Complex networks and link streams for the empirical analysis of large software

Large software may be modeled as graphs in several ways. For instance, nodes may represent modules, objects or functions, and links may encode dependencies between them, calls, heritage, etc. One may then study a large software through such graphs, called complex networks because they have no strong mathematical properties. Studying them sheds much light on the structure of the considered software. If one turns to the analysis of the dynamics of large software, like execution traces, then the considered graphs evolve over time. This raises challenging issues, as there is currently no clear way to study such objects. We develop a new approach consisting in modeling traces as link streams, i.e. series of triplets (t,a,b) meaning that a and b interacted at time t. For instance, such a triplet may model a call between two modules at run time. Analyzing such streams directly turns out to be much easier and powerful than transforming them into dynamic graphs that poorly capture their dynamics. We present our work on this topic, with directions for applications in software analysis.