Don't instantiate an element with a coordinate system when there isn't a way to get its location (#913)

CHANGELOG.md

@@ -1,4 +1,4 @@
-## 0.8.1-dev1
+## 0.8.1-dev2
 
 ### Enhancements
 
@@ -10,6 +10,7 @@
 
 * Fixed `auto` strategy detected scanned document as having extractable text and using `fast` strategy, resulting in no output.
 * Fix list detection in MS Word documents.
+* Don't instantiate an element with a coordinate system when there isn't a way to get its location data.
 
 ## 0.8.0
 

test_unstructured/file_utils/test_filetype.py

@@ -4,6 +4,9 @@ import zipfile
 
 import magic
 import pytest
+from PIL import Image
+from unstructured_inference.inference import layout
+from unstructured_inference.inference.layoutelement import LocationlessLayoutElement
 
 from unstructured.file_utils import filetype
 from unstructured.file_utils.filetype import (
@@ -12,6 +15,7 @@ from unstructured.file_utils.filetype import (
     _is_text_file_a_csv,
     _is_text_file_a_json,
     detect_filetype,
+    document_to_element_list,
 )
 
 FILE_DIRECTORY = pathlib.Path(__file__).parent.resolve()
@@ -26,6 +30,29 @@ XLSX_MIME_TYPES = [
 ]
 
 
+class MockPageLayout(layout.PageLayout):
+    def __init__(self, number: int, image: Image):
+        self.number = number
+        self.image = image
+
+    @property
+    def elements(self):
+        return [
+            LocationlessLayoutElement(
+                type="Headline",
+                text="Charlie Brown and the Great Pumpkin",
+            ),
+        ]
+
+
+class MockDocumentLayout(layout.DocumentLayout):
+    @property
+    def pages(self):
+        return [
+            MockPageLayout(number=1, image=Image.new("1", (1, 1))),
+        ]
+
+
 @pytest.mark.parametrize(
     ("file", "expected"),
     [
@@ -436,3 +463,9 @@ def test_detect_filetype_skips_escape_commas_for_csv(tmpdir):
 
     with open(filename, "rb") as f:
         assert detect_filetype(file=f) == FileType.CSV
+
+
+def test_document_to_element_list_omits_coord_system_when_coord_points_absent():
+    layout_elem_absent_coordinates = MockDocumentLayout()
+    elements = document_to_element_list(layout_elem_absent_coordinates)
+    assert elements[0].metadata.coordinates is None

unstructured/__version__.py

@@ -1 +1 @@
-__version__ = "0.8.1-dev1"  # pragma: no cover
+__version__ = "0.8.1-dev2"  # pragma: no cover

unstructured/file_utils/filetype.py

@@ -451,7 +451,7 @@ def document_to_element_list(
     for i, page in enumerate(document.pages):
         page_elements: List[Element] = []
         for layout_element in page.elements:
-            if hasattr(page, "image"):
+            if hasattr(page, "image") and hasattr(layout_element, "coordinates"):
                 image_format = page.image.format
                 coordinate_system = PixelSpace(width=page.image.width, height=page.image.height)
             else:

unstructured/partition/pdf.py

@@ -326,12 +326,16 @@ def _process_pdfminer_pages(
                 _text = clean_extra_whitespace(_text)
                 if _text.strip():
                     text_segments.append(_text)
-                    element = element_from_text(_text)
                     coordinate_system = PixelSpace(
                         width=width,
                         height=height,
                     )
                     points = ((x1, y1), (x1, y2), (x2, y2), (x2, y1))
+                    element = element_from_text(
+                        _text,
+                        coordinates=points,
+                        coordinate_system=coordinate_system,
+                    )
                     coordinates_metadata = CoordinatesMetadata(
                         points=points,
                         system=coordinate_system,

unstructured/partition/text.py

@@ -1,7 +1,8 @@
 import re
-from typing import IO, Callable, List, Optional
+from typing import IO, Callable, List, Optional, Tuple
 
 from unstructured.cleaners.core import clean_bullets, group_broken_paragraphs
+from unstructured.documents.coordinates import CoordinateSystem
 from unstructured.documents.elements import (
     Address,
     Element,
@@ -143,14 +144,26 @@ def partition_text(
     return elements
 
 
-def element_from_text(text: str) -> Element:
+def element_from_text(
+    text: str,
+    coordinates: Optional[Tuple[Tuple[float, float], ...]] = None,
+    coordinate_system: Optional[CoordinateSystem] = None,
+) -> Element:
     if is_bulleted_text(text):
-        return ListItem(text=clean_bullets(text))
+        return ListItem(
+            text=clean_bullets(text),
+            coordinates=coordinates,
+            coordinate_system=coordinate_system,
+        )
     elif is_us_city_state_zip(text):
-        return Address(text=text)
+        return Address(text=text, coordinates=coordinates, coordinate_system=coordinate_system)
     elif is_possible_narrative_text(text):
-        return NarrativeText(text=text)
+        return NarrativeText(
+            text=text,
+            coordinates=coordinates,
+            coordinate_system=coordinate_system,
+        )
     elif is_possible_title(text):
-        return Title(text=text)
+        return Title(text=text, coordinates=coordinates, coordinate_system=coordinate_system)
     else:
-        return Text(text=text)
+        return Text(text=text, coordinates=coordinates, coordinate_system=coordinate_system)