Turtle parser accepts literal subjects, which is invalid RDF

[miguelvaara]

While running a script including this kind of loop,

for s, p, o in g.triples((None, None, skos.Concept)):
  some code

..an accidentally added literal as a subject is surprisingly allowed (originally from the loaded turtle formatted file). None as a subject can read types (in this case literals) outside the standard. An assumption is that it is the parser that allows non-standard subject types.

Python: 3.6.9
RDFLib: 5.0.0

[osma]

To clarify: The problem is with the Turtle parser, not the .triples method (where it surfaced). The parser seems to happily accept a literal in the subject position, even though this is not allowed in RDF.

Contents of test.ttl:

"example" <http://example.org/p> <http://example.org/o> .

Code:

>>> import rdflib
RDFLib Version: 5.0.0
>>> g = rdflib.Graph()
>>> g.parse('test.ttl', format='turtle')
<Graph identifier=Nec6e885a6e8e4d36bc35621f099c2b49 (<class 'rdflib.graph.Graph'>)>
>>> for triple in g.triples((None, None, None)):
...     print(triple)
... 
(rdflib.term.Literal('example'), rdflib.term.URIRef('http://example.org/p'), rdflib.term.URIRef('http://example.org/o'))

The N-Triples parser fails on the same file.

[osma]

Related to #895 - it appears that rdflib is not very strict about RDF and in some cases (but not always) supports generalized RDF datasets where literals can be in the subject position.