docling 2.56.1__py3-none-any.whl → 2.57.0__py3-none-any.whl

docling/backend/docx/drawingml/utils.py

@@ -0,0 +1,131 @@
+import os
+import shutil
+import subprocess
+from pathlib import Path
+from tempfile import mkdtemp
+from typing import Callable, Optional
+
+import pypdfium2
+from docx.document import Document
+from PIL import Image, ImageChops
+
+
+def get_libreoffice_cmd(raise_if_unavailable: bool = False) -> Optional[str]:
+    """Return the libreoffice cmd and optionally test it."""
+
+    libreoffice_cmd = (
+        shutil.which("libreoffice")
+        or shutil.which("soffice")
+        or (
+            "/Applications/LibreOffice.app/Contents/MacOS/soffice"
+            if os.path.isfile("/Applications/LibreOffice.app/Contents/MacOS/soffice")
+            else None
+        )
+    )
+
+    if raise_if_unavailable:
+        if libreoffice_cmd is None:
+            raise RuntimeError("Libreoffice not found")
+
+        # The following test will raise if the libreoffice_cmd cannot be used
+        subprocess.run(
+            [
+                libreoffice_cmd,
+                "-h",
+            ],
+            stdout=subprocess.DEVNULL,
+            stderr=subprocess.DEVNULL,
+            check=True,
+        )
+
+    return libreoffice_cmd
+
+
+def get_docx_to_pdf_converter() -> Optional[Callable]:
+    """
+    Detects the best available DOCX to PDF tool and returns a conversion function.
+    The returned function accepts (input_path, output_path).
+    Returns None if no tool is available.
+    """
+
+    # Try LibreOffice
+    libreoffice_cmd = get_libreoffice_cmd()
+
+    if libreoffice_cmd:
+
+        def convert_with_libreoffice(input_path, output_path):
+            subprocess.run(
+                [
+                    libreoffice_cmd,
+                    "--headless",
+                    "--convert-to",
+                    "pdf",
+                    "--outdir",
+                    os.path.dirname(output_path),
+                    input_path,
+                ],
+                stdout=subprocess.DEVNULL,
+                stderr=subprocess.DEVNULL,
+                check=True,
+            )
+
+            expected_output = os.path.join(
+                os.path.dirname(output_path),
+                os.path.splitext(os.path.basename(input_path))[0] + ".pdf",
+            )
+            if expected_output != output_path:
+                os.rename(expected_output, output_path)
+
+        return convert_with_libreoffice
+
+    ## Space for other DOCX to PDF converters if available
+
+    # No tools found
+    return None
+
+
+def crop_whitespace(image: Image.Image, bg_color=None, padding=0) -> Image.Image:
+    if bg_color is None:
+        bg_color = image.getpixel((0, 0))
+
+    bg = Image.new(image.mode, image.size, bg_color)
+    diff = ImageChops.difference(image, bg)
+    bbox = diff.getbbox()
+
+    if bbox:
+        left, upper, right, lower = bbox
+        left = max(0, left - padding)
+        upper = max(0, upper - padding)
+        right = min(image.width, right + padding)
+        lower = min(image.height, lower + padding)
+        return image.crop((left, upper, right, lower))
+    else:
+        return image
+
+
+def get_pil_from_dml_docx(
+    docx: Document, converter: Optional[Callable]
+) -> Optional[Image.Image]:
+    if converter is None:
+        return None
+
+    temp_dir = Path(mkdtemp())
+    temp_docx = Path(temp_dir / "drawing_only.docx")
+    temp_pdf = Path(temp_dir / "drawing_only.pdf")
+
+    # 1) Save docx temporarily
+    docx.save(str(temp_docx))
+
+    # 2) Export to PDF
+    converter(temp_docx, temp_pdf)
+
+    # 3) Load PDF as PNG
+    pdf = pypdfium2.PdfDocument(temp_pdf)
+    page = pdf[0]
+    image = crop_whitespace(page.render(scale=2).to_pil())
+    page.close()
+    pdf.close()
+
+    shutil.rmtree(temp_dir, ignore_errors=True)
+
+    return image

docling/backend/msword_backend.py

@@ -1,8 +1,9 @@
 import logging
 import re
+from copy import deepcopy
 from io import BytesIO
 from pathlib import Path
-from typing import Any, List, Optional, Union
+from typing import Any, Callable, List, Optional, Union
 
 from docling_core.types.doc import (
     DocItemLabel,
@@ -33,6 +34,11 @@ from pydantic import AnyUrl
 from typing_extensions import override
 
