Flora Parser Bugs

[abhishekreddyallu]

FLORA parser workflows have optional-file guard and dataset-shape inconsistencies

1. OpenEA attribute file guard is inverted

Location: ontoaligner/ontology/kg.py:1028

Current code:

if not attribute_path:
            with open(attribute_path, 'r', encoding='UTF-8') as attribute_file:

This skips valid attribute files and tries to open None when no attribute file is provided.

Observed before fix:

Case 1: attribute path is provided
BUG: no FileNotFoundError, so attribute_path was ignored

Case 2: attribute path is None
BUG: TypeError expected str, bytes or os.PathLike object, not NoneType

Fix:

if attribute_path:
      with open(attribute_path, 'r', encoding='UTF-8') as attribute_file:

---

2. OpenEA dataset output shape breaks FLORAEncoder

Location: ontoaligner/ontology/kg.py:1644-1645

Current code:

"source": source_kg,
"target": target_kg,

But FLORAEncoder expects:

kwargs["source"][0]["graph"]
kwargs["target"][0]["graph"]

Observed before fix:

source type: <class 'dict'>
target type: <class 'dict'>
FAIL: KeyError 0

Fix:

"source": [source_kg],
"target": [target_kg],

Verified after fix:

source type: <class 'list'>
target type: <class 'list'>
PASS: FLORAEncoder accepted OpenEA dataset output
encoded length: 2
encoded[0] type: <class 'ontoaligner.ontology.kg.Graph'>
encoded[1] type: <class 'ontoaligner.ontology.kg.Graph'>

---

3. DBpedia optional translated names and attributes use inverted guards

Locations: ontoaligner/ontology/kg.py:1188, ontoaligner/ontology/kg.py:1222

Current code:

if (not translated_name_path) and ent_ids_path.endswith('_1'):
    ent_name_trans = {}
                with open(translated_name_path, encoding='UTF-8') as f2:

and:

if not att_triples_path:
   for triple in parse_turtle_triples(att_triples_path):

This ignores provided optional files and crashes when optional paths are missing.

Observed before fix:

translated label loaded: False
attribute triple loaded: False
FAIL: TypeError expected str, bytes or os.PathLike object, not NoneType

Fix:

if translated_name_path and ent_ids_path.endswith('_1'):

and:

if att_triples_path:

Verified after fix:

translated label loaded: True
attribute triple loaded: True
PASS: optional paths did not crash

---

4. Graph attribute detection checks subjects instead of literal objects

Location: ontoaligner/ontology/kg.py:298-300

Current code:

def is_attribute(self, pred):
    """
    Return ``True`` if at least one object of ``pred`` is a literal.

    Args:
        pred (str): Predicate IRI to test.

    Returns:
        bool: ``True`` if the predicate has any literal objects.
    """
    if pred in self.relindex:
        for literal in self.relindex[pred]:
            if is_literal(literal):
                return True
    return False

Graph.relindex is structured as:

{predicate: {subject: {objects}}}

So this loop:

for literal in self.relindex[pred]:

actually iterates over the subjects, not the object values. The variable name literal is misleading.

Observed before fix:

relindex[predicate]: {'http://example.org/material1': {'"Steel"'}}
is_attribute(predicate): False
AssertionError: BUG: Graph.is_attribute() returned False even though the predicate has a literal object.

The graph contains a literal object:

"Steel"

but is_attribute() checks the subject:

http://example.org/material1

so the predicate is incorrectly treated as non-attribute.

Fix:

Iterate one level deeper and test the actual objects:

def is_attribute(self, pred):
    """
    Return ``True`` if at least one object of ``pred`` is a literal.

    Args:
        pred (str): Predicate IRI to test.

    Returns:
        bool: ``True`` if the predicate has any literal objects.
    """
    if pred in self.relindex:
        for subject in self.relindex[pred]:
            for literal in self.relindex[pred][subject]:
                if is_literal(literal):
                    return True
    return False

Verified after fix:

relindex[predicate]: {'http://example.org/material1': {'"Steel"'}}
is_attribute(predicate): True
PASS: Graph.is_attribute() correctly detects literal objects.