According to https://en.wikipedia.org/wiki/Bzip2 the reference implementation of bzip2 writes an end-of-stream marker (also called stream footer) with a magic number and a stream checksum to the file.

Python does not do so. The files can still be read by all bzip2 compatible software I tried. For completeness and better error detection however, writing this marker (optionally maybe) would be useful.

Thank you, I have added Serhiy and Thomas for this issue but I have removed the versions 3.7, 3.6, 3.5 and 2.7.

Maybe it's a new feature for the next versions of Python.

The bz2 library and the bzip2 utility generate identical files. For example:

>>> import bz2
>>> with bz2.open('bz2test.txt.bz2', mode='wb') as f:
...     f.write(b'foobar')
... 
6

generates this file:

$ hexdump bz2test.txt.bz2
0000000 42 5a 68 39 31 41 59 26 53 59 52 c0 3d c1 00 00
0000010 01 01 80 31 00 90 00 20 00 21 83 41 9a 09 88 1c
0000020 5d c9 14 e1 42 41 4b 00 f7 04                  
000002a

and

$ echo -n 'foobar' | bzip2 -c > bzip2test.txt.bz2

generates this file:

$ hexdump bzip2test.txt.bz2 
0000000 42 5a 68 39 31 41 59 26 53 59 52 c0 3d c1 00 00
0000010 01 01 80 31 00 90 00 20 00 21 83 41 9a 09 88 1c
0000020 5d c9 14 e1 42 41 4b 00 f7 04                  
000002a

The StreamFooter is there, it is just difficult to see because it isn't byte-aligned. It starts at byte 0x20 bit 6.  If you take 0x5dc914e14241 and shift it right two bits, you get the StreamFooter FooterMagic value: 0x177245385090. It is followed by a 32-bit CRC and two zero-bits of padding to byte-align the end of the file.

Sorry, the StreamFooter starts at byte 0x1f bit 1, not 0x20 bit 6.

Sorry, I forgot about the bit alignment when I was checking for the footer. This can be closed then.