Unicode encoding for filesystem in Mac OS X not correct in Python?

Question

Having a bit of struggle with Unicode file names in OS X and Python. I am trying to use filenames as input for a regular expression later in the code, but the encoding used in the filenames seem to be different from what sys.getfilesystemencoding() tells me. Take the following code:

#!/usr/bin/env python
# coding=utf-8

import sys,os
print sys.getfilesystemencoding()

p = u'/temp/s/'
s = u'åäö'
print 's', [ord(c) for c in s], s
s2 = s.encode(sys.getfilesystemencoding())
print 's2', [ord(c) for c in s2], s2
os.mkdir(p+s)
for d in os.listdir(p):
  print 'dir', [ord(c) for c in d], d

It outputs the following:

utf-8
s [229, 228, 246] åäö
s2 [195, 165, 195, 164, 195, 182] åäö
dir [97, 778, 97, 776, 111, 776] åäö

So, file system encoding is utf-8, but when I encode my filename åäö using that, it will not be the same as if I create a dir name with the same string. I expect that when I use my string åäö to create a dir, and read it's name back, it should use the same codes as if I applied the encoding directly.

If we look at the code points 97, 778, 97, 776, 111, 776, it's basically ASCII characters with added diacritic, e.g. o + ¨ = ö, which makes it two characters, not one. How can I avoid this discrepancy, is there an encoding scheme in Python that matches this behaviour by OS X, and why is not getfilesystemencoding() giving me the right result?

Or have I messed up?

Answer 1

MacOS X uses a special kind of decomposed UTF-8 to store filenames. If you need to e.g. read in filenames and write them to a "normal" UTF-8 file, you must normalize them :

filename = unicodedata.normalize('NFC', unicode(filename, 'utf-8')).encode('utf-8')

from here: https://web.archive.org/web/20120423075412/http://boodebr.org/main/python/all-about-python-and-unicode

Answer 2

getfilesystemencoding() is giving you the correct response (the encoding), but it does not tell you the unicode normalisation form.

In particular, the HFS+ filesystem uses UTF-8 encoding, and a normalisation form close to "D" (which requires composed characters like ö to be decomposed into o¨). HFS+ is also tied to the normalisation form as it existed in Unicode version 3.2—as detailed in Apple's documentation for the HFS+ format.

Python's unicodedata.normalize method converts between forms, and if you prefix the call with the ucd_3_2_0 object, you can constrain it to Unicode version 3.2:

filename = unicodedata.ucd_3_2_0.normalize('NFC', unicode(filename, 'utf-8')).encode('utf-8')