Data types are not being written on serialization

[tcmitchell]

When dateTime types are written to RDF/XML the data type is not included. This probably happens on other types besides dateTime but I don't have specific examples.

We need to modify SBOL2Serialize.py to include the data type of Literals when it is present.

This is one of those issues that is going to pop up because SBOL 2 requires a special serialization and we cannot use the default provided by RDFLib. SBOL 3 fixes this issue.

[bbartley]

Hi,
I'm not sure that data types are supposed to be serialized. There are a couple of serialization examples of XML DateTimes in the 2.3.0 spec that show datetime serialization without the rdf:DataType. See secs. 12.7 and 13.1.1

[tcmitchell]

SBOL 2.3.0 section 13.1.1 is pretty convincing that data types should NOT be included. It says:

The serialization of an Activity MUST have the following form

And then gives the following example, which does not have data type specified:

<prov:Activity rdf:about="...">
    ... properties inherited from identified ...
    zero or more <prov:qualifiedAssociation>
                    <prov:Association rdf:about="...">...</prov:Association>
                 </prov:qualifiedAssociation> elements
    zero or more <prov:qualifiedUsage>
                    <prov:Usage rdf:about="...">...</prov:Usage>
                </prov:qualifiedUsage> elements
    zero or more <sbol:type rdf:resource="... "/> elements
    zero or one <prov:startedAtTime>...</prov:startedAtTime> element
    zero or one <prov:endedAtTime>...</prov:startedAtTime> element
    zero or more <prov:wasInformedBy rdf:resource="..."/> element
</prov:Activity>

It sounds like the behavior we have now, which is not including the datatype, is the specified syntax, and including it may be a bug.

Additionally, all examples of endedAtTime in SBOLTestSuite/SBOL2* do not include a datatype attribute.

[tcmitchell]

Here is a candidate fix in case we want it in the future:

diff --git a/sbol2/SBOL2Serialize.py b/sbol2/SBOL2Serialize.py
index c012d7d..2300f3d 100644
--- a/sbol2/SBOL2Serialize.py
+++ b/sbol2/SBOL2Serialize.py
@@ -111,7 +111,11 @@ def serialize_sboll2(g):
                 QName(rdfNS, 'resource'): obj.toPython()
             })
         elif isinstance(obj, Literal):
-            elem = etree.SubElement(element, prefixify(predicate, prefixes, True))
+            attrib = None
+            if obj.datatype:
+                attrib = {QName(rdfNS, 'datatype'): obj.datatype.toPython()}
+            elem = etree.SubElement(element, prefixify(predicate, prefixes, True),
+                                    attrib=attrib)
             elem.text = obj
         else:
             raise Exception()

[cjmyers]

I would not infer from the lack of datatype in the serialization that it is not allowed. This is indeed one of the problems with including serialization examples in the document, which we no longer do in SBOL3. I would interpret the lack of mention is not that is disallowed, but that it does not say if is or is not required. Namely, it is optional. I think though it is good to include as it is more precise.