 from docling.backend.abstract_backend import DeclarativeDocumentBackend
+from docling.backend.docx.drawingml.utils import (
+    get_docx_to_pdf_converter,
+    get_libreoffice_cmd,
+    get_pil_from_dml_docx,
+)
 from docling.backend.docx.latex.omml import oMath2Latex
 from docling.datamodel.base_models import InputFormat
 from docling.datamodel.document import InputDocument
@@ -64,6 +70,9 @@ class MsWordDocumentBackend(DeclarativeDocumentBackend):
         self.equation_bookends: str = "<eq>{EQ}</eq>"
         # Track processed textbox elements to avoid duplication
         self.processed_textbox_elements: List[int] = []
+        self.docx_to_pdf_converter: Optional[Callable] = None
+        self.docx_to_pdf_converter_init = False
+        self.display_drawingml_warning = True
 
         for i in range(-1, self.max_levels):
             self.parents[i] = None
@@ -80,18 +89,11 @@ class MsWordDocumentBackend(DeclarativeDocumentBackend):
             "indents": [None],
         }
 
-        self.docx_obj = None
-        try:
-            if isinstance(self.path_or_stream, BytesIO):
-                self.docx_obj = Document(self.path_or_stream)
-            elif isinstance(self.path_or_stream, Path):
-                self.docx_obj = Document(str(self.path_or_stream))
-
+        self.docx_obj = self.load_msword_file(
+            path_or_stream=self.path_or_stream, document_hash=self.document_hash
+        )
+        if self.docx_obj:
             self.valid = True
-        except Exception as e:
-            raise RuntimeError(
-                f"MsWordDocumentBackend could not load document with hash {self.document_hash}"
-            ) from e
 
     @override
     def is_valid(self) -> bool:
@@ -139,6 +141,22 @@ class MsWordDocumentBackend(DeclarativeDocumentBackend):
                 f"Cannot convert doc with {self.document_hash} because the backend failed to init."
             )
 
+    @staticmethod
+    def load_msword_file(
+        path_or_stream: Union[BytesIO, Path], document_hash: str
+    ) -> DocxDocument:
+        try:
+            if isinstance(path_or_stream, BytesIO):
+                return Document(path_or_stream)
+            elif isinstance(path_or_stream, Path):
+                return Document(str(path_or_stream))
+            else:
+                return None
+        except Exception as e:
+            raise RuntimeError(
+                f"MsWordDocumentBackend could not load document with hash {document_hash}"
+            ) from e
+
     def _update_history(
         self,
         name: str,
@@ -195,6 +213,7 @@ class MsWordDocumentBackend(DeclarativeDocumentBackend):
             }
             xpath_expr = etree.XPath(".//a:blip", namespaces=namespaces)
             drawing_blip = xpath_expr(element)
+            drawingml_els = element.findall(".//w:drawing", namespaces=namespaces)
 
             # Check for textbox content - check multiple textbox formats
             # Only process if the element hasn't been processed before
@@ -274,6 +293,26 @@ class MsWordDocumentBackend(DeclarativeDocumentBackend):
                 ):
                     te1 = self._handle_text_elements(element, docx_obj, doc)
                     added_elements.extend(te1)
+            # Check for DrawingML elements
+            elif drawingml_els:
+                if (
+                    self.docx_to_pdf_converter is None
+                    and self.docx_to_pdf_converter_init is False
+                ):
+                    self.docx_to_pdf_converter = get_docx_to_pdf_converter()
+                    self.docx_to_pdf_converter_init = True
+
+                if self.docx_to_pdf_converter is None:
+                    if self.display_drawingml_warning:
+                        if self.docx_to_pdf_converter is None:
+                            _log.warning(
+                                "Found DrawingML elements in document, but no DOCX to PDF converters. "
+                                "If you want these exported, make sure you have "
+                                "LibreOffice binary in PATH or specify its path with DOCLING_LIBREOFFICE_CMD."
+                            )
+                            self.display_drawingml_warning = False
+                else:
+                    self._handle_drawingml(doc=doc, drawingml_els=drawingml_els)
             # Check for the sdt containers, like table of contents
             elif tag_name in ["sdt"]:
                 sdt_content = element.find(".//w:sdtContent", namespaces=namespaces)
@@ -1381,3 +1420,39 @@ class MsWordDocumentBackend(DeclarativeDocumentBackend):
                 )
                 elem_ref.append(p3.get_ref())
         return elem_ref
+
+    def _handle_drawingml(self, doc: DoclingDocument, drawingml_els: Any):
+        # 1) Make an empty copy of the original document
+        dml_doc = self.load_msword_file(self.path_or_stream, self.document_hash)
+        body = dml_doc._element.body
+        for child in list(body):
+            body.remove(child)
+
+        # 2) Add DrawingML to empty document
+        new_para = dml_doc.add_paragraph()
+        new_r = new_para.add_run()
+        for dml in drawingml_els:
+            new_r._r.append(deepcopy(dml))
+
+        # 3) Export DOCX->PDF->PNG and save it in DoclingDocument
+        level = self._get_level()
+        try:
+            pil_image = get_pil_from_dml_docx(
+                dml_doc, converter=self.docx_to_pdf_converter
+            )
+            if pil_image is None:
+                raise UnidentifiedImageError
+
+            doc.add_picture(
+                parent=self.parents[level - 1],
+                image=ImageRef.from_pil(image=pil_image, dpi=72),
+                caption=None,
+            )
+        except (UnidentifiedImageError, OSError):
+            _log.warning("Warning: DrawingML image cannot be loaded by Pillow")
+            doc.add_picture(
+                parent=self.parents[level - 1],
+                caption=None,
+            )
+
+        return

