 +371,11 @@ wheels = [
 name = "cmake"
 version = "3.31.10"
 source = { registry = "https://pypi.org/simple" }
+resolution-markers = [
+    "python_full_version >= '3.12' and sys_platform != 'win32'",
+    "python_full_version == '3.11.*' and sys_platform != 'win32'",
+    "python_full_version < '3.11' and sys_platform != 'win32'",
+]
 sdist = { url = "https://files.pythonhosted.org/packages/37/7b/fbadb3f4fe90ad6ef57f9f5f9e4f721af8e86376fbdf11da2c6ed099830e/cmake-3.31.10.tar.gz", hash = "sha256:ec3d14a0e72e401b3665034dc37901df17f0b4e9c5b163be6cfedfb93470ac0f", size = 34499, upload-time = "2025-11-20T17:07:54.664Z" }
 wheels = [
     { url = "https://files.pythonhosted.org/packages/99/3b/6ed408a99709808df4014bead522e075f0e8d1100e6d886eb5bc30fee04e/cmake-3.31.10-py3-none-macosx_10_10_universal2.whl", hash = "sha256:ad697643a00d9ba85179590a383c4f7401169b55ebf4b8b2938daf28c6bdeb6d", size = 48001731, upload-time = "2025-11-20T17:06:57.473Z" },
@@ -382,9 +393,22 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/ba/19/a3014beb02b344599e7a879f9c91d5e388aa76924b1369e13ffd535d613d/cmake-3.31.10-py3-none-musllinux_1_1_s390x.whl", hash = "sha256:b331984de38dbda22d676f8812c8905526341ba7b397fe8c359255ff4d051193", size = 27972718, upload-time = "2025-11-20T17:07:37.492Z" },
     { url = "https://files.pythonhosted.org/packages/b0/53/63b70d583c93352beeb692f0ec6bf4dede96f92df85c009577a663be304a/cmake-3.31.10-py3-none-musllinux_1_1_x86_64.whl", hash = "sha256:678fc23db37cc69f01e18eb28790450ecc9401fd2fcd43364cc18f92330c12c2", size = 29497491, upload-time = "2025-11-20T17:07:40.316Z" },
     { url = "https://files.pythonhosted.org/packages/fc/a3/9d3a3881e70b947abe2d779e4c37d1aa9ac7f8339354e78d6036d520a220/cmake-3.31.10-py3-none-musllinux_1_2_armv7l.whl", hash = "sha256:e135d5f1e59f1dc1c80eea87321977d6c0eff86cb601c28e0965a1dd457ea587", size = 33183971, upload-time = "2025-11-20T17:07:43.392Z" },
-    { url = "https://files.pythonhosted.org/packages/06/f2/8e13be79373ef808659e52680643e0388057060e2627757b90df72c50681/cmake-3.31.10-py3-none-win32.whl", hash = "sha256:b059a1810a2ce766b3e531bdc8d730bc192e260a9fa7dec7a0eb7a053d6063c7", size = 33416497, upload-time = "2025-11-20T17:07:46.467Z" },
-    { url = "https://files.pythonhosted.org/packages/84/4b/e433ab430c580ec8e3d54cb2ec5c7ea76ae0b1e6ef2c2998c854f21d7780/cmake-3.31.10-py3-none-win_amd64.whl", hash = "sha256:f1ea1fe826355560e8976c3d5794d9357444209bc0e0d56676c71e6a571fd474", size = 36630429, upload-time = "2025-11-20T17:07:49.445Z" },
-    { url = "https://files.pythonhosted.org/packages/9f/c3/4eb3288ccd779556fee4b7b0955bdb14545bbae83f85f7c819dde8013708/cmake-3.31.10-py3-none-win_arm64.whl", hash = "sha256:422a54711aa977af19d59b8f6010354cdda0b72a2e6d702b6d892e3e2cdf98a2", size = 35457178, upload-time = "2025-11-20T17:07:52.367Z" },
+]
+
+[[package]]
+name = "cmake"
+version = "4.4.3"
+source = { registry = "https://pypi.org/simple" }
+resolution-markers = [
+    "python_full_version >= '3.12' and sys_platform == 'win32'",
+    "python_full_version == '3.11.*' and sys_platform == 'win32'",
+    "python_full_version < '3.11' and sys_platform == 'win32'",
+]
+sdist = { url = "https://files.pythonhosted.org/packages/02/65/6cd38c1ee4292d4e91a7700ef827afdad365f981589db1750a2b74c84685/cmake-4.4.3.tar.gz", hash = "sha256:b0c2703ec0a624649dd184c0ac5ee4c7f5fe5ef35eb82f9da44f5a0903adb2b6", size = 34936, upload-time = "2026-08-30T12:26:40.394Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/8a/32/9462fc9022ebe960c05b87ca3d50cd93e267d1fcec99bea23add85cd3b76/cmake-4.4.3-py3-none-win32.whl", hash = "sha256:a21c9f60d530d678659f100445b9c0828467a67d4bb5e8eb76343aca91e80e90", size = 38773799, upload-time = "2026-08-30T12:26:31.624Z" },
+    { url = "https://files.pythonhosted.org/packages/90/8c/e872d51e4cc7cd3fc159bd273baf5faab2de5a4f290e2728d14d551e0029/cmake-4.4.3-py3-none-win_amd64.whl", hash = "sha256:708ad6b662ee89d3eb2eac02a1afba20ca653139cbe769fed046124d0c6bc050", size = 42324327, upload-time = "2026-08-30T12:26:34.963Z" },
