Graphs don't serialize new content correctly

[jeroenh]

Consider the following:

>>> g = rdflib.Graph()
>>> g.add((u"#foo",u"#bar",u"#foobar"))
>>> g.serialize()
'<?xml version="1.0" encoding="UTF-8"?>\n<rdf:RDF\n   xmlns:ns1="#"\n   xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#"\n>\n</rdf:RDF>\n'

My new triplet is not there.

Ah you say, but you should commit your graph! But that also does not work:

>>> g = rdflib.Graph()
>>> g.add((u"#foo",u"#bar",u"#foobar"))
>>> g.commit()
>>> g.serialize()
'<?xml version="1.0" encoding="UTF-8"?>\n<rdf:RDF\n   xmlns:ns1="#"\n   xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#"\n>\n</rdf:RDF>\n'

What is going on here? Note that in both cases the Namespace does get added to the graph, so clearly something is happening.

[wikier]

Wrong usage of the API, subjects can not be plain text variables.

Therefore you would need to use something like:

>>> g = rdflib.Graph()
>>> g.add((rdflib.URIRef(u"#foo"), u"#bar", u"#foobar")) 
>>> g.serialize(base="http://example.org")

Maybe we would need to improve both the documentation and the feedback provided to the user on such situations...

[jeroenh]

Okay, it makes sense to combine this with #227 also and add some more sensible valuechecking in the add method. Not only should it reject plain strings, it should also reject triplets which have the subject and predicate as unicode objects.

[gromgull]

This is the same underlying issue as #200 again.

@wikier - actually no part of the triple (or quad, see #167) should be a string, it may work sometimes, but will come back and bit you sooner or later. "unicode object has no .n3()" method for instance...

We could "auto-box" non rdflib.term.Node objects though - but only in rdflib4, where a literal/python mapping overhaul is planned anyway.

[gromgull]

I made #230 for the possibility of auto-boxing objects.