When exporting a document containing non-ASCII characters (for example in the description field), the XML export done through nuxeo-core-io is done with the header
<?xml version="1.0" encoding="UTF-8"?>
but the caracters contained are actually using the latin-1 charset.
This makes the XML file illegal and prevents any further reimport.
This was tested through nuxeo-shell.
The restlet export (exportTree?format=XML) also uses incorrect charset but of a different nature, an export for "café" gives "caf?\u00a9" (with an actual question mark).