docling/pipeline/vlm_pipeline.py

@@ -6,6 +6,7 @@ from typing import List, Optional, Union, cast
 
 from docling_core.types.doc import (
     BoundingBox,
+    ContentLayer,
     DocItem,
     DoclingDocument,
     ImageRef,
@@ -251,9 +252,9 @@ class VlmPipeline(PaginatedPipeline):
                 # No code blocks found, return original text
                 return text
 
-        for pg_idx, page in enumerate(conv_res.pages):
-            page_no = pg_idx + 1  # FIXME: might be incorrect
+        page_docs = []
 
+        for pg_idx, page in enumerate(conv_res.pages):
             predicted_text = ""
             if page.predictions.vlm_response:
                 predicted_text = page.predictions.vlm_response.text + "\n\n"
@@ -273,6 +274,24 @@ class VlmPipeline(PaginatedPipeline):
             )
             page_doc = backend.convert()
 
+            # Modify provenance in place for all items in the page document
+            for item, level in page_doc.iterate_items(
+                with_groups=True,
+                traverse_pictures=True,
+                included_content_layers=set(ContentLayer),
+            ):
+                if isinstance(item, DocItem):
+                    item.prov = [
+                        ProvenanceItem(
+                            page_no=pg_idx + 1,
+                            bbox=BoundingBox(
+                                t=0.0, b=0.0, l=0.0, r=0.0
+                            ),  # FIXME: would be nice not to have to "fake" it
+                            charspan=[0, 0],
+                        )
+                    ]
+
+            # Add page metadata to the page document before concatenation
             if page.image is not None:
                 pg_width = page.image.width
                 pg_height = page.image.height
@@ -280,27 +299,18 @@ class VlmPipeline(PaginatedPipeline):
                 pg_width = 1
                 pg_height = 1
 
-            conv_res.document.add_page(
-                page_no=page_no,
+            page_doc.add_page(
+                page_no=pg_idx + 1,
                 size=Size(width=pg_width, height=pg_height),
                 image=ImageRef.from_pil(image=page.image, dpi=72)
                 if page.image
                 else None,
             )
 
-            for item, level in page_doc.iterate_items():
-                item.prov = [
-                    ProvenanceItem(
-                        page_no=pg_idx + 1,
-                        bbox=BoundingBox(
-                            t=0.0, b=0.0, l=0.0, r=0.0
-                        ),  # FIXME: would be nice not to have to "fake" it
-                        charspan=[0, 0],
-                    )
-                ]
-                conv_res.document.append_child_item(child=item)
+            page_docs.append(page_doc)
 
-        return conv_res.document
+        final_doc = DoclingDocument.concatenate(docs=page_docs)
+        return final_doc
 
     def _turn_html_into_doc(self, conv_res):
         def _extract_html_code(text):
@@ -328,9 +338,9 @@ class VlmPipeline(PaginatedPipeline):
                 # No code blocks found, return original text
                 return text
 
-        for pg_idx, page in enumerate(conv_res.pages):
-            page_no = pg_idx + 1  # FIXME: might be incorrect
+        page_docs = []
 
