Fix barcode detection bugs and improve robustness

- Fix decode_with_variants() unreachable code bug (results was defined after return)
- Add robust decode variants: grayscale, scale-up, rotations (90°, 180°, 270°)
- Use LANCZOS resampling for better quality in newer Pillow versions
- Improve UPC-A/EAN-13 validation with leading zero normalization
- Remove arbitrary coordinate offset (+ page_num * 1000) for cleaner positioning
- Add comprehensive error handling and fallbacks
- Enhanced barcode detection with multiple image processing variants

pdf_comparator.py

@@ -571,35 +571,82 @@ def ean_like_checksum_ok(digits: str) -> bool:
     return True
 
 def validate_symbology(symbology: str, data: bytes) -> bool:
+    """Validate barcode symbology with improved UPC-A/EAN-13 handling"""
     try:
         text = data.decode('utf-8', errors='ignore')
     except Exception:
         return False
+    
     sym = (symbology or '').upper()
-    if sym in ("EAN13","EAN-13","EAN8","EAN-8","UPCA","UPC-A"):
-        return ean_like_checksum_ok(re.sub(r"\D", "", text))
-    if sym in ("QRCODE","QRCODEMODEL2","QR-CODE"):
+    
+    if sym in ("EAN13", "EAN-13", "EAN8", "EAN-8", "UPCA", "UPC-A"):
+        # Extract digits only
+        digits = re.sub(r"\D", "", text)
+        
+        # Handle UPC-A reported as EAN-13 with leading 0
+        if sym in ("EAN13", "EAN-13") and len(digits) == 13 and digits.startswith('0'):
+            # Remove leading 0 to get UPC-A format
+            digits = digits[1:]
+        
+        return ean_like_checksum_ok(digits)
+    
+    if sym in ("QRCODE", "QRCODEMODEL2", "QR-CODE"):
         return len(text) > 0
+    
+    # Default validation for other symbologies
     return len(text) > 0
 
 def boxes_from_rect(x: int, y: int, w: int, h: int) -> Box:
     return Box(y, x, y + h, x + w, w * h)
 
 def decode_with_variants(img: Image.Image):
+    """Robust barcode/QR code detection with multiple variants"""
     if not HAS_BARCODE:
         return []
-        results = []
+    
+    results = []
+    
     def do_decode(pil_img):
         try:
             dec = zbar_decode(pil_img)
-            if dec: results.extend(dec)
+            if dec: 
+                results.extend(dec)
         except Exception:
             pass
+    
+    # Try original image
     do_decode(img)
-    if not results:      do_decode(img.convert('L'))
-    if not results:      do_decode(img.resize((img.width*2, img.height*2), Image.BICUBIC))
+    
+    # Try grayscale conversion
+    if not results:
+        do_decode(img.convert('L'))
+    
+    # Try scaling up (better for small barcodes)
+    if not results:
+        try:
+            # Use LANCZOS for better quality in newer Pillow versions
+            scaled_img = img.resize((img.width*2, img.height*2), Image.LANCZOS)
+            do_decode(scaled_img)
+        except AttributeError:
+            # Fallback for older Pillow versions
+            scaled_img = img.resize((img.width*2, img.height*2), Image.BICUBIC)
+            do_decode(scaled_img)
+    
+    # Try different color modes
     if not results and img.mode != 'RGB':
         do_decode(img.convert('RGB'))
+    
+    # Try rotated versions (common for QR codes)
+    if not results:
+        for angle in [90, 180, 270]:
+            try:
+                rotated = img.rotate(angle, expand=True)
+                do_decode(rotated)
+                if results:
+                    break
+            except Exception:
+                continue
+    
     return results
     
 def extract_pdf_images(pdf_path: str, max_pages: int = 5, dpi: int = 300) -> List[Image.Image]:
@@ -688,8 +735,9 @@ def find_barcode_boxes_and_info_from_pdf(
                     width = int(rect.width * scale_x)
                     height = int(rect.height * scale_y)
                 else:
+                    # Use PDF coordinates directly without arbitrary offset
                     left = rect.left
-                    top = rect.top + (page_num * 1000)
+                    top = rect.top
                     width = rect.width
                     height = rect.height