os.lisdir, gets unicode, returns unicode... USUALLY?!?!?

[gabor]

hi,

from the documentation (http://docs.python.org/lib/os-file-dir.html) for
os.listdir:

"On Windows NT/2k/XP and Unix, if path is a Unicode object, the result
will be a list of Unicode objects."

i'm on Unix. (linux, ubuntu edgy)

so it seems that it does not always return unicode filenames.

it seems that it tries to interpret the filenames using the filesystem's
encoding, and if that fails, it simply returns the filename as byte-string.

so you get back let's say an array of 21 filenames, from which 3 are
byte-strings, and the rest unicode strings.

after digging around, i found this in the source code:

#ifdef Py_USING_UNICODE
if (arg_is_unicode) {
PyObject *w;

w = PyUnicode_FromEncodedObject(v,
Py_FileSystemDefaultEncoding,
"strict");
if (w != NULL) {
Py_DECREF(v);
v = w;
}
else {
/* fall back to the original byte string, as
discussed in patch #683592 */
PyErr_Clear();
}
}
#endif

so if the to-unicode-conversion fails, it falls back to the original
byte-string. i went and have read the patch-discussion.

and now i'm not sure what to do.
i know that:

1. the documentation is completely wrong. it does not always return
unicode filenames
2. it's true that the documentation does not specify what happens if the
filename is not in the filesystem-encoding, but i simply expected that i
get an Unicode-exception, as everywhere else. you see, exceptions are
ok, i can deal with them. but this is just plain wrong. from now on,
EVERYWHERE where i use os.listdir, i will have to go through all the
filenames in it, and check if they are unicode-strings or not.

so basically i'd like to ask here: am i reading something incorrectly?
or am i using os.listdir the "wrong way"? how do other people deal with
this?

p.s: one additional note. if you code expects os.listdir to return
unicode, that usually means that all your code uses unicode strings.
which in turn means, that those filenames will somehow later interact
with unicode strings. which means that that byte-string-filename will
probably get auto-converted to unicode at a later point, and that
auto-conversion will VERY probably fail, because the auto-convert only
happens using 'ascii' as the encoding, and if it was not possible to
decode the filename inside listdir, it's quite probable that it also
will not work using 'ascii' as the charset.


gabor

 

[Terry Reedy]

so if the to-unicode-conversion fails, it falls back to the original
byte-string. i went and have read the patch-discussion.

and now i'm not sure what to do.
i know that:

1. the documentation is completely wrong. it does not always return
unicode filenames

Unless someone says otherwise, report the discrepancy between doc and code
as a bug on the SF tracker. I have no idea of what the resolution should
be ;-).

tjr

 

[=?ISO-8859-1?Q?=22Martin_v=2E_L=F6wis=22?=]

so basically i'd like to ask here: am i reading something incorrectly?

You are reading it correctly. This is how it behaves.

or am i using os.listdir the "wrong way"? how do other people deal with
this?

You didn't say why the behavior causes a problem for you - you only
explained what the behavior is.

Most people use os.listdir in a way like this:

for name in os.listdir(path):
full = os.path.join(path, name)
attrib = os.stat(full)
if some-condition:
f = open(full)
...

All this code will typically work just fine with the current behavior,
so people typically don't see any problem.

Regards,
Martin

 

[gabor]

You didn't say why the behavior causes a problem for you - you only
explained what the behavior is.

Most people use os.listdir in a way like this:

for name in os.listdir(path):
full = os.path.join(path, name)
attrib = os.stat(full)
if some-condition:
f = open(full)
...

All this code will typically work just fine with the current behavior,
so people typically don't see any problem.

i am sorry, but it will not work. actually this is exactly what i did,
and it did not work. it dies in the os.path.join call, where file_name
is converted into unicode. and python uses 'ascii' as the charset in
such cases. but, because listdir already failed to decode the file_name
with the filesystem-encoding, it usually also fails when tried with 'ascii'.

example:
Traceback (most recent call last):
File "<stdin>", line 1, in ?
File "/usr/lib/python2.4/posixpath.py", line 65, in join
path += '/' + b
UnicodeDecodeError: 'ascii' codec can't decode byte 0xe7 in position 1:
ordinal not in range(128)

gabor

 

[=?ISO-8859-1?Q?=22Martin_v=2E_L=F6wis=22?=]

i am sorry, but it will not work. actually this is exactly what i did,
and it did not work. it dies in the os.path.join call, where file_name
is converted into unicode. and python uses 'ascii' as the charset in
such cases. but, because listdir already failed to decode the file_name
with the filesystem-encoding, it usually also fails when tried with
'ascii'.

