fix: pptx line break and space handling (#1664)

Signed-off-by: Martin Wind <martin.wind@im-c.at>

docling/backend/mspowerpoint_backend.py

@@ -20,6 +20,7 @@ from docling_core.types.doc.document import ContentLayer
 from PIL import Image, UnidentifiedImageError
 from pptx import Presentation
 from pptx.enum.shapes import MSO_SHAPE_TYPE, PP_PLACEHOLDER
+from pptx.oxml.text import CT_TextLineBreak
 
 from docling.backend.abstract_backend import (
     DeclarativeDocumentBackend,
@@ -120,136 +121,91 @@ class MsPowerpointDocumentBackend(DeclarativeDocumentBackend, PaginatedDocumentB
 
         return prov
 
-    def handle_text_elements(self, shape, parent_slide, slide_ind, doc, slide_size):  # noqa: C901
+    def handle_text_elements(self, shape, parent_slide, slide_ind, doc, slide_size):
-        is_a_list = False
         is_list_group_created = False
         enum_list_item_value = 0
         new_list = None
-        bullet_type = "None"
-        list_label = GroupLabel.LIST
         doc_label = DocItemLabel.LIST_ITEM
         prov = self.generate_prov(shape, slide_ind, shape.text.strip(), slide_size)
 
-        # Identify if shape contains lists
+        def is_list_item(paragraph):
-        for paragraph in shape.text_frame.paragraphs:
+            """Check if the paragraph is a list item."""
-            # Check if paragraph is a bullet point using the `element` XML
             p = paragraph._element
             if (
                 p.find(".//a:buChar", namespaces={"a": self.namespaces["a"]})
                 is not None
             ):
-                bullet_type = "Bullet"
+                return (True, "Bullet")
-                is_a_list = True
             elif (
                 p.find(".//a:buAutoNum", namespaces={"a": self.namespaces["a"]})
                 is not None
             ):
-                bullet_type = "Numbered"
+                return (True, "Numbered")
-                is_a_list = True
+            elif paragraph.level > 0:
-            else:
-                is_a_list = False
-
-            if paragraph.level > 0:
                 # Most likely a sub-list
-                is_a_list = True
+                return (True, "None")
-
-            if is_a_list:
-                # Determine if this is an unordered list or an ordered list.
-                # Set GroupLabel.ORDERED_LIST when it fits.
-                if bullet_type == "Numbered":
-                    list_label = GroupLabel.ORDERED_LIST
-
-            if is_a_list:
-                _log.debug("LIST DETECTED!")
             else:
-                _log.debug("No List")
+                return (False, "None")
-
-        # If there is a list inside of the shape, create a new docling list to assign list items to
-        # if is_a_list:
-        #     new_list = doc.add_group(
-        #         label=list_label, name=f"list", parent=parent_slide
-        #     )
 
         # Iterate through paragraphs to build up text
         for paragraph in shape.text_frame.paragraphs:
-            # p_text = paragraph.text.strip()
+            is_a_list, bullet_type = is_list_item(paragraph)
             p = paragraph._element
-            enum_list_item_value += 1
-            inline_paragraph_text = ""
-            inline_list_item_text = ""
 
-            for e in p.iterfind(".//a:r", namespaces={"a": self.namespaces["a"]}):
+            # Convert line breaks to spaces and accumulate text
-                if len(e.text.strip()) > 0:
+            p_text = ""
-                    e_is_a_list_item = False
+            for e in p.content_children:
-                    is_numbered = False
+                if isinstance(e, CT_TextLineBreak):
-                    if (
+                    p_text += " "
-                        p.find(".//a:buChar", namespaces={"a": self.namespaces["a"]})
+                else:
-                        is not None
+                    p_text += e.text
-                    ):
-                        bullet_type = "Bullet"
-                        e_is_a_list_item = True
-                    elif (
-                        p.find(".//a:buAutoNum", namespaces={"a": self.namespaces["a"]})
-                        is not None
-                    ):
-                        bullet_type = "Numbered"
-                        is_numbered = True
-                        e_is_a_list_item = True
-                    else:
-                        e_is_a_list_item = False
 
