RDF/XML behaves weird if RDF is the default namespace

[joernhees]

Noticed this while trying to parse mozilla addon install.rdf files. They use xmlns="http://www.w3.org/1999/02/22-rdf-syntax-ns#" so they don't have to prefix things like <Description> or about with <rdf:Description> or rdf:about.

Parsing such a file the current rdflib will not understand the rdf basics anymore and introduce BNodes to capture the xml.

Examples:

Here we define the rdf namespace as commonly done:

In [64]: my_data = '''
<rdf:RDF
  xmlns:rdf='http://www.w3.org/1999/02/22-rdf-syntax-ns#'
  xmlns:rdfs='http://www.w3.org/2000/01/rdf-schema#'
>
  <rdf:Description rdf:about="urn:mozilla:install-manifest">
    <rdfs:label>Example</rdfs:label>
    <rdfs:comment>This is really just an example.</rdfs:comment>
  </rdf:Description>
</rdf:RDF>
'''

In [65]: g = rdflib.Graph()

In [66]: g.parse(data=my_data)
Out[66]: <Graph identifier=N84e0685a0c7c4dd2be6a59ad9cbadac9 (<class 'rdflib.graph.Graph'>)>

In [67]: list(g)
Out[67]:
[(rdflib.term.URIRef(u'urn:mozilla:install-manifest'),
  rdflib.term.URIRef(u'http://www.w3.org/2000/01/rdf-schema#label'),
  rdflib.term.Literal(u'Example')),
 (rdflib.term.URIRef(u'urn:mozilla:install-manifest'),
  rdflib.term.URIRef(u'http://www.w3.org/2000/01/rdf-schema#comment'),
  rdflib.term.Literal(u'This is really just an example.'))]

this is ok.

let's use something other than rdf, e.g. fdr:

In [76]: my_data = '''
   ....: <fdr:RDF
   ....:   xmlns:fdr='http://www.w3.org/1999/02/22-rdf-syntax-ns#'
   ....:   xmlns:rdfs='http://www.w3.org/2000/01/rdf-schema#'
   ....: >
   ....:   <fdr:Description fdr:about="urn:mozilla:install-manifest">
   ....:     <rdfs:label>Example</rdfs:label>
   ....:     <rdfs:comment>This is really just an example.</rdfs:comment>
   ....:   </fdr:Description>
   ....: </fdr:RDF>
   ....: '''

In [77]: g = rdflib.Graph()

In [78]: g.parse(data=my_data)
Out[78]: <Graph identifier=N877052e9ac414a52841a8157ab590fb2 (<class 'rdflib.graph.Graph'>)>

In [79]: list(g)
Out[79]:
[(rdflib.term.URIRef(u'urn:mozilla:install-manifest'),
  rdflib.term.URIRef(u'http://www.w3.org/2000/01/rdf-schema#label'),
  rdflib.term.Literal(u'Example')),
 (rdflib.term.URIRef(u'urn:mozilla:install-manifest'),
  rdflib.term.URIRef(u'http://www.w3.org/2000/01/rdf-schema#comment'),
  rdflib.term.Literal(u'This is really just an example.'))]

this looks ok as well.

now let's make rdf the default namespace so we don't need to prefix everything with rdf (2nd line):

In [68]: my_data = '''
<RDF
  xmlns='http://www.w3.org/1999/02/22-rdf-syntax-ns#'
  xmlns:rdfs='http://www.w3.org/2000/01/rdf-schema#'
>
  <Description about="urn:mozilla:install-manifest">
    <rdfs:label>Example</rdfs:label>
    <rdfs:comment>This is really just an example.</rdfs:comment>
  </Description>
</RDF>
'''

In [69]: g = rdflib.Graph()

In [70]: g.parse(data=my_data)
Out[70]: <Graph identifier=Nc29cd64640354624874a897aea509b12 (<class 'rdflib.graph.Graph'>)>

In [71]: list(g)
Out[71]:
[(rdflib.term.BNode('N6da6e3fa36bc4a898e3ef99e67e29cc7'),
  rdflib.term.URIRef(u'about'),
  rdflib.term.Literal(u'urn:mozilla:install-manifest')),
 (rdflib.term.BNode('N6da6e3fa36bc4a898e3ef99e67e29cc7'),
  rdflib.term.URIRef(u'http://www.w3.org/2000/01/rdf-schema#label'),
  rdflib.term.Literal(u'Example')),
 (rdflib.term.BNode('N6da6e3fa36bc4a898e3ef99e67e29cc7'),
  rdflib.term.URIRef(u'http://www.w3.org/2000/01/rdf-schema#comment'),
  rdflib.term.Literal(u'This is really just an example.'))]

this looks as if rdflib doesn't understand rdf anymore, as it introduces a BNode above all :(

[uholzer]

I think you always have to prefix the rdf:about attribute. This is because the parent element of an unprefixed attribute defines its meaning. In fact, an unprefixed attribute is always in the empty namespace.