+        for pg_idx, page in enumerate(conv_res.pages):
             predicted_text = ""
             if page.predictions.vlm_response:
                 predicted_text = page.predictions.vlm_response.text + "\n\n"
@@ -341,7 +351,7 @@ class VlmPipeline(PaginatedPipeline):
             out_doc = InputDocument(
                 path_or_stream=response_bytes,
                 filename=conv_res.input.file.name,
-                format=InputFormat.MD,
+                format=InputFormat.HTML,
                 backend=HTMLDocumentBackend,
             )
             backend = HTMLDocumentBackend(
@@ -350,6 +360,24 @@ class VlmPipeline(PaginatedPipeline):
             )
             page_doc = backend.convert()
 
+            # Modify provenance in place for all items in the page document
+            for item, level in page_doc.iterate_items(
+                with_groups=True,
+                traverse_pictures=True,
+                included_content_layers=set(ContentLayer),
+            ):
+                if isinstance(item, DocItem):
+                    item.prov = [
+                        ProvenanceItem(
+                            page_no=pg_idx + 1,
+                            bbox=BoundingBox(
+                                t=0.0, b=0.0, l=0.0, r=0.0
+                            ),  # FIXME: would be nice not to have to "fake" it
+                            charspan=[0, 0],
+                        )
+                    ]
+
+            # Add page metadata to the page document before concatenation
             if page.image is not None:
                 pg_width = page.image.width
                 pg_height = page.image.height
@@ -357,27 +385,19 @@ class VlmPipeline(PaginatedPipeline):
                 pg_width = 1
                 pg_height = 1
 
-            conv_res.document.add_page(
-                page_no=page_no,
+            page_doc.add_page(
+                page_no=pg_idx + 1,
                 size=Size(width=pg_width, height=pg_height),
                 image=ImageRef.from_pil(image=page.image, dpi=72)
                 if page.image
                 else None,
             )
 
-            for item, level in page_doc.iterate_items():
-                item.prov = [
-                    ProvenanceItem(
-                        page_no=pg_idx + 1,
-                        bbox=BoundingBox(
-                            t=0.0, b=0.0, l=0.0, r=0.0
-                        ),  # FIXME: would be nice not to have to "fake" it
-                        charspan=[0, 0],
-                    )
-                ]
-                conv_res.document.append_child_item(child=item)
+            page_docs.append(page_doc)
 
-        return conv_res.document
+        # Concatenate all page documents to preserve hierarchy
+        final_doc = DoclingDocument.concatenate(docs=page_docs)
+        return final_doc
 
     @classmethod
     def get_default_options(cls) -> VlmPipelineOptions:

{docling-2.56.1.dist-info → docling-2.57.0.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: docling
-Version: 2.56.1
+Version: 2.57.0
 Summary: SDK and CLI for parsing PDF, DOCX, HTML, and more, to a unified document representation for powering downstream workflows such as gen AI applications.
 Author-email: Christoph Auer <cau@zurich.ibm.com>, Michele Dolfi <dol@zurich.ibm.com>, Maxim Lysak <mly@zurich.ibm.com>, Nikos Livathinos <nli@zurich.ibm.com>, Ahmed Nassar <ahn@zurich.ibm.com>, Panos Vagenas <pva@zurich.ibm.com>, Peter Staar <taa@zurich.ibm.com>
 License-Expression: MIT

