UnicodeDecodeError from bzr version when log file has non-ascii path

[Mark Hammond]

consider the following on Windows:

c:\>set BZR_HOME=\temp\test_©har
[do some bzr stuff...]
c:\>dir %BZR_HOME%
 Directory of O:\temp\test_©har
...
01/01/2009 07:57 PM 5,233 .bzr.log
c:\>bzr version
...
  File "O:\src\bazaar\bzr\bzr.work\bzrlib\version.py", line 78, in show_version
    to_file.write(trace._bzr_log_filename + '\n')
  File "o:\src\python-2.5-svn\lib\codecs.py", line 303, in write
    data, consumed = self.encode(object, self.errors)
  File "o:\src\python-2.5-svn\lib\encodings\cp850.py", line 12, in encode
    return codecs.charmap_encode(input,errors,encoding_map)
UnicodeDecodeError: 'ascii' codec can't decode byte 0xa9 in position 11: ordinal not in range(128)

[Mark Hammond (mhammond)]

This is fixed with:

=== modified file 'bzrlib/version.py'
--- bzrlib/version.py 2008-04-02 00:14:00 +0000
+++ bzrlib/version.py 2009-01-01 09:10:31 +0000
@@ -75,7 +75,10 @@
             config_dir = config_dir.decode(osutils.get_user_encoding())
         to_file.write(" Bazaar configuration: %s\n" % (config_dir,))
         to_file.write(" Bazaar log file: ")
- to_file.write(trace._bzr_log_filename + '\n')
+ log_fn = trace._bzr_log_filename
+ if not isinstance(log_fn, unicode):
+ log_fn = log_fn.decode(sys.getfilesystemencoding())
+ to_file.write(log_fn + '\n')
     if show_copyright:
         to_file.write('\n')
         to_file.write(bzrlib.__copyright__ + '\n')

[Robert Collins (lifeless)]

On Thu, 2009-01-01 at 09:12 +0000, Mark Hammond wrote:
> This is fixed with:

Looks sensible to me; do you think its possible to write a test case for
it?

-Rob
--
GPG key available at: <http://www.robertcollins.net/keys.txt>.

[Mark Hammond (mhammond)]

> Looks sensible to me; do you think its possible to write a test case for
> it?

I expect so, given how simple the repro is. I'd expect it might also be possible to reproduce on Linux, so long as the bytes fetched from os.environ() aren't encodable in the stdout encoding. That might be a better first step, so any tests and failure cases can be sure to be portable.

[Vincent Ladeuil (vila)]

marking 'in progress' but leaving the unassigned because a patch is available in the comments (but a test is lacking).