Bug in longturtle serialization

[mschiedon]

The longturtle serializer fails to emit a whitespace separator between a predicate and a list of objects if one of these objects is a blank node (and the blank node cannot be 'inlined', i.e. is used more than once). The problem can be reproduced using this Python code:

from rdflib import Graph

input = '''\
@prefix ex: <https://example.org/> .

ex:1 a ex:Thing ;
    ex:relatedTo ex:3, _:bnode0 .

ex:2 a ex:Thing ;
    ex:relatedTo _:bnode0 .

_:bnode0 a ex:Thing .
'''

graph = Graph().parse(data=input, format='turtle')
output = graph.serialize(format='longturtle')
print(output)
assert output.find('relatedTo_:') == -1, \
    'Missing whitespace separation between predicate' \
    ' and the first blank node of a list of objects.'

The resulting Turtle with the bug looks like below. Note the missing space between the predicate ex:relatedTo and blank node _:n40fef3a41a034be9a7116df126afd613b1 for the ex:1 case. The ex:2 case does correctly use a space separator when serializing because it's a single object and not a list.

PREFIX ex: <https://example.org/>

ex:1
    a ex:Thing ;
    ex:relatedTo_:n40fef3a41a034be9a7116df126afd613b1 ,
        ex:3 ;
.

ex:2
    a ex:Thing ;
    ex:relatedTo _:n40fef3a41a034be9a7116df126afd613b1 ;
.

_:n40fef3a41a034be9a7116df126afd613b1
    a ex:Thing ;
.

I believe the issue might be solved by adding an additional indent in the longturtle.py source code on this line, as shown in the code below.

    def objectList(self, objects):
        count = len(objects)
        if count == 0:
            return
        depthmod = (count == 1) and 0 or 1
        self.depth += depthmod
        first_nl = False
        if count > 1:
            if not isinstance(objects[0], BNode):
                self.write("\n" + self.indent(1))
                # BUG: Gave below line an extra indent.
                first_nl = True
        self.path(objects[0], OBJECT, newline=first_nl)
        for obj in objects[1:]:
            self.write(" ,")
            if not isinstance(obj, BNode):
                self.write("\n" + self.indent(1))
            self.path(obj, OBJECT, newline=True)
        self.depth -= depthmod

[nicholascar]

I think this issue has been addressed by PR #2700 but that fix is currently only in the HEAD of this repo, not an RDFlib release yet. It should appear in 7.0.1 or 7.1.0 in the next few weeks when we make that release which will fix a bunch of small things.

[mschiedon]

I think this issue has been addressed by PR #2700 but that fix is currently only in the HEAD of this repo, not an RDFlib release yet. It should appear in 7.0.1 or 7.1.0 in the next few weeks when we make that release which will fix a bunch of small things.

Excellent, thank you! I can confirm this addresses the issue. Looking forward to the next rdflib release then 👍

[danielpcampagna]

Are there any updates on this issue?

[mschiedon]

@nicholascar The release notes of rdflib 7.1.0 mention this:

2024-03-12 - Fix LongTurtle multi-BN object serialization bug #2700

The code changes from that fix ought to include something like this:

But I don't think I see that fix reflected in the code of the 7.1.0 release.

Could you please comment/investigate?

[edmondchuc]

I think the changes made in that PR were inadvertently reverted by beda4fa in PR #2731.

From memory, mypy was failing and complaining that the else block was unreachable code. We may have to re-submit the PR again and add some annotations to instruct mypy to ignore the error.

[mschiedon]

I think the changes made in that PR were inadvertently reverted by beda4fa in PR #2731.

From memory, mypy was failing and complaining that the else block was unreachable code. We may have to re-submit the PR again and add some annotations to instruct mypy to ignore the error.

Yes, that may be the commit where the fix was regressed. If the mypy complaint was about the elif in below code, that was probably a legitimate complaint, since something that is not a BNode would include an IRI and Literal, so the elif would then never trigger for the Literal.

            if not isinstance(objects[0], BNode):
                self.write("\n" + self.indent(1))
            elif isinstance(objects[0], Literal):
                self.write(" ")

Note that the above code would not have fixed the missing space separator between the predicate and the blank node identifier. The code from #2700 would, which I would hope (expect) mypy would not have an issue with:

            if not isinstance(objects[0], BNode):
                self.write("\n" + self.indent(1))
            else:
                self.write(" ")