Incorrect UTF-8-smashing of conflict localisations

When a property conflict is presented, some localised messages will be
mutilated. Example (French locale):

Actualisé à la révision 1.
Conflit sur la propriété 'p' découvert sur 'f'.
Tentative de d'ajout d'une nouvelle propri?\195?\169t?\195?\169 'p'
mais elle existe d?\195?\169j?\195?\160.
<<<<<<< (valeur locale de la propri?\195?\169t?\195?\169)
y=======
x>>>>>>> (valeur ext?\195?\169rieure de la propri?\195?\169t?\195?\169)

This happens because append_prop_conflict calls
svn_utf_cstring_from_utf8_fuzzy on the whole string including the
preamble and the conflict delimiters, converting any non-ASCII bytes
to ?\ddd.

Sorry about not having a patch ready, but I'm not sure about what was
intended. Should the property values always be converted to ASCII in
that way, just the bytes that are not well-formed UTF-8, or the entire
property value iff it is not entirely UTF-8?
Received on 2013-04-14 14:21:34 CEST