Logged In: YES
user_id=33168

The attached patch fixes the problem and all tests pass on
Linux. But I'm not really sure if this should be fixed or
not. Perhaps someone more familiar with unicode and
filenames (like Martin or Marc-Andre?) could provide
feedback. I don't know if this could create any problems on
Windows.

Changing to a patch.

Logged In: YES
user_id=12579

I wrote a slightly different patch that converts the
os.curdir to Unicode inside glob but the patch here is just
as good. The nnorwitz patch works well for me on Windows
with Unicode file names:

>>> glob.glob("*")
['a.bat', 'abc', 'ascii', 'b.bat', 'fileobject.c',
'fileobject.c.diff', 'Gr\xfc\xdf-Gott', 'pep-0277.txt',
'posixmodule.c', 'posixmodule.c.diff', 'uni.py',
'winunichanges.zip', 'Ge??-sa?', '????????????', '??????',
'???', '????G\xdf', '???']

>>> glob.glob(u"*")
[u'a.bat', u'abc', u'ascii', u'b.bat', u'fileobject.c',
u'fileobject.c.diff', u'Gr\xfc\xdf-Gott', u'pep-0277.txt',
u'posixmodule.c', u'posixmodule.c.diff', u'uni.py',
u'winunichanges.zip',
u'\u0393\u03b5\u03b9\u03ac-\u03c3\u03b1\u03c2',
u'\u0417\u0434\u0440\u0430\u0432\u0441\u0442\u0432\u0443\u0439\u0442\u0435',
u'\u05d4\u05e9\u05e7\u05e6\u05e5\u05e1',
u'\u306b\u307d\u3093',
u'\u66e8\u05e9\u3093\u0434\u0393\xdf', u'\u66e8\u66e9\u66eb']

Here is my patch if you are interested:

--- glob.py     Wed Jun 06 06:24:38 2001
+++ g:\Python23\Lib\glob.py     Sun Aug 01 23:50:43 2004
@@ -19,7 +19,10 @@
             return []
     dirname, basename = os.path.split(pathname)
     if not dirname:
-        return glob1(os.curdir, basename)
+        # Use the current directory but match the argument
+        # string form, either unicode or string.
+        dirname = type(dirname)(os.curdir)
+        return glob1(dirname, basename)
     elif has_magic(dirname):
         list = glob(dirname)
     else:
@@ -40,7 +43,7 @@
     return result

 def glob1(dirname, pattern):
-    if not dirname: dirname = os.curdir
+    assert dirname
     try:
         names = os.listdir(dirname)
     except os.error:

Logged In: YES
user_id=21627

The patch in itself is fine - it is policy that file name
functions return Unicode strings for unicode arguments.

The test is flawed, though: there is no guarantee that the
string and the Unicode result compare equal (see
nyamatongwe's message) - there isn't even a guarantee that
they compare without raising an exception.

Also, there is currently no guarantee that
unicode-in-unicode-out works on all platforms. For example,
on Windows 95, this hasn't been implemented (and neither on
OS/2). So I would:

a) drop the test checking for equality
b) consider returns-not-unicode a failure only if os.listdir
does return unicode for unicode arguments.

Shouldn't the conversion to Unicode use the filesystem encoding?

gbrandl: you are right, it should.

Committed corrected patch as rev. 54197, 54198 (2.5).

2.5.1 (r251:54863, Jul 10 2008, 17:24:48)

Fails for me with 2.5.1 on Linux, OS X, and Windows.

>>> glob.glob("*")
['t.txt', 't\xd0\xb4.txt', 't\xe2\xbd\x94.txt']
>>> glob.glob(u"*")
['t.txt', 't\xd0\xb4.txt', 't\xe2\xbd\x94.txt']
>>> glob.glob(u"./*")
[u'./t.txt', u'./t\u0434.txt', u'./t\u2f54.txt']