from_n3 erroneously unescapes \xhh

[joernhees]

Exploring #546 made me realize that from_n3 is more liberal than the following method based on the full n3 parser (should just be used for testing!), which is obviously much slower (one full graph with store etc. for parsing) and ignores cases such as quoted graphs or lists that from_n3 can actually parse:

In [1]: import rdflib
INFO:rdflib:RDFLib Version: 4.2.2-dev

In [2]: def own_from_n3(s):
   ...:     n3_s = '<foo:s> <foo:p> ' + s.encode('utf-8') + '.\n'
   ...:     g = rdflib.Graph()
   ...:     g.parse(data=n3_s, format='n3')
   ...:     return list(g)[0][2]
   ...:

This is based on #548 already, so the trailing slashes work in both versions:

In [3]: s = '"trailing slash\\\\"'

In [4]: rdflib.util.from_n3(s)
Out[4]: rdflib.term.Literal(u'trailing slash\\')

In [5]: own_from_n3(s)
Out[5]: rdflib.term.Literal(u'trailing slash\\')

from_n3 is at least incorrectly unescaping \\x6f to a u'\xf6' aka u'ö'. The n3 spec explicitly mentions that the \xhh escapes are unused: http://www.w3.org/TeamSubmission/n3/#escaping :

n [6]: s = u'"jörn"'

In [7]: rdflib.util.from_n3(s)
Out[7]: rdflib.term.Literal(u'j\xf6rn')

In [8]: own_from_n3(s)
Out[8]: rdflib.term.Literal(u'j\xf6rn')

In [9]: s = '"j\\xf6rn"'

In [10]: rdflib.util.from_n3(s)
Out[10]: rdflib.term.Literal(u'j\xf6rn')

In [11]: own_from_n3(s)
  File "<string>", line unknown
BadSyntax


In [12]: s = '"j\\u00f6rn"'

In [13]: rdflib.util.from_n3(s)
Out[13]: rdflib.term.Literal(u'j\xf6rn')

In [14]: own_from_n3(s)
Out[14]: rdflib.term.Literal(u'j\xf6rn')

In [15]: s = '"j\\U000000f6rn"'

In [16]: rdflib.util.from_n3(s)
Out[16]: rdflib.term.Literal(u'j\xf6rn')

In [17]: own_from_n3(s)
Out[17]: rdflib.term.Literal(u'j\xf6rn')

In [18]: s = u'"j\\u00f6öörn"'

In [19]: rdflib.util.from_n3(s)
Out[19]: rdflib.term.Literal(u'j\xf6\xf6\xf6rn')

In [20]: own_from_n3(s)
Out[20]: rdflib.term.Literal(u'j\xf6\xf6\xf6rn')

further inconsistencies between from_n3 and the full n3 parser should be investigated ... Also i'd like to introduce some tests for this as well...

[joernhees]

I also guess this goes all the way back to things mentioned in #192