-                    if e_is_a_list_item:
+            if is_a_list:
-                        if len(inline_paragraph_text) > 0:
+                enum_marker = ""
-                            # output accumulated inline text:
+                enumerated = bullet_type == "Numbered"
-                            doc.add_text(
+
-                                label=doc_label,
+                if not is_list_group_created:
-                                parent=parent_slide,
+                    new_list = doc.add_group(
-                                text=inline_paragraph_text,
+                        label=GroupLabel.ORDERED_LIST
-                                prov=prov,
+                        if enumerated
-                            )
+                        else GroupLabel.LIST,
-                        # Set marker and enumerated arguments if this is an enumeration element.
+                        name="list",
-                        inline_list_item_text += e.text
+                        parent=parent_slide,
-                        # print(e.text)
+                    )
-                    else:
+                    is_list_group_created = True
-                        # Assign proper label to the text, depending if it's a Title or Section Header
+                    enum_list_item_value = 0
-                        # For other types of text, assign - PARAGRAPH
+
-                        doc_label = DocItemLabel.PARAGRAPH
+                if enumerated:
-                        if shape.is_placeholder:
+                    enum_list_item_value += 1
-                            placeholder_type = shape.placeholder_format.type
+                    enum_marker = str(enum_list_item_value) + "."
-                            if placeholder_type in [
+
-                                PP_PLACEHOLDER.CENTER_TITLE,
+                doc.add_list_item(
-                                PP_PLACEHOLDER.TITLE,
+                    marker=enum_marker,
-                            ]:
+                    enumerated=enumerated,
-                                # It's a title
+                    parent=new_list,
-                                doc_label = DocItemLabel.TITLE
+                    text=p_text,
-                            elif placeholder_type == PP_PLACEHOLDER.SUBTITLE:
+                    prov=prov,
-                                DocItemLabel.SECTION_HEADER
+                )
-                        enum_list_item_value = 0
+            else:  # is paragraph not a list item
-                        inline_paragraph_text += e.text
+                # Assign proper label to the text, depending if it's a Title or Section Header
+                # For other types of text, assign - PARAGRAPH
+                doc_label = DocItemLabel.PARAGRAPH
+                if shape.is_placeholder:
+                    placeholder_type = shape.placeholder_format.type
+                    if placeholder_type in [
+                        PP_PLACEHOLDER.CENTER_TITLE,
+                        PP_PLACEHOLDER.TITLE,
+                    ]:
+                        # It's a title
+                        doc_label = DocItemLabel.TITLE
+                    elif placeholder_type == PP_PLACEHOLDER.SUBTITLE:
+                        DocItemLabel.SECTION_HEADER
 
-            if len(inline_paragraph_text) > 0:
                 # output accumulated inline text:
                 doc.add_text(
                     label=doc_label,
                     parent=parent_slide,
-                    text=inline_paragraph_text,
+                    text=p_text,
-                    prov=prov,
-                )
-
-            if len(inline_list_item_text) > 0:
-                enum_marker = ""
-                if is_numbered:
-                    enum_marker = str(enum_list_item_value) + "."
-                if not is_list_group_created:
-                    new_list = doc.add_group(
-                        label=list_label, name="list", parent=parent_slide
-                    )
-                    is_list_group_created = True
-                doc.add_list_item(
-                    marker=enum_marker,
-                    enumerated=is_numbered,
-                    parent=new_list,
-                    text=inline_list_item_text,
                     prov=prov,
                 )
         return

tests/data/groundtruth/docling_v2/powerpoint_bad_text.pptx.itxt

@@ -0,0 +1,3 @@
+item-0 at level 0: unspecified: group _root_
+  item-1 at level 1: chapter: group slide-0
+    item-2 at level 2: title: X-Library The fully customisable ... llection exclusively for our customers

tests/data/groundtruth/docling_v2/powerpoint_bad_text.pptx.json

@@ -0,0 +1,86 @@
+{
+  "schema_name": "DoclingDocument",
+  "version": "1.3.0",
+  "name": "powerpoint_bad_text",
+  "origin": {
+    "mimetype": "application/vnd.ms-powerpoint",
+    "binary_hash": 1443005848482130016,
+    "filename": "powerpoint_bad_text.pptx"
+  },
+  "furniture": {
+    "self_ref": "#/furniture",
+    "children": [],
+    "content_layer": "furniture",
+    "name": "_root_",
+    "label": "unspecified"
+  },
+  "body": {
+    "self_ref": "#/body",
+    "children": [
+      {
+        "$ref": "#/groups/0"
+      }
+    ],
+    "content_layer": "body",
+    "name": "_root_",
+    "label": "unspecified"
+  },
+  "groups": [
+    {
+      "self_ref": "#/groups/0",
+      "parent": {
+        "$ref": "#/body"
+      },
+      "children": [
+        {
+          "$ref": "#/texts/0"
+        }
+      ],
+      "content_layer": "body",
+      "name": "slide-0",
+      "label": "chapter"
+    }
+  ],
+  "texts": [
+    {
+      "self_ref": "#/texts/0",
+      "parent": {
+        "$ref": "#/groups/0"
+      },
+      "children": [],
+      "content_layer": "body",
+      "label": "title",
+      "prov": [
+        {
+          "page_no": 1,
+          "bbox": {
+            "l": 1041400.0,
+            "t": 4582390.0,
+            "r": 8083550.0,
+            "b": 1689099.0,
+            "coord_origin": "BOTTOMLEFT"
+          },
+          "charspan": [
+            0,
+            118
+          ]
+        }
+      ],
+      "orig": "X-Library The fully customisable and copyright-free standard content template collection exclusively for our customers",
+      "text": "X-Library The fully customisable and copyright-free standard content template collection exclusively for our customers"
+    }
+  ],
+  "pictures": [],
+  "tables": [],
+  "key_value_items": [],
+  "form_items": [],
+  "pages": {
+    "1": {
+      "size": {
+        "width": 12190413.0,
+        "height": 6858000.0
+      },
+      "page_no": 1
+    }
+  }
+}

tests/data/groundtruth/docling_v2/powerpoint_bad_text.pptx.md

@@ -0,0 +1 @@
+# X-Library The fully customisable and copyright-free standard content template collection exclusively for our customers