The RDF/XML writer (commons 0019-mizan-integrationsFR-016): an ontology's triples as RDF/XML 1.1, as text, in one canonical form, so that a tool that reads only RDF/XML (the OWL API, Protege, older triple stores) reads the same graph as the Turtle, and a generated file can be checked for staleness.
The form is flat: one rdf:Description for each subject, in the order of their N-Triples terms, its rdf:type first and its other predicates by IRI, each predicate's objects in the order of their N-Triples terms (the order the kernel's Turtle writer uses), so the same triples give the same bytes whatever the order they were stated in. A predicate's IRI is split into a namespace and an XML name: at a declared prefix where one fits, and otherwise after the last character that an XML name may not hold, the namespace then getting a prefix ns1, ns2 in the order of the namespaces' IRIs. A plain string (an xsd:string) has no rdf:datatype; a string in a language has xml:lang; a blank node is an rdf:nodeID. What XML 1.0 cannot say is a finding, never silently changed: a predicate with no XML name (its IRI ends in a digit or in a character an XML name may not hold) and a control character in a literal.