From d291d4899154ffb710962125e750c576db46aa5d Mon Sep 17 00:00:00 2001
From: "James R. Barlow" <jim@purplerock.ca>
Date: Mon, 16 Apr 2018 12:15:19 -0700
Subject: [PATCH] PNG palette: parse PDF string from leptonica instead

Seems better to accept whatever leptonica rather than make detailed
assumptions about how it encodes the palette.

Experimented with setting FlateDecode on the palette but it seems to
expand it.
---
 src/ocrmypdf/leptonica.py | 12 +++++-------
 src/ocrmypdf/pipeline.py  | 12 +++++++-----
 2 files changed, 12 insertions(+), 12 deletions(-)

diff --git a/src/ocrmypdf/leptonica.py b/src/ocrmypdf/leptonica.py
index 3a8cefe1..d9a48262 100644
--- a/src/ocrmypdf/leptonica.py
+++ b/src/ocrmypdf/leptonica.py
@@ -27,7 +27,6 @@ import logging
 import warnings
 from tempfile import TemporaryFile
 from ctypes.util import find_library
-from binascii import unhexlify
 from .lib._leptonica import ffi
 from functools import lru_cache
 from enum import Enum
@@ -518,12 +517,11 @@ class CompressedData:
             return getattr(self._compdata, name)
         raise AttributeError(name)
 
-    def get_palette(self):
-        buf = ffi.buffer(self._compdata.cmapdatahex, 
-                         self._compdata.ncolors * 7 + 3)
-        hex_str = bytes(buf).decode('ascii')
-        hex_str = hex_str[1:-1].replace(' ', '')
-        return unhexlify(hex_str)
+    def get_palette_pdf_string(self):
+        "Returns palette pre-formatted for use in PDF"
+        buflen = len('< ') + len(' rrggbb') * self._compdata.ncolors + len('>')
+        buf = ffi.buffer(self._compdata.cmapdatahex, buflen)
+        return bytes(buf)
 
     @staticmethod
     def _destroy(compdata):
diff --git a/src/ocrmypdf/pipeline.py b/src/ocrmypdf/pipeline.py
index 2240b945..04bdfd6f 100644
--- a/src/ocrmypdf/pipeline.py
+++ b/src/ocrmypdf/pipeline.py
@@ -1176,10 +1176,12 @@ def optimize_pdf(
         im_obj.Height = compdata.h
 
         if compdata.ncolors > 0:
-            palette_stream = pikepdf.Stream(pike, compdata.get_palette())
+            palette_pdf_string = compdata.get_palette_pdf_string()
+            log.info(palette_pdf_string)
+            palette_data = pikepdf.Object.parse(palette_pdf_string)
+            palette_stream = pikepdf.Stream(pike, bytes(palette_data))
             palette = [pikepdf.Name('/Indexed'), pikepdf.Name('/DeviceRGB'),
                        compdata.ncolors - 1, palette_stream]
-            #cs = pike.make_indirect(palette)
             cs = palette
         else:
             if compdata.spp == 1:
@@ -1192,13 +1194,13 @@ def optimize_pdf(
 
         im_obj.write(compdata.read(), pikepdf.Name('/FlateDecode'), predictor)
 
-        
-
+    # Not object_stream_mode + preserve_pdfa generates noncompliant PDFs
     pike.save(output_file, preserve_pdfa=True)
+
     input_size = Path(input_file).stat().st_size
     output_size = Path(output_file).stat().st_size
     improvement = input_size / output_size
-    log.info("Optimize reduced size by {:.1f}".format(improvement))
+    log.info("Optimize reduced size by {:.3f}".format(improvement))
 
 
 def merge_sidecars(