Ah, right. So yes, it will typically fail immediately - just as you
wanted it to do, anyway; the advantage with this failure is that you
can also find out what specific file name is causing the problem
(whereas when listdir failed completely, you could not easily find
out the cause of the failure).

How would you propose listdir should behave?

Regards,
Martin

 

[Fredrik Lundh]

get an Unicode-exception, as everywhere else. you see, exceptions are
ok, i can deal with them.

p.s: one additional note. if you code expects os.listdir to return
unicode, that usually means that all your code uses unicode strings.
which in turn means, that those filenames will somehow later interact
with unicode strings. which means that that byte-string-filename will
probably get auto-converted to unicode at a later point, and that
auto-conversion will VERY probably fail

it will raise an exception, most likely. didn't you just say that
exceptions were ok?

</F>

 

[Laurent Pointal]

gabor a écrit :

hi,

from the documentation (http://docs.python.org/lib/os-file-dir.html) for
os.listdir:

"On Windows NT/2k/XP and Unix, if path is a Unicode object, the result
will be a list of Unicode objects."

Maybe, for each filename, you can test if it is an unicode string, and
if not, convert it to unicode using the encoding indicated by
sys.getfilesystemencoding().

Have a try.

A+

Laurent.

 

[Johan von Boisman]

gabor a écrit :

Maybe, for each filename, you can test if it is an unicode string, and
if not, convert it to unicode using the encoding indicated by
sys.getfilesystemencoding().

Have a try.

A+

Laurent.

Strange coincident, as I was wrestling with this problem only yesterday.

I found this most illuminating discussion on the topic with
contributions from Mr Lövis and others:

http://www.thescripts.com/forum/thread41954.html

/johan

 

[gabor]

Laurent Pointal wrote:

gabor a écrit :

Maybe, for each filename, you can test if it is an unicode string, and
if not, convert it to unicode using the encoding indicated by
sys.getfilesystemencoding().

Have a try.

A+

Laurent.

gabor a écrit :

Maybe, for each filename, you can test if it is an unicode string, and
if not, convert it to unicode using the encoding indicated by
sys.getfilesystemencoding().

i don't think it would work. because os.listdir already tried, and
failed (that's why we got a byte-string and not an unicode-string)

gabor

 

[Leo Kislov]

Ah, right. So yes, it will typically fail immediately - just as you
wanted it to do, anyway; the advantage with this failure is that you
can also find out what specific file name is causing the problem
(whereas when listdir failed completely, you could not easily find
out the cause of the failure).

How would you propose listdir should behave?

How about returning two lists, first list contains unicode names, the
second list contains undecodable names:

files, troublesome = os.listdir(separate_errors=True)

and make separate_errors=True by default in python 3.0 ?

-- Leo

 

[gabor]

it will raise an exception, most likely. didn't you just say that
exceptions were ok?

yes, but it's raised at the wrong place imho

(just to clarify: simply pointing out this behavior in the documentation
is also one of the possible solutions)

for me the current behavior seems as if file-reading would work like this:

a = open('foo.txt')
data = a.read()
a.close()

print data

gabor

 

[=?ISO-8859-1?Q?=22Martin_v=2E_L=F6wis=22?=]

How about returning two lists, first list contains unicode names, the
second list contains undecodable names:

files, troublesome = os.listdir(separate_errors=True)

and make separate_errors=True by default in python 3.0 ?

That would be quite an incompatible change, no?

Regards,
Martin

 

[Fredrik Lundh]

That would be quite an incompatible change, no?

it also violates a fundamental design rule for the standard library.

</F>

 

[Leo Kislov]

That would be quite an incompatible change, no?

Yeah, that was idea-dump. Actually it is possible to make this idea
mostly backward compatible by making os.listdir() return only unicode
names and os.binlistdir() return only binary directory entries.
Unfortunately the same trick will not work for getcwd.

Another idea is to map all 256 bytes to unicode private code points.
When a file name cannot be fully decoded the undecoded bytes will be
mapped to specially allocated code points. Unfortunately this idea
seems to leak if the program later wants to write such unicode string
to a file. Python will have to throw an exception since we don't know
if it is ok to write broken string to a file. So we are back to square
one, programs need to deal with filesystem garbage

-- Leo