miscs (Interro02) : horizontal cutting resolution

copienator/commands/splitting_int.py

@@ -80,15 +80,9 @@ def _save_cropped_page(
     y1: float,
     output_path: Path,
 ) -> None:
-    page = document[page_number]
-    rotated_rectangle = page.rect * page.transformation_matrix
-    visual_crop = pymupdf.Rect(
-        rotated_rectangle.x0 + x0,
-        y0,
-        rotated_rectangle.x0 + x1,
-        y1,
-    )
-    unrotated_clip = visual_crop * page.derotation_matrix
+    # The source has been normalized by _prepare_split_pages: no rotation or
+    # CropBox translation remains in the coordinates passed to show_pdf_page.
+    visual_crop = pymupdf.Rect(x0, y0, x1, y1)
     cropped = pymupdf.open()
     try:
         target_page = cropped.new_page(width=visual_crop.width, height=visual_crop.height)
@@ -96,14 +90,32 @@ def _save_cropped_page(
             target_page.rect,
             document,
             page_number,
-            rotate=-page.rotation,
-            clip=unrotated_clip,
+            clip=visual_crop,
         )
         cropped.save(output_path)
     finally:
         cropped.close()
 
 
+def _prepare_split_pages(document: pymupdf.Document) -> list[pymupdf.Rect]:
+    """Use the label preview's full-page coordinates; retain visible bounds.
+
+    Work only on the in-memory document. Baking rotation into the content after
+    restoring the MediaBox avoids show_pdf_page's rotated CropBox offsets.
+    Intersecting with the saved bounds later preserves prior margin cropping.
+    """
+    visible_bounds = []
+    for page in document:
+        crop = page.cropbox
+        media = page.mediabox
+        full_crop = pymupdf.Rect(media.x0, 0, media.x1, media.height)
+        page.set_cropbox(full_crop)
+        crop -= (full_crop.x0, full_crop.y0, full_crop.x0, full_crop.y0)
+        visible_bounds.append(crop * page.rotation_matrix)
+        page.remove_rotation()
+    return visible_bounds
+
+
 def _render_split_outputs(
     input_pdf: Path,
     coords_list: list[Coordinate],
@@ -112,6 +124,7 @@ def _render_split_outputs(
     """Render every current answer into an otherwise empty staging directory."""
     document = pymupdf.open(input_pdf)
     try:
+        visible_bounds = _prepare_split_pages(document)
         parsed = _parse_coordinates(coords_list)
         parts_by_label: defaultdict[str, list[Path]] = defaultdict(list)
         with tempfile.TemporaryDirectory(prefix="copienator-split-") as temp_directory:
@@ -160,16 +173,20 @@ def _render_split_outputs(
                     page = document[page_number]
                     y0 = (y_start / 1000) * page.rect.height if page_number == start_page else 0
                     y1 = (end_y / 1000) * page.rect.height if page_number == end_page else page.rect.height
-                    if y1 <= y0 + 1:
+                    clip = pymupdf.Rect(
+                        fraction_x0 * page.rect.width, y0,
+                        fraction_x1 * page.rect.width, y1,
+                    ) & visible_bounds[page_number] & page.rect
+                    if clip.is_empty or clip.height <= 1 or clip.width <= 1:
                         continue
                     part_path = temporary / f"part-{index}-{page_number}.pdf"
                     _save_cropped_page(
                         document,
                         page_number,
-                        fraction_x0 * page.rect.width,
-                        y0,
-                        fraction_x1 * page.rect.width,
-                        y1,
+                        clip.x0,
+                        clip.y0,
+                        clip.x1,
+                        clip.y1,
                         part_path,
                     )
                     parts_by_label[clean_label].append(part_path)
@@ -264,4 +281,3 @@ def main(argv: Sequence[str] | None = None) -> int:
 
 if __name__ == "__main__":
     raise SystemExit(main())
-