[MERGE] UTF-8 encoding in binary diffs

[Aaron Bentley]

Aaron Bentley has voted +0.
Status is now: Waiting
Comment:
Thanks.  I'm surprised your test case uses utf-8 for the filenames. 
That could conceivably break on non-utf8 filesystems.  Better to use 
unicode strings, I think.  But then you do need to make sure the test is 
skipped on non-unicode filesystems.

cStringIO doesn't trigger the bug because of bugs in the unicode 
handling of cStringIO.

For details, see: 
http://bundlebuggy.aaronbentley.com/request/%3Cd06a5cd30707060557u7b55fbe7s1b3bb839d45185f9%40mail.gmail.com%3E