[kmaglione]

The RDF spec disagrees:

http://www.w3.org/TR/REC-rdf-syntax/#eventterm-attribute-URI

For attributes without namespaces, "if local-name is ID, about, resource, parseType or type, set to a string value of the concatenation of the RDF namespace URI reference and the value of the local-name accessor".

The parsers/rdfxml.py plugin has code to handle this case, in its convert method. It's broken, however, because it converts the attribute name to a URIRef too early, which breaks the lookups which rely on it being a string. The following change produces the correct behavior:

--- rdfxml.py   2015-03-04 12:02:02.555737327 -0800
+++ -   2015-03-05 14:16:19.582665301 -0800
@@ -235,9 +235,9 @@
         atts = {}
         for (n, v) in attrs.items(): #attrs._attrs.iteritems(): #
             if n[0] is None:
-                att = URIRef(n[1])
+                att = n[1]
             else:
-                att = URIRef("".join(n))
+                att = "".join(n)
             if att.startswith(XMLNS) or att[0:3].lower()=="xml":
                 pass
             elif att in UNQUALIFIED:

[joernhees]

hmm, several thoughts on this:

here's an example install.rdf:
https://github.com/protz/GMail-Conversation-View/blob/master/install.rdf

let's put it through the W3C RDF validator:
http://www.w3.org/RDF/Validator/rdfval?URI=https%3A%2F%2Fraw.githubusercontent.com%2Fprotz%2FGMail-Conversation-View%2Fmaster%2Finstall.rdf&PARSE=Parse+URI%3A+&TRIPLES_AND_GRAPH=PRINT_TRIPLES&FORMAT=PNG_EMBED

Error: {W102} unqualified use of rdf:about is deprecated.[Line = 7, Column = 53]

Also looking at its output it becomes quite apparent that it does not interpret the install.rdf correctly (similar to rdflib it inserts a blank node above and doesn't seem to understand rdf:Description, rdf:about, etc.) @kmaglione, @magopian. So i'd really recommend to change the default install.rdf

There are several reasons we should still deal with this:

1. it seems that jena's riot can handle it (with a deprecation warning though):

   $ riot install.rdf
   14:33:56 WARN  riot                 :: [line: 7, col: 53] {W102} unqualified use of rdf:about is deprecated.
   <urn:mozilla:install-manifest> <http://www.mozilla.org/2004/em-rdf#id> "gconversation@xulforum.org" .
   <urn:mozilla:install-manifest> <http://www.mozilla.org/2004/em-rdf#version> "2.9pre" .
   <urn:mozilla:install-manifest> <http://www.mozilla.org/2004/em-rdf#targetApplication> _:BX2D4e60610dX3A14bef4addc9X3AX2D7fff .
   

2. @kmaglione's link above can actually also be found in a section of our own rdfxml parser:
   https://github.com/RDFLib/rdflib/blob/master/rdflib/plugins/parsers/rdfxml.py#L22
   To me this seems as if we wanted to have handling for this, but as he explains the lookup in convert fails as it tries to find URIRef('about') in a dict which only contains 'about'.

3. Shockingly @kmaglione's fix (see fix RDF/XML problem with unqualified use of rdf:about #470) breaks our prominent doctest example in https://github.com/RDFLib/rdflib/blob/master/rdflib/__init__.py#L21 . Interestingly his fix is right and the test is just wrong, as there isn't 9 but only 4 statements in http://www.w3.org/2000/10/swap/test/meet/blue.rdf . Seems as if for a very long time rdflib wasn't actually able to correctly parse http://www.w3.org/2000/10/swap/test/meet/blue.rdf :(

[kmaglione]

So i'd really recommend to change the default install.rdf

Unfortunately, there isn't such a thing. People copy install.rdf templates from all over the place. They mostly come from other add-ons, or are generated by frameworks.

[uholzer]

@kmaglione: Thanks for clearing up and fixing the bug.

The W3C validator is rather strict to reject the install.rdf, however, validators are supposed to be strict. Also, I found the same problem in the install.rdf of a Firefox extension I wrote myself. So yes, people copy templates from all over the place, for example (sarcasm) from the Mozilla Developer Network.