{docling-2.56.1.dist-info → docling-2.57.0.dist-info}/RECORD

@@ -15,12 +15,13 @@ docling/backend/md_backend.py,sha256=TWboEPHl93pqI_Go1a3XpP-KpzI3d17xo5ZW42Ul0kY
 docling/backend/mets_gbs_backend.py,sha256=EA8sY6tbmGiysKGYPPZiNlK-i7Adn8bLTo-7Ym15hTU,12774
 docling/backend/msexcel_backend.py,sha256=GOuA-MlShpzFmCmJq3-Z28iquwWUg4k8v-AT4O-aAQI,19305
 docling/backend/mspowerpoint_backend.py,sha256=wJgB2JStEPfD7MPpWQlpPN7bffPxaHFUnKD4wj8SLxU,15114
-docling/backend/msword_backend.py,sha256=Jfd57hzG8iFVAzqsOAHe5jG8LCHAIBXJhQCW0tESnMM,54405
+docling/backend/msword_backend.py,sha256=L44vFoSHOtbX-S_lSb8EKW-nzwL_ptVPhNV74ydmwqE,57457
 docling/backend/noop_backend.py,sha256=EOPbD86FzZPX-K_DpNrJh0_lC0bZz--4DpG-OagDNGY,1688
 docling/backend/pdf_backend.py,sha256=Wcd1NSrAMjXK8VicTki5p-j-JLofklt07eF0kIG17_0,3361
 docling/backend/pypdfium2_backend.py,sha256=AYhWs9S8W_TkAK0-OkRmUNf4HUZl26FP7-XYjwU5zDk,14209
 docling/backend/webvtt_backend.py,sha256=9xPcfWVLuqhEAFrkv8aU36qHnSgjeINZAXT_C9C6XJA,19165
 docling/backend/docx/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+docling/backend/docx/drawingml/utils.py,sha256=E9Iq8_052eEV5L1IN3ZqFX9eBidH56DKNlh6Tk7Do0I,3640
 docling/backend/docx/latex/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 docling/backend/docx/latex/latex_dict.py,sha256=tFJp4ScT_AkY2ON7nLEa560p601Jq2glcZvMKxxjn7w,6593
 docling/backend/docx/latex/omml.py,sha256=4vh9FCbXh-Tb6KJGqNwzlMUMYEnnJgBtBI24dwy6t2U,12416
@@ -88,7 +89,7 @@ docling/pipeline/extraction_vlm_pipeline.py,sha256=veUOTe8nGdnduZKaGn1RRb-NfU1H6
 docling/pipeline/simple_pipeline.py,sha256=FSL_ucDd9k0D9DjNKMUkyCULIU8a057dvWfLEPmAc2A,2287
 docling/pipeline/standard_pdf_pipeline.py,sha256=xOge0zP5wli51n_6QLrFHQlwwvsivI7OMt00tht3my4,10479
 docling/pipeline/threaded_standard_pdf_pipeline.py,sha256=i67G5AOW7PIFCe5JS2sdBmPAKvAH6ScxIBhjwOGZcrI,28183
-docling/pipeline/vlm_pipeline.py,sha256=oMcdgzym_UQbVN3bajux_hENY40XGOnb6NU6Kwje2Os,15376
+docling/pipeline/vlm_pipeline.py,sha256=HSbSoGZyy4eIK8eOL2g_NymrHg8r-DrB2buggJQAqHU,16189
 docling/utils/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 docling/utils/accelerator_utils.py,sha256=DSajLxVx1JEVT0zt5de26llciLNlVfIDfSa2zYCFJzQ,2909
 docling/utils/api_image_request.py,sha256=kQDmTvQT6M2IgXnGYeoNflI6sLUG6WTCcEft94CRwWg,5379
@@ -102,9 +103,9 @@ docling/utils/orientation.py,sha256=jTyLxyT31FlOodZoBMlADHNQK2lAWKYVs5z7pXd_6Cg,
 docling/utils/profiling.py,sha256=YaMGoB9MMZpagF9mb5ndoHj8Lpb9aIdb7El-Pl7IcFs,1753
 docling/utils/utils.py,sha256=kJtIYuzXeOyJHYlxmLAo7dGM5rEsDa1i84qEsUj1nio,1908
 docling/utils/visualization.py,sha256=tY2ylE2aiQKkmzlSLnFW-HTfFyqUUMguW18ldd1PLfo,2868
-docling-2.56.1.dist-info/licenses/LICENSE,sha256=mBb7ErEcM8VS9OhiGHnQ2kk75HwPhr54W1Oiz3965MY,1088
-docling-2.56.1.dist-info/METADATA,sha256=LgkCRBKjI20tarzBBeFpRi0vo_fZ4Sh6rD6K8yQkrY8,11364
-docling-2.56.1.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-docling-2.56.1.dist-info/entry_points.txt,sha256=hzVlbeE0aMSTQ9S0-NTYN0Hmgsn6qL_EA2qX4UbkAuY,149
-docling-2.56.1.dist-info/top_level.txt,sha256=vkIywP-USjFyYo1AIRQbWQQaL3xB5jf8vkCYdTIfNic,8
-docling-2.56.1.dist-info/RECORD,,
+docling-2.57.0.dist-info/licenses/LICENSE,sha256=mBb7ErEcM8VS9OhiGHnQ2kk75HwPhr54W1Oiz3965MY,1088
+docling-2.57.0.dist-info/METADATA,sha256=oDfwFunLJTLSDVastMVq9JkUpIgeKOOVX1MZb6rtqcE,11364
+docling-2.57.0.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+docling-2.57.0.dist-info/entry_points.txt,sha256=hzVlbeE0aMSTQ9S0-NTYN0Hmgsn6qL_EA2qX4UbkAuY,149
+docling-2.57.0.dist-info/top_level.txt,sha256=vkIywP-USjFyYo1AIRQbWQQaL3xB5jf8vkCYdTIfNic,8
+docling-2.57.0.dist-info/RECORD,,