Comment 6 for bug 659763

From the code point of view, the user name is decoded with the so-called user_encoding which should apply to the user *input*, so arguably we shouldn't use it for a file content... even if this sounds like the best bet in most of the cases.

One workaround that doesn't require any bzr change is to use user names that can be decoded with the declared encoding which is: 'ANSI_X3.4-1968'. I.e. ascii.

Note that you're using bzr-2.0.3 and that most recent versions establish the user name in a slightly different way (and it's unclear to me how these versions will behave here :-/).