+    { url = "https://files.pythonhosted.org/packages/49/0f/b5dcd56529c38990c3d854646e74c1eb5c325c858fe84455ec5616ac7bcd/cmake-4.4.3-py3-none-win_arm64.whl", hash = "sha256:c0d2ccd8f83605fedfd688bb37b4cb4008a7ddf6c32321b9c804168031a3cc74", size = 40494945, upload-time = "2026-08-30T12:26:38.195Z" },
 ]
 
 [[package]]
@@ -401,7 +425,8 @@ name = "contourpy"
 version = "1.3.2"
 source = { registry = "https://pypi.org/simple" }
 resolution-markers = [
-    "python_full_version < '3.11'",
+    "python_full_version < '3.11' and sys_platform != 'win32'",
+    "python_full_version < '3.11' and sys_platform == 'win32'",
 ]
 dependencies = [
     { name = "numpy", version = "2.2.6", source = { registry = "https://pypi.org/simple" } },
@@ -471,8 +496,10 @@ name = "contourpy"
 version = "1.3.3"
 source = { registry = "https://pypi.org/simple" }
 resolution-markers = [
-    "python_full_version >= '3.12'",
-    "python_full_version == '3.11.*'",
+    "python_full_version >= '3.12' and sys_platform != 'win32'",
+    "python_full_version >= '3.12' and sys_platform == 'win32'",
+    "python_full_version == '3.11.*' and sys_platform != 'win32'",
+    "python_full_version == '3.11.*' and sys_platform == 'win32'",
 ]
 dependencies = [
     { name = "numpy", version = "2.4.2", source = { registry = "https://pypi.org/simple" } },
@@ -703,7 +730,7 @@ wheels = [
 
 [[package]]
 name = "docling-parse"
-version = "7.22.0"
+version = "7.22.1"
 source = { editable = "." }
 dependencies = [
     { name = "docling-core" },
@@ -716,7 +743,8 @@ dependencies = [
 build = [
     { name = "cibuildwheel", version = "2.23.3", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version < '3.11'" },
     { name = "cibuildwheel", version = "3.3.1", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.11'" },
-    { name = "cmake" },
+    { name = "cmake", version = "3.31.10", source = { registry = "https://pypi.org/simple" }, marker = "sys_platform != 'win32'" },
+    { name = "cmake", version = "4.4.3", source = { registry = "https://pypi.org/simple" }, marker = "sys_platform == 'win32'" },
     { name = "delocate" },
     { name = "pybind11" },
     { name = "setuptools" },
@@ -755,7 +783,8 @@ requires-dist = [
 build = [
     { name = "cibuildwheel", marker = "python_full_version < '3.11'", specifier = ">=2.19.2,<3.0.0" },
     { name = "cibuildwheel", marker = "python_full_version >= '3.11'", specifier = ">=2.19.2,<4.0.0" },
-    { name = "cmake", specifier = ">=3.27.0,<4.0.0" },
+    { name = "cmake", marker = "sys_platform != 'win32'", specifier = ">=3.27.0,<5.0.0" },
+    { name = "cmake", marker = "sys_platform == 'win32'", specifier = ">=4.2.0,<5.0.0" },
     { name = "delocate", specifier = ">=0.11.0,<1.0.0" },
     { name = "pybind11", specifier = ">=2.13.6" },
     { name = "setuptools", specifier = ">=77.0.3" },
@@ -1696,7 +1725,8 @@ name = "numpy"
 version = "2.2.6"
 source = { registry = "https://pypi.org/simple" }
 resolution-markers = [
-    "python_full_version < '3.11'",
+    "python_full_version < '3.11' and sys_platform != 'win32'",
+    "python_full_version < '3.11' and sys_platform == 'win32'",
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/76/21/7d2a95e4bba9dc13d043ee156a356c0a8f0c6309dff6b21b4d71a073b8a8/numpy-2.2.6.tar.gz", hash = "sha256:e29554e2bef54a90aa5cc07da6ce955accb83f21ab5de01a62c8478897b264fd", size = 20276440, upload-time = "2025-05-17T22:38:04.611Z" }
 wheels = [
@@ -1761,8 +1791,10 @@ name = "numpy"
 version = "2.4.2"
 source = { registry = "https://pypi.org/simple" }
 resolution-markers = [
-    "python_full_version >= '3.12'",
-    "python_full_version == '3.11.*'",
+    "python_full_version >= '3.12' and sys_platform != 'win32'",
+    "python_full_version >= '3.12' and sys_platform == 'win32'",
+    "python_full_version == '3.11.*' and sys_platform != 'win32'",
+    "python_full_version == '3.11.*' and sys_platform == 'win32'",
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/57/fd/0005efbd0af48e55eb3c7208af93f2862d4b1a56cd78e84309a2d959208d/numpy-2.4.2.tar.gz", hash = "sha256:659a6107e31a83c4e33f763942275fd278b21d095094044eb35569e86a21ddae", size = 20723651, upload-time = "2026-01-31T23:13:10.135Z" }
 wheels = [