From 559d42cd4aed08333145c776878c7134bba2acf9 Mon Sep 17 00:00:00 2001
From: Johannes 'josch' Schauer <josch@mister-muffin.de>
Date: Wed, 29 Apr 2020 16:15:55 +0200
Subject: [PATCH] magick.py: use our own deflate algo for bit-by-bit
 reproducible output independent of the compressor used

---
 magick.py | 47 ++++++++++++++++++++++++++++++++++++++++++++++-
 test.sh   | 26 +++++++++++++-------------
 2 files changed, 59 insertions(+), 14 deletions(-)

diff --git a/magick.py b/magick.py
index 8299cf1..72a43e3 100755
--- a/magick.py
+++ b/magick.py
@@ -71,6 +71,51 @@ def palettize(img, pal):
     return result
 
 
+# we cannot use zlib.compress() because different compressors may compress the
+# same data differently, for example by using different optimizations on
+# different architectures:
+# https://lists.fedoraproject.org/archives/list/devel@lists.fedoraproject.org/thread/R7GD4L5Z6HELCDAL2RDESWR2F3ZXHWVX/
+#
+# to make the compressed representation of the uncompressed data bit-by-bit
+# identical on all platforms we make use of the compression method 0, that is,
+# no compression at all :)
+def compress(data):
+    # two-byte zlib header (rfc1950)
+    # common header for lowest compression level
+    # bits 0-3: Compression info, base-2 logarithm of the LZ77 window size,
+    #           minus eight -- 7 indicates a 32K window size
+    # bits 4-7: Compression method -- 8 is deflate
+    # bits 8-9: Compression level -- 0 is fastest
+    # bit 10:   preset dictionary -- 0 is none
+    # bits 11-15: check bits so that the 16-bit unsigned integer stored in MSB
+    #             order is a multiple of 31
+    result = b"\x78\x01"
+    # content is stored in deflate format (rfc1951)
+    # maximum chunk size is the largest 16 bit unsigned integer
+    chunksize = 0xFFFF
+    for i in range(0, len(data), chunksize):
+        # bits 0-4 are unused
+        # bits 5-6 indicate compression method -- 0 is no compression
+        # bit 7 indicates the last chunk
+        if i * chunksize < len(data) - chunksize:
+            result += b"\x00"
+        else:
+            # last chunck
+            result += b"\x01"
+        chunk = data[i : i + chunksize]
+        # the chunk length as little endian 16 bit unsigned integer
+        result += struct.pack("<H", len(chunk))
+        # the one's complement of the chunk length
+        # one's complement is all bits inverted which is the result of
+        # xor with 0xffff for a 16 bit unsigned integer
+        result += struct.pack("<H", len(chunk) ^ 0xFFFF)
+        result += chunk
+    # adler32 checksum of the uncompressed data as big endian 32 bit unsigned
+    # integer
+    result += struct.pack(">I", zlib.adler32(data))
+    return result
+
+
 def write_png(data, path, bitdepth, colortype, palette=None):
     with open(path, "wb") as f:
         f.write(b"\x89PNG\r\n\x1A\n")
@@ -124,7 +169,7 @@ def write_png(data, path, bitdepth, colortype, palette=None):
                     raw += struct.pack(">B", val)
             else:
                 raise Exception()
-        compressed = zlib.compress(raw)
+        compressed = compress(raw)
         block = b"IDAT" + compressed
         f.write(
             struct.pack(">I", len(compressed))
diff --git a/test.sh b/test.sh
index f4d50e5..88ec374 100755
--- a/test.sh
+++ b/test.sh
@@ -97,19 +97,19 @@ trap error EXIT
 python3 magick.py "$tempdir"
 
 cat << END | ( cd "$tempdir"; md5sum --check --status - )
-7ed200c092c726c68e889514fff0d8f1  alpha.png
-bf56e00465b98fb738f6edd2c58dac3b  gray16.png
-f93c3e3c11dad3f8c11db4fd2d01c2cc  gray1.png
-d63167c66e8a65bd5c15f68c8d554c48  gray2.png
-6bceb845d9c9946adad1526954973945  gray4.png
-7ba8152b9146eb7d9d50529189baf7db  gray8.png
-75130ec7635919e40c7396d45899ddbe  inverse.png
-bd1a2c9f9dfc51a827eafa3877cf7f83  normal16.png
-329eac79fec2e1bc30d7a50ba2e3f2a5  normal.png
-9ffd3f592b399f9f9c23892db82369cd  palette1.png
-6d3a39fe5f2efea5975f048d11a5cb02  palette2.png
-57add39e5c278249b64ab23314a41c39  palette4.png
-192a3b298d812a156e6f248238d2bb52  palette8.png
+cc611e80cde3b9b7adb7723801a4e5d4  alpha.png
+706175887af8ca1a33cfd93449f970df  gray16.png
+198e3333d7dc6158f94d1007e75c5bd3  gray1.png
+795ff9c5c59cdc95f5a0b9191a7247d9  gray2.png
+722223ba74be9cba1af4a549076b70d3  gray4.png
+2320216faa5a10bf0f5f04ebce07f8e1  gray8.png
+35a47d6ae6de8c9d0b31aa0cda8648f3  inverse.png
+6ad810399058a87d8145d8d9a7734da5  normal16.png
+c8d2e1f116f31ecdeae050524efca7b6  normal.png
+c69edd665ee3546d9eb57fe10cd8095a  palette1.png
+504ecab83662aa2394042a62b1d2eafb  palette2.png
+90d1bbe7b26e210d8846cf8692a98b72  palette4.png
+50bf09eb3571901f0bf642b9a733038c  palette8.png
 END
 
 # use img2pdfprog environment variable if it is set