fix(excel): merge two separated tables#2626

docling/backend/msexcel_backend.py

@@ -406,7 +406,7 @@ def _find_true_data_bounds(self, sheet: Worksheet) -> DataRegion:
         return DataRegion(min_row, max_row, min_col, max_col)
 
     def _find_data_tables(self, sheet: Worksheet) -> list[ExcelTable]:
-        """Find all compact rectangular data tables in an Excel worksheet.
+        """Find all rectangular data tables in an Excel worksheet. Two non-empty cells forms a data table if they are adjancent.
 
         Args:
             sheet: The Excel worksheet to be parsed.
@@ -437,7 +437,7 @@ def _find_data_tables(self, sheet: Worksheet) -> list[ExcelTable]:
 
                 # If the cell starts a new table, find its bounds
                 table_bounds, visited_cells = self._find_table_bounds(
-                    sheet, ri, rj, bounds.max_row, bounds.max_col
+                    sheet, ri, rj, bounds
                 )
 
                 visited.update(visited_cells)  # Mark these cells as visited
@@ -450,26 +450,48 @@ def _find_table_bounds(
         sheet: Worksheet,
         start_row: int,
         start_col: int,
-        max_row: int,
-        max_col: int,
+        bounds: DataRegion,
     ) -> tuple[ExcelTable, set[tuple[int, int]]]:
-        """Determine the bounds of a compact rectangular table.
+        """Determine the bounds of a rectangular table. Two non-empty cells forms a data table if they are adjancent.
 
         Args:
             sheet: The Excel worksheet to be parsed.
             start_row: The row number of the starting cell.
             start_col: The column number of the starting cell.
-            max_row: Maximum row boundary from true data bounds.
-            max_col: Maximum column boundary from true data bounds.
+            bounds: boundary from true data bounds.
 
         Returns:
             A tuple with an Excel table and a set of cell coordinates.
         """
         _log.debug("find_table_bounds")
 
+        max_row = bounds.max_row
+        max_col = bounds.max_col
+
         table_max_row = self._find_table_bottom(sheet, start_row, start_col, max_row)
         table_max_col = self._find_table_right(sheet, start_row, start_col, max_col)
 
+        # Expand table on the left, bounds indexing starts with + 1
+        for rj in range(start_col, bounds.min_col - 1, -1):
+            if self._is_column_empty(sheet, start_row + 2, table_max_row + 1, rj):
+                break
+            else:
+                start_col = rj - 1
+
+        # Expand table on the right
+        for rj in range(table_max_col + 1, max_col):
+            if self._is_column_empty(sheet, start_row + 2, table_max_row + 1, rj + 1):
+                break
+            else:
+                table_max_col = rj
+
+        # Expand the table on the bottom
+        for ri in range(table_max_row + 1, max_row):
+            if self._is_row_empty(sheet, ri + 1, start_col + 1, table_max_col + 1):
+                break
+            else:
+                table_max_row = ri
+
         # Collect the data within the bounds
         data = []
         visited_cells: set[tuple[int, int]] = set()
@@ -556,7 +578,6 @@ def _find_table_bottom(
                 (mr for mr in sheet.merged_cells.ranges if cell.coordinate in mr),
                 None,
             )
-
             if cell.value is None and not merged_range:
                 break  # Stop if the cell is empty and not merged
 
@@ -684,3 +705,31 @@ def _get_sheet_content_layer(sheet: Worksheet) -> Optional[ContentLayer]:
             if sheet.sheet_state == Worksheet.SHEETSTATE_VISIBLE
             else ContentLayer.INVISIBLE
         )
+
+    @staticmethod
+    def _is_column_empty(
+        sheet: Worksheet, start_row: int, end_row: int, col: int
+    ) -> bool:
+        for (value,) in sheet.iter_rows(
+            min_row=start_row,
+            max_row=end_row,
+            min_col=col,
+            max_col=col,
+            values_only=True,
+        ):
+            if value not in (None, ""):
+                return False  # Found a non-empty value
+        return True  # All cells were empty
+
+    @staticmethod
+    def _is_row_empty(sheet: Worksheet, row: int, start_col: int, end_col: int) -> bool:
+        for (value,) in sheet.iter_cols(
+            min_row=row,
+            max_row=row,
+            min_col=start_col,
+            max_col=end_col,
+            values_only=True,
+        ):
+            if value not in (None, ""):
+                return False  # Found a non-empty value
+        return True  # All cells were empty

tests/data/groundtruth/docling_v2/xlsx_02_sample_sales_data.xlsm.json

@@ -1,11 +1,11 @@
 {
   "schema_name": "DoclingDocument",
-  "version": "1.7.0",
-  "name": "sample_sales_data",
+  "version": "1.8.0",
+  "name": "xlsx_02_sample_sales_data",
   "origin": {
     "mimetype": "application/vnd.openxmlformats-officedocument.spreadsheetml.sheet",
     "binary_hash": 14806485565397602516,
-    "filename": "sample_sales_data.xlsm"
+    "filename": "xlsx_02_sample_sales_data.xlsm"
   },
   "furniture": {
     "self_ref": "#/furniture",

tests/data/groundtruth/docling_v2/xlsx_03_chartsheet.xlsx.json

@@ -1,10 +1,10 @@
 {
   "schema_name": "DoclingDocument",
-  "version": "1.7.0",
+  "version": "1.8.0",
   "name": "xlsx_03_chartsheet",
   "origin": {
     "mimetype": "application/vnd.openxmlformats-officedocument.spreadsheetml.sheet",
-    "binary_hash": 548415533138925042,
+    "binary_hash": 472193488349663234,
     "filename": "xlsx_03_chartsheet.xlsx"
   },
   "furniture": {
@@ -855,4 +855,4 @@
       "page_no": 2
     }
   }
-}
+}

tests/data/groundtruth/docling_v2/xlsx_04_inflated.xlsx.json

@@ -1,11 +1,11 @@
 {
   "schema_name": "DoclingDocument",
-  "version": "1.7.0",
-  "name": "test-02",
+  "version": "1.8.0",
+  "name": "xlsx_04_inflated",
   "origin": {
     "mimetype": "application/vnd.openxmlformats-officedocument.spreadsheetml.sheet",
-    "binary_hash": 13967282146026440806,
-    "filename": "test-02.xlsx"
+    "binary_hash": 8997038978642400831,
+    "filename": "xlsx_04_inflated.xlsx"
   },
   "furniture": {
     "self_ref": "#/furniture",