This is listed in TODO, and the problematic part of xml2po.py is the following snippet in replaceNodeContentsWithText function:

            # Lets add document DTD so entities are resolved
            dtd = doc.intSubset()
            tmp = ''
            if expand_entities: # FIXME: we get a "Segmentation fault" in libxml2.parseMemory() when we include DTD otherwise
                tmp = dtd.serialize('utf-8')
            tmp = tmp + '<%s>%s</%s>' % (starttag, text, endtag)


It seems now this was a problem with libxml2 (yet I suspected it was my code), since it now works for me without any problems on libxml2 2.6.21 (anyone knows when the change happened?).


I am just adding it to Bugzilla since it's a common request, along with a very simple patch that should apply to many different xml2po versions (for those not wishing to recompile g-d-u from CVS).