Code reuse, code querying and computer aided programming are some of the main research challenges in software engineering. Therefore, we have introduced CodeOntology as an approach to leverage recent advances in the Semantic Web area and the impressive amount of open source code freely available online, to provide a semantic view of software systems by extracting structured information from source code and by performing named entity disambiguation on the comments provided within the code, in order to link the corresponding entities to pertinent DBpedia resources. In this paper, we focus on the expressiveness of this framework by showing how CodeOntology can be used for static code analysis, semantic component search and code reuse.
Recent research in software engineering is focusing towards graph-based
approaches to model software architecture and software process. For instance, in
[
Details about the implementation of the parser and the design of the ontology
have been provided in [
The parser provided by CodeOntology is capable of analyzing the structure of Java projects to generate RDF triples. The input of this process can either be (i) a text le containing Java source code, (ii) the root directory of a Java project or (iii) a JAR le aggregating many Java class les. The output mainly contains triples about structural information common to all object-oriented programming languages, like class hierarchy or the RDF serialization of the underlying structure of each class. The parser has been successfully applied to the OpenJDK 8 source code1, generating a data set which consists of about 2 million RDF triples. Figure 1 shows some information about the structure of this data set.
As we can see, actual source code as literals and literal comments are less than 10% of the total number of extracted RDF triples, while structural information about source code covers almost the 80% of the data set. This knowledge base is available at https://doi.org/10.5281/zenodo.818116 and can be queried at http://codeontology.org/sparql.
The parser supports both Maven and Gradle projects. This allows to download the dependencies of the input project in the form of JAR les that can be optionally analyzed and serialized into RDF triples. The parser, along with a detailed tutorial about how to apply it on di erent kinds of Java projects, is available on GitHub2. 1 http://openjdk.java.net/ 2 https://github.com/codeontology/parser
CodeOntology allows to leverage a powerful language like SPARQL to run highly expressive queries over source code. When a method m references another resource r, then the parser is able to serialize this information into a triple of the form: m woc:references r. We can use this property to easily select recursive methods by means of the following simple query.
SELECT ? method WHERE { ? method a woc : Method ;
woc : references + ? method . } A more interesting example which makes use of the same property is given by the following query, which selects the classes that turn out to be the most referenced ones by the methods of the other classes.
SELECT ? class ( COUNT ( DISTINCT ? anotherClass ) as ? count ) WHERE { ? class a woc : Class . ? method a woc : Method ; woc : isDeclaredBy ? anotherClass ; woc : references ? class .
FILTER (? class != ? anotherClass ) } GROUP BY ? class ORDER BY DESC (? count ) Unsurprisingly, the most referenced class in OpenJDK is the java.lang.String class, followed by the classes java.lang.Object and java.io.IOException.
Another important use case of CodeOntology is undoubtedly the semantic retrieval of software components. For instance, we can exploit DBpedia links to select all methods computing the cube root of a parameter of type double. SELECT ? method WHERE { ? method a woc : Method ; woc : hasParameter / woc : hasType woc : Double ; dul : associatedWith dbpedia : Cube_root .
The execution of this query against the data set extracted from OpenJDK yields two methods, namely the method cbrt(double) declared by the class java. lang.Math and the method cbrt(double), declared by the class java.lang. StrictMath. Another example is given by the following query, which selects all resources associated with public-key cryptography and, in particular, with RSA. SELECT ?r WHERE { ?r dul : associatedWith dbpedia : Public - key_cryptography , dbpedia : RSA_ \( cryptosystem \) .
CodeOntology allows also to run other interesting queries for di erent purposes,
such as detecting the implementation of a speci ed design pattern or computing
}
}
software metrics, like the well-known CK metrics [
Sort classes by the number of subclasses Select classes implementing the Singleton pattern Select classes implementing the Builder pattern Select classes implementing the Factory pattern Select methods to read/write Zip les
Select methods to read an image at a speci ed URL
CodeOntology is an open community-shared resource which aims at enabling the RDF representation, from coarse to ne grain, of the structure and relations found within source code. This way, it is possible to precisely search speci c software components using expressive SPARQL queries, some of which will be showcased during the demo. A video showing how to build and use the parser, as well as the execution of some query examples, is available at: https://www. youtube.com/watch?v=bd6pvUDy8kA.
Acknowledgments. This work was supported in part by a 2015 Google Faculty Research Award and Sardegna Ricerche (project OKgraph, CRP 120).