From 71fbda8bf68fffa6c1d6c4cb7bb6f86f20a1ab9c Mon Sep 17 00:00:00 2001
From: "James R. Barlow" <jim@purplerock.ca>
Date: Sat, 20 Feb 2016 01:20:44 -0800
Subject: [PATCH] Adjust page orientation parsing to deal with change in Tess
 3.04.01

---
 ocrmypdf/main.py      | 17 +++++++++--------
 ocrmypdf/tesseract.py | 14 +++++++++++++-
 2 files changed, 22 insertions(+), 9 deletions(-)

diff --git a/ocrmypdf/main.py b/ocrmypdf/main.py
index 2dd53ccf..a1121261 100755
--- a/ocrmypdf/main.py
+++ b/ocrmypdf/main.py
@@ -524,9 +524,9 @@ def orient_page(
 
     direction = {
         0: '⇧',
-        90: '⇦',
+        90: '⇨',
         180: '⇩',
-        270: '⇨'
+        270: '⇦'
     }
 
     log.info(
@@ -544,8 +544,8 @@ def orient_page(
         reader = pypdf.PdfFileReader(page_pdf)
         page = reader.pages[0]
 
-        # Rotate opposite of orientation
-        rotated_page = page.rotateClockwise(orient_conf.angle)
+        # angle is a clockwise angle, so rotating ccw will correct the error
+        rotated_page = page.rotateCounterClockwise(orient_conf.angle)
         writer.addPage(rotated_page)
         with open(output_file, 'wb') as out:
             writer.write(out)
@@ -788,11 +788,12 @@ def add_text_layer(
 
     page_text = pdf_text.getPage(0)
 
-    # The text page always will be oriented up
+    # The text page always will be oriented up by this stage
     # but if lossless_reconstruction, pdf_image may have a rotation applied
-    # we can't just merge the pages, because a page can only have one /Rotate
-    # tag, so the differential rotation must be corrected.
-    # Also, pdf_image may not have its mediabox nailed to (0, 0)
+    # We have to eliminate the /Rotate tag (because it applies to the whole
+    # page) and rotate the image layer to match the text page
+    # Also, pdf_image may not have its mediabox nailed to (0, 0), so may need
+    # translation
     page_image = pdf_image.getPage(0)
     rotation = page_image.get('/Rotate', 0)
 
diff --git a/ocrmypdf/tesseract.py b/ocrmypdf/tesseract.py
index 0a205621..ab1a417b 100644
--- a/ocrmypdf/tesseract.py
+++ b/ocrmypdf/tesseract.py
@@ -107,8 +107,20 @@ def get_orientation(input_file, language: list, timeout: float, log):
             if len(parts) == 2:
                 osd[parts[0].strip()] = parts[1].strip()
 
+        angle = int(osd.get('Orientation in degrees', 0))
+        if 'Orientation' in osd:
+            # Tesseract < 3.04.01
+            # reports "Orientation in degrees" as a counterclockwise angle
+            # We keep it clockwise
+            assert 'Rotate' not in osd
+            angle = -angle % 360
+        else:
+            # Tesseract == 3.04.01, hopefully also Tesseract > 3.04.01
+            # reports "Orientation in degrees" as a clockwise angle
+            assert 'Rotate' in osd
+
         oc = OrientationConfidence(
-            angle=int(osd.get('Orientation in degrees', 0)),
+            angle=angle,
             confidence=float(osd.get('Orientation confidence', 0)))
         return oc