Fixed in upstream github.com/s3tools/s3cmd
master branch with commit
Author: Matt Domsch <email@example.com
Date: Mon Apr 7 22:48:14 2014 -0500
better MIME magic library handling
The different mime magic libraries expect input to their from_file(),
file(), and id_filename() functions, respectively, to take either a
filesystem-encoded (generally UTF-8) string, or a unicode filename.
Different versions of the libraries each expect different inputs for
filename though. This is annoying.
Here, we call these functions, first with a UTF-8 encoded string
filename. If that fails with a UnicodeDecodeError, we try again
passing a unicode filename.
Also, delete mime_magic_buffer() everywhere, and the introspection of
gzip files to see what type of object is inside. It doesn't matter to
the S3 web server - it needs to be type application/x-gzip, not
type=application/tar encoding=gzip (as mimetypes would tell us). We
stopped using the encoding value here as HTTP Content-Encoding in
commit 44e35